projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
vsock/test: fix SEQPACKET message bounds test
[linux-2.6-block.git] / tools / perf / builtin-trace.c
CommitLineData
a598bb5e
ACM		 1/*
2 * builtin-trace.c
3 *
4 * Builtin 'trace' command:
5 *
6 * Display a continuously updated trace of any workload, CPU, specific PID,
7 * system wide, etc. Default format is loosely strace like, but any other
8 * event may be specified using --event.
9 *
10 * Copyright (C) 2012, 2013, 2014, 2015 Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
11 *
12 * Initially based on the 'trace' prototype by Thomas Gleixner:
13 *
14 * http://lwn.net/Articles/415728/ ("Announcing a new utility: 'trace'")
a598bb5e
ACM		 15 */
16
aeb00b1a 17#include "util/record.h"
988bdb31 18#include <api/fs/tracing_path.h>
d891f2b7 19#ifdef HAVE_LIBBPF_SUPPORT
744fafc7 20#include <bpf/bpf.h>
3d6dfae8 21#include <bpf/libbpf.h>
5e6da6be
IR		 22#ifdef HAVE_BPF_SKEL
23#include "bpf_skel/augmented_raw_syscalls.skel.h"
24#endif
d891f2b7 25#endif
ff7a4f98 26#include "util/bpf_map.h"
c3e78a34 27#include "util/rlimit.h"
514f1c67 28#include "builtin.h"
9ea42ba4 29#include "util/cgroup.h"
752fde44 30#include "util/color.h"
ac96287c 31#include "util/config.h"
7c304ee0 32#include "util/debug.h"
4a3cec84 33#include "util/dso.h"
092bd3cd 34#include "util/env.h"
5ab8c689 35#include "util/event.h"
ca125277
ACM		 36#include "util/evsel.h"
37#include "util/evsel_fprintf.h"
ea49e01c 38#include "util/synthetic-events.h"
514f1c67 39#include "util/evlist.h"
22ac4318 40#include "util/evswitch.h"
e0fcfb08 41#include "util/mmap.h"
8520a98d 42#include <subcmd/pager.h>
4b6ab94e 43#include <subcmd/exec-cmd.h>
752fde44 44#include "util/machine.h"
1101f69a 45#include "util/map.h"
daecf9e0 46#include "util/symbol.h"
9a3993d4 47#include "util/path.h"
6810fc91 48#include "util/session.h"
752fde44 49#include "util/thread.h"
4b6ab94e 50#include <subcmd/parse-options.h>
2ae3a312 51#include "util/strlist.h"
bdc89661 52#include "util/intlist.h"
514f1c67 53#include "util/thread_map.h"
bf2575c1 54#include "util/stat.h"
4a3cec84 55#include "util/tool.h"
2da39f1c 56#include "util/util.h"
fd5cead2 57#include "trace/beauty/beauty.h"
97978b3e 58#include "trace-event.h"
9aca7f17 59#include "util/parse-events.h"
9b7c7728 60#include "util/tracepoint.h"
566a0885 61#include "callchain.h"
fea01392 62#include "print_binary.h"
a067558e 63#include "string2.h"
fd0db102 64#include "syscalltbl.h"
96c14451 65#include "rb_resort.h"
c1a604df 66#include "../perf.h"
514f1c67 67
a43783ae 68#include <errno.h>
fd20e811 69#include <inttypes.h>
4208735d 70#include <poll.h>
9607ad3a 71#include <signal.h>
514f1c67 72#include <stdlib.h>
017037ff 73#include <string.h>
8dd2a131 74#include <linux/err.h>
997bba8c 75#include <linux/filter.h>
877a7a11 76#include <linux/kernel.h>
39878d49 77#include <linux/random.h>
c6d4a494 78#include <linux/stringify.h>
bd48c63e 79#include <linux/time64.h>
7f7c536f 80#include <linux/zalloc.h>
bafae98e 81#include <fcntl.h>
4bcc4cff 82#include <sys/sysmacros.h>
514f1c67 83
3052ba56 84#include <linux/ctype.h>
7728fa0c 85#include <perf/mmap.h>
3d689ed6 86
378ef0f5
IR		 87#ifdef HAVE_LIBTRACEEVENT
88#include <traceevent/event-parse.h>
89#endif
90
c188e7ac
ACM		 91#ifndef O_CLOEXEC
92# define O_CLOEXEC 02000000
93#endif
94
83a51694
ACM		 95#ifndef F_LINUX_SPECIFIC_BASE
96# define F_LINUX_SPECIFIC_BASE 1024
97#endif
98
eadcab4c
LY		 99#define RAW_SYSCALL_ARGS_NUM 6
100
3f41b778
ACM		 101/*
102 * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100
103 */
888ca854
ACM		 104struct syscall_arg_fmt {
105 size_t (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
3f41b778 106 bool (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val);
888ca854
ACM		 107 unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
108 void *parm;
109 const char *name;
9597945d 110 u16 nr_entries; // for arrays
888ca854
ACM		 111 bool show_zero;
112};
113
114struct syscall_fmt {
115 const char *name;
116 const char *alias;
117 struct {
118 const char *sys_enter,
119 *sys_exit;
120 } bpf_prog_name;
eadcab4c 121 struct syscall_arg_fmt arg[RAW_SYSCALL_ARGS_NUM];
888ca854
ACM		 122 u8 nr_args;
123 bool errpid;
124 bool timeout;
125 bool hexret;
126};
127
d1d438a3
ACM		 128struct trace {
129 struct perf_tool tool;
fd0db102 130 struct syscalltbl *sctbl;
d1d438a3 131 struct {
d1d438a3
ACM		 132 struct syscall *table;
133 struct {
32dcd021 134 struct evsel *sys_enter,
5e6da6be
IR		 135 *sys_exit,
136 *bpf_output;
d1d438a3
ACM		 137 } events;
138 } syscalls;
5e6da6be
IR		 139#ifdef HAVE_BPF_SKEL
140 struct augmented_raw_syscalls_bpf *skel;
141#endif
d1d438a3 142 struct record_opts opts;
63503dba 143 struct evlist *evlist;
d1d438a3
ACM		 144 struct machine *host;
145 struct thread *current;
9ea42ba4 146 struct cgroup *cgroup;
d1d438a3
ACM		 147 u64 base_time;
148 FILE *output;
149 unsigned long nr_events;
5067a8cd
ACM		 150 unsigned long nr_events_printed;
151 unsigned long max_events;
22ac4318 152 struct evswitch evswitch;
d1d438a3
ACM		 153 struct strlist *ev_qualifier;
154 struct {
155 size_t nr;
156 int *entries;
157 } ev_qualifier_ids;
d1d438a3
ACM		 158 struct {
159 size_t nr;
160 pid_t *entries;
744fafc7 161 struct bpf_map *map;
d1d438a3
ACM		 162 } filter_pids;
163 double duration_filter;
164 double runtime_ms;
165 struct {
166 u64 vfs_getname,
167 proc_getname;
168 } stats;
c6d4a494 169 unsigned int max_stack;
5cf9c84e 170 unsigned int min_stack;
b9b6a2ea 171 int raw_augmented_syscalls_args_size;
cd26ea6d 172 bool raw_augmented_syscalls;
79d725cd 173 bool fd_path_disabled;
b9b6a2ea 174 bool sort_events;
d1d438a3
ACM		 175 bool not_ev_qualifier;
176 bool live;
177 bool full_time;
178 bool sched;
179 bool multiple_threads;
180 bool summary;
181 bool summary_only;
b88b14db 182 bool errno_summary;
0a6545bd 183 bool failure_only;
d1d438a3 184 bool show_comm;
591421e1 185 bool print_sample;
d1d438a3
ACM		 186 bool show_tool_stats;
187 bool trace_syscalls;
f11b2803 188 bool libtraceevent_print;
44621819 189 bool kernel_syscallchains;
9ed45d59 190 s16 args_alignment;
b036146f 191 bool show_tstamp;
42e4a52d 192 bool show_duration;
e7c634fc 193 bool show_zeros;
9d6dc178 194 bool show_arg_names;
c65c83ff 195 bool show_string_prefix;
d1d438a3
ACM		 196 bool force;
197 bool vfs_getname;
198 int trace_pgfaults;
7e035929 199 char *perfconfig_events;
028713aa
JO		 200 struct {
201 struct ordered_events data;
202 u64 last;
203 } oe;
d1d438a3 204};
a1c2552d 205
77170988
ACM		 206struct tp_field {
207 int offset;
208 union {
209 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
210 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
211 };
212};
213
214#define TP_UINT_FIELD(bits) \
215static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
216{ \
55d43bca
DA		 217 u##bits value; \
218 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
219 return value; \
77170988
ACM		 220}
221
222TP_UINT_FIELD(8);
223TP_UINT_FIELD(16);
224TP_UINT_FIELD(32);
225TP_UINT_FIELD(64);
226
227#define TP_UINT_FIELD__SWAPPED(bits) \
228static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
229{ \
55d43bca
DA		 230 u##bits value; \
231 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
77170988
ACM		 232 return bswap_##bits(value);\
233}
234
235TP_UINT_FIELD__SWAPPED(16);
236TP_UINT_FIELD__SWAPPED(32);
237TP_UINT_FIELD__SWAPPED(64);
238
aa823f58 239static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
77170988 240{
aa823f58 241 field->offset = offset;
77170988 242
aa823f58 243 switch (size) {
77170988
ACM		 244 case 1:
245 field->integer = tp_field__u8;
246 break;
247 case 2:
248 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
249 break;
250 case 4:
251 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
252 break;
253 case 8:
254 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
255 break;
256 default:
257 return -1;
258 }
259
260 return 0;
261}
262
2c92f982 263static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
aa823f58
ACM		 264{
265 return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
266}
267
77170988
ACM		 268static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
269{
270 return sample->raw_data + field->offset;
271}
272
aa823f58 273static int __tp_field__init_ptr(struct tp_field *field, int offset)
77170988 274{
aa823f58 275 field->offset = offset;
77170988
ACM		 276 field->pointer = tp_field__ptr;
277 return 0;
278}
279
2c92f982 280static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
aa823f58
ACM		 281{
282 return __tp_field__init_ptr(field, format_field->offset);
283}
284
77170988
ACM		 285struct syscall_tp {
286 struct tp_field id;
287 union {
288 struct tp_field args, ret;
289 };
290};
291
2b00bb62
ACM		 292/*
293 * The evsel->priv as used by 'perf trace'
294 * sc: for raw_syscalls:sys_{enter,exit} and syscalls:sys_{enter,exit}_SYSCALLNAME
295 * fmt: for all the other tracepoints
296 */
297struct evsel_trace {
298 struct syscall_tp sc;
299 struct syscall_arg_fmt *fmt;
300};
301
302static struct evsel_trace *evsel_trace__new(void)
303{
304 return zalloc(sizeof(struct evsel_trace));
305}
306
307static void evsel_trace__delete(struct evsel_trace *et)
308{
309 if (et == NULL)
310 return;
311
312 zfree(&et->fmt);
313 free(et);
314}
315
fecd9907
ACM		 316/*
317 * Used with raw_syscalls:sys_{enter,exit} and with the
318 * syscalls:sys_{enter,exit}_SYSCALL tracepoints
319 */
320static inline struct syscall_tp *__evsel__syscall_tp(struct evsel *evsel)
321{
2b00bb62 322 struct evsel_trace *et = evsel->priv;
fecd9907 323
2b00bb62 324 return &et->sc;
fecd9907
ACM		 325}
326
8b913df5
ACM		 327static struct syscall_tp *evsel__syscall_tp(struct evsel *evsel)
328{
329 if (evsel->priv == NULL) {
2b00bb62
ACM		 330 evsel->priv = evsel_trace__new();
331 if (evsel->priv == NULL)
332 return NULL;
8b913df5
ACM		 333 }
334
335 return __evsel__syscall_tp(evsel);
336}
337
fecd9907
ACM		 338/*
339 * Used with all the other tracepoints.
340 */
341static inline struct syscall_arg_fmt *__evsel__syscall_arg_fmt(struct evsel *evsel)
342{
2b00bb62 343 struct evsel_trace *et = evsel->priv;
fecd9907 344
2b00bb62 345 return et->fmt;
fecd9907
ACM		 346}
347
8b913df5
ACM		 348static struct syscall_arg_fmt *evsel__syscall_arg_fmt(struct evsel *evsel)
349{
2b00bb62
ACM		 350 struct evsel_trace *et = evsel->priv;
351
8b913df5 352 if (evsel->priv == NULL) {
2b00bb62
ACM		 353 et = evsel->priv = evsel_trace__new();
354
355 if (et == NULL)
356 return NULL;
357 }
358
359 if (et->fmt == NULL) {
360 et->fmt = calloc(evsel->tp_format->format.nr_fields, sizeof(struct syscall_arg_fmt));
361 if (et->fmt == NULL)
362 goto out_delete;
8b913df5
ACM		 363 }
364
365 return __evsel__syscall_arg_fmt(evsel);
2b00bb62
ACM		 366
367out_delete:
368 evsel_trace__delete(evsel->priv);
369 evsel->priv = NULL;
370 return NULL;
8b913df5
ACM		 371}
372
794bca26 373static int evsel__init_tp_uint_field(struct evsel *evsel, struct tp_field *field, const char *name)
77170988 374{
efc0cdc9 375 struct tep_format_field *format_field = evsel__field(evsel, name);
77170988
ACM		 376
377 if (format_field == NULL)
378 return -1;
379
380 return tp_field__init_uint(field, format_field, evsel->needs_swap);
381}
382
383#define perf_evsel__init_sc_tp_uint_field(evsel, name) \
fecd9907 384 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
794bca26 385 evsel__init_tp_uint_field(evsel, &sc->name, #name); })
77170988 386
794bca26 387static int evsel__init_tp_ptr_field(struct evsel *evsel, struct tp_field *field, const char *name)
77170988 388{
efc0cdc9 389 struct tep_format_field *format_field = evsel__field(evsel, name);
77170988
ACM		 390
391 if (format_field == NULL)
392 return -1;
393
394 return tp_field__init_ptr(field, format_field);
395}
396
397#define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
fecd9907 398 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
794bca26 399 evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
77170988 400
5eb2dd2a 401static void evsel__delete_priv(struct evsel *evsel)
77170988 402{
04662523 403 zfree(&evsel->priv);
5eb2dd2a 404 evsel__delete(evsel);
77170988
ACM		 405}
406
794bca26 407static int evsel__init_syscall_tp(struct evsel *evsel)
d32855fa 408{
8b913df5 409 struct syscall_tp *sc = evsel__syscall_tp(evsel);
d32855fa 410
8b913df5 411 if (sc != NULL) {
794bca26
ACM		 412 if (evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
413 evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
8b913df5 414 return -ENOENT;
5e6da6be 415
d32855fa
ACM		 416 return 0;
417 }
418
419 return -ENOMEM;
d32855fa
ACM		 420}
421
794bca26 422static int evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)
d3d1c4bd 423{
8b913df5 424 struct syscall_tp *sc = evsel__syscall_tp(evsel);
d3d1c4bd 425
8b913df5 426 if (sc != NULL) {
efc0cdc9 427 struct tep_format_field *syscall_id = evsel__field(tp, "id");
b9b6a2ea 428 if (syscall_id == NULL)
efc0cdc9 429 syscall_id = evsel__field(tp, "__syscall_nr");
8b913df5
ACM		 430 if (syscall_id == NULL ||
431 __tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
432 return -EINVAL;
d3d1c4bd
ACM		 433
434 return 0;
435 }
436
437 return -ENOMEM;
d3d1c4bd
ACM		 438}
439
794bca26 440static int evsel__init_augmented_syscall_tp_args(struct evsel *evsel)
d3d1c4bd 441{
fecd9907 442 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
d3d1c4bd
ACM		 443
444 return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
445}
446
794bca26 447static int evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)
b1a9e253 448{
fecd9907 449 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
b1a9e253
ACM		 450
451 return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
452}
453
794bca26 454static int evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)
96695d44 455{
8b913df5 456 if (evsel__syscall_tp(evsel) != NULL) {
96695d44 457 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
8b913df5 458 return -ENOENT;
96695d44
NK		 459
460 evsel->handler = handler;
461 return 0;
462 }
463
464 return -ENOMEM;
96695d44
NK		 465}
466
32dcd021 467static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
77170988 468{
8f6725a2 469 struct evsel *evsel = evsel__newtp("raw_syscalls", direction);
77170988 470
9aca7f17 471 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
8dd2a131 472 if (IS_ERR(evsel))
8f6725a2 473 evsel = evsel__newtp("syscalls", direction);
9aca7f17 474
8dd2a131
JO		 475 if (IS_ERR(evsel))
476 return NULL;
477
794bca26 478 if (evsel__init_raw_syscall_tp(evsel, handler))
8dd2a131 479 goto out_delete;
77170988
ACM		 480
481 return evsel;
482
483out_delete:
5eb2dd2a 484 evsel__delete_priv(evsel);
77170988
ACM		 485 return NULL;
486}
487
488#define perf_evsel__sc_tp_uint(evsel, name, sample) \
fecd9907 489 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
77170988
ACM		 490 fields->name.integer(&fields->name, sample); })
491
492#define perf_evsel__sc_tp_ptr(evsel, name, sample) \
fecd9907 493 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
77170988
ACM		 494 fields->name.pointer(&fields->name, sample); })
495
97c2a780
ACM		 496size_t strarray__scnprintf_suffix(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_suffix, int val)
497{
498 int idx = val - sa->offset;
499
500 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
501 size_t printed = scnprintf(bf, size, intfmt, val);
502 if (show_suffix)
503 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
504 return printed;
505 }
506
507 return scnprintf(bf, size, "%s%s", sa->entries[idx], show_suffix ? sa->prefix : "");
508}
509
c65c83ff 510size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
0ae79636
ACM		 511{
512 int idx = val - sa->offset;
1f115cb7 513
9614b8d6
ACM		 514 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
515 size_t printed = scnprintf(bf, size, intfmt, val);
516 if (show_prefix)
517 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
518 return printed;
519 }
1f115cb7 520
c65c83ff 521 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
03e3adc9
ACM		 522}
523
975b7c2f
ACM		 524static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
525 const char *intfmt,
526 struct syscall_arg *arg)
1f115cb7 527{
c65c83ff 528 return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
1f115cb7
ACM		 529}
530
975b7c2f
ACM		 531static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
532 struct syscall_arg *arg)
533{
534 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
535}
536
1f115cb7
ACM		 537#define SCA_STRARRAY syscall_arg__scnprintf_strarray
538
d066da97
ACM		 539bool syscall_arg__strtoul_strarray(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
540{
541 return strarray__strtoul(arg->parm, bf, size, ret);
542}
543
e0712baa
ACM		 544bool syscall_arg__strtoul_strarray_flags(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
545{
546 return strarray__strtoul_flags(arg->parm, bf, size, ret);
547}
548
1a8a90b8
ACM		 549bool syscall_arg__strtoul_strarrays(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
550{
551 return strarrays__strtoul(arg->parm, bf, size, ret);
552}
553
f5b91dbb
ACM		 554size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)
555{
556 return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);
557}
558
1f2d085e 559size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
83a51694 560{
9614b8d6 561 size_t printed;
83a51694
ACM		 562 int i;
563
564 for (i = 0; i < sas->nr_entries; ++i) {
565 struct strarray *sa = sas->entries[i];
1f2d085e 566 int idx = val - sa->offset;
83a51694
ACM		 567
568 if (idx >= 0 && idx < sa->nr_entries) {
569 if (sa->entries[idx] == NULL)
570 break;
1f2d085e 571 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
83a51694
ACM		 572 }
573 }
574
9614b8d6
ACM		 575 printed = scnprintf(bf, size, intfmt, val);
576 if (show_prefix)
577 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
578 return printed;
1f2d085e
ACM		 579}
580
d0a3a104
ACM		 581bool strarray__strtoul(struct strarray *sa, char *bf, size_t size, u64 *ret)
582{
583 int i;
584
585 for (i = 0; i < sa->nr_entries; ++i) {
586 if (sa->entries[i] && strncmp(sa->entries[i], bf, size) == 0 && sa->entries[i][size] == '\0') {
587 *ret = sa->offset + i;
588 return true;
589 }
590 }
591
592 return false;
593}
594
154c978d
ACM		 595bool strarray__strtoul_flags(struct strarray *sa, char *bf, size_t size, u64 *ret)
596{
597 u64 val = 0;
598 char *tok = bf, *sep, *end;
599
600 *ret = 0;
601
602 while (size != 0) {
603 int toklen = size;
604
605 sep = memchr(tok, '|', size);
606 if (sep != NULL) {
607 size -= sep - tok + 1;
608
609 end = sep - 1;
610 while (end > tok && isspace(*end))
611 --end;
612
613 toklen = end - tok + 1;
614 }
615
616 while (isspace(*tok))
617 ++tok;
618
619 if (isalpha(*tok) || *tok == '_') {
620 if (!strarray__strtoul(sa, tok, toklen, &val))
621 return false;
96b73141
CZ		 622 } else
623 val = strtoul(tok, NULL, 0);
154c978d
ACM		 624
625 *ret |= (1 << (val - 1));
626
627 if (sep == NULL)
628 break;
629 tok = sep + 1;
630 }
631
632 return true;
633}
634
d0a3a104
ACM		 635bool strarrays__strtoul(struct strarrays *sas, char *bf, size_t size, u64 *ret)
636{
637 int i;
638
639 for (i = 0; i < sas->nr_entries; ++i) {
640 struct strarray *sa = sas->entries[i];
641
642 if (strarray__strtoul(sa, bf, size, ret))
643 return true;
644 }
645
646 return false;
647}
648
1f2d085e
ACM		 649size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
650 struct syscall_arg *arg)
651{
652 return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
83a51694
ACM		 653}
654
48e1f91a
ACM		 655#ifndef AT_FDCWD
656#define AT_FDCWD -100
657#endif
658
75b757ca
ACM		 659static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
660 struct syscall_arg *arg)
661{
662 int fd = arg->val;
c65c83ff 663 const char *prefix = "AT_FD";
75b757ca
ACM		 664
665 if (fd == AT_FDCWD)
c65c83ff 666 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
75b757ca
ACM		 667
668 return syscall_arg__scnprintf_fd(bf, size, arg);
669}
670
671#define SCA_FDAT syscall_arg__scnprintf_fd_at
672
673static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
674 struct syscall_arg *arg);
675
676#define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
677
2c2b1623 678size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
13d4ff3e 679{
01533e97 680 return scnprintf(bf, size, "%#lx", arg->val);
13d4ff3e
ACM		 681}
682
ce05539f
ACM		 683size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
684{
685 if (arg->val == 0)
686 return scnprintf(bf, size, "NULL");
687 return syscall_arg__scnprintf_hex(bf, size, arg);
688}
689
2c2b1623 690size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
a1c2552d
ACM		 691{
692 return scnprintf(bf, size, "%d", arg->val);
693}
694
5dde91ed
ACM		 695size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
696{
697 return scnprintf(bf, size, "%ld", arg->val);
698}
699
9597945d
ACM		 700static size_t syscall_arg__scnprintf_char_array(char *bf, size_t size, struct syscall_arg *arg)
701{
702 // XXX Hey, maybe for sched:sched_switch prev/next comm fields we can
703 // fill missing comms using thread__set_comm()...
