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 <api/fs/tracing_path.h> 19#ifdef HAVE_LIBBPF_SUPPORT 20#include <bpf/bpf.h> 21#include <bpf/libbpf.h> 22#ifdef HAVE_BPF_SKEL 23#include "bpf_skel/augmented_raw_syscalls.skel.h" 24#endif 25#endif 26#include "util/bpf_map.h" 27#include "util/rlimit.h" 28#include "builtin.h" 29#include "util/cgroup.h" 30#include "util/color.h" 31#include "util/config.h" 32#include "util/debug.h" 33#include "util/dso.h" 34#include "util/env.h" 35#include "util/event.h" 36#include "util/evsel.h" 37#include "util/evsel_fprintf.h" 38#include "util/synthetic-events.h" 39#include "util/evlist.h" 40#include "util/evswitch.h" 41#include "util/mmap.h" 42#include <subcmd/pager.h> 43#include <subcmd/exec-cmd.h> 44#include "util/machine.h" 45#include "util/map.h" 46#include "util/symbol.h" 47#include "util/path.h" 48#include "util/session.h" 49#include "util/thread.h" 50#include <subcmd/parse-options.h> 51#include "util/strlist.h" 52#include "util/intlist.h" 53#include "util/thread_map.h" 54#include "util/stat.h" 55#include "util/tool.h" 56#include "util/util.h" 57#include "trace/beauty/beauty.h" 58#include "trace-event.h" 59#include "util/parse-events.h" 60#include "util/tracepoint.h" 61#include "callchain.h" 62#include "print_binary.h" 63#include "string2.h" 64#include "syscalltbl.h" 65#include "rb_resort.h" 66#include "../perf.h" 67 68#include <errno.h> 69#include <inttypes.h> 70#include <poll.h> 71#include <signal.h> 72#include <stdlib.h> 73#include <string.h> 74#include <linux/err.h> 75#include <linux/filter.h> 76#include <linux/kernel.h> 77#include <linux/list_sort.h> 78#include <linux/random.h> 79#include <linux/stringify.h> 80#include <linux/time64.h> 81#include <linux/zalloc.h> 82#include <fcntl.h> 83#include <sys/sysmacros.h> 84 85#include <linux/ctype.h> 86#include <perf/mmap.h> 87 88#ifdef HAVE_LIBTRACEEVENT 89#include <traceevent/event-parse.h> 90#endif 91 92#ifndef O_CLOEXEC 93# define O_CLOEXEC 02000000 94#endif 95 96#ifndef F_LINUX_SPECIFIC_BASE 97# define F_LINUX_SPECIFIC_BASE 1024 98#endif 99 100#define RAW_SYSCALL_ARGS_NUM 6 101 102/* 103 * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100 104 */ 105struct syscall_arg_fmt { 106 size_t (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg); 107 bool (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val); 108 unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val); 109 void *parm; 110 const char *name; 111 u16 nr_entries; // for arrays 112 bool show_zero; 113}; 114 115struct syscall_fmt { 116 const char *name; 117 const char *alias; 118 struct { 119 const char *sys_enter, 120 *sys_exit; 121 } bpf_prog_name; 122 struct syscall_arg_fmt arg[RAW_SYSCALL_ARGS_NUM]; 123 u8 nr_args; 124 bool errpid; 125 bool timeout; 126 bool hexret; 127}; 128 129struct trace { 130 struct perf_tool tool; 131 struct syscalltbl *sctbl; 132 struct { 133 struct syscall *table; 134 struct { 135 struct evsel *sys_enter, 136 *sys_exit, 137 *bpf_output; 138 } events; 139 } syscalls; 140#ifdef HAVE_BPF_SKEL 141 struct augmented_raw_syscalls_bpf *skel; 142#endif 143 struct record_opts opts; 144 struct evlist *evlist; 145 struct machine *host; 146 struct thread *current; 147 struct cgroup *cgroup; 148 u64 base_time; 149 FILE *output; 150 unsigned long nr_events; 151 unsigned long nr_events_printed; 152 unsigned long max_events; 153 struct evswitch evswitch; 154 struct strlist *ev_qualifier; 155 struct { 156 size_t nr; 157 int *entries; 158 } ev_qualifier_ids; 159 struct { 160 size_t nr; 161 pid_t *entries; 162 struct bpf_map *map; 163 } filter_pids; 164 double duration_filter; 165 double runtime_ms; 166 struct { 167 u64 vfs_getname, 168 proc_getname; 169 } stats; 170 unsigned int max_stack; 171 unsigned int min_stack; 172 int raw_augmented_syscalls_args_size; 173 bool raw_augmented_syscalls; 174 bool fd_path_disabled; 175 bool sort_events; 176 bool not_ev_qualifier; 177 bool live; 178 bool full_time; 179 bool sched; 180 bool multiple_threads; 181 bool summary; 182 bool summary_only; 183 bool errno_summary; 184 bool failure_only; 185 bool show_comm; 186 bool print_sample; 187 bool show_tool_stats; 188 bool trace_syscalls; 189 bool libtraceevent_print; 190 bool kernel_syscallchains; 191 s16 args_alignment; 192 bool show_tstamp; 193 bool show_duration; 194 bool show_zeros; 195 bool show_arg_names; 196 bool show_string_prefix; 197 bool force; 198 bool vfs_getname; 199 int trace_pgfaults; 200 char *perfconfig_events; 201 struct { 202 struct ordered_events data; 203 u64 last; 204 } oe; 205}; 206 207struct tp_field { 208 int offset; 209 union { 210 u64 (*integer)(struct tp_field *field, struct perf_sample *sample); 211 void *(*pointer)(struct tp_field *field, struct perf_sample *sample); 212 }; 213}; 214 215#define TP_UINT_FIELD(bits) \ 216static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \ 217{ \ 218 u##bits value; \ 219 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \ 220 return value; \ 221} 222 223TP_UINT_FIELD(8); 224TP_UINT_FIELD(16); 225TP_UINT_FIELD(32); 226TP_UINT_FIELD(64); 227 228#define TP_UINT_FIELD__SWAPPED(bits) \ 229static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \ 230{ \ 231 u##bits value; \ 232 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \ 233 return bswap_##bits(value);\ 234} 235 236TP_UINT_FIELD__SWAPPED(16); 237TP_UINT_FIELD__SWAPPED(32); 238TP_UINT_FIELD__SWAPPED(64); 239 240static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap) 241{ 242 field->offset = offset; 243 244 switch (size) { 245 case 1: 246 field->integer = tp_field__u8; 247 break; 248 case 2: 249 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16; 250 break; 251 case 4: 252 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32; 253 break; 254 case 8: 255 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64; 256 break; 257 default: 258 return -1; 259 } 260 261 return 0; 262} 263 264static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap) 265{ 266 return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap); 267} 268 269static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample) 270{ 271 return sample->raw_data + field->offset; 272} 273 274static int __tp_field__init_ptr(struct tp_field *field, int offset) 275{ 276 field->offset = offset; 277 field->pointer = tp_field__ptr; 278 return 0; 279} 280 281static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field) 282{ 283 return __tp_field__init_ptr(field, format_field->offset); 284} 285 286struct syscall_tp { 287 struct tp_field id; 288 union { 289 struct tp_field args, ret; 290 }; 291}; 292 293/* 294 * The evsel->priv as used by 'perf trace' 295 * sc: for raw_syscalls:sys_{enter,exit} and syscalls:sys_{enter,exit}_SYSCALLNAME 296 * fmt: for all the other tracepoints 297 */ 298struct evsel_trace { 299 struct syscall_tp sc; 300 struct syscall_arg_fmt *fmt; 301}; 302 303static struct evsel_trace *evsel_trace__new(void) 304{ 305 return zalloc(sizeof(struct evsel_trace)); 306} 307 308static void evsel_trace__delete(struct evsel_trace *et) 309{ 310 if (et == NULL) 311 return; 312 313 zfree(&et->fmt); 314 free(et); 315} 316 317/* 318 * Used with raw_syscalls:sys_{enter,exit} and with the 319 * syscalls:sys_{enter,exit}_SYSCALL tracepoints 320 */ 321static inline struct syscall_tp *__evsel__syscall_tp(struct evsel *evsel) 322{ 323 struct evsel_trace *et = evsel->priv; 324 325 return &et->sc; 326} 327 328static struct syscall_tp *evsel__syscall_tp(struct evsel *evsel) 329{ 330 if (evsel->priv == NULL) { 331 evsel->priv = evsel_trace__new(); 332 if (evsel->priv == NULL) 333 return NULL; 334 } 335 336 return __evsel__syscall_tp(evsel); 337} 338 339/* 340 * Used with all the other tracepoints. 341 */ 342static inline struct syscall_arg_fmt *__evsel__syscall_arg_fmt(struct evsel *evsel) 343{ 344 struct evsel_trace *et = evsel->priv; 345 346 return et->fmt; 347} 348 349static struct syscall_arg_fmt *evsel__syscall_arg_fmt(struct evsel *evsel) 350{ 351 struct evsel_trace *et = evsel->priv; 352 353 if (evsel->priv == NULL) { 354 et = evsel->priv = evsel_trace__new(); 355 356 if (et == NULL) 357 return NULL; 358 } 359 360 if (et->fmt == NULL) { 361 et->fmt = calloc(evsel->tp_format->format.nr_fields, sizeof(struct syscall_arg_fmt)); 362 if (et->fmt == NULL) 363 goto out_delete; 364 } 365 366 return __evsel__syscall_arg_fmt(evsel); 367 368out_delete: 369 evsel_trace__delete(evsel->priv); 370 evsel->priv = NULL; 371 return NULL; 372} 373 374static int evsel__init_tp_uint_field(struct evsel *evsel, struct tp_field *field, const char *name) 375{ 376 struct tep_format_field *format_field = evsel__field(evsel, name); 377 378 if (format_field == NULL) 379 return -1; 380 381 return tp_field__init_uint(field, format_field, evsel->needs_swap); 382} 383 384#define perf_evsel__init_sc_tp_uint_field(evsel, name) \ 385 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\ 386 evsel__init_tp_uint_field(evsel, &sc->name, #name); }) 387 388static int evsel__init_tp_ptr_field(struct evsel *evsel, struct tp_field *field, const char *name) 389{ 390 struct tep_format_field *format_field = evsel__field(evsel, name); 391 392 if (format_field == NULL) 393 return -1; 394 395 return tp_field__init_ptr(field, format_field); 396} 397 398#define perf_evsel__init_sc_tp_ptr_field(evsel, name) \ 399 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\ 400 evsel__init_tp_ptr_field(evsel, &sc->name, #name); }) 401 402static void evsel__delete_priv(struct evsel *evsel) 403{ 404 zfree(&evsel->priv); 405 evsel__delete(evsel); 406} 407 408static int evsel__init_syscall_tp(struct evsel *evsel) 409{ 410 struct syscall_tp *sc = evsel__syscall_tp(evsel); 411 412 if (sc != NULL) { 413 if (evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") && 414 evsel__init_tp_uint_field(evsel, &sc->id, "nr")) 415 return -ENOENT; 416 417 return 0; 418 } 419 420 return -ENOMEM; 421} 422 423static int evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp) 424{ 425 struct syscall_tp *sc = evsel__syscall_tp(evsel); 426 427 if (sc != NULL) { 428 struct tep_format_field *syscall_id = evsel__field(tp, "id"); 429 if (syscall_id == NULL) 430 syscall_id = evsel__field(tp, "__syscall_nr"); 431 if (syscall_id == NULL || 432 __tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap)) 433 return -EINVAL; 434 435 return 0; 436 } 437 438 return -ENOMEM; 439} 440 441static int evsel__init_augmented_syscall_tp_args(struct evsel *evsel) 442{ 443 struct syscall_tp *sc = __evsel__syscall_tp(evsel); 444 445 return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)); 446} 447 448static int evsel__init_augmented_syscall_tp_ret(struct evsel *evsel) 449{ 450 struct syscall_tp *sc = __evsel__syscall_tp(evsel); 451 452 return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap); 453} 454 455static int evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler) 456{ 457 if (evsel__syscall_tp(evsel) != NULL) { 458 if (perf_evsel__init_sc_tp_uint_field(evsel, id)) 459 return -ENOENT; 460 461 evsel->handler = handler; 462 return 0; 463 } 464 465 return -ENOMEM; 466} 467 468static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler) 469{ 470 struct evsel *evsel = evsel__newtp("raw_syscalls", direction); 471 472 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */ 473 if (IS_ERR(evsel)) 474 evsel = evsel__newtp("syscalls", direction); 475 476 if (IS_ERR(evsel)) 477 return NULL; 478 479 if (evsel__init_raw_syscall_tp(evsel, handler)) 480 goto out_delete; 481 482 return evsel; 483 484out_delete: 485 evsel__delete_priv(evsel); 486 return NULL; 487} 488 489#define perf_evsel__sc_tp_uint(evsel, name, sample) \ 490 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \ 491 fields->name.integer(&fields->name, sample); }) 492 493#define perf_evsel__sc_tp_ptr(evsel, name, sample) \ 494 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \ 495 fields->name.pointer(&fields->name, sample); }) 496 497size_t strarray__scnprintf_suffix(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_suffix, int val) 498{ 499 int idx = val - sa->offset; 500 501 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) { 502 size_t printed = scnprintf(bf, size, intfmt, val); 503 if (show_suffix) 504 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix); 505 return printed; 506 } 507 508 return scnprintf(bf, size, "%s%s", sa->entries[idx], show_suffix ? sa->prefix : ""); 509} 510 511size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val) 512{ 513 int idx = val - sa->offset; 514 515 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) { 516 size_t printed = scnprintf(bf, size, intfmt, val); 517 if (show_prefix) 518 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix); 519 return printed; 520 } 521 522 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]); 523} 524 525static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size, 526 const char *intfmt, 527 struct syscall_arg *arg) 528{ 529 return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val); 530} 531 532static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size, 533 struct syscall_arg *arg) 534{ 535 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg); 536} 537 538#define SCA_STRARRAY syscall_arg__scnprintf_strarray 539 540bool syscall_arg__strtoul_strarray(char *bf, size_t size, struct syscall_arg *arg, u64 *ret) 541{ 542 return strarray__strtoul(arg->parm, bf, size, ret); 543} 544 545bool syscall_arg__strtoul_strarray_flags(char *bf, size_t size, struct syscall_arg *arg, u64 *ret) 546{ 547 return strarray__strtoul_flags(arg->parm, bf, size, ret); 548} 549 550bool syscall_arg__strtoul_strarrays(char *bf, size_t size, struct syscall_arg *arg, u64 *ret) 551{ 552 return strarrays__strtoul(arg->parm, bf, size, ret); 553} 554 555size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg) 556{ 557 return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val); 558} 559 560size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val) 561{ 562 size_t printed; 563 int i; 564 565 for (i = 0; i < sas->nr_entries; ++i) { 566 struct strarray *sa = sas->entries[i]; 567 int idx = val - sa->offset; 568 569 if (idx >= 0 && idx < sa->nr_entries) { 570 if (sa->entries[idx] == NULL) 571 break; 572 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]); 573 } 574 } 575 576 printed = scnprintf(bf, size, intfmt, val); 577 if (show_prefix) 578 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix); 579 return printed; 580} 581 582bool strarray__strtoul(struct strarray *sa, char *bf, size_t size, u64 *ret) 583{ 584 int i; 585 586 for (i = 0; i < sa->nr_entries; ++i) { 587 if (sa->entries[i] && strncmp(sa->entries[i], bf, size) == 0 && sa->entries[i][size] == '\0') { 588 *ret = sa->offset + i; 589 return true; 590 } 591 } 592 593 return false; 594} 595 596bool strarray__strtoul_flags(struct strarray *sa, char *bf, size_t size, u64 *ret) 597{ 598 u64 val = 0; 599 char *tok = bf, *sep, *end; 600 601 *ret = 0; 602 603 while (size != 0) { 604 int toklen = size; 605 606 sep = memchr(tok, '|', size); 607 if (sep != NULL) { 608 size -= sep - tok + 1; 609 610 end = sep - 1; 611 while (end > tok && isspace(*end)) 612 --end; 613 614 toklen = end - tok + 1; 615 } 616 617 while (isspace(*tok)) 618 ++tok; 619 620 if (isalpha(*tok) || *tok == '_') { 621 if (!strarray__strtoul(sa, tok, toklen, &val)) 622 return false; 623 } else 624 val = strtoul(tok, NULL, 0); 625 626 *ret |= (1 << (val - 1)); 627 628 if (sep == NULL) 629 break; 630 tok = sep + 1; 631 } 632 633 return true; 634} 635 636bool strarrays__strtoul(struct strarrays *sas, char *bf, size_t size, u64 *ret) 637{ 638 int i; 639 640 for (i = 0; i < sas->nr_entries; ++i) { 641 struct strarray *sa = sas->entries[i]; 642 643 if (strarray__strtoul(sa, bf, size, ret)) 644 return true; 645 } 646 647 return false; 648} 649 650size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size, 651 struct syscall_arg *arg) 652{ 653 return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val); 654} 655 656#ifndef AT_FDCWD 657#define AT_FDCWD -100 658#endif 659 660static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size, 661 struct syscall_arg *arg) 662{ 663 int fd = arg->val; 664 const char *prefix = "AT_FD"; 665 666 if (fd == AT_FDCWD) 667 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD"); 668 669 return syscall_arg__scnprintf_fd(bf, size, arg); 670} 671 672#define SCA_FDAT syscall_arg__scnprintf_fd_at 673 674static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size, 675 struct syscall_arg *arg); 676 677#define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd 678 679size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg) 680{ 681 return scnprintf(bf, size, "%#lx", arg->val); 682} 683 684size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg) 685{ 686 if (arg->val == 0) 687 return scnprintf(bf, size, "NULL"); 688 return syscall_arg__scnprintf_hex(bf, size, arg); 689} 690 691size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg) 692{ 693 return scnprintf(bf, size, "%d", arg->val); 694} 695 696size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg) 697{ 698 return scnprintf(bf, size, "%ld", arg->val); 699} 700 701static size_t syscall_arg__scnprintf_char_array(char *bf, size_t size, struct syscall_arg *arg) 702{ 703 // XXX Hey, maybe for sched:sched_switch prev/next comm fields we can 704 // fill missing comms using thread__set_comm()... 705 // here or in a special syscall_arg__scnprintf_pid_sched_tp... 706 return scnprintf(bf, size, "\"%-.