brintos

brintos / linux-shallow public Read only

0
0
Text · 157.3 KiB · d3f11b9 Raw
5496 lines · c
1/*2 * builtin-trace.c3 *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 other8 * 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_SUPPORT20#include <bpf/bpf.h>21#include <bpf/libbpf.h>22#include <bpf/btf.h>23#ifdef HAVE_BPF_SKEL24#include "bpf_skel/augmented_raw_syscalls.skel.h"25#endif26#endif27#include "util/bpf_map.h"28#include "util/rlimit.h"29#include "builtin.h"30#include "util/cgroup.h"31#include "util/color.h"32#include "util/config.h"33#include "util/debug.h"34#include "util/dso.h"35#include "util/env.h"36#include "util/event.h"37#include "util/evsel.h"38#include "util/evsel_fprintf.h"39#include "util/synthetic-events.h"40#include "util/evlist.h"41#include "util/evswitch.h"42#include "util/mmap.h"43#include <subcmd/pager.h>44#include <subcmd/exec-cmd.h>45#include "util/machine.h"46#include "util/map.h"47#include "util/symbol.h"48#include "util/path.h"49#include "util/session.h"50#include "util/thread.h"51#include <subcmd/parse-options.h>52#include "util/strlist.h"53#include "util/intlist.h"54#include "util/thread_map.h"55#include "util/stat.h"56#include "util/tool.h"57#include "util/util.h"58#include "trace/beauty/beauty.h"59#include "trace-event.h"60#include "util/parse-events.h"61#include "util/tracepoint.h"62#include "callchain.h"63#include "print_binary.h"64#include "string2.h"65#include "syscalltbl.h"66#include "rb_resort.h"67#include "../perf.h"68#include "trace_augment.h"69 70#include <errno.h>71#include <inttypes.h>72#include <poll.h>73#include <signal.h>74#include <stdlib.h>75#include <string.h>76#include <linux/err.h>77#include <linux/filter.h>78#include <linux/kernel.h>79#include <linux/list_sort.h>80#include <linux/random.h>81#include <linux/stringify.h>82#include <linux/time64.h>83#include <linux/zalloc.h>84#include <fcntl.h>85#include <sys/sysmacros.h>86 87#include <linux/ctype.h>88#include <perf/mmap.h>89 90#ifdef HAVE_LIBTRACEEVENT91#include <traceevent/event-parse.h>92#endif93 94#ifndef O_CLOEXEC95# define O_CLOEXEC		0200000096#endif97 98#ifndef F_LINUX_SPECIFIC_BASE99# define F_LINUX_SPECIFIC_BASE	1024100#endif101 102#define RAW_SYSCALL_ARGS_NUM	6103 104/*105 * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100106 *107 * We have to explicitely mark the direction of the flow of data, if from the108 * kernel to user space or the other way around, since the BPF collector we109 * have so far copies only from user to kernel space, mark the arguments that110 * go that direction, so that we don´t end up collecting the previous contents111 * for syscall args that goes from kernel to user space.112 */113struct syscall_arg_fmt {114	size_t	   (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);115	bool	   (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val);116	unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);117	void	   *parm;118	const char *name;119	u16	   nr_entries; // for arrays120	bool	   from_user;121	bool	   show_zero;122#ifdef HAVE_LIBBPF_SUPPORT123	const struct btf_type *type;124	int	   type_id; /* used in btf_dump */125#endif126};127 128struct syscall_fmt {129	const char *name;130	const char *alias;131	struct {132		const char *sys_enter,133			   *sys_exit;134	}	   bpf_prog_name;135	struct syscall_arg_fmt arg[RAW_SYSCALL_ARGS_NUM];136	u8	   nr_args;137	bool	   errpid;138	bool	   timeout;139	bool	   hexret;140};141 142struct trace {143	struct perf_tool	tool;144	struct syscalltbl	*sctbl;145	struct {146		struct syscall  *table;147		struct {148			struct evsel *sys_enter,149				*sys_exit,150				*bpf_output;151		}		events;152	} syscalls;153#ifdef HAVE_BPF_SKEL154	struct augmented_raw_syscalls_bpf *skel;155#endif156#ifdef HAVE_LIBBPF_SUPPORT157	struct btf		*btf;158#endif159	struct record_opts	opts;160	struct evlist	*evlist;161	struct machine		*host;162	struct thread		*current;163	struct cgroup		*cgroup;164	u64			base_time;165	FILE			*output;166	unsigned long		nr_events;167	unsigned long		nr_events_printed;168	unsigned long		max_events;169	struct evswitch		evswitch;170	struct strlist		*ev_qualifier;171	struct {172		size_t		nr;173		int		*entries;174	}			ev_qualifier_ids;175	struct {176		size_t		nr;177		pid_t		*entries;178		struct bpf_map  *map;179	}			filter_pids;180	double			duration_filter;181	double			runtime_ms;182	struct {183		u64		vfs_getname,184				proc_getname;185	} stats;186	unsigned int		max_stack;187	unsigned int		min_stack;188	int			raw_augmented_syscalls_args_size;189	bool			raw_augmented_syscalls;190	bool			fd_path_disabled;191	bool			sort_events;192	bool			not_ev_qualifier;193	bool			live;194	bool			full_time;195	bool			sched;196	bool			multiple_threads;197	bool			summary;198	bool			summary_only;199	bool			errno_summary;200	bool			failure_only;201	bool			show_comm;202	bool			print_sample;203	bool			show_tool_stats;204	bool			trace_syscalls;205	bool			libtraceevent_print;206	bool			kernel_syscallchains;207	s16			args_alignment;208	bool			show_tstamp;209	bool			show_duration;210	bool			show_zeros;211	bool			show_arg_names;212	bool			show_string_prefix;213	bool			force;214	bool			vfs_getname;215	bool			force_btf;216	int			trace_pgfaults;217	char			*perfconfig_events;218	struct {219		struct ordered_events	data;220		u64			last;221	} oe;222};223 224static void trace__load_vmlinux_btf(struct trace *trace __maybe_unused)225{226#ifdef HAVE_LIBBPF_SUPPORT227	if (trace->btf != NULL)228		return;229 230	trace->btf = btf__load_vmlinux_btf();231	if (verbose > 0) {232		fprintf(trace->output, trace->btf ? "vmlinux BTF loaded\n" :233						    "Failed to load vmlinux BTF\n");234	}235#endif236}237 238struct tp_field {239	int offset;240	union {241		u64 (*integer)(struct tp_field *field, struct perf_sample *sample);242		void *(*pointer)(struct tp_field *field, struct perf_sample *sample);243	};244};245 246#define TP_UINT_FIELD(bits) \247static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \248{ \249	u##bits value; \250	memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \251	return value;  \252}253 254TP_UINT_FIELD(8);255TP_UINT_FIELD(16);256TP_UINT_FIELD(32);257TP_UINT_FIELD(64);258 259#define TP_UINT_FIELD__SWAPPED(bits) \260static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \261{ \262	u##bits value; \263	memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \264	return bswap_##bits(value);\265}266 267TP_UINT_FIELD__SWAPPED(16);268TP_UINT_FIELD__SWAPPED(32);269TP_UINT_FIELD__SWAPPED(64);270 271static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)272{273	field->offset = offset;274 275	switch (size) {276	case 1:277		field->integer = tp_field__u8;278		break;279	case 2:280		field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;281		break;282	case 4:283		field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;284		break;285	case 8:286		field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;287		break;288	default:289		return -1;290	}291 292	return 0;293}294 295static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)296{297	return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);298}299 300static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)301{302	return sample->raw_data + field->offset;303}304 305static int __tp_field__init_ptr(struct tp_field *field, int offset)306{307	field->offset = offset;308	field->pointer = tp_field__ptr;309	return 0;310}311 312static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)313{314	return __tp_field__init_ptr(field, format_field->offset);315}316 317struct syscall_tp {318	struct tp_field id;319	union {320		struct tp_field args, ret;321	};322};323 324/*325 * The evsel->priv as used by 'perf trace'326 * sc:	for raw_syscalls:sys_{enter,exit} and syscalls:sys_{enter,exit}_SYSCALLNAME327 * fmt: for all the other tracepoints328 */329struct evsel_trace {330	struct syscall_tp	sc;331	struct syscall_arg_fmt  *fmt;332};333 334static struct evsel_trace *evsel_trace__new(void)335{336	return zalloc(sizeof(struct evsel_trace));337}338 339static void evsel_trace__delete(struct evsel_trace *et)340{341	if (et == NULL)342		return;343 344	zfree(&et->fmt);345	free(et);346}347 348/*349 * Used with raw_syscalls:sys_{enter,exit} and with the350 * syscalls:sys_{enter,exit}_SYSCALL tracepoints351 */352static inline struct syscall_tp *__evsel__syscall_tp(struct evsel *evsel)353{354	struct evsel_trace *et = evsel->priv;355 356	return &et->sc;357}358 359static struct syscall_tp *evsel__syscall_tp(struct evsel *evsel)360{361	if (evsel->priv == NULL) {362		evsel->priv = evsel_trace__new();363		if (evsel->priv == NULL)364			return NULL;365	}366 367	return __evsel__syscall_tp(evsel);368}369 370/*371 * Used with all the other tracepoints.372 */373static inline struct syscall_arg_fmt *__evsel__syscall_arg_fmt(struct evsel *evsel)374{375	struct evsel_trace *et = evsel->priv;376 377	return et->fmt;378}379 380static struct syscall_arg_fmt *evsel__syscall_arg_fmt(struct evsel *evsel)381{382	struct evsel_trace *et = evsel->priv;383 384	if (evsel->priv == NULL) {385		et = evsel->priv = evsel_trace__new();386 387		if (et == NULL)388			return NULL;389	}390 391	if (et->fmt == NULL) {392		et->fmt = calloc(evsel->tp_format->format.nr_fields, sizeof(struct syscall_arg_fmt));393		if (et->fmt == NULL)394			goto out_delete;395	}396 397	return __evsel__syscall_arg_fmt(evsel);398 399out_delete:400	evsel_trace__delete(evsel->priv);401	evsel->priv = NULL;402	return NULL;403}404 405static int evsel__init_tp_uint_field(struct evsel *evsel, struct tp_field *field, const char *name)406{407	struct tep_format_field *format_field = evsel__field(evsel, name);408 409	if (format_field == NULL)410		return -1;411 412	return tp_field__init_uint(field, format_field, evsel->needs_swap);413}414 415#define perf_evsel__init_sc_tp_uint_field(evsel, name) \416	({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\417	   evsel__init_tp_uint_field(evsel, &sc->name, #name); })418 419static int evsel__init_tp_ptr_field(struct evsel *evsel, struct tp_field *field, const char *name)420{421	struct tep_format_field *format_field = evsel__field(evsel, name);422 423	if (format_field == NULL)424		return -1;425 426	return tp_field__init_ptr(field, format_field);427}428 429#define perf_evsel__init_sc_tp_ptr_field(evsel, name) \430	({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\431	   evsel__init_tp_ptr_field(evsel, &sc->name, #name); })432 433static void evsel__delete_priv(struct evsel *evsel)434{435	zfree(&evsel->priv);436	evsel__delete(evsel);437}438 439static int evsel__init_syscall_tp(struct evsel *evsel)440{441	struct syscall_tp *sc = evsel__syscall_tp(evsel);442 443	if (sc != NULL) {444		if (evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&445		    evsel__init_tp_uint_field(evsel, &sc->id, "nr"))446			return -ENOENT;447 448		return 0;449	}450 451	return -ENOMEM;452}453 454static int evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)455{456	struct syscall_tp *sc = evsel__syscall_tp(evsel);457 458	if (sc != NULL) {459		struct tep_format_field *syscall_id = evsel__field(tp, "id");460		if (syscall_id == NULL)461			syscall_id = evsel__field(tp, "__syscall_nr");462		if (syscall_id == NULL ||463		    __tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))464			return -EINVAL;465 466		return 0;467	}468 469	return -ENOMEM;470}471 472static int evsel__init_augmented_syscall_tp_args(struct evsel *evsel)473{474	struct syscall_tp *sc = __evsel__syscall_tp(evsel);475 476	return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));477}478 479static int evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)480{481	struct syscall_tp *sc = __evsel__syscall_tp(evsel);482 483	return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);484}485 486static int evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)487{488	if (evsel__syscall_tp(evsel) != NULL) {489		if (perf_evsel__init_sc_tp_uint_field(evsel, id))490			return -ENOENT;491 492		evsel->handler = handler;493		return 0;494	}495 496	return -ENOMEM;497}498 499static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)500{501	struct evsel *evsel = evsel__newtp("raw_syscalls", direction);502 503	/* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */504	if (IS_ERR(evsel))505		evsel = evsel__newtp("syscalls", direction);506 507	if (IS_ERR(evsel))508		return NULL;509 510	if (evsel__init_raw_syscall_tp(evsel, handler))511		goto out_delete;512 513	return evsel;514 515out_delete:516	evsel__delete_priv(evsel);517	return NULL;518}519 520#define perf_evsel__sc_tp_uint(evsel, name, sample) \521	({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \522	   fields->name.integer(&fields->name, sample); })523 524#define perf_evsel__sc_tp_ptr(evsel, name, sample) \525	({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \526	   fields->name.pointer(&fields->name, sample); })527 528size_t strarray__scnprintf_suffix(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_suffix, int val)529{530	int idx = val - sa->offset;531 532	if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {533		size_t printed = scnprintf(bf, size, intfmt, val);534		if (show_suffix)535			printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);536		return printed;537	}538 539	return scnprintf(bf, size, "%s%s", sa->entries[idx], show_suffix ? sa->prefix : "");540}541 542size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)543{544	int idx = val - sa->offset;545 546	if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {547		size_t printed = scnprintf(bf, size, intfmt, val);548		if (show_prefix)549			printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);550		return printed;551	}552 553	return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);554}555 556static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,557						const char *intfmt,558					        struct syscall_arg *arg)559{560	return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);561}562 563static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,564					      struct syscall_arg *arg)565{566	return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);567}568 569#define SCA_STRARRAY syscall_arg__scnprintf_strarray570 571bool syscall_arg__strtoul_strarray(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)572{573	return strarray__strtoul(arg->parm, bf, size, ret);574}575 576bool syscall_arg__strtoul_strarray_flags(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)577{578	return strarray__strtoul_flags(arg->parm, bf, size, ret);579}580 581bool syscall_arg__strtoul_strarrays(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)582{583	return strarrays__strtoul(arg->parm, bf, size, ret);584}585 586size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)587{588	return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);589}590 591size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)592{593	size_t printed;594	int i;595 596	for (i = 0; i < sas->nr_entries; ++i) {597		struct strarray *sa = sas->entries[i];598		int idx = val - sa->offset;599 600		if (idx >= 0 && idx < sa->nr_entries) {601			if (sa->entries[idx] == NULL)602				break;603			return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);604		}605	}606 607	printed = scnprintf(bf, size, intfmt, val);608	if (show_prefix)609		printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);610	return printed;611}612 613bool strarray__strtoul(struct strarray *sa, char *bf, size_t size, u64 *ret)614{615	int i;616 617	for (i = 0; i < sa->nr_entries; ++i) {618		if (sa->entries[i] && strncmp(sa->entries[i], bf, size) == 0 && sa->entries[i][size] == '\0') {619			*ret = sa->offset + i;620			return true;621		}622	}623 624	return false;625}626 627bool strarray__strtoul_flags(struct strarray *sa, char *bf, size_t size, u64 *ret)628{629	u64 val = 0;630	char *tok = bf, *sep, *end;631 632	*ret = 0;633 634	while (size != 0) {635		int toklen = size;636 637		sep = memchr(tok, '|', size);638		if (sep != NULL) {639			size -= sep - tok + 1;640 641			end = sep - 1;642			while (end > tok && isspace(*end))643				--end;644 645			toklen = end - tok + 1;646		}647 648		while (isspace(*tok))649			++tok;650 651		if (isalpha(*tok) || *tok == '_') {652			if (!strarray__strtoul(sa, tok, toklen, &val))653				return false;654		} else655			val = strtoul(tok, NULL, 0);656 657		*ret |= (1 << (val - 1));658 659		if (sep == NULL)660			break;661		tok = sep + 1;662	}663 664	return true;665}666 667bool strarrays__strtoul(struct strarrays *sas, char *bf, size_t size, u64 *ret)668{669	int i;670 671	for (i = 0; i < sas->nr_entries; ++i) {672		struct strarray *sa = sas->entries[i];673 674		if (strarray__strtoul(sa, bf, size, ret))675			return true;676	}677 678	return false;679}680 681size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,682					struct syscall_arg *arg)683{684	return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);685}686 687#ifndef AT_FDCWD688#define AT_FDCWD	-100689#endif690 691static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,692					   struct syscall_arg *arg)693{694	int fd = arg->val;695	const char *prefix = "AT_FD";696 697	if (fd == AT_FDCWD)698		return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");699 700	return syscall_arg__scnprintf_fd(bf, size, arg);701}702 703#define SCA_FDAT syscall_arg__scnprintf_fd_at704 705static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,706					      struct syscall_arg *arg);707 708#define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd709 710size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)711{712	return scnprintf(bf, size, "%#lx", arg->val);713}714 715size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)716{717	if (arg->val == 0)718		return scnprintf(bf, size, "NULL");719	return syscall_arg__scnprintf_hex(bf, size, arg);720}721 722size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)723{724	return scnprintf(bf, size, "%d", arg->val);725}726 727size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)728{729	return scnprintf(bf, size, "%ld", arg->val);730}731 732static size_t syscall_arg__scnprintf_char_array(char *bf, size_t size, struct syscall_arg *arg)733{734	// XXX Hey, maybe for sched:sched_switch prev/next comm fields we can735	//     fill missing comms using thread__set_comm()...736	//     here or in a special syscall_arg__scnprintf_pid_sched_tp...737	return scnprintf(bf, size, "\"%-.