704 // here or in a special syscall_arg__scnprintf_pid_sched_tp...
c5e006cd 705 return scnprintf(bf, size, "\"%-.*s\"", arg->fmt->nr_entries ?: arg->len, arg->val);
9597945d
ACM		 706}
707
708#define SCA_CHAR_ARRAY syscall_arg__scnprintf_char_array
709
729a7841
ACM		 710static const char *bpf_cmd[] = {
711 "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
ea0056f0
WL		 712 "MAP_GET_NEXT_KEY", "PROG_LOAD", "OBJ_PIN", "OBJ_GET", "PROG_ATTACH",
713 "PROG_DETACH", "PROG_TEST_RUN", "PROG_GET_NEXT_ID", "MAP_GET_NEXT_ID",
714 "PROG_GET_FD_BY_ID", "MAP_GET_FD_BY_ID", "OBJ_GET_INFO_BY_FD",
715 "PROG_QUERY", "RAW_TRACEPOINT_OPEN", "BTF_LOAD", "BTF_GET_FD_BY_ID",
716 "TASK_FD_QUERY", "MAP_LOOKUP_AND_DELETE_ELEM", "MAP_FREEZE",
717 "BTF_GET_NEXT_ID", "MAP_LOOKUP_BATCH", "MAP_LOOKUP_AND_DELETE_BATCH",
718 "MAP_UPDATE_BATCH", "MAP_DELETE_BATCH", "LINK_CREATE", "LINK_UPDATE",
719 "LINK_GET_FD_BY_ID", "LINK_GET_NEXT_ID", "ENABLE_STATS", "ITER_CREATE",
720 "LINK_DETACH", "PROG_BIND_MAP",
729a7841 721};
2e3d7fac 722static DEFINE_STRARRAY(bpf_cmd, "BPF_");
729a7841 723
f6af0956
ACM		 724static const char *fsmount_flags[] = {
725 [1] = "CLOEXEC",
726};
727static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");
728
dcc6fd64
ACM		 729#include "trace/beauty/generated/fsconfig_arrays.c"
730
731static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");
732
03e3adc9 733static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
2e3d7fac 734static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
eac032c5 735
1f115cb7 736static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
2e3d7fac 737static DEFINE_STRARRAY(itimers, "ITIMER_");
1f115cb7 738
b62bee1b
ACM		 739static const char *keyctl_options[] = {
740 "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
741 "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
742 "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
743 "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
744 "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
745};
2e3d7fac 746static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
b62bee1b 747
efe6b882
ACM		 748static const char *whences[] = { "SET", "CUR", "END",
749#ifdef SEEK_DATA
750"DATA",
751#endif
752#ifdef SEEK_HOLE
753"HOLE",
754#endif
755};
2e3d7fac 756static DEFINE_STRARRAY(whences, "SEEK_");
f9da0b0c 757
80f587d5
ACM		 758static const char *fcntl_cmds[] = {
759 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
e000e5e3
ACM		 760 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
761 "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
762 "GETOWNER_UIDS",
80f587d5 763};
2e3d7fac 764static DEFINE_STRARRAY(fcntl_cmds, "F_");
80f587d5 765
83a51694
ACM		 766static const char *fcntl_linux_specific_cmds[] = {
767 "SETLEASE", "GETLEASE", "NOTIFY", [5] = "CANCELLK", "DUPFD_CLOEXEC",
768 "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
64e4561d 769 "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
83a51694
ACM		 770};
771
2e3d7fac 772static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
83a51694
ACM		 773
774static struct strarray *fcntl_cmds_arrays[] = {
775 &strarray__fcntl_cmds,
776 &strarray__fcntl_linux_specific_cmds,
777};
778
779static DEFINE_STRARRAYS(fcntl_cmds_arrays);
780
c045bf02
ACM		 781static const char *rlimit_resources[] = {
782 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
783 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
784 "RTTIME",
785};
2e3d7fac 786static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
c045bf02 787
eb5b1b14 788static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
2e3d7fac 789static DEFINE_STRARRAY(sighow, "SIG_");
eb5b1b14 790
4f8c1b74
DA		 791static const char *clockid[] = {
792 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
28ebb87c
ACM		 793 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
794 "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
4f8c1b74 795};
2e3d7fac 796static DEFINE_STRARRAY(clockid, "CLOCK_");
4f8c1b74 797
51108999
ACM		 798static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
799 struct syscall_arg *arg)
800{
c65c83ff
ACM		 801 bool show_prefix = arg->show_string_prefix;
802 const char *suffix = "_OK";
51108999
ACM		 803 size_t printed = 0;
804 int mode = arg->val;
805
806 if (mode == F_OK) /* 0 */
c65c83ff 807 return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
51108999
ACM		 808#define P_MODE(n) \
809 if (mode & n##_OK) { \
c65c83ff 810 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
51108999
ACM		 811 mode &= ~n##_OK; \
812 }
813
814 P_MODE(R);
815 P_MODE(W);
816 P_MODE(X);
817#undef P_MODE
818
819 if (mode)
820 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
821
822 return printed;
823}
824
825#define SCA_ACCMODE syscall_arg__scnprintf_access_mode
826
f994592d
ACM		 827static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
828 struct syscall_arg *arg);
829
830#define SCA_FILENAME syscall_arg__scnprintf_filename
831
46cce19b
ACM		 832static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
833 struct syscall_arg *arg)
834{
c65c83ff
ACM		 835 bool show_prefix = arg->show_string_prefix;
836 const char *prefix = "O_";
46cce19b
ACM		 837 int printed = 0, flags = arg->val;
838
839#define P_FLAG(n) \
840 if (flags & O_##n) { \
c65c83ff 841 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
46cce19b
ACM		 842 flags &= ~O_##n; \
843 }
844
845 P_FLAG(CLOEXEC);
846 P_FLAG(NONBLOCK);
847#undef P_FLAG
848
849 if (flags)
850 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
851
852 return printed;
853}
854
855#define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
856
a355a61e
ACM		 857#ifndef GRND_NONBLOCK
858#define GRND_NONBLOCK 0x0001
859#endif
860#ifndef GRND_RANDOM
861#define GRND_RANDOM 0x0002
862#endif
863
39878d49
ACM		 864static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
865 struct syscall_arg *arg)
866{
c65c83ff
ACM		 867 bool show_prefix = arg->show_string_prefix;
868 const char *prefix = "GRND_";
39878d49
ACM		 869 int printed = 0, flags = arg->val;
870
871#define P_FLAG(n) \
872 if (flags & GRND_##n) { \
c65c83ff 873 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
39878d49
ACM		 874 flags &= ~GRND_##n; \
875 }
876
877 P_FLAG(RANDOM);
878 P_FLAG(NONBLOCK);
879#undef P_FLAG
880
881 if (flags)
882 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
883
884 return printed;
885}
886
887#define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
888
82d4a110
ACM		 889#define STRARRAY(name, array) \
890 { .scnprintf = SCA_STRARRAY, \
d066da97 891 .strtoul = STUL_STRARRAY, \
82d4a110 892 .parm = &strarray__##array, }
453350dd 893
f6af0956
ACM		 894#define STRARRAY_FLAGS(name, array) \
895 { .scnprintf = SCA_STRARRAY_FLAGS, \
e0712baa 896 .strtoul = STUL_STRARRAY_FLAGS, \
f6af0956
ACM		 897 .parm = &strarray__##array, }
898
092bd3cd 899#include "trace/beauty/arch_errno_names.c"
ea8dc3ce 900#include "trace/beauty/eventfd.c"
d5d71e86 901#include "trace/beauty/futex_op.c"
3258abe0 902#include "trace/beauty/futex_val3.c"
df4cb167 903#include "trace/beauty/mmap.c"
ba2f22cf 904#include "trace/beauty/mode_t.c"
a30e6259 905#include "trace/beauty/msg_flags.c"
8f48df69 906#include "trace/beauty/open_flags.c"
62de344e 907#include "trace/beauty/perf_event_open.c"
d5d71e86 908#include "trace/beauty/pid.c"
a3bca91f 909#include "trace/beauty/sched_policy.c"
f5cd95ea 910#include "trace/beauty/seccomp.c"
12199d8e 911#include "trace/beauty/signum.c"
bbf86c43 912#include "trace/beauty/socket_type.c"
7206b900 913#include "trace/beauty/waitid_options.c"
a3bca91f 914
60995604 915static const struct syscall_fmt syscall_fmts[] = {
1f63139c 916 { .name = "access",
82d4a110 917 .arg = { [1] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, },
fb7068e7
ACM		 918 { .name = "arch_prctl",
919 .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
920 [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
02ef2884 921 { .name = "bind",
ef969ca6 922 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
247dd65b
ACM		 923 [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ },
924 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
1f63139c 925 { .name = "bpf",
82d4a110 926 .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
beccb2b5 927 { .name = "brk", .hexret = true,
ce05539f 928 .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
1f63139c 929 { .name = "clock_gettime",
82d4a110 930 .arg = { [0] = STRARRAY(clk_id, clockid), }, },
6ac73820
ACM		 931 { .name = "clock_nanosleep",
932 .arg = { [2] = { .scnprintf = SCA_TIMESPEC, /* rqtp */ }, }, },
33396a3a
ACM		 933 { .name = "clone", .errpid = true, .nr_args = 5,
934 .arg = { [0] = { .name = "flags", .scnprintf = SCA_CLONE_FLAGS, },
935 [1] = { .name = "child_stack", .scnprintf = SCA_HEX, },
936 [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
937 [3] = { .name = "child_tidptr", .scnprintf = SCA_HEX, },
938 [4] = { .name = "tls", .scnprintf = SCA_HEX, }, }, },
1f63139c 939 { .name = "close",
82d4a110 940 .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
d5a7e661 941 { .name = "connect",
ef969ca6
ACM		 942 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
943 [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ },
1d862752 944 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
1f63139c 945 { .name = "epoll_ctl",
82d4a110 946 .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
1f63139c 947 { .name = "eventfd2",
82d4a110 948 .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
1f63139c 949 { .name = "fchmodat",
82d4a110 950 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1f63139c 951 { .name = "fchownat",
82d4a110 952 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1f63139c 953 { .name = "fcntl",
82c38338
ACM		 954 .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
955 .strtoul = STUL_STRARRAYS,
39cc355b
ACM		 956 .parm = &strarrays__fcntl_cmds_arrays,
957 .show_zero = true, },
82d4a110 958 [2] = { .scnprintf = SCA_FCNTL_ARG, /* arg */ }, }, },
1f63139c 959 { .name = "flock",
82d4a110 960 .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
dcc6fd64
ACM		 961 { .name = "fsconfig",
962 .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },
f6af0956
ACM		 963 { .name = "fsmount",
964 .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),
965 [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },
693bd394
ACM		 966 { .name = "fspick",
967 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
968 [1] = { .scnprintf = SCA_FILENAME, /* path */ },
969 [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },
1f63139c
ACM		 970 { .name = "fstat", .alias = "newfstat", },
971 { .name = "fstatat", .alias = "newfstatat", },
972 { .name = "futex",
3258abe0
ACM		 973 .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
974 [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
1f63139c 975 { .name = "futimesat",
82d4a110 976 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1f63139c 977 { .name = "getitimer",
82d4a110 978 .arg = { [0] = STRARRAY(which, itimers), }, },
c65f1070 979 { .name = "getpid", .errpid = true, },
d1d438a3 980 { .name = "getpgid", .errpid = true, },
c65f1070 981 { .name = "getppid", .errpid = true, },
1f63139c 982 { .name = "getrandom",
82d4a110 983 .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
1f63139c 984 { .name = "getrlimit",
82d4a110 985 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
f1c1e45e
ACM		 986 { .name = "getsockopt",
987 .arg = { [1] = STRARRAY(level, socket_level), }, },
2d1073de 988 { .name = "gettid", .errpid = true, },
1f63139c 989 { .name = "ioctl",
82d4a110 990 .arg = {
844ae5b4
ACM		 991#if defined(__i386__) || defined(__x86_64__)
992/*
993 * FIXME: Make this available to all arches.
994 */
1cc47f2d 995 [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
82d4a110 996 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
844ae5b4 997#else
82d4a110 998 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
844ae5b4 999#endif
1de3038d
ACM		 1000 { .name = "kcmp", .nr_args = 5,
1001 .arg = { [0] = { .name = "pid1", .scnprintf = SCA_PID, },
1002 [1] = { .name = "pid2", .scnprintf = SCA_PID, },
1003 [2] = { .name = "type", .scnprintf = SCA_KCMP_TYPE, },
1004 [3] = { .name = "idx1", .scnprintf = SCA_KCMP_IDX, },
1005 [4] = { .name = "idx2", .scnprintf = SCA_KCMP_IDX, }, }, },
1f63139c 1006 { .name = "keyctl",
82d4a110 1007 .arg = { [0] = STRARRAY(option, keyctl_options), }, },
1f63139c 1008 { .name = "kill",
82d4a110 1009 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1f63139c 1010 { .name = "linkat",
82d4a110 1011 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1f63139c 1012 { .name = "lseek",
82d4a110 1013 .arg = { [2] = STRARRAY(whence, whences), }, },
1f63139c
ACM		 1014 { .name = "lstat", .alias = "newlstat", },
1015 { .name = "madvise",
82d4a110
ACM		 1016 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1017 [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
1f63139c 1018 { .name = "mkdirat",
82d4a110 1019 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1f63139c 1020 { .name = "mknodat",
82d4a110 1021 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
beccb2b5 1022 { .name = "mmap", .hexret = true,
54265664
JO		 1023/* The standard mmap maps to old_mmap on s390x */
1024#if defined(__s390x__)
1025 .alias = "old_mmap",
1026#endif
ce05539f 1027 .arg = { [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
27198a89
ACM		 1028 [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */
1029 .strtoul = STUL_STRARRAY_FLAGS,
1030 .parm = &strarray__mmap_flags, },
a6631340 1031 [5] = { .scnprintf = SCA_HEX, /* offset */ }, }, },
73d141ad 1032 { .name = "mount",
23c07a23
ACM		 1033 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
1034 [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
73d141ad 1035 .mask_val = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
566e3022
ACM		 1036 { .name = "move_mount",
1037 .arg = { [0] = { .scnprintf = SCA_FDAT, /* from_dfd */ },
1038 [1] = { .scnprintf = SCA_FILENAME, /* from_pathname */ },
1039 [2] = { .scnprintf = SCA_FDAT, /* to_dfd */ },
1040 [3] = { .scnprintf = SCA_FILENAME, /* to_pathname */ },
1041 [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },
1f63139c 1042 { .name = "mprotect",
82d4a110
ACM		 1043 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1044 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ }, }, },
1f63139c 1045 { .name = "mq_unlink",
82d4a110 1046 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
ae685380 1047 { .name = "mremap", .hexret = true,
ce05539f 1048 .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
1f63139c 1049 { .name = "name_to_handle_at",
82d4a110 1050 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1f63139c 1051 { .name = "newfstatat",
82d4a110 1052 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1f63139c 1053 { .name = "open",
82d4a110 1054 .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1f63139c 1055 { .name = "open_by_handle_at",
82d4a110
ACM		 1056 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1057 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1f63139c 1058 { .name = "openat",
82d4a110
ACM		 1059 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1060 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1f63139c 1061 { .name = "perf_event_open",
a9cd6c67
ACM		 1062 .arg = { [0] = { .scnprintf = SCA_PERF_ATTR, /* attr */ },
1063 [2] = { .scnprintf = SCA_INT, /* cpu */ },
82d4a110
ACM		 1064 [3] = { .scnprintf = SCA_FD, /* group_fd */ },
1065 [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
1f63139c 1066 { .name = "pipe2",
82d4a110 1067 .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
83bc9c37
ACM		 1068 { .name = "pkey_alloc",
1069 .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS, /* access_rights */ }, }, },
1070 { .name = "pkey_free",
1071 .arg = { [0] = { .scnprintf = SCA_INT, /* key */ }, }, },
1072 { .name = "pkey_mprotect",
1073 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1074 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
1075 [3] = { .scnprintf = SCA_INT, /* pkey */ }, }, },
1f63139c
ACM		 1076 { .name = "poll", .timeout = true, },
1077 { .name = "ppoll", .timeout = true, },
fb7068e7 1078 { .name = "prctl",
d7a07b29
ACM		 1079 .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */
1080 .strtoul = STUL_STRARRAY,
1081 .parm = &strarray__prctl_options, },
d688d037
ACM		 1082 [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
1083 [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
1f63139c
ACM		 1084 { .name = "pread", .alias = "pread64", },
1085 { .name = "preadv", .alias = "pread", },
1086 { .name = "prlimit64",
82d4a110 1087 .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
1f63139c
ACM		 1088 { .name = "pwrite", .alias = "pwrite64", },
1089 { .name = "readlinkat",
82d4a110 1090 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1f63139c 1091 { .name = "recvfrom",
82d4a110 1092 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1f63139c 1093 { .name = "recvmmsg",
82d4a110 1094 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1f63139c 1095 { .name = "recvmsg",
82d4a110 1096 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1f63139c 1097 { .name = "renameat",
436651ca
ACM		 1098 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1099 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
84a83541
ACM		 1100 { .name = "renameat2",
1101 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
ca7ff2c8
ACM		 1102 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
1103 [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
1f63139c 1104 { .name = "rt_sigaction",
82d4a110 1105 .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1f63139c 1106 { .name = "rt_sigprocmask",
82d4a110 1107 .arg = { [0] = STRARRAY(how, sighow), }, },
1f63139c 1108 { .name = "rt_sigqueueinfo",
82d4a110 1109 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1f63139c 1110 { .name = "rt_tgsigqueueinfo",
82d4a110 1111 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1f63139c 1112 { .name = "sched_setscheduler",
82d4a110 1113 .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
1f63139c 1114 { .name = "seccomp",
82d4a110
ACM		 1115 .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP, /* op */ },
1116 [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
1f63139c 1117 { .name = "select", .timeout = true, },
e4b00e93 1118 { .name = "sendfile", .alias = "sendfile64", },
1f63139c 1119 { .name = "sendmmsg",
82d4a110 1120 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1f63139c 1121 { .name = "sendmsg",
82d4a110 1122 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1f63139c 1123 { .name = "sendto",
6ebb6862
ACM		 1124 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
1125 [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
c65f1070 1126 { .name = "set_tid_address", .errpid = true, },
1f63139c 1127 { .name = "setitimer",
82d4a110 1128 .arg = { [0] = STRARRAY(which, itimers), }, },
1f63139c 1129 { .name = "setrlimit",
82d4a110 1130 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
0826b7fd
ACM		 1131 { .name = "setsockopt",
1132 .arg = { [1] = STRARRAY(level, socket_level), }, },
1f63139c 1133 { .name = "socket",
82d4a110 1134 .arg = { [0] = STRARRAY(family, socket_families),
162d3edb
ACM		 1135 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1136 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1f63139c 1137 { .name = "socketpair",
82d4a110 1138 .arg = { [0] = STRARRAY(family, socket_families),
162d3edb
ACM		 1139 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1140 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1f63139c
ACM		 1141 { .name = "stat", .alias = "newstat", },
1142 { .name = "statx",
82d4a110
ACM		 1143 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fdat */ },
1144 [2] = { .scnprintf = SCA_STATX_FLAGS, /* flags */ } ,
1145 [3] = { .scnprintf = SCA_STATX_MASK, /* mask */ }, }, },
1f63139c 1146 { .name = "swapoff",
82d4a110 1147 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1f63139c 1148 { .name = "swapon",
82d4a110 1149 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1f63139c 1150 { .name = "symlinkat",
82d4a110 1151 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
a9a187a7
ACM		 1152 { .name = "sync_file_range",
1153 .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },
1f63139c 1154 { .name = "tgkill",
82d4a110 1155 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1f63139c 1156 { .name = "tkill",
82d4a110 1157 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
476c92ca
ACM		 1158 { .name = "umount2", .alias = "umount",
1159 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
1f63139c
ACM		 1160 { .name = "uname", .alias = "newuname", },
1161 { .name = "unlinkat",
82d4a110 1162 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1f63139c 1163 { .name = "utimensat",
82d4a110 1164 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
11c8e39f 1165 { .name = "wait4", .errpid = true,
82d4a110 1166 .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
11c8e39f 1167 { .name = "waitid", .errpid = true,
82d4a110 1168 .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
514f1c67
ACM		 1169};
1170
1171static int syscall_fmt__cmp(const void *name, const void *fmtp)
1172{
1173 const struct syscall_fmt *fmt = fmtp;
1174 return strcmp(name, fmt->name);
1175}
1176
60995604
IR		 1177static const struct syscall_fmt *__syscall_fmt__find(const struct syscall_fmt *fmts,
1178 const int nmemb,
1179 const char *name)
bcddbfc5
ACM		 1180{
1181 return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1182}
1183
60995604 1184static const struct syscall_fmt *syscall_fmt__find(const char *name)
514f1c67
ACM		 1185{
1186 const int nmemb = ARRAY_SIZE(syscall_fmts);
bcddbfc5 1187 return __syscall_fmt__find(syscall_fmts, nmemb, name);
514f1c67
ACM		 1188}
1189
60995604
IR		 1190static const struct syscall_fmt *__syscall_fmt__find_by_alias(const struct syscall_fmt *fmts,
1191 const int nmemb, const char *alias)
f932184e 1192{
bcddbfc5 1193 int i;
f932184e
ACM		 1194
1195 for (i = 0; i < nmemb; ++i) {
bcddbfc5
ACM		 1196 if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0)
1197 return &fmts[i];
f932184e
ACM		 1198 }
1199
1200 return NULL;
1201}
1202
60995604 1203static const struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
bcddbfc5
ACM		 1204{
1205 const int nmemb = ARRAY_SIZE(syscall_fmts);
1206 return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias);
1207}
1208
6a648b53
ACM		 1209/*
1210 * is_exit: is this "exit" or "exit_group"?