*s\"", arg->fmt->nr_entries ?: arg->len, arg->val); 707} 708 709#define SCA_CHAR_ARRAY syscall_arg__scnprintf_char_array 710 711static const char *bpf_cmd[] = { 712 "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM", 713 "MAP_GET_NEXT_KEY", "PROG_LOAD", "OBJ_PIN", "OBJ_GET", "PROG_ATTACH", 714 "PROG_DETACH", "PROG_TEST_RUN", "PROG_GET_NEXT_ID", "MAP_GET_NEXT_ID", 715 "PROG_GET_FD_BY_ID", "MAP_GET_FD_BY_ID", "OBJ_GET_INFO_BY_FD", 716 "PROG_QUERY", "RAW_TRACEPOINT_OPEN", "BTF_LOAD", "BTF_GET_FD_BY_ID", 717 "TASK_FD_QUERY", "MAP_LOOKUP_AND_DELETE_ELEM", "MAP_FREEZE", 718 "BTF_GET_NEXT_ID", "MAP_LOOKUP_BATCH", "MAP_LOOKUP_AND_DELETE_BATCH", 719 "MAP_UPDATE_BATCH", "MAP_DELETE_BATCH", "LINK_CREATE", "LINK_UPDATE", 720 "LINK_GET_FD_BY_ID", "LINK_GET_NEXT_ID", "ENABLE_STATS", "ITER_CREATE", 721 "LINK_DETACH", "PROG_BIND_MAP", 722}; 723static DEFINE_STRARRAY(bpf_cmd, "BPF_"); 724 725static const char *fsmount_flags[] = { 726 [1] = "CLOEXEC", 727}; 728static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_"); 729 730#include "trace/beauty/generated/fsconfig_arrays.c" 731 732static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_"); 733 734static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", }; 735static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1); 736 737static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", }; 738static DEFINE_STRARRAY(itimers, "ITIMER_"); 739 740static const char *keyctl_options[] = { 741 "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN", 742 "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ", 743 "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT", 744 "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT", 745 "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT", 746}; 747static DEFINE_STRARRAY(keyctl_options, "KEYCTL_"); 748 749static const char *whences[] = { "SET", "CUR", "END", 750#ifdef SEEK_DATA 751"DATA", 752#endif 753#ifdef SEEK_HOLE 754"HOLE", 755#endif 756}; 757static DEFINE_STRARRAY(whences, "SEEK_"); 758 759static const char *fcntl_cmds[] = { 760 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK", 761 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64", 762 "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX", 763 "GETOWNER_UIDS", 764}; 765static DEFINE_STRARRAY(fcntl_cmds, "F_"); 766 767static const char *fcntl_linux_specific_cmds[] = { 768 "SETLEASE", "GETLEASE", "NOTIFY", "DUPFD_QUERY", [5] = "CANCELLK", "DUPFD_CLOEXEC", 769 "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS", 770 "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT", 771}; 772 773static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE); 774 775static struct strarray *fcntl_cmds_arrays[] = { 776 &strarray__fcntl_cmds, 777 &strarray__fcntl_linux_specific_cmds, 778}; 779 780static DEFINE_STRARRAYS(fcntl_cmds_arrays); 781 782static const char *rlimit_resources[] = { 783 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE", 784 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO", 785 "RTTIME", 786}; 787static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_"); 788 789static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", }; 790static DEFINE_STRARRAY(sighow, "SIG_"); 791 792static const char *clockid[] = { 793 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID", 794 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME", 795 "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI" 796}; 797static DEFINE_STRARRAY(clockid, "CLOCK_"); 798 799static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size, 800 struct syscall_arg *arg) 801{ 802 bool show_prefix = arg->show_string_prefix; 803 const char *suffix = "_OK"; 804 size_t printed = 0; 805 int mode = arg->val; 806 807 if (mode == F_OK) /* 0 */ 808 return scnprintf(bf, size, "F%s", show_prefix ? suffix : ""); 809#define P_MODE(n) \ 810 if (mode & n##_OK) { \ 811 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \ 812 mode &= ~n##_OK; \ 813 } 814 815 P_MODE(R); 816 P_MODE(W); 817 P_MODE(X); 818#undef P_MODE 819 820 if (mode) 821 printed += scnprintf(bf + printed, size - printed, "|%#x", mode); 822 823 return printed; 824} 825 826#define SCA_ACCMODE syscall_arg__scnprintf_access_mode 827 828static size_t syscall_arg__scnprintf_filename(char *bf, size_t size, 829 struct syscall_arg *arg); 830 831#define SCA_FILENAME syscall_arg__scnprintf_filename 832 833static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size, 834 struct syscall_arg *arg) 835{ 836 bool show_prefix = arg->show_string_prefix; 837 const char *prefix = "O_"; 838 int printed = 0, flags = arg->val; 839 840#define P_FLAG(n) \ 841 if (flags & O_##n) { \ 842 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \ 843 flags &= ~O_##n; \ 844 } 845 846 P_FLAG(CLOEXEC); 847 P_FLAG(NONBLOCK); 848#undef P_FLAG 849 850 if (flags) 851 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags); 852 853 return printed; 854} 855 856#define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags 857 858#ifndef GRND_NONBLOCK 859#define GRND_NONBLOCK 0x0001 860#endif 861#ifndef GRND_RANDOM 862#define GRND_RANDOM 0x0002 863#endif 864 865static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size, 866 struct syscall_arg *arg) 867{ 868 bool show_prefix = arg->show_string_prefix; 869 const char *prefix = "GRND_"; 870 int printed = 0, flags = arg->val; 871 872#define P_FLAG(n) \ 873 if (flags & GRND_##n) { \ 874 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \ 875 flags &= ~GRND_##n; \ 876 } 877 878 P_FLAG(RANDOM); 879 P_FLAG(NONBLOCK); 880#undef P_FLAG 881 882 if (flags) 883 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags); 884 885 return printed; 886} 887 888#define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags 889 890#define STRARRAY(name, array) \ 891 { .scnprintf = SCA_STRARRAY, \ 892 .strtoul = STUL_STRARRAY, \ 893 .parm = &strarray__##array, } 894 895#define STRARRAY_FLAGS(name, array) \ 896 { .scnprintf = SCA_STRARRAY_FLAGS, \ 897 .strtoul = STUL_STRARRAY_FLAGS, \ 898 .parm = &strarray__##array, } 899 900#include "trace/beauty/arch_errno_names.c" 901#include "trace/beauty/eventfd.c" 902#include "trace/beauty/futex_op.c" 903#include "trace/beauty/futex_val3.c" 904#include "trace/beauty/mmap.c" 905#include "trace/beauty/mode_t.c" 906#include "trace/beauty/msg_flags.c" 907#include "trace/beauty/open_flags.c" 908#include "trace/beauty/perf_event_open.c" 909#include "trace/beauty/pid.c" 910#include "trace/beauty/sched_policy.c" 911#include "trace/beauty/seccomp.c" 912#include "trace/beauty/signum.c" 913#include "trace/beauty/socket_type.c" 914#include "trace/beauty/waitid_options.c" 915 916static const struct syscall_fmt syscall_fmts[] = { 917 { .name = "access", 918 .arg = { [1] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, }, 919 { .name = "arch_prctl", 920 .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ }, 921 [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, }, 922 { .name = "bind", 923 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ }, 924 [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ }, 925 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, }, 926 { .name = "bpf", 927 .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, }, 928 { .name = "brk", .hexret = true, 929 .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, }, 930 { .name = "clock_gettime", 931 .arg = { [0] = STRARRAY(clk_id, clockid), }, }, 932 { .name = "clock_nanosleep", 933 .arg = { [2] = { .scnprintf = SCA_TIMESPEC, /* rqtp */ }, }, }, 934 { .name = "clone", .errpid = true, .nr_args = 5, 935 .arg = { [0] = { .name = "flags", .scnprintf = SCA_CLONE_FLAGS, }, 936 [1] = { .name = "child_stack", .scnprintf = SCA_HEX, }, 937 [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, }, 938 [3] = { .name = "child_tidptr", .scnprintf = SCA_HEX, }, 939 [4] = { .name = "tls", .scnprintf = SCA_HEX, }, }, }, 940 { .name = "close", 941 .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, }, 942 { .name = "connect", 943 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ }, 944 [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ }, 945 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, }, 946 { .name = "epoll_ctl", 947 .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, }, 948 { .name = "eventfd2", 949 .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, }, 950 { .name = "faccessat", 951 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, 952 [1] = { .scnprintf = SCA_FILENAME, /* pathname */ }, 953 [2] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, }, 954 { .name = "faccessat2", 955 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, 956 [1] = { .scnprintf = SCA_FILENAME, /* pathname */ }, 957 [2] = { .scnprintf = SCA_ACCMODE, /* mode */ }, 958 [3] = { .scnprintf = SCA_FACCESSAT2_FLAGS, /* flags */ }, }, }, 959 { .name = "fchmodat", 960 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, }, 961 { .name = "fchownat", 962 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, }, 963 { .name = "fcntl", 964 .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */ 965 .strtoul = STUL_STRARRAYS, 966 .parm = &strarrays__fcntl_cmds_arrays, 967 .show_zero = true, }, 968 [2] = { .scnprintf = SCA_FCNTL_ARG, /* arg */ }, }, }, 969 { .name = "flock", 970 .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, }, 971 { .name = "fsconfig", 972 .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, }, 973 { .name = "fsmount", 974 .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags), 975 [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, }, 976 { .name = "fspick", 977 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, 978 [1] = { .scnprintf = SCA_FILENAME, /* path */ }, 979 [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, }, 980 { .name = "fstat", .alias = "newfstat", }, 981 { .name = "futex", 982 .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ }, 983 [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, }, 984 { .name = "futimesat", 985 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, }, 986 { .name = "getitimer", 987 .arg = { [0] = STRARRAY(which, itimers), }, }, 988 { .name = "getpid", .errpid = true, }, 989 { .name = "getpgid", .errpid = true, }, 990 { .name = "getppid", .errpid = true, }, 991 { .name = "getrandom", 992 .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, }, 993 { .name = "getrlimit", 994 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, }, 995 { .name = "getsockopt", 996 .arg = { [1] = STRARRAY(level, socket_level), }, }, 997 { .name = "gettid", .errpid = true, }, 998 { .name = "ioctl", 999 .arg = { 1000#if defined(__i386__) || defined(__x86_64__) 1001/* 1002 * FIXME: Make this available to all arches. 1003 */ 1004 [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ }, 1005 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, }, 1006#else 1007 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, }, 1008#endif 1009 { .name = "kcmp", .nr_args = 5, 1010 .arg = { [0] = { .name = "pid1", .scnprintf = SCA_PID, }, 1011 [1] = { .name = "pid2", .scnprintf = SCA_PID, }, 1012 [2] = { .name = "type", .scnprintf = SCA_KCMP_TYPE, }, 1013 [3] = { .name = "idx1", .scnprintf = SCA_KCMP_IDX, }, 1014 [4] = { .name = "idx2", .scnprintf = SCA_KCMP_IDX, }, }, }, 1015 { .name = "keyctl", 1016 .arg = { [0] = STRARRAY(option, keyctl_options), }, }, 1017 { .name = "kill", 1018 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, }, 1019 { .name = "linkat", 1020 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, }, 1021 { .name = "lseek", 1022 .arg = { [2] = STRARRAY(whence, whences), }, }, 1023 { .name = "lstat", .alias = "newlstat", }, 1024 { .name = "madvise", 1025 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ }, 1026 [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, }, 1027 { .name = "mkdirat", 1028 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, }, 1029 { .name = "mknodat", 1030 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, }, 1031 { .name = "mmap", .hexret = true, 1032/* The standard mmap maps to old_mmap on s390x */ 1033#if defined(__s390x__) 1034 .alias = "old_mmap", 1035#endif 1036 .arg = { [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ }, 1037 [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */ 1038 .strtoul = STUL_STRARRAY_FLAGS, 1039 .parm = &strarray__mmap_flags, }, 1040 [5] = { .scnprintf = SCA_HEX, /* offset */ }, }, }, 1041 { .name = "mount", 1042 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ }, 1043 [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */ 1044 .mask_val = SCAMV_MOUNT_FLAGS, /* flags */ }, }, }, 1045 { .name = "move_mount", 1046 .arg = { [0] = { .scnprintf = SCA_FDAT, /* from_dfd */ }, 1047 [1] = { .scnprintf = SCA_FILENAME, /* from_pathname */ }, 1048 [2] = { .scnprintf = SCA_FDAT, /* to_dfd */ }, 1049 [3] = { .scnprintf = SCA_FILENAME, /* to_pathname */ }, 1050 [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, }, 1051 { .name = "mprotect", 1052 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ }, 1053 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ }, }, }, 1054 { .name = "mq_unlink", 1055 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, }, 1056 { .name = "mremap", .hexret = true, 1057 .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, }, 1058 { .name = "name_to_handle_at", 1059 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, }, 1060 { .name = "nanosleep", 1061 .arg = { [0] = { .scnprintf = SCA_TIMESPEC, /* req */ }, }, }, 1062 { .name = "newfstatat", .alias = "fstatat", 1063 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, 1064 [1] = { .scnprintf = SCA_FILENAME, /* pathname */ }, 1065 [3] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ }, }, }, 1066 { .name = "open", 1067 .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, }, 1068 { .name = "open_by_handle_at", 1069 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, 1070 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, }, 1071 { .name = "openat", 1072 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, 1073 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, }, 1074 { .name = "perf_event_open", 1075 .arg = { [0] = { .scnprintf = SCA_PERF_ATTR, /* attr */ }, 1076 [2] = { .scnprintf = SCA_INT, /* cpu */ }, 1077 [3] = { .scnprintf = SCA_FD, /* group_fd */ }, 1078 [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, }, 1079 { .name = "pipe2", 1080 .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, }, 1081 { .name = "pkey_alloc", 1082 .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS, /* access_rights */ }, }, }, 1083 { .name = "pkey_free", 1084 .arg = { [0] = { .scnprintf = SCA_INT, /* key */ }, }, }, 1085 { .name = "pkey_mprotect", 1086 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ }, 1087 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ }, 1088 [3] = { .scnprintf = SCA_INT, /* pkey */ }, }, }, 1089 { .name = "poll", .timeout = true, }, 1090 { .name = "ppoll", .timeout = true, }, 1091 { .name = "prctl", 1092 .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */ 1093 .strtoul = STUL_STRARRAY, 1094 .parm = &strarray__prctl_options, }, 1095 [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ }, 1096 [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, }, 1097 { .name = "pread", .alias = "pread64", }, 1098 { .name = "preadv", .alias = "pread", }, 1099 { .name = "prlimit64", 1100 .arg = { [1] = STRARRAY(resource, rlimit_resources), }, }, 1101 { .name = "pwrite", .alias = "pwrite64", }, 1102 { .name = "readlinkat", 1103 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, }, 1104 { .name = "recvfrom", 1105 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, }, 1106 { .name = "recvmmsg", 1107 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, }, 1108 { .name = "recvmsg", 1109 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, }, 1110 { .name = "renameat", 1111 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ }, 1112 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, }, 1113 { .name = "renameat2", 1114 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ }, 1115 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, 1116 [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, }, 1117 { .name = "rt_sigaction", 1118 .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, }, 1119 { .name = "rt_sigprocmask", 1120 .arg = { [0] = STRARRAY(how, sighow), }, }, 1121 { .name = "rt_sigqueueinfo", 1122 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, }, 1123 { .name = "rt_tgsigqueueinfo", 1124 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, }, 1125 { .name = "sched_setscheduler", 1126 .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, }, 1127 { .name = "seccomp", 1128 .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP, /* op */ }, 1129 [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, }, 1130 { .name = "select", .timeout = true, }, 1131 { .name = "sendfile", .alias = "sendfile64", }, 1132 { .name = "sendmmsg", 1133 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, }, 1134 { .name = "sendmsg", 1135 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, }, 1136 { .name = "sendto", 1137 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, 1138 [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, }, 1139 { .name = "set_tid_address", .errpid = true, }, 1140 { .name = "setitimer", 1141 .arg = { [0] = STRARRAY(which, itimers), }, }, 1142 { .name = "setrlimit", 1143 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, }, 1144 { .name = "setsockopt", 1145 .arg = { [1] = STRARRAY(level, socket_level), }, }, 1146 { .name = "socket", 1147 .arg = { [0] = STRARRAY(family, socket_families), 1148 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ }, 1149 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, }, 1150 { .name = "socketpair", 1151 .arg = { [0] = STRARRAY(family, socket_families), 1152 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ }, 1153 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, }, 1154 { .name = "stat", .alias = "newstat", }, 1155 { .name = "statx", 1156 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fdat */ }, 1157 [2] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ } , 1158 [3] = { .scnprintf = SCA_STATX_MASK, /* mask */ }, }, }, 1159 { .name = "swapoff", 1160 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, }, 1161 { .name = "swapon", 1162 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, }, 1163 { .name = "symlinkat", 1164 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, }, 1165 { .name = "sync_file_range", 1166 .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, }, 1167 { .name = "tgkill", 1168 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, }, 1169 { .name = "tkill", 1170 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, }, 1171 { .name = "umount2", .alias = "umount", 1172 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, }, 1173 { .name = "uname", .alias = "newuname", }, 1174 { .name = "unlinkat", 1175 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, 1176 [1] = { .scnprintf = SCA_FILENAME, /* pathname */ }, 1177 [2] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ }, }, }, 1178 { .name = "utimensat", 1179 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, }, 1180 { .name = "wait4", .errpid = true, 1181 .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, }, 1182 { .name = "waitid", .errpid = true, 1183 .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, }, 1184}; 1185 1186static int syscall_fmt__cmp(const void *name, const void *fmtp) 1187{ 1188 const struct syscall_fmt *fmt = fmtp; 1189 return strcmp(name, fmt->name); 1190} 1191 1192static const struct syscall_fmt *__syscall_fmt__find(const struct syscall_fmt *fmts, 1193 const int nmemb, 1194 const char *name) 1195{ 1196 return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp); 1197} 1198 1199static const struct syscall_fmt *syscall_fmt__find(const char *name) 1200{ 1201 const int nmemb = ARRAY_SIZE(syscall_fmts); 1202 return __syscall_fmt__find(syscall_fmts, nmemb, name); 1203} 1204 1205static const struct syscall_fmt *__syscall_fmt__find_by_alias(const struct syscall_fmt *fmts, 1206 const int nmemb, const char *alias) 1207{ 1208 int i; 1209 1210 for (i = 0; i < nmemb; ++i) { 1211 if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0) 1212 return &fmts[i]; 1213 } 1214 1215 return NULL; 1216} 1217 1218static const struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias) 1219{ 1220 const int nmemb = ARRAY_SIZE(syscall_fmts); 1221 return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias); 1222} 1223 1224/* 1225 * is_exit: is this "exit" or "exit_group"? 1226 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter. 1227 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc. 1228 * nonexistent: Just a hole in the syscall table, syscall id not allocated 1229 */ 1230struct syscall { 1231 struct tep_event *tp_format; 1232 int nr_args; 1233 int args_size; 1234 struct { 1235 struct bpf_program *sys_enter, 1236 *sys_exit; 1237 } bpf_prog; 1238 bool is_exit; 1239 bool is_open; 1240 bool nonexistent; 1241 struct tep_format_field *args; 1242 const char *name; 1243 const struct syscall_fmt *fmt; 1244 struct syscall_arg_fmt *arg_fmt; 1245}; 1246 1247/* 1248 * We need to have this 'calculated' boolean because in some cases we really 1249 * don't know what is the duration of a syscall, for instance, when we start 1250 * a session and some threads are waiting for a syscall to finish, say 'poll', 1251 * in which case all we can do is to print "( ? ) for duration and for the 1252 * start timestamp. 