*s\"", arg->fmt->nr_entries ?: arg->len, arg->val);738}739 740#define SCA_CHAR_ARRAY syscall_arg__scnprintf_char_array741 742static const char *bpf_cmd[] = {743	"MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",744	"MAP_GET_NEXT_KEY", "PROG_LOAD", "OBJ_PIN", "OBJ_GET", "PROG_ATTACH",745	"PROG_DETACH", "PROG_TEST_RUN", "PROG_GET_NEXT_ID", "MAP_GET_NEXT_ID",746	"PROG_GET_FD_BY_ID", "MAP_GET_FD_BY_ID", "OBJ_GET_INFO_BY_FD",747	"PROG_QUERY", "RAW_TRACEPOINT_OPEN", "BTF_LOAD", "BTF_GET_FD_BY_ID",748	"TASK_FD_QUERY", "MAP_LOOKUP_AND_DELETE_ELEM", "MAP_FREEZE",749	"BTF_GET_NEXT_ID", "MAP_LOOKUP_BATCH", "MAP_LOOKUP_AND_DELETE_BATCH",750	"MAP_UPDATE_BATCH", "MAP_DELETE_BATCH", "LINK_CREATE", "LINK_UPDATE",751	"LINK_GET_FD_BY_ID", "LINK_GET_NEXT_ID", "ENABLE_STATS", "ITER_CREATE",752	"LINK_DETACH", "PROG_BIND_MAP",753};754static DEFINE_STRARRAY(bpf_cmd, "BPF_");755 756static const char *fsmount_flags[] = {757	[1] = "CLOEXEC",758};759static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");760 761#include "trace/beauty/generated/fsconfig_arrays.c"762 763static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");764 765static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };766static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);767 768static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };769static DEFINE_STRARRAY(itimers, "ITIMER_");770 771static const char *keyctl_options[] = {772	"GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",773	"SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",774	"INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",775	"ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",776	"INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",777};778static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");779 780static const char *whences[] = { "SET", "CUR", "END",781#ifdef SEEK_DATA782"DATA",783#endif784#ifdef SEEK_HOLE785"HOLE",786#endif787};788static DEFINE_STRARRAY(whences, "SEEK_");789 790static const char *fcntl_cmds[] = {791	"DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",792	"SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",793	"SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",794	"GETOWNER_UIDS",795};796static DEFINE_STRARRAY(fcntl_cmds, "F_");797 798static const char *fcntl_linux_specific_cmds[] = {799	"SETLEASE", "GETLEASE", "NOTIFY", "DUPFD_QUERY", [5] = "CANCELLK", "DUPFD_CLOEXEC",800	"SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",801	"GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",802};803 804static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);805 806static struct strarray *fcntl_cmds_arrays[] = {807	&strarray__fcntl_cmds,808	&strarray__fcntl_linux_specific_cmds,809};810 811static DEFINE_STRARRAYS(fcntl_cmds_arrays);812 813static const char *rlimit_resources[] = {814	"CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",815	"MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",816	"RTTIME",817};818static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");819 820static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };821static DEFINE_STRARRAY(sighow, "SIG_");822 823static const char *clockid[] = {824	"REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",825	"MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",826	"REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"827};828static DEFINE_STRARRAY(clockid, "CLOCK_");829 830static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,831						 struct syscall_arg *arg)832{833	bool show_prefix = arg->show_string_prefix;834	const char *suffix = "_OK";835	size_t printed = 0;836	int mode = arg->val;837 838	if (mode == F_OK) /* 0 */839		return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");840#define	P_MODE(n) \841	if (mode & n##_OK) { \842		printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \843		mode &= ~n##_OK; \844	}845 846	P_MODE(R);847	P_MODE(W);848	P_MODE(X);849#undef P_MODE850 851	if (mode)852		printed += scnprintf(bf + printed, size - printed, "|%#x", mode);853 854	return printed;855}856 857#define SCA_ACCMODE syscall_arg__scnprintf_access_mode858 859static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,860					      struct syscall_arg *arg);861 862#define SCA_FILENAME syscall_arg__scnprintf_filename863 864// 'argname' is just documentational at this point, to remove the previous comment with that info865#define SCA_FILENAME_FROM_USER(argname) \866	  { .scnprintf	= SCA_FILENAME, \867	    .from_user	= true, }868 869static size_t syscall_arg__scnprintf_buf(char *bf, size_t size, struct syscall_arg *arg);870 871#define SCA_BUF syscall_arg__scnprintf_buf872 873static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,874						struct syscall_arg *arg)875{876	bool show_prefix = arg->show_string_prefix;877	const char *prefix = "O_";878	int printed = 0, flags = arg->val;879 880#define	P_FLAG(n) \881	if (flags & O_##n) { \882		printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \883		flags &= ~O_##n; \884	}885 886	P_FLAG(CLOEXEC);887	P_FLAG(NONBLOCK);888#undef P_FLAG889 890	if (flags)891		printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);892 893	return printed;894}895 896#define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags897 898#ifndef GRND_NONBLOCK899#define GRND_NONBLOCK	0x0001900#endif901#ifndef GRND_RANDOM902#define GRND_RANDOM	0x0002903#endif904 905static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,906						   struct syscall_arg *arg)907{908	bool show_prefix = arg->show_string_prefix;909	const char *prefix = "GRND_";910	int printed = 0, flags = arg->val;911 912#define	P_FLAG(n) \913	if (flags & GRND_##n) { \914		printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \915		flags &= ~GRND_##n; \916	}917 918	P_FLAG(RANDOM);919	P_FLAG(NONBLOCK);920#undef P_FLAG921 922	if (flags)923		printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);924 925	return printed;926}927 928#define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags929 930#ifdef HAVE_LIBBPF_SUPPORT931static void syscall_arg_fmt__cache_btf_enum(struct syscall_arg_fmt *arg_fmt, struct btf *btf, char *type)932{933	int id;934 935	type = strstr(type, "enum ");936	if (type == NULL)937		return;938 939	type += 5; // skip "enum " to get the enumeration name940 941	id = btf__find_by_name(btf, type);942	if (id < 0)943		return;944 945	arg_fmt->type = btf__type_by_id(btf, id);946}947 948static bool syscall_arg__strtoul_btf_enum(char *bf, size_t size, struct syscall_arg *arg, u64 *val)949{950	const struct btf_type *bt = arg->fmt->type;951	struct btf *btf = arg->trace->btf;952	struct btf_enum *be = btf_enum(bt);953 954	for (int i = 0; i < btf_vlen(bt); ++i, ++be) {955		const char *name = btf__name_by_offset(btf, be->name_off);956		int max_len = max(size, strlen(name));957 958		if (strncmp(name, bf, max_len) == 0) {959			*val = be->val;960			return true;961		}962	}963 964	return false;965}966 967static bool syscall_arg__strtoul_btf_type(char *bf, size_t size, struct syscall_arg *arg, u64 *val)968{969	const struct btf_type *bt;970	char *type = arg->type_name;971	struct btf *btf;972 973	trace__load_vmlinux_btf(arg->trace);974 975	btf = arg->trace->btf;976	if (btf == NULL)977		return false;978 979	if (arg->fmt->type == NULL) {980		// See if this is an enum981		syscall_arg_fmt__cache_btf_enum(arg->fmt, btf, type);982	}983 984	// Now let's see if we have a BTF type resolved985	bt = arg->fmt->type;986	if (bt == NULL)987		return false;988 989	// If it is an enum:990	if (btf_is_enum(arg->fmt->type))991		return syscall_arg__strtoul_btf_enum(bf, size, arg, val);992 993	return false;994}995 996static size_t btf_enum_scnprintf(const struct btf_type *type, struct btf *btf, char *bf, size_t size, int val)997{998	struct btf_enum *be = btf_enum(type);999	const int nr_entries = btf_vlen(type);1000 1001	for (int i = 0; i < nr_entries; ++i, ++be) {1002		if (be->val == val) {1003			return scnprintf(bf, size, "%s",1004					 btf__name_by_offset(btf, be->name_off));1005		}1006	}1007 1008	return 0;1009}1010 1011struct trace_btf_dump_snprintf_ctx {1012	char   *bf;1013	size_t printed, size;1014};1015 1016static void trace__btf_dump_snprintf(void *vctx, const char *fmt, va_list args)1017{1018	struct trace_btf_dump_snprintf_ctx *ctx = vctx;1019 1020	ctx->printed += vscnprintf(ctx->bf + ctx->printed, ctx->size - ctx->printed, fmt, args);1021}1022 1023static size_t btf_struct_scnprintf(const struct btf_type *type, struct btf *btf, char *bf, size_t size, struct syscall_arg *arg)1024{1025	struct trace_btf_dump_snprintf_ctx ctx = {1026		.bf   = bf,1027		.size = size,1028	};1029	struct augmented_arg *augmented_arg = arg->augmented.args;1030	int type_id = arg->fmt->type_id, consumed;1031	struct btf_dump *btf_dump;1032 1033	LIBBPF_OPTS(btf_dump_opts, dump_opts);1034	LIBBPF_OPTS(btf_dump_type_data_opts, dump_data_opts);1035 1036	if (arg == NULL || arg->augmented.args == NULL)1037		return 0;1038 1039	dump_data_opts.compact	  = true;1040	dump_data_opts.skip_names = !arg->trace->show_arg_names;1041 1042	btf_dump = btf_dump__new(btf, trace__btf_dump_snprintf, &ctx, &dump_opts);1043	if (btf_dump == NULL)1044		return 0;1045 1046	/* pretty print the struct data here */1047	if (btf_dump__dump_type_data(btf_dump, type_id, arg->augmented.args->value, type->size, &dump_data_opts) == 0)1048		return 0;1049 1050	consumed = sizeof(*augmented_arg) + augmented_arg->size;1051	arg->augmented.args = ((void *)arg->augmented.args) + consumed;1052	arg->augmented.size -= consumed;1053 1054	btf_dump__free(btf_dump);1055 1056	return ctx.printed;1057}1058 1059static size_t trace__btf_scnprintf(struct trace *trace, struct syscall_arg *arg, char *bf,1060				   size_t size, int val, char *type)1061{1062	struct syscall_arg_fmt *arg_fmt = arg->fmt;1063 1064	if (trace->btf == NULL)1065		return 0;1066 1067	if (arg_fmt->type == NULL) {1068		// Check if this is an enum and if we have the BTF type for it.1069		syscall_arg_fmt__cache_btf_enum(arg_fmt, trace->btf, type);1070	}1071 1072	// Did we manage to find a BTF type for the syscall/tracepoint argument?1073	if (arg_fmt->type == NULL)1074		return 0;1075 1076	if (btf_is_enum(arg_fmt->type))1077		return btf_enum_scnprintf(arg_fmt->type, trace->btf, bf, size, val);1078	else if (btf_is_struct(arg_fmt->type) || btf_is_union(arg_fmt->type))1079		return btf_struct_scnprintf(arg_fmt->type, trace->btf, bf, size, arg);1080 1081	return 0;1082}1083 1084#else // HAVE_LIBBPF_SUPPORT1085static size_t trace__btf_scnprintf(struct trace *trace __maybe_unused, struct syscall_arg *arg __maybe_unused,1086				   char *bf __maybe_unused, size_t size __maybe_unused, int val __maybe_unused,1087				   char *type __maybe_unused)1088{1089	return 0;1090}1091 1092static bool syscall_arg__strtoul_btf_type(char *bf __maybe_unused, size_t size __maybe_unused,1093					  struct syscall_arg *arg __maybe_unused, u64 *val __maybe_unused)1094{1095	return false;1096}1097#endif // HAVE_LIBBPF_SUPPORT1098 1099#define STUL_BTF_TYPE syscall_arg__strtoul_btf_type1100 1101#define STRARRAY(name, array) \1102	  { .scnprintf	= SCA_STRARRAY, \1103	    .strtoul	= STUL_STRARRAY, \1104	    .parm	= &strarray__##array, }1105 1106#define STRARRAY_FLAGS(name, array) \1107	  { .scnprintf	= SCA_STRARRAY_FLAGS, \1108	    .strtoul	= STUL_STRARRAY_FLAGS, \1109	    .parm	= &strarray__##array, }1110 1111#include "trace/beauty/arch_errno_names.c"1112#include "trace/beauty/eventfd.c"1113#include "trace/beauty/futex_op.c"1114#include "trace/beauty/futex_val3.c"1115#include "trace/beauty/mmap.c"1116#include "trace/beauty/mode_t.c"1117#include "trace/beauty/msg_flags.c"1118#include "trace/beauty/open_flags.c"1119#include "trace/beauty/perf_event_open.c"1120#include "trace/beauty/pid.c"1121#include "trace/beauty/sched_policy.c"1122#include "trace/beauty/seccomp.c"1123#include "trace/beauty/signum.c"1124#include "trace/beauty/socket_type.c"1125#include "trace/beauty/waitid_options.c"1126 1127static const struct syscall_fmt syscall_fmts[] = {1128	{ .name	    = "access",1129	  .arg = { [1] = { .scnprintf = SCA_ACCMODE,  /* mode */ }, }, },1130	{ .name	    = "arch_prctl",1131	  .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },1132		   [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },1133	{ .name	    = "bind",1134	  .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },1135		   [1] = SCA_SOCKADDR_FROM_USER(umyaddr),1136		   [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },1137	{ .name	    = "bpf",1138	  .arg = { [0] = STRARRAY(cmd, bpf_cmd),1139		   [1] = { .from_user = true /* attr */, }, } },1140	{ .name	    = "brk",	    .hexret = true,1141	  .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },1142	{ .name     = "clock_gettime",1143	  .arg = { [0] = STRARRAY(clk_id, clockid), }, },1144	{ .name	    = "clock_nanosleep",1145	  .arg = { [2] = SCA_TIMESPEC_FROM_USER(req), }, },1146	{ .name	    = "clone",	    .errpid = true, .nr_args = 5,1147	  .arg = { [0] = { .name = "flags",	    .scnprintf = SCA_CLONE_FLAGS, },1148		   [1] = { .name = "child_stack",   .scnprintf = SCA_HEX, },1149		   [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },1150		   [3] = { .name = "child_tidptr",  .scnprintf = SCA_HEX, },1151		   [4] = { .name = "tls",	    .scnprintf = SCA_HEX, }, }, },1152	{ .name	    = "close",1153	  .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },1154	{ .name	    = "connect",1155	  .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },1156		   [1] = SCA_SOCKADDR_FROM_USER(servaddr),1157		   [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },1158	{ .name	    = "epoll_ctl",1159	  .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },1160	{ .name	    = "eventfd2",1161	  .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },1162	{ .name     = "faccessat",1163	  .arg = { [0] = { .scnprintf = SCA_FDAT,	  /* dirfd */ },1164		   [1] = SCA_FILENAME_FROM_USER(pathname),1165		   [2] = { .scnprintf = SCA_ACCMODE,	  /* mode */ }, }, },1166	{ .name     = "faccessat2",1167	  .arg = { [0] = { .scnprintf = SCA_FDAT,	  /* dirfd */ },1168		   [1] = SCA_FILENAME_FROM_USER(pathname),1169		   [2] = { .scnprintf = SCA_ACCMODE,	  /* mode */ },1170		   [3] = { .scnprintf = SCA_FACCESSAT2_FLAGS, /* flags */ }, }, },1171	{ .name	    = "fchmodat",1172	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },1173	{ .name	    = "fchownat",1174	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },1175	{ .name	    = "fcntl",1176	  .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD,  /* cmd */1177			   .strtoul   = STUL_STRARRAYS,1178			   .parm      = &strarrays__fcntl_cmds_arrays,1179			   .show_zero = true, },1180		   [2] = { .scnprintf =  SCA_FCNTL_ARG, /* arg */ }, }, },1181	{ .name	    = "flock",1182	  .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },1183	{ .name     = "fsconfig",1184	  .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },1185	{ .name     = "fsmount",1186	  .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),1187		   [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },1188	{ .name     = "fspick",1189	  .arg = { [0] = { .scnprintf = SCA_FDAT,	  /* dfd */ },1190		   [1] = SCA_FILENAME_FROM_USER(path),1191		   [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },1192	{ .name	    = "fstat", .alias = "newfstat", },1193	{ .name	    = "futex",1194	  .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },1195		   [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },1196	{ .name	    = "futimesat",1197	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },1198	{ .name	    = "getitimer",1199	  .arg = { [0] = STRARRAY(which, itimers), }, },1200	{ .name	    = "getpid",	    .errpid = true, },1201	{ .name	    = "getpgid",    .errpid = true, },1202	{ .name	    = "getppid",    .errpid = true, },1203	{ .name	    = "getrandom",1204	  .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },1205	{ .name	    = "getrlimit",1206	  .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },1207	{ .name	    = "getsockopt",1208	  .arg = { [1] = STRARRAY(level, socket_level), }, },1209	{ .name	    = "gettid",	    .errpid = true, },1210	{ .name	    = "ioctl",1211	  .arg = {1212#if defined(__i386__) || defined(__x86_64__)1213/*1214 * FIXME: Make this available to all arches.1215 */1216		   [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },1217		   [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },1218#else1219		   [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },1220#endif1221	{ .name	    = "kcmp",	    .nr_args = 5,1222	  .arg = { [0] = { .name = "pid1",	.scnprintf = SCA_PID, },1223		   [1] = { .name = "pid2",	.scnprintf = SCA_PID, },1224		   [2] = { .name = "type",	.scnprintf = SCA_KCMP_TYPE, },1225		   [3] = { .name = "idx1",	.scnprintf = SCA_KCMP_IDX, },1226		   [4] = { .name = "idx2",	.scnprintf = SCA_KCMP_IDX, }, }, },1227	{ .name	    = "keyctl",1228	  .arg = { [0] = STRARRAY(option, keyctl_options), }, },1229	{ .name	    = "kill",1230	  .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },1231	{ .name	    = "linkat",1232	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },1233	{ .name	    = "lseek",1234	  .arg = { [2] = STRARRAY(whence, whences), }, },1235	{ .name	    = "lstat", .alias = "newlstat", },1236	{ .name     = "madvise",1237	  .arg = { [0] = { .scnprintf = SCA_HEX,      /* start */ },1238		   [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },1239	{ .name	    = "mkdirat",1240	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },1241	{ .name	    = "mknodat",1242	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },1243	{ .name	    = "mmap",	    .hexret = true,1244/* The standard mmap maps to old_mmap on s390x */1245#if defined(__s390x__)1246	.alias = "old_mmap",1247#endif1248	  .arg = { [2] = { .scnprintf = SCA_MMAP_PROT, .show_zero = true, /* prot */ },1249		   [3] = { .scnprintf = SCA_MMAP_FLAGS,	/* flags */1250			   .strtoul   = STUL_STRARRAY_FLAGS,1251			   .parm      = &strarray__mmap_flags, },1252		   [5] = { .scnprintf = SCA_HEX,	/* offset */ }, }, },1253	{ .name	    = "mount",1254	  .arg = { [0] = SCA_FILENAME_FROM_USER(devname),1255		   [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */1256			   .mask_val  = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },1257	{ .name	    = "move_mount",1258	  .arg = { [0] = { .scnprintf = SCA_FDAT,	/* from_dfd */ },1259		   [1] = SCA_FILENAME_FROM_USER(pathname),1260		   [2] = { .scnprintf = SCA_FDAT,	/* to_dfd */ },1261		   [3] = SCA_FILENAME_FROM_USER(pathname),1262		   [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },1263	{ .name	    = "mprotect",1264	  .arg = { [0] = { .scnprintf = SCA_HEX,	/* start */ },1265		   [2] = { .scnprintf = SCA_MMAP_PROT, .show_zero = true, /* prot */ }, }, },1266	{ .name	    = "mq_unlink",1267	  .arg = { [0] = SCA_FILENAME_FROM_USER(u_name), }, },1268	{ .name	    = "mremap",	    .hexret = true,1269	  .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },1270	{ .name	    = "name_to_handle_at",1271	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },1272	{ .name	    = "nanosleep",1273	  .arg = { [0] = SCA_TIMESPEC_FROM_USER(req), }, },1274	{ .name	    = "newfstatat", .alias = "fstatat",1275	  .arg = { [0] = { .scnprintf = SCA_FDAT,	  /* dirfd */ },1276		   [1] = SCA_FILENAME_FROM_USER(pathname),1277		   [3] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ }, }, },1278	{ .name	    = "open",1279	  .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },1280	{ .name	    = "open_by_handle_at",1281	  .arg = { [0] = { .scnprintf = SCA_FDAT,	/* dfd */ },1282		   [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },1283	{ .name	    = "openat",1284	  .arg = { [0] = { .scnprintf = SCA_FDAT,	/* dfd */ },1285		   [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },1286	{ .name	    = "perf_event_open",1287	  .arg = { [0] = SCA_PERF_ATTR_FROM_USER(attr),1288		   [2] = { .scnprintf = SCA_INT,	/* cpu */ },1289		   [3] = { .scnprintf = SCA_FD,		/* group_fd */ },1290		   [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },1291	{ .name	    = "pipe2",1292	  .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },1293	{ .name	    = "pkey_alloc",1294	  .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS,	/* access_rights */ }, }, },1295	{ .name	    = "pkey_free",1296	  .arg = { [0] = { .scnprintf = SCA_INT,	/* key */ }, }, },1297	{ .name	    = "pkey_mprotect",1298	  .arg = { [0] = { .scnprintf = SCA_HEX,	/* start */ },1299		   [2] = { .scnprintf = SCA_MMAP_PROT, .show_zero = true, /* prot */ },1300		   [3] = { .scnprintf = SCA_INT,	/* pkey */ }, }, },1301	{ .name	    = "poll", .timeout = true, },1302	{ .name	    = "ppoll", .timeout = true, },1303	{ .name	    = "prctl",1304	  .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */1305			   .strtoul   = STUL_STRARRAY,1306			   .parm      = &strarray__prctl_options, },1307		   [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },1308		   [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },1309	{ .name	    = "pread", .alias = "pread64", },1310	{ .name	    = "preadv", .alias = "pread", },1311	{ .name	    = "prlimit64",1312	  .arg = { [1] = STRARRAY(resource, rlimit_resources),1313		   [2] = { .from_user = true /* new_rlim */, }, }, },1314	{ .name	    = "pwrite", .alias = "pwrite64", },1315	{ .name	    = "readlinkat",1316	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },1317	{ .name	    = "recvfrom",1318	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },1319	{ .name	    = "recvmmsg",1320	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },1321	{ .name	    = "recvmsg",1322	  .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },1323	{ .name	    = "renameat",1324	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },1325		   [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },1326	{ .name	    = "renameat2",1327	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },1328		   [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },1329		   [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },1330	{ .name	    = "rseq",	    .errpid = true,1331	  .arg = { [0] = { .from_user = true /* rseq */, }, }, },1332	{ .name	    = "rt_sigaction",1333	  .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },1334	{ .name	    = "rt_sigprocmask",1335	  .arg = { [0] = STRARRAY(how, sighow), }, },1336	{ .name	    = "rt_sigqueueinfo",1337	  .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },1338	{ .name	    = "rt_tgsigqueueinfo",1339	  .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },1340	{ .name	    = "sched_setscheduler",1341	  .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },1342	{ .name	    = "seccomp",1343	  .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP,	   /* op */ },1344		   [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },1345	{ .name	    = "select", .timeout = true, },1346	{ .name	    = "sendfile", .alias = "sendfile64", },1347	{ .name	    = "sendmmsg",1348	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },1349	{ .name	    = "sendmsg",1350	  .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },1351	{ .name	    = "sendto",1352	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },1353		   [4] = SCA_SOCKADDR_FROM_USER(addr), }, },1354	{ .name	    = "set_robust_list",	    .errpid = true,1355	  .arg = { [0] = { .from_user = true /* head */, }, }, },1356	{ .name	    = "set_tid_address", .errpid = true, },1357	{ .name	    = "setitimer",1358	  .arg = { [0] = STRARRAY(which, itimers), }, },1359	{ .name	    = "setrlimit",1360	  .arg = { [0] = STRARRAY(resource, rlimit_resources),1361		   [1] = { .from_user = true /* rlim */, }, }, },1362	{ .name	    = "setsockopt",1363	  .arg = { [1] = STRARRAY(level, socket_level), }, },1364	{ .name	    = "socket",1365	  .arg = { [0] = STRARRAY(family, socket_families),1366		   [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },1367		   [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },1368	{ .name	    = "socketpair",1369	  .arg = { [0] = STRARRAY(family, socket_families),1370		   [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },1371		   [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },1372	{ .name	    = "stat", .alias = "newstat", },1373	{ .name	    = "statx",1374	  .arg = { [0] = { .scnprintf = SCA_FDAT,	 /* fdat */ },1375		   [2] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ } ,1376		   [3] = { .scnprintf = SCA_STATX_MASK,	 /* mask */ }, }, },1377	{ .name	    = "swapoff",1378	  .arg = { [0] = SCA_FILENAME_FROM_USER(specialfile), }, },1379	{ .name	    = "swapon",1380	  .arg = { [0] = SCA_FILENAME_FROM_USER(specialfile), }, },1381	{ .name	    = "symlinkat",1382	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },1383	{ .name	    = "sync_file_range",1384	  .