1211 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
7a983a0f 1212 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
b8b1033f 1213 * nonexistent: Just a hole in the syscall table, syscall id not allocated
6a648b53 1214 */
514f1c67 1215struct syscall {
97fbf3f0 1216 struct tep_event *tp_format;
f208bd8d 1217 int nr_args;
7a983a0f 1218 int args_size;
6ff8fff4
ACM		 1219 struct {
1220 struct bpf_program *sys_enter,
1221 *sys_exit;
1222 } bpf_prog;
6a648b53
ACM		 1223 bool is_exit;
1224 bool is_open;
b8b1033f 1225 bool nonexistent;
2c92f982 1226 struct tep_format_field *args;
514f1c67 1227 const char *name;
60995604 1228 const struct syscall_fmt *fmt;
82d4a110 1229 struct syscall_arg_fmt *arg_fmt;
514f1c67
ACM		 1230};
1231
fd2b2975
ACM		 1232/*
1233 * We need to have this 'calculated' boolean because in some cases we really
1234 * don't know what is the duration of a syscall, for instance, when we start
1235 * a session and some threads are waiting for a syscall to finish, say 'poll',
1236 * in which case all we can do is to print "( ? ) for duration and for the
1237 * start timestamp.
1238 */
1239static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
60c907ab
ACM		 1240{
1241 double duration = (double)t / NSEC_PER_MSEC;
1242 size_t printed = fprintf(fp, "(");
1243
fd2b2975 1244 if (!calculated)
522283fe 1245 printed += fprintf(fp, " ");
fd2b2975 1246 else if (duration >= 1.0)
60c907ab
ACM		 1247 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1248 else if (duration >= 0.01)
1249 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1250 else
1251 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
c24ff998 1252 return printed + fprintf(fp, "): ");
60c907ab
ACM		 1253}
1254
f994592d
ACM		 1255/**
1256 * filename.ptr: The filename char pointer that will be vfs_getname'd
1257 * filename.entry_str_pos: Where to insert the string translated from
1258 * filename.ptr by the vfs_getname tracepoint/kprobe.
84486caa
ACM		 1259 * ret_scnprintf: syscall args may set this to a different syscall return
1260 * formatter, for instance, fcntl may return fds, file flags, etc.
f994592d 1261 */
752fde44
ACM		 1262struct thread_trace {
1263 u64 entry_time;
752fde44 1264 bool entry_pending;
efd5745e 1265 unsigned long nr_events;
a2ea67d7 1266 unsigned long pfmaj, pfmin;
752fde44 1267 char *entry_str;
1302d88e 1268 double runtime_ms;
7ee57434 1269 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
f994592d
ACM		 1270 struct {
1271 unsigned long ptr;
7f4f8001
ACM		 1272 short int entry_str_pos;
1273 bool pending_open;
1274 unsigned int namelen;
1275 char *name;
f994592d 1276 } filename;
75b757ca 1277 struct {
f4a74fcb
ACM		 1278 int max;
1279 struct file *table;
1280 } files;
bf2575c1
DA		 1281
1282 struct intlist *syscall_stats;
752fde44
ACM		 1283};
1284
1285static struct thread_trace *thread_trace__new(void)
1286{
75b757ca
ACM		 1287 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
1288
7a6d49dc 1289 if (ttrace) {
f4a74fcb 1290 ttrace->files.max = -1;
7a6d49dc
LY		 1291 ttrace->syscall_stats = intlist__new(NULL);
1292 }
bf2575c1 1293
75b757ca 1294 return ttrace;
752fde44
ACM		 1295}
1296
fcca1faf
ACM		 1297static void thread_trace__free_files(struct thread_trace *ttrace);
1298
9de251cb
ACM		 1299static void thread_trace__delete(void *pttrace)
1300{
1301 struct thread_trace *ttrace = pttrace;
1302
1303 if (!ttrace)
1304 return;
1305
1306 intlist__delete(ttrace->syscall_stats);
1307 ttrace->syscall_stats = NULL;
fcca1faf 1308 thread_trace__free_files(ttrace);
9de251cb
ACM		 1309 zfree(&ttrace->entry_str);
1310 free(ttrace);
1311}
1312
c24ff998 1313static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
752fde44 1314{
efd5745e
ACM		 1315 struct thread_trace *ttrace;
1316
752fde44
ACM		 1317 if (thread == NULL)
1318 goto fail;
1319
89dceb22
NK		 1320 if (thread__priv(thread) == NULL)
1321 thread__set_priv(thread, thread_trace__new());
48000a1a 1322
89dceb22 1323 if (thread__priv(thread) == NULL)
752fde44
ACM		 1324 goto fail;
1325
89dceb22 1326 ttrace = thread__priv(thread);
efd5745e
ACM		 1327 ++ttrace->nr_events;
1328
1329 return ttrace;
752fde44 1330fail:
c24ff998 1331 color_fprintf(fp, PERF_COLOR_RED,
752fde44
ACM		 1332 "WARNING: not enough memory, dropping samples!\n");
1333 return NULL;
1334}
1335
84486caa
ACM		 1336
1337void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
7ee57434 1338 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
84486caa
ACM		 1339{
1340 struct thread_trace *ttrace = thread__priv(arg->thread);
1341
1342 ttrace->ret_scnprintf = ret_scnprintf;
1343}
1344
598d02c5
SF		 1345#define TRACE_PFMAJ (1 << 0)
1346#define TRACE_PFMIN (1 << 1)
1347
e4d44e83
ACM		 1348static const size_t trace__entry_str_size = 2048;
1349
fcca1faf
ACM		 1350static void thread_trace__free_files(struct thread_trace *ttrace)
1351{
1352 for (int i = 0; i < ttrace->files.max; ++i) {
1353 struct file *file = ttrace->files.table + i;
1354 zfree(&file->pathname);
1355 }
1356
1357 zfree(&ttrace->files.table);
1358 ttrace->files.max = -1;
1359}
1360
d7e13484 1361static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
75b757ca 1362{
05107486
ACM		 1363 if (fd < 0)
1364 return NULL;
1365
f4a74fcb
ACM		 1366 if (fd > ttrace->files.max) {
1367 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
75b757ca 1368
f4a74fcb 1369 if (nfiles == NULL)
d7e13484 1370 return NULL;
75b757ca 1371
f4a74fcb
ACM		 1372 if (ttrace->files.max != -1) {
1373 memset(nfiles + ttrace->files.max + 1, 0,
1374 (fd - ttrace->files.max) * sizeof(struct file));
75b757ca 1375 } else {
f4a74fcb 1376 memset(nfiles, 0, (fd + 1) * sizeof(struct file));
75b757ca
ACM		 1377 }
1378
f4a74fcb
ACM		 1379 ttrace->files.table = nfiles;
1380 ttrace->files.max = fd;
75b757ca
ACM		 1381 }
1382
d7e13484
ACM		 1383 return ttrace->files.table + fd;
1384}
1385
2d473389
ACM		 1386struct file *thread__files_entry(struct thread *thread, int fd)
1387{
1388 return thread_trace__files_entry(thread__priv(thread), fd);
1389}
1390
d7e13484
ACM		 1391static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1392{
1393 struct thread_trace *ttrace = thread__priv(thread);
1394 struct file *file = thread_trace__files_entry(ttrace, fd);
1395
1396 if (file != NULL) {
4bcc4cff
ACM		 1397 struct stat st;
1398 if (stat(pathname, &st) == 0)
1399 file->dev_maj = major(st.st_rdev);
d7e13484
ACM		 1400 file->pathname = strdup(pathname);
1401 if (file->pathname)
1402 return 0;
1403 }
75b757ca 1404
d7e13484 1405 return -1;
75b757ca
ACM		 1406}
1407
97119f37
ACM		 1408static int thread__read_fd_path(struct thread *thread, int fd)
1409{
1410 char linkname[PATH_MAX], pathname[PATH_MAX];
1411 struct stat st;
1412 int ret;
1413
ee84a303 1414 if (thread__pid(thread) == thread__tid(thread)) {
97119f37 1415 scnprintf(linkname, sizeof(linkname),
ee84a303 1416 "/proc/%d/fd/%d", thread__pid(thread), fd);
97119f37
ACM		 1417 } else {
1418 scnprintf(linkname, sizeof(linkname),
ee84a303
IR		 1419 "/proc/%d/task/%d/fd/%d",
1420 thread__pid(thread), thread__tid(thread), fd);
97119f37
ACM		 1421 }
1422
1423 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1424 return -1;
1425
1426 ret = readlink(linkname, pathname, sizeof(pathname));
1427
1428 if (ret < 0 || ret > st.st_size)
1429 return -1;
1430
1431 pathname[ret] = '\0';
1432 return trace__set_fd_pathname(thread, fd, pathname);
1433}
1434
c522739d
ACM		 1435static const char *thread__fd_path(struct thread *thread, int fd,
1436 struct trace *trace)
75b757ca 1437{
89dceb22 1438 struct thread_trace *ttrace = thread__priv(thread);
75b757ca 1439
79d725cd 1440 if (ttrace == NULL || trace->fd_path_disabled)
75b757ca
ACM		 1441 return NULL;
1442
1443 if (fd < 0)
1444 return NULL;
1445
f4a74fcb 1446 if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
c522739d
ACM		 1447 if (!trace->live)
1448 return NULL;
1449 ++trace->stats.proc_getname;
cdcd1e6b 1450 if (thread__read_fd_path(thread, fd))
c522739d
ACM		 1451 return NULL;
1452 }
75b757ca 1453
f4a74fcb 1454 return ttrace->files.table[fd].pathname;
75b757ca
ACM		 1455}
1456
fc65eb82 1457size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
75b757ca
ACM		 1458{
1459 int fd = arg->val;
1460 size_t printed = scnprintf(bf, size, "%d", fd);
c522739d 1461 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
75b757ca
ACM		 1462
1463 if (path)
1464 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1465
1466 return printed;
1467}
1468
0a2f7540
ACM		 1469size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1470{
1471 size_t printed = scnprintf(bf, size, "%d", fd);
1472 struct thread *thread = machine__find_thread(trace->host, pid, pid);
1473
1474 if (thread) {
1475 const char *path = thread__fd_path(thread, fd, trace);
1476
1477 if (path)
1478 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1479
1480 thread__put(thread);
1481 }
1482
1483 return printed;
1484}
1485
75b757ca
ACM		 1486static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1487 struct syscall_arg *arg)
1488{
1489 int fd = arg->val;
1490 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
89dceb22 1491 struct thread_trace *ttrace = thread__priv(arg->thread);
75b757ca 1492
f4a74fcb
ACM		 1493 if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1494 zfree(&ttrace->files.table[fd].pathname);
75b757ca
ACM		 1495
1496 return printed;
1497}
1498
f994592d
ACM		 1499static void thread__set_filename_pos(struct thread *thread, const char *bf,
1500 unsigned long ptr)
1501{
1502 struct thread_trace *ttrace = thread__priv(thread);
1503
1504 ttrace->filename.ptr = ptr;
1505 ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1506}
1507
75d1e306
ACM		 1508static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1509{
1510 struct augmented_arg *augmented_arg = arg->augmented.args;
8195168e
ACM		 1511 size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1512 /*
1513 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1514 * we would have two strings, each prefixed by its size.
1515 */
1516 int consumed = sizeof(*augmented_arg) + augmented_arg->size;
75d1e306 1517
016f327c 1518 arg->augmented.args = ((void *)arg->augmented.args) + consumed;
8195168e 1519 arg->augmented.size -= consumed;
75d1e306 1520
8195168e 1521 return printed;
75d1e306
ACM		 1522}
1523
f994592d
ACM		 1524static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1525 struct syscall_arg *arg)
1526{
1527 unsigned long ptr = arg->val;
1528
75d1e306
ACM		 1529 if (arg->augmented.args)
1530 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1531
f994592d
ACM		 1532 if (!arg->trace->vfs_getname)
1533 return scnprintf(bf, size, "%#x", ptr);
1534
1535 thread__set_filename_pos(arg->thread, bf, ptr);
1536 return 0;
1537}
1538
ae9ed035
ACM		 1539static bool trace__filter_duration(struct trace *trace, double t)
1540{
1541 return t < (trace->duration_filter * NSEC_PER_MSEC);
1542}
1543
fd2b2975 1544static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
752fde44
ACM		 1545{
1546 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1547
60c907ab 1548 return fprintf(fp, "%10.3f ", ts);
752fde44
ACM		 1549}
1550
fd2b2975
ACM		 1551/*
1552 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1553 * using ttrace->entry_time for a thread that receives a sys_exit without
1554 * first having received a sys_enter ("poll" issued before tracing session
1555 * starts, lost sys_enter exit due to ring buffer overflow).
1556 */
1557static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1558{
1559 if (tstamp > 0)
1560 return __trace__fprintf_tstamp(trace, tstamp, fp);
1561
1562 return fprintf(fp, " ? ");
1563}
1564
de9f498d 1565static pid_t workload_pid = -1;
92ea0720
IR		 1566static volatile sig_atomic_t done = false;
1567static volatile sig_atomic_t interrupted = false;
f15eb531 1568
de9f498d 1569static void sighandler_interrupt(int sig __maybe_unused)
f15eb531 1570{
de9f498d
CD		 1571 done = interrupted = true;
1572}
1573
1574static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
1575 void *context __maybe_unused)
1576{
1577 if (info->si_pid == workload_pid)
1578 done = true;
f15eb531
NK		 1579}
1580
6dcbd212 1581static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
752fde44 1582{
6dcbd212 1583 size_t printed = 0;
752fde44 1584
50c95cbd
ACM		 1585 if (trace->multiple_threads) {
1586 if (trace->show_comm)
1902efe7 1587 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
ee84a303 1588 printed += fprintf(fp, "%d ", thread__tid(thread));
50c95cbd 1589 }
752fde44
ACM		 1590
1591 return printed;
1592}
1593
6dcbd212
ACM		 1594static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1595 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1596{
b036146f
ACM		 1597 size_t printed = 0;
1598
1599 if (trace->show_tstamp)
1600 printed = trace__fprintf_tstamp(trace, tstamp, fp);
42e4a52d
ACM		 1601 if (trace->show_duration)
1602 printed += fprintf_duration(duration, duration_calculated, fp);
6dcbd212
ACM		 1603 return printed + trace__fprintf_comm_tid(trace, thread, fp);
1604}
1605
c24ff998 1606static int trace__process_event(struct trace *trace, struct machine *machine,
162f0bef 1607 union perf_event *event, struct perf_sample *sample)
752fde44
ACM		 1608{
1609 int ret = 0;
1610
1611 switch (event->header.type) {
1612 case PERF_RECORD_LOST:
c24ff998 1613 color_fprintf(trace->output, PERF_COLOR_RED,
752fde44 1614 "LOST %" PRIu64 " events!\n", event->lost.lost);
162f0bef 1615 ret = machine__process_lost_event(machine, event, sample);
3ed5ca2e 1616 break;
752fde44 1617 default:
162f0bef 1618 ret = machine__process_event(machine, event, sample);
752fde44
ACM		 1619 break;
1620 }
1621
1622 return ret;
1623}
1624
c24ff998 1625static int trace__tool_process(struct perf_tool *tool,
752fde44 1626 union perf_event *event,
162f0bef 1627 struct perf_sample *sample,
752fde44
ACM		 1628 struct machine *machine)
1629{
c24ff998 1630 struct trace *trace = container_of(tool, struct trace, tool);
162f0bef 1631 return trace__process_event(trace, machine, event, sample);
752fde44
ACM		 1632}
1633
caf8a0d0
ACM		 1634static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1635{
1636 struct machine *machine = vmachine;
1637
1638 if (machine->kptr_restrict_warned)
1639 return NULL;
1640
1641 if (symbol_conf.kptr_restrict) {
1642 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
d06e5fad 1643 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
caf8a0d0
ACM		 1644 "Kernel samples will not be resolved.\n");
1645 machine->kptr_restrict_warned = true;
1646 return NULL;
1647 }
1648
1649 return machine__resolve_kernel_addr(vmachine, addrp, modp);
1650}
1651
63503dba 1652static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
752fde44 1653{
0a7e6d1b 1654 int err = symbol__init(NULL);
752fde44
ACM		 1655
1656 if (err)
1657 return err;
1658
8fb598e5
DA		 1659 trace->host = machine__new_host();
1660 if (trace->host == NULL)
1661 return -ENOMEM;
752fde44 1662
9de251cb
ACM		 1663 thread__set_priv_destructor(thread_trace__delete);
1664
cbd5c178
AV		 1665 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1666 if (err < 0)
1667 goto out;
706c3da4 1668
a33fbd56 1669 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
84111b9c
NK		 1670 evlist->core.threads, trace__tool_process,
1671 true, false, 1);
cbd5c178 1672out:
752fde44
ACM		 1673 if (err)
1674 symbol__exit();
1675
1676 return err;
1677}
1678
33974a41
AV		 1679static void trace__symbols__exit(struct trace *trace)
1680{
1681 machine__exit(trace->host);
1682 trace->host = NULL;
1683
1684 symbol__exit();
1685}
1686
5e58fcfa 1687static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
13d4ff3e 1688{
5e58fcfa 1689 int idx;
13d4ff3e 1690
eadcab4c 1691 if (nr_args == RAW_SYSCALL_ARGS_NUM && sc->fmt && sc->fmt->nr_args != 0)
332337da
ACM		 1692 nr_args = sc->fmt->nr_args;
1693
5e58fcfa 1694 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
82d4a110 1695 if (sc->arg_fmt == NULL)
13d4ff3e
ACM		 1696 return -1;
1697
5e58fcfa
ACM		 1698 for (idx = 0; idx < nr_args; ++idx) {
1699 if (sc->fmt)
82d4a110 1700 sc->arg_fmt[idx] = sc->fmt->arg[idx];
5e58fcfa 1701 }
82d4a110 1702
5e58fcfa
ACM		 1703 sc->nr_args = nr_args;
1704 return 0;
1705}
1706
60995604 1707static const struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {
df604bfd
ACM		 1708 { .name = "msr", .scnprintf = SCA_X86_MSR, .strtoul = STUL_X86_MSR, },
1709 { .name = "vector", .scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, },
5d88099b
ACM		 1710};
1711
1712static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)
1713{
1714 const struct syscall_arg_fmt *fmt = fmtp;
1715 return strcmp(name, fmt->name);
1716}
1717
60995604
IR		 1718static const struct syscall_arg_fmt *
1719__syscall_arg_fmt__find_by_name(const struct syscall_arg_fmt *fmts, const int nmemb,
1720 const char *name)
5d88099b
ACM		 1721{
1722 return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);
1723}
1724
60995604 1725static const struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)
5d88099b
ACM		 1726{
1727 const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);
1728 return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);
1729}
1730
8d1d4ff5
ACM		 1731static struct tep_format_field *
1732syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field)
5e58fcfa 1733{
8d1d4ff5
ACM		 1734 struct tep_format_field *last_field = NULL;
1735 int len;
5e58fcfa 1736
8d1d4ff5 1737 for (; field; field = field->next, ++arg) {
7a983a0f
ACM		 1738 last_field = field;
1739
8d1d4ff5 1740 if (arg->scnprintf)
5e58fcfa 1741 continue;
1f115cb7 1742
dea87bfb
ACM		 1743 len = strlen(field->name);
1744
82d4a110 1745 if (strcmp(field->type, "const char *") == 0 &&
dea87bfb
ACM		 1746 ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1747 strstr(field->name, "path") != NULL))
8d1d4ff5 1748 arg->scnprintf = SCA_FILENAME;
ce05539f 1749 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
8d1d4ff5 1750 arg->scnprintf = SCA_PTR;
d1d438a3 1751 else if (strcmp(field->type, "pid_t") == 0)
8d1d4ff5 1752 arg->scnprintf = SCA_PID;
ba2f22cf 1753 else if (strcmp(field->type, "umode_t") == 0)
8d1d4ff5 1754 arg->scnprintf = SCA_MODE_T;
c5e006cd 1755 else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) {
9597945d
ACM		 1756 arg->scnprintf = SCA_CHAR_ARRAY;
1757 arg->nr_entries = field->arraylen;
1758 } else if ((strcmp(field->type, "int") == 0 ||
b6565c90
ACM		 1759 strcmp(field->type, "unsigned int") == 0 ||
1760 strcmp(field->type, "long") == 0) &&
dea87bfb 1761 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
b6565c90
ACM		 1762 /*
1763 * /sys/kernel/tracing/events/syscalls/sys_enter*
818448e9 1764 * grep -E 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
b6565c90
ACM		 1765 * 65 int
1766 * 23 unsigned int
1767 * 7 unsigned long
1768 */
8d1d4ff5 1769 arg->scnprintf = SCA_FD;
60995604
IR		 1770 } else {
1771 const struct syscall_arg_fmt *fmt =
1772 syscall_arg_fmt__find_by_name(field->name);
5d88099b 1773
3f41b778 1774 if (fmt) {
5d88099b 1775 arg->scnprintf = fmt->scnprintf;
3f41b778
ACM		 1776 arg->strtoul = fmt->strtoul;
1777 }
b6565c90 1778 }
13d4ff3e
ACM		 1779 }
1780
8d1d4ff5
ACM		 1781 return last_field;
1782}
1783
1784static int syscall__set_arg_fmts(struct syscall *sc)
1785{
1786 struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args);
1787
7a983a0f
ACM		 1788 if (last_field)
1789 sc->args_size = last_field->offset + last_field->size;
1790
13d4ff3e
ACM		 1791 return 0;
1792}
1793
514f1c67
ACM		 1794static int trace__read_syscall_info(struct trace *trace, int id)
1795{
1796 char tp_name[128];
1797 struct syscall *sc;
fd0db102 1798 const char *name = syscalltbl__name(trace->sctbl, id);
3a531260 1799
d21cb73a 1800#ifdef HAVE_SYSCALL_TABLE_SUPPORT
30a910d7 1801 if (trace->syscalls.table == NULL) {
3e70008a 1802 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
30a910d7 1803 if (trace->syscalls.table == NULL)
5d2bd889 1804 return -ENOMEM;
514f1c67 1805 }
d21cb73a
ACM		 1806#else
1807 if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) {
1808 // When using libaudit we don't know beforehand what is the max syscall id
1809 struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1810
1811 if (table == NULL)
1812 return -ENOMEM;
514f1c67 1813
f3013f7e
JS		 1814 // Need to memset from offset 0 and +1 members if brand new
1815 if (trace->syscalls.table == NULL)
1816 memset(table, 0, (id + 1) * sizeof(*sc));
1817 else
1818 memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc));
d21cb73a
ACM		 1819
1820 trace->syscalls.table = table;
1821 trace->sctbl->syscalls.max_id = id;
1822 }
1823#endif
514f1c67 1824 sc = trace->syscalls.table + id;
b8b1033f 1825 if (sc->nonexistent)
d4223e17 1826 return -EEXIST;
2ae3a312 1827
b8b1033f
ACM		 1828 if (name == NULL) {
1829 sc->nonexistent = true;
d4223e17 1830 return -EEXIST;
b8b1033f
ACM		 1831 }
1832
1833 sc->name = name;
3a531260 1834 sc->fmt = syscall_fmt__find(sc->name);
514f1c67 1835
aec1930b 1836 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
97978b3e 1837 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
aec1930b 1838
8dd2a131 1839 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
aec1930b 1840 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
97978b3e 1841 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
aec1930b 1842 }
514f1c67 1843
03e9a5d8
LY		 1844 /*
1845 * Fails to read trace point format via sysfs node, so the trace point
1846 * doesn't exist. Set the 'nonexistent' flag as true.