1253 */ 1254static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp) 1255{ 1256 double duration = (double)t / NSEC_PER_MSEC; 1257 size_t printed = fprintf(fp, "("); 1258 1259 if (!calculated) 1260 printed += fprintf(fp, " "); 1261 else if (duration >= 1.0) 1262 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration); 1263 else if (duration >= 0.01) 1264 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration); 1265 else 1266 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration); 1267 return printed + fprintf(fp, "): "); 1268} 1269 1270/** 1271 * filename.ptr: The filename char pointer that will be vfs_getname'd 1272 * filename.entry_str_pos: Where to insert the string translated from 1273 * filename.ptr by the vfs_getname tracepoint/kprobe. 1274 * ret_scnprintf: syscall args may set this to a different syscall return 1275 * formatter, for instance, fcntl may return fds, file flags, etc. 1276 */ 1277struct thread_trace { 1278 u64 entry_time; 1279 bool entry_pending; 1280 unsigned long nr_events; 1281 unsigned long pfmaj, pfmin; 1282 char *entry_str; 1283 double runtime_ms; 1284 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg); 1285 struct { 1286 unsigned long ptr; 1287 short int entry_str_pos; 1288 bool pending_open; 1289 unsigned int namelen; 1290 char *name; 1291 } filename; 1292 struct { 1293 int max; 1294 struct file *table; 1295 } files; 1296 1297 struct intlist *syscall_stats; 1298}; 1299 1300static struct thread_trace *thread_trace__new(void) 1301{ 1302 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace)); 1303 1304 if (ttrace) { 1305 ttrace->files.max = -1; 1306 ttrace->syscall_stats = intlist__new(NULL); 1307 } 1308 1309 return ttrace; 1310} 1311 1312static void thread_trace__free_files(struct thread_trace *ttrace); 1313 1314static void thread_trace__delete(void *pttrace) 1315{ 1316 struct thread_trace *ttrace = pttrace; 1317 1318 if (!ttrace) 1319 return; 1320 1321 intlist__delete(ttrace->syscall_stats); 1322 ttrace->syscall_stats = NULL; 1323 thread_trace__free_files(ttrace); 1324 zfree(&ttrace->entry_str); 1325 free(ttrace); 1326} 1327 1328static struct thread_trace *thread__trace(struct thread *thread, FILE *fp) 1329{ 1330 struct thread_trace *ttrace; 1331 1332 if (thread == NULL) 1333 goto fail; 1334 1335 if (thread__priv(thread) == NULL) 1336 thread__set_priv(thread, thread_trace__new()); 1337 1338 if (thread__priv(thread) == NULL) 1339 goto fail; 1340 1341 ttrace = thread__priv(thread); 1342 ++ttrace->nr_events; 1343 1344 return ttrace; 1345fail: 1346 color_fprintf(fp, PERF_COLOR_RED, 1347 "WARNING: not enough memory, dropping samples!\n"); 1348 return NULL; 1349} 1350 1351 1352void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg, 1353 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg)) 1354{ 1355 struct thread_trace *ttrace = thread__priv(arg->thread); 1356 1357 ttrace->ret_scnprintf = ret_scnprintf; 1358} 1359 1360#define TRACE_PFMAJ (1 << 0) 1361#define TRACE_PFMIN (1 << 1) 1362 1363static const size_t trace__entry_str_size = 2048; 1364 1365static void thread_trace__free_files(struct thread_trace *ttrace) 1366{ 1367 for (int i = 0; i < ttrace->files.max; ++i) { 1368 struct file *file = ttrace->files.table + i; 1369 zfree(&file->pathname); 1370 } 1371 1372 zfree(&ttrace->files.table); 1373 ttrace->files.max = -1; 1374} 1375 1376static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd) 1377{ 1378 if (fd < 0) 1379 return NULL; 1380 1381 if (fd > ttrace->files.max) { 1382 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file)); 1383 1384 if (nfiles == NULL) 1385 return NULL; 1386 1387 if (ttrace->files.max != -1) { 1388 memset(nfiles + ttrace->files.max + 1, 0, 1389 (fd - ttrace->files.max) * sizeof(struct file)); 1390 } else { 1391 memset(nfiles, 0, (fd + 1) * sizeof(struct file)); 1392 } 1393 1394 ttrace->files.table = nfiles; 1395 ttrace->files.max = fd; 1396 } 1397 1398 return ttrace->files.table + fd; 1399} 1400 1401struct file *thread__files_entry(struct thread *thread, int fd) 1402{ 1403 return thread_trace__files_entry(thread__priv(thread), fd); 1404} 1405 1406static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname) 1407{ 1408 struct thread_trace *ttrace = thread__priv(thread); 1409 struct file *file = thread_trace__files_entry(ttrace, fd); 1410 1411 if (file != NULL) { 1412 struct stat st; 1413 if (stat(pathname, &st) == 0) 1414 file->dev_maj = major(st.st_rdev); 1415 file->pathname = strdup(pathname); 1416 if (file->pathname) 1417 return 0; 1418 } 1419 1420 return -1; 1421} 1422 1423static int thread__read_fd_path(struct thread *thread, int fd) 1424{ 1425 char linkname[PATH_MAX], pathname[PATH_MAX]; 1426 struct stat st; 1427 int ret; 1428 1429 if (thread__pid(thread) == thread__tid(thread)) { 1430 scnprintf(linkname, sizeof(linkname), 1431 "/proc/%d/fd/%d", thread__pid(thread), fd); 1432 } else { 1433 scnprintf(linkname, sizeof(linkname), 1434 "/proc/%d/task/%d/fd/%d", 1435 thread__pid(thread), thread__tid(thread), fd); 1436 } 1437 1438 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname)) 1439 return -1; 1440 1441 ret = readlink(linkname, pathname, sizeof(pathname)); 1442 1443 if (ret < 0 || ret > st.st_size) 1444 return -1; 1445 1446 pathname[ret] = '\0'; 1447 return trace__set_fd_pathname(thread, fd, pathname); 1448} 1449 1450static const char *thread__fd_path(struct thread *thread, int fd, 1451 struct trace *trace) 1452{ 1453 struct thread_trace *ttrace = thread__priv(thread); 1454 1455 if (ttrace == NULL || trace->fd_path_disabled) 1456 return NULL; 1457 1458 if (fd < 0) 1459 return NULL; 1460 1461 if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) { 1462 if (!trace->live) 1463 return NULL; 1464 ++trace->stats.proc_getname; 1465 if (thread__read_fd_path(thread, fd)) 1466 return NULL; 1467 } 1468 1469 return ttrace->files.table[fd].pathname; 1470} 1471 1472size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg) 1473{ 1474 int fd = arg->val; 1475 size_t printed = scnprintf(bf, size, "%d", fd); 1476 const char *path = thread__fd_path(arg->thread, fd, arg->trace); 1477 1478 if (path) 1479 printed += scnprintf(bf + printed, size - printed, "<%s>", path); 1480 1481 return printed; 1482} 1483 1484size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size) 1485{ 1486 size_t printed = scnprintf(bf, size, "%d", fd); 1487 struct thread *thread = machine__find_thread(trace->host, pid, pid); 1488 1489 if (thread) { 1490 const char *path = thread__fd_path(thread, fd, trace); 1491 1492 if (path) 1493 printed += scnprintf(bf + printed, size - printed, "<%s>", path); 1494 1495 thread__put(thread); 1496 } 1497 1498 return printed; 1499} 1500 1501static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size, 1502 struct syscall_arg *arg) 1503{ 1504 int fd = arg->val; 1505 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg); 1506 struct thread_trace *ttrace = thread__priv(arg->thread); 1507 1508 if (ttrace && fd >= 0 && fd <= ttrace->files.max) 1509 zfree(&ttrace->files.table[fd].pathname); 1510 1511 return printed; 1512} 1513 1514static void thread__set_filename_pos(struct thread *thread, const char *bf, 1515 unsigned long ptr) 1516{ 1517 struct thread_trace *ttrace = thread__priv(thread); 1518 1519 ttrace->filename.ptr = ptr; 1520 ttrace->filename.entry_str_pos = bf - ttrace->entry_str; 1521} 1522 1523static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size) 1524{ 1525 struct augmented_arg *augmented_arg = arg->augmented.args; 1526 size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value); 1527 /* 1528 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls 1529 * we would have two strings, each prefixed by its size. 1530 */ 1531 int consumed = sizeof(*augmented_arg) + augmented_arg->size; 1532 1533 arg->augmented.args = ((void *)arg->augmented.args) + consumed; 1534 arg->augmented.size -= consumed; 1535 1536 return printed; 1537} 1538 1539static size_t syscall_arg__scnprintf_filename(char *bf, size_t size, 1540 struct syscall_arg *arg) 1541{ 1542 unsigned long ptr = arg->val; 1543 1544 if (arg->augmented.args) 1545 return syscall_arg__scnprintf_augmented_string(arg, bf, size); 1546 1547 if (!arg->trace->vfs_getname) 1548 return scnprintf(bf, size, "%#x", ptr); 1549 1550 thread__set_filename_pos(arg->thread, bf, ptr); 1551 return 0; 1552} 1553 1554static bool trace__filter_duration(struct trace *trace, double t) 1555{ 1556 return t < (trace->duration_filter * NSEC_PER_MSEC); 1557} 1558 1559static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp) 1560{ 1561 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC; 1562 1563 return fprintf(fp, "%10.3f ", ts); 1564} 1565 1566/* 1567 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are 1568 * using ttrace->entry_time for a thread that receives a sys_exit without 1569 * first having received a sys_enter ("poll" issued before tracing session 1570 * starts, lost sys_enter exit due to ring buffer overflow). 1571 */ 1572static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp) 1573{ 1574 if (tstamp > 0) 1575 return __trace__fprintf_tstamp(trace, tstamp, fp); 1576 1577 return fprintf(fp, " ? "); 1578} 1579 1580static pid_t workload_pid = -1; 1581static volatile sig_atomic_t done = false; 1582static volatile sig_atomic_t interrupted = false; 1583 1584static void sighandler_interrupt(int sig __maybe_unused) 1585{ 1586 done = interrupted = true; 1587} 1588 1589static void sighandler_chld(int sig __maybe_unused, siginfo_t *info, 1590 void *context __maybe_unused) 1591{ 1592 if (info->si_pid == workload_pid) 1593 done = true; 1594} 1595 1596static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp) 1597{ 1598 size_t printed = 0; 1599 1600 if (trace->multiple_threads) { 1601 if (trace->show_comm) 1602 printed += fprintf(fp, "%.14s/", thread__comm_str(thread)); 1603 printed += fprintf(fp, "%d ", thread__tid(thread)); 1604 } 1605 1606 return printed; 1607} 1608 1609static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread, 1610 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp) 1611{ 1612 size_t printed = 0; 1613 1614 if (trace->show_tstamp) 1615 printed = trace__fprintf_tstamp(trace, tstamp, fp); 1616 if (trace->show_duration) 1617 printed += fprintf_duration(duration, duration_calculated, fp); 1618 return printed + trace__fprintf_comm_tid(trace, thread, fp); 1619} 1620 1621static int trace__process_event(struct trace *trace, struct machine *machine, 1622 union perf_event *event, struct perf_sample *sample) 1623{ 1624 int ret = 0; 1625 1626 switch (event->header.type) { 1627 case PERF_RECORD_LOST: 1628 color_fprintf(trace->output, PERF_COLOR_RED, 1629 "LOST %" PRIu64 " events!\n", event->lost.lost); 1630 ret = machine__process_lost_event(machine, event, sample); 1631 break; 1632 default: 1633 ret = machine__process_event(machine, event, sample); 1634 break; 1635 } 1636 1637 return ret; 1638} 1639 1640static int trace__tool_process(struct perf_tool *tool, 1641 union perf_event *event, 1642 struct perf_sample *sample, 1643 struct machine *machine) 1644{ 1645 struct trace *trace = container_of(tool, struct trace, tool); 1646 return trace__process_event(trace, machine, event, sample); 1647} 1648 1649static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp) 1650{ 1651 struct machine *machine = vmachine; 1652 1653 if (machine->kptr_restrict_warned) 1654 return NULL; 1655 1656 if (symbol_conf.kptr_restrict) { 1657 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n" 1658 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n" 1659 "Kernel samples will not be resolved.\n"); 1660 machine->kptr_restrict_warned = true; 1661 return NULL; 1662 } 1663 1664 return machine__resolve_kernel_addr(vmachine, addrp, modp); 1665} 1666 1667static int trace__symbols_init(struct trace *trace, struct evlist *evlist) 1668{ 1669 int err = symbol__init(NULL); 1670 1671 if (err) 1672 return err; 1673 1674 trace->host = machine__new_host(); 1675 if (trace->host == NULL) 1676 return -ENOMEM; 1677 1678 thread__set_priv_destructor(thread_trace__delete); 1679 1680 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr); 1681 if (err < 0) 1682 goto out; 1683 1684 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target, 1685 evlist->core.threads, trace__tool_process, 1686 true, false, 1); 1687out: 1688 if (err) 1689 symbol__exit(); 1690 1691 return err; 1692} 1693 1694static void trace__symbols__exit(struct trace *trace) 1695{ 1696 machine__exit(trace->host); 1697 trace->host = NULL; 1698 1699 symbol__exit(); 1700} 1701 1702static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args) 1703{ 1704 int idx; 1705 1706 if (nr_args == RAW_SYSCALL_ARGS_NUM && sc->fmt && sc->fmt->nr_args != 0) 1707 nr_args = sc->fmt->nr_args; 1708 1709 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt)); 1710 if (sc->arg_fmt == NULL) 1711 return -1; 1712 1713 for (idx = 0; idx < nr_args; ++idx) { 1714 if (sc->fmt) 1715 sc->arg_fmt[idx] = sc->fmt->arg[idx]; 1716 } 1717 1718 sc->nr_args = nr_args; 1719 return 0; 1720} 1721 1722static const struct syscall_arg_fmt syscall_arg_fmts__by_name[] = { 1723 { .name = "msr", .scnprintf = SCA_X86_MSR, .strtoul = STUL_X86_MSR, }, 1724 { .name = "vector", .scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, }, 1725}; 1726 1727static int syscall_arg_fmt__cmp(const void *name, const void *fmtp) 1728{ 1729 const struct syscall_arg_fmt *fmt = fmtp; 1730 return strcmp(name, fmt->name); 1731} 1732 1733static const struct syscall_arg_fmt * 1734__syscall_arg_fmt__find_by_name(const struct syscall_arg_fmt *fmts, const int nmemb, 1735 const char *name) 1736{ 1737 return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp); 1738} 1739 1740static const struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name) 1741{ 1742 const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name); 1743 return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name); 1744} 1745 1746static struct tep_format_field * 1747syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field) 1748{ 1749 struct tep_format_field *last_field = NULL; 1750 int len; 1751 1752 for (; field; field = field->next, ++arg) { 1753 last_field = field; 1754 1755 if (arg->scnprintf) 1756 continue; 1757 1758 len = strlen(field->name); 1759 1760 if (strcmp(field->type, "const char *") == 0 && 1761 ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) || 1762 strstr(field->name, "path") != NULL)) 1763 arg->scnprintf = SCA_FILENAME; 1764 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr")) 1765 arg->scnprintf = SCA_PTR; 1766 else if (strcmp(field->type, "pid_t") == 0) 1767 arg->scnprintf = SCA_PID; 1768 else if (strcmp(field->type, "umode_t") == 0) 1769 arg->scnprintf = SCA_MODE_T; 1770 else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) { 1771 arg->scnprintf = SCA_CHAR_ARRAY; 1772 arg->nr_entries = field->arraylen; 1773 } else if ((strcmp(field->type, "int") == 0 || 1774 strcmp(field->type, "unsigned int") == 0 || 1775 strcmp(field->type, "long") == 0) && 1776 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) { 1777 /* 1778 * /sys/kernel/tracing/events/syscalls/sys_enter* 1779 * grep -E 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c 1780 * 65 int 1781 * 23 unsigned int 1782 * 7 unsigned long 1783 */ 1784 arg->scnprintf = SCA_FD; 1785 } else { 1786 const struct syscall_arg_fmt *fmt = 1787 syscall_arg_fmt__find_by_name(field->name); 1788 1789 if (fmt) { 1790 arg->scnprintf = fmt->scnprintf; 1791 arg->strtoul = fmt->strtoul; 1792 } 1793 } 1794 } 1795 1796 return last_field; 1797} 1798 1799static int syscall__set_arg_fmts(struct syscall *sc) 1800{ 1801 struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args); 1802 1803 if (last_field) 1804 sc->args_size = last_field->offset + last_field->size; 1805 1806 return 0; 1807} 1808 1809static int trace__read_syscall_info(struct trace *trace, int id) 1810{ 1811 char tp_name[128]; 1812 struct syscall *sc; 1813 const char *name = syscalltbl__name(trace->sctbl, id); 1814 1815#ifdef HAVE_SYSCALL_TABLE_SUPPORT 1816 if (trace->syscalls.table == NULL) { 1817 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc)); 1818 if (trace->syscalls.table == NULL) 1819 return -ENOMEM; 1820 } 1821#else 1822 if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) { 1823 // When using libaudit we don't know beforehand what is the max syscall id 1824 struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc)); 1825 1826 if (table == NULL) 1827 return -ENOMEM; 1828 1829 // Need to memset from offset 0 and +1 members if brand new 1830 if (trace->syscalls.table == NULL) 1831 memset(table, 0, (id + 1) * sizeof(*sc)); 1832 else 1833 memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc)); 1834 1835 trace->syscalls.table = table; 1836 trace->sctbl->syscalls.max_id = id; 1837 } 1838#endif 1839 sc = trace->syscalls.table + id; 1840 if (sc->nonexistent) 1841 return -EEXIST; 1842 1843 if (name == NULL) { 1844 sc->nonexistent = true; 1845 return -EEXIST; 1846 } 1847 1848 sc->name = name; 1849 sc->fmt = syscall_fmt__find(sc->name); 1850 1851 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name); 1852 sc->tp_format = trace_event__tp_format("syscalls", tp_name); 1853 1854 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) { 1855 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias); 1856 sc->tp_format = trace_event__tp_format("syscalls", tp_name); 1857 } 1858 1859 /* 1860 * Fails to read trace point format via sysfs node, so the trace point 1861 * doesn't exist. Set the 'nonexistent' flag as true. 1862 */ 1863 if (IS_ERR(sc->tp_format)) { 1864 sc->nonexistent = true; 1865 return PTR_ERR(sc->tp_format); 1866 } 1867 1868 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ? 