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },1385	{ .name	    = "tgkill",1386	  .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },1387	{ .name	    = "tkill",1388	  .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },1389	{ .name     = "umount2", .alias = "umount",1390	  .arg = { [0] = SCA_FILENAME_FROM_USER(name), }, },1391	{ .name	    = "uname", .alias = "newuname", },1392	{ .name	    = "unlinkat",1393	  .arg = { [0] = { .scnprintf = SCA_FDAT,	  /* dfd */ },1394		   [1] = SCA_FILENAME_FROM_USER(pathname),1395		   [2] = { .scnprintf = SCA_FS_AT_FLAGS,  /* flags */ }, }, },1396	{ .name	    = "utimensat",1397	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },1398	{ .name	    = "wait4",	    .errpid = true,1399	  .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },1400	{ .name	    = "waitid",	    .errpid = true,1401	  .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },1402	{ .name	    = "write",1403	  .arg = { [1] = { .scnprintf = SCA_BUF /* buf */, .from_user = true, }, }, },1404};1405 1406static int syscall_fmt__cmp(const void *name, const void *fmtp)1407{1408	const struct syscall_fmt *fmt = fmtp;1409	return strcmp(name, fmt->name);1410}1411 1412static const struct syscall_fmt *__syscall_fmt__find(const struct syscall_fmt *fmts,1413						     const int nmemb,1414						     const char *name)1415{1416	return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);1417}1418 1419static const struct syscall_fmt *syscall_fmt__find(const char *name)1420{1421	const int nmemb = ARRAY_SIZE(syscall_fmts);1422	return __syscall_fmt__find(syscall_fmts, nmemb, name);1423}1424 1425static const struct syscall_fmt *__syscall_fmt__find_by_alias(const struct syscall_fmt *fmts,1426							      const int nmemb, const char *alias)1427{1428	int i;1429 1430	for (i = 0; i < nmemb; ++i) {1431		if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0)1432			return &fmts[i];1433	}1434 1435	return NULL;1436}1437 1438static const struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)1439{1440	const int nmemb = ARRAY_SIZE(syscall_fmts);1441	return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias);1442}1443 1444/*1445 * is_exit: is this "exit" or "exit_group"?1446 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.1447 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.1448 * nonexistent: Just a hole in the syscall table, syscall id not allocated1449 */1450struct syscall {1451	struct tep_event    *tp_format;1452	int		    nr_args;1453	int		    args_size;1454	struct {1455		struct bpf_program *sys_enter,1456				   *sys_exit;1457	}		    bpf_prog;1458	bool		    is_exit;1459	bool		    is_open;1460	bool		    nonexistent;1461	bool		    use_btf;1462	struct tep_format_field *args;1463	const char	    *name;1464	const struct syscall_fmt  *fmt;1465	struct syscall_arg_fmt *arg_fmt;1466};1467 1468/*1469 * We need to have this 'calculated' boolean because in some cases we really1470 * don't know what is the duration of a syscall, for instance, when we start1471 * a session and some threads are waiting for a syscall to finish, say 'poll',1472 * in which case all we can do is to print "( ? ) for duration and for the1473 * start timestamp.1474 */1475static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)1476{1477	double duration = (double)t / NSEC_PER_MSEC;1478	size_t printed = fprintf(fp, "(");1479 1480	if (!calculated)1481		printed += fprintf(fp, "         ");1482	else if (duration >= 1.0)1483		printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);1484	else if (duration >= 0.01)1485		printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);1486	else1487		printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);1488	return printed + fprintf(fp, "): ");1489}1490 1491/**1492 * filename.ptr: The filename char pointer that will be vfs_getname'd1493 * filename.entry_str_pos: Where to insert the string translated from1494 *                         filename.ptr by the vfs_getname tracepoint/kprobe.1495 * ret_scnprintf: syscall args may set this to a different syscall return1496 *                formatter, for instance, fcntl may return fds, file flags, etc.1497 */1498struct thread_trace {1499	u64		  entry_time;1500	bool		  entry_pending;1501	unsigned long	  nr_events;1502	unsigned long	  pfmaj, pfmin;1503	char		  *entry_str;1504	double		  runtime_ms;1505	size_t		  (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);1506        struct {1507		unsigned long ptr;1508		short int     entry_str_pos;1509		bool	      pending_open;1510		unsigned int  namelen;1511		char	      *name;1512	} filename;1513	struct {1514		int	      max;1515		struct file   *table;1516	} files;1517 1518	struct intlist *syscall_stats;1519};1520 1521static struct thread_trace *thread_trace__new(void)1522{1523	struct thread_trace *ttrace =  zalloc(sizeof(struct thread_trace));1524 1525	if (ttrace) {1526		ttrace->files.max = -1;1527		ttrace->syscall_stats = intlist__new(NULL);1528	}1529 1530	return ttrace;1531}1532 1533static void thread_trace__free_files(struct thread_trace *ttrace);1534 1535static void thread_trace__delete(void *pttrace)1536{1537	struct thread_trace *ttrace = pttrace;1538 1539	if (!ttrace)1540		return;1541 1542	intlist__delete(ttrace->syscall_stats);1543	ttrace->syscall_stats = NULL;1544	thread_trace__free_files(ttrace);1545	zfree(&ttrace->entry_str);1546	free(ttrace);1547}1548 1549static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)1550{1551	struct thread_trace *ttrace;1552 1553	if (thread == NULL)1554		goto fail;1555 1556	if (thread__priv(thread) == NULL)1557		thread__set_priv(thread, thread_trace__new());1558 1559	if (thread__priv(thread) == NULL)1560		goto fail;1561 1562	ttrace = thread__priv(thread);1563	++ttrace->nr_events;1564 1565	return ttrace;1566fail:1567	color_fprintf(fp, PERF_COLOR_RED,1568		      "WARNING: not enough memory, dropping samples!\n");1569	return NULL;1570}1571 1572 1573void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,1574				    size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))1575{1576	struct thread_trace *ttrace = thread__priv(arg->thread);1577 1578	ttrace->ret_scnprintf = ret_scnprintf;1579}1580 1581#define TRACE_PFMAJ		(1 << 0)1582#define TRACE_PFMIN		(1 << 1)1583 1584static const size_t trace__entry_str_size = 2048;1585 1586static void thread_trace__free_files(struct thread_trace *ttrace)1587{1588	for (int i = 0; i < ttrace->files.max; ++i) {1589		struct file *file = ttrace->files.table + i;1590		zfree(&file->pathname);1591	}1592 1593	zfree(&ttrace->files.table);1594	ttrace->files.max  = -1;1595}1596 1597static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)1598{1599	if (fd < 0)1600		return NULL;1601 1602	if (fd > ttrace->files.max) {1603		struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));1604 1605		if (nfiles == NULL)1606			return NULL;1607 1608		if (ttrace->files.max != -1) {1609			memset(nfiles + ttrace->files.max + 1, 0,1610			       (fd - ttrace->files.max) * sizeof(struct file));1611		} else {1612			memset(nfiles, 0, (fd + 1) * sizeof(struct file));1613		}1614 1615		ttrace->files.table = nfiles;1616		ttrace->files.max   = fd;1617	}1618 1619	return ttrace->files.table + fd;1620}1621 1622struct file *thread__files_entry(struct thread *thread, int fd)1623{1624	return thread_trace__files_entry(thread__priv(thread), fd);1625}1626 1627static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)1628{1629	struct thread_trace *ttrace = thread__priv(thread);1630	struct file *file = thread_trace__files_entry(ttrace, fd);1631 1632	if (file != NULL) {1633		struct stat st;1634		if (stat(pathname, &st) == 0)1635			file->dev_maj = major(st.st_rdev);1636		file->pathname = strdup(pathname);1637		if (file->pathname)1638			return 0;1639	}1640 1641	return -1;1642}1643 1644static int thread__read_fd_path(struct thread *thread, int fd)1645{1646	char linkname[PATH_MAX], pathname[PATH_MAX];1647	struct stat st;1648	int ret;1649 1650	if (thread__pid(thread) == thread__tid(thread)) {1651		scnprintf(linkname, sizeof(linkname),1652			  "/proc/%d/fd/%d", thread__pid(thread), fd);1653	} else {1654		scnprintf(linkname, sizeof(linkname),1655			  "/proc/%d/task/%d/fd/%d",1656			  thread__pid(thread), thread__tid(thread), fd);1657	}1658 1659	if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))1660		return -1;1661 1662	ret = readlink(linkname, pathname, sizeof(pathname));1663 1664	if (ret < 0 || ret > st.st_size)1665		return -1;1666 1667	pathname[ret] = '\0';1668	return trace__set_fd_pathname(thread, fd, pathname);1669}1670 1671static const char *thread__fd_path(struct thread *thread, int fd,1672				   struct trace *trace)1673{1674	struct thread_trace *ttrace = thread__priv(thread);1675 1676	if (ttrace == NULL || trace->fd_path_disabled)1677		return NULL;1678 1679	if (fd < 0)1680		return NULL;1681 1682	if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {1683		if (!trace->live)1684			return NULL;1685		++trace->stats.proc_getname;1686		if (thread__read_fd_path(thread, fd))1687			return NULL;1688	}1689 1690	return ttrace->files.table[fd].pathname;1691}1692 1693size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)1694{1695	int fd = arg->val;1696	size_t printed = scnprintf(bf, size, "%d", fd);1697	const char *path = thread__fd_path(arg->thread, fd, arg->trace);1698 1699	if (path)1700		printed += scnprintf(bf + printed, size - printed, "<%s>", path);1701 1702	return printed;1703}1704 1705size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)1706{1707        size_t printed = scnprintf(bf, size, "%d", fd);1708	struct thread *thread = machine__find_thread(trace->host, pid, pid);1709 1710	if (thread) {1711		const char *path = thread__fd_path(thread, fd, trace);1712 1713		if (path)1714			printed += scnprintf(bf + printed, size - printed, "<%s>", path);1715 1716		thread__put(thread);1717	}1718 1719        return printed;1720}1721 1722static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,1723					      struct syscall_arg *arg)1724{1725	int fd = arg->val;1726	size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);1727	struct thread_trace *ttrace = thread__priv(arg->thread);1728 1729	if (ttrace && fd >= 0 && fd <= ttrace->files.max)1730		zfree(&ttrace->files.table[fd].pathname);1731 1732	return printed;1733}1734 1735static void thread__set_filename_pos(struct thread *thread, const char *bf,1736				     unsigned long ptr)1737{1738	struct thread_trace *ttrace = thread__priv(thread);1739 1740	ttrace->filename.ptr = ptr;1741	ttrace->filename.entry_str_pos = bf - ttrace->entry_str;1742}1743 1744static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)1745{1746	struct augmented_arg *augmented_arg = arg->augmented.args;1747	size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);1748	/*1749	 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls1750	 * we would have two strings, each prefixed by its size.1751	 */1752	int consumed = sizeof(*augmented_arg) + augmented_arg->size;1753 1754	arg->augmented.args = ((void *)arg->augmented.args) + consumed;1755	arg->augmented.size -= consumed;1756 1757	return printed;1758}1759 1760static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,1761					      struct syscall_arg *arg)1762{1763	unsigned long ptr = arg->val;1764 1765	if (arg->augmented.args)1766		return syscall_arg__scnprintf_augmented_string(arg, bf, size);1767 1768	if (!arg->trace->vfs_getname)1769		return scnprintf(bf, size, "%#x", ptr);1770 1771	thread__set_filename_pos(arg->thread, bf, ptr);1772	return 0;1773}1774 1775#define MAX_CONTROL_CHAR 311776#define MAX_ASCII 1271777 1778static size_t syscall_arg__scnprintf_buf(char *bf, size_t size, struct syscall_arg *arg)1779{1780	struct augmented_arg *augmented_arg = arg->augmented.args;1781	unsigned char *orig = (unsigned char *)augmented_arg->value;1782	size_t printed = 0;1783	int consumed;1784 1785	if (augmented_arg == NULL)1786		return 0;1787 1788	for (int j = 0; j < augmented_arg->size; ++j) {1789		bool control_char = orig[j] <= MAX_CONTROL_CHAR || orig[j] >= MAX_ASCII;1790		/* print control characters (0~31 and 127), and non-ascii characters in \(digits) */1791		printed += scnprintf(bf + printed, size - printed, control_char ? "\\%d" : "%c", (int)orig[j]);1792	}1793 1794	consumed = sizeof(*augmented_arg) + augmented_arg->size;1795	arg->augmented.args = ((void *)arg->augmented.args) + consumed;1796	arg->augmented.size -= consumed;1797 1798	return printed;1799}1800 1801static bool trace__filter_duration(struct trace *trace, double t)1802{1803	return t < (trace->duration_filter * NSEC_PER_MSEC);1804}1805 1806static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)1807{1808	double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;1809 1810	return fprintf(fp, "%10.3f ", ts);1811}1812 1813/*1814 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are1815 * using ttrace->entry_time for a thread that receives a sys_exit without1816 * first having received a sys_enter ("poll" issued before tracing session1817 * starts, lost sys_enter exit due to ring buffer overflow).1818 */1819static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)1820{1821	if (tstamp > 0)1822		return __trace__fprintf_tstamp(trace, tstamp, fp);1823 1824	return fprintf(fp, "         ? ");1825}1826 1827static pid_t workload_pid = -1;1828static volatile sig_atomic_t done = false;1829static volatile sig_atomic_t interrupted = false;1830 1831static void sighandler_interrupt(int sig __maybe_unused)1832{1833	done = interrupted = true;1834}1835 1836static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,1837			    void *context __maybe_unused)1838{1839	if (info->si_pid == workload_pid)1840		done = true;1841}1842 1843static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)1844{1845	size_t printed = 0;1846 1847	if (trace->multiple_threads) {1848		if (trace->show_comm)1849			printed += fprintf(fp, "%.14s/", thread__comm_str(thread));1850		printed += fprintf(fp, "%d ", thread__tid(thread));1851	}1852 1853	return printed;1854}1855 1856static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,1857					u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)1858{1859	size_t printed = 0;1860 1861	if (trace->show_tstamp)1862		printed = trace__fprintf_tstamp(trace, tstamp, fp);1863	if (trace->show_duration)1864		printed += fprintf_duration(duration, duration_calculated, fp);1865	return printed + trace__fprintf_comm_tid(trace, thread, fp);1866}1867 1868static int trace__process_event(struct trace *trace, struct machine *machine,1869				union perf_event *event, struct perf_sample *sample)1870{1871	int ret = 0;1872 1873	switch (event->header.type) {1874	case PERF_RECORD_LOST:1875		color_fprintf(trace->output, PERF_COLOR_RED,1876			      "LOST %" PRIu64 " events!\n", event->lost.lost);1877		ret = machine__process_lost_event(machine, event, sample);1878		break;1879	default:1880		ret = machine__process_event(machine, event, sample);1881		break;1882	}1883 1884	return ret;1885}1886 1887static int trace__tool_process(const struct perf_tool *tool,1888			       union perf_event *event,1889			       struct perf_sample *sample,1890			       struct machine *machine)1891{1892	struct trace *trace = container_of(tool, struct trace, tool);1893	return trace__process_event(trace, machine, event, sample);1894}1895 1896static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)1897{1898	struct machine *machine = vmachine;1899 1900	if (machine->kptr_restrict_warned)1901		return NULL;1902 1903	if (symbol_conf.kptr_restrict) {1904		pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"1905			   "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"1906			   "Kernel samples will not be resolved.\n");1907		machine->kptr_restrict_warned = true;1908		return NULL;1909	}1910 1911	return machine__resolve_kernel_addr(vmachine, addrp, modp);1912}1913 1914static int trace__symbols_init(struct trace *trace, struct evlist *evlist)1915{1916	int err = symbol__init(NULL);1917 1918	if (err)1919		return err;1920 1921	trace->host = machine__new_host();1922	if (trace->host == NULL)1923		return -ENOMEM;1924 1925	thread__set_priv_destructor(thread_trace__delete);1926 1927	err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);1928	if (err < 0)1929		goto out;1930 1931	err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,1932					    evlist->core.threads, trace__tool_process,1933					    true, false, 1);1934out:1935	if (err)1936		symbol__exit();1937 1938	return err;1939}1940 1941static void trace__symbols__exit(struct trace *trace)1942{1943	machine__exit(trace->host);1944	trace->host = NULL;1945 1946	symbol__exit();1947}1948 1949static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)1950{1951	int idx;1952 1953	if (nr_args == RAW_SYSCALL_ARGS_NUM && sc->fmt && sc->fmt->nr_args != 0)1954		nr_args = sc->fmt->nr_args;1955 1956	sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));1957	if (sc->arg_fmt == NULL)1958		return -1;1959 1960	for (idx = 0; idx < nr_args; ++idx) {1961		if (sc->fmt)1962			sc->arg_fmt[idx] = sc->fmt->arg[idx];1963	}1964 1965	sc->nr_args = nr_args;1966	return 0;1967}1968 1969static const struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {1970	{ .name = "msr",	.scnprintf = SCA_X86_MSR,	  .strtoul = STUL_X86_MSR,	   },1971	{ .name = "vector",	.scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, },1972};1973 1974static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)1975{1976       const struct syscall_arg_fmt *fmt = fmtp;1977       return strcmp(name, fmt->name);1978}1979 1980static const struct syscall_arg_fmt *1981__syscall_arg_fmt__find_by_name(const struct syscall_arg_fmt *fmts, const int nmemb,1982				const char *name)1983{1984       return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);1985}1986 1987static const struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)1988{1989       const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);1990       return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);1991}1992 1993static struct tep_format_field *1994syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field,1995			    bool *use_btf)1996{1997	struct tep_format_field *last_field = NULL;1998	int len;1999 2000	for (; field; field = field->next, ++arg) {2001		last_field = field;2002 2003		if (arg->scnprintf)2004			continue;2005 2006		len = strlen(field->name);2007 2008		// As far as heuristics (or intention) goes this seems to hold true, and makes sense!2009		if ((field->flags & TEP_FIELD_IS_POINTER) && strstarts(field->type, "const "))2010			arg->from_user = true;2011 2012		if (strcmp(field->type, "const char *") == 0 &&2013		    ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||2014		     strstr(field->name, "path") != NULL)) {2015			arg->scnprintf = SCA_FILENAME;2016		} else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))2017			arg->scnprintf = SCA_PTR;2018		else if (strcmp(field->type, "pid_t") == 0)2019			arg->scnprintf = SCA_PID;2020		else if (strcmp(field->type, "umode_t") == 0)2021			arg->scnprintf = SCA_MODE_T;2022		else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) {2023			arg->scnprintf = SCA_CHAR_ARRAY;2024			arg->nr_entries = field->arraylen;2025		} else if ((strcmp(field->type, "int") == 0 ||2026			  strcmp(field->type, "unsigned int") == 0 ||2027			  strcmp(field->type, "long") == 0) &&2028			 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {2029			/*2030			 * /sys/kernel/tracing/events/syscalls/sys_enter*2031			 * grep -E 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c2032			 * 65 int2033			 * 23 unsigned int2034			 * 7 unsigned long2035			 */2036			arg->scnprintf = SCA_FD;2037		} else if (strstr(field->type, "enum") && use_btf != NULL) {2038			*use_btf = true;2039			arg->strtoul = STUL_BTF_TYPE;2040		} else {2041			const struct syscall_arg_fmt *fmt =2042				syscall_arg_fmt__find_by_name(field->name);2043 2044			if (fmt) {2045				arg->scnprintf = fmt->scnprintf;2046				arg->strtoul   = fmt->strtoul;2047			}2048		}2049	}2050 2051	return last_field;2052}2053 2054static int syscall__set_arg_fmts(struct syscall *sc)2055{2056	struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args,2057									  &sc->use_btf);2058 2059	if (last_field)2060		sc->args_size = last_field->offset + last_field->size;2061 2062	return 0;2063}2064 2065static int trace__read_syscall_info(struct trace *trace, int id)2066{2067	char tp_name[128];2068	struct syscall *sc;2069	const char *name = syscalltbl__name(trace->sctbl, id);2070	int err;2071 2072#ifdef HAVE_SYSCALL_TABLE_SUPPORT2073	if (trace->syscalls.table == NULL) {2074		trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));2075		if (trace->syscalls.table == NULL)2076			return -ENOMEM;2077	}2078#else2079	if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) {2080		// When using libaudit we don't know beforehand what is the max syscall id2081		struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));2082 2083		if (table == NULL)2084			return -ENOMEM;2085 2086		// Need to memset from offset 0 and +1 members if brand new2087		if (trace->syscalls.table == NULL)2088			memset(table, 0, (id + 1) * sizeof(*sc));2089		else2090			memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc));2091 2092		trace->syscalls.table	      = table;2093		trace->sctbl->syscalls.max_id = id;2094	}2095#endif2096	sc = trace->syscalls.table + id;2097	if (sc->nonexistent)2098		return -EEXIST;2099 2100	if (name == NULL) {2101		sc->nonexistent = true;2102		return -EEXIST;2103	}2104 2105	sc->name = name;2106	sc->fmt  = syscall_fmt__find(sc->name);2107 2108	snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);2109	sc->tp_format = trace_event__tp_format("syscalls", tp_name);2110 2111	if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {2112		snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);2113		sc->tp_format = trace_event__tp_format("syscalls", tp_name);2114	}2115 2116	/*2117	 * Fails to read trace point format via sysfs node, so the trace point2118	 * doesn't exist.  