1847 */
1848 if (IS_ERR(sc->tp_format)) {
1849 sc->nonexistent = true;
1850 return PTR_ERR(sc->tp_format);
1851 }
1852
eadcab4c
LY		 1853 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ?
1854 RAW_SYSCALL_ARGS_NUM : sc->tp_format->format.nr_fields))
5d2bd889 1855 return -ENOMEM;
5e58fcfa 1856
f208bd8d 1857 sc->args = sc->tp_format->format.fields;
c42de706
TS		 1858 /*
1859 * We need to check and discard the first variable '__syscall_nr'
1860 * or 'nr' that mean the syscall number. It is needless here.
1861 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1862 */
1863 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
f208bd8d
ACM		 1864 sc->args = sc->args->next;
1865 --sc->nr_args;
1866 }
1867
5089f20e 1868 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
6a648b53 1869 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
5089f20e 1870
13d4ff3e 1871 return syscall__set_arg_fmts(sc);
514f1c67
ACM		 1872}
1873
794bca26 1874static int evsel__init_tp_arg_scnprintf(struct evsel *evsel)
947b843c 1875{
8b913df5 1876 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
947b843c 1877
8b913df5
ACM		 1878 if (fmt != NULL) {
1879 syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields);
947b843c
ACM		 1880 return 0;
1881 }
1882
1883 return -ENOMEM;
1884}
1885
83e69b92
ACM		 1886static int intcmp(const void *a, const void *b)
1887{
1888 const int *one = a, *another = b;
1889
1890 return *one - *another;
1891}
1892
d0cc439b
ACM		 1893static int trace__validate_ev_qualifier(struct trace *trace)
1894{
a4066d64 1895 int err = 0;
04c41bcb 1896 bool printed_invalid_prefix = false;
d0cc439b 1897 struct str_node *pos;
99f26f85 1898 size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
d0cc439b 1899
99f26f85 1900 trace->ev_qualifier_ids.entries = malloc(nr_allocated *
8b3ce757
ACM		 1901 sizeof(trace->ev_qualifier_ids.entries[0]));
1902
1903 if (trace->ev_qualifier_ids.entries == NULL) {
1904 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1905 trace->output);
1906 err = -EINVAL;
1907 goto out;
1908 }
1909
602a1f4d 1910 strlist__for_each_entry(pos, trace->ev_qualifier) {
d0cc439b 1911 const char *sc = pos->s;
27702bcf 1912 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
d0cc439b 1913
8b3ce757 1914 if (id < 0) {
27702bcf
ACM		 1915 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1916 if (id >= 0)
1917 goto matches;
1918
04c41bcb
ACM		 1919 if (!printed_invalid_prefix) {
1920 pr_debug("Skipping unknown syscalls: ");
1921 printed_invalid_prefix = true;
d0cc439b 1922 } else {
04c41bcb 1923 pr_debug(", ");
d0cc439b
ACM		 1924 }
1925
04c41bcb
ACM		 1926 pr_debug("%s", sc);
1927 continue;
d0cc439b 1928 }
27702bcf 1929matches:
99f26f85 1930 trace->ev_qualifier_ids.entries[nr_used++] = id;
27702bcf
ACM		 1931 if (match_next == -1)
1932 continue;
1933
1934 while (1) {
1935 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1936 if (id < 0)
1937 break;
99f26f85 1938 if (nr_allocated == nr_used) {
27702bcf
ACM		 1939 void *entries;
1940
1941 nr_allocated += 8;
1942 entries = realloc(trace->ev_qualifier_ids.entries,
1943 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1944 if (entries == NULL) {
1945 err = -ENOMEM;
1946 fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1947 goto out_free;
1948 }
1949 trace->ev_qualifier_ids.entries = entries;
1950 }
99f26f85 1951 trace->ev_qualifier_ids.entries[nr_used++] = id;
27702bcf 1952 }
d0cc439b
ACM		 1953 }
1954
99f26f85 1955 trace->ev_qualifier_ids.nr = nr_used;
83e69b92 1956 qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
8b3ce757 1957out:
04c41bcb
ACM		 1958 if (printed_invalid_prefix)
1959 pr_debug("\n");
d0cc439b 1960 return err;
04c41bcb
ACM		 1961out_free:
1962 zfree(&trace->ev_qualifier_ids.entries);
1963 trace->ev_qualifier_ids.nr = 0;
1964 goto out;
d0cc439b
ACM		 1965}
1966
3803a229
ACM		 1967static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1968{
1969 bool in_ev_qualifier;
1970
1971 if (trace->ev_qualifier_ids.nr == 0)
1972 return true;
1973
1974 in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1975 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1976
1977 if (in_ev_qualifier)
1978 return !trace->not_ev_qualifier;
1979
1980 return trace->not_ev_qualifier;
1981}
1982
55d43bca
DA		 1983/*
1984 * args is to be interpreted as a series of longs but we need to handle
1985 * 8-byte unaligned accesses. args points to raw_data within the event
1986 * and raw_data is guaranteed to be 8-byte unaligned because it is
1987 * preceded by raw_size which is a u32. So we need to copy args to a temp
1988 * variable to read it. Most notably this avoids extended load instructions
1989 * on unaligned addresses
1990 */
325f5091 1991unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
f9f83b33
ACM		 1992{
1993 unsigned long val;
325f5091 1994 unsigned char *p = arg->args + sizeof(unsigned long) * idx;
f9f83b33
ACM		 1995
1996 memcpy(&val, p, sizeof(val));
1997 return val;
1998}
1999
c51bdfec
ACM		 2000static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
2001 struct syscall_arg *arg)
2002{
2003 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
2004 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
2005
2006 return scnprintf(bf, size, "arg%d: ", arg->idx);
2007}
2008
496fd346
ACM		 2009/*
2010 * Check if the value is in fact zero, i.e. mask whatever needs masking, such
2011 * as mount 'flags' argument that needs ignoring some magic flag, see comment
2012 * in tools/perf/trace/beauty/mount_flags.c
2013 */
3e0c9b2c 2014static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)
496fd346 2015{
3e0c9b2c
ACM		 2016 if (fmt && fmt->mask_val)
2017 return fmt->mask_val(arg, val);
496fd346
ACM		 2018
2019 return val;
2020}
2021
3e0c9b2c
ACM		 2022static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,
2023 struct syscall_arg *arg, unsigned long val)
d032d79e 2024{
3e0c9b2c 2025 if (fmt && fmt->scnprintf) {
d032d79e 2026 arg->val = val;
3e0c9b2c
ACM		 2027 if (fmt->parm)
2028 arg->parm = fmt->parm;
2029 return fmt->scnprintf(bf, size, arg);
d032d79e
ACM		 2030 }
2031 return scnprintf(bf, size, "%ld", val);
2032}
2033
752fde44 2034static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
7a983a0f
ACM		 2035 unsigned char *args, void *augmented_args, int augmented_args_size,
2036 struct trace *trace, struct thread *thread)
514f1c67 2037{
514f1c67 2038 size_t printed = 0;
55d43bca 2039 unsigned long val;
d032d79e
ACM		 2040 u8 bit = 1;
2041 struct syscall_arg arg = {
2042 .args = args,
7a983a0f
ACM		 2043 .augmented = {
2044 .size = augmented_args_size,
2045 .args = augmented_args,
2046 },
d032d79e
ACM		 2047 .idx = 0,
2048 .mask = 0,
2049 .trace = trace,
2050 .thread = thread,
c65c83ff 2051 .show_string_prefix = trace->show_string_prefix,
d032d79e 2052 };
84486caa
ACM		 2053 struct thread_trace *ttrace = thread__priv(thread);
2054
2055 /*
2056 * Things like fcntl will set this in its 'cmd' formatter to pick the
2057 * right formatter for the return value (an fd? file flags?), which is
2058 * not needed for syscalls that always return a given type, say an fd.
2059 */
2060 ttrace->ret_scnprintf = NULL;
514f1c67 2061
f208bd8d 2062 if (sc->args != NULL) {
2c92f982 2063 struct tep_format_field *field;
6e7eeb51 2064
f208bd8d 2065 for (field = sc->args; field;
01533e97
ACM		 2066 field = field->next, ++arg.idx, bit <<= 1) {
2067 if (arg.mask & bit)
6e7eeb51 2068 continue;
55d43bca 2069
888ca854 2070 arg.fmt = &sc->arg_fmt[arg.idx];
f9f83b33 2071 val = syscall_arg__val(&arg, arg.idx);
496fd346
ACM		 2072 /*
2073 * Some syscall args need some mask, most don't and
2074 * return val untouched.
2075 */
3e0c9b2c 2076 val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);
55d43bca 2077
4aa58232
ACM		 2078 /*
2079 * Suppress this argument if its value is zero and
2080 * and we don't have a string associated in an
2081 * strarray for it.
2082 */
55d43bca 2083 if (val == 0 &&
e7c634fc 2084 !trace->show_zeros &&
82d4a110 2085 !(sc->arg_fmt &&
d47737d5
ACM		 2086 (sc->arg_fmt[arg.idx].show_zero ||
2087 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
82d4a110
ACM		 2088 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
2089 sc->arg_fmt[arg.idx].parm))
22ae5cf1
ACM		 2090 continue;
2091
9d6dc178
ACM		 2092 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2093
2094 if (trace->show_arg_names)
2095 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2096
3e0c9b2c
ACM		 2097 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],
2098 bf + printed, size - printed, &arg, val);
514f1c67 2099 }
4c4d6e51
ACM		 2100 } else if (IS_ERR(sc->tp_format)) {
2101 /*
2102 * If we managed to read the tracepoint /format file, then we
2103 * may end up not having any args, like with gettid(), so only
2104 * print the raw args when we didn't manage to read it.
2105 */
332337da 2106 while (arg.idx < sc->nr_args) {
d032d79e
ACM		 2107 if (arg.mask & bit)
2108 goto next_arg;
2109 val = syscall_arg__val(&arg, arg.idx);
c51bdfec
ACM		 2110 if (printed)
2111 printed += scnprintf(bf + printed, size - printed, ", ");
2112 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
3e0c9b2c 2113 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);
d032d79e
ACM		 2114next_arg:
2115 ++arg.idx;
2116 bit <<= 1;
514f1c67
ACM		 2117 }
2118 }
2119
2120 return printed;
2121}
2122
32dcd021 2123typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
0c82adcf 2124 union perf_event *event,
ba3d7dee
ACM		 2125 struct perf_sample *sample);
2126
2127static struct syscall *trace__syscall_info(struct trace *trace,
32dcd021 2128 struct evsel *evsel, int id)
ba3d7dee 2129{
5d2bd889 2130 int err = 0;
ba3d7dee
ACM		 2131
2132 if (id < 0) {
adaa18bf
ACM		 2133
2134 /*
2135 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
2136 * before that, leaving at a higher verbosity level till that is
2137 * explained. Reproduced with plain ftrace with:
2138 *
2139 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
2140 * grep "NR -1 " /t/trace_pipe
2141 *
2142 * After generating some load on the machine.
2143 */
2144 if (verbose > 1) {
2145 static u64 n;
2146 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
8ab2e96d 2147 id, evsel__name(evsel), ++n);
adaa18bf 2148 }
ba3d7dee
ACM		 2149 return NULL;
2150 }
2151
b8b1033f
ACM		 2152 err = -EINVAL;
2153
d21cb73a
ACM		 2154#ifdef HAVE_SYSCALL_TABLE_SUPPORT
2155 if (id > trace->sctbl->syscalls.max_id) {
2156#else
2157 if (id >= trace->sctbl->syscalls.max_id) {
2158 /*
2159 * With libaudit we don't know beforehand what is the max_id,
2160 * so we let trace__read_syscall_info() figure that out as we
2161 * go on reading syscalls.
2162 */
2163 err = trace__read_syscall_info(trace, id);
2164 if (err)
2165#endif
ba3d7dee 2166 goto out_cant_read;
d21cb73a 2167 }
ba3d7dee 2168
30a910d7
ACM		 2169 if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
2170 (err = trace__read_syscall_info(trace, id)) != 0)
ba3d7dee
ACM		 2171 goto out_cant_read;
2172
03e9a5d8 2173 if (trace->syscalls.table && trace->syscalls.table[id].nonexistent)
b8b1033f 2174 goto out_cant_read;
b8b1033f 2175
ba3d7dee
ACM		 2176 return &trace->syscalls.table[id];
2177
2178out_cant_read:
bb963e16 2179 if (verbose > 0) {
5d2bd889
ACM		 2180 char sbuf[STRERR_BUFSIZE];
2181 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
30a910d7 2182 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
7c304ee0
ACM		 2183 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
2184 fputs(" information\n", trace->output);
2185 }
ba3d7dee
ACM		 2186 return NULL;
2187}
2188
8eded45f
ACM		 2189struct syscall_stats {
2190 struct stats stats;
2191 u64 nr_failures;
b88b14db
ACM		 2192 int max_errno;
2193 u32 *errnos;
8eded45f
ACM		 2194};
2195
b88b14db
ACM		 2196static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace,
2197 int id, struct perf_sample *sample, long err, bool errno_summary)
bf2575c1
DA		 2198{
2199 struct int_node *inode;
8eded45f 2200 struct syscall_stats *stats;
bf2575c1
DA		 2201 u64 duration = 0;
2202
2203 inode = intlist__findnew(ttrace->syscall_stats, id);
2204 if (inode == NULL)
2205 return;
2206
2207 stats = inode->priv;
2208 if (stats == NULL) {
e3e7572f 2209 stats = zalloc(sizeof(*stats));
bf2575c1
DA		 2210 if (stats == NULL)
2211 return;
8eded45f 2212
8eded45f 2213 init_stats(&stats->stats);
bf2575c1
DA		 2214 inode->priv = stats;
2215 }
2216
2217 if (ttrace->entry_time && sample->time > ttrace->entry_time)
2218 duration = sample->time - ttrace->entry_time;
2219
8eded45f
ACM		 2220 update_stats(&stats->stats, duration);
2221
b88b14db 2222 if (err < 0) {
8eded45f 2223 ++stats->nr_failures;
b88b14db
ACM		 2224
2225 if (!errno_summary)
2226 return;
2227
2228 err = -err;
2229 if (err > stats->max_errno) {
2230 u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32));
2231
2232 if (new_errnos) {
2233 memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32));
2234 } else {
2235 pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n",
ee84a303
IR		 2236 thread__comm_str(thread), thread__pid(thread),
2237 thread__tid(thread));
b88b14db
ACM		 2238 return;
2239 }
2240
2241 stats->errnos = new_errnos;
2242 stats->max_errno = err;
2243 }
2244
2245 ++stats->errnos[err - 1];
2246 }
bf2575c1
DA		 2247}
2248
522283fe 2249static int trace__printf_interrupted_entry(struct trace *trace)
e596663e
ACM		 2250{
2251 struct thread_trace *ttrace;
e596663e 2252 size_t printed;
172bf02d 2253 int len;
e596663e 2254
0a6545bd 2255 if (trace->failure_only || trace->current == NULL)
e596663e
ACM		 2256 return 0;
2257
2258 ttrace = thread__priv(trace->current);
2259
2260 if (!ttrace->entry_pending)
2261 return 0;
2262
522283fe 2263 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
172bf02d
ACM		 2264 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
2265
2266 if (len < trace->args_alignment - 4)
2267 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
e596663e 2268
172bf02d
ACM		 2269 printed += fprintf(trace->output, " ...\n");
2270
2271 ttrace->entry_pending = false;
5067a8cd
ACM		 2272 ++trace->nr_events_printed;
2273
e596663e
ACM		 2274 return printed;
2275}
2276
32dcd021 2277static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
591421e1
ACM		 2278 struct perf_sample *sample, struct thread *thread)
2279{
2280 int printed = 0;
2281
2282 if (trace->print_sample) {
2283 double ts = (double)sample->time / NSEC_PER_MSEC;
2284
2285 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
8ab2e96d 2286 evsel__name(evsel), ts,
591421e1
ACM		 2287 thread__comm_str(thread),
2288 sample->pid, sample->tid, sample->cpu);
2289 }
2290
2291 return printed;
2292}
2293
b9b6a2ea 2294static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
8a041f86
ACM		 2295{
2296 void *augmented_args = NULL;
cd26ea6d
ACM		 2297 /*
2298 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
b9b6a2ea
ACM		 2299 * and there we get all 6 syscall args plus the tracepoint common fields
2300 * that gets calculated at the start and the syscall_nr (another long).
2301 * So we check if that is the case and if so don't look after the
2302 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
2303 * which is fixed.
cd26ea6d
ACM		 2304 *
2305 * We'll revisit this later to pass s->args_size to the BPF augmenter
2306 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
2307 * copies only what we need for each syscall, like what happens when we
2308 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
2309 * traffic to just what is needed for each syscall.
2310 */
b9b6a2ea 2311 int args_size = raw_augmented_args_size ?: sc->args_size;
8a041f86 2312
cd26ea6d 2313 *augmented_args_size = sample->raw_size - args_size;
8a041f86 2314 if (*augmented_args_size > 0)
cd26ea6d 2315 augmented_args = sample->raw_data + args_size;
8a041f86
ACM		 2316
2317 return augmented_args;
2318}
2319
f2ebf8ff
RM		 2320static void syscall__exit(struct syscall *sc)
2321{
2322 if (!sc)
2323 return;
2324
9997d5dd 2325 zfree(&sc->arg_fmt);
f2ebf8ff
RM		 2326}
2327
32dcd021 2328static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
0c82adcf 2329 union perf_event *event __maybe_unused,
ba3d7dee
ACM		 2330 struct perf_sample *sample)
2331{
752fde44 2332 char *msg;
ba3d7dee 2333 void *args;
4b8a240e 2334 int printed = 0;
2ae3a312 2335 struct thread *thread;
b91fc39f 2336 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
db2da3f8
ACM		 2337 int augmented_args_size = 0;
2338 void *augmented_args = NULL;
bf2575c1 2339 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2ae3a312
ACM		 2340 struct thread_trace *ttrace;
2341
2342 if (sc == NULL)
2343 return -1;
ba3d7dee 2344
8fb598e5 2345 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
c24ff998 2346 ttrace = thread__trace(thread, trace->output);
2ae3a312 2347 if (ttrace == NULL)
b91fc39f 2348 goto out_put;
ba3d7dee 2349
591421e1
ACM		 2350 trace__fprintf_sample(trace, evsel, sample, thread);
2351
77170988 2352 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
752fde44
ACM		 2353
2354 if (ttrace->entry_str == NULL) {
e4d44e83 2355 ttrace->entry_str = malloc(trace__entry_str_size);
752fde44 2356 if (!ttrace->entry_str)
b91fc39f 2357 goto out_put;
752fde44
ACM		 2358 }
2359
5cf9c84e 2360 if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
522283fe 2361 trace__printf_interrupted_entry(trace);
db2da3f8
ACM		 2362 /*
2363 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2364 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2365 * this breaks syscall__augmented_args() check for augmented args, as we calculate
2366 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2367 * so when handling, say the openat syscall, we end up getting 6 args for the
2368 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2369 * thinking that the extra 2 u64 args are the augmented filename, so just check
2370 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2371 */
2372 if (evsel != trace->syscalls.events.sys_enter)
b9b6a2ea 2373 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
752fde44
ACM		 2374 ttrace->entry_time = sample->time;
2375 msg = ttrace->entry_str;
e4d44e83 2376 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
752fde44 2377
e4d44e83 2378 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
db2da3f8 2379 args, augmented_args, augmented_args_size, trace, thread);
752fde44 2380
5089f20e 2381 if (sc->is_exit) {
0a6545bd 2382 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
4b8a240e
ACM		 2383 int alignment = 0;
2384
fd2b2975 2385 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
4b8a240e
ACM		 2386 printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2387 if (trace->args_alignment > printed)
2388 alignment = trace->args_alignment - printed;
2389 fprintf(trace->output, "%*s= ?\n", alignment, " ");
ae9ed035 2390 }
7f4f8001 2391 } else {
752fde44 2392 ttrace->entry_pending = true;
7f4f8001
ACM		 2393 /* See trace__vfs_getname & trace__sys_exit */
2394 ttrace->filename.pending_open = false;
2395 }
ba3d7dee 2396
f3b623b8
ACM		 2397 if (trace->current != thread) {
2398 thread__put(trace->current);
2399 trace->current = thread__get(thread);
2400 }
b91fc39f
ACM		 2401 err = 0;
2402out_put:
2403 thread__put(thread);
2404 return err;
ba3d7dee
ACM		 2405}
2406
32dcd021 2407static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
a98392bb
ACM		 2408 struct perf_sample *sample)
2409{
a98392bb
ACM		 2410 struct thread_trace *ttrace;
2411 struct thread *thread;
f3acd886
ACM		 2412 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2413 struct syscall *sc = trace__syscall_info(trace, evsel, id);
a98392bb 2414 char msg[1024];
7a983a0f
ACM		 2415 void *args, *augmented_args = NULL;
2416 int augmented_args_size;
a98392bb 2417
a98392bb
ACM		 2418 if (sc == NULL)
2419 return -1;
2420
2421 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2422 ttrace = thread__trace(thread, trace->output);
2423 /*
2424 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2425 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2426 */
2427 if (ttrace == NULL)
2428 goto out_put;
2429
f3acd886 2430 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
b9b6a2ea 2431 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
7a983a0f 2432 syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
a98392bb
ACM		 2433 fprintf(trace->output, "%s", msg);
2434 err = 0;
2435out_put:
2436 thread__put(thread);
2437 return err;
2438}
2439
32dcd021 2440static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
5cf9c84e
ACM		 2441 struct perf_sample *sample,
2442 struct callchain_cursor *cursor)
202ff968
ACM		 2443{
2444 struct addr_location al;
1fc632ce
JO		 2445 int max_stack = evsel->core.attr.sample_max_stack ?