1869 RAW_SYSCALL_ARGS_NUM : sc->tp_format->format.nr_fields)) 1870 return -ENOMEM; 1871 1872 sc->args = sc->tp_format->format.fields; 1873 /* 1874 * We need to check and discard the first variable '__syscall_nr' 1875 * or 'nr' that mean the syscall number. It is needless here. 1876 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels. 1877 */ 1878 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) { 1879 sc->args = sc->args->next; 1880 --sc->nr_args; 1881 } 1882 1883 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit"); 1884 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat"); 1885 1886 return syscall__set_arg_fmts(sc); 1887} 1888 1889static int evsel__init_tp_arg_scnprintf(struct evsel *evsel) 1890{ 1891 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel); 1892 1893 if (fmt != NULL) { 1894 syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields); 1895 return 0; 1896 } 1897 1898 return -ENOMEM; 1899} 1900 1901static int intcmp(const void *a, const void *b) 1902{ 1903 const int *one = a, *another = b; 1904 1905 return *one - *another; 1906} 1907 1908static int trace__validate_ev_qualifier(struct trace *trace) 1909{ 1910 int err = 0; 1911 bool printed_invalid_prefix = false; 1912 struct str_node *pos; 1913 size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier); 1914 1915 trace->ev_qualifier_ids.entries = malloc(nr_allocated * 1916 sizeof(trace->ev_qualifier_ids.entries[0])); 1917 1918 if (trace->ev_qualifier_ids.entries == NULL) { 1919 fputs("Error:\tNot enough memory for allocating events qualifier ids\n", 1920 trace->output); 1921 err = -EINVAL; 1922 goto out; 1923 } 1924 1925 strlist__for_each_entry(pos, trace->ev_qualifier) { 1926 const char *sc = pos->s; 1927 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1; 1928 1929 if (id < 0) { 1930 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next); 1931 if (id >= 0) 1932 goto matches; 1933 1934 if (!printed_invalid_prefix) { 1935 pr_debug("Skipping unknown syscalls: "); 1936 printed_invalid_prefix = true; 1937 } else { 1938 pr_debug(", "); 1939 } 1940 1941 pr_debug("%s", sc); 1942 continue; 1943 } 1944matches: 1945 trace->ev_qualifier_ids.entries[nr_used++] = id; 1946 if (match_next == -1) 1947 continue; 1948 1949 while (1) { 1950 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next); 1951 if (id < 0) 1952 break; 1953 if (nr_allocated == nr_used) { 1954 void *entries; 1955 1956 nr_allocated += 8; 1957 entries = realloc(trace->ev_qualifier_ids.entries, 1958 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0])); 1959 if (entries == NULL) { 1960 err = -ENOMEM; 1961 fputs("\nError:\t Not enough memory for parsing\n", trace->output); 1962 goto out_free; 1963 } 1964 trace->ev_qualifier_ids.entries = entries; 1965 } 1966 trace->ev_qualifier_ids.entries[nr_used++] = id; 1967 } 1968 } 1969 1970 trace->ev_qualifier_ids.nr = nr_used; 1971 qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp); 1972out: 1973 if (printed_invalid_prefix) 1974 pr_debug("\n"); 1975 return err; 1976out_free: 1977 zfree(&trace->ev_qualifier_ids.entries); 1978 trace->ev_qualifier_ids.nr = 0; 1979 goto out; 1980} 1981 1982static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id) 1983{ 1984 bool in_ev_qualifier; 1985 1986 if (trace->ev_qualifier_ids.nr == 0) 1987 return true; 1988 1989 in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries, 1990 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL; 1991 1992 if (in_ev_qualifier) 1993 return !trace->not_ev_qualifier; 1994 1995 return trace->not_ev_qualifier; 1996} 1997 1998/* 1999 * args is to be interpreted as a series of longs but we need to handle 2000 * 8-byte unaligned accesses. args points to raw_data within the event 2001 * and raw_data is guaranteed to be 8-byte unaligned because it is 2002 * preceded by raw_size which is a u32. So we need to copy args to a temp 2003 * variable to read it. Most notably this avoids extended load instructions 2004 * on unaligned addresses 2005 */ 2006unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx) 2007{ 2008 unsigned long val; 2009 unsigned char *p = arg->args + sizeof(unsigned long) * idx; 2010 2011 memcpy(&val, p, sizeof(val)); 2012 return val; 2013} 2014 2015static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size, 2016 struct syscall_arg *arg) 2017{ 2018 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name) 2019 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name); 2020 2021 return scnprintf(bf, size, "arg%d: ", arg->idx); 2022} 2023 2024/* 2025 * Check if the value is in fact zero, i.e. mask whatever needs masking, such 2026 * as mount 'flags' argument that needs ignoring some magic flag, see comment 2027 * in tools/perf/trace/beauty/mount_flags.c 2028 */ 2029static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val) 2030{ 2031 if (fmt && fmt->mask_val) 2032 return fmt->mask_val(arg, val); 2033 2034 return val; 2035} 2036 2037static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size, 2038 struct syscall_arg *arg, unsigned long val) 2039{ 2040 if (fmt && fmt->scnprintf) { 2041 arg->val = val; 2042 if (fmt->parm) 2043 arg->parm = fmt->parm; 2044 return fmt->scnprintf(bf, size, arg); 2045 } 2046 return scnprintf(bf, size, "%ld", val); 2047} 2048 2049static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size, 2050 unsigned char *args, void *augmented_args, int augmented_args_size, 2051 struct trace *trace, struct thread *thread) 2052{ 2053 size_t printed = 0; 2054 unsigned long val; 2055 u8 bit = 1; 2056 struct syscall_arg arg = { 2057 .args = args, 2058 .augmented = { 2059 .size = augmented_args_size, 2060 .args = augmented_args, 2061 }, 2062 .idx = 0, 2063 .mask = 0, 2064 .trace = trace, 2065 .thread = thread, 2066 .show_string_prefix = trace->show_string_prefix, 2067 }; 2068 struct thread_trace *ttrace = thread__priv(thread); 2069 2070 /* 2071 * Things like fcntl will set this in its 'cmd' formatter to pick the 2072 * right formatter for the return value (an fd? file flags?), which is 2073 * not needed for syscalls that always return a given type, say an fd. 2074 */ 2075 ttrace->ret_scnprintf = NULL; 2076 2077 if (sc->args != NULL) { 2078 struct tep_format_field *field; 2079 2080 for (field = sc->args; field; 2081 field = field->next, ++arg.idx, bit <<= 1) { 2082 if (arg.mask & bit) 2083 continue; 2084 2085 arg.fmt = &sc->arg_fmt[arg.idx]; 2086 val = syscall_arg__val(&arg, arg.idx); 2087 /* 2088 * Some syscall args need some mask, most don't and 2089 * return val untouched. 2090 */ 2091 val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val); 2092 2093 /* 2094 * Suppress this argument if its value is zero and 2095 * and we don't have a string associated in an 2096 * strarray for it. 2097 */ 2098 if (val == 0 && 2099 !trace->show_zeros && 2100 !(sc->arg_fmt && 2101 (sc->arg_fmt[arg.idx].show_zero || 2102 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY || 2103 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) && 2104 sc->arg_fmt[arg.idx].parm)) 2105 continue; 2106 2107 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : ""); 2108 2109 if (trace->show_arg_names) 2110 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name); 2111 2112 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], 2113 bf + printed, size - printed, &arg, val); 2114 } 2115 } else if (IS_ERR(sc->tp_format)) { 2116 /* 2117 * If we managed to read the tracepoint /format file, then we 2118 * may end up not having any args, like with gettid(), so only 2119 * print the raw args when we didn't manage to read it. 2120 */ 2121 while (arg.idx < sc->nr_args) { 2122 if (arg.mask & bit) 2123 goto next_arg; 2124 val = syscall_arg__val(&arg, arg.idx); 2125 if (printed) 2126 printed += scnprintf(bf + printed, size - printed, ", "); 2127 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg); 2128 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val); 2129next_arg: 2130 ++arg.idx; 2131 bit <<= 1; 2132 } 2133 } 2134 2135 return printed; 2136} 2137 2138typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel, 2139 union perf_event *event, 2140 struct perf_sample *sample); 2141 2142static struct syscall *trace__syscall_info(struct trace *trace, 2143 struct evsel *evsel, int id) 2144{ 2145 int err = 0; 2146 2147 if (id < 0) { 2148 2149 /* 2150 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried 2151 * before that, leaving at a higher verbosity level till that is 2152 * explained. Reproduced with plain ftrace with: 2153 * 2154 * echo 1 > /t/events/raw_syscalls/sys_exit/enable 2155 * grep "NR -1 " /t/trace_pipe 2156 * 2157 * After generating some load on the machine. 2158 */ 2159 if (verbose > 1) { 2160 static u64 n; 2161 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n", 2162 id, evsel__name(evsel), ++n); 2163 } 2164 return NULL; 2165 } 2166 2167 err = -EINVAL; 2168 2169#ifdef HAVE_SYSCALL_TABLE_SUPPORT 2170 if (id > trace->sctbl->syscalls.max_id) { 2171#else 2172 if (id >= trace->sctbl->syscalls.max_id) { 2173 /* 2174 * With libaudit we don't know beforehand what is the max_id, 2175 * so we let trace__read_syscall_info() figure that out as we 2176 * go on reading syscalls. 2177 */ 2178 err = trace__read_syscall_info(trace, id); 2179 if (err) 2180#endif 2181 goto out_cant_read; 2182 } 2183 2184 if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) && 2185 (err = trace__read_syscall_info(trace, id)) != 0) 2186 goto out_cant_read; 2187 2188 if (trace->syscalls.table && trace->syscalls.table[id].nonexistent) 2189 goto out_cant_read; 2190 2191 return &trace->syscalls.table[id]; 2192 2193out_cant_read: 2194 if (verbose > 0) { 2195 char sbuf[STRERR_BUFSIZE]; 2196 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf))); 2197 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL) 2198 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name); 2199 fputs(" information\n", trace->output); 2200 } 2201 return NULL; 2202} 2203 2204struct syscall_stats { 2205 struct stats stats; 2206 u64 nr_failures; 2207 int max_errno; 2208 u32 *errnos; 2209}; 2210 2211static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace, 2212 int id, struct perf_sample *sample, long err, bool errno_summary) 2213{ 2214 struct int_node *inode; 2215 struct syscall_stats *stats; 2216 u64 duration = 0; 2217 2218 inode = intlist__findnew(ttrace->syscall_stats, id); 2219 if (inode == NULL) 2220 return; 2221 2222 stats = inode->priv; 2223 if (stats == NULL) { 2224 stats = zalloc(sizeof(*stats)); 2225 if (stats == NULL) 2226 return; 2227 2228 init_stats(&stats->stats); 2229 inode->priv = stats; 2230 } 2231 2232 if (ttrace->entry_time && sample->time > ttrace->entry_time) 2233 duration = sample->time - ttrace->entry_time; 2234 2235 update_stats(&stats->stats, duration); 2236 2237 if (err < 0) { 2238 ++stats->nr_failures; 2239 2240 if (!errno_summary) 2241 return; 2242 2243 err = -err; 2244 if (err > stats->max_errno) { 2245 u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32)); 2246 2247 if (new_errnos) { 2248 memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32)); 2249 } else { 2250 pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n", 2251 thread__comm_str(thread), thread__pid(thread), 2252 thread__tid(thread)); 2253 return; 2254 } 2255 2256 stats->errnos = new_errnos; 2257 stats->max_errno = err; 2258 } 2259 2260 ++stats->errnos[err - 1]; 2261 } 2262} 2263 2264static int trace__printf_interrupted_entry(struct trace *trace) 2265{ 2266 struct thread_trace *ttrace; 2267 size_t printed; 2268 int len; 2269 2270 if (trace->failure_only || trace->current == NULL) 2271 return 0; 2272 2273 ttrace = thread__priv(trace->current); 2274 2275 if (!ttrace->entry_pending) 2276 return 0; 2277 2278 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output); 2279 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str); 2280 2281 if (len < trace->args_alignment - 4) 2282 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " "); 2283 2284 printed += fprintf(trace->output, " ...\n"); 2285 2286 ttrace->entry_pending = false; 2287 ++trace->nr_events_printed; 2288 2289 return printed; 2290} 2291 2292static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel, 2293 struct perf_sample *sample, struct thread *thread) 2294{ 2295 int printed = 0; 2296 2297 if (trace->print_sample) { 2298 double ts = (double)sample->time / NSEC_PER_MSEC; 2299 2300 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n", 2301 evsel__name(evsel), ts, 2302 thread__comm_str(thread), 2303 sample->pid, sample->tid, sample->cpu); 2304 } 2305 2306 return printed; 2307} 2308 2309static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size) 2310{ 2311 void *augmented_args = NULL; 2312 /* 2313 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter 2314 * and there we get all 6 syscall args plus the tracepoint common fields 2315 * that gets calculated at the start and the syscall_nr (another long). 2316 * So we check if that is the case and if so don't look after the 2317 * sc->args_size but always after the full raw_syscalls:sys_enter payload, 2318 * which is fixed. 2319 * 2320 * We'll revisit this later to pass s->args_size to the BPF augmenter 2321 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it 2322 * copies only what we need for each syscall, like what happens when we 2323 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace 2324 * traffic to just what is needed for each syscall. 2325 */ 2326 int args_size = raw_augmented_args_size ?: sc->args_size; 2327 2328 *augmented_args_size = sample->raw_size - args_size; 2329 if (*augmented_args_size > 0) 2330 augmented_args = sample->raw_data + args_size; 2331 2332 return augmented_args; 2333} 2334 2335static void syscall__exit(struct syscall *sc) 2336{ 2337 if (!sc) 2338 return; 2339 2340 zfree(&sc->arg_fmt); 2341} 2342 2343static int trace__sys_enter(struct trace *trace, struct evsel *evsel, 2344 union perf_event *event __maybe_unused, 2345 struct perf_sample *sample) 2346{ 2347 char *msg; 2348 void *args; 2349 int printed = 0; 2350 struct thread *thread; 2351 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1; 2352 int augmented_args_size = 0; 2353 void *augmented_args = NULL; 2354 struct syscall *sc = trace__syscall_info(trace, evsel, id); 2355 struct thread_trace *ttrace; 2356 2357 if (sc == NULL) 2358 return -1; 2359 2360 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid); 2361 ttrace = thread__trace(thread, trace->output); 2362 if (ttrace == NULL) 2363 goto out_put; 2364 2365 trace__fprintf_sample(trace, evsel, sample, thread); 2366 2367 args = perf_evsel__sc_tp_ptr(evsel, args, sample); 2368 2369 if (ttrace->entry_str == NULL) { 2370 ttrace->entry_str = malloc(trace__entry_str_size); 2371 if (!ttrace->entry_str) 2372 goto out_put; 2373 } 2374 2375 if (!(trace->duration_filter || trace->summary_only || trace->min_stack)) 2376 trace__printf_interrupted_entry(trace); 2377 /* 2378 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible 2379 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments 2380 * this breaks syscall__augmented_args() check for augmented args, as we calculate 2381 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file, 2382 * so when handling, say the openat syscall, we end up getting 6 args for the 2383 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly 2384 * thinking that the extra 2 u64 args are the augmented filename, so just check 2385 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one. 2386 */ 2387 if (evsel != trace->syscalls.events.sys_enter) 2388 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size); 2389 ttrace->entry_time = sample->time; 2390 msg = ttrace->entry_str; 2391 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name); 2392 2393 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed, 2394 args, augmented_args, augmented_args_size, trace, thread); 2395 2396 if (sc->is_exit) { 2397 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) { 2398 int alignment = 0; 2399 2400 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output); 2401 printed = fprintf(trace->output, "%s)", ttrace->entry_str); 2402 if (trace->args_alignment > printed) 2403 alignment = trace->args_alignment - printed; 2404 fprintf(trace->output, "%*s= ?\n", alignment, " "); 2405 } 2406 } else { 2407 ttrace->entry_pending = true; 2408 /* See trace__vfs_getname & trace__sys_exit */ 2409 ttrace->filename.pending_open = false; 2410 } 2411 2412 if (trace->current != thread) { 2413 thread__put(trace->current); 2414 trace->current = thread__get(thread); 2415 } 2416 err = 0; 2417out_put: 2418 thread__put(thread); 2419 return err; 2420} 2421 2422static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel, 2423 struct perf_sample *sample) 2424{ 2425 struct thread_trace *ttrace; 2426 struct thread *thread; 2427 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1; 2428 struct syscall *sc = trace__syscall_info(trace, evsel, id); 2429 char msg[1024]; 2430 void *args, *augmented_args = NULL; 2431 int augmented_args_size; 2432 2433 if (sc == NULL) 2434 return -1; 2435 2436 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid); 2437 ttrace = thread__trace(thread, trace->output); 2438 /* 2439 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args() 2440 * and the rest of the beautifiers accessing it via struct syscall_arg touches it. 2441 */ 2442 if (ttrace == NULL) 2443 goto out_put; 2444 2445 args = perf_evsel__sc_tp_ptr(evsel, args, sample); 2446 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size); 2447 syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread); 2448 fprintf(trace->output, "%s", msg); 2449 err = 0; 2450out_put: 2451 thread__put(thread); 2452 return err; 2453} 2454 2455static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel, 2456 struct perf_sample *sample, 2457 struct callchain_cursor *cursor) 2458{ 2459 struct addr_location al; 2460 int max_stack = evsel->core.attr.sample_max_stack ? 2461 evsel->core.attr.sample_max_stack : 2462 trace->max_stack; 2463 int err = -1; 2464 2465 addr_location__init(&al); 2466 if (machine__resolve(trace->host, &al, sample) < 0) 2467 goto out; 2468 2469 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack); 2470out: 2471 addr_location__exit(&al); 2472 return err; 2473} 2474 2475static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample) 2476{ 2477 /* TODO: user-configurable print_opts */ 2478 const unsigned int print_opts = EVSEL__PRINT_SYM | 2479 EVSEL__PRINT_DSO | 2480 EVSEL__PRINT_UNKNOWN_AS_ADDR; 2481 2482 return sample__fprintf_callchain(sample, 38, print_opts, get_tls_callchain_cursor(), symbol_conf.bt_stop_list, trace->output); 2483} 2484 2485static const char *errno_to_name(struct evsel *evsel, int err) 2486{ 2487 struct perf_env *env = evsel__env(evsel); 2488 2489 return perf_env__arch_strerrno(env, err); 2490} 2491 2492static int trace__sys_exit(struct trace *trace, struct evsel *evsel, 2493 union perf_event *event __maybe_unused, 2494 struct perf_sample *sample) 2495{ 2496 long ret; 2497 u64 duration = 0; 2498 bool duration_calculated = false; 2499 struct thread *thread; 2500 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0; 2501 int alignment = trace->args_alignment; 2502 struct syscall *sc = trace__syscall_info(trace, evsel, id); 2503 struct thread_trace *ttrace; 2504 2505 if (sc == NULL) 2506 return -1; 2507 2508 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid); 2509 ttrace = thread__trace(thread, trace->output); 2510 if (ttrace == NULL) 2511 goto out_put; 2512 2513 trace__fprintf_sample(trace, evsel, sample, thread); 2514 2515 ret = perf_evsel__sc_tp_uint(evsel, ret, sample); 2516 2517 if (trace->summary) 2518 thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary); 2519 2520 if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) { 2521 trace__set_fd_pathname(thread, ret, ttrace->filename.name); 2522 ttrace->filename.pending_open = false; 2523 ++trace->stats.vfs_getname; 2524 } 2525 2526 if (ttrace->entry_time) { 2527 duration = sample->time - ttrace->entry_time; 2528 if (trace__filter_duration(trace, duration)) 2529 goto out; 2530 duration_calculated = true; 2531 } else if (trace->duration_filter) 2532 goto out; 2533 2534 if (sample->callchain) { 2535 struct callchain_cursor *cursor = get_tls_callchain_cursor(); 2536 2537 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor); 2538 if (callchain_ret == 0) { 2539 if (cursor->nr < trace->min_stack) 2540 goto out; 2541 callchain_ret = 1; 2542 } 2543 } 2544 2545 if (trace->summary_only || (ret >= 0 && trace->failure_only)) 2546 goto out; 2547 2548 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output); 2549 2550 if (ttrace->entry_pending) { 2551 printed = fprintf(trace->output, "%s", ttrace->entry_str); 2552 } else { 2553 printed += fprintf(trace->output, " ... ["); 2554 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued"); 2555 printed += 9; 2556 printed += fprintf(trace->output, "]: %s()", sc->name); 2557 } 2558 2559 printed++; /* the closing ')' */ 2560 2561 if (alignment > printed) 2562 alignment -= printed; 2563 else 2564 alignment = 0; 2565 2566 fprintf(trace->output, ")%*s= ", alignment, " "); 2567 2568 if (sc->fmt == NULL) { 2569 if (ret < 0) 2570 goto errno_print; 2571signed_print: 2572 fprintf(trace->output, "%ld", ret); 2573 } else if (ret < 0) { 2574errno_print: { 2575 char bf[STRERR_BUFSIZE]; 2576 const char *emsg = str_error_r(-ret, bf, sizeof(bf)), 2577 *e = errno_to_name(evsel, -ret); 2578 2579 fprintf(trace->output, "-1 %s (%s)", e, emsg); 2580 } 2581 } else if (ret == 0 && sc->fmt->timeout) 2582 fprintf(trace->output, "0 (Timeout)"); 2583 else if (ttrace->ret_scnprintf) { 2584 char bf[1024]; 2585 struct syscall_arg arg = { 2586 .val = ret, 2587 .thread = thread, 2588 .trace = trace, 2589 }; 2590 ttrace->ret_scnprintf(bf, sizeof(bf), &arg); 2591 ttrace->ret_scnprintf = NULL; 2592 fprintf(trace->output, "%s", bf); 2593 } else if (sc->fmt->hexret) 2594 fprintf(trace->output, "%#lx", ret); 2595 else if (sc->fmt->errpid) { 2596 struct thread *child = machine__find_thread(trace->host, ret, ret); 2597 2598 if (child != NULL) { 2599 fprintf(trace->output, "%ld", ret); 2600 if (thread__comm_set(child)) 2601 fprintf(trace->output, " (%s)", thread__comm_str(child)); 2602 thread__put(child); 2603 } 2604 } else 2605 goto signed_print; 2606 2607 fputc('\n', trace->output); 2608 2609 /* 2610 * We only consider an 'event' for the sake of --max-events a non-filtered 2611 * sys_enter + sys_exit and other tracepoint events. 