Set the 'nonexistent' flag as true.2119	 */2120	if (IS_ERR(sc->tp_format)) {2121		sc->nonexistent = true;2122		return PTR_ERR(sc->tp_format);2123	}2124 2125	if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ?2126					RAW_SYSCALL_ARGS_NUM : sc->tp_format->format.nr_fields))2127		return -ENOMEM;2128 2129	sc->args = sc->tp_format->format.fields;2130	/*2131	 * We need to check and discard the first variable '__syscall_nr'2132	 * or 'nr' that mean the syscall number. It is needless here.2133	 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.2134	 */2135	if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {2136		sc->args = sc->args->next;2137		--sc->nr_args;2138	}2139 2140	sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");2141	sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");2142 2143	err = syscall__set_arg_fmts(sc);2144 2145	/* after calling syscall__set_arg_fmts() we'll know whether use_btf is true */2146	if (sc->use_btf)2147		trace__load_vmlinux_btf(trace);2148 2149	return err;2150}2151 2152static int evsel__init_tp_arg_scnprintf(struct evsel *evsel, bool *use_btf)2153{2154	struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);2155 2156	if (fmt != NULL) {2157		syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields, use_btf);2158		return 0;2159	}2160 2161	return -ENOMEM;2162}2163 2164static int intcmp(const void *a, const void *b)2165{2166	const int *one = a, *another = b;2167 2168	return *one - *another;2169}2170 2171static int trace__validate_ev_qualifier(struct trace *trace)2172{2173	int err = 0;2174	bool printed_invalid_prefix = false;2175	struct str_node *pos;2176	size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);2177 2178	trace->ev_qualifier_ids.entries = malloc(nr_allocated *2179						 sizeof(trace->ev_qualifier_ids.entries[0]));2180 2181	if (trace->ev_qualifier_ids.entries == NULL) {2182		fputs("Error:\tNot enough memory for allocating events qualifier ids\n",2183		       trace->output);2184		err = -EINVAL;2185		goto out;2186	}2187 2188	strlist__for_each_entry(pos, trace->ev_qualifier) {2189		const char *sc = pos->s;2190		int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;2191 2192		if (id < 0) {2193			id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);2194			if (id >= 0)2195				goto matches;2196 2197			if (!printed_invalid_prefix) {2198				pr_debug("Skipping unknown syscalls: ");2199				printed_invalid_prefix = true;2200			} else {2201				pr_debug(", ");2202			}2203 2204			pr_debug("%s", sc);2205			continue;2206		}2207matches:2208		trace->ev_qualifier_ids.entries[nr_used++] = id;2209		if (match_next == -1)2210			continue;2211 2212		while (1) {2213			id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);2214			if (id < 0)2215				break;2216			if (nr_allocated == nr_used) {2217				void *entries;2218 2219				nr_allocated += 8;2220				entries = realloc(trace->ev_qualifier_ids.entries,2221						  nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));2222				if (entries == NULL) {2223					err = -ENOMEM;2224					fputs("\nError:\t Not enough memory for parsing\n", trace->output);2225					goto out_free;2226				}2227				trace->ev_qualifier_ids.entries = entries;2228			}2229			trace->ev_qualifier_ids.entries[nr_used++] = id;2230		}2231	}2232 2233	trace->ev_qualifier_ids.nr = nr_used;2234	qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);2235out:2236	if (printed_invalid_prefix)2237		pr_debug("\n");2238	return err;2239out_free:2240	zfree(&trace->ev_qualifier_ids.entries);2241	trace->ev_qualifier_ids.nr = 0;2242	goto out;2243}2244 2245static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)2246{2247	bool in_ev_qualifier;2248 2249	if (trace->ev_qualifier_ids.nr == 0)2250		return true;2251 2252	in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,2253				  trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;2254 2255	if (in_ev_qualifier)2256	       return !trace->not_ev_qualifier;2257 2258	return trace->not_ev_qualifier;2259}2260 2261/*2262 * args is to be interpreted as a series of longs but we need to handle2263 * 8-byte unaligned accesses. args points to raw_data within the event2264 * and raw_data is guaranteed to be 8-byte unaligned because it is2265 * preceded by raw_size which is a u32. So we need to copy args to a temp2266 * variable to read it. Most notably this avoids extended load instructions2267 * on unaligned addresses2268 */2269unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)2270{2271	unsigned long val;2272	unsigned char *p = arg->args + sizeof(unsigned long) * idx;2273 2274	memcpy(&val, p, sizeof(val));2275	return val;2276}2277 2278static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,2279				      struct syscall_arg *arg)2280{2281	if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)2282		return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);2283 2284	return scnprintf(bf, size, "arg%d: ", arg->idx);2285}2286 2287/*2288 * Check if the value is in fact zero, i.e. mask whatever needs masking, such2289 * as mount 'flags' argument that needs ignoring some magic flag, see comment2290 * in tools/perf/trace/beauty/mount_flags.c2291 */2292static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)2293{2294	if (fmt && fmt->mask_val)2295		return fmt->mask_val(arg, val);2296 2297	return val;2298}2299 2300static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,2301					     struct syscall_arg *arg, unsigned long val)2302{2303	if (fmt && fmt->scnprintf) {2304		arg->val = val;2305		if (fmt->parm)2306			arg->parm = fmt->parm;2307		return fmt->scnprintf(bf, size, arg);2308	}2309	return scnprintf(bf, size, "%ld", val);2310}2311 2312static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,2313				      unsigned char *args, void *augmented_args, int augmented_args_size,2314				      struct trace *trace, struct thread *thread)2315{2316	size_t printed = 0, btf_printed;2317	unsigned long val;2318	u8 bit = 1;2319	struct syscall_arg arg = {2320		.args	= args,2321		.augmented = {2322			.size = augmented_args_size,2323			.args = augmented_args,2324		},2325		.idx	= 0,2326		.mask	= 0,2327		.trace  = trace,2328		.thread = thread,2329		.show_string_prefix = trace->show_string_prefix,2330	};2331	struct thread_trace *ttrace = thread__priv(thread);2332	void *default_scnprintf;2333 2334	/*2335	 * Things like fcntl will set this in its 'cmd' formatter to pick the2336	 * right formatter for the return value (an fd? file flags?), which is2337	 * not needed for syscalls that always return a given type, say an fd.2338	 */2339	ttrace->ret_scnprintf = NULL;2340 2341	if (sc->args != NULL) {2342		struct tep_format_field *field;2343 2344		for (field = sc->args; field;2345		     field = field->next, ++arg.idx, bit <<= 1) {2346			if (arg.mask & bit)2347				continue;2348 2349			arg.fmt = &sc->arg_fmt[arg.idx];2350			val = syscall_arg__val(&arg, arg.idx);2351			/*2352			 * Some syscall args need some mask, most don't and2353			 * return val untouched.2354			 */2355			val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);2356 2357			/*2358			 * Suppress this argument if its value is zero and show_zero2359			 * property isn't set.2360			 *2361			 * If it has a BTF type, then override the zero suppression knob2362			 * as the common case is for zero in an enum to have an associated entry.2363			 */2364			if (val == 0 && !trace->show_zeros &&2365			    !(sc->arg_fmt && sc->arg_fmt[arg.idx].show_zero) &&2366			    !(sc->arg_fmt && sc->arg_fmt[arg.idx].strtoul == STUL_BTF_TYPE))2367				continue;2368 2369			printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");2370 2371			if (trace->show_arg_names)2372				printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);2373 2374			default_scnprintf = sc->arg_fmt[arg.idx].scnprintf;2375 2376			if (trace->force_btf || default_scnprintf == NULL || default_scnprintf == SCA_PTR) {2377				btf_printed = trace__btf_scnprintf(trace, &arg, bf + printed,2378								   size - printed, val, field->type);2379				if (btf_printed) {2380					printed += btf_printed;2381					continue;2382				}2383			}2384 2385			printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],2386								  bf + printed, size - printed, &arg, val);2387		}2388	} else if (IS_ERR(sc->tp_format)) {2389		/*2390		 * If we managed to read the tracepoint /format file, then we2391		 * may end up not having any args, like with gettid(), so only2392		 * print the raw args when we didn't manage to read it.2393		 */2394		while (arg.idx < sc->nr_args) {2395			if (arg.mask & bit)2396				goto next_arg;2397			val = syscall_arg__val(&arg, arg.idx);2398			if (printed)2399				printed += scnprintf(bf + printed, size - printed, ", ");2400			printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);2401			printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);2402next_arg:2403			++arg.idx;2404			bit <<= 1;2405		}2406	}2407 2408	return printed;2409}2410 2411typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,2412				  union perf_event *event,2413				  struct perf_sample *sample);2414 2415static struct syscall *trace__syscall_info(struct trace *trace,2416					   struct evsel *evsel, int id)2417{2418	int err = 0;2419 2420	if (id < 0) {2421 2422		/*2423		 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried2424		 * before that, leaving at a higher verbosity level till that is2425		 * explained. Reproduced with plain ftrace with:2426		 *2427		 * echo 1 > /t/events/raw_syscalls/sys_exit/enable2428		 * grep "NR -1 " /t/trace_pipe2429		 *2430		 * After generating some load on the machine.2431 		 */2432		if (verbose > 1) {2433			static u64 n;2434			fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",2435				id, evsel__name(evsel), ++n);2436		}2437		return NULL;2438	}2439 2440	err = -EINVAL;2441 2442#ifdef HAVE_SYSCALL_TABLE_SUPPORT2443	if (id > trace->sctbl->syscalls.max_id) {2444#else2445	if (id >= trace->sctbl->syscalls.max_id) {2446		/*2447		 * With libaudit we don't know beforehand what is the max_id,2448		 * so we let trace__read_syscall_info() figure that out as we2449		 * go on reading syscalls.2450		 */2451		err = trace__read_syscall_info(trace, id);2452		if (err)2453#endif2454		goto out_cant_read;2455	}2456 2457	if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&2458	    (err = trace__read_syscall_info(trace, id)) != 0)2459		goto out_cant_read;2460 2461	if (trace->syscalls.table && trace->syscalls.table[id].nonexistent)2462		goto out_cant_read;2463 2464	return &trace->syscalls.table[id];2465 2466out_cant_read:2467	if (verbose > 0) {2468		char sbuf[STRERR_BUFSIZE];2469		fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));2470		if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)2471			fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);2472		fputs(" information\n", trace->output);2473	}2474	return NULL;2475}2476 2477struct syscall_stats {2478	struct stats stats;2479	u64	     nr_failures;2480	int	     max_errno;2481	u32	     *errnos;2482};2483 2484static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace,2485				 int id, struct perf_sample *sample, long err, bool errno_summary)2486{2487	struct int_node *inode;2488	struct syscall_stats *stats;2489	u64 duration = 0;2490 2491	inode = intlist__findnew(ttrace->syscall_stats, id);2492	if (inode == NULL)2493		return;2494 2495	stats = inode->priv;2496	if (stats == NULL) {2497		stats = zalloc(sizeof(*stats));2498		if (stats == NULL)2499			return;2500 2501		init_stats(&stats->stats);2502		inode->priv = stats;2503	}2504 2505	if (ttrace->entry_time && sample->time > ttrace->entry_time)2506		duration = sample->time - ttrace->entry_time;2507 2508	update_stats(&stats->stats, duration);2509 2510	if (err < 0) {2511		++stats->nr_failures;2512 2513		if (!errno_summary)2514			return;2515 2516		err = -err;2517		if (err > stats->max_errno) {2518			u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32));2519 2520			if (new_errnos) {2521				memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32));2522			} else {2523				pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n",2524					 thread__comm_str(thread), thread__pid(thread),2525					 thread__tid(thread));2526				return;2527			}2528 2529			stats->errnos = new_errnos;2530			stats->max_errno = err;2531		}2532 2533		++stats->errnos[err - 1];2534	}2535}2536 2537static int trace__printf_interrupted_entry(struct trace *trace)2538{2539	struct thread_trace *ttrace;2540	size_t printed;2541	int len;2542 2543	if (trace->failure_only || trace->current == NULL)2544		return 0;2545 2546	ttrace = thread__priv(trace->current);2547 2548	if (!ttrace->entry_pending)2549		return 0;2550 2551	printed  = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);2552	printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);2553 2554	if (len < trace->args_alignment - 4)2555		printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");2556 2557	printed += fprintf(trace->output, " ...\n");2558 2559	ttrace->entry_pending = false;2560	++trace->nr_events_printed;2561 2562	return printed;2563}2564 2565static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,2566				 struct perf_sample *sample, struct thread *thread)2567{2568	int printed = 0;2569 2570	if (trace->print_sample) {2571		double ts = (double)sample->time / NSEC_PER_MSEC;2572 2573		printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",2574				   evsel__name(evsel), ts,2575				   thread__comm_str(thread),2576				   sample->pid, sample->tid, sample->cpu);2577	}2578 2579	return printed;2580}2581 2582static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)2583{2584	void *augmented_args = NULL;2585	/*2586	 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter2587	 * and there we get all 6 syscall args plus the tracepoint common fields2588	 * that gets calculated at the start and the syscall_nr (another long).2589	 * So we check if that is the case and if so don't look after the2590	 * sc->args_size but always after the full raw_syscalls:sys_enter payload,2591	 * which is fixed.2592	 *2593	 * We'll revisit this later to pass s->args_size to the BPF augmenter2594	 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it2595	 * copies only what we need for each syscall, like what happens when we2596	 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace2597	 * traffic to just what is needed for each syscall.2598	 */2599	int args_size = raw_augmented_args_size ?: sc->args_size;2600 2601	*augmented_args_size = sample->raw_size - args_size;2602	if (*augmented_args_size > 0)2603		augmented_args = sample->raw_data + args_size;2604 2605	return augmented_args;2606}2607 2608static void syscall__exit(struct syscall *sc)2609{2610	if (!sc)2611		return;2612 2613	zfree(&sc->arg_fmt);2614}2615 2616static int trace__sys_enter(struct trace *trace, struct evsel *evsel,2617			    union perf_event *event __maybe_unused,2618			    struct perf_sample *sample)2619{2620	char *msg;2621	void *args;2622	int printed = 0;2623	struct thread *thread;2624	int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;2625	int augmented_args_size = 0;2626	void *augmented_args = NULL;2627	struct syscall *sc = trace__syscall_info(trace, evsel, id);2628	struct thread_trace *ttrace;2629 2630	if (sc == NULL)2631		return -1;2632 2633	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);2634	ttrace = thread__trace(thread, trace->output);2635	if (ttrace == NULL)2636		goto out_put;2637 2638	trace__fprintf_sample(trace, evsel, sample, thread);2639 2640	args = perf_evsel__sc_tp_ptr(evsel, args, sample);2641 2642	if (ttrace->entry_str == NULL) {2643		ttrace->entry_str = malloc(trace__entry_str_size);2644		if (!ttrace->entry_str)2645			goto out_put;2646	}2647 2648	if (!(trace->duration_filter || trace->summary_only || trace->min_stack))2649		trace__printf_interrupted_entry(trace);2650	/*2651	 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible2652	 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments2653	 * this breaks syscall__augmented_args() check for augmented args, as we calculate2654	 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,2655	 * so when handling, say the openat syscall, we end up getting 6 args for the2656	 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly2657	 * thinking that the extra 2 u64 args are the augmented filename, so just check2658	 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.2659	 */2660	if (evsel != trace->syscalls.events.sys_enter)2661		augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);2662	ttrace->entry_time = sample->time;2663	msg = ttrace->entry_str;2664	printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);2665 2666	printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,2667					   args, augmented_args, augmented_args_size, trace, thread);2668 2669	if (sc->is_exit) {2670		if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {2671			int alignment = 0;2672 2673			trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);2674			printed = fprintf(trace->output, "%s)", ttrace->entry_str);2675			if (trace->args_alignment > printed)2676				alignment = trace->args_alignment - printed;2677			fprintf(trace->output, "%*s= ?