2446 evsel->core.attr.sample_max_stack :
3a9e9a47 2447 trace->max_stack;
0dd5041c 2448 int err = -1;
5cf9c84e 2449
0dd5041c 2450 addr_location__init(&al);
4291bf5c 2451 if (machine__resolve(trace->host, &al, sample) < 0)
0dd5041c 2452 goto out;
5cf9c84e 2453
4291bf5c 2454 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
0dd5041c
IR		 2455out:
2456 addr_location__exit(&al);
4291bf5c 2457 return err;
5cf9c84e
ACM		 2458}
2459
2460static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2461{
202ff968 2462 /* TODO: user-configurable print_opts */
e20ab86e
ACM		 2463 const unsigned int print_opts = EVSEL__PRINT_SYM |
2464 EVSEL__PRINT_DSO |
2465 EVSEL__PRINT_UNKNOWN_AS_ADDR;
202ff968 2466
8ab12a20 2467 return sample__fprintf_callchain(sample, 38, print_opts, get_tls_callchain_cursor(), symbol_conf.bt_stop_list, trace->output);
202ff968
ACM		 2468}
2469
32dcd021 2470static const char *errno_to_name(struct evsel *evsel, int err)
092bd3cd 2471{
6e6d1d65 2472 struct perf_env *env = evsel__env(evsel);
092bd3cd
HB		 2473 const char *arch_name = perf_env__arch(env);
2474
2475 return arch_syscalls__strerrno(arch_name, err);
2476}
2477
32dcd021 2478static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
0c82adcf 2479 union perf_event *event __maybe_unused,
ba3d7dee
ACM		 2480 struct perf_sample *sample)
2481{
2c82c3ad 2482 long ret;
60c907ab 2483 u64 duration = 0;
fd2b2975 2484 bool duration_calculated = false;
2ae3a312 2485 struct thread *thread;
4b8a240e
ACM		 2486 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2487 int alignment = trace->args_alignment;
bf2575c1 2488 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2ae3a312
ACM		 2489 struct thread_trace *ttrace;
2490
2491 if (sc == NULL)
2492 return -1;
ba3d7dee 2493
8fb598e5 2494 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
c24ff998 2495 ttrace = thread__trace(thread, trace->output);
2ae3a312 2496 if (ttrace == NULL)
b91fc39f 2497 goto out_put;
ba3d7dee 2498
591421e1
ACM		 2499 trace__fprintf_sample(trace, evsel, sample, thread);
2500
77170988 2501 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
ba3d7dee 2502
8eded45f 2503 if (trace->summary)
b88b14db 2504 thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary);
8eded45f 2505
79d725cd 2506 if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
7f4f8001
ACM		 2507 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2508 ttrace->filename.pending_open = false;
c522739d
ACM		 2509 ++trace->stats.vfs_getname;
2510 }
2511
ae9ed035 2512 if (ttrace->entry_time) {
60c907ab 2513 duration = sample->time - ttrace->entry_time;
ae9ed035
ACM		 2514 if (trace__filter_duration(trace, duration))
2515 goto out;
fd2b2975 2516 duration_calculated = true;
ae9ed035
ACM		 2517 } else if (trace->duration_filter)
2518 goto out;
60c907ab 2519
5cf9c84e 2520 if (sample->callchain) {
8ab12a20
IR		 2521 struct callchain_cursor *cursor = get_tls_callchain_cursor();
2522
2523 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
5cf9c84e 2524 if (callchain_ret == 0) {
8ab12a20 2525 if (cursor->nr < trace->min_stack)
5cf9c84e
ACM		 2526 goto out;
2527 callchain_ret = 1;
2528 }
2529 }
2530
0a6545bd 2531 if (trace->summary_only || (ret >= 0 && trace->failure_only))
fd2eabaf
DA		 2532 goto out;
2533
fd2b2975 2534 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
752fde44
ACM		 2535
2536 if (ttrace->entry_pending) {
4b8a240e 2537 printed = fprintf(trace->output, "%s", ttrace->entry_str);
752fde44 2538 } else {
ac6e022c 2539 printed += fprintf(trace->output, " ... [");
c24ff998 2540 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
ac6e022c
ACM		 2541 printed += 9;
2542 printed += fprintf(trace->output, "]: %s()", sc->name);
752fde44
ACM		 2543 }
2544
4b8a240e
ACM		 2545 printed++; /* the closing ')' */
2546
2547 if (alignment > printed)
2548 alignment -= printed;
2549 else
2550 alignment = 0;
2551
2552 fprintf(trace->output, ")%*s= ", alignment, " ");
2553
da3c9a44 2554 if (sc->fmt == NULL) {
1f63139c
ACM		 2555 if (ret < 0)
2556 goto errno_print;
da3c9a44 2557signed_print:
4b8a240e 2558 fprintf(trace->output, "%ld", ret);
1f63139c
ACM		 2559 } else if (ret < 0) {
2560errno_print: {
942a91ed 2561 char bf[STRERR_BUFSIZE];
c8b5f2c9 2562 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
092bd3cd 2563 *e = errno_to_name(evsel, -ret);
ba3d7dee 2564
2c83dfae 2565 fprintf(trace->output, "-1 %s (%s)", e, emsg);
1f63139c 2566 }
da3c9a44 2567 } else if (ret == 0 && sc->fmt->timeout)
2c83dfae 2568 fprintf(trace->output, "0 (Timeout)");
84486caa
ACM		 2569 else if (ttrace->ret_scnprintf) {
2570 char bf[1024];
7ee57434
ACM		 2571 struct syscall_arg arg = {
2572 .val = ret,
2573 .thread = thread,
2574 .trace = trace,
2575 };
2576 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
84486caa 2577 ttrace->ret_scnprintf = NULL;
4b8a240e 2578 fprintf(trace->output, "%s", bf);
84486caa 2579 } else if (sc->fmt->hexret)
4b8a240e 2580 fprintf(trace->output, "%#lx", ret);
11c8e39f
ACM		 2581 else if (sc->fmt->errpid) {
2582 struct thread *child = machine__find_thread(trace->host, ret, ret);
2583
2584 if (child != NULL) {
4b8a240e 2585 fprintf(trace->output, "%ld", ret);
ee84a303 2586 if (thread__comm_set(child))
11c8e39f
ACM		 2587 fprintf(trace->output, " (%s)", thread__comm_str(child));
2588 thread__put(child);
2589 }
2590 } else
da3c9a44 2591 goto signed_print;
ba3d7dee 2592
c24ff998 2593 fputc('\n', trace->output);
566a0885 2594
5067a8cd
ACM		 2595 /*
2596 * We only consider an 'event' for the sake of --max-events a non-filtered
2597 * sys_enter + sys_exit and other tracepoint events.
2598 */
2599 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2600 interrupted = true;
2601
5cf9c84e
ACM		 2602 if (callchain_ret > 0)
2603 trace__fprintf_callchain(trace, sample);
2604 else if (callchain_ret < 0)
8ab2e96d 2605 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
ae9ed035 2606out:
752fde44 2607 ttrace->entry_pending = false;
b91fc39f
ACM		 2608 err = 0;
2609out_put:
2610 thread__put(thread);
2611 return err;
ba3d7dee
ACM		 2612}
2613
32dcd021 2614static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
0c82adcf 2615 union perf_event *event __maybe_unused,
c522739d
ACM		 2616 struct perf_sample *sample)
2617{
f994592d
ACM		 2618 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2619 struct thread_trace *ttrace;
2620 size_t filename_len, entry_str_len, to_move;
2621 ssize_t remaining_space;
2622 char *pos;
efc0cdc9 2623 const char *filename = evsel__rawptr(evsel, sample, "pathname");
f994592d
ACM		 2624
2625 if (!thread)
2626 goto out;
2627
2628 ttrace = thread__priv(thread);
2629 if (!ttrace)
ef65e96e 2630 goto out_put;
f994592d 2631
7f4f8001 2632 filename_len = strlen(filename);
39f0e7a8 2633 if (filename_len == 0)
ef65e96e 2634 goto out_put;
7f4f8001
ACM		 2635
2636 if (ttrace->filename.namelen < filename_len) {
2637 char *f = realloc(ttrace->filename.name, filename_len + 1);
2638
2639 if (f == NULL)
ef65e96e 2640 goto out_put;
7f4f8001
ACM		 2641
2642 ttrace->filename.namelen = filename_len;
2643 ttrace->filename.name = f;
2644 }
2645
2646 strcpy(ttrace->filename.name, filename);
2647 ttrace->filename.pending_open = true;
2648
f994592d 2649 if (!ttrace->filename.ptr)
ef65e96e 2650 goto out_put;
f994592d
ACM		 2651
2652 entry_str_len = strlen(ttrace->entry_str);
2653 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2654 if (remaining_space <= 0)
ef65e96e 2655 goto out_put;
f994592d 2656
f994592d
ACM		 2657 if (filename_len > (size_t)remaining_space) {
2658 filename += filename_len - remaining_space;
2659 filename_len = remaining_space;
2660 }
2661
2662 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2663 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2664 memmove(pos + filename_len, pos, to_move);
2665 memcpy(pos, filename, filename_len);
2666
2667 ttrace->filename.ptr = 0;
2668 ttrace->filename.entry_str_pos = 0;
ef65e96e
ACM		 2669out_put:
2670 thread__put(thread);
f994592d 2671out:
c522739d
ACM		 2672 return 0;
2673}
2674
32dcd021 2675static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
0c82adcf 2676 union perf_event *event __maybe_unused,
1302d88e
ACM		 2677 struct perf_sample *sample)
2678{
efc0cdc9 2679 u64 runtime = evsel__intval(evsel, sample, "runtime");
1302d88e 2680 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
8fb598e5 2681 struct thread *thread = machine__findnew_thread(trace->host,
314add6b
AH		 2682 sample->pid,
2683 sample->tid);
c24ff998 2684 struct thread_trace *ttrace = thread__trace(thread, trace->output);
1302d88e
ACM		 2685
2686 if (ttrace == NULL)
2687 goto out_dump;
2688
2689 ttrace->runtime_ms += runtime_ms;
2690 trace->runtime_ms += runtime_ms;
ef65e96e 2691out_put:
b91fc39f 2692 thread__put(thread);
1302d88e
ACM		 2693 return 0;
2694
2695out_dump:
c24ff998 2696 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
1302d88e 2697 evsel->name,
efc0cdc9
ACM		 2698 evsel__strval(evsel, sample, "comm"),
2699 (pid_t)evsel__intval(evsel, sample, "pid"),
1302d88e 2700 runtime,
efc0cdc9 2701 evsel__intval(evsel, sample, "vruntime"));
ef65e96e 2702 goto out_put;
1302d88e
ACM		 2703}
2704
923d0c9a
ACM		 2705static int bpf_output__printer(enum binary_printer_ops op,
2706 unsigned int val, void *extra __maybe_unused, FILE *fp)
1d6c9407 2707{
1d6c9407
WN		 2708 unsigned char ch = (unsigned char)val;
2709
2710 switch (op) {
2711 case BINARY_PRINT_CHAR_DATA:
923d0c9a 2712 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
1d6c9407
WN		 2713 case BINARY_PRINT_DATA_BEGIN:
2714 case BINARY_PRINT_LINE_BEGIN:
2715 case BINARY_PRINT_ADDR:
2716 case BINARY_PRINT_NUM_DATA:
2717 case BINARY_PRINT_NUM_PAD:
2718 case BINARY_PRINT_SEP:
2719 case BINARY_PRINT_CHAR_PAD:
2720 case BINARY_PRINT_LINE_END:
2721 case BINARY_PRINT_DATA_END:
2722 default:
2723 break;
2724 }
923d0c9a
ACM		 2725
2726 return 0;
1d6c9407
WN		 2727}
2728
2729static void bpf_output__fprintf(struct trace *trace,
2730 struct perf_sample *sample)
2731{
923d0c9a
ACM		 2732 binary__fprintf(sample->raw_data, sample->raw_size, 8,
2733 bpf_output__printer, NULL, trace->output);
5067a8cd 2734 ++trace->nr_events_printed;
1d6c9407
WN		 2735}
2736
f11b2803
ACM		 2737static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,
2738 struct thread *thread, void *augmented_args, int augmented_args_size)
2739{
2740 char bf[2048];
2741 size_t size = sizeof(bf);
2742 struct tep_format_field *field = evsel->tp_format->format.fields;
fecd9907 2743 struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel);
f11b2803
ACM		 2744 size_t printed = 0;
2745 unsigned long val;
2746 u8 bit = 1;
2747 struct syscall_arg syscall_arg = {
2748 .augmented = {
2749 .size = augmented_args_size,
2750 .args = augmented_args,
2751 },
2752 .idx = 0,
2753 .mask = 0,
2754 .trace = trace,
2755 .thread = thread,
2756 .show_string_prefix = trace->show_string_prefix,
2757 };
2758
2759 for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {
2760 if (syscall_arg.mask & bit)
2761 continue;
2762
c5e006cd 2763 syscall_arg.len = 0;
f11b2803 2764 syscall_arg.fmt = arg;
c5e006cd
ACM		 2765 if (field->flags & TEP_FIELD_IS_ARRAY) {
2766 int offset = field->offset;
2767
2768 if (field->flags & TEP_FIELD_IS_DYNAMIC) {
2769 offset = format_field__intval(field, sample, evsel->needs_swap);
2770 syscall_arg.len = offset >> 16;
2771 offset &= 0xffff;
1634bad3 2772 if (tep_field_is_relative(field->flags))
7c689c83 2773 offset += field->offset + field->size;
c5e006cd
ACM		 2774 }
2775
2776 val = (uintptr_t)(sample->raw_data + offset);
2777 } else
f11b2803
ACM		 2778 val = format_field__intval(field, sample, evsel->needs_swap);
2779 /*
2780 * Some syscall args need some mask, most don't and
2781 * return val untouched.
2782 */
2783 val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);
2784
2785 /*
2786 * Suppress this argument if its value is zero and
632f5c22 2787 * we don't have a string associated in an
f11b2803
ACM		 2788 * strarray for it.
2789 */
2790 if (val == 0 &&
2791 !trace->show_zeros &&
2792 !((arg->show_zero ||
2793 arg->scnprintf == SCA_STRARRAY ||
2794 arg->scnprintf == SCA_STRARRAYS) &&
2795 arg->parm))
2796 continue;
2797
2798 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2799
888964a0 2800 if (trace->show_arg_names)
f11b2803
ACM		 2801 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2802
2803 printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);
2804 }
2805
2806 return printed + fprintf(trace->output, "%s", bf);
2807}
2808
32dcd021 2809static int trace__event_handler(struct trace *trace, struct evsel *evsel,
14a052df
ACM		 2810 union perf_event *event __maybe_unused,
2811 struct perf_sample *sample)
2812{
a9c5e6c1 2813 struct thread *thread;
7ad35615 2814 int callchain_ret = 0;
a9c5e6c1
ACM		 2815 /*
2816 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2817 * this event's max_events having been hit and this is an entry coming
2818 * from the ring buffer that we should discard, since the max events
2819 * have already been considered/printed.
2820 */
2821 if (evsel->disabled)
2822 return 0;
2823
2824 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
7ad35615
ACM		 2825
2826 if (sample->callchain) {
8ab12a20
IR		 2827 struct callchain_cursor *cursor = get_tls_callchain_cursor();
2828
2829 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
7ad35615 2830 if (callchain_ret == 0) {
8ab12a20 2831 if (cursor->nr < trace->min_stack)
7ad35615
ACM		 2832 goto out;
2833 callchain_ret = 1;
2834 }
2835 }
2836
522283fe 2837 trace__printf_interrupted_entry(trace);
14a052df 2838 trace__fprintf_tstamp(trace, sample->time, trace->output);
0808921a 2839
42e4a52d 2840 if (trace->trace_syscalls && trace->show_duration)
0808921a
ACM		 2841 fprintf(trace->output, "( ): ");
2842
c4191e55
ACM		 2843 if (thread)
2844 trace__fprintf_comm_tid(trace, thread, trace->output);
2845
5e6da6be 2846 if (evsel == trace->syscalls.events.bpf_output) {
1cdf618f
ACM		 2847 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2848 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2849
2850 if (sc) {
c4191e55
ACM		 2851 fprintf(trace->output, "%s(", sc->name);
2852 trace__fprintf_sys_enter(trace, evsel, sample);
2853 fputc(')', trace->output);
2854 goto newline;
1cdf618f
ACM		 2855 }
2856
2857 /*
2858 * XXX: Not having the associated syscall info or not finding/adding
2859 * the thread should never happen, but if it does...
2860 * fall thru and print it as a bpf_output event.
2861 */
2862 }
2863
311baaf9 2864 fprintf(trace->output, "%s(", evsel->name);
14a052df 2865
c754c382 2866 if (evsel__is_bpf_output(evsel)) {
1cdf618f 2867 bpf_output__fprintf(trace, sample);
1d6c9407 2868 } else if (evsel->tp_format) {
a98392bb
ACM		 2869 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2870 trace__fprintf_sys_enter(trace, evsel, sample)) {
f11b2803
ACM		 2871 if (trace->libtraceevent_print) {
2872 event_format__fprintf(evsel->tp_format, sample->cpu,
2873 sample->raw_data, sample->raw_size,
2874 trace->output);
2875 } else {
2876 trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);
2877 }
a98392bb 2878 }
14a052df
ACM		 2879 }
2880
1cdf618f 2881newline:
311baaf9 2882 fprintf(trace->output, ")\n");
202ff968 2883
7ad35615
ACM		 2884 if (callchain_ret > 0)
2885 trace__fprintf_callchain(trace, sample);
2886 else if (callchain_ret < 0)
8ab2e96d 2887 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
db25bf98
ACM		 2888
2889 ++trace->nr_events_printed;
2890
2891 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2892 evsel__disable(evsel);
2893 evsel__close(evsel);
2894 }
7ad35615 2895out:
a937c665 2896 thread__put(thread);
14a052df
ACM		 2897 return 0;
2898}
2899
598d02c5
SF		 2900static void print_location(FILE *f, struct perf_sample *sample,
2901 struct addr_location *al,
2902 bool print_dso, bool print_sym)
2903{
2904
bb963e16 2905 if ((verbose > 0 || print_dso) && al->map)
63df0e4b 2906 fprintf(f, "%s@", map__dso(al->map)->long_name);
598d02c5 2907
bb963e16 2908 if ((verbose > 0 || print_sym) && al->sym)
4414a3c5 2909 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
598d02c5
SF		 2910 al->addr - al->sym->start);
2911 else if (al->map)
4414a3c5 2912 fprintf(f, "0x%" PRIx64, al->addr);
598d02c5 2913 else
4414a3c5 2914 fprintf(f, "0x%" PRIx64, sample->addr);
598d02c5
SF		 2915}
2916
2917static int trace__pgfault(struct trace *trace,
32dcd021 2918 struct evsel *evsel,
473398a2 2919 union perf_event *event __maybe_unused,
598d02c5
SF		 2920 struct perf_sample *sample)
2921{
2922 struct thread *thread;
598d02c5
SF		 2923 struct addr_location al;
2924 char map_type = 'd';
a2ea67d7 2925 struct thread_trace *ttrace;
b91fc39f 2926 int err = -1;
1df54290 2927 int callchain_ret = 0;
598d02c5 2928
0dd5041c 2929 addr_location__init(&al);
598d02c5 2930 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1df54290
ACM		 2931
2932 if (sample->callchain) {
8ab12a20
IR		 2933 struct callchain_cursor *cursor = get_tls_callchain_cursor();
2934
2935 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
1df54290 2936 if (callchain_ret == 0) {
8ab12a20 2937 if (cursor->nr < trace->min_stack)
1df54290
ACM		 2938 goto out_put;
2939 callchain_ret = 1;
2940 }
2941 }
2942
a2ea67d7
SF		 2943 ttrace = thread__trace(thread, trace->output);
2944 if (ttrace == NULL)
b91fc39f 2945 goto out_put;
a2ea67d7 2946
1fc632ce 2947 if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
a2ea67d7
SF		 2948 ttrace->pfmaj++;
2949 else
2950 ttrace->pfmin++;
2951
2952 if (trace->summary_only)
b91fc39f 2953 goto out;
598d02c5 2954
4546263d 2955 thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
598d02c5 2956
fd2b2975 2957 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
598d02c5
SF		 2958
2959 fprintf(trace->output, "%sfault [",
1fc632ce 2960 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
598d02c5
SF		 2961 "maj" : "min");
2962
2963 print_location(trace->output, sample, &al, false, true);
2964
2965 fprintf(trace->output, "] => ");
2966
117d3c24 2967 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
598d02c5
SF		 2968
2969 if (!al.map) {
4546263d 2970 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
598d02c5
SF		 2971
2972 if (al.map)
2973 map_type = 'x';
2974 else
2975 map_type = '?';
2976 }
2977
2978 print_location(trace->output, sample, &al, true, false);
2979
2980 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
0c3a6ef4 2981
1df54290
ACM		 2982 if (callchain_ret > 0)
2983 trace__fprintf_callchain(trace, sample);
2984 else if (callchain_ret < 0)
8ab2e96d 2985 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
5067a8cd
ACM		 2986
2987 ++trace->nr_events_printed;
b91fc39f
ACM		 2988out:
2989 err = 0;
2990out_put:
2991 thread__put(thread);
0dd5041c 2992 addr_location__exit(&al);
b91fc39f 2993 return err;
598d02c5
SF		 2994}
2995
e6001980 2996static void trace__set_base_time(struct trace *trace,
32dcd021 2997 struct evsel *evsel,
e6001980
ACM		 2998 struct perf_sample *sample)
2999{
8a07a809
ACM		 3000 /*
3001 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
3002 * and don't use sample->time unconditionally, we may end up having
3003 * some other event in the future without PERF_SAMPLE_TIME for good
3004 * reason, i.e. we may not be interested in its timestamps, just in
3005 * it taking place, picking some piece of information when it
3006 * appears in our event stream (vfs_getname comes to mind).