2612 */ 2613 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX) 2614 interrupted = true; 2615 2616 if (callchain_ret > 0) 2617 trace__fprintf_callchain(trace, sample); 2618 else if (callchain_ret < 0) 2619 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel)); 2620out: 2621 ttrace->entry_pending = false; 2622 err = 0; 2623out_put: 2624 thread__put(thread); 2625 return err; 2626} 2627 2628static int trace__vfs_getname(struct trace *trace, struct evsel *evsel, 2629 union perf_event *event __maybe_unused, 2630 struct perf_sample *sample) 2631{ 2632 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid); 2633 struct thread_trace *ttrace; 2634 size_t filename_len, entry_str_len, to_move; 2635 ssize_t remaining_space; 2636 char *pos; 2637 const char *filename = evsel__rawptr(evsel, sample, "pathname"); 2638 2639 if (!thread) 2640 goto out; 2641 2642 ttrace = thread__priv(thread); 2643 if (!ttrace) 2644 goto out_put; 2645 2646 filename_len = strlen(filename); 2647 if (filename_len == 0) 2648 goto out_put; 2649 2650 if (ttrace->filename.namelen < filename_len) { 2651 char *f = realloc(ttrace->filename.name, filename_len + 1); 2652 2653 if (f == NULL) 2654 goto out_put; 2655 2656 ttrace->filename.namelen = filename_len; 2657 ttrace->filename.name = f; 2658 } 2659 2660 strcpy(ttrace->filename.name, filename); 2661 ttrace->filename.pending_open = true; 2662 2663 if (!ttrace->filename.ptr) 2664 goto out_put; 2665 2666 entry_str_len = strlen(ttrace->entry_str); 2667 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */ 2668 if (remaining_space <= 0) 2669 goto out_put; 2670 2671 if (filename_len > (size_t)remaining_space) { 2672 filename += filename_len - remaining_space; 2673 filename_len = remaining_space; 2674 } 2675 2676 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */ 2677 pos = ttrace->entry_str + ttrace->filename.entry_str_pos; 2678 memmove(pos + filename_len, pos, to_move); 2679 memcpy(pos, filename, filename_len); 2680 2681 ttrace->filename.ptr = 0; 2682 ttrace->filename.entry_str_pos = 0; 2683out_put: 2684 thread__put(thread); 2685out: 2686 return 0; 2687} 2688 2689static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel, 2690 union perf_event *event __maybe_unused, 2691 struct perf_sample *sample) 2692{ 2693 u64 runtime = evsel__intval(evsel, sample, "runtime"); 2694 double runtime_ms = (double)runtime / NSEC_PER_MSEC; 2695 struct thread *thread = machine__findnew_thread(trace->host, 2696 sample->pid, 2697 sample->tid); 2698 struct thread_trace *ttrace = thread__trace(thread, trace->output); 2699 2700 if (ttrace == NULL) 2701 goto out_dump; 2702 2703 ttrace->runtime_ms += runtime_ms; 2704 trace->runtime_ms += runtime_ms; 2705out_put: 2706 thread__put(thread); 2707 return 0; 2708 2709out_dump: 2710 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n", 2711 evsel->name, 2712 evsel__strval(evsel, sample, "comm"), 2713 (pid_t)evsel__intval(evsel, sample, "pid"), 2714 runtime, 2715 evsel__intval(evsel, sample, "vruntime")); 2716 goto out_put; 2717} 2718 2719static int bpf_output__printer(enum binary_printer_ops op, 2720 unsigned int val, void *extra __maybe_unused, FILE *fp) 2721{ 2722 unsigned char ch = (unsigned char)val; 2723 2724 switch (op) { 2725 case BINARY_PRINT_CHAR_DATA: 2726 return fprintf(fp, "%c", isprint(ch) ? ch : '.'); 2727 case BINARY_PRINT_DATA_BEGIN: 2728 case BINARY_PRINT_LINE_BEGIN: 2729 case BINARY_PRINT_ADDR: 2730 case BINARY_PRINT_NUM_DATA: 2731 case BINARY_PRINT_NUM_PAD: 2732 case BINARY_PRINT_SEP: 2733 case BINARY_PRINT_CHAR_PAD: 2734 case BINARY_PRINT_LINE_END: 2735 case BINARY_PRINT_DATA_END: 2736 default: 2737 break; 2738 } 2739 2740 return 0; 2741} 2742 2743static void bpf_output__fprintf(struct trace *trace, 2744 struct perf_sample *sample) 2745{ 2746 binary__fprintf(sample->raw_data, sample->raw_size, 8, 2747 bpf_output__printer, NULL, trace->output); 2748 ++trace->nr_events_printed; 2749} 2750 2751static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample, 2752 struct thread *thread, void *augmented_args, int augmented_args_size) 2753{ 2754 char bf[2048]; 2755 size_t size = sizeof(bf); 2756 struct tep_format_field *field = evsel->tp_format->format.fields; 2757 struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel); 2758 size_t printed = 0; 2759 unsigned long val; 2760 u8 bit = 1; 2761 struct syscall_arg syscall_arg = { 2762 .augmented = { 2763 .size = augmented_args_size, 2764 .args = augmented_args, 2765 }, 2766 .idx = 0, 2767 .mask = 0, 2768 .trace = trace, 2769 .thread = thread, 2770 .show_string_prefix = trace->show_string_prefix, 2771 }; 2772 2773 for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) { 2774 if (syscall_arg.mask & bit) 2775 continue; 2776 2777 syscall_arg.len = 0; 2778 syscall_arg.fmt = arg; 2779 if (field->flags & TEP_FIELD_IS_ARRAY) { 2780 int offset = field->offset; 2781 2782 if (field->flags & TEP_FIELD_IS_DYNAMIC) { 2783 offset = format_field__intval(field, sample, evsel->needs_swap); 2784 syscall_arg.len = offset >> 16; 2785 offset &= 0xffff; 2786 if (tep_field_is_relative(field->flags)) 2787 offset += field->offset + field->size; 2788 } 2789 2790 val = (uintptr_t)(sample->raw_data + offset); 2791 } else 2792 val = format_field__intval(field, sample, evsel->needs_swap); 2793 /* 2794 * Some syscall args need some mask, most don't and 2795 * return val untouched. 2796 */ 2797 val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val); 2798 2799 /* 2800 * Suppress this argument if its value is zero and 2801 * we don't have a string associated in an 2802 * strarray for it. 2803 */ 2804 if (val == 0 && 2805 !trace->show_zeros && 2806 !((arg->show_zero || 2807 arg->scnprintf == SCA_STRARRAY || 2808 arg->scnprintf == SCA_STRARRAYS) && 2809 arg->parm)) 2810 continue; 2811 2812 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : ""); 2813 2814 if (trace->show_arg_names) 2815 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name); 2816 2817 printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val); 2818 } 2819 2820 return printed + fprintf(trace->output, "%s", bf); 2821} 2822 2823static int trace__event_handler(struct trace *trace, struct evsel *evsel, 2824 union perf_event *event __maybe_unused, 2825 struct perf_sample *sample) 2826{ 2827 struct thread *thread; 2828 int callchain_ret = 0; 2829 /* 2830 * Check if we called perf_evsel__disable(evsel) due to, for instance, 2831 * this event's max_events having been hit and this is an entry coming 2832 * from the ring buffer that we should discard, since the max events 2833 * have already been considered/printed. 2834 */ 2835 if (evsel->disabled) 2836 return 0; 2837 2838 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid); 2839 2840 if (sample->callchain) { 2841 struct callchain_cursor *cursor = get_tls_callchain_cursor(); 2842 2843 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor); 2844 if (callchain_ret == 0) { 2845 if (cursor->nr < trace->min_stack) 2846 goto out; 2847 callchain_ret = 1; 2848 } 2849 } 2850 2851 trace__printf_interrupted_entry(trace); 2852 trace__fprintf_tstamp(trace, sample->time, trace->output); 2853 2854 if (trace->trace_syscalls && trace->show_duration) 2855 fprintf(trace->output, "( ): "); 2856 2857 if (thread) 2858 trace__fprintf_comm_tid(trace, thread, trace->output); 2859 2860 if (evsel == trace->syscalls.events.bpf_output) { 2861 int id = perf_evsel__sc_tp_uint(evsel, id, sample); 2862 struct syscall *sc = trace__syscall_info(trace, evsel, id); 2863 2864 if (sc) { 2865 fprintf(trace->output, "%s(", sc->name); 2866 trace__fprintf_sys_enter(trace, evsel, sample); 2867 fputc(')', trace->output); 2868 goto newline; 2869 } 2870 2871 /* 2872 * XXX: Not having the associated syscall info or not finding/adding 2873 * the thread should never happen, but if it does... 2874 * fall thru and print it as a bpf_output event. 2875 */ 2876 } 2877 2878 fprintf(trace->output, "%s(", evsel->name); 2879 2880 if (evsel__is_bpf_output(evsel)) { 2881 bpf_output__fprintf(trace, sample); 2882 } else if (evsel->tp_format) { 2883 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) || 2884 trace__fprintf_sys_enter(trace, evsel, sample)) { 2885 if (trace->libtraceevent_print) { 2886 event_format__fprintf(evsel->tp_format, sample->cpu, 2887 sample->raw_data, sample->raw_size, 2888 trace->output); 2889 } else { 2890 trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0); 2891 } 2892 } 2893 } 2894 2895newline: 2896 fprintf(trace->output, ")\n"); 2897 2898 if (callchain_ret > 0) 2899 trace__fprintf_callchain(trace, sample); 2900 else if (callchain_ret < 0) 2901 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel)); 2902 2903 ++trace->nr_events_printed; 2904 2905 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) { 2906 evsel__disable(evsel); 2907 evsel__close(evsel); 2908 } 2909out: 2910 thread__put(thread); 2911 return 0; 2912} 2913 2914static void print_location(FILE *f, struct perf_sample *sample, 2915 struct addr_location *al, 2916 bool print_dso, bool print_sym) 2917{ 2918 2919 if ((verbose > 0 || print_dso) && al->map) 2920 fprintf(f, "%s@", dso__long_name(map__dso(al->map))); 2921 2922 if ((verbose > 0 || print_sym) && al->sym) 2923 fprintf(f, "%s+0x%" PRIx64, al->sym->name, 2924 al->addr - al->sym->start); 2925 else if (al->map) 2926 fprintf(f, "0x%" PRIx64, al->addr); 2927 else 2928 fprintf(f, "0x%" PRIx64, sample->addr); 2929} 2930 2931static int trace__pgfault(struct trace *trace, 2932 struct evsel *evsel, 2933 union perf_event *event __maybe_unused, 2934 struct perf_sample *sample) 2935{ 2936 struct thread *thread; 2937 struct addr_location al; 2938 char map_type = 'd'; 2939 struct thread_trace *ttrace; 2940 int err = -1; 2941 int callchain_ret = 0; 2942 2943 addr_location__init(&al); 2944 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid); 2945 2946 if (sample->callchain) { 2947 struct callchain_cursor *cursor = get_tls_callchain_cursor(); 2948 2949 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor); 2950 if (callchain_ret == 0) { 2951 if (cursor->nr < trace->min_stack) 2952 goto out_put; 2953 callchain_ret = 1; 2954 } 2955 } 2956 2957 ttrace = thread__trace(thread, trace->output); 2958 if (ttrace == NULL) 2959 goto out_put; 2960 2961 if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ) 2962 ttrace->pfmaj++; 2963 else 2964 ttrace->pfmin++; 2965 2966 if (trace->summary_only) 2967 goto out; 2968 2969 thread__find_symbol(thread, sample->cpumode, sample->ip, &al); 2970 2971 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output); 2972 2973 fprintf(trace->output, "%sfault [", 2974 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ? 2975 "maj" : "min"); 2976 2977 print_location(trace->output, sample, &al, false, true); 2978 2979 fprintf(trace->output, "] => "); 2980 2981 thread__find_symbol(thread, sample->cpumode, sample->addr, &al); 2982 2983 if (!al.map) { 2984 thread__find_symbol(thread, sample->cpumode, sample->addr, &al); 2985 2986 if (al.map) 2987 map_type = 'x'; 2988 else 2989 map_type = '?'; 2990 } 2991 2992 print_location(trace->output, sample, &al, true, false); 2993 2994 fprintf(trace->output, " (%c%c)\n", map_type, al.level); 2995 2996 if (callchain_ret > 0) 2997 trace__fprintf_callchain(trace, sample); 2998 else if (callchain_ret < 0) 2999 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel)); 3000 3001 ++trace->nr_events_printed; 3002out: 3003 err = 0; 3004out_put: 3005 thread__put(thread); 3006 addr_location__exit(&al); 3007 return err; 3008} 3009 3010static void trace__set_base_time(struct trace *trace, 3011 struct evsel *evsel, 3012 struct perf_sample *sample) 3013{ 3014 /* 3015 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust 3016 * and don't use sample->time unconditionally, we may end up having 3017 * some other event in the future without PERF_SAMPLE_TIME for good 3018 * reason, i.e. we may not be interested in its timestamps, just in 3019 * it taking place, picking some piece of information when it 3020 * appears in our event stream (vfs_getname comes to mind). 3021 */ 3022 if (trace->base_time == 0 && !trace->full_time && 3023 (evsel->core.attr.sample_type & PERF_SAMPLE_TIME)) 3024 trace->base_time = sample->time; 3025} 3026 3027static int trace__process_sample(struct perf_tool *tool, 3028 union perf_event *event, 3029 struct perf_sample *sample, 3030 struct evsel *evsel, 3031 struct machine *machine __maybe_unused) 3032{ 3033 struct trace *trace = container_of(tool, struct trace, tool); 3034 struct thread *thread; 3035 int err = 0; 3036 3037 tracepoint_handler handler = evsel->handler; 3038 3039 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid); 3040 if (thread && thread__is_filtered(thread)) 3041 goto out; 3042 3043 trace__set_base_time(trace, evsel, sample); 3044 3045 if (handler) { 3046 ++trace->nr_events; 3047 handler(trace, evsel, event, sample); 3048 } 3049out: 3050 thread__put(thread); 3051 return err; 3052} 3053 3054static int trace__record(struct trace *trace, int argc, const char **argv) 3055{ 3056 unsigned int rec_argc, i, j; 3057 const char **rec_argv; 3058 const char * const record_args[] = { 3059 "record", 3060 "-R", 3061 "-m", "1024", 3062 "-c", "1", 3063 }; 3064 pid_t pid = getpid(); 3065 char *filter = asprintf__tp_filter_pids(1, &pid); 3066 const char * const sc_args[] = { "-e", }; 3067 unsigned int sc_args_nr = ARRAY_SIZE(sc_args); 3068 const char * const majpf_args[] = { "-e", "major-faults" }; 3069 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args); 3070 const char * const minpf_args[] = { "-e", "minor-faults" }; 3071 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args); 3072 int err = -1; 3073 3074 /* +3 is for the event string below and the pid filter */ 3075 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 + 3076 majpf_args_nr + minpf_args_nr + argc; 3077 rec_argv = calloc(rec_argc + 1, sizeof(char *)); 3078 3079 if (rec_argv == NULL || filter == NULL) 3080 goto out_free; 3081 3082 j = 0; 3083 for (i = 0; i < ARRAY_SIZE(record_args); i++) 3084 rec_argv[j++] = record_args[i]; 3085 3086 if (trace->trace_syscalls) { 3087 for (i = 0; i < sc_args_nr; i++) 3088 rec_argv[j++] = sc_args[i]; 3089 3090 /* event string may be different for older kernels - e.g., RHEL6 */ 3091 if (is_valid_tracepoint("raw_syscalls:sys_enter")) 3092 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit"; 3093 else if (is_valid_tracepoint("syscalls:sys_enter")) 3094 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit"; 3095 else { 3096 pr_err("Neither raw_syscalls nor syscalls events exist.\n"); 3097 goto out_free; 3098 } 3099 } 3100 3101 rec_argv[j++] = "--filter"; 3102 rec_argv[j++] = filter; 3103 3104 if (trace->trace_pgfaults & TRACE_PFMAJ) 3105 for (i = 0; i < majpf_args_nr; i++) 3106 rec_argv[j++] = majpf_args[i]; 3107 3108 if (trace->trace_pgfaults & TRACE_PFMIN) 3109 for (i = 0; i < minpf_args_nr; i++) 3110 rec_argv[j++] = minpf_args[i]; 3111 3112 for (i = 0; i < (unsigned int)argc; i++) 3113 rec_argv[j++] = argv[i]; 3114 3115 err = cmd_record(j, rec_argv); 3116out_free: 3117 free(filter); 3118 free(rec_argv); 3119 return err; 3120} 3121 3122static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp); 3123 3124static bool evlist__add_vfs_getname(struct evlist *evlist) 3125{ 3126 bool found = false; 3127 struct evsel *evsel, *tmp; 3128 struct parse_events_error err; 3129 int ret; 3130 3131 parse_events_error__init(&err); 3132 ret = parse_events(evlist, "probe:vfs_getname*", &err); 3133 parse_events_error__exit(&err); 3134 if (ret) 3135 return false; 3136 3137 evlist__for_each_entry_safe(evlist, evsel, tmp) { 3138 if (!strstarts(evsel__name(evsel), "probe:vfs_getname")) 3139 continue; 3140 3141 if (evsel__field(evsel, "pathname")) { 3142 evsel->handler = trace__vfs_getname; 3143 found = true; 3144 continue; 3145 } 3146 3147 list_del_init(&evsel->core.node); 3148 evsel->evlist = NULL; 3149 evsel__delete(evsel); 3150 } 3151 3152 return found; 3153} 3154 3155static struct evsel *evsel__new_pgfault(u64 config) 3156{ 3157 struct evsel *evsel; 3158 struct perf_event_attr attr = { 3159 .type = PERF_TYPE_SOFTWARE, 3160 .mmap_data = 1, 3161 }; 3162 3163 attr.config = config; 3164 attr.sample_period = 1; 3165 3166 event_attr_init(&attr); 3167 3168 evsel = evsel__new(&attr); 3169 if (evsel) 3170 evsel->handler = trace__pgfault; 3171 3172 return evsel; 3173} 3174 3175static void evlist__free_syscall_tp_fields(struct evlist *evlist) 3176{ 3177 struct evsel *evsel; 3178 3179 evlist__for_each_entry(evlist, evsel) { 3180 evsel_trace__delete(evsel->priv); 3181 evsel->priv = NULL; 3182 } 3183} 3184 3185static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample) 3186{ 3187 const u32 type = event->header.type; 3188 struct evsel *evsel; 3189 3190 if (type != PERF_RECORD_SAMPLE) { 3191 trace__process_event(trace, trace->host, event, sample); 3192 return; 3193 } 3194 3195 evsel = evlist__id2evsel(trace->evlist, sample->id); 3196 if (evsel == NULL) { 3197 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id); 3198 return; 3199 } 3200 3201 if (evswitch__discard(&trace->evswitch, evsel)) 3202 return; 3203 3204 trace__set_base_time(trace, evsel, sample); 3205 3206 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT && 3207 sample->raw_data == NULL) { 3208 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n", 3209 evsel__name(evsel), sample->tid, 3210 sample->cpu, sample->raw_size); 3211 } else { 3212 tracepoint_handler handler = evsel->handler; 3213 handler(trace, evsel, event, sample); 3214 } 3215 3216 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX) 3217 interrupted = true; 3218} 3219 3220static int trace__add_syscall_newtp(struct trace *trace) 3221{ 3222 int ret = -1; 3223 struct evlist *evlist = trace->evlist; 3224 struct evsel *sys_enter, *sys_exit; 3225 3226 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter); 3227 if (sys_enter == NULL) 3228 goto out; 3229 3230 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args)) 3231 goto out_delete_sys_enter; 3232 3233 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit); 3234 if (sys_exit == NULL) 3235 goto out_delete_sys_enter; 3236 3237 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret)) 3238 goto out_delete_sys_exit; 3239 3240 evsel__config_callchain(sys_enter, &trace->opts, &callchain_param); 3241 evsel__config_callchain(sys_exit, &trace->opts, &callchain_param); 3242 3243 evlist__add(evlist, sys_enter); 3244 evlist__add(evlist, sys_exit); 3245 3246 if (callchain_param.enabled && !trace->kernel_syscallchains) { 3247 /* 3248 * We're interested only in the user space callchain 3249 * leading to the syscall, allow overriding that for 3250 * debugging reasons using --kernel_syscall_callchains 3251 */ 3252 sys_exit->core.attr.exclude_callchain_kernel = 1; 3253 } 3254 3255 trace->syscalls.events.sys_enter = sys_enter; 3256 trace->syscalls.events.sys_exit = sys_exit; 3257 3258 ret = 0; 3259out: 3260 return ret; 3261 3262out_delete_sys_exit: 3263 evsel__delete_priv(sys_exit); 3264out_delete_sys_enter: 3265 evsel__delete_priv(sys_enter); 3266 goto out; 3267} 3268 3269static int trace__set_ev_qualifier_tp_filter(struct trace *trace) 3270{ 3271 int err = -1; 3272 struct evsel *sys_exit; 3273 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier, 3274 trace->ev_qualifier_ids.nr, 3275 trace->ev_qualifier_ids.entries); 3276 3277 if (filter == NULL) 3278 goto out_enomem; 3279 3280 if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) { 3281 sys_exit = trace->syscalls.events.sys_exit; 3282 err = evsel__append_tp_filter(sys_exit, filter); 3283 } 3284 3285 free(filter); 3286out: 3287 return err; 3288out_enomem: 3289 errno = ENOMEM; 3290 goto out; 3291} 3292 3293#ifdef HAVE_BPF_SKEL 3294static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name) 3295{ 3296 struct bpf_program *pos, *prog = NULL; 3297 const char *sec_name; 3298 3299 if (trace->skel->obj == NULL) 3300 return NULL; 3301 3302 bpf_object__for_each_program(pos, trace->skel->obj) { 3303 sec_name = bpf_program__section_name(pos); 3304 if (sec_name && !strcmp(sec_name, name)) { 3305 prog = pos; 3306 break; 3307 } 3308 } 3309 3310 return prog; 3311} 3312 3313static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc, 3314 const char *prog_name, const char *type) 3315{ 3316 struct bpf_program *prog; 3317 3318 if (prog_name == NULL) { 3319 char default_prog_name[256]; 3320 scnprintf(default_prog_name, sizeof(default_prog_name), "tp/syscalls/sys_%s_%s", type, sc->name); 3321 prog = trace__find_bpf_program_by_title(trace, default_prog_name); 3322 if (prog != NULL) 3323 goto out_found; 3324 if (sc->fmt && sc->fmt->alias) { 3325 scnprintf(default_prog_name, sizeof(default_prog_name), "tp/syscalls/sys_%s_%s", type, sc->fmt->alias); 3326 prog = trace__find_bpf_program_by_title(trace, default_prog_name); 3327 if (prog != NULL) 3328 goto out_found; 3329 } 3330 goto out_unaugmented; 3331 } 3332 3333 prog = trace__find_bpf_program_by_title(trace, prog_name); 3334 3335 if (prog != NULL) { 3336out_found: 3337 return prog; 3338 } 3339 3340 pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n", 3341 prog_name, type, sc->name); 3342out_unaugmented: 3343 return trace->skel->progs.syscall_unaugmented; 3344} 3345 3346static void trace__init_syscall_bpf_progs(struct trace *trace, int id) 3347{ 3348 struct syscall *sc = trace__syscall_info(trace, NULL, id); 3349 3350 if (sc == NULL) 3351 return; 3352 3353 sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter"); 3354 sc->bpf_prog.sys_exit = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit : NULL, "exit"); 3355} 3356 3357static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id) 3358{ 3359 struct syscall *sc = trace__syscall_info(trace, NULL, id); 3360 return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->skel->progs.syscall_unaugmented); 3361} 3362 3363static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id) 3364{ 3365 struct syscall *sc = trace__syscall_info(trace, NULL, id); 3366 return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->skel->progs.syscall_unaugmented); 3367} 3368 3369static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc) 3370{ 3371 struct tep_format_field *field, *candidate_field; 3372 int id; 3373 3374 /* 3375 * We're only interested in syscalls that have a pointer: 3376 */ 3377 for (field = sc->args; field; field = field->next) { 3378 if (field->flags & TEP_FIELD_IS_POINTER) 3379 goto try_to_find_pair; 3380 } 3381 3382 return NULL; 3383 3384try_to_find_pair: 3385 for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) { 3386 struct syscall *pair = trace__syscall_info(trace, NULL, id); 3387 struct bpf_program *pair_prog; 3388 bool is_candidate = false; 3389 3390 if (pair == NULL || pair == sc || 3391 pair->bpf_prog.sys_enter == trace->skel->progs.syscall_unaugmented) 3392 continue; 3393 3394 for (field = sc->args, candidate_field = pair->args; 3395 field && candidate_field; field = field->next, candidate_field = candidate_field->next) { 3396 bool is_pointer = field->flags & TEP_FIELD_IS_POINTER, 3397 candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER; 3398 3399 if (is_pointer) { 3400 if (!candidate_is_pointer) { 3401 // The candidate just doesn't copies our pointer arg, might copy other pointers we want. 3402 continue; 3403 } 3404 } else { 3405 if (candidate_is_pointer) { 3406 // The candidate might copy a pointer we don't have, skip it. 3407 goto next_candidate; 3408 } 3409 continue; 3410 } 3411 3412 if (strcmp(field->type, candidate_field->type)) 3413 goto next_candidate; 3414 3415 /* 3416 * This is limited in the BPF program but sys_write 3417 * uses "const char *" for its "buf" arg so we need to 3418 * use some heuristic that is kinda future proof... 3419 */ 3420 if (strcmp(field->type, "const char *") == 0 && 3421 !(strstr(field->name, "name") || 3422 strstr(field->name, "path") || 3423 strstr(field->name, "file") || 3424 strstr(field->name, "root") || 3425 strstr(field->name, "description"))) 3426 goto next_candidate; 3427 3428 is_candidate = true; 3429 } 3430 3431 if (!is_candidate) 3432 goto next_candidate; 3433 3434 /* 3435 * Check if the tentative pair syscall augmenter has more pointers, if it has, 3436 * then it may be collecting that and we then can't use it, as it would collect 3437 * more than what is common to the two syscalls. 3438 */ 3439 if (candidate_field) { 3440 for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next) 3441 if (candidate_field->flags & TEP_FIELD_IS_POINTER) 3442 goto next_candidate; 3443 } 3444 3445 pair_prog = pair->bpf_prog.sys_enter; 3446 /* 3447 * If the pair isn't enabled, then its bpf_prog.sys_enter will not 3448 * have been searched for, so search it here and if it returns the 3449 * unaugmented one, then ignore it, otherwise we'll reuse that BPF 3450 * program for a filtered syscall on a non-filtered one. 3451 * 3452 * For instance, we have "!syscalls:sys_enter_renameat" and that is 3453 * useful for "renameat2". 3454 */ 3455 if (pair_prog == NULL) { 3456 pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter"); 3457 if (pair_prog == trace->skel->progs.syscall_unaugmented) 3458 goto next_candidate; 3459 } 3460 3461 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name); 3462 return pair_prog; 3463 next_candidate: 3464 continue; 3465 } 3466 3467 return NULL; 3468} 3469 3470static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace) 3471{ 3472 int map_enter_fd = bpf_map__fd(trace->skel->maps.syscalls_sys_enter); 3473 int map_exit_fd = bpf_map__fd(trace->skel->maps.syscalls_sys_exit); 3474 int err = 0, key; 3475 3476 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) { 3477 int prog_fd; 3478 3479 if (!trace__syscall_enabled(trace, key)) 3480 continue; 3481 3482 trace__init_syscall_bpf_progs(trace, key); 3483 3484 // It'll get at least the "!raw_syscalls:unaugmented" 3485 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key); 3486 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY); 3487 if (err) 3488 break; 3489 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key); 3490 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY); 3491 if (err) 3492 break; 3493 } 3494 3495 /* 3496 * Now lets do a second pass looking for enabled syscalls without 3497 * an augmenter that have a signature that is a superset of another 3498 * syscall with an augmenter so that we can auto-reuse it. 3499 * 3500 * I.e. if we have an augmenter for the "open" syscall that has 3501 * this signature: 3502 * 3503 * int open(const char *pathname, int flags, mode_t mode); 3504 * 3505 * I.e. that will collect just the first string argument, then we 3506 * can reuse it for the 'creat' syscall, that has this signature: 3507 * 3508 * int creat(const char *pathname, mode_t mode); 3509 * 3510 * and for: 3511 * 3512 * int stat(const char *pathname, struct stat *statbuf); 3513 * int lstat(const char *pathname, struct stat *statbuf); 3514 * 3515 * Because the 'open' augmenter will collect the first arg as a string, 3516 * and leave alone all the other args, which already helps with 3517 * beautifying 'stat' and 'lstat''s pathname arg. 3518 * 3519 * Then, in time, when 'stat' gets an augmenter that collects both 3520 * first and second arg (this one on the raw_syscalls:sys_exit prog 3521 * array tail call, then that one will be used. 3522 */ 3523 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) { 3524 struct syscall *sc = trace__syscall_info(trace, NULL, key); 3525 struct bpf_program *pair_prog; 3526 int prog_fd; 3527 3528 if (sc == NULL || sc->bpf_prog.sys_enter == NULL) 3529 continue; 3530 3531 /* 3532 * For now we're just reusing the sys_enter prog, and if it 3533 * already has an augmenter, we don't need to find one. 3534 */ 3535 if (sc->bpf_prog.sys_enter != trace->skel->progs.syscall_unaugmented) 3536 continue; 3537 3538 /* 3539 * Look at all the other syscalls for one that has a signature 3540 * that is close enough that we can share: 3541 */ 3542 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc); 3543 if (pair_prog == NULL) 3544 continue; 3545 3546 sc->bpf_prog.sys_enter = pair_prog; 3547 3548 /* 3549 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter 3550 * with the fd for the program we're reusing: 3551 */ 3552 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter); 3553 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY); 3554 if (err) 3555 break; 3556 } 3557 3558 return err; 3559} 3560#endif // HAVE_BPF_SKEL 3561 3562static int trace__set_ev_qualifier_filter(struct trace *trace) 3563{ 3564 if (trace->syscalls.events.sys_enter) 3565 return trace__set_ev_qualifier_tp_filter(trace); 3566 return 0; 3567} 3568 3569static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused, 3570 size_t npids __maybe_unused, pid_t *pids __maybe_unused) 3571{ 3572 int err = 0; 3573#ifdef HAVE_LIBBPF_SUPPORT 3574 bool value = true; 3575 int map_fd = bpf_map__fd(map); 3576 size_t i; 3577 3578 for (i = 0; i < npids; ++i) { 3579 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY); 3580 if (err) 3581 break; 3582 } 3583#endif 3584 return err; 3585} 3586 3587static int trace__set_filter_loop_pids(struct trace *trace) 3588{ 3589 unsigned int nr = 1, err; 3590 pid_t pids[32] = { 3591 getpid(), 3592 }; 3593 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]); 3594 3595 while (thread && nr < ARRAY_SIZE(pids)) { 3596 struct thread *parent = machine__find_thread(trace->host, 3597 thread__ppid(thread), 3598 thread__ppid(thread)); 3599 3600 if (parent == NULL) 3601 break; 3602 3603 if (!strcmp(thread__comm_str(parent), "sshd") || 3604 strstarts(thread__comm_str(parent), "gnome-terminal")) { 3605 pids[nr++] = thread__tid(parent); 3606 break; 3607 } 3608 thread = parent; 3609 } 3610 3611 err = evlist__append_tp_filter_pids(trace->evlist, nr, pids); 3612 if (!err && trace->filter_pids.map) 3613 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids); 3614 3615 return err; 3616} 3617 3618static int trace__set_filter_pids(struct trace *trace) 3619{ 3620 int err = 0; 3621 /* 3622 * Better not use !target__has_task() here because we need to cover the 3623 * case where no threads were specified in the command line, but a 3624 * workload was, and in that case we will fill in the thread_map when 3625 * we fork the workload in evlist__prepare_workload. 3626 */ 3627 if (trace->filter_pids.nr > 0) { 3628 err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr, 3629 trace->filter_pids.entries); 3630 if (!err && trace->filter_pids.map) { 3631 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr, 3632 trace->filter_pids.entries); 3633 } 3634 } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) { 3635 err = trace__set_filter_loop_pids(trace); 3636 } 3637 3638 return err; 3639} 3640 3641static int __trace__deliver_event(struct trace *trace, union perf_event *event) 3642{ 3643 struct evlist *evlist = trace->evlist; 3644 struct perf_sample sample; 3645 int err = evlist__parse_sample(evlist, event, &sample); 3646 3647 if (err) 3648 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err); 3649 else 3650 trace__handle_event(trace, event, &sample); 3651 3652 return 0; 3653} 3654 3655static int __trace__flush_events(struct trace *trace) 3656{ 3657 u64 first = ordered_events__first_time(&trace->oe.data); 3658 u64 flush = trace->oe.last - NSEC_PER_SEC; 3659 3660 /* Is there some thing to flush.. */ 3661 if (first && first < flush) 3662 return ordered_events__flush_time(&trace->oe.data, flush); 3663 3664 return 0; 3665} 3666 3667static int trace__flush_events(struct trace *trace) 3668{ 3669 return !trace->sort_events ? 0 : __trace__flush_events(trace); 3670} 3671 3672static int trace__deliver_event(struct trace *trace, union perf_event *event) 3673{ 3674 int err; 3675 3676 if (!trace->sort_events) 3677 return __trace__deliver_event(trace, event); 3678 3679 err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last); 3680 if (err && err != -1) 3681 return err; 3682 3683 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0, NULL); 3684 if (err) 3685 return err; 3686 3687 return trace__flush_events(trace); 3688} 3689 3690static int ordered_events__deliver_event(struct ordered_events *oe, 3691 struct ordered_event *event) 3692{ 3693 struct trace *trace = container_of(oe, struct trace, oe.data); 3694 3695 return __trace__deliver_event(trace, event->event); 3696} 3697 3698static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg) 3699{ 3700 struct tep_format_field *field; 3701 struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel); 3702 3703 if (evsel->tp_format == NULL || fmt == NULL) 3704 return NULL; 3705 3706 for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt) 3707 if (strcmp(field->name, arg) == 0) 3708 return fmt; 3709 3710 return NULL; 3711} 3712 3713static int trace__expand_filter(struct trace *trace __maybe_unused, struct evsel *evsel) 3714{ 3715 char *tok, *left = evsel->filter, *new_filter = evsel->filter; 3716 3717 while ((tok = strpbrk(left, "=<>!")) != NULL) { 3718 char *right = tok + 1, *right_end; 3719 3720 if (*right == '=') 3721 ++right; 3722 3723 while (isspace(*right)) 3724 ++right; 3725 3726 if (*right == '\0') 3727 break; 3728 3729 while (!isalpha(*left)) 3730 if (++left == tok) { 3731 /* 3732 * Bail out, can't find the name of the argument that is being 3733 * used in the filter, let it try to set this filter, will fail later. 3734 */ 3735 return 0; 3736 } 3737 3738 right_end = right + 1; 3739 while (isalnum(*right_end) || *right_end == '_' || *right_end == '|') 3740 ++right_end; 3741 3742 if (isalpha(*right)) { 3743 struct syscall_arg_fmt *fmt; 3744 int left_size = tok - left, 3745 right_size = right_end - right; 3746 char arg[128]; 3747 3748 while (isspace(left[left_size - 1])) 3749 --left_size; 3750 3751 scnprintf(arg, sizeof(arg), "%.*s", left_size, left); 3752 3753 fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg); 3754 if (fmt == NULL) { 3755 pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n", 3756 arg, evsel->name, evsel->filter); 3757 return -1; 3758 } 3759 3760 pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ", 3761 arg, (int)(right - tok), tok, right_size, right); 3762 3763 if (fmt->strtoul) { 3764 u64 val; 3765 struct syscall_arg syscall_arg = { 3766 .parm = fmt->parm, 3767 }; 3768 3769 if (fmt->strtoul(right, right_size, &syscall_arg, &val)) { 3770 char *n, expansion[19]; 3771 int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val); 3772 int expansion_offset = right - new_filter; 3773 3774 pr_debug("%s", expansion); 3775 3776 if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) { 3777 pr_debug(" out of memory!\n"); 3778 free(new_filter); 3779 return -1; 3780 } 3781 if (new_filter != evsel->filter) 3782 free(new_filter); 3783 left = n + expansion_offset + expansion_lenght; 3784 new_filter = n; 3785 } else { 3786 pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n", 3787 right_size, right, arg, evsel->name, evsel->filter); 3788 return -1; 3789 } 3790 } else { 3791 pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n", 3792 arg, evsel->name, evsel->filter); 3793 return -1; 3794 } 3795 3796 pr_debug("\n"); 3797 } else { 3798 left = right_end; 3799 } 3800 } 3801 3802 if (new_filter != evsel->filter) { 3803 pr_debug("New filter for %s: %s\n", evsel->name, new_filter); 3804 evsel__set_filter(evsel, new_filter); 3805 free(new_filter); 3806 } 3807 3808 return 0; 3809} 3810 3811static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel) 3812{ 3813 struct evlist *evlist = trace->evlist; 3814 struct evsel *evsel; 3815 3816 evlist__for_each_entry(evlist, evsel) { 3817 if (evsel->filter == NULL) 3818 continue; 3819 3820 if (trace__expand_filter(trace, evsel)) { 3821 *err_evsel = evsel; 3822 return -1; 3823 } 3824 } 3825 3826 return 0; 3827} 3828 3829static int trace__run(struct trace *trace, int argc, const char **argv) 3830{ 3831 struct evlist *evlist = trace->evlist; 3832 struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL; 3833 int err = -1, i; 3834 unsigned long before; 3835 const bool forks = argc > 0; 3836 bool draining = false; 3837 3838 trace->live = true; 3839 3840 if (!trace->raw_augmented_syscalls) { 3841 if (trace->trace_syscalls && trace__add_syscall_newtp(trace)) 3842 goto out_error_raw_syscalls; 3843 3844 if (trace->trace_syscalls) 3845 trace->vfs_getname = evlist__add_vfs_getname(evlist); 3846 } 3847 3848 if ((trace->trace_pgfaults & TRACE_PFMAJ)) { 3849 pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ); 3850 if (pgfault_maj == NULL) 3851 goto out_error_mem; 3852 evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param); 3853 evlist__add(evlist, pgfault_maj); 3854 } 3855 3856 if ((trace->trace_pgfaults & TRACE_PFMIN)) { 3857 pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN); 3858 if (pgfault_min == NULL) 3859 goto out_error_mem; 3860 evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param); 3861 evlist__add(evlist, pgfault_min); 3862 } 3863 3864 /* Enable ignoring missing threads when -u/-p option is defined. */ 3865 trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid; 3866 3867 if (trace->sched && 3868 evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime)) 3869 goto out_error_sched_stat_runtime; 3870 /* 3871 * If a global cgroup was set, apply it to all the events without an 3872 * explicit cgroup. I.e.: 3873 * 3874 * trace -G A -e sched:*switch 3875 * 3876 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc 3877 * _and_ sched:sched_switch to the 'A' cgroup, while: 3878 * 3879 * trace -e sched:*switch -G A 3880 * 3881 * will only set the sched:sched_switch event to the 'A' cgroup, all the 3882 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without" 3883 * a cgroup (on the root cgroup, sys wide, etc). 3884 * 3885 * Multiple cgroups: 3886 * 3887 * trace -G A -e sched:*switch -G B 3888 * 3889 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes 3890 * to the 'B' cgroup. 