\n", alignment, " ");2678		}2679	} else {2680		ttrace->entry_pending = true;2681		/* See trace__vfs_getname & trace__sys_exit */2682		ttrace->filename.pending_open = false;2683	}2684 2685	if (trace->current != thread) {2686		thread__put(trace->current);2687		trace->current = thread__get(thread);2688	}2689	err = 0;2690out_put:2691	thread__put(thread);2692	return err;2693}2694 2695static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,2696				    struct perf_sample *sample)2697{2698	struct thread_trace *ttrace;2699	struct thread *thread;2700	int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;2701	struct syscall *sc = trace__syscall_info(trace, evsel, id);2702	char msg[1024];2703	void *args, *augmented_args = NULL;2704	int augmented_args_size;2705 2706	if (sc == NULL)2707		return -1;2708 2709	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);2710	ttrace = thread__trace(thread, trace->output);2711	/*2712	 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()2713	 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.2714	 */2715	if (ttrace == NULL)2716		goto out_put;2717 2718	args = perf_evsel__sc_tp_ptr(evsel, args, sample);2719	augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);2720	syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);2721	fprintf(trace->output, "%s", msg);2722	err = 0;2723out_put:2724	thread__put(thread);2725	return err;2726}2727 2728static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,2729				    struct perf_sample *sample,2730				    struct callchain_cursor *cursor)2731{2732	struct addr_location al;2733	int max_stack = evsel->core.attr.sample_max_stack ?2734			evsel->core.attr.sample_max_stack :2735			trace->max_stack;2736	int err = -1;2737 2738	addr_location__init(&al);2739	if (machine__resolve(trace->host, &al, sample) < 0)2740		goto out;2741 2742	err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);2743out:2744	addr_location__exit(&al);2745	return err;2746}2747 2748static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)2749{2750	/* TODO: user-configurable print_opts */2751	const unsigned int print_opts = EVSEL__PRINT_SYM |2752				        EVSEL__PRINT_DSO |2753				        EVSEL__PRINT_UNKNOWN_AS_ADDR;2754 2755	return sample__fprintf_callchain(sample, 38, print_opts, get_tls_callchain_cursor(), symbol_conf.bt_stop_list, trace->output);2756}2757 2758static const char *errno_to_name(struct evsel *evsel, int err)2759{2760	struct perf_env *env = evsel__env(evsel);2761 2762	return perf_env__arch_strerrno(env, err);2763}2764 2765static int trace__sys_exit(struct trace *trace, struct evsel *evsel,2766			   union perf_event *event __maybe_unused,2767			   struct perf_sample *sample)2768{2769	long ret;2770	u64 duration = 0;2771	bool duration_calculated = false;2772	struct thread *thread;2773	int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;2774	int alignment = trace->args_alignment;2775	struct syscall *sc = trace__syscall_info(trace, evsel, id);2776	struct thread_trace *ttrace;2777 2778	if (sc == NULL)2779		return -1;2780 2781	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);2782	ttrace = thread__trace(thread, trace->output);2783	if (ttrace == NULL)2784		goto out_put;2785 2786	trace__fprintf_sample(trace, evsel, sample, thread);2787 2788	ret = perf_evsel__sc_tp_uint(evsel, ret, sample);2789 2790	if (trace->summary)2791		thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary);2792 2793	if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {2794		trace__set_fd_pathname(thread, ret, ttrace->filename.name);2795		ttrace->filename.pending_open = false;2796		++trace->stats.vfs_getname;2797	}2798 2799	if (ttrace->entry_time) {2800		duration = sample->time - ttrace->entry_time;2801		if (trace__filter_duration(trace, duration))2802			goto out;2803		duration_calculated = true;2804	} else if (trace->duration_filter)2805		goto out;2806 2807	if (sample->callchain) {2808		struct callchain_cursor *cursor = get_tls_callchain_cursor();2809 2810		callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);2811		if (callchain_ret == 0) {2812			if (cursor->nr < trace->min_stack)2813				goto out;2814			callchain_ret = 1;2815		}2816	}2817 2818	if (trace->summary_only || (ret >= 0 && trace->failure_only))2819		goto out;2820 2821	trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);2822 2823	if (ttrace->entry_pending) {2824		printed = fprintf(trace->output, "%s", ttrace->entry_str);2825	} else {2826		printed += fprintf(trace->output, " ... [");2827		color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");2828		printed += 9;2829		printed += fprintf(trace->output, "]: %s()", sc->name);2830	}2831 2832	printed++; /* the closing ')' */2833 2834	if (alignment > printed)2835		alignment -= printed;2836	else2837		alignment = 0;2838 2839	fprintf(trace->output, ")%*s= ", alignment, " ");2840 2841	if (sc->fmt == NULL) {2842		if (ret < 0)2843			goto errno_print;2844signed_print:2845		fprintf(trace->output, "%ld", ret);2846	} else if (ret < 0) {2847errno_print: {2848		char bf[STRERR_BUFSIZE];2849		const char *emsg = str_error_r(-ret, bf, sizeof(bf)),2850			   *e = errno_to_name(evsel, -ret);2851 2852		fprintf(trace->output, "-1 %s (%s)", e, emsg);2853	}2854	} else if (ret == 0 && sc->fmt->timeout)2855		fprintf(trace->output, "0 (Timeout)");2856	else if (ttrace->ret_scnprintf) {2857		char bf[1024];2858		struct syscall_arg arg = {2859			.val	= ret,2860			.thread	= thread,2861			.trace	= trace,2862		};2863		ttrace->ret_scnprintf(bf, sizeof(bf), &arg);2864		ttrace->ret_scnprintf = NULL;2865		fprintf(trace->output, "%s", bf);2866	} else if (sc->fmt->hexret)2867		fprintf(trace->output, "%#lx", ret);2868	else if (sc->fmt->errpid) {2869		struct thread *child = machine__find_thread(trace->host, ret, ret);2870 2871		if (child != NULL) {2872			fprintf(trace->output, "%ld", ret);2873			if (thread__comm_set(child))2874				fprintf(trace->output, " (%s)", thread__comm_str(child));2875			thread__put(child);2876		}2877	} else2878		goto signed_print;2879 2880	fputc('\n', trace->output);2881 2882	/*2883	 * We only consider an 'event' for the sake of --max-events a non-filtered2884	 * sys_enter + sys_exit and other tracepoint events.2885	 */2886	if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)2887		interrupted = true;2888 2889	if (callchain_ret > 0)2890		trace__fprintf_callchain(trace, sample);2891	else if (callchain_ret < 0)2892		pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));2893out:2894	ttrace->entry_pending = false;2895	err = 0;2896out_put:2897	thread__put(thread);2898	return err;2899}2900 2901static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,2902			      union perf_event *event __maybe_unused,2903			      struct perf_sample *sample)2904{2905	struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);2906	struct thread_trace *ttrace;2907	size_t filename_len, entry_str_len, to_move;2908	ssize_t remaining_space;2909	char *pos;2910	const char *filename = evsel__rawptr(evsel, sample, "pathname");2911 2912	if (!thread)2913		goto out;2914 2915	ttrace = thread__priv(thread);2916	if (!ttrace)2917		goto out_put;2918 2919	filename_len = strlen(filename);2920	if (filename_len == 0)2921		goto out_put;2922 2923	if (ttrace->filename.namelen < filename_len) {2924		char *f = realloc(ttrace->filename.name, filename_len + 1);2925 2926		if (f == NULL)2927			goto out_put;2928 2929		ttrace->filename.namelen = filename_len;2930		ttrace->filename.name = f;2931	}2932 2933	strcpy(ttrace->filename.name, filename);2934	ttrace->filename.pending_open = true;2935 2936	if (!ttrace->filename.ptr)2937		goto out_put;2938 2939	entry_str_len = strlen(ttrace->entry_str);2940	remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */2941	if (remaining_space <= 0)2942		goto out_put;2943 2944	if (filename_len > (size_t)remaining_space) {2945		filename += filename_len - remaining_space;2946		filename_len = remaining_space;2947	}2948 2949	to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */2950	pos = ttrace->entry_str + ttrace->filename.entry_str_pos;2951	memmove(pos + filename_len, pos, to_move);2952	memcpy(pos, filename, filename_len);2953 2954	ttrace->filename.ptr = 0;2955	ttrace->filename.entry_str_pos = 0;2956out_put:2957	thread__put(thread);2958out:2959	return 0;2960}2961 2962static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,2963				     union perf_event *event __maybe_unused,2964				     struct perf_sample *sample)2965{2966        u64 runtime = evsel__intval(evsel, sample, "runtime");2967	double runtime_ms = (double)runtime / NSEC_PER_MSEC;2968	struct thread *thread = machine__findnew_thread(trace->host,2969							sample->pid,2970							sample->tid);2971	struct thread_trace *ttrace = thread__trace(thread, trace->output);2972 2973	if (ttrace == NULL)2974		goto out_dump;2975 2976	ttrace->runtime_ms += runtime_ms;2977	trace->runtime_ms += runtime_ms;2978out_put:2979	thread__put(thread);2980	return 0;2981 2982out_dump:2983	fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",2984	       evsel->name,2985	       evsel__strval(evsel, sample, "comm"),2986	       (pid_t)evsel__intval(evsel, sample, "pid"),2987	       runtime,2988	       evsel__intval(evsel, sample, "vruntime"));2989	goto out_put;2990}2991 2992static int bpf_output__printer(enum binary_printer_ops op,2993			       unsigned int val, void *extra __maybe_unused, FILE *fp)2994{2995	unsigned char ch = (unsigned char)val;2996 2997	switch (op) {2998	case BINARY_PRINT_CHAR_DATA:2999		return fprintf(fp, "%c", isprint(ch) ? ch : '.');3000	case BINARY_PRINT_DATA_BEGIN:3001	case BINARY_PRINT_LINE_BEGIN:3002	case BINARY_PRINT_ADDR:3003	case BINARY_PRINT_NUM_DATA:3004	case BINARY_PRINT_NUM_PAD:3005	case BINARY_PRINT_SEP:3006	case BINARY_PRINT_CHAR_PAD:3007	case BINARY_PRINT_LINE_END:3008	case BINARY_PRINT_DATA_END:3009	default:3010		break;3011	}3012 3013	return 0;3014}3015 3016static void bpf_output__fprintf(struct trace *trace,3017				struct perf_sample *sample)3018{3019	binary__fprintf(sample->raw_data, sample->raw_size, 8,3020			bpf_output__printer, NULL, trace->output);3021	++trace->nr_events_printed;3022}3023 3024static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,3025				       struct thread *thread, void *augmented_args, int augmented_args_size)3026{3027	char bf[2048];3028	size_t size = sizeof(bf);3029	struct tep_format_field *field = evsel->tp_format->format.fields;3030	struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel);3031	size_t printed = 0, btf_printed;3032	unsigned long val;3033	u8 bit = 1;3034	struct syscall_arg syscall_arg = {3035		.augmented = {3036			.size = augmented_args_size,3037			.args = augmented_args,3038		},3039		.idx	= 0,3040		.mask	= 0,3041		.trace  = trace,3042		.thread = thread,3043		.show_string_prefix = trace->show_string_prefix,3044	};3045 3046	for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {3047		if (syscall_arg.mask & bit)3048			continue;3049 3050		syscall_arg.len = 0;3051		syscall_arg.fmt = arg;3052		if (field->flags & TEP_FIELD_IS_ARRAY) {3053			int offset = field->offset;3054 3055			if (field->flags & TEP_FIELD_IS_DYNAMIC) {3056				offset = format_field__intval(field, sample, evsel->needs_swap);3057				syscall_arg.len = offset >> 16;3058				offset &= 0xffff;3059				if (tep_field_is_relative(field->flags))3060					offset += field->offset + field->size;3061			}3062 3063			val = (uintptr_t)(sample->raw_data + offset);3064		} else3065			val = format_field__intval(field, sample, evsel->needs_swap);3066		/*3067		 * Some syscall args need some mask, most don't and3068		 * return val untouched.3069		 */3070		val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);3071 3072		/* Suppress this argument if its value is zero and show_zero property isn't set. */3073		if (val == 0 && !trace->show_zeros && !arg->show_zero && arg->strtoul != STUL_BTF_TYPE)3074			continue;3075 3076		printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");3077 3078		if (trace->show_arg_names)3079			printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);3080 3081		btf_printed = trace__btf_scnprintf(trace, &syscall_arg, bf + printed, size - printed, val, field->type);3082		if (btf_printed) {3083			printed += btf_printed;3084			continue;3085		}3086 3087		printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);3088	}3089 3090	return printed + fprintf(trace->output, "%s", bf);3091}3092 3093static int trace__event_handler(struct trace *trace, struct evsel *evsel,3094				union perf_event *event __maybe_unused,3095				struct perf_sample *sample)3096{3097	struct thread *thread;3098	int callchain_ret = 0;3099	/*3100	 * Check if we called perf_evsel__disable(evsel) due to, for instance,3101	 * this event's max_events having been hit and this is an entry coming3102	 * from the ring buffer that we should discard, since the max events3103	 * have already been considered/printed.3104	 */3105	if (evsel->disabled)3106		return 0;3107 3108	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);3109 3110	if (sample->callchain) {3111		struct callchain_cursor *cursor = get_tls_callchain_cursor();3112 3113		callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);3114		if (callchain_ret == 0) {3115			if (cursor->nr < trace->min_stack)3116				goto out;3117			callchain_ret = 1;3118		}3119	}3120 3121	trace__printf_interrupted_entry(trace);3122	trace__fprintf_tstamp(trace, sample->time, trace->output);3123 3124	if (trace->trace_syscalls && trace->show_duration)3125		fprintf(trace->output, "(         ): ");3126 3127	if (thread)3128		trace__fprintf_comm_tid(trace, thread, trace->output);3129 3130	if (evsel == trace->syscalls.events.bpf_output) {3131		int id = perf_evsel__sc_tp_uint(evsel, id, sample);3132		struct syscall *sc = trace__syscall_info(trace, evsel, id);3133 3134		if (sc) {3135			fprintf(trace->output, "%s(", sc->name);3136			trace__fprintf_sys_enter(trace, evsel, sample);3137			fputc(')', trace->output);3138			goto newline;3139		}3140 3141		/*3142		 * XXX: Not having the associated syscall info or not finding/adding3143		 * 	the thread should never happen, but if it does...3144		 * 	fall thru and print it as a bpf_output event.3145		 */3146	}3147 3148	fprintf(trace->output, "%s(", evsel->name);3149 3150	if (evsel__is_bpf_output(evsel)) {3151		bpf_output__fprintf(trace, sample);3152	} else if (evsel->tp_format) {3153		if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||3154		    trace__fprintf_sys_enter(trace, evsel, sample)) {3155			if (trace->libtraceevent_print) {3156				event_format__fprintf(evsel->tp_format, sample->cpu,3157						      sample->raw_data, sample->raw_size,3158						      trace->output);3159			} else {3160				trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);3161			}3162		}3163	}3164 3165newline:3166	fprintf(trace->output, ")\n");3167 3168	if (callchain_ret > 0)3169		trace__fprintf_callchain(trace, sample);3170	else if (callchain_ret < 0)3171		pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));3172 3173	++trace->nr_events_printed;3174 3175	if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {3176		evsel__disable(evsel);3177		evsel__close(evsel);3178	}3179out:3180	thread__put(thread);3181	return 0;3182}3183 3184static void print_location(FILE *f, struct perf_sample *sample,3185			   struct addr_location *al,3186			   bool print_dso, bool print_sym)3187{3188 3189	if ((verbose > 0 || print_dso) && al->map)3190		fprintf(f, "%s@", dso__long_name(map__dso(al->map)));3191 3192	if ((verbose > 0 || print_sym) && al->sym)3193		fprintf(f, "%s+0x%" PRIx64, al->sym->name,3194			al->addr - al->sym->start);3195	else if (al->map)3196		fprintf(f, "0x%" PRIx64, al->addr);3197	else3198		fprintf(f, "0x%" PRIx64, sample->addr);3199}3200 3201static int trace__pgfault(struct trace *trace,3202			  struct evsel *evsel,3203			  union perf_event *event __maybe_unused,3204			  struct perf_sample *sample)3205{3206	struct thread *thread;3207	struct addr_location al;3208	char map_type = 'd';3209	struct thread_trace *ttrace;3210	int err = -1;3211	int callchain_ret = 0;3212 3213	addr_location__init(&al);3214	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);3215 3216	if (sample->callchain) {3217		struct callchain_cursor *cursor = get_tls_callchain_cursor();3218 3219		callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);3220		if (callchain_ret == 0) {3221			if (cursor->nr < trace->min_stack)3222				goto out_put;3223			callchain_ret = 1;3224		}3225	}3226 3227	ttrace = thread__trace(thread, trace->output);3228	if (ttrace == NULL)3229		goto out_put;3230 3231	if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)3232		ttrace->pfmaj++;3233	else3234		ttrace->pfmin++;3235 3236	if (trace->summary_only)3237		goto out;3238 3239	thread__find_symbol(thread, sample->cpumode, sample->ip, &al);3240 3241	trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);3242 3243	fprintf(trace->output, "%sfault [",3244		evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?3245		"maj" : "min");3246 3247	print_location(trace->output, sample, &al, false, true);3248 3249	fprintf(trace->output, "] => ");3250 3251	thread__find_symbol(thread, sample->cpumode, sample->addr, &al);3252 3253	if (!al.map) {3254		thread__find_symbol(thread, sample->cpumode, sample->addr, &al);3255 3256		if (al.map)3257			map_type = 'x';3258		else3259			map_type = '?';3260	}3261 3262	print_location(trace->output, sample, &al, true, false);3263 3264	fprintf(trace->output, " (%c%c)\n", map_type, al.level);3265 3266	if (callchain_ret > 0)3267		trace__fprintf_callchain(trace, sample);3268	else if (callchain_ret < 0)3269		pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));3270 3271	++trace->nr_events_printed;3272out:3273	err = 0;3274out_put:3275	thread__put(thread);3276	addr_location__exit(&al);3277	return err;3278}3279 3280static void trace__set_base_time(struct trace *trace,3281				 struct evsel *evsel,3282				 struct perf_sample *sample)3283{3284	/*3285	 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust3286	 * and don't use sample->time unconditionally, we may end up having3287	 * some other event in the future without PERF_SAMPLE_TIME for good3288	 * reason, i.e. we may not be interested in its timestamps, just in3289	 * it taking place, picking some piece of information when it3290	 * appears in our event stream (vfs_getname comes to mind).3291	 */3292	if (trace->base_time == 0 && !trace->full_time &&3293	    (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))3294		trace->base_time = sample->time;3295}3296 3297static int trace__process_sample(const struct perf_tool *tool,3298				 union perf_event *event,3299				 struct perf_sample *sample,3300				 struct evsel *evsel,3301				 struct machine *machine __maybe_unused)3302{3303	struct trace *trace = container_of(tool, struct trace, tool);3304	struct thread *thread;3305	int err = 0;3306 3307	tracepoint_handler handler = evsel->handler;3308 3309	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);3310	if (thread && thread__is_filtered(thread))3311		goto out;3312 3313	trace__set_base_time(trace, evsel, sample);3314 3315	if (handler) {3316		++trace->nr_events;3317		handler(trace, evsel, event, sample);3318	}3319out:3320	thread__put(thread);3321	return err;3322}3323 3324static int trace__record(struct trace *trace, int argc, const char **argv)3325{3326	unsigned int rec_argc, i, j;3327	const char **rec_argv;3328	const char * const record_args[] = {3329		"record",3330		"-R",3331		"-m", "1024",3332		"-c", "1",3333	};3334	pid_t pid = getpid();3335	char *filter = asprintf__tp_filter_pids(1, &pid);3336	const char * const sc_args[] = { "-e", };3337	unsigned int sc_args_nr = ARRAY_SIZE(sc_args);3338	const char * const majpf_args[] = { "-e", "major-faults" };3339	unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);3340	const char * const minpf_args[] = { "-e", "minor-faults" };3341	unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);3342	int err = -1;3343 3344	/* +3 is for the event string below and the pid filter */3345	rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 +3346		majpf_args_nr + minpf_args_nr + argc;3347	rec_argv = calloc(rec_argc + 1, sizeof(char *));3348 3349	if (rec_argv == NULL || filter == NULL)3350		goto out_free;3351 3352	j = 0;3353	for (i = 0; i < ARRAY_SIZE(record_args); i++)3354		rec_argv[j++] = record_args[i];3355 3356	if (trace->trace_syscalls) {3357		for (i = 0; i < sc_args_nr; i++)3358			rec_argv[j++] = sc_args[i];3359 3360		/* event string may be different for older kernels - e.g., RHEL6 */3361		if (is_valid_tracepoint("raw_syscalls:sys_enter"))3362			rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";3363		else if (is_valid_tracepoint("syscalls:sys_enter"))3364			rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";3365		else {3366			pr_err("Neither raw_syscalls nor syscalls events exist.\n");3367			goto out_free;3368		}3369	}3370 3371	rec_argv[j++] = "--filter";3372	rec_argv[j++] = filter;3373 3374	if (trace->trace_pgfaults & TRACE_PFMAJ)3375		for (i = 0; i < majpf_args_nr; i++)3376			rec_argv[j++] = majpf_args[i];3377 3378	if (trace->trace_pgfaults & TRACE_PFMIN)3379		for (i = 0; i < minpf_args_nr; i++)3380			rec_argv[j++] = minpf_args[i];3381 3382	for (i = 0; i < (unsigned int)argc; i++)3383		rec_argv[j++] = argv[i];3384 3385	err = cmd_record(j, rec_argv);3386out_free:3387	free(filter);3388	free(rec_argv);3389	return err;3390}3391 3392static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);3393 3394static bool evlist__add_vfs_getname(struct evlist *evlist)3395{3396	bool found = false;3397	struct evsel *evsel, *tmp;3398	struct parse_events_error err;3399	int ret;3400 3401	parse_events_error__init(&err);3402	ret = parse_events(evlist, "probe:vfs_getname*", &err);3403	parse_events_error__exit(&err);3404	if (ret)3405		return false;3406 3407	evlist__for_each_entry_safe(evlist, evsel, tmp) {3408		if (!strstarts(evsel__name(evsel), "probe:vfs_getname"))3409			continue;3410 3411		if (evsel__field(evsel, "pathname")) {3412			evsel->handler = trace__vfs_getname;3413			found = true;3414			continue;3415		}3416 3417		list_del_init(&evsel->core.node);3418		evsel->evlist = NULL;3419		evsel__delete(evsel);3420	}3421 3422	return found;3423}3424 3425static struct evsel *evsel__new_pgfault(u64 config)3426{3427	struct evsel *evsel;3428	struct perf_event_attr attr = {3429		.type = PERF_TYPE_SOFTWARE,3430		.mmap_data = 1,3431	};3432 3433	attr.config = config;3434	attr.sample_period = 1;3435 3436	event_attr_init(&attr);3437 3438	evsel = evsel__new(&attr);3439	if (evsel)3440		evsel->handler = trace__pgfault;3441 3442	return evsel;3443}3444 3445static void evlist__free_syscall_tp_fields(struct evlist *evlist)3446{3447	struct evsel *evsel;3448 3449	evlist__for_each_entry(evlist, evsel) {3450		evsel_trace__delete(evsel->priv);3451		evsel->priv = NULL;3452	}3453}3454 3455static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)3456{3457	const u32 type = event->header.type;3458	struct evsel *evsel;3459 3460	if (type != PERF_RECORD_SAMPLE) {3461		trace__process_event(trace, trace->host, event, sample);3462		return;3463	}3464 3465	evsel = evlist__id2evsel(trace->evlist, sample->id);3466	if (evsel == NULL) {3467		fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);3468		return;3469	}3470 3471	if (evswitch__discard(&trace->evswitch, evsel))3472		return;3473 3474	trace__set_base_time(trace, evsel, sample);3475 3476	if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&3477	    sample->raw_data == NULL) {3478		fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",3479		       evsel__name(evsel), sample->tid,3480		       sample->cpu, sample->raw_size);3481	} else {3482		tracepoint_handler handler = evsel->handler;3483		handler(trace, evsel, event, sample);3484	}3485 3486	if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)3487		interrupted = true;3488}3489 3490static int trace__add_syscall_newtp(struct trace *trace)3491{3492	int ret = -1;3493	struct evlist *evlist = trace->evlist;3494	struct evsel *sys_enter, *sys_exit;3495 3496	sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);3497	if (sys_enter == NULL)3498		goto out;3499 3500	if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))3501		goto out_delete_sys_enter;3502 3503	sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);3504	