3007 */
3008 if (trace->base_time == 0 && !trace->full_time &&
1fc632ce 3009 (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
e6001980
ACM		 3010 trace->base_time = sample->time;
3011}
3012
6810fc91 3013static int trace__process_sample(struct perf_tool *tool,
0c82adcf 3014 union perf_event *event,
6810fc91 3015 struct perf_sample *sample,
32dcd021 3016 struct evsel *evsel,
6810fc91
DA		 3017 struct machine *machine __maybe_unused)
3018{
3019 struct trace *trace = container_of(tool, struct trace, tool);
aa07df6e 3020 struct thread *thread;
6810fc91
DA		 3021 int err = 0;
3022
744a9719 3023 tracepoint_handler handler = evsel->handler;
6810fc91 3024
aa07df6e
DA		 3025 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
3026 if (thread && thread__is_filtered(thread))
ef65e96e 3027 goto out;
bdc89661 3028
e6001980 3029 trace__set_base_time(trace, evsel, sample);
6810fc91 3030
3160565f
DA		 3031 if (handler) {
3032 ++trace->nr_events;
0c82adcf 3033 handler(trace, evsel, event, sample);
3160565f 3034 }
ef65e96e
ACM		 3035out:
3036 thread__put(thread);
6810fc91
DA		 3037 return err;
3038}
3039
1e28fe0a 3040static int trace__record(struct trace *trace, int argc, const char **argv)
5e2485b1
DA		 3041{
3042 unsigned int rec_argc, i, j;
3043 const char **rec_argv;
3044 const char * const record_args[] = {
3045 "record",
3046 "-R",
3047 "-m", "1024",
3048 "-c", "1",
5e2485b1 3049 };
7fbfe22c
ACM		 3050 pid_t pid = getpid();
3051 char *filter = asprintf__tp_filter_pids(1, &pid);
1e28fe0a
SF		 3052 const char * const sc_args[] = { "-e", };
3053 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
3054 const char * const majpf_args[] = { "-e", "major-faults" };
3055 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
3056 const char * const minpf_args[] = { "-e", "minor-faults" };
3057 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
7fbfe22c 3058 int err = -1;
1e28fe0a 3059
7fbfe22c
ACM		 3060 /* +3 is for the event string below and the pid filter */
3061 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 +
1e28fe0a 3062 majpf_args_nr + minpf_args_nr + argc;
5e2485b1
DA		 3063 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3064
7fbfe22c
ACM		 3065 if (rec_argv == NULL || filter == NULL)
3066 goto out_free;
5e2485b1 3067
1e28fe0a 3068 j = 0;
5e2485b1 3069 for (i = 0; i < ARRAY_SIZE(record_args); i++)
1e28fe0a
SF		 3070 rec_argv[j++] = record_args[i];
3071
e281a960
SF		 3072 if (trace->trace_syscalls) {
3073 for (i = 0; i < sc_args_nr; i++)
3074 rec_argv[j++] = sc_args[i];
3075
3076 /* event string may be different for older kernels - e.g., RHEL6 */
3077 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
3078 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
3079 else if (is_valid_tracepoint("syscalls:sys_enter"))
3080 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
3081 else {
3082 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
7fbfe22c 3083 goto out_free;
e281a960 3084 }
9aca7f17 3085 }
9aca7f17 3086
7fbfe22c
ACM		 3087 rec_argv[j++] = "--filter";
3088 rec_argv[j++] = filter;
3089
1e28fe0a
SF		 3090 if (trace->trace_pgfaults & TRACE_PFMAJ)
3091 for (i = 0; i < majpf_args_nr; i++)
3092 rec_argv[j++] = majpf_args[i];
3093
3094 if (trace->trace_pgfaults & TRACE_PFMIN)
3095 for (i = 0; i < minpf_args_nr; i++)
3096 rec_argv[j++] = minpf_args[i];
3097
3098 for (i = 0; i < (unsigned int)argc; i++)
3099 rec_argv[j++] = argv[i];
5e2485b1 3100
7fbfe22c
ACM		 3101 err = cmd_record(j, rec_argv);
3102out_free:
3103 free(filter);
3104 free(rec_argv);
3105 return err;
5e2485b1
DA		 3106}
3107
bf2575c1
DA		 3108static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
3109
a1cf3a75 3110static bool evlist__add_vfs_getname(struct evlist *evlist)
c522739d 3111{
6ab3bc24 3112 bool found = false;
32dcd021 3113 struct evsel *evsel, *tmp;
a910e466
IR		 3114 struct parse_events_error err;
3115 int ret;
8dd2a131 3116
07eafd4e 3117 parse_events_error__init(&err);
a910e466 3118 ret = parse_events(evlist, "probe:vfs_getname*", &err);
07eafd4e
IR		 3119 parse_events_error__exit(&err);
3120 if (ret)
08c98776 3121 return false;
c522739d 3122
6ab3bc24 3123 evlist__for_each_entry_safe(evlist, evsel, tmp) {
8ab2e96d 3124 if (!strstarts(evsel__name(evsel), "probe:vfs_getname"))
6ab3bc24
ACM		 3125 continue;
3126
efc0cdc9 3127 if (evsel__field(evsel, "pathname")) {
6ab3bc24
ACM		 3128 evsel->handler = trace__vfs_getname;
3129 found = true;
3130 continue;
3131 }
3132
b27c4ece 3133 list_del_init(&evsel->core.node);
6ab3bc24 3134 evsel->evlist = NULL;
5eb2dd2a 3135 evsel__delete(evsel);
c522739d
ACM		 3136 }
3137
6ab3bc24 3138 return found;
c522739d
ACM		 3139}
3140
8f6725a2 3141static struct evsel *evsel__new_pgfault(u64 config)
598d02c5 3142{
32dcd021 3143 struct evsel *evsel;
598d02c5
SF		 3144 struct perf_event_attr attr = {
3145 .type = PERF_TYPE_SOFTWARE,
3146 .mmap_data = 1,
598d02c5
SF		 3147 };
3148
3149 attr.config = config;
0524798c 3150 attr.sample_period = 1;
598d02c5
SF		 3151
3152 event_attr_init(&attr);
3153
365c3ae7 3154 evsel = evsel__new(&attr);
0ae537cb
ACM		 3155 if (evsel)
3156 evsel->handler = trace__pgfault;
598d02c5 3157
0ae537cb 3158 return evsel;
598d02c5
SF		 3159}
3160
3cb4d5e0
RM		 3161static void evlist__free_syscall_tp_fields(struct evlist *evlist)
3162{
3163 struct evsel *evsel;
3164
3165 evlist__for_each_entry(evlist, evsel) {
7962ef13
ACM		 3166 evsel_trace__delete(evsel->priv);
3167 evsel->priv = NULL;
3cb4d5e0
RM		 3168 }
3169}
3170
ddbb1b13
ACM		 3171static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
3172{
3173 const u32 type = event->header.type;
32dcd021 3174 struct evsel *evsel;
ddbb1b13 3175
ddbb1b13
ACM		 3176 if (type != PERF_RECORD_SAMPLE) {
3177 trace__process_event(trace, trace->host, event, sample);
3178 return;
3179 }
3180
3ccf8a7b 3181 evsel = evlist__id2evsel(trace->evlist, sample->id);
ddbb1b13
ACM		 3182 if (evsel == NULL) {
3183 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
3184 return;
3185 }
3186
22ac4318
ACM		 3187 if (evswitch__discard(&trace->evswitch, evsel))
3188 return;
3189
e6001980
ACM		 3190 trace__set_base_time(trace, evsel, sample);
3191
1fc632ce 3192 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
ddbb1b13
ACM		 3193 sample->raw_data == NULL) {
3194 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
8ab2e96d 3195 evsel__name(evsel), sample->tid,
ddbb1b13
ACM		 3196 sample->cpu, sample->raw_size);
3197 } else {
3198 tracepoint_handler handler = evsel->handler;
3199 handler(trace, evsel, event, sample);
3200 }
5067a8cd
ACM		 3201
3202 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
3203 interrupted = true;
ddbb1b13
ACM		 3204}
3205
c27366f0
ACM		 3206static int trace__add_syscall_newtp(struct trace *trace)
3207{
3208 int ret = -1;
63503dba 3209 struct evlist *evlist = trace->evlist;
32dcd021 3210 struct evsel *sys_enter, *sys_exit;
c27366f0 3211
63f11c80 3212 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
c27366f0
ACM		 3213 if (sys_enter == NULL)
3214 goto out;
3215
3216 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
3217 goto out_delete_sys_enter;
3218
63f11c80 3219 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
c27366f0
ACM		 3220 if (sys_exit == NULL)
3221 goto out_delete_sys_enter;
3222
3223 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
3224 goto out_delete_sys_exit;
3225
6ec17b4e
ACM		 3226 evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
3227 evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
08e26396 3228
a1cf3a75
JO		 3229 evlist__add(evlist, sys_enter);
3230 evlist__add(evlist, sys_exit);
c27366f0 3231
2ddd5c04 3232 if (callchain_param.enabled && !trace->kernel_syscallchains) {
44621819
ACM		 3233 /*
3234 * We're interested only in the user space callchain
3235 * leading to the syscall, allow overriding that for
3236 * debugging reasons using --kernel_syscall_callchains
3237 */
1fc632ce 3238 sys_exit->core.attr.exclude_callchain_kernel = 1;
44621819
ACM		 3239 }
3240
8b3ce757
ACM		 3241 trace->syscalls.events.sys_enter = sys_enter;
3242 trace->syscalls.events.sys_exit = sys_exit;
c27366f0
ACM		 3243
3244 ret = 0;
3245out:
3246 return ret;
3247
3248out_delete_sys_exit:
5eb2dd2a 3249 evsel__delete_priv(sys_exit);
c27366f0 3250out_delete_sys_enter:
5eb2dd2a 3251 evsel__delete_priv(sys_enter);
c27366f0
ACM		 3252 goto out;
3253}
3254
246fbe03 3255static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
19867b61
ACM		 3256{
3257 int err = -1;
32dcd021 3258 struct evsel *sys_exit;
19867b61
ACM		 3259 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
3260 trace->ev_qualifier_ids.nr,
3261 trace->ev_qualifier_ids.entries);
3262
3263 if (filter == NULL)
3264 goto out_enomem;
3265
ad681adf 3266 if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) {
b15d0a4c 3267 sys_exit = trace->syscalls.events.sys_exit;
ad681adf 3268 err = evsel__append_tp_filter(sys_exit, filter);
b15d0a4c 3269 }
19867b61
ACM		 3270
3271 free(filter);
3272out:
3273 return err;
3274out_enomem:
3275 errno = ENOMEM;
3276 goto out;
3277}
c27366f0 3278
5e6da6be 3279#ifdef HAVE_BPF_SKEL
5834da7f
ACM		 3280static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
3281{
b098f336
KFL		 3282 struct bpf_program *pos, *prog = NULL;
3283 const char *sec_name;
3284
5e6da6be 3285 if (trace->skel->obj == NULL)
5834da7f
ACM		 3286 return NULL;
3287
5e6da6be 3288 bpf_object__for_each_program(pos, trace->skel->obj) {
b098f336
KFL		 3289 sec_name = bpf_program__section_name(pos);
3290 if (sec_name && !strcmp(sec_name, name)) {
3291 prog = pos;
3292 break;
3293 }
3294 }
3295
3296 return prog;
5834da7f
ACM		 3297}
3298
6ff8fff4
ACM		 3299static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
3300 const char *prog_name, const char *type)
3301{
3302 struct bpf_program *prog;
3303
8b8044e5
ACM		 3304 if (prog_name == NULL) {
3305 char default_prog_name[256];
5e6da6be 3306 scnprintf(default_prog_name, sizeof(default_prog_name), "tp/syscalls/sys_%s_%s", type, sc->name);
8b8044e5
ACM		 3307 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3308 if (prog != NULL)
3309 goto out_found;
3310 if (sc->fmt && sc->fmt->alias) {
5e6da6be 3311 scnprintf(default_prog_name, sizeof(default_prog_name), "tp/syscalls/sys_%s_%s", type, sc->fmt->alias);
8b8044e5
ACM		 3312 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3313 if (prog != NULL)
3314 goto out_found;
3315 }
6ff8fff4 3316 goto out_unaugmented;
8b8044e5 3317 }
6ff8fff4
ACM		 3318
3319 prog = trace__find_bpf_program_by_title(trace, prog_name);
8b8044e5
ACM		 3320
3321 if (prog != NULL) {
3322out_found:
6ff8fff4 3323 return prog;
8b8044e5 3324 }
6ff8fff4
ACM		 3325
3326 pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
3327 prog_name, type, sc->name);
3328out_unaugmented:
5e6da6be 3329 return trace->skel->progs.syscall_unaugmented;
6ff8fff4
ACM		 3330}
3331
3332static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
3333{
3334 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3335
3336 if (sc == NULL)
3337 return;
3338
8b8044e5
ACM		 3339 sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3340 sc->bpf_prog.sys_exit = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit : NULL, "exit");
6ff8fff4
ACM		 3341}
3342
3803a229
ACM		 3343static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
3344{
3345 struct syscall *sc = trace__syscall_info(trace, NULL, id);
5e6da6be 3346 return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->skel->progs.syscall_unaugmented);
3803a229
ACM		 3347}
3348
3349static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
3350{
3351 struct syscall *sc = trace__syscall_info(trace, NULL, id);
5e6da6be 3352 return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->skel->progs.syscall_unaugmented);
3803a229
ACM		 3353}
3354
ad4153f9
ACM		 3355static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
3356{
3357 struct tep_format_field *field, *candidate_field;
3358 int id;
3359
3360 /*
3361 * We're only interested in syscalls that have a pointer:
3362 */
3363 for (field = sc->args; field; field = field->next) {
3364 if (field->flags & TEP_FIELD_IS_POINTER)
3365 goto try_to_find_pair;
3366 }
3367
3368 return NULL;
3369
3370try_to_find_pair:
3371 for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) {
3372 struct syscall *pair = trace__syscall_info(trace, NULL, id);
3373 struct bpf_program *pair_prog;
3374 bool is_candidate = false;
3375
3376 if (pair == NULL || pair == sc ||
5e6da6be 3377 pair->bpf_prog.sys_enter == trace->skel->progs.syscall_unaugmented)
ad4153f9
ACM		 3378 continue;
3379
3380 for (field = sc->args, candidate_field = pair->args;
3381 field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3382 bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3383 candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3384
3385 if (is_pointer) {
3386 if (!candidate_is_pointer) {
3387 // The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3388 continue;
3389 }
3390 } else {
3391 if (candidate_is_pointer) {
3392 // The candidate might copy a pointer we don't have, skip it.
3393 goto next_candidate;
3394 }
3395 continue;
3396 }
3397
3398 if (strcmp(field->type, candidate_field->type))
3399 goto next_candidate;
3400
64917f4d
ACM		 3401 /*
3402 * This is limited in the BPF program but sys_write
3403 * uses "const char *" for its "buf" arg so we need to
3404 * use some heuristic that is kinda future proof...
3405 */
3406 if (strcmp(field->type, "const char *") == 0 &&
3407 !(strstr(field->name, "name") ||
3408 strstr(field->name, "path") ||
3409 strstr(field->name, "file") ||
3410 strstr(field->name, "root") ||
3411 strstr(field->name, "description")))
3412 goto next_candidate;
3413
ad4153f9
ACM		 3414 is_candidate = true;
3415 }
3416
3417 if (!is_candidate)
3418 goto next_candidate;
3419
3420 /*
3421 * Check if the tentative pair syscall augmenter has more pointers, if it has,
3422 * then it may be collecting that and we then can't use it, as it would collect
3423 * more than what is common to the two syscalls.
3424 */
3425 if (candidate_field) {
3426 for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3427 if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3428 goto next_candidate;
3429 }
3430
3431 pair_prog = pair->bpf_prog.sys_enter;
3432 /*
3433 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3434 * have been searched for, so search it here and if it returns the
3435 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3436 * program for a filtered syscall on a non-filtered one.
3437 *
3438 * For instance, we have "!syscalls:sys_enter_renameat" and that is
3439 * useful for "renameat2".
3440 */
3441 if (pair_prog == NULL) {
3442 pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
5e6da6be 3443 if (pair_prog == trace->skel->progs.syscall_unaugmented)
ad4153f9
ACM		 3444 goto next_candidate;
3445 }
3446
3447 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3448 return pair_prog;
3449 next_candidate:
3450 continue;
3451 }
3452
3453 return NULL;
3454}
3455
3803a229
ACM		 3456static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3457{
5e6da6be
IR		 3458 int map_enter_fd = bpf_map__fd(trace->skel->maps.syscalls_sys_enter);
3459 int map_exit_fd = bpf_map__fd(trace->skel->maps.syscalls_sys_exit);
3803a229
ACM		 3460 int err = 0, key;
3461
3462 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3463 int prog_fd;
3464
3465 if (!trace__syscall_enabled(trace, key))
3466 continue;
3467
3468 trace__init_syscall_bpf_progs(trace, key);
3469
3470 // It'll get at least the "!raw_syscalls:unaugmented"
3471 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3472 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3473 if (err)
3474 break;
3475 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3476 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3477 if (err)
3478 break;
3479 }
3480
ad4153f9
ACM		 3481 /*
3482 * Now lets do a second pass looking for enabled syscalls without
3483 * an augmenter that have a signature that is a superset of another
3484 * syscall with an augmenter so that we can auto-reuse it.
3485 *
3486 * I.e. if we have an augmenter for the "open" syscall that has
3487 * this signature:
3488 *
3489 * int open(const char *pathname, int flags, mode_t mode);
3490 *
3491 * I.e. that will collect just the first string argument, then we
3492 * can reuse it for the 'creat' syscall, that has this signature:
3493 *
3494 * int creat(const char *pathname, mode_t mode);
3495 *
3496 * and for:
3497 *
3498 * int stat(const char *pathname, struct stat *statbuf);
3499 * int lstat(const char *pathname, struct stat *statbuf);
3500 *
3501 * Because the 'open' augmenter will collect the first arg as a string,
3502 * and leave alone all the other args, which already helps with
3503 * beautifying 'stat' and 'lstat''s pathname arg.
3504 *
3505 * Then, in time, when 'stat' gets an augmenter that collects both
3506 * first and second arg (this one on the raw_syscalls:sys_exit prog
3507 * array tail call, then that one will be used.
3508 */
3509 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3510 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3511 struct bpf_program *pair_prog;
3512 int prog_fd;
3513
3514 if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3515 continue;
3516
3517 /*
3518 * For now we're just reusing the sys_enter prog, and if it
3519 * already has an augmenter, we don't need to find one.
3520 */
5e6da6be 3521 if (sc->bpf_prog.sys_enter != trace->skel->progs.syscall_unaugmented)
ad4153f9
ACM		 3522 continue;
3523
3524 /*
3525 * Look at all the other syscalls for one that has a signature
3526 * that is close enough that we can share:
3527 */
3528 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3529 if (pair_prog == NULL)
3530 continue;
3531
3532 sc->bpf_prog.sys_enter = pair_prog;
3533
3534 /*
3535 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3536 * with the fd for the program we're reusing:
3537 */
3538 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3539 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3540 if (err)
3541 break;
3542 }
3543
3803a229
ACM		 3544 return err;
3545}
5e6da6be 3546#endif // HAVE_BPF_SKEL
b27b38ed 3547
246fbe03
ACM		 3548static int trace__set_ev_qualifier_filter(struct trace *trace)
3549{
f76214f9
ACM		 3550 if (trace->syscalls.events.sys_enter)
3551 return trace__set_ev_qualifier_tp_filter(trace);
3552 return 0;
246fbe03
ACM		 3553}
3554
a9964c43
ACM		 3555static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3556 size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3557{
3558 int err = 0;
3559#ifdef HAVE_LIBBPF_SUPPORT
3560 bool value = true;
3561 int map_fd = bpf_map__fd(map);
3562 size_t i;
3563
3564 for (i = 0; i < npids; ++i) {
3565 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3566 if (err)
3567 break;
3568 }
3569#endif
3570 return err;
3571}
3572
dd1a5037
ACM		 3573static int trace__set_filter_loop_pids(struct trace *trace)
3574{
a9964c43 3575 unsigned int nr = 1, err;
dd1a5037
ACM		 3576 pid_t pids[32] = {
3577 getpid(),
3578 };
082ab9a1
ACM		 3579 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3580
3581 while (thread && nr < ARRAY_SIZE(pids)) {
ee84a303
IR		 3582 struct thread *parent = machine__find_thread(trace->host,
3583 thread__ppid(thread),
3584 thread__ppid(thread));
082ab9a1
ACM		 3585
3586 if (parent == NULL)
3587 break;
3588
ca2da70c
ACM		 3589 if (!strcmp(thread__comm_str(parent), "sshd") ||
3590 strstarts(thread__comm_str(parent), "gnome-terminal")) {
ee84a303 3591 pids[nr++] = thread__tid(parent);
082ab9a1
ACM		 3592 break;
3593 }
3594 thread = parent;
3595 }
dd1a5037 3596
24bf91a7 3597 err = evlist__append_tp_filter_pids(trace->evlist, nr, pids);
a9964c43
ACM		 3598 if (!err && trace->filter_pids.map)
3599 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3600
3601 return err;
3602}
3603
3604static int trace__set_filter_pids(struct trace *trace)
3605{
3606 int err = 0;
3607 /*
3608 * Better not use !target__has_task() here because we need to cover the
3609 * case where no threads were specified in the command line, but a
3610 * workload was, and in that case we will fill in the thread_map when
7b392ef0 3611 * we fork the workload in evlist__prepare_workload.
a9964c43
ACM		 3612 */
3613 if (trace->filter_pids.nr > 0) {
24bf91a7
ACM		 3614 err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3615 trace->filter_pids.entries);
a9964c43
ACM		 3616 if (!err && trace->filter_pids.map) {
3617 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3618 trace->filter_pids.entries);
3619 }
a2f354e3 3620 } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
a9964c43
ACM		 3621 err = trace__set_filter_loop_pids(trace);
3622 }
3623
3624 return err;
dd1a5037
ACM		 3625}
3626
7ba61524 3627static int __trace__deliver_event(struct trace *trace, union perf_event *event)
1f44b3e2 3628{
63503dba 3629 struct evlist *evlist = trace->evlist;
1f44b3e2 3630 struct perf_sample sample;
2a6599cd 3631 int err = evlist__parse_sample(evlist, event, &sample);
1f44b3e2 3632
1f44b3e2
JO		 3633 if (err)
3634 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3635 else
3636 trace__handle_event(trace, event, &sample);
3637
3638 return 0;
3639}
3640
a761a8d1 3641static int __trace__flush_events(struct trace *trace)
028713aa
JO		 3642{
3643 u64 first = ordered_events__first_time(&trace->oe.data);
3644 u64 flush = trace->oe.last - NSEC_PER_SEC;
3645
3646 /* Is there some thing to flush.. */
3647 if (first && first < flush)
3648 return ordered_events__flush_time(&trace->oe.data, flush);
3649
3650 return 0;
3651}
3652
a761a8d1
ACM		 3653static int trace__flush_events(struct trace *trace)
3654{
3655 return !trace->sort_events ? 0 : __trace__flush_events(trace);
3656}
3657
7ba61524 3658static int trace__deliver_event(struct trace *trace, union perf_event *event)
028713aa 3659{
028713aa
JO		 3660 int err;
3661
a761a8d1
ACM		 3662 if (!trace->sort_events)
3663 return __trace__deliver_event(trace, event);
3664
2a6599cd 3665 err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
028713aa
JO		 3666 if (err && err != -1)
3667 return err;
3668
2292083f 3669 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0, NULL);
028713aa
JO		 3670 if (err)
3671 return err;
3672
7ba61524 3673 return trace__flush_events(trace);
028713aa
JO		 3674}
3675
3676static int ordered_events__deliver_event(struct ordered_events *oe,
3677 struct ordered_event *event)
3678{
3679 struct trace *trace = container_of(oe, struct trace, oe.data);
3680
7ba61524 3681 return __trace__deliver_event(trace, event->event);
028713aa
JO		 3682}
3683
794bca26 3684static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg)
90df0249
ACM		 3685{
3686 struct tep_format_field *field;
fecd9907 3687 struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel);
90df0249
ACM		 3688
3689 if (evsel->tp_format == NULL || fmt == NULL)
3690 return NULL;
3691
3692 for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt)
3693 if (strcmp(field->name, arg) == 0)
3694 return fmt;
3695
3696 return NULL;
3697}
3698
3699static int trace__expand_filter(struct trace *trace __maybe_unused, struct evsel *evsel)
3700{
3701 char *tok, *left = evsel->filter, *new_filter = evsel->filter;
3702
3703 while ((tok = strpbrk(left, "=<>!")) != NULL) {
3704 char *right = tok + 1, *right_end;
3705
3706 if (*right == '=')
3707 ++right;
3708
3709 while (isspace(*right))
3710 ++right;
3711
3712 if (*right == '\0')
3713 break;
3714
3715 while (!isalpha(*left))
3716 if (++left == tok) {
3717 /*
3718 * Bail out, can't find the name of the argument that is being
3719 * used in the filter, let it try to set this filter, will fail later.