3891 * 3892 * evlist__set_default_cgroup() grabs a reference of the passed cgroup 3893 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL. 3894 */ 3895 if (trace->cgroup) 3896 evlist__set_default_cgroup(trace->evlist, trace->cgroup); 3897 3898 err = evlist__create_maps(evlist, &trace->opts.target); 3899 if (err < 0) { 3900 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n"); 3901 goto out_delete_evlist; 3902 } 3903 3904 err = trace__symbols_init(trace, evlist); 3905 if (err < 0) { 3906 fprintf(trace->output, "Problems initializing symbol libraries!\n"); 3907 goto out_delete_evlist; 3908 } 3909 3910 evlist__config(evlist, &trace->opts, &callchain_param); 3911 3912 if (forks) { 3913 err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL); 3914 if (err < 0) { 3915 fprintf(trace->output, "Couldn't run the workload!\n"); 3916 goto out_delete_evlist; 3917 } 3918 workload_pid = evlist->workload.pid; 3919 } 3920 3921 err = evlist__open(evlist); 3922 if (err < 0) 3923 goto out_error_open; 3924#ifdef HAVE_BPF_SKEL 3925 if (trace->syscalls.events.bpf_output) { 3926 struct perf_cpu cpu; 3927 3928 /* 3929 * Set up the __augmented_syscalls__ BPF map to hold for each 3930 * CPU the bpf-output event's file descriptor. 3931 */ 3932 perf_cpu_map__for_each_cpu(cpu, i, trace->syscalls.events.bpf_output->core.cpus) { 3933 bpf_map__update_elem(trace->skel->maps.__augmented_syscalls__, 3934 &cpu.cpu, sizeof(int), 3935 xyarray__entry(trace->syscalls.events.bpf_output->core.fd, 3936 cpu.cpu, 0), 3937 sizeof(__u32), BPF_ANY); 3938 } 3939 } 3940#endif 3941 err = trace__set_filter_pids(trace); 3942 if (err < 0) 3943 goto out_error_mem; 3944 3945#ifdef HAVE_BPF_SKEL 3946 if (trace->skel && trace->skel->progs.sys_enter) 3947 trace__init_syscalls_bpf_prog_array_maps(trace); 3948#endif 3949 3950 if (trace->ev_qualifier_ids.nr > 0) { 3951 err = trace__set_ev_qualifier_filter(trace); 3952 if (err < 0) 3953 goto out_errno; 3954 3955 if (trace->syscalls.events.sys_exit) { 3956 pr_debug("event qualifier tracepoint filter: %s\n", 3957 trace->syscalls.events.sys_exit->filter); 3958 } 3959 } 3960 3961 /* 3962 * If the "close" syscall is not traced, then we will not have the 3963 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the 3964 * fd->pathname table and were ending up showing the last value set by 3965 * syscalls opening a pathname and associating it with a descriptor or 3966 * reading it from /proc/pid/fd/ in cases where that doesn't make 3967 * sense. 3968 * 3969 * So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is 3970 * not in use. 3971 */ 3972 trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close")); 3973 3974 err = trace__expand_filters(trace, &evsel); 3975 if (err) 3976 goto out_delete_evlist; 3977 err = evlist__apply_filters(evlist, &evsel); 3978 if (err < 0) 3979 goto out_error_apply_filters; 3980 3981 err = evlist__mmap(evlist, trace->opts.mmap_pages); 3982 if (err < 0) 3983 goto out_error_mmap; 3984 3985 if (!target__none(&trace->opts.target) && !trace->opts.target.initial_delay) 3986 evlist__enable(evlist); 3987 3988 if (forks) 3989 evlist__start_workload(evlist); 3990 3991 if (trace->opts.target.initial_delay) { 3992 usleep(trace->opts.target.initial_delay * 1000); 3993 evlist__enable(evlist); 3994 } 3995 3996 trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 || 3997 perf_thread_map__nr(evlist->core.threads) > 1 || 3998 evlist__first(evlist)->core.attr.inherit; 3999 4000 /* 4001 * Now that we already used evsel->core.attr to ask the kernel to setup the 4002 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in 4003 * trace__resolve_callchain(), allowing per-event max-stack settings 4004 * to override an explicitly set --max-stack global setting. 4005 */ 4006 evlist__for_each_entry(evlist, evsel) { 4007 if (evsel__has_callchain(evsel) && 4008 evsel->core.attr.sample_max_stack == 0) 4009 evsel->core.attr.sample_max_stack = trace->max_stack; 4010 } 4011again: 4012 before = trace->nr_events; 4013 4014 for (i = 0; i < evlist->core.nr_mmaps; i++) { 4015 union perf_event *event; 4016 struct mmap *md; 4017 4018 md = &evlist->mmap[i]; 4019 if (perf_mmap__read_init(&md->core) < 0) 4020 continue; 4021 4022 while ((event = perf_mmap__read_event(&md->core)) != NULL) { 4023 ++trace->nr_events; 4024 4025 err = trace__deliver_event(trace, event); 4026 if (err) 4027 goto out_disable; 4028 4029 perf_mmap__consume(&md->core); 4030 4031 if (interrupted) 4032 goto out_disable; 4033 4034 if (done && !draining) { 4035 evlist__disable(evlist); 4036 draining = true; 4037 } 4038 } 4039 perf_mmap__read_done(&md->core); 4040 } 4041 4042 if (trace->nr_events == before) { 4043 int timeout = done ? 100 : -1; 4044 4045 if (!draining && evlist__poll(evlist, timeout) > 0) { 4046 if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0) 4047 draining = true; 4048 4049 goto again; 4050 } else { 4051 if (trace__flush_events(trace)) 4052 goto out_disable; 4053 } 4054 } else { 4055 goto again; 4056 } 4057 4058out_disable: 4059 thread__zput(trace->current); 4060 4061 evlist__disable(evlist); 4062 4063 if (trace->sort_events) 4064 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL); 4065 4066 if (!err) { 4067 if (trace->summary) 4068 trace__fprintf_thread_summary(trace, trace->output); 4069 4070 if (trace->show_tool_stats) { 4071 fprintf(trace->output, "Stats:\n " 4072 " vfs_getname : %" PRIu64 "\n" 4073 " proc_getname: %" PRIu64 "\n", 4074 trace->stats.vfs_getname, 4075 trace->stats.proc_getname); 4076 } 4077 } 4078 4079out_delete_evlist: 4080 trace__symbols__exit(trace); 4081 evlist__free_syscall_tp_fields(evlist); 4082 evlist__delete(evlist); 4083 cgroup__put(trace->cgroup); 4084 trace->evlist = NULL; 4085 trace->live = false; 4086 return err; 4087{ 4088 char errbuf[BUFSIZ]; 4089 4090out_error_sched_stat_runtime: 4091 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime"); 4092 goto out_error; 4093 4094out_error_raw_syscalls: 4095 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)"); 4096 goto out_error; 4097 4098out_error_mmap: 4099 evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf)); 4100 goto out_error; 4101 4102out_error_open: 4103 evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf)); 4104 4105out_error: 4106 fprintf(trace->output, "%s\n", errbuf); 4107 goto out_delete_evlist; 4108 4109out_error_apply_filters: 4110 fprintf(trace->output, 4111 "Failed to set filter \"%s\" on event %s with %d (%s)\n", 4112 evsel->filter, evsel__name(evsel), errno, 4113 str_error_r(errno, errbuf, sizeof(errbuf))); 4114 goto out_delete_evlist; 4115} 4116out_error_mem: 4117 fprintf(trace->output, "Not enough memory to run!\n"); 4118 goto out_delete_evlist; 4119 4120out_errno: 4121 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno)); 4122 goto out_delete_evlist; 4123} 4124 4125static int trace__replay(struct trace *trace) 4126{ 4127 const struct evsel_str_handler handlers[] = { 4128 { "probe:vfs_getname", trace__vfs_getname, }, 4129 }; 4130 struct perf_data data = { 4131 .path = input_name, 4132 .mode = PERF_DATA_MODE_READ, 4133 .force = trace->force, 4134 }; 4135 struct perf_session *session; 4136 struct evsel *evsel; 4137 int err = -1; 4138 4139 trace->tool.sample = trace__process_sample; 4140 trace->tool.mmap = perf_event__process_mmap; 4141 trace->tool.mmap2 = perf_event__process_mmap2; 4142 trace->tool.comm = perf_event__process_comm; 4143 trace->tool.exit = perf_event__process_exit; 4144 trace->tool.fork = perf_event__process_fork; 4145 trace->tool.attr = perf_event__process_attr; 4146 trace->tool.tracing_data = perf_event__process_tracing_data; 4147 trace->tool.build_id = perf_event__process_build_id; 4148 trace->tool.namespaces = perf_event__process_namespaces; 4149 4150 trace->tool.ordered_events = true; 4151 trace->tool.ordering_requires_timestamps = true; 4152 4153 /* add tid to output */ 4154 trace->multiple_threads = true; 4155 4156 session = perf_session__new(&data, &trace->tool); 4157 if (IS_ERR(session)) 4158 return PTR_ERR(session); 4159 4160 if (trace->opts.target.pid) 4161 symbol_conf.pid_list_str = strdup(trace->opts.target.pid); 4162 4163 if (trace->opts.target.tid) 4164 symbol_conf.tid_list_str = strdup(trace->opts.target.tid); 4165 4166 if (symbol__init(&session->header.env) < 0) 4167 goto out; 4168 4169 trace->host = &session->machines.host; 4170 4171 err = perf_session__set_tracepoints_handlers(session, handlers); 4172 if (err) 4173 goto out; 4174 4175 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter"); 4176 trace->syscalls.events.sys_enter = evsel; 4177 /* older kernels have syscalls tp versus raw_syscalls */ 4178 if (evsel == NULL) 4179 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter"); 4180 4181 if (evsel && 4182 (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 || 4183 perf_evsel__init_sc_tp_ptr_field(evsel, args))) { 4184 pr_err("Error during initialize raw_syscalls:sys_enter event\n"); 4185 goto out; 4186 } 4187 4188 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit"); 4189 trace->syscalls.events.sys_exit = evsel; 4190 if (evsel == NULL) 4191 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit"); 4192 if (evsel && 4193 (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 || 4194 perf_evsel__init_sc_tp_uint_field(evsel, ret))) { 4195 pr_err("Error during initialize raw_syscalls:sys_exit event\n"); 4196 goto out; 4197 } 4198 4199 evlist__for_each_entry(session->evlist, evsel) { 4200 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE && 4201 (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ || 4202 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN || 4203 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS)) 4204 evsel->handler = trace__pgfault; 4205 } 4206 4207 setup_pager(); 4208 4209 err = perf_session__process_events(session); 4210 if (err) 4211 pr_err("Failed to process events, error %d", err); 4212 4213 else if (trace->summary) 4214 trace__fprintf_thread_summary(trace, trace->output); 4215 4216out: 4217 perf_session__delete(session); 4218 4219 return err; 4220} 4221 4222static size_t trace__fprintf_threads_header(FILE *fp) 4223{ 4224 size_t printed; 4225 4226 printed = fprintf(fp, "\n Summary of events:\n\n"); 4227 4228 return printed; 4229} 4230 4231DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs, 4232 struct syscall_stats *stats; 4233 double msecs; 4234 int syscall; 4235) 4236{ 4237 struct int_node *source = rb_entry(nd, struct int_node, rb_node); 4238 struct syscall_stats *stats = source->priv; 4239 4240 entry->syscall = source->i; 4241 entry->stats = stats; 4242 entry->msecs = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0; 4243} 4244 4245static size_t thread__dump_stats(struct thread_trace *ttrace, 4246 struct trace *trace, FILE *fp) 4247{ 4248 size_t printed = 0; 4249 struct syscall *sc; 4250 struct rb_node *nd; 4251 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats); 4252 4253 if (syscall_stats == NULL) 4254 return 0; 4255 4256 printed += fprintf(fp, "\n"); 4257 4258 printed += fprintf(fp, " syscall calls errors total min avg max stddev\n"); 4259 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n"); 4260 printed += fprintf(fp, " --------------- -------- ------ -------- --------- --------- --------- ------\n"); 4261 4262 resort_rb__for_each_entry(nd, syscall_stats) { 4263 struct syscall_stats *stats = syscall_stats_entry->stats; 4264 if (stats) { 4265 double min = (double)(stats->stats.min) / NSEC_PER_MSEC; 4266 double max = (double)(stats->stats.max) / NSEC_PER_MSEC; 4267 double avg = avg_stats(&stats->stats); 4268 double pct; 4269 u64 n = (u64)stats->stats.n; 4270 4271 pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0; 4272 avg /= NSEC_PER_MSEC; 4273 4274 sc = &trace->syscalls.table[syscall_stats_entry->syscall]; 4275 printed += fprintf(fp, " %-15s", sc->name); 4276 printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f", 4277 n, stats->nr_failures, syscall_stats_entry->msecs, min, avg); 4278 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct); 4279 4280 if (trace->errno_summary && stats->nr_failures) { 4281 int e; 4282 4283 for (e = 0; e < stats->max_errno; ++e) { 4284 if (stats->errnos[e] != 0) 4285 fprintf(fp, "\t\t\t\t%s: %d\n", perf_env__arch_strerrno(trace->host->env, e + 1), stats->errnos[e]); 4286 } 4287 } 4288 } 4289 } 4290 4291 resort_rb__delete(syscall_stats); 4292 printed += fprintf(fp, "\n\n"); 4293 4294 return printed; 4295} 4296 4297static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace) 4298{ 4299 size_t printed = 0; 4300 struct thread_trace *ttrace = thread__priv(thread); 4301 double ratio; 4302 4303 if (ttrace == NULL) 4304 return 0; 4305 4306 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0; 4307 4308 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread__tid(thread)); 4309 printed += fprintf(fp, "%lu events, ", ttrace->nr_events); 4310 printed += fprintf(fp, "%.1f%%", ratio); 4311 if (ttrace->pfmaj) 4312 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj); 4313 if (ttrace->pfmin) 4314 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin); 4315 if (trace->sched) 4316 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms); 4317 else if (fputc('\n', fp) != EOF) 4318 ++printed; 4319 4320 printed += thread__dump_stats(ttrace, trace, fp); 4321 4322 return printed; 4323} 4324 4325static unsigned long thread__nr_events(struct thread_trace *ttrace) 4326{ 4327 return ttrace ? ttrace->nr_events : 0; 4328} 4329 4330static int trace_nr_events_cmp(void *priv __maybe_unused, 4331 const struct list_head *la, 4332 const struct list_head *lb) 4333{ 4334 struct thread_list *a = list_entry(la, struct thread_list, list); 4335 struct thread_list *b = list_entry(lb, struct thread_list, list); 4336 unsigned long a_nr_events = thread__nr_events(thread__priv(a->thread)); 4337 unsigned long b_nr_events = thread__nr_events(thread__priv(b->thread)); 4338 4339 if (a_nr_events != b_nr_events) 4340 return a_nr_events < b_nr_events ? -1 : 1; 4341 4342 /* Identical number of threads, place smaller tids first. */ 4343 return thread__tid(a->thread) < thread__tid(b->thread) 4344 ? -1 4345 : (thread__tid(a->thread) > thread__tid(b->thread) ? 1 : 0); 4346} 4347 4348static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp) 4349{ 4350 size_t printed = trace__fprintf_threads_header(fp); 4351 LIST_HEAD(threads); 4352 4353 if (machine__thread_list(trace->host, &threads) == 0) { 4354 struct thread_list *pos; 4355 4356 list_sort(NULL, &threads, trace_nr_events_cmp); 4357 4358 list_for_each_entry(pos, &threads, list) 4359 printed += trace__fprintf_thread(fp, pos->thread, trace); 4360 } 4361 thread_list__delete(&threads); 4362 return printed; 4363} 4364 4365static int trace__set_duration(const struct option *opt, const char *str, 4366 int unset __maybe_unused) 4367{ 4368 struct trace *trace = opt->value; 4369 4370 trace->duration_filter = atof(str); 4371 return 0; 4372} 4373 4374static int trace__set_filter_pids_from_option(const struct option *opt, const char *str, 4375 int unset __maybe_unused) 4376{ 4377 int ret = -1; 4378 size_t i; 4379 struct trace *trace = opt->value; 4380 /* 4381 * FIXME: introduce a intarray class, plain parse csv and create a 4382 * { int nr, int entries[] } struct... 4383 */ 4384 struct intlist *list = intlist__new(str); 4385 4386 if (list == NULL) 4387 return -1; 4388 4389 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1; 4390 trace->filter_pids.entries = calloc(i, sizeof(pid_t)); 4391 4392 if (trace->filter_pids.entries == NULL) 4393 goto out; 4394 4395 trace->filter_pids.entries[0] = getpid(); 4396 4397 for (i = 1; i < trace->filter_pids.nr; ++i) 4398 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i; 4399 4400 intlist__delete(list); 4401 ret = 0; 4402out: 4403 return ret; 4404} 4405 4406static int trace__open_output(struct trace *trace, const char *filename) 4407{ 4408 struct stat st; 4409 4410 if (!stat(filename, &st) && st.st_size) { 4411 char oldname[PATH_MAX]; 4412 4413 scnprintf(oldname, sizeof(oldname), "%s.old", filename); 4414 unlink(oldname); 4415 rename(filename, oldname); 4416 } 4417 4418 trace->output = fopen(filename, "w"); 4419 4420 return trace->output == NULL ? -errno : 0; 4421} 4422 4423static int parse_pagefaults(const struct option *opt, const char *str, 4424 int unset __maybe_unused) 4425{ 4426 int *trace_pgfaults = opt->value; 4427 4428 if (strcmp(str, "all") == 0) 4429 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN; 4430 else if (strcmp(str, "maj") == 0) 4431 *trace_pgfaults |= TRACE_PFMAJ; 4432 else if (strcmp(str, "min") == 0) 4433 *trace_pgfaults |= TRACE_PFMIN; 4434 else 4435 return -1; 4436 4437 return 0; 4438} 4439 4440static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler) 4441{ 4442 struct evsel *evsel; 4443 4444 evlist__for_each_entry(evlist, evsel) { 4445 if (evsel->handler == NULL) 4446 evsel->handler = handler; 4447 } 4448} 4449 4450static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name) 4451{ 4452 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel); 4453 4454 if (fmt) { 4455 const struct syscall_fmt *scfmt = syscall_fmt__find(name); 4456 4457 if (scfmt) { 4458 int skip = 0; 4459 4460 if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 || 4461 strcmp(evsel->tp_format->format.fields->name, "nr") == 0) 4462 ++skip; 4463 4464 memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt)); 4465 } 4466 } 4467} 4468 4469static int evlist__set_syscall_tp_fields(struct evlist *evlist) 4470{ 4471 struct evsel *evsel; 4472 4473 evlist__for_each_entry(evlist, evsel) { 4474 if (evsel->priv || !evsel->tp_format) 4475 continue; 4476 4477 if (strcmp(evsel->tp_format->system, "syscalls")) { 4478 evsel__init_tp_arg_scnprintf(evsel); 4479 continue; 4480 } 4481 4482 if (evsel__init_syscall_tp(evsel)) 4483 return -1; 4484 4485 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) { 4486 struct syscall_tp *sc = __evsel__syscall_tp(evsel); 4487 4488 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64))) 4489 return -1; 4490 4491 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1); 4492 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) { 4493 struct syscall_tp *sc = __evsel__syscall_tp(evsel); 4494 4495 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap)) 4496 return -1; 4497 4498 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1); 4499 } 4500 } 4501 4502 return 0; 4503} 4504 4505/* 4506 * XXX: Hackish, just splitting the combined -e+--event (syscalls 4507 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use 4508 * existing facilities unchanged (trace->ev_qualifier + parse_options()). 4509 * 4510 * It'd be better to introduce a parse_options() variant that would return a 4511 * list with the terms it didn't match to an event... 