if (sys_exit == NULL)3505		goto out_delete_sys_enter;3506 3507	if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))3508		goto out_delete_sys_exit;3509 3510	evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);3511	evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);3512 3513	evlist__add(evlist, sys_enter);3514	evlist__add(evlist, sys_exit);3515 3516	if (callchain_param.enabled && !trace->kernel_syscallchains) {3517		/*3518		 * We're interested only in the user space callchain3519		 * leading to the syscall, allow overriding that for3520		 * debugging reasons using --kernel_syscall_callchains3521		 */3522		sys_exit->core.attr.exclude_callchain_kernel = 1;3523	}3524 3525	trace->syscalls.events.sys_enter = sys_enter;3526	trace->syscalls.events.sys_exit  = sys_exit;3527 3528	ret = 0;3529out:3530	return ret;3531 3532out_delete_sys_exit:3533	evsel__delete_priv(sys_exit);3534out_delete_sys_enter:3535	evsel__delete_priv(sys_enter);3536	goto out;3537}3538 3539static int trace__set_ev_qualifier_tp_filter(struct trace *trace)3540{3541	int err = -1;3542	struct evsel *sys_exit;3543	char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,3544						trace->ev_qualifier_ids.nr,3545						trace->ev_qualifier_ids.entries);3546 3547	if (filter == NULL)3548		goto out_enomem;3549 3550	if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) {3551		sys_exit = trace->syscalls.events.sys_exit;3552		err = evsel__append_tp_filter(sys_exit, filter);3553	}3554 3555	free(filter);3556out:3557	return err;3558out_enomem:3559	errno = ENOMEM;3560	goto out;3561}3562 3563#ifdef HAVE_BPF_SKEL3564static int syscall_arg_fmt__cache_btf_struct(struct syscall_arg_fmt *arg_fmt, struct btf *btf, char *type)3565{3566       int id;3567 3568	if (arg_fmt->type != NULL)3569		return -1;3570 3571       id = btf__find_by_name(btf, type);3572       if (id < 0)3573		return -1;3574 3575       arg_fmt->type    = btf__type_by_id(btf, id);3576       arg_fmt->type_id = id;3577 3578       return 0;3579}3580 3581static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)3582{3583	struct bpf_program *pos, *prog = NULL;3584	const char *sec_name;3585 3586	if (trace->skel->obj == NULL)3587		return NULL;3588 3589	bpf_object__for_each_program(pos, trace->skel->obj) {3590		sec_name = bpf_program__section_name(pos);3591		if (sec_name && !strcmp(sec_name, name)) {3592			prog = pos;3593			break;3594		}3595	}3596 3597	return prog;3598}3599 3600static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,3601							const char *prog_name, const char *type)3602{3603	struct bpf_program *prog;3604 3605	if (prog_name == NULL) {3606		char default_prog_name[256];3607		scnprintf(default_prog_name, sizeof(default_prog_name), "tp/syscalls/sys_%s_%s", type, sc->name);3608		prog = trace__find_bpf_program_by_title(trace, default_prog_name);3609		if (prog != NULL)3610			goto out_found;3611		if (sc->fmt && sc->fmt->alias) {3612			scnprintf(default_prog_name, sizeof(default_prog_name), "tp/syscalls/sys_%s_%s", type, sc->fmt->alias);3613			prog = trace__find_bpf_program_by_title(trace, default_prog_name);3614			if (prog != NULL)3615				goto out_found;3616		}3617		goto out_unaugmented;3618	}3619 3620	prog = trace__find_bpf_program_by_title(trace, prog_name);3621 3622	if (prog != NULL) {3623out_found:3624		return prog;3625	}3626 3627	pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",3628		 prog_name, type, sc->name);3629out_unaugmented:3630	return trace->skel->progs.syscall_unaugmented;3631}3632 3633static void trace__init_syscall_bpf_progs(struct trace *trace, int id)3634{3635	struct syscall *sc = trace__syscall_info(trace, NULL, id);3636 3637	if (sc == NULL)3638		return;3639 3640	sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");3641	sc->bpf_prog.sys_exit  = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit  : NULL,  "exit");3642}3643 3644static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)3645{3646	struct syscall *sc = trace__syscall_info(trace, NULL, id);3647	return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->skel->progs.syscall_unaugmented);3648}3649 3650static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)3651{3652	struct syscall *sc = trace__syscall_info(trace, NULL, id);3653	return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->skel->progs.syscall_unaugmented);3654}3655 3656static int trace__bpf_sys_enter_beauty_map(struct trace *trace, int key, unsigned int *beauty_array)3657{3658	struct tep_format_field *field;3659	struct syscall *sc = trace__syscall_info(trace, NULL, key);3660	const struct btf_type *bt;3661	char *struct_offset, *tmp, name[32];3662	bool can_augment = false;3663	int i, cnt;3664 3665	if (sc == NULL)3666		return -1;3667 3668	trace__load_vmlinux_btf(trace);3669	if (trace->btf == NULL)3670		return -1;3671 3672	for (i = 0, field = sc->args; field; ++i, field = field->next) {3673		// XXX We're only collecting pointer payloads _from_ user space3674		if (!sc->arg_fmt[i].from_user)3675			continue;3676 3677		struct_offset = strstr(field->type, "struct ");3678		if (struct_offset == NULL)3679			struct_offset = strstr(field->type, "union ");3680		else3681			struct_offset++; // "union" is shorter3682 3683		if (field->flags & TEP_FIELD_IS_POINTER && struct_offset) { /* struct or union (think BPF's attr arg) */3684			struct_offset += 6;3685 3686			/* for 'struct foo *', we only want 'foo' */3687			for (tmp = struct_offset, cnt = 0; *tmp != ' ' && *tmp != '\0'; ++tmp, ++cnt) {3688			}3689 3690			strncpy(name, struct_offset, cnt);3691			name[cnt] = '\0';3692 3693			/* cache struct's btf_type and type_id */3694			if (syscall_arg_fmt__cache_btf_struct(&sc->arg_fmt[i], trace->btf, name))3695				continue;3696 3697			bt = sc->arg_fmt[i].type;3698			beauty_array[i] = bt->size;3699			can_augment = true;3700		} else if (field->flags & TEP_FIELD_IS_POINTER && /* string */3701			   strcmp(field->type, "const char *") == 0 &&3702			   (strstr(field->name, "name") ||3703			    strstr(field->name, "path") ||3704			    strstr(field->name, "file") ||3705			    strstr(field->name, "root") ||3706			    strstr(field->name, "key") ||3707			    strstr(field->name, "special") ||3708			    strstr(field->name, "type") ||3709			    strstr(field->name, "description"))) {3710			beauty_array[i] = 1;3711			can_augment = true;3712		} else if (field->flags & TEP_FIELD_IS_POINTER && /* buffer */3713			   strstr(field->type, "char *") &&3714			   (strstr(field->name, "buf") ||3715			    strstr(field->name, "val") ||3716			    strstr(field->name, "msg"))) {3717			int j;3718			struct tep_format_field *field_tmp;3719 3720			/* find the size of the buffer that appears in pairs with buf */3721			for (j = 0, field_tmp = sc->args; field_tmp; ++j, field_tmp = field_tmp->next) {3722				if (!(field_tmp->flags & TEP_FIELD_IS_POINTER) && /* only integers */3723				    (strstr(field_tmp->name, "count") ||3724				     strstr(field_tmp->name, "siz") ||  /* size, bufsiz */3725				     (strstr(field_tmp->name, "len") && strcmp(field_tmp->name, "filename")))) {3726					 /* filename's got 'len' in it, we don't want that */3727					beauty_array[i] = -(j + 1);3728					can_augment = true;3729					break;3730				}3731			}3732		}3733	}3734 3735	if (can_augment)3736		return 0;3737 3738	return -1;3739}3740 3741static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)3742{3743	struct tep_format_field *field, *candidate_field;3744	/*3745	 * We're only interested in syscalls that have a pointer:3746	 */3747	for (field = sc->args; field; field = field->next) {3748		if (field->flags & TEP_FIELD_IS_POINTER)3749			goto try_to_find_pair;3750	}3751 3752	return NULL;3753 3754try_to_find_pair:3755	for (int i = 0; i < trace->sctbl->syscalls.nr_entries; ++i) {3756		int id = syscalltbl__id_at_idx(trace->sctbl, i);3757		struct syscall *pair = trace__syscall_info(trace, NULL, id);3758		struct bpf_program *pair_prog;3759		bool is_candidate = false;3760 3761		if (pair == NULL || pair == sc ||3762		    pair->bpf_prog.sys_enter == trace->skel->progs.syscall_unaugmented)3763			continue;3764 3765		for (field = sc->args, candidate_field = pair->args;3766		     field && candidate_field; field = field->next, candidate_field = candidate_field->next) {3767			bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,3768			     candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;3769 3770			if (is_pointer) {3771			       if (!candidate_is_pointer) {3772					// The candidate just doesn't copies our pointer arg, might copy other pointers we want.3773					continue;3774			       }3775			} else {3776				if (candidate_is_pointer) {3777					// The candidate might copy a pointer we don't have, skip it.3778					goto next_candidate;3779				}3780				continue;3781			}3782 3783			if (strcmp(field->type, candidate_field->type))3784				goto next_candidate;3785 3786			/*3787			 * This is limited in the BPF program but sys_write3788			 * uses "const char *" for its "buf" arg so we need to3789			 * use some heuristic that is kinda future proof...3790			 */3791			if (strcmp(field->type, "const char *") == 0 &&3792			    !(strstr(field->name, "name") ||3793			      strstr(field->name, "path") ||3794			      strstr(field->name, "file") ||3795			      strstr(field->name, "root") ||3796			      strstr(field->name, "description")))3797				goto next_candidate;3798 3799			is_candidate = true;3800		}3801 3802		if (!is_candidate)3803			goto next_candidate;3804 3805		/*3806		 * Check if the tentative pair syscall augmenter has more pointers, if it has,3807		 * then it may be collecting that and we then can't use it, as it would collect3808		 * more than what is common to the two syscalls.3809		 */3810		if (candidate_field) {3811			for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)3812				if (candidate_field->flags & TEP_FIELD_IS_POINTER)3813					goto next_candidate;3814		}3815 3816		pair_prog = pair->bpf_prog.sys_enter;3817		/*3818		 * If the pair isn't enabled, then its bpf_prog.sys_enter will not3819		 * have been searched for, so search it here and if it returns the3820		 * unaugmented one, then ignore it, otherwise we'll reuse that BPF3821		 * program for a filtered syscall on a non-filtered one.3822		 *3823		 * For instance, we have "!syscalls:sys_enter_renameat" and that is3824		 * useful for "renameat2".3825		 */3826		if (pair_prog == NULL) {3827			pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");3828			if (pair_prog == trace->skel->progs.syscall_unaugmented)3829				goto next_candidate;3830		}3831 3832		pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);3833		return pair_prog;3834	next_candidate:3835		continue;3836	}3837 3838	return NULL;3839}3840 3841static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)3842{3843	int map_enter_fd = bpf_map__fd(trace->skel->maps.syscalls_sys_enter);3844	int map_exit_fd  = bpf_map__fd(trace->skel->maps.syscalls_sys_exit);3845	int beauty_map_fd = bpf_map__fd(trace->skel->maps.beauty_map_enter);3846	int err = 0;3847	unsigned int beauty_array[6];3848 3849	for (int i = 0; i < trace->sctbl->syscalls.nr_entries; ++i) {3850		int prog_fd, key = syscalltbl__id_at_idx(trace->sctbl, i);3851 3852		if (!trace__syscall_enabled(trace, key))3853			continue;3854 3855		trace__init_syscall_bpf_progs(trace, key);3856 3857		// It'll get at least the "!raw_syscalls:unaugmented"3858		prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);3859		err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);3860		if (err)3861			break;3862		prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);3863		err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);3864		if (err)3865			break;3866 3867		/* use beauty_map to tell BPF how many bytes to collect, set beauty_map's value here */3868		memset(beauty_array, 0, sizeof(beauty_array));3869		err = trace__bpf_sys_enter_beauty_map(trace, key, (unsigned int *)beauty_array);3870		if (err)3871			continue;3872		err = bpf_map_update_elem(beauty_map_fd, &key, beauty_array, BPF_ANY);3873		if (err)3874			break;3875	}3876 3877	/*3878	 * Now lets do a second pass looking for enabled syscalls without3879	 * an augmenter that have a signature that is a superset of another3880	 * syscall with an augmenter so that we can auto-reuse it.3881	 *3882	 * I.e. if we have an augmenter for the "open" syscall that has3883	 * this signature:3884	 *3885	 *   int open(const char *pathname, int flags, mode_t mode);3886	 *3887	 * I.e. that will collect just the first string argument, then we3888	 * can reuse it for the 'creat' syscall, that has this signature:3889	 *3890	 *   int creat(const char *pathname, mode_t mode);3891	 *3892	 * and for:3893	 *3894	 *   int stat(const char *pathname, struct stat *statbuf);3895	 *   int lstat(const char *pathname, struct stat *statbuf);3896	 *3897	 * Because the 'open' augmenter will collect the first arg as a string,3898	 * and leave alone all the other args, which already helps with3899	 * beautifying 'stat' and 'lstat''s pathname arg.3900	 *3901	 * Then, in time, when 'stat' gets an augmenter that collects both3902	 * first and second arg (this one on the raw_syscalls:sys_exit prog3903	 * array tail call, then that one will be used.3904	 */3905	for (int i = 0; i < trace->sctbl->syscalls.nr_entries; ++i) {3906		int key = syscalltbl__id_at_idx(trace->sctbl, i);3907		struct syscall *sc = trace__syscall_info(trace, NULL, key);3908		struct bpf_program *pair_prog;3909		int prog_fd;3910 3911		if (sc == NULL || sc->bpf_prog.sys_enter == NULL)3912			continue;3913 3914		/*3915		 * For now we're just reusing the sys_enter prog, and if it3916		 * already has an augmenter, we don't need to find one.3917		 */3918		if (sc->bpf_prog.sys_enter != trace->skel->progs.syscall_unaugmented)3919			continue;3920 3921		/*3922		 * Look at all the other syscalls for one that has a signature3923		 * that is close enough that we can share:3924		 */3925		pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);3926		if (pair_prog == NULL)3927			continue;3928 3929		sc->bpf_prog.sys_enter = pair_prog;3930 3931		/*3932		 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter3933		 * with the fd for the program we're reusing:3934		 */3935		prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);3936		err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);3937		if (err)3938			break;3939	}3940 3941	return err;3942}3943#endif // HAVE_BPF_SKEL3944 3945static int trace__set_ev_qualifier_filter(struct trace *trace)3946{3947	if (trace->syscalls.events.sys_enter)3948		return trace__set_ev_qualifier_tp_filter(trace);3949	return 0;3950}3951 3952static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,3953				    size_t npids __maybe_unused, pid_t *pids __maybe_unused)3954{3955	int err = 0;3956#ifdef HAVE_LIBBPF_SUPPORT3957	bool value = true;3958	int map_fd = bpf_map__fd(map);3959	size_t i;3960 3961	for (i = 0; i < npids; ++i) {3962		err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);3963		if (err)3964			break;3965	}3966#endif3967	return err;3968}3969 3970static int trace__set_filter_loop_pids(struct trace *trace)3971{3972	unsigned int nr = 1, err;3973	pid_t pids[32] = {3974		getpid(),3975	};3976	struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);3977 3978	while (thread && nr < ARRAY_SIZE(pids)) {3979		struct thread *parent = machine__find_thread(trace->host,3980							     thread__ppid(thread),3981							     thread__ppid(thread));3982 3983		if (parent == NULL)3984			break;3985 3986		if (!strcmp(thread__comm_str(parent), "sshd") ||3987		    strstarts(thread__comm_str(parent), "gnome-terminal")) {3988			pids[nr++] = thread__tid(parent);3989			break;3990		}3991		thread = parent;3992	}3993 3994	err = evlist__append_tp_filter_pids(trace->evlist, nr, pids);3995	if (!err && trace->filter_pids.map)3996		err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);3997 3998	return err;3999}4000 4001static int trace__set_filter_pids(struct trace *trace)4002{4003	int err = 0;4004	/*4005	 * Better not use !target__has_task() here because we need to cover the4006	 * case where no threads were specified in the command line, but a4007	 * workload was, and in that case we will fill in the thread_map when4008	 * we fork the workload in evlist__prepare_workload.4009	 */4010	if (trace->filter_pids.nr > 0) {4011		err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr,4012						    trace->filter_pids.entries);4013		if (!err && trace->filter_pids.map) {4014			err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,4015						       trace->filter_pids.entries);4016		}4017	} else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {4018		err = trace__set_filter_loop_pids(trace);4019	}4020 4021	return err;4022}4023 4024static int __trace__deliver_event(struct trace *trace, union perf_event *event)4025{4026	struct evlist *evlist = trace->evlist;4027	struct perf_sample sample;4028	int err = evlist__parse_sample(evlist, event, &sample);4029 4030	if (err)4031		fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);4032	else4033		trace__handle_event(trace, event, &sample);4034 4035	return 0;4036}4037 4038static int __trace__flush_events(struct trace *trace)4039{4040	u64 first = ordered_events__first_time(&trace->oe.data);4041	u64 flush = trace->oe.last - NSEC_PER_SEC;4042 4043	/* Is there some thing to flush.. */4044	if (first && first < flush)4045		return ordered_events__flush_time(&trace->oe.data, flush);4046 4047	return 0;4048}4049 4050static int trace__flush_events(struct trace *trace)4051{4052	return !trace->sort_events ? 0 : __trace__flush_events(trace);4053}4054 4055static int trace__deliver_event(struct trace *trace, union perf_event *event)4056{4057	int err;4058 4059	if (!trace->sort_events)4060		return __trace__deliver_event(trace, event);4061 4062	err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);4063	if (err && err != -1)4064		return err;4065 4066	err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0, NULL);4067	if (err)4068		return err;4069 4070	return trace__flush_events(trace);4071}4072 4073static int ordered_events__deliver_event(struct ordered_events *oe,4074					 struct ordered_event *event)4075{4076	struct trace *trace = container_of(oe, struct trace, oe.data);4077 4078	return __trace__deliver_event(trace, event->event);4079}4080 4081static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg,4082								   char **type)4083{4084	struct tep_format_field *field;4085	struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel);4086 4087	if (evsel->tp_format == NULL || fmt == NULL)4088		return NULL;4089 4090	for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt)4091		if (strcmp(field->name, arg) == 0) {4092			*type = field->type;4093			return fmt;4094		}4095 4096	return NULL;4097}4098 4099static int trace__expand_filter(struct trace *trace, struct evsel *evsel)4100{4101	char *tok, *left = evsel->filter, *new_filter = evsel->filter;4102 4103	while ((tok = strpbrk(left, "=<>!")) != NULL) {4104		char *right = tok + 1, *right_end;4105 4106		if (*right == '=')4107			++right;4108 4109		while (isspace(*right))4110			++right;4111 4112		if (*right == '\0')4113			break;4114 4115		while (!isalpha(*left))4116			if (++left == tok) {4117				/*4118				 * Bail out, can't find the name of the argument that is being4119				 * used in the filter, let it try to set this filter, will fail later.4120				 */4121				return 0;4122			}4123 4124		right_end = right + 1;4125		while (isalnum(*right_end) || *right_end == '_' || *right_end == '|')4126			++right_end;4127 4128		if (isalpha(*right)) {4129			struct syscall_arg_fmt *fmt;4130			int left_size = tok - left,4131			    right_size = right_end - right;4132			char arg[128], *type;4133 4134			while (isspace(left[left_size - 1]))4135				--left_size;4136 4137			scnprintf(arg, sizeof(arg), "%.*s", left_size, left);4138 4139			fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg, &type);4140			if (fmt == NULL) {4141				pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n",4142				       arg, evsel->name, evsel->filter);4143				return -1;4144			}4145 4146			pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ",4147				 arg, (int)(right - tok), tok, right_size, right);4148 4149			if (fmt->strtoul) {4150				u64 val;4151				struct syscall_arg syscall_arg = {4152					.trace = trace,4153					.fmt   = fmt,4154					.type_name = type,4155					.parm = fmt->parm,4156				};4157 4158				if (fmt->strtoul(right, right_size, &syscall_arg, &val)) {4159					char *n, expansion[19];4160					int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val);4161					int expansion_offset = right - new_filter;4162 4163					pr_debug("%s", expansion);4164 4165					if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) {4166						pr_debug(" out of memory!