3720 */
3721 return 0;
3722 }
3723
3724 right_end = right + 1;
154c978d 3725 while (isalnum(*right_end) || *right_end == '_' || *right_end == '|')
90df0249
ACM		 3726 ++right_end;
3727
3728 if (isalpha(*right)) {
3729 struct syscall_arg_fmt *fmt;
3730 int left_size = tok - left,
3731 right_size = right_end - right;
3732 char arg[128];
3733
3734 while (isspace(left[left_size - 1]))
3735 --left_size;
3736
3737 scnprintf(arg, sizeof(arg), "%.*s", left_size, left);
3738
794bca26 3739 fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg);
90df0249 3740 if (fmt == NULL) {
3cdc8db9
ACM		 3741 pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n",
3742 arg, evsel->name, evsel->filter);
90df0249
ACM		 3743 return -1;
3744 }
3745
3746 pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ",
3747 arg, (int)(right - tok), tok, right_size, right);
3748
3749 if (fmt->strtoul) {
3750 u64 val;
9afec87e
ACM		 3751 struct syscall_arg syscall_arg = {
3752 .parm = fmt->parm,
3753 };
3754
3755 if (fmt->strtoul(right, right_size, &syscall_arg, &val)) {
90df0249
ACM		 3756 char *n, expansion[19];
3757 int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val);
3758 int expansion_offset = right - new_filter;
3759
3760 pr_debug("%s", expansion);
3761
3762 if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) {
3763 pr_debug(" out of memory!\n");
3764 free(new_filter);
3765 return -1;
3766 }
3767 if (new_filter != evsel->filter)
3768 free(new_filter);
3769 left = n + expansion_offset + expansion_lenght;
3770 new_filter = n;
3771 } else {
3772 pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3773 right_size, right, arg, evsel->name, evsel->filter);
3774 return -1;
3775 }
3776 } else {
3777 pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3778 arg, evsel->name, evsel->filter);
3779 return -1;
3780 }
3781
3782 pr_debug("\n");
3783 } else {
3784 left = right_end;
3785 }
3786 }
3787
3788 if (new_filter != evsel->filter) {
3789 pr_debug("New filter for %s: %s\n", evsel->name, new_filter);
ad681adf 3790 evsel__set_filter(evsel, new_filter);
90df0249
ACM		 3791 free(new_filter);
3792 }
3793
3794 return 0;
3795}
3796
3797static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel)
3798{
3799 struct evlist *evlist = trace->evlist;
3800 struct evsel *evsel;
3801
3802 evlist__for_each_entry(evlist, evsel) {
3803 if (evsel->filter == NULL)
3804 continue;
3805
3806 if (trace__expand_filter(trace, evsel)) {
3807 *err_evsel = evsel;
3808 return -1;
3809 }
3810 }
3811
3812 return 0;
3813}
3814
f15eb531 3815static int trace__run(struct trace *trace, int argc, const char **argv)
514f1c67 3816{
63503dba 3817 struct evlist *evlist = trace->evlist;
32dcd021 3818 struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
efd5745e
ACM		 3819 int err = -1, i;
3820 unsigned long before;
f15eb531 3821 const bool forks = argc > 0;
46fb3c21 3822 bool draining = false;
514f1c67 3823
75b757ca
ACM		 3824 trace->live = true;
3825
0df50e0b
ACM		 3826 if (!trace->raw_augmented_syscalls) {
3827 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3828 goto out_error_raw_syscalls;
514f1c67 3829
0df50e0b 3830 if (trace->trace_syscalls)
a1cf3a75 3831 trace->vfs_getname = evlist__add_vfs_getname(evlist);
0df50e0b 3832 }
c522739d 3833
0ae537cb 3834 if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
8f6725a2 3835 pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
0ae537cb
ACM		 3836 if (pgfault_maj == NULL)
3837 goto out_error_mem;
6ec17b4e 3838 evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
a1cf3a75 3839 evlist__add(evlist, pgfault_maj);
e2726d99 3840 }
598d02c5 3841
0ae537cb 3842 if ((trace->trace_pgfaults & TRACE_PFMIN)) {
8f6725a2 3843 pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
0ae537cb
ACM		 3844 if (pgfault_min == NULL)
3845 goto out_error_mem;
6ec17b4e 3846 evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
a1cf3a75 3847 evlist__add(evlist, pgfault_min);
0ae537cb 3848 }
598d02c5 3849
b4515ad6
GL		 3850 /* Enable ignoring missing threads when -u/-p option is defined. */
3851 trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid;
3852
1302d88e 3853 if (trace->sched &&
e251abee 3854 evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime))
2cc990ba 3855 goto out_error_sched_stat_runtime;
9ea42ba4
ACM		 3856 /*
3857 * If a global cgroup was set, apply it to all the events without an
3858 * explicit cgroup. I.e.:
3859 *
3860 * trace -G A -e sched:*switch
3861 *
3862 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3863 * _and_ sched:sched_switch to the 'A' cgroup, while:
3864 *
3865 * trace -e sched:*switch -G A
3866 *
3867 * will only set the sched:sched_switch event to the 'A' cgroup, all the
3868 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3869 * a cgroup (on the root cgroup, sys wide, etc).
3870 *
3871 * Multiple cgroups:
3872 *
3873 * trace -G A -e sched:*switch -G B
3874 *
3875 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
3876 * to the 'B' cgroup.
3877 *
3878 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
3879 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
3880 */
3881 if (trace->cgroup)
3882 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
3883
7748bb71 3884 err = evlist__create_maps(evlist, &trace->opts.target);
514f1c67 3885 if (err < 0) {
c24ff998 3886 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
514f1c67
ACM		 3887 goto out_delete_evlist;
3888 }
3889
752fde44
ACM		 3890 err = trace__symbols_init(trace, evlist);
3891 if (err < 0) {
c24ff998 3892 fprintf(trace->output, "Problems initializing symbol libraries!\n");
03ad9747 3893 goto out_delete_evlist;
752fde44
ACM		 3894 }
3895
78e1bc25 3896 evlist__config(evlist, &trace->opts, &callchain_param);
fde54b78 3897
f15eb531 3898 if (forks) {
7b392ef0 3899 err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL);
f15eb531 3900 if (err < 0) {
c24ff998 3901 fprintf(trace->output, "Couldn't run the workload!\n");
03ad9747 3902 goto out_delete_evlist;
f15eb531 3903 }
de9f498d 3904 workload_pid = evlist->workload.pid;
f15eb531
NK		 3905 }
3906
474ddc4c 3907 err = evlist__open(evlist);
a8f23d8f
ACM		 3908 if (err < 0)
3909 goto out_error_open;
5e6da6be 3910#ifdef HAVE_BPF_SKEL
83a0943b 3911 if (trace->syscalls.events.bpf_output) {
5e6da6be 3912 struct perf_cpu cpu;
514f1c67 3913
5e6da6be
IR		 3914 /*
3915 * Set up the __augmented_syscalls__ BPF map to hold for each
3916 * CPU the bpf-output event's file descriptor.
3917 */
3918 perf_cpu_map__for_each_cpu(cpu, i, trace->syscalls.events.bpf_output->core.cpus) {
3919 bpf_map__update_elem(trace->skel->maps.__augmented_syscalls__,
3920 &cpu.cpu, sizeof(int),
3921 xyarray__entry(trace->syscalls.events.bpf_output->core.fd,
3922 cpu.cpu, 0),
3923 sizeof(__u32), BPF_ANY);
3924 }
3925 }
3926#endif
a9964c43 3927 err = trace__set_filter_pids(trace);
94ad89bc
ACM		 3928 if (err < 0)
3929 goto out_error_mem;
3930
5e6da6be 3931#ifdef HAVE_BPF_SKEL
83a0943b 3932 if (trace->skel && trace->skel->progs.sys_enter)
3803a229 3933 trace__init_syscalls_bpf_prog_array_maps(trace);
5e6da6be 3934#endif
3803a229 3935
19867b61
ACM		 3936 if (trace->ev_qualifier_ids.nr > 0) {
3937 err = trace__set_ev_qualifier_filter(trace);
3938 if (err < 0)
3939 goto out_errno;
19867b61 3940
0df50e0b
ACM		 3941 if (trace->syscalls.events.sys_exit) {
3942 pr_debug("event qualifier tracepoint filter: %s\n",
3943 trace->syscalls.events.sys_exit->filter);
3944 }
2e5e5f87 3945 }
19867b61 3946
79d725cd
ACM		 3947 /*
3948 * If the "close" syscall is not traced, then we will not have the
3949 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
3950 * fd->pathname table and were ending up showing the last value set by
3951 * syscalls opening a pathname and associating it with a descriptor or
3952 * reading it from /proc/pid/fd/ in cases where that doesn't make
3953 * sense.
3954 *
3955 * So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
3956 * not in use.
3957 */
3958 trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
3959
90df0249
ACM		 3960 err = trace__expand_filters(trace, &evsel);
3961 if (err)
3962 goto out_delete_evlist;
24bf91a7 3963 err = evlist__apply_filters(evlist, &evsel);
94ad89bc
ACM		 3964 if (err < 0)
3965 goto out_error_apply_filters;
241b057c 3966
9521b5f2 3967 err = evlist__mmap(evlist, trace->opts.mmap_pages);
e09b18d4
ACM		 3968 if (err < 0)
3969 goto out_error_mmap;
514f1c67 3970
cb4b9e68 3971 if (!target__none(&trace->opts.target) && !trace->opts.target.initial_delay)
1c87f165 3972 evlist__enable(evlist);
cb24d01d 3973
f15eb531 3974 if (forks)
7b392ef0 3975 evlist__start_workload(evlist);
f15eb531 3976
cb4b9e68
CD		 3977 if (trace->opts.target.initial_delay) {
3978 usleep(trace->opts.target.initial_delay * 1000);
1c87f165 3979 evlist__enable(evlist);
e36b7821
AB		 3980 }
3981
a2f354e3 3982 trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
fd3f518f
IR		 3983 perf_thread_map__nr(evlist->core.threads) > 1 ||
3984 evlist__first(evlist)->core.attr.inherit;
bd3dda9a
ACM		 3985
3986 /*
1fc632ce
JO		 3987 * Now that we already used evsel->core.attr to ask the kernel to setup the
3988 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
bd3dda9a 3989 * trace__resolve_callchain(), allowing per-event max-stack settings
adba1634 3990 * to override an explicitly set --max-stack global setting.
bd3dda9a
ACM		 3991 */
3992 evlist__for_each_entry(evlist, evsel) {
27de9b2b 3993 if (evsel__has_callchain(evsel) &&
1fc632ce
JO		 3994 evsel->core.attr.sample_max_stack == 0)
3995 evsel->core.attr.sample_max_stack = trace->max_stack;
bd3dda9a 3996 }
514f1c67 3997again:
efd5745e 3998 before = trace->nr_events;
514f1c67 3999
c976ee11 4000 for (i = 0; i < evlist->core.nr_mmaps; i++) {
514f1c67 4001 union perf_event *event;
a5830532 4002 struct mmap *md;
514f1c67 4003
d7f55c62 4004 md = &evlist->mmap[i];
7c4d4182 4005 if (perf_mmap__read_init(&md->core) < 0)
d7f55c62
KL		 4006 continue;
4007
151ed5d7 4008 while ((event = perf_mmap__read_event(&md->core)) != NULL) {
efd5745e 4009 ++trace->nr_events;
514f1c67 4010
7ba61524 4011 err = trace__deliver_event(trace, event);
028713aa
JO		 4012 if (err)
4013 goto out_disable;
514f1c67 4014
7728fa0c 4015 perf_mmap__consume(&md->core);
20c5f10e 4016
ba209f85
ACM		 4017 if (interrupted)
4018 goto out_disable;
02ac5421
ACM		 4019
4020 if (done && !draining) {
e74676de 4021 evlist__disable(evlist);
02ac5421
ACM		 4022 draining = true;
4023 }
514f1c67 4024 }
32fdc2ca 4025 perf_mmap__read_done(&md->core);
514f1c67
ACM		 4026 }
4027
efd5745e 4028 if (trace->nr_events == before) {
ba209f85 4029 int timeout = done ? 100 : -1;
f15eb531 4030
80ab2987 4031 if (!draining && evlist__poll(evlist, timeout) > 0) {
f4009e7b 4032 if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
46fb3c21
ACM		 4033 draining = true;
4034
ba209f85 4035 goto again;
028713aa 4036 } else {
7ba61524 4037 if (trace__flush_events(trace))
028713aa 4038 goto out_disable;
46fb3c21 4039 }
ba209f85
ACM		 4040 } else {
4041 goto again;
f15eb531
NK		 4042 }
4043
ba209f85 4044out_disable:
f3b623b8
ACM		 4045 thread__zput(trace->current);
4046
e74676de 4047 evlist__disable(evlist);
514f1c67 4048
a761a8d1
ACM		 4049 if (trace->sort_events)
4050 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
028713aa 4051
c522739d
ACM		 4052 if (!err) {
4053 if (trace->summary)
4054 trace__fprintf_thread_summary(trace, trace->output);
4055
4056 if (trace->show_tool_stats) {
4057 fprintf(trace->output, "Stats:\n "
4058 " vfs_getname : %" PRIu64 "\n"
4059 " proc_getname: %" PRIu64 "\n",
4060 trace->stats.vfs_getname,
4061 trace->stats.proc_getname);
4062 }
4063 }
bf2575c1 4064
514f1c67 4065out_delete_evlist:
33974a41 4066 trace__symbols__exit(trace);
3cb4d5e0 4067 evlist__free_syscall_tp_fields(evlist);
c12995a5 4068 evlist__delete(evlist);
9ea42ba4 4069 cgroup__put(trace->cgroup);
14a052df 4070 trace->evlist = NULL;
75b757ca 4071 trace->live = false;
514f1c67 4072 return err;
6ef068cb
ACM		 4073{
4074 char errbuf[BUFSIZ];
a8f23d8f 4075
2cc990ba 4076out_error_sched_stat_runtime:
988bdb31 4077 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
2cc990ba
ACM		 4078 goto out_error;
4079
801c67b0 4080out_error_raw_syscalls:
988bdb31 4081 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
a8f23d8f
ACM		 4082 goto out_error;
4083
e09b18d4 4084out_error_mmap:
d1f249ec 4085 evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
e09b18d4
ACM		 4086 goto out_error;
4087
a8f23d8f 4088out_error_open:
d1f249ec 4089 evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
a8f23d8f
ACM		 4090
4091out_error:
6ef068cb 4092 fprintf(trace->output, "%s\n", errbuf);
87f91868 4093 goto out_delete_evlist;
94ad89bc
ACM		 4094
4095out_error_apply_filters:
4096 fprintf(trace->output,
4097 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
8ab2e96d 4098 evsel->filter, evsel__name(evsel), errno,
c8b5f2c9 4099 str_error_r(errno, errbuf, sizeof(errbuf)));
94ad89bc 4100 goto out_delete_evlist;
514f1c67 4101}
5ed08dae
ACM		 4102out_error_mem:
4103 fprintf(trace->output, "Not enough memory to run!\n");
4104 goto out_delete_evlist;
19867b61
ACM		 4105
4106out_errno:
4107 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
4108 goto out_delete_evlist;
a8f23d8f 4109}
514f1c67 4110
6810fc91
DA		 4111static int trace__replay(struct trace *trace)
4112{
32dcd021 4113 const struct evsel_str_handler handlers[] = {
c522739d 4114 { "probe:vfs_getname", trace__vfs_getname, },
6810fc91 4115 };
8ceb41d7 4116 struct perf_data data = {
2d4f2799
JO		 4117 .path = input_name,
4118 .mode = PERF_DATA_MODE_READ,
4119 .force = trace->force,
f5fc1412 4120 };
6810fc91 4121 struct perf_session *session;
32dcd021 4122 struct evsel *evsel;
6810fc91
DA		 4123 int err = -1;
4124
4125 trace->tool.sample = trace__process_sample;
4126 trace->tool.mmap = perf_event__process_mmap;
384c671e 4127 trace->tool.mmap2 = perf_event__process_mmap2;
6810fc91
DA		 4128 trace->tool.comm = perf_event__process_comm;
4129 trace->tool.exit = perf_event__process_exit;
4130 trace->tool.fork = perf_event__process_fork;
4131 trace->tool.attr = perf_event__process_attr;
f3b3614a 4132 trace->tool.tracing_data = perf_event__process_tracing_data;
6810fc91 4133 trace->tool.build_id = perf_event__process_build_id;
f3b3614a 4134 trace->tool.namespaces = perf_event__process_namespaces;
6810fc91 4135
0a8cb85c 4136 trace->tool.ordered_events = true;
6810fc91
DA		 4137 trace->tool.ordering_requires_timestamps = true;
4138
4139 /* add tid to output */
4140 trace->multiple_threads = true;
4141
2681bd85 4142 session = perf_session__new(&data, &trace->tool);
6ef81c55
MI		 4143 if (IS_ERR(session))
4144 return PTR_ERR(session);
6810fc91 4145
aa07df6e
DA		 4146 if (trace->opts.target.pid)
4147 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
4148
4149 if (trace->opts.target.tid)
4150 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
4151
0a7e6d1b 4152 if (symbol__init(&session->header.env) < 0)
cb2ffae2
NK		 4153 goto out;
4154
8fb598e5
DA		 4155 trace->host = &session->machines.host;
4156
6810fc91
DA		 4157 err = perf_session__set_tracepoints_handlers(session, handlers);
4158 if (err)
4159 goto out;
4160
b02736f7 4161 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter");
4b335e1e 4162 trace->syscalls.events.sys_enter = evsel;
9aca7f17
DA		 4163 /* older kernels have syscalls tp versus raw_syscalls */
4164 if (evsel == NULL)
b02736f7 4165 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter");
003824e8 4166
e281a960 4167 if (evsel &&
794bca26 4168 (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
e281a960 4169 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
003824e8
NK		 4170 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
4171 goto out;
4172 }
4173
b02736f7 4174 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit");
4b335e1e 4175 trace->syscalls.events.sys_exit = evsel;
9aca7f17 4176 if (evsel == NULL)
b02736f7 4177 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit");
e281a960 4178 if (evsel &&
794bca26 4179 (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
e281a960 4180 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
003824e8 4181 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
6810fc91
DA		 4182 goto out;
4183 }
4184
e5cadb93 4185 evlist__for_each_entry(session->evlist, evsel) {
1fc632ce
JO		 4186 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
4187 (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
4188 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
4189 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
1e28fe0a
SF		 4190 evsel->handler = trace__pgfault;
4191 }
4192
6810fc91
DA		 4193 setup_pager();
4194
b7b61cbe 4195 err = perf_session__process_events(session);
6810fc91
DA		 4196 if (err)
4197 pr_err("Failed to process events, error %d", err);
4198
bf2575c1
DA		 4199 else if (trace->summary)
4200 trace__fprintf_thread_summary(trace, trace->output);
4201
6810fc91
DA		 4202out:
4203 perf_session__delete(session);
4204
4205 return err;
4206}
4207
1302d88e
ACM		 4208static size_t trace__fprintf_threads_header(FILE *fp)
4209{
4210 size_t printed;
4211
99ff7150 4212 printed = fprintf(fp, "\n Summary of events:\n\n");
bf2575c1
DA		 4213
4214 return printed;
4215}
4216
b535d523 4217DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
8eded45f
ACM		 4218 struct syscall_stats *stats;
4219 double msecs;
4220 int syscall;
b535d523
ACM		 4221)
4222{
4223 struct int_node *source = rb_entry(nd, struct int_node, rb_node);
8eded45f 4224 struct syscall_stats *stats = source->priv;
b535d523
ACM		 4225
4226 entry->syscall = source->i;
4227 entry->stats = stats;
8eded45f 4228 entry->msecs = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0;
b535d523
ACM		 4229}
4230
bf2575c1
DA		 4231static size_t thread__dump_stats(struct thread_trace *ttrace,
4232 struct trace *trace, FILE *fp)
4233{
bf2575c1
DA		 4234 size_t printed = 0;
4235 struct syscall *sc;
b535d523
ACM		 4236 struct rb_node *nd;
4237 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
bf2575c1 4238
b535d523 4239 if (syscall_stats == NULL)
bf2575c1
DA		 4240 return 0;
4241
4242 printed += fprintf(fp, "\n");
4243
8eded45f
ACM		 4244 printed += fprintf(fp, " syscall calls errors total min avg max stddev\n");
4245 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
4246 printed += fprintf(fp, " --------------- -------- ------ -------- --------- --------- --------- ------\n");
99ff7150 4247
98a91837 4248 resort_rb__for_each_entry(nd, syscall_stats) {
8eded45f 4249 struct syscall_stats *stats = syscall_stats_entry->stats;
bf2575c1 4250 if (stats) {
8eded45f
ACM		 4251 double min = (double)(stats->stats.min) / NSEC_PER_MSEC;
4252 double max = (double)(stats->stats.max) / NSEC_PER_MSEC;
4253 double avg = avg_stats(&stats->stats);
bf2575c1 4254 double pct;
8eded45f 4255 u64 n = (u64)stats->stats.n;
bf2575c1 4256
8eded45f 4257 pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0;
bf2575c1
DA		 4258 avg /= NSEC_PER_MSEC;
4259
b535d523 4260 sc = &trace->syscalls.table[syscall_stats_entry->syscall];
99ff7150 4261 printed += fprintf(fp, " %-15s", sc->name);
8eded45f
ACM		 4262 printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f",
4263 n, stats->nr_failures, syscall_stats_entry->msecs, min, avg);
27a778b5 4264 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
b88b14db
ACM		 4265
4266 if (trace->errno_summary && stats->nr_failures) {
4267 const char *arch_name = perf_env__arch(trace->host->env);
4268 int e;
4269
4270 for (e = 0; e < stats->max_errno; ++e) {
4271 if (stats->errnos[e] != 0)
4272 fprintf(fp, "\t\t\t\t%s: %d\n", arch_syscalls__strerrno(arch_name, e + 1), stats->errnos[e]);
4273 }
4274 }
bf2575c1 4275 }
bf2575c1
DA		 4276 }
4277
b535d523 4278 resort_rb__delete(syscall_stats);
bf2575c1 4279 printed += fprintf(fp, "\n\n");
1302d88e
ACM		 4280
4281 return printed;
4282}
4283
96c14451 4284static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
896cbb56 4285{
96c14451 4286 size_t printed = 0;
89dceb22 4287 struct thread_trace *ttrace = thread__priv(thread);
896cbb56
DA		 4288 double ratio;
4289
4290 if (ttrace == NULL)
4291 return 0;
4292
4293 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
4294
ee84a303 4295 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread__tid(thread));
99ff7150 4296 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
15e65c69 4297 printed += fprintf(fp, "%.1f%%", ratio);
a2ea67d7
SF		 4298 if (ttrace->pfmaj)
4299 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
4300 if (ttrace->pfmin)
4301 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
03548ebf
ACM		 4302 if (trace->sched)
4303 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
4304 else if (fputc('\n', fp) != EOF)
4305 ++printed;
4306
bf2575c1 4307 printed += thread__dump_stats(ttrace, trace, fp);
896cbb56 4308
96c14451
ACM		 4309 return printed;
4310}
896cbb56 4311
96c14451
ACM		 4312static unsigned long thread__nr_events(struct thread_trace *ttrace)
4313{
4314 return ttrace ? ttrace->nr_events : 0;
4315}
4316
ee84a303
IR		 4317DEFINE_RESORT_RB(threads,
4318 (thread__nr_events(thread__priv(a->thread)) <
4319 thread__nr_events(thread__priv(b->thread))),
96c14451
ACM		 4320 struct thread *thread;
4321)
4322{
7ee227f6 4323 entry->thread = rb_entry(nd, struct thread_rb_node, rb_node)->thread;
896cbb56
DA		 4324}
4325
1302d88e
ACM		 4326static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
4327{
96c14451
ACM		 4328 size_t printed = trace__fprintf_threads_header(fp);
4329 struct rb_node *nd;
91e467bc 4330 int i;
1302d88e 4331
91e467bc
KL		 4332 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
4333 DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
96c14451 4334
91e467bc
KL		 4335 if (threads == NULL) {
4336 fprintf(fp, "%s", "Error sorting output by nr_events!\n");
4337 return 0;
4338 }
896cbb56 4339
91e467bc
KL		 4340 resort_rb__for_each_entry(nd, threads)
4341 printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
96c14451 4342
91e467bc
KL		 4343 resort_rb__delete(threads);
4344 }
96c14451 4345 return printed;
1302d88e
ACM		 4346}
4347
ae9ed035
ACM		 4348static int trace__set_duration(const struct option *opt, const char *str,
4349 int unset __maybe_unused)
4350{
4351 struct trace *trace = opt->value;
4352
4353 trace->duration_filter = atof(str);
4354 return 0;
4355}
4356
6a0b3aba
ACM		 4357static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
4358 int unset __maybe_unused)
f078c385
ACM		 4359{
4360 int ret = -1;
4361 size_t i;
4362 struct trace *trace = opt->value;
4363 /*
4364 * FIXME: introduce a intarray class, plain parse csv and create a
4365 * { int nr, int entries[] } struct...
