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