4512 */ 4513static int trace__parse_events_option(const struct option *opt, const char *str, 4514 int unset __maybe_unused) 4515{ 4516 struct trace *trace = (struct trace *)opt->value; 4517 const char *s = str; 4518 char *sep = NULL, *lists[2] = { NULL, NULL, }; 4519 int len = strlen(str) + 1, err = -1, list, idx; 4520 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR); 4521 char group_name[PATH_MAX]; 4522 const struct syscall_fmt *fmt; 4523 4524 if (strace_groups_dir == NULL) 4525 return -1; 4526 4527 if (*s == '!') { 4528 ++s; 4529 trace->not_ev_qualifier = true; 4530 } 4531 4532 while (1) { 4533 if ((sep = strchr(s, ',')) != NULL) 4534 *sep = '\0'; 4535 4536 list = 0; 4537 if (syscalltbl__id(trace->sctbl, s) >= 0 || 4538 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) { 4539 list = 1; 4540 goto do_concat; 4541 } 4542 4543 fmt = syscall_fmt__find_by_alias(s); 4544 if (fmt != NULL) { 4545 list = 1; 4546 s = fmt->name; 4547 } else { 4548 path__join(group_name, sizeof(group_name), strace_groups_dir, s); 4549 if (access(group_name, R_OK) == 0) 4550 list = 1; 4551 } 4552do_concat: 4553 if (lists[list]) { 4554 sprintf(lists[list] + strlen(lists[list]), ",%s", s); 4555 } else { 4556 lists[list] = malloc(len); 4557 if (lists[list] == NULL) 4558 goto out; 4559 strcpy(lists[list], s); 4560 } 4561 4562 if (!sep) 4563 break; 4564 4565 *sep = ','; 4566 s = sep + 1; 4567 } 4568 4569 if (lists[1] != NULL) { 4570 struct strlist_config slist_config = { 4571 .dirname = strace_groups_dir, 4572 }; 4573 4574 trace->ev_qualifier = strlist__new(lists[1], &slist_config); 4575 if (trace->ev_qualifier == NULL) { 4576 fputs("Not enough memory to parse event qualifier", trace->output); 4577 goto out; 4578 } 4579 4580 if (trace__validate_ev_qualifier(trace)) 4581 goto out; 4582 trace->trace_syscalls = true; 4583 } 4584 4585 err = 0; 4586 4587 if (lists[0]) { 4588 struct parse_events_option_args parse_events_option_args = { 4589 .evlistp = &trace->evlist, 4590 }; 4591 struct option o = { 4592 .value = &parse_events_option_args, 4593 }; 4594 err = parse_events_option(&o, lists[0], 0); 4595 } 4596out: 4597 free(strace_groups_dir); 4598 free(lists[0]); 4599 free(lists[1]); 4600 if (sep) 4601 *sep = ','; 4602 4603 return err; 4604} 4605 4606static int trace__parse_cgroups(const struct option *opt, const char *str, int unset) 4607{ 4608 struct trace *trace = opt->value; 4609 4610 if (!list_empty(&trace->evlist->core.entries)) { 4611 struct option o = { 4612 .value = &trace->evlist, 4613 }; 4614 return parse_cgroups(&o, str, unset); 4615 } 4616 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str); 4617 4618 return 0; 4619} 4620 4621static int trace__config(const char *var, const char *value, void *arg) 4622{ 4623 struct trace *trace = arg; 4624 int err = 0; 4625 4626 if (!strcmp(var, "trace.add_events")) { 4627 trace->perfconfig_events = strdup(value); 4628 if (trace->perfconfig_events == NULL) { 4629 pr_err("Not enough memory for %s\n", "trace.add_events"); 4630 return -1; 4631 } 4632 } else if (!strcmp(var, "trace.show_timestamp")) { 4633 trace->show_tstamp = perf_config_bool(var, value); 4634 } else if (!strcmp(var, "trace.show_duration")) { 4635 trace->show_duration = perf_config_bool(var, value); 4636 } else if (!strcmp(var, "trace.show_arg_names")) { 4637 trace->show_arg_names = perf_config_bool(var, value); 4638 if (!trace->show_arg_names) 4639 trace->show_zeros = true; 4640 } else if (!strcmp(var, "trace.show_zeros")) { 4641 bool new_show_zeros = perf_config_bool(var, value); 4642 if (!trace->show_arg_names && !new_show_zeros) { 4643 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n"); 4644 goto out; 4645 } 4646 trace->show_zeros = new_show_zeros; 4647 } else if (!strcmp(var, "trace.show_prefix")) { 4648 trace->show_string_prefix = perf_config_bool(var, value); 4649 } else if (!strcmp(var, "trace.no_inherit")) { 4650 trace->opts.no_inherit = perf_config_bool(var, value); 4651 } else if (!strcmp(var, "trace.args_alignment")) { 4652 int args_alignment = 0; 4653 if (perf_config_int(&args_alignment, var, value) == 0) 4654 trace->args_alignment = args_alignment; 4655 } else if (!strcmp(var, "trace.tracepoint_beautifiers")) { 4656 if (strcasecmp(value, "libtraceevent") == 0) 4657 trace->libtraceevent_print = true; 4658 else if (strcasecmp(value, "libbeauty") == 0) 4659 trace->libtraceevent_print = false; 4660 } 4661out: 4662 return err; 4663} 4664 4665static void trace__exit(struct trace *trace) 4666{ 4667 int i; 4668 4669 strlist__delete(trace->ev_qualifier); 4670 zfree(&trace->ev_qualifier_ids.entries); 4671 if (trace->syscalls.table) { 4672 for (i = 0; i <= trace->sctbl->syscalls.max_id; i++) 4673 syscall__exit(&trace->syscalls.table[i]); 4674 zfree(&trace->syscalls.table); 4675 } 4676 syscalltbl__delete(trace->sctbl); 4677 zfree(&trace->perfconfig_events); 4678} 4679 4680#ifdef HAVE_BPF_SKEL 4681static int bpf__setup_bpf_output(struct evlist *evlist) 4682{ 4683 int err = parse_event(evlist, "bpf-output/no-inherit=1,name=__augmented_syscalls__/"); 4684 4685 if (err) 4686 pr_debug("ERROR: failed to create the \"__augmented_syscalls__\" bpf-output event\n"); 4687 4688 return err; 4689} 4690#endif 4691 4692int cmd_trace(int argc, const char **argv) 4693{ 4694 const char *trace_usage[] = { 4695 "perf trace [<options>] [<command>]", 4696 "perf trace [<options>] -- <command> [<options>]", 4697 "perf trace record [<options>] [<command>]", 4698 "perf trace record [<options>] -- <command> [<options>]", 4699 NULL 4700 }; 4701 struct trace trace = { 4702 .opts = { 4703 .target = { 4704 .uid = UINT_MAX, 4705 .uses_mmap = true, 4706 }, 4707 .user_freq = UINT_MAX, 4708 .user_interval = ULLONG_MAX, 4709 .no_buffering = true, 4710 .mmap_pages = UINT_MAX, 4711 }, 4712 .output = stderr, 4713 .show_comm = true, 4714 .show_tstamp = true, 4715 .show_duration = true, 4716 .show_arg_names = true, 4717 .args_alignment = 70, 4718 .trace_syscalls = false, 4719 .kernel_syscallchains = false, 4720 .max_stack = UINT_MAX, 4721 .max_events = ULONG_MAX, 4722 }; 4723 const char *output_name = NULL; 4724 const struct option trace_options[] = { 4725 OPT_CALLBACK('e', "event", &trace, "event", 4726 "event/syscall selector. use 'perf list' to list available events", 4727 trace__parse_events_option), 4728 OPT_CALLBACK(0, "filter", &trace.evlist, "filter", 4729 "event filter", parse_filter), 4730 OPT_BOOLEAN(0, "comm", &trace.show_comm, 4731 "show the thread COMM next to its id"), 4732 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"), 4733 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace", 4734 trace__parse_events_option), 4735 OPT_STRING('o', "output", &output_name, "file", "output file name"), 4736 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"), 4737 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid", 4738 "trace events on existing process id"), 4739 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid", 4740 "trace events on existing thread id"), 4741 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids", 4742 "pids to filter (by the kernel)", trace__set_filter_pids_from_option), 4743 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide, 4744 "system-wide collection from all CPUs"), 4745 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu", 4746 "list of cpus to monitor"), 4747 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit, 4748 "child tasks do not inherit counters"), 4749 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages", 4750 "number of mmap data pages", evlist__parse_mmap_pages), 4751 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user", 4752 "user to profile"), 4753 OPT_CALLBACK(0, "duration", &trace, "float", 4754 "show only events with duration > N.M ms", 4755 trace__set_duration), 4756 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"), 4757 OPT_INCR('v', "verbose", &verbose, "be more verbose"), 4758 OPT_BOOLEAN('T', "time", &trace.full_time, 4759 "Show full timestamp, not time relative to first start"), 4760 OPT_BOOLEAN(0, "failure", &trace.failure_only, 4761 "Show only syscalls that failed"), 4762 OPT_BOOLEAN('s', "summary", &trace.summary_only, 4763 "Show only syscall summary with statistics"), 4764 OPT_BOOLEAN('S', "with-summary", &trace.summary, 4765 "Show all syscalls and summary with statistics"), 4766 OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary, 4767 "Show errno stats per syscall, use with -s or -S"), 4768 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min", 4769 "Trace pagefaults", parse_pagefaults, "maj"), 4770 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"), 4771 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"), 4772 OPT_CALLBACK(0, "call-graph", &trace.opts, 4773 "record_mode[,record_size]", record_callchain_help, 4774 &record_parse_callchain_opt), 4775 OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print, 4776 "Use libtraceevent to print the tracepoint arguments."), 4777 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains, 4778 "Show the kernel callchains on the syscall exit path"), 4779 OPT_ULONG(0, "max-events", &trace.max_events, 4780 "Set the maximum number of events to print, exit after that is reached. "), 4781 OPT_UINTEGER(0, "min-stack", &trace.min_stack, 4782 "Set the minimum stack depth when parsing the callchain, " 4783 "anything below the specified depth will be ignored."), 4784 OPT_UINTEGER(0, "max-stack", &trace.max_stack, 4785 "Set the maximum stack depth when parsing the callchain, " 4786 "anything beyond the specified depth will be ignored. " 4787 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 4788 OPT_BOOLEAN(0, "sort-events", &trace.sort_events, 4789 "Sort batch of events before processing, use if getting out of order events"), 4790 OPT_BOOLEAN(0, "print-sample", &trace.print_sample, 4791 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"), 4792 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout, 4793 "per thread proc mmap processing timeout in ms"), 4794 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only", 4795 trace__parse_cgroups), 4796 OPT_INTEGER('D', "delay", &trace.opts.target.initial_delay, 4797 "ms to wait before starting measurement after program " 4798 "start"), 4799 OPTS_EVSWITCH(&trace.evswitch), 4800 OPT_END() 4801 }; 4802 bool __maybe_unused max_stack_user_set = true; 4803 bool mmap_pages_user_set = true; 4804 struct evsel *evsel; 4805 const char * const trace_subcommands[] = { "record", NULL }; 4806 int err = -1; 4807 char bf[BUFSIZ]; 4808 struct sigaction sigchld_act; 4809 4810 signal(SIGSEGV, sighandler_dump_stack); 4811 signal(SIGFPE, sighandler_dump_stack); 4812 signal(SIGINT, sighandler_interrupt); 4813 4814 memset(&sigchld_act, 0, sizeof(sigchld_act)); 4815 sigchld_act.sa_flags = SA_SIGINFO; 4816 sigchld_act.sa_sigaction = sighandler_chld; 4817 sigaction(SIGCHLD, &sigchld_act, NULL); 4818 4819 trace.evlist = evlist__new(); 4820 trace.sctbl = syscalltbl__new(); 4821 4822 if (trace.evlist == NULL || trace.sctbl == NULL) { 4823 pr_err("Not enough memory to run!\n"); 4824 err = -ENOMEM; 4825 goto out; 4826 } 4827 4828 /* 4829 * Parsing .perfconfig may entail creating a BPF event, that may need 4830 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting 4831 * is too small. This affects just this process, not touching the 4832 * global setting. If it fails we'll get something in 'perf trace -v' 4833 * to help diagnose the problem. 4834 */ 4835 rlimit__bump_memlock(); 4836 4837 err = perf_config(trace__config, &trace); 4838 if (err) 4839 goto out; 4840 4841 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands, 4842 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION); 4843 4844 /* 4845 * Here we already passed thru trace__parse_events_option() and it has 4846 * already figured out if -e syscall_name, if not but if --event 4847 * foo:bar was used, the user is interested _just_ in those, say, 4848 * tracepoint events, not in the strace-like syscall-name-based mode. 4849 * 4850 * This is important because we need to check if strace-like mode is 4851 * needed to decided if we should filter out the eBPF 4852 * __augmented_syscalls__ code, if it is in the mix, say, via 4853 * .perfconfig trace.add_events, and filter those out. 4854 */ 4855 if (!trace.trace_syscalls && !trace.trace_pgfaults && 4856 trace.evlist->core.nr_entries == 0 /* Was --events used? */) { 4857 trace.trace_syscalls = true; 4858 } 4859 /* 4860 * Now that we have --verbose figured out, lets see if we need to parse 4861 * events from .perfconfig, so that if those events fail parsing, say some 4862 * BPF program fails, then we'll be able to use --verbose to see what went 4863 * wrong in more detail. 4864 */ 4865 if (trace.perfconfig_events != NULL) { 4866 struct parse_events_error parse_err; 4867 4868 parse_events_error__init(&parse_err); 4869 err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err); 4870 if (err) 4871 parse_events_error__print(&parse_err, trace.perfconfig_events); 4872 parse_events_error__exit(&parse_err); 4873 if (err) 4874 goto out; 4875 } 4876 4877 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) { 4878 usage_with_options_msg(trace_usage, trace_options, 4879 "cgroup monitoring only available in system-wide mode"); 4880 } 4881 4882#ifdef HAVE_BPF_SKEL 4883 if (!trace.trace_syscalls) 4884 goto skip_augmentation; 4885 4886 if ((argc >= 1) && (strcmp(argv[0], "record") == 0)) { 4887 pr_debug("Syscall augmentation fails with record, disabling augmentation"); 4888 goto skip_augmentation; 4889 } 4890 4891 trace.skel = augmented_raw_syscalls_bpf__open(); 4892 if (!trace.skel) { 4893 pr_debug("Failed to open augmented syscalls BPF skeleton"); 4894 } else { 4895 /* 4896 * Disable attaching the BPF programs except for sys_enter and 4897 * sys_exit that tail call into this as necessary. 4898 */ 4899 struct bpf_program *prog; 4900 4901 bpf_object__for_each_program(prog, trace.skel->obj) { 4902 if (prog != trace.skel->progs.sys_enter && prog != trace.skel->progs.sys_exit) 4903 bpf_program__set_autoattach(prog, /*autoattach=*/false); 4904 } 4905 4906 err = augmented_raw_syscalls_bpf__load(trace.skel); 4907 4908 if (err < 0) { 4909 libbpf_strerror(err, bf, sizeof(bf)); 4910 pr_debug("Failed to load augmented syscalls BPF skeleton: %s\n", bf); 4911 } else { 4912 augmented_raw_syscalls_bpf__attach(trace.skel); 4913 trace__add_syscall_newtp(&trace); 4914 } 4915 } 4916 4917 err = bpf__setup_bpf_output(trace.evlist); 4918 if (err) { 4919 libbpf_strerror(err, bf, sizeof(bf)); 4920 pr_err("ERROR: Setup BPF output event failed: %s\n", bf); 4921 goto out; 4922 } 4923 trace.syscalls.events.bpf_output = evlist__last(trace.evlist); 4924 assert(evsel__name_is(trace.syscalls.events.bpf_output, "__augmented_syscalls__")); 4925skip_augmentation: 4926#endif 4927 err = -1; 4928 4929 if (trace.trace_pgfaults) { 4930 trace.opts.sample_address = true; 4931 trace.opts.sample_time = true; 4932 } 4933 4934 if (trace.opts.mmap_pages == UINT_MAX) 4935 mmap_pages_user_set = false; 4936 4937 if (trace.max_stack == UINT_MAX) { 4938 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack(); 4939 max_stack_user_set = false; 4940 } 4941 4942#ifdef HAVE_DWARF_UNWIND_SUPPORT 4943 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) { 4944 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false); 4945 } 4946#endif 4947 4948 if (callchain_param.enabled) { 4949 if (!mmap_pages_user_set && geteuid() == 0) 4950 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4; 4951 4952 symbol_conf.use_callchain = true; 4953 } 4954 4955 if (trace.evlist->core.nr_entries > 0) { 4956 evlist__set_default_evsel_handler(trace.evlist, trace__event_handler); 4957 if (evlist__set_syscall_tp_fields(trace.evlist)) { 4958 perror("failed to set syscalls:* tracepoint fields"); 4959 goto out; 4960 } 4961 } 4962 4963 if (trace.sort_events) { 4964 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace); 4965 ordered_events__set_copy_on_queue(&trace.oe.data, true); 4966 } 4967 4968 /* 4969 * If we are augmenting syscalls, then combine what we put in the 4970 * __augmented_syscalls__ BPF map with what is in the 4971 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF, 4972 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit. 4973 * 4974 * We'll switch to look at two BPF maps, one for sys_enter and the 4975 * other for sys_exit when we start augmenting the sys_exit paths with 4976 * buffers that are being copied from kernel to userspace, think 'read' 4977 * syscall. 4978 */ 4979 if (trace.syscalls.events.bpf_output) { 4980 evlist__for_each_entry(trace.evlist, evsel) { 4981 bool raw_syscalls_sys_exit = evsel__name_is(evsel, "raw_syscalls:sys_exit"); 4982 4983 if (raw_syscalls_sys_exit) { 4984 trace.raw_augmented_syscalls = true; 4985 goto init_augmented_syscall_tp; 4986 } 4987 4988 if (trace.syscalls.events.bpf_output->priv == NULL && 4989 strstr(evsel__name(evsel), "syscalls:sys_enter")) { 4990 struct evsel *augmented = trace.syscalls.events.bpf_output; 4991 if (evsel__init_augmented_syscall_tp(augmented, evsel) || 4992 evsel__init_augmented_syscall_tp_args(augmented)) 4993 goto out; 4994 /* 4995 * Augmented is __augmented_syscalls__ BPF_OUTPUT event 4996 * Above we made sure we can get from the payload the tp fields 4997 * that we get from syscalls:sys_enter tracefs format file. 4998 */ 4999 augmented->handler = trace__sys_enter; 5000 /* 5001 * Now we do the same for the *syscalls:sys_enter event so that 5002 * if we handle it directly, i.e. if the BPF prog returns 0 so 5003 * as not to filter it, then we'll handle it just like we would 5004 * for the BPF_OUTPUT one: 5005 */ 5006 if (evsel__init_augmented_syscall_tp(evsel, evsel) || 5007 evsel__init_augmented_syscall_tp_args(evsel)) 5008 goto out; 5009 evsel->handler = trace__sys_enter; 5010 } 5011 5012 if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) { 5013 struct syscall_tp *sc; 5014init_augmented_syscall_tp: 5015 if (evsel__init_augmented_syscall_tp(evsel, evsel)) 5016 goto out; 5017 sc = __evsel__syscall_tp(evsel); 5018 /* 5019 * For now with BPF raw_augmented we hook into 5020 * raw_syscalls:sys_enter and there we get all 5021 * 6 syscall args plus the tracepoint common 5022 * fields and the syscall_nr (another long). 5023 * So we check if that is the case and if so 5024 * don't look after the sc->args_size but 5025 * always after the full raw_syscalls:sys_enter 5026 * payload, which is fixed. 5027 * 5028 * We'll revisit this later to pass 5029 * s->args_size to the BPF augmenter (now 5030 * tools/perf/examples/bpf/augmented_raw_syscalls.c, 5031 * so that it copies only what we need for each 5032 * syscall, like what happens when we use 5033 * syscalls:sys_enter_NAME, so that we reduce 5034 * the kernel/userspace traffic to just what is 5035 * needed for each syscall. 5036 */ 5037 if (trace.raw_augmented_syscalls) 5038 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset; 5039 evsel__init_augmented_syscall_tp_ret(evsel); 5040 evsel->handler = trace__sys_exit; 5041 } 5042 } 5043 } 5044 5045 if ((argc >= 1) && (strcmp(argv[0], "record") == 0)) 5046 return trace__record(&trace, argc-1, &argv[1]); 5047 5048 /* Using just --errno-summary will trigger --summary */ 5049 if (trace.errno_summary && !trace.summary && !trace.summary_only) 5050 trace.summary_only = true; 5051 5052 /* summary_only implies summary option, but don't overwrite summary if set */ 5053 if (trace.summary_only) 5054 trace.summary = trace.summary_only; 5055 5056 if (output_name != NULL) { 5057 err = trace__open_output(&trace, output_name); 5058 if (err < 0) { 5059 perror("failed to create output file"); 5060 goto out; 5061 } 5062 } 5063 5064 err = evswitch__init(&trace.evswitch, trace.evlist, stderr); 5065 if (err) 5066 goto out_close; 5067 5068 err = target__validate(&trace.opts.target); 5069 if (err) { 5070 target__strerror(&trace.opts.target, err, bf, sizeof(bf)); 5071 fprintf(trace.output, "%s", bf); 5072 goto out_close; 5073 } 5074 5075 err = target__parse_uid(&trace.opts.target); 5076 if (err) { 5077 target__strerror(&trace.opts.target, err, bf, sizeof(bf)); 5078 fprintf(trace.output, "%s", bf); 5079 goto out_close; 5080 } 5081 5082 if (!argc && target__none(&trace.opts.target)) 5083 trace.opts.target.system_wide = true; 5084 5085 if (input_name) 5086 err = trace__replay(&trace); 5087 else 5088 err = trace__run(&trace, argc, argv); 5089 5090out_close: 5091 if (output_name != NULL) 5092 fclose(trace.output); 5093out: 5094 trace__exit(&trace); 5095#ifdef HAVE_BPF_SKEL 5096 augmented_raw_syscalls_bpf__destroy(trace.skel); 5097#endif 5098 return err; 5099} 5100