\n");4167						free(new_filter);4168						return -1;4169					}4170					if (new_filter != evsel->filter)4171						free(new_filter);4172					left = n + expansion_offset + expansion_lenght;4173					new_filter = n;4174				} else {4175					pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n",4176					       right_size, right, arg, evsel->name, evsel->filter);4177					return -1;4178				}4179			} else {4180				pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n",4181				       arg, evsel->name, evsel->filter);4182				return -1;4183			}4184 4185			pr_debug("\n");4186		} else {4187			left = right_end;4188		}4189	}4190 4191	if (new_filter != evsel->filter) {4192		pr_debug("New filter for %s: %s\n", evsel->name, new_filter);4193		evsel__set_filter(evsel, new_filter);4194		free(new_filter);4195	}4196 4197	return 0;4198}4199 4200static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel)4201{4202	struct evlist *evlist = trace->evlist;4203	struct evsel *evsel;4204 4205	evlist__for_each_entry(evlist, evsel) {4206		if (evsel->filter == NULL)4207			continue;4208 4209		if (trace__expand_filter(trace, evsel)) {4210			*err_evsel = evsel;4211			return -1;4212		}4213	}4214 4215	return 0;4216}4217 4218static int trace__run(struct trace *trace, int argc, const char **argv)4219{4220	struct evlist *evlist = trace->evlist;4221	struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;4222	int err = -1, i;4223	unsigned long before;4224	const bool forks = argc > 0;4225	bool draining = false;4226 4227	trace->live = true;4228 4229	if (!trace->raw_augmented_syscalls) {4230		if (trace->trace_syscalls && trace__add_syscall_newtp(trace))4231			goto out_error_raw_syscalls;4232 4233		if (trace->trace_syscalls)4234			trace->vfs_getname = evlist__add_vfs_getname(evlist);4235	}4236 4237	if ((trace->trace_pgfaults & TRACE_PFMAJ)) {4238		pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);4239		if (pgfault_maj == NULL)4240			goto out_error_mem;4241		evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);4242		evlist__add(evlist, pgfault_maj);4243	}4244 4245	if ((trace->trace_pgfaults & TRACE_PFMIN)) {4246		pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);4247		if (pgfault_min == NULL)4248			goto out_error_mem;4249		evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);4250		evlist__add(evlist, pgfault_min);4251	}4252 4253	/* Enable ignoring missing threads when -u/-p option is defined. */4254	trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid;4255 4256	if (trace->sched &&4257	    evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime))4258		goto out_error_sched_stat_runtime;4259	/*4260	 * If a global cgroup was set, apply it to all the events without an4261	 * explicit cgroup. I.e.:4262	 *4263	 * 	trace -G A -e sched:*switch4264	 *4265	 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc4266	 * _and_ sched:sched_switch to the 'A' cgroup, while:4267	 *4268	 * trace -e sched:*switch -G A4269	 *4270	 * will only set the sched:sched_switch event to the 'A' cgroup, all the4271	 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"4272	 * a cgroup (on the root cgroup, sys wide, etc).4273	 *4274	 * Multiple cgroups:4275	 *4276	 * trace -G A -e sched:*switch -G B4277	 *4278	 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes4279	 * to the 'B' cgroup.4280	 *4281	 * evlist__set_default_cgroup() grabs a reference of the passed cgroup4282	 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.4283	 */4284	if (trace->cgroup)4285		evlist__set_default_cgroup(trace->evlist, trace->cgroup);4286 4287	err = evlist__create_maps(evlist, &trace->opts.target);4288	if (err < 0) {4289		fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");4290		goto out_delete_evlist;4291	}4292 4293	err = trace__symbols_init(trace, evlist);4294	if (err < 0) {4295		fprintf(trace->output, "Problems initializing symbol libraries!\n");4296		goto out_delete_evlist;4297	}4298 4299	evlist__config(evlist, &trace->opts, &callchain_param);4300 4301	if (forks) {4302		err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL);4303		if (err < 0) {4304			fprintf(trace->output, "Couldn't run the workload!\n");4305			goto out_delete_evlist;4306		}4307		workload_pid = evlist->workload.pid;4308	}4309 4310	err = evlist__open(evlist);4311	if (err < 0)4312		goto out_error_open;4313#ifdef HAVE_BPF_SKEL4314	if (trace->syscalls.events.bpf_output) {4315		struct perf_cpu cpu;4316 4317		/*4318		 * Set up the __augmented_syscalls__ BPF map to hold for each4319		 * CPU the bpf-output event's file descriptor.4320		 */4321		perf_cpu_map__for_each_cpu(cpu, i, trace->syscalls.events.bpf_output->core.cpus) {4322			bpf_map__update_elem(trace->skel->maps.__augmented_syscalls__,4323					&cpu.cpu, sizeof(int),4324					xyarray__entry(trace->syscalls.events.bpf_output->core.fd,4325						       cpu.cpu, 0),4326					sizeof(__u32), BPF_ANY);4327		}4328	}4329#endif4330	err = trace__set_filter_pids(trace);4331	if (err < 0)4332		goto out_error_mem;4333 4334#ifdef HAVE_BPF_SKEL4335	if (trace->skel && trace->skel->progs.sys_enter)4336		trace__init_syscalls_bpf_prog_array_maps(trace);4337#endif4338 4339	if (trace->ev_qualifier_ids.nr > 0) {4340		err = trace__set_ev_qualifier_filter(trace);4341		if (err < 0)4342			goto out_errno;4343 4344		if (trace->syscalls.events.sys_exit) {4345			pr_debug("event qualifier tracepoint filter: %s\n",4346				 trace->syscalls.events.sys_exit->filter);4347		}4348	}4349 4350	/*4351	 * If the "close" syscall is not traced, then we will not have the4352	 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the4353	 * fd->pathname table and were ending up showing the last value set by4354	 * syscalls opening a pathname and associating it with a descriptor or4355	 * reading it from /proc/pid/fd/ in cases where that doesn't make4356	 * sense.4357	 *4358	 *  So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is4359	 *  not in use.4360	 */4361	trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));4362 4363	err = trace__expand_filters(trace, &evsel);4364	if (err)4365		goto out_delete_evlist;4366	err = evlist__apply_filters(evlist, &evsel, &trace->opts.target);4367	if (err < 0)4368		goto out_error_apply_filters;4369 4370	err = evlist__mmap(evlist, trace->opts.mmap_pages);4371	if (err < 0)4372		goto out_error_mmap;4373 4374	if (!target__none(&trace->opts.target) && !trace->opts.target.initial_delay)4375		evlist__enable(evlist);4376 4377	if (forks)4378		evlist__start_workload(evlist);4379 4380	if (trace->opts.target.initial_delay) {4381		usleep(trace->opts.target.initial_delay * 1000);4382		evlist__enable(evlist);4383	}4384 4385	trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||4386		perf_thread_map__nr(evlist->core.threads) > 1 ||4387		evlist__first(evlist)->core.attr.inherit;4388 4389	/*4390	 * Now that we already used evsel->core.attr to ask the kernel to setup the4391	 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in4392	 * trace__resolve_callchain(), allowing per-event max-stack settings4393	 * to override an explicitly set --max-stack global setting.4394	 */4395	evlist__for_each_entry(evlist, evsel) {4396		if (evsel__has_callchain(evsel) &&4397		    evsel->core.attr.sample_max_stack == 0)4398			evsel->core.attr.sample_max_stack = trace->max_stack;4399	}4400again:4401	before = trace->nr_events;4402 4403	for (i = 0; i < evlist->core.nr_mmaps; i++) {4404		union perf_event *event;4405		struct mmap *md;4406 4407		md = &evlist->mmap[i];4408		if (perf_mmap__read_init(&md->core) < 0)4409			continue;4410 4411		while ((event = perf_mmap__read_event(&md->core)) != NULL) {4412			++trace->nr_events;4413 4414			err = trace__deliver_event(trace, event);4415			if (err)4416				goto out_disable;4417 4418			perf_mmap__consume(&md->core);4419 4420			if (interrupted)4421				goto out_disable;4422 4423			if (done && !draining) {4424				evlist__disable(evlist);4425				draining = true;4426			}4427		}4428		perf_mmap__read_done(&md->core);4429	}4430 4431	if (trace->nr_events == before) {4432		int timeout = done ? 100 : -1;4433 4434		if (!draining && evlist__poll(evlist, timeout) > 0) {4435			if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)4436				draining = true;4437 4438			goto again;4439		} else {4440			if (trace__flush_events(trace))4441				goto out_disable;4442		}4443	} else {4444		goto again;4445	}4446 4447out_disable:4448	thread__zput(trace->current);4449 4450	evlist__disable(evlist);4451 4452	if (trace->sort_events)4453		ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);4454 4455	if (!err) {4456		if (trace->summary)4457			trace__fprintf_thread_summary(trace, trace->output);4458 4459		if (trace->show_tool_stats) {4460			fprintf(trace->output, "Stats:\n "4461					       " vfs_getname : %" PRIu64 "\n"4462					       " proc_getname: %" PRIu64 "\n",4463				trace->stats.vfs_getname,4464				trace->stats.proc_getname);4465		}4466	}4467 4468out_delete_evlist:4469	trace__symbols__exit(trace);4470	evlist__free_syscall_tp_fields(evlist);4471	evlist__delete(evlist);4472	cgroup__put(trace->cgroup);4473	trace->evlist = NULL;4474	trace->live = false;4475	return err;4476{4477	char errbuf[BUFSIZ];4478 4479out_error_sched_stat_runtime:4480	tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");4481	goto out_error;4482 4483out_error_raw_syscalls:4484	tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");4485	goto out_error;4486 4487out_error_mmap:4488	evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));4489	goto out_error;4490 4491out_error_open:4492	evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));4493 4494out_error:4495	fprintf(trace->output, "%s\n", errbuf);4496	goto out_delete_evlist;4497 4498out_error_apply_filters:4499	fprintf(trace->output,4500		"Failed to set filter \"%s\" on event %s with %d (%s)\n",4501		evsel->filter, evsel__name(evsel), errno,4502		str_error_r(errno, errbuf, sizeof(errbuf)));4503	goto out_delete_evlist;4504}4505out_error_mem:4506	fprintf(trace->output, "Not enough memory to run!\n");4507	goto out_delete_evlist;4508 4509out_errno:4510	fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));4511	goto out_delete_evlist;4512}4513 4514static int trace__replay(struct trace *trace)4515{4516	const struct evsel_str_handler handlers[] = {4517		{ "probe:vfs_getname",	     trace__vfs_getname, },4518	};4519	struct perf_data data = {4520		.path  = input_name,4521		.mode  = PERF_DATA_MODE_READ,4522		.force = trace->force,4523	};4524	struct perf_session *session;4525	struct evsel *evsel;4526	int err = -1;4527 4528	trace->tool.sample	  = trace__process_sample;4529	trace->tool.mmap	  = perf_event__process_mmap;4530	trace->tool.mmap2	  = perf_event__process_mmap2;4531	trace->tool.comm	  = perf_event__process_comm;4532	trace->tool.exit	  = perf_event__process_exit;4533	trace->tool.fork	  = perf_event__process_fork;4534	trace->tool.attr	  = perf_event__process_attr;4535	trace->tool.tracing_data  = perf_event__process_tracing_data;4536	trace->tool.build_id	  = perf_event__process_build_id;4537	trace->tool.namespaces	  = perf_event__process_namespaces;4538 4539	trace->tool.ordered_events = true;4540	trace->tool.ordering_requires_timestamps = true;4541 4542	/* add tid to output */4543	trace->multiple_threads = true;4544 4545	session = perf_session__new(&data, &trace->tool);4546	if (IS_ERR(session))4547		return PTR_ERR(session);4548 4549	if (trace->opts.target.pid)4550		symbol_conf.pid_list_str = strdup(trace->opts.target.pid);4551 4552	if (trace->opts.target.tid)4553		symbol_conf.tid_list_str = strdup(trace->opts.target.tid);4554 4555	if (symbol__init(&session->header.env) < 0)4556		goto out;4557 4558	trace->host = &session->machines.host;4559 4560	err = perf_session__set_tracepoints_handlers(session, handlers);4561	if (err)4562		goto out;4563 4564	evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter");4565	trace->syscalls.events.sys_enter = evsel;4566	/* older kernels have syscalls tp versus raw_syscalls */4567	if (evsel == NULL)4568		evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter");4569 4570	if (evsel &&4571	    (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||4572	    perf_evsel__init_sc_tp_ptr_field(evsel, args))) {4573		pr_err("Error during initialize raw_syscalls:sys_enter event\n");4574		goto out;4575	}4576 4577	evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit");4578	trace->syscalls.events.sys_exit = evsel;4579	if (evsel == NULL)4580		evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit");4581	if (evsel &&4582	    (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||4583	    perf_evsel__init_sc_tp_uint_field(evsel, ret))) {4584		pr_err("Error during initialize raw_syscalls:sys_exit event\n");4585		goto out;4586	}4587 4588	evlist__for_each_entry(session->evlist, evsel) {4589		if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&4590		    (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||4591		     evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||4592		     evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))4593			evsel->handler = trace__pgfault;4594	}4595 4596	setup_pager();4597 4598	err = perf_session__process_events(session);4599	if (err)4600		pr_err("Failed to process events, error %d", err);4601 4602	else if (trace->summary)4603		trace__fprintf_thread_summary(trace, trace->output);4604 4605out:4606	perf_session__delete(session);4607 4608	return err;4609}4610 4611static size_t trace__fprintf_threads_header(FILE *fp)4612{4613	size_t printed;4614 4615	printed  = fprintf(fp, "\n Summary of events:\n\n");4616 4617	return printed;4618}4619 4620DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,4621	struct syscall_stats *stats;4622	double		     msecs;4623	int		     syscall;4624)4625{4626	struct int_node *source = rb_entry(nd, struct int_node, rb_node);4627	struct syscall_stats *stats = source->priv;4628 4629	entry->syscall = source->i;4630	entry->stats   = stats;4631	entry->msecs   = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0;4632}4633 4634static size_t thread__dump_stats(struct thread_trace *ttrace,4635				 struct trace *trace, FILE *fp)4636{4637	size_t printed = 0;4638	struct syscall *sc;4639	struct rb_node *nd;4640	DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);4641 4642	if (syscall_stats == NULL)4643		return 0;4644 4645	printed += fprintf(fp, "\n");4646 4647	printed += fprintf(fp, "   syscall            calls  errors  total       min       avg       max       stddev\n");4648	printed += fprintf(fp, "                                     (msec)    (msec)    (msec)    (msec)        (%%)\n");4649	printed += fprintf(fp, "   --------------- --------  ------ -------- --------- --------- ---------     ------\n");4650 4651	resort_rb__for_each_entry(nd, syscall_stats) {4652		struct syscall_stats *stats = syscall_stats_entry->stats;4653		if (stats) {4654			double min = (double)(stats->stats.min) / NSEC_PER_MSEC;4655			double max = (double)(stats->stats.max) / NSEC_PER_MSEC;4656			double avg = avg_stats(&stats->stats);4657			double pct;4658			u64 n = (u64)stats->stats.n;4659 4660			pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0;4661			avg /= NSEC_PER_MSEC;4662 4663			sc = &trace->syscalls.table[syscall_stats_entry->syscall];4664			printed += fprintf(fp, "   %-15s", sc->name);4665			printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f",4666					   n, stats->nr_failures, syscall_stats_entry->msecs, min, avg);4667			printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);4668 4669			if (trace->errno_summary && stats->nr_failures) {4670				int e;4671 4672				for (e = 0; e < stats->max_errno; ++e) {4673					if (stats->errnos[e] != 0)4674						fprintf(fp, "\t\t\t\t%s: %d\n", perf_env__arch_strerrno(trace->host->env, e + 1), stats->errnos[e]);4675				}4676			}4677		}4678	}4679 4680	resort_rb__delete(syscall_stats);4681	printed += fprintf(fp, "\n\n");4682 4683	return printed;4684}4685 4686static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)4687{4688	size_t printed = 0;4689	struct thread_trace *ttrace = thread__priv(thread);4690	double ratio;4691 4692	if (ttrace == NULL)4693		return 0;4694 4695	ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;4696 4697	printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread__tid(thread));4698	printed += fprintf(fp, "%lu events, ", ttrace->nr_events);4699	printed += fprintf(fp, "%.1f%%", ratio);4700	if (ttrace->pfmaj)4701		printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);4702	if (ttrace->pfmin)4703		printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);4704	if (trace->sched)4705		printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);4706	else if (fputc('\n', fp) != EOF)4707		++printed;4708 4709	printed += thread__dump_stats(ttrace, trace, fp);4710 4711	return printed;4712}4713 4714static unsigned long thread__nr_events(struct thread_trace *ttrace)4715{4716	return ttrace ? ttrace->nr_events : 0;4717}4718 4719static int trace_nr_events_cmp(void *priv __maybe_unused,4720			       const struct list_head *la,4721			       const struct list_head *lb)4722{4723	struct thread_list *a = list_entry(la, struct thread_list, list);4724	struct thread_list *b = list_entry(lb, struct thread_list, list);4725	unsigned long a_nr_events = thread__nr_events(thread__priv(a->thread));4726	unsigned long b_nr_events = thread__nr_events(thread__priv(b->thread));4727 4728	if (a_nr_events != b_nr_events)4729		return a_nr_events < b_nr_events ? -1 : 1;4730 4731	/* Identical number of threads, place smaller tids first. */4732	return thread__tid(a->thread) < thread__tid(b->thread)4733		? -14734		: (thread__tid(a->thread) > thread__tid(b->thread) ? 1 : 0);4735}4736 4737static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)4738{4739	size_t printed = trace__fprintf_threads_header(fp);4740	LIST_HEAD(threads);4741 4742	if (machine__thread_list(trace->host, &threads) == 0) {4743		struct thread_list *pos;4744 4745		list_sort(NULL, &threads, trace_nr_events_cmp);4746 4747		list_for_each_entry(pos, &threads, list)4748			printed += trace__fprintf_thread(fp, pos->thread, trace);4749	}4750	thread_list__delete(&threads);4751	return printed;4752}4753 4754static int trace__set_duration(const struct option *opt, const char *str,4755			       int unset __maybe_unused)4756{4757	struct trace *trace = opt->value;4758 4759	trace->duration_filter = atof(str);4760	return 0;4761}4762 4763static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,4764					      int unset __maybe_unused)4765{4766	int ret = -1;4767	size_t i;4768	struct trace *trace = opt->value;4769	/*4770	 * FIXME: introduce a intarray class, plain parse csv and create a4771	 * { int nr, int entries[] } struct...4772	 */4773	struct intlist *list = intlist__new(str);4774 4775	if (list == NULL)4776		return -1;4777 4778	i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;4779	trace->filter_pids.entries = calloc(i, sizeof(pid_t));4780 4781	if (trace->filter_pids.entries == NULL)4782		goto out;4783 4784	trace->filter_pids.entries[0] = getpid();4785 4786	for (i = 1; i < trace->filter_pids.nr; ++i)4787		trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;4788 4789	intlist__delete(list);4790	ret = 0;4791out:4792	return ret;4793}4794 4795static int trace__open_output(struct trace *trace, const char *filename)4796{4797	struct stat st;4798 4799	if (!stat(filename, &st) && st.st_size) {4800		char oldname[PATH_MAX];4801 4802		scnprintf(oldname, sizeof(oldname), "%s.old", filename);4803		unlink(oldname);4804		rename(filename, oldname);4805	}4806 4807	trace->output = fopen(filename, "w");4808 4809	return trace->output == NULL ? -errno : 0;4810}4811 4812static int parse_pagefaults(const struct option *opt, const char *str,4813			    int unset __maybe_unused)4814{4815	int *trace_pgfaults = opt->value;4816 4817	if (strcmp(str, "all") == 0)4818		*trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;4819	else if (strcmp(str, "maj") == 0)4820		*trace_pgfaults |= TRACE_PFMAJ;4821	else if (strcmp(str, "min") == 0)4822		*trace_pgfaults |= TRACE_PFMIN;4823	else4824		return -1;4825 4826	return 0;4827}4828 4829static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)4830{4831	struct evsel *evsel;4832 4833	evlist__for_each_entry(evlist, evsel) {4834		if (evsel->handler == NULL)4835			evsel->handler = handler;4836	}4837}4838 4839static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name)4840{4841	struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);4842 4843	if (fmt) {4844		const struct syscall_fmt *scfmt = syscall_fmt__find(name);4845 4846		if (scfmt) {4847			int skip = 0;4848 4849			if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 ||4850			    strcmp(evsel->tp_format->format.fields->name, "nr") == 0)4851				++skip;4852 4853			memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt));4854		}4855	}4856}4857 4858static int evlist__set_syscall_tp_fields(struct evlist *evlist, bool *use_btf)4859{4860	struct evsel *evsel;4861 4862	evlist__for_each_entry(evlist, evsel) {4863		if (evsel->priv || !evsel->tp_format)4864			continue;4865 4866		if (strcmp(evsel->tp_format->system, "syscalls")) {4867			evsel__init_tp_arg_scnprintf(evsel, use_btf);4868			continue;4869		}4870 4871		if (evsel__init_syscall_tp(evsel))4872			return -1;4873 4874		if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {4875			struct syscall_tp *sc = __evsel__syscall_tp(evsel);4876 4877			if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))4878				return -1;4879 4880			evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1);4881		} else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {4882			struct syscall_tp *sc = __evsel__syscall_tp(evsel);4883 4884			if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))4885				return -1;4886 4887			evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1);4888		}4889	}4890 4891	return 0;4892}4893 4894/*4895 * XXX: