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