Hackish, just splitting the combined -e+--event (syscalls4896 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use4897 * existing facilities unchanged (trace->ev_qualifier + parse_options()).4898 *4899 * It'd be better to introduce a parse_options() variant that would return a4900 * list with the terms it didn't match to an event...4901 */4902static int trace__parse_events_option(const struct option *opt, const char *str,4903				      int unset __maybe_unused)4904{4905	struct trace *trace = (struct trace *)opt->value;4906	const char *s = str;4907	char *sep = NULL, *lists[2] = { NULL, NULL, };4908	int len = strlen(str) + 1, err = -1, list, idx;4909	char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);4910	char group_name[PATH_MAX];4911	const struct syscall_fmt *fmt;4912 4913	if (strace_groups_dir == NULL)4914		return -1;4915 4916	if (*s == '!') {4917		++s;4918		trace->not_ev_qualifier = true;4919	}4920 4921	while (1) {4922		if ((sep = strchr(s, ',')) != NULL)4923			*sep = '\0';4924 4925		list = 0;4926		if (syscalltbl__id(trace->sctbl, s) >= 0 ||4927		    syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {4928			list = 1;4929			goto do_concat;4930		}4931 4932		fmt = syscall_fmt__find_by_alias(s);4933		if (fmt != NULL) {4934			list = 1;4935			s = fmt->name;4936		} else {4937			path__join(group_name, sizeof(group_name), strace_groups_dir, s);4938			if (access(group_name, R_OK) == 0)4939				list = 1;4940		}4941do_concat:4942		if (lists[list]) {4943			sprintf(lists[list] + strlen(lists[list]), ",%s", s);4944		} else {4945			lists[list] = malloc(len);4946			if (lists[list] == NULL)4947				goto out;4948			strcpy(lists[list], s);4949		}4950 4951		if (!sep)4952			break;4953 4954		*sep = ',';4955		s = sep + 1;4956	}4957 4958	if (lists[1] != NULL) {4959		struct strlist_config slist_config = {4960			.dirname = strace_groups_dir,4961		};4962 4963		trace->ev_qualifier = strlist__new(lists[1], &slist_config);4964		if (trace->ev_qualifier == NULL) {4965			fputs("Not enough memory to parse event qualifier", trace->output);4966			goto out;4967		}4968 4969		if (trace__validate_ev_qualifier(trace))4970			goto out;4971		trace->trace_syscalls = true;4972	}4973 4974	err = 0;4975 4976	if (lists[0]) {4977		struct parse_events_option_args parse_events_option_args = {4978			.evlistp = &trace->evlist,4979		};4980		struct option o = {4981			.value = &parse_events_option_args,4982		};4983		err = parse_events_option(&o, lists[0], 0);4984	}4985out:4986	free(strace_groups_dir);4987	free(lists[0]);4988	free(lists[1]);4989	if (sep)4990		*sep = ',';4991 4992	return err;4993}4994 4995static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)4996{4997	struct trace *trace = opt->value;4998 4999	if (!list_empty(&trace->evlist->core.entries)) {5000		struct option o = {5001			.value = &trace->evlist,5002		};5003		return parse_cgroups(&o, str, unset);5004	}5005	trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);5006 5007	return 0;5008}5009 5010static int trace__config(const char *var, const char *value, void *arg)5011{5012	struct trace *trace = arg;5013	int err = 0;5014 5015	if (!strcmp(var, "trace.add_events")) {5016		trace->perfconfig_events = strdup(value);5017		if (trace->perfconfig_events == NULL) {5018			pr_err("Not enough memory for %s\n", "trace.add_events");5019			return -1;5020		}5021	} else if (!strcmp(var, "trace.show_timestamp")) {5022		trace->show_tstamp = perf_config_bool(var, value);5023	} else if (!strcmp(var, "trace.show_duration")) {5024		trace->show_duration = perf_config_bool(var, value);5025	} else if (!strcmp(var, "trace.show_arg_names")) {5026		trace->show_arg_names = perf_config_bool(var, value);5027		if (!trace->show_arg_names)5028			trace->show_zeros = true;5029	} else if (!strcmp(var, "trace.show_zeros")) {5030		bool new_show_zeros = perf_config_bool(var, value);5031		if (!trace->show_arg_names && !new_show_zeros) {5032			pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");5033			goto out;5034		}5035		trace->show_zeros = new_show_zeros;5036	} else if (!strcmp(var, "trace.show_prefix")) {5037		trace->show_string_prefix = perf_config_bool(var, value);5038	} else if (!strcmp(var, "trace.no_inherit")) {5039		trace->opts.no_inherit = perf_config_bool(var, value);5040	} else if (!strcmp(var, "trace.args_alignment")) {5041		int args_alignment = 0;5042		if (perf_config_int(&args_alignment, var, value) == 0)5043			trace->args_alignment = args_alignment;5044	} else if (!strcmp(var, "trace.tracepoint_beautifiers")) {5045		if (strcasecmp(value, "libtraceevent") == 0)5046			trace->libtraceevent_print = true;5047		else if (strcasecmp(value, "libbeauty") == 0)5048			trace->libtraceevent_print = false;5049	}5050out:5051	return err;5052}5053 5054static void trace__exit(struct trace *trace)5055{5056	int i;5057 5058	strlist__delete(trace->ev_qualifier);5059	zfree(&trace->ev_qualifier_ids.entries);5060	if (trace->syscalls.table) {5061		for (i = 0; i <= trace->sctbl->syscalls.max_id; i++)5062			syscall__exit(&trace->syscalls.table[i]);5063		zfree(&trace->syscalls.table);5064	}5065	syscalltbl__delete(trace->sctbl);5066	zfree(&trace->perfconfig_events);5067}5068 5069#ifdef HAVE_BPF_SKEL5070static int bpf__setup_bpf_output(struct evlist *evlist)5071{5072	int err = parse_event(evlist, "bpf-output/no-inherit=1,name=__augmented_syscalls__/");5073 5074	if (err)5075		pr_debug("ERROR: failed to create the \"__augmented_syscalls__\" bpf-output event\n");5076 5077	return err;5078}5079#endif5080 5081int cmd_trace(int argc, const char **argv)5082{5083	const char *trace_usage[] = {5084		"perf trace [<options>] [<command>]",5085		"perf trace [<options>] -- <command> [<options>]",5086		"perf trace record [<options>] [<command>]",5087		"perf trace record [<options>] -- <command> [<options>]",5088		NULL5089	};5090	struct trace trace = {5091		.opts = {5092			.target = {5093				.uid	   = UINT_MAX,5094				.uses_mmap = true,5095			},5096			.user_freq     = UINT_MAX,5097			.user_interval = ULLONG_MAX,5098			.no_buffering  = true,5099			.mmap_pages    = UINT_MAX,5100		},5101		.output = stderr,5102		.show_comm = true,5103		.show_tstamp = true,5104		.show_duration = true,5105		.show_arg_names = true,5106		.args_alignment = 70,5107		.trace_syscalls = false,5108		.kernel_syscallchains = false,5109		.max_stack = UINT_MAX,5110		.max_events = ULONG_MAX,5111	};5112	const char *output_name = NULL;5113	const struct option trace_options[] = {5114	OPT_CALLBACK('e', "event", &trace, "event",5115		     "event/syscall selector. use 'perf list' to list available events",5116		     trace__parse_events_option),5117	OPT_CALLBACK(0, "filter", &trace.evlist, "filter",5118		     "event filter", parse_filter),5119	OPT_BOOLEAN(0, "comm", &trace.show_comm,5120		    "show the thread COMM next to its id"),5121	OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),5122	OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",5123		     trace__parse_events_option),5124	OPT_STRING('o', "output", &output_name, "file", "output file name"),5125	OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),5126	OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",5127		    "trace events on existing process id"),5128	OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",5129		    "trace events on existing thread id"),5130	OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",5131		     "pids to filter (by the kernel)", trace__set_filter_pids_from_option),5132	OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,5133		    "system-wide collection from all CPUs"),5134	OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",5135		    "list of cpus to monitor"),5136	OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,5137		    "child tasks do not inherit counters"),5138	OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",5139		     "number of mmap data pages", evlist__parse_mmap_pages),5140	OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",5141		   "user to profile"),5142	OPT_CALLBACK(0, "duration", &trace, "float",5143		     "show only events with duration > N.M ms",5144		     trace__set_duration),5145	OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),5146	OPT_INCR('v', "verbose", &verbose, "be more verbose"),5147	OPT_BOOLEAN('T', "time", &trace.full_time,5148		    "Show full timestamp, not time relative to first start"),5149	OPT_BOOLEAN(0, "failure", &trace.failure_only,5150		    "Show only syscalls that failed"),5151	OPT_BOOLEAN('s', "summary", &trace.summary_only,5152		    "Show only syscall summary with statistics"),5153	OPT_BOOLEAN('S', "with-summary", &trace.summary,5154		    "Show all syscalls and summary with statistics"),5155	OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary,5156		    "Show errno stats per syscall, use with -s or -S"),5157	OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",5158		     "Trace pagefaults", parse_pagefaults, "maj"),5159	OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),5160	OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),5161	OPT_CALLBACK(0, "call-graph", &trace.opts,5162		     "record_mode[,record_size]", record_callchain_help,5163		     &record_parse_callchain_opt),5164	OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,5165		    "Use libtraceevent to print the tracepoint arguments."),5166	OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,5167		    "Show the kernel callchains on the syscall exit path"),5168	OPT_ULONG(0, "max-events", &trace.max_events,5169		"Set the maximum number of events to print, exit after that is reached. "),5170	OPT_UINTEGER(0, "min-stack", &trace.min_stack,5171		     "Set the minimum stack depth when parsing the callchain, "5172		     "anything below the specified depth will be ignored."),5173	OPT_UINTEGER(0, "max-stack", &trace.max_stack,5174		     "Set the maximum stack depth when parsing the callchain, "5175		     "anything beyond the specified depth will be ignored. "5176		     "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),5177	OPT_BOOLEAN(0, "sort-events", &trace.sort_events,5178			"Sort batch of events before processing, use if getting out of order events"),5179	OPT_BOOLEAN(0, "print-sample", &trace.print_sample,5180			"print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),5181	OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,5182			"per thread proc mmap processing timeout in ms"),5183	OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",5184		     trace__parse_cgroups),5185	OPT_INTEGER('D', "delay", &trace.opts.target.initial_delay,5186		     "ms to wait before starting measurement after program "5187		     "start"),5188	OPT_BOOLEAN(0, "force-btf", &trace.force_btf, "Prefer btf_dump general pretty printer"5189		       "to customized ones"),5190	OPTS_EVSWITCH(&trace.evswitch),5191	OPT_END()5192	};5193	bool __maybe_unused max_stack_user_set = true;5194	bool mmap_pages_user_set = true;5195	struct evsel *evsel;5196	const char * const trace_subcommands[] = { "record", NULL };5197	int err = -1;5198	char bf[BUFSIZ];5199	struct sigaction sigchld_act;5200 5201	signal(SIGSEGV, sighandler_dump_stack);5202	signal(SIGFPE, sighandler_dump_stack);5203	signal(SIGINT, sighandler_interrupt);5204 5205	memset(&sigchld_act, 0, sizeof(sigchld_act));5206	sigchld_act.sa_flags = SA_SIGINFO;5207	sigchld_act.sa_sigaction = sighandler_chld;5208	sigaction(SIGCHLD, &sigchld_act, NULL);5209 5210	trace.evlist = evlist__new();5211	trace.sctbl = syscalltbl__new();5212 5213	if (trace.evlist == NULL || trace.sctbl == NULL) {5214		pr_err("Not enough memory to run!\n");5215		err = -ENOMEM;5216		goto out;5217	}5218 5219	/*5220	 * Parsing .perfconfig may entail creating a BPF event, that may need5221	 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting5222	 * is too small. This affects just this process, not touching the5223	 * global setting. If it fails we'll get something in 'perf trace -v'5224	 * to help diagnose the problem.5225	 */5226	rlimit__bump_memlock();5227 5228	err = perf_config(trace__config, &trace);5229	if (err)5230		goto out;5231 5232	argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,5233				 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);5234 5235	/*5236	 * Here we already passed thru trace__parse_events_option() and it has5237	 * already figured out if -e syscall_name, if not but if --event5238	 * foo:bar was used, the user is interested _just_ in those, say,5239	 * tracepoint events, not in the strace-like syscall-name-based mode.5240	 *5241	 * This is important because we need to check if strace-like mode is5242	 * needed to decided if we should filter out the eBPF5243	 * __augmented_syscalls__ code, if it is in the mix, say, via5244	 * .perfconfig trace.add_events, and filter those out.5245	 */5246	if (!trace.trace_syscalls && !trace.trace_pgfaults &&5247	    trace.evlist->core.nr_entries == 0 /* Was --events used? */) {5248		trace.trace_syscalls = true;5249	}5250	/*5251	 * Now that we have --verbose figured out, lets see if we need to parse5252	 * events from .perfconfig, so that if those events fail parsing, say some5253	 * BPF program fails, then we'll be able to use --verbose to see what went5254	 * wrong in more detail.5255	 */5256	if (trace.perfconfig_events != NULL) {5257		struct parse_events_error parse_err;5258 5259		parse_events_error__init(&parse_err);5260		err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);5261		if (err)5262			parse_events_error__print(&parse_err, trace.perfconfig_events);5263		parse_events_error__exit(&parse_err);5264		if (err)5265			goto out;5266	}5267 5268	if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {5269		usage_with_options_msg(trace_usage, trace_options,5270				       "cgroup monitoring only available in system-wide mode");5271	}5272 5273#ifdef HAVE_BPF_SKEL5274	if (!trace.trace_syscalls)5275		goto skip_augmentation;5276 5277	if ((argc >= 1) && (strcmp(argv[0], "record") == 0)) {5278		pr_debug("Syscall augmentation fails with record, disabling augmentation");5279		goto skip_augmentation;5280	}5281 5282	trace.skel = augmented_raw_syscalls_bpf__open();5283	if (!trace.skel) {5284		pr_debug("Failed to open augmented syscalls BPF skeleton");5285	} else {5286		/*5287		 * Disable attaching the BPF programs except for sys_enter and5288		 * sys_exit that tail call into this as necessary.5289		 */5290		struct bpf_program *prog;5291 5292		bpf_object__for_each_program(prog, trace.skel->obj) {5293			if (prog != trace.skel->progs.sys_enter && prog != trace.skel->progs.sys_exit)5294				bpf_program__set_autoattach(prog, /*autoattach=*/false);5295		}5296 5297		err = augmented_raw_syscalls_bpf__load(trace.skel);5298 5299		if (err < 0) {5300			libbpf_strerror(err, bf, sizeof(bf));5301			pr_debug("Failed to load augmented syscalls BPF skeleton: %s\n", bf);5302		} else {5303			augmented_raw_syscalls_bpf__attach(trace.skel);5304			trace__add_syscall_newtp(&trace);5305		}5306	}5307 5308	err = bpf__setup_bpf_output(trace.evlist);5309	if (err) {5310		libbpf_strerror(err, bf, sizeof(bf));5311		pr_err("ERROR: Setup BPF output event failed: %s\n", bf);5312		goto out;5313	}5314	trace.syscalls.events.bpf_output = evlist__last(trace.evlist);5315	assert(evsel__name_is(trace.syscalls.events.bpf_output, "__augmented_syscalls__"));5316skip_augmentation:5317#endif5318	err = -1;5319 5320	if (trace.trace_pgfaults) {5321		trace.opts.sample_address = true;5322		trace.opts.sample_time = true;5323	}5324 5325	if (trace.opts.mmap_pages == UINT_MAX)5326		mmap_pages_user_set = false;5327 5328	if (trace.max_stack == UINT_MAX) {5329		trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();5330		max_stack_user_set = false;5331	}5332 5333#ifdef HAVE_DWARF_UNWIND_SUPPORT5334	if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {5335		record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);5336	}5337#endif5338 5339	if (callchain_param.enabled) {5340		if (!mmap_pages_user_set && geteuid() == 0)5341			trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;5342 5343		symbol_conf.use_callchain = true;5344	}5345 5346	if (trace.evlist->core.nr_entries > 0) {5347		bool use_btf = false;5348 5349		evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);5350		if (evlist__set_syscall_tp_fields(trace.evlist, &use_btf)) {5351			perror("failed to set syscalls:* tracepoint fields");5352			goto out;5353		}5354 5355		if (use_btf)5356			trace__load_vmlinux_btf(&trace);5357	}5358 5359	if (trace.sort_events) {5360		ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);5361		ordered_events__set_copy_on_queue(&trace.oe.data, true);5362	}5363 5364	/*5365	 * If we are augmenting syscalls, then combine what we put in the5366	 * __augmented_syscalls__ BPF map with what is in the5367	 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,5368	 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.5369	 *5370	 * We'll switch to look at two BPF maps, one for sys_enter and the5371	 * other for sys_exit when we start augmenting the sys_exit paths with5372	 * buffers that are being copied from kernel to userspace, think 'read'5373	 * syscall.5374	 */5375	if (trace.syscalls.events.bpf_output) {5376		evlist__for_each_entry(trace.evlist, evsel) {5377			bool raw_syscalls_sys_exit = evsel__name_is(evsel, "raw_syscalls:sys_exit");5378 5379			if (raw_syscalls_sys_exit) {5380				trace.raw_augmented_syscalls = true;5381				goto init_augmented_syscall_tp;5382			}5383 5384			if (trace.syscalls.events.bpf_output->priv == NULL &&5385			    strstr(evsel__name(evsel), "syscalls:sys_enter")) {5386				struct evsel *augmented = trace.syscalls.events.bpf_output;5387				if (evsel__init_augmented_syscall_tp(augmented, evsel) ||5388				    evsel__init_augmented_syscall_tp_args(augmented))5389					goto out;5390				/*5391				 * Augmented is __augmented_syscalls__ BPF_OUTPUT event5392				 * Above we made sure we can get from the payload the tp fields5393				 * that we get from syscalls:sys_enter tracefs format file.5394				 */5395				augmented->handler = trace__sys_enter;5396				/*5397				 * Now we do the same for the *syscalls:sys_enter event so that5398				 * if we handle it directly, i.e. if the BPF prog returns 0 so5399				 * as not to filter it, then we'll handle it just like we would5400				 * for the BPF_OUTPUT one:5401				 */5402				if (evsel__init_augmented_syscall_tp(evsel, evsel) ||5403				    evsel__init_augmented_syscall_tp_args(evsel))5404					goto out;5405				evsel->handler = trace__sys_enter;5406			}5407 5408			if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) {5409				struct syscall_tp *sc;5410init_augmented_syscall_tp:5411				if (evsel__init_augmented_syscall_tp(evsel, evsel))5412					goto out;5413				sc = __evsel__syscall_tp(evsel);5414				/*5415				 * For now with BPF raw_augmented we hook into5416				 * raw_syscalls:sys_enter and there we get all5417				 * 6 syscall args plus the tracepoint common5418				 * fields and the syscall_nr (another long).5419				 * So we check if that is the case and if so5420				 * don't look after the sc->args_size but5421				 * always after the full raw_syscalls:sys_enter5422				 * payload, which is fixed.5423				 *5424				 * We'll revisit this later to pass5425				 * s->args_size to the BPF augmenter (now5426				 * tools/perf/examples/bpf/augmented_raw_syscalls.c,5427				 * so that it copies only what we need for each5428				 * syscall, like what happens when we use5429				 * syscalls:sys_enter_NAME, so that we reduce5430				 * the kernel/userspace traffic to just what is5431				 * needed for each syscall.5432				 */5433				if (trace.raw_augmented_syscalls)5434					trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;5435				evsel__init_augmented_syscall_tp_ret(evsel);5436				evsel->handler = trace__sys_exit;5437			}5438		}5439	}5440 5441	if ((argc >= 1) && (strcmp(argv[0], "record") == 0))5442		return trace__record(&trace, argc-1, &argv[1]);5443 5444	/* Using just --errno-summary will trigger --summary */5445	if (trace.errno_summary && !trace.summary && !trace.summary_only)5446		trace.summary_only = true;5447 5448	/* summary_only implies summary option, but don't overwrite summary if set */5449	if (trace.summary_only)5450		trace.summary = trace.summary_only;5451 5452	if (output_name != NULL) {5453		err = trace__open_output(&trace, output_name);5454		if (err < 0) {5455			perror("failed to create output file");5456			goto out;5457		}5458	}5459 5460	err = evswitch__init(&trace.evswitch, trace.evlist, stderr);5461	if (err)5462		goto out_close;5463 5464	err = target__validate(&trace.opts.target);5465	if (err) {5466		target__strerror(&trace.opts.target, err, bf, sizeof(bf));5467		fprintf(trace.output, "%s", bf);5468		goto out_close;5469	}5470 5471	err = target__parse_uid(&trace.opts.target);5472	if (err) {5473		target__strerror(&trace.opts.target, err, bf, sizeof(bf));5474		fprintf(trace.output, "%s", bf);5475		goto out_close;5476	}5477 5478	if (!argc && target__none(&trace.opts.target))5479		trace.opts.target.system_wide = true;5480 5481	if (input_name)5482		err = trace__replay(&trace);5483	else5484		err = trace__run(&trace, argc, argv);5485 5486out_close:5487	if (output_name != NULL)5488		fclose(trace.output);5489out:5490	trace__exit(&trace);5491#ifdef HAVE_BPF_SKEL5492	augmented_raw_syscalls_bpf__destroy(trace.skel);5493#endif5494	return err;5495}5496