4366 */
4367 struct intlist *list = intlist__new(str);
4368
4369 if (list == NULL)
4370 return -1;
4371
4372 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
4373 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
4374
4375 if (trace->filter_pids.entries == NULL)
4376 goto out;
4377
4378 trace->filter_pids.entries[0] = getpid();
4379
4380 for (i = 1; i < trace->filter_pids.nr; ++i)
4381 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
4382
4383 intlist__delete(list);
4384 ret = 0;
4385out:
4386 return ret;
4387}
4388
c24ff998
ACM		 4389static int trace__open_output(struct trace *trace, const char *filename)
4390{
4391 struct stat st;
4392
4393 if (!stat(filename, &st) && st.st_size) {
4394 char oldname[PATH_MAX];
4395
4396 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
4397 unlink(oldname);
4398 rename(filename, oldname);
4399 }
4400
4401 trace->output = fopen(filename, "w");
4402
4403 return trace->output == NULL ? -errno : 0;
4404}
4405
598d02c5
SF		 4406static int parse_pagefaults(const struct option *opt, const char *str,
4407 int unset __maybe_unused)
4408{
4409 int *trace_pgfaults = opt->value;
4410
4411 if (strcmp(str, "all") == 0)
4412 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
4413 else if (strcmp(str, "maj") == 0)
4414 *trace_pgfaults |= TRACE_PFMAJ;
4415 else if (strcmp(str, "min") == 0)
4416 *trace_pgfaults |= TRACE_PFMIN;
4417 else
4418 return -1;
4419
4420 return 0;
4421}
4422
206d635a 4423static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)
14a052df 4424{
32dcd021 4425 struct evsel *evsel;
14a052df 4426
206d635a
ACM		 4427 evlist__for_each_entry(evlist, evsel) {
4428 if (evsel->handler == NULL)
4429 evsel->handler = handler;
4430 }
14a052df
ACM		 4431}
4432
362222f8
ACM		 4433static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name)
4434{
4435 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
4436
4437 if (fmt) {
60995604 4438 const struct syscall_fmt *scfmt = syscall_fmt__find(name);
362222f8
ACM		 4439
4440 if (scfmt) {
4441 int skip = 0;
4442
4443 if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 ||
4444 strcmp(evsel->tp_format->format.fields->name, "nr") == 0)
4445 ++skip;
4446
4447 memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt));
4448 }
4449 }
4450}
4451
63503dba 4452static int evlist__set_syscall_tp_fields(struct evlist *evlist)
d32855fa 4453{
32dcd021 4454 struct evsel *evsel;
d32855fa
ACM		 4455
4456 evlist__for_each_entry(evlist, evsel) {
4457 if (evsel->priv || !evsel->tp_format)
4458 continue;
4459
947b843c 4460 if (strcmp(evsel->tp_format->system, "syscalls")) {
794bca26 4461 evsel__init_tp_arg_scnprintf(evsel);
d32855fa 4462 continue;
947b843c 4463 }
d32855fa 4464
794bca26 4465 if (evsel__init_syscall_tp(evsel))
d32855fa
ACM		 4466 return -1;
4467
4468 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
fecd9907 4469 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
d32855fa
ACM		 4470
4471 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
4472 return -1;
362222f8
ACM		 4473
4474 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1);
d32855fa 4475 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
fecd9907 4476 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
d32855fa
ACM		 4477
4478 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
4479 return -1;
362222f8
ACM		 4480
4481 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1);
d32855fa
ACM		 4482 }
4483 }
4484
4485 return 0;
4486}
4487
017037ff
ACM		 4488/*
4489 * XXX: Hackish, just splitting the combined -e+--event (syscalls
4490 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
4491 * existing facilities unchanged (trace->ev_qualifier + parse_options()).
4492 *
4493 * It'd be better to introduce a parse_options() variant that would return a
4494 * list with the terms it didn't match to an event...
4495 */
4496static int trace__parse_events_option(const struct option *opt, const char *str,
4497 int unset __maybe_unused)
4498{
4499 struct trace *trace = (struct trace *)opt->value;
4500 const char *s = str;
4501 char *sep = NULL, *lists[2] = { NULL, NULL, };
27702bcf 4502 int len = strlen(str) + 1, err = -1, list, idx;
017037ff
ACM		 4503 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
4504 char group_name[PATH_MAX];
60995604 4505 const struct syscall_fmt *fmt;
017037ff
ACM		 4506
4507 if (strace_groups_dir == NULL)
4508 return -1;
4509
4510 if (*s == '!') {
4511 ++s;
4512 trace->not_ev_qualifier = true;
4513 }
4514
4515 while (1) {
4516 if ((sep = strchr(s, ',')) != NULL)
4517 *sep = '\0';
4518
4519 list = 0;
27702bcf
ACM		 4520 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
4521 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
017037ff 4522 list = 1;
f932184e
ACM		 4523 goto do_concat;
4524 }
4525
4526 fmt = syscall_fmt__find_by_alias(s);
4527 if (fmt != NULL) {
4528 list = 1;
4529 s = fmt->name;
017037ff
ACM		 4530 } else {
4531 path__join(group_name, sizeof(group_name), strace_groups_dir, s);
4532 if (access(group_name, R_OK) == 0)
4533 list = 1;
4534 }
f932184e 4535do_concat:
017037ff
ACM		 4536 if (lists[list]) {
4537 sprintf(lists[list] + strlen(lists[list]), ",%s", s);
4538 } else {
4539 lists[list] = malloc(len);
4540 if (lists[list] == NULL)
4541 goto out;
4542 strcpy(lists[list], s);
4543 }
4544
4545 if (!sep)
4546 break;
4547
4548 *sep = ',';
4549 s = sep + 1;
4550 }
4551
4552 if (lists[1] != NULL) {
4553 struct strlist_config slist_config = {
4554 .dirname = strace_groups_dir,
4555 };
4556
4557 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4558 if (trace->ev_qualifier == NULL) {
4559 fputs("Not enough memory to parse event qualifier", trace->output);
4560 goto out;
4561 }
4562
4563 if (trace__validate_ev_qualifier(trace))
4564 goto out;
b912885a 4565 trace->trace_syscalls = true;
017037ff
ACM		 4566 }
4567
4568 err = 0;
4569
4570 if (lists[0]) {
411ad22e
IR		 4571 struct parse_events_option_args parse_events_option_args = {
4572 .evlistp = &trace->evlist,
4573 };
a6293f36 4574 struct option o = {
411ad22e 4575 .value = &parse_events_option_args,
a6293f36 4576 };
017037ff
ACM		 4577 err = parse_events_option(&o, lists[0], 0);
4578 }
4579out:
659ede7d
RM		 4580 free(strace_groups_dir);
4581 free(lists[0]);
4582 free(lists[1]);
017037ff
ACM		 4583 if (sep)
4584 *sep = ',';
4585
4586 return err;
4587}
4588
9ea42ba4
ACM		 4589static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4590{
4591 struct trace *trace = opt->value;
4592
a6293f36
SI		 4593 if (!list_empty(&trace->evlist->core.entries)) {
4594 struct option o = {
4595 .value = &trace->evlist,
4596 };
4597 return parse_cgroups(&o, str, unset);
4598 }
9ea42ba4
ACM		 4599 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
4600
4601 return 0;
4602}
4603
ac96287c
ACM		 4604static int trace__config(const char *var, const char *value, void *arg)
4605{
e7c634fc 4606 struct trace *trace = arg;
ac96287c
ACM		 4607 int err = 0;
4608
4609 if (!strcmp(var, "trace.add_events")) {
7e035929
ACM		 4610 trace->perfconfig_events = strdup(value);
4611 if (trace->perfconfig_events == NULL) {
4612 pr_err("Not enough memory for %s\n", "trace.add_events");
4613 return -1;
4614 }
b036146f
ACM		 4615 } else if (!strcmp(var, "trace.show_timestamp")) {
4616 trace->show_tstamp = perf_config_bool(var, value);
42e4a52d
ACM		 4617 } else if (!strcmp(var, "trace.show_duration")) {
4618 trace->show_duration = perf_config_bool(var, value);
9d6dc178
ACM		 4619 } else if (!strcmp(var, "trace.show_arg_names")) {
4620 trace->show_arg_names = perf_config_bool(var, value);
4621 if (!trace->show_arg_names)
4622 trace->show_zeros = true;
e7c634fc 4623 } else if (!strcmp(var, "trace.show_zeros")) {
9d6dc178
ACM		 4624 bool new_show_zeros = perf_config_bool(var, value);
4625 if (!trace->show_arg_names && !new_show_zeros) {
4626 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4627 goto out;
4628 }
4629 trace->show_zeros = new_show_zeros;
c65c83ff
ACM		 4630 } else if (!strcmp(var, "trace.show_prefix")) {
4631 trace->show_string_prefix = perf_config_bool(var, value);
d32de87e
ACM		 4632 } else if (!strcmp(var, "trace.no_inherit")) {
4633 trace->opts.no_inherit = perf_config_bool(var, value);
9ed45d59
ACM		 4634 } else if (!strcmp(var, "trace.args_alignment")) {
4635 int args_alignment = 0;
4636 if (perf_config_int(&args_alignment, var, value) == 0)
4637 trace->args_alignment = args_alignment;
f11b2803
ACM		 4638 } else if (!strcmp(var, "trace.tracepoint_beautifiers")) {
4639 if (strcasecmp(value, "libtraceevent") == 0)
4640 trace->libtraceevent_print = true;
4641 else if (strcasecmp(value, "libbeauty") == 0)
4642 trace->libtraceevent_print = false;
ac96287c 4643 }
9d6dc178 4644out:
ac96287c
ACM		 4645 return err;
4646}
4647
6c7f0ab0
RM		 4648static void trace__exit(struct trace *trace)
4649{
f2ebf8ff
RM		 4650 int i;
4651
6c7f0ab0 4652 strlist__delete(trace->ev_qualifier);
9997d5dd 4653 zfree(&trace->ev_qualifier_ids.entries);
f2ebf8ff
RM		 4654 if (trace->syscalls.table) {
4655 for (i = 0; i <= trace->sctbl->syscalls.max_id; i++)
4656 syscall__exit(&trace->syscalls.table[i]);
9997d5dd 4657 zfree(&trace->syscalls.table);
f2ebf8ff 4658 }
6c7f0ab0
RM		 4659 syscalltbl__delete(trace->sctbl);
4660 zfree(&trace->perfconfig_events);
4661}
4662
5e6da6be
IR		 4663#ifdef HAVE_BPF_SKEL
4664static int bpf__setup_bpf_output(struct evlist *evlist)
4665{
4666 int err = parse_event(evlist, "bpf-output/no-inherit=1,name=__augmented_syscalls__/");
4667
4668 if (err)
4669 pr_debug("ERROR: failed to create the \"__augmented_syscalls__\" bpf-output event\n");
4670
4671 return err;
4672}
4673#endif
4674
b0ad8ea6 4675int cmd_trace(int argc, const char **argv)
514f1c67 4676{
6fdd9cb7 4677 const char *trace_usage[] = {
f15eb531
NK		 4678 "perf trace [<options>] [<command>]",
4679 "perf trace [<options>] -- <command> [<options>]",
5e2485b1
DA		 4680 "perf trace record [<options>] [<command>]",
4681 "perf trace record [<options>] -- <command> [<options>]",
514f1c67
ACM		 4682 NULL
4683 };
4684 struct trace trace = {
514f1c67
ACM		 4685 .opts = {
4686 .target = {
4687 .uid = UINT_MAX,
4688 .uses_mmap = true,
4689 },
4690 .user_freq = UINT_MAX,
4691 .user_interval = ULLONG_MAX,
509051ea 4692 .no_buffering = true,
38d5447d 4693 .mmap_pages = UINT_MAX,
514f1c67 4694 },
007d66a0 4695 .output = stderr,
50c95cbd 4696 .show_comm = true,
b036146f 4697 .show_tstamp = true,
42e4a52d 4698 .show_duration = true,
9d6dc178 4699 .show_arg_names = true,
9ed45d59 4700 .args_alignment = 70,
b912885a 4701 .trace_syscalls = false,
44621819 4702 .kernel_syscallchains = false,
05614993 4703 .max_stack = UINT_MAX,
5067a8cd 4704 .max_events = ULONG_MAX,
514f1c67 4705 };
c24ff998 4706 const char *output_name = NULL;
514f1c67 4707 const struct option trace_options[] = {
017037ff
ACM		 4708 OPT_CALLBACK('e', "event", &trace, "event",
4709 "event/syscall selector. use 'perf list' to list available events",
4710 trace__parse_events_option),
d4097f19
ACM		 4711 OPT_CALLBACK(0, "filter", &trace.evlist, "filter",
4712 "event filter", parse_filter),
50c95cbd
ACM		 4713 OPT_BOOLEAN(0, "comm", &trace.show_comm,
4714 "show the thread COMM next to its id"),
c522739d 4715 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
017037ff
ACM		 4716 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4717 trace__parse_events_option),
c24ff998 4718 OPT_STRING('o', "output", &output_name, "file", "output file name"),
6810fc91 4719 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
514f1c67
ACM		 4720 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4721 "trace events on existing process id"),
ac9be8ee 4722 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
514f1c67 4723 "trace events on existing thread id"),
fa0e4ffe 4724 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
6a0b3aba 4725 "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
ac9be8ee 4726 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
514f1c67 4727 "system-wide collection from all CPUs"),
ac9be8ee 4728 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
514f1c67 4729 "list of cpus to monitor"),
6810fc91 4730 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
514f1c67 4731 "child tasks do not inherit counters"),
994a1f78 4732 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
25f84702 4733 "number of mmap data pages", evlist__parse_mmap_pages),
ac9be8ee 4734 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
514f1c67 4735 "user to profile"),
ae9ed035
ACM		 4736 OPT_CALLBACK(0, "duration", &trace, "float",
4737 "show only events with duration > N.M ms",
4738 trace__set_duration),
1302d88e 4739 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
7c304ee0 4740 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4bb09192
DA		 4741 OPT_BOOLEAN('T', "time", &trace.full_time,
4742 "Show full timestamp, not time relative to first start"),
0a6545bd
ACM		 4743 OPT_BOOLEAN(0, "failure", &trace.failure_only,
4744 "Show only syscalls that failed"),
fd2eabaf
DA		 4745 OPT_BOOLEAN('s', "summary", &trace.summary_only,
4746 "Show only syscall summary with statistics"),
4747 OPT_BOOLEAN('S', "with-summary", &trace.summary,
4748 "Show all syscalls and summary with statistics"),
b88b14db
ACM		 4749 OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary,
4750 "Show errno stats per syscall, use with -s or -S"),
598d02c5
SF		 4751 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4752 "Trace pagefaults", parse_pagefaults, "maj"),
e281a960 4753 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
e366a6d8 4754 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
566a0885
MW		 4755 OPT_CALLBACK(0, "call-graph", &trace.opts,
4756 "record_mode[,record_size]", record_callchain_help,
4757 &record_parse_callchain_opt),
f11b2803
ACM		 4758 OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,
4759 "Use libtraceevent to print the tracepoint arguments."),
44621819
ACM		 4760 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4761 "Show the kernel callchains on the syscall exit path"),
5067a8cd
ACM		 4762 OPT_ULONG(0, "max-events", &trace.max_events,
4763 "Set the maximum number of events to print, exit after that is reached. "),
5cf9c84e
ACM		 4764 OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4765 "Set the minimum stack depth when parsing the callchain, "
4766 "anything below the specified depth will be ignored."),
c6d4a494
ACM		 4767 OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4768 "Set the maximum stack depth when parsing the callchain, "
4769 "anything beyond the specified depth will be ignored. "
4cb93446 4770 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
a761a8d1
ACM		 4771 OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4772 "Sort batch of events before processing, use if getting out of order events"),
591421e1
ACM		 4773 OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4774 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
3fcb10e4 4775 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
9d9cad76 4776 "per thread proc mmap processing timeout in ms"),
9ea42ba4
ACM		 4777 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4778 trace__parse_cgroups),
cb4b9e68 4779 OPT_INTEGER('D', "delay", &trace.opts.target.initial_delay,
e36b7821
AB		 4780 "ms to wait before starting measurement after program "
4781 "start"),
22ac4318 4782 OPTS_EVSWITCH(&trace.evswitch),
514f1c67
ACM		 4783 OPT_END()
4784 };
ccd62a89 4785 bool __maybe_unused max_stack_user_set = true;
f3e459d1 4786 bool mmap_pages_user_set = true;
32dcd021 4787 struct evsel *evsel;
6fdd9cb7 4788 const char * const trace_subcommands[] = { "record", NULL };
78e890ea 4789 int err = -1;
32caf0d1 4790 char bf[BUFSIZ];
de9f498d 4791 struct sigaction sigchld_act;
514f1c67 4792
4d08cb80
ACM		 4793 signal(SIGSEGV, sighandler_dump_stack);
4794 signal(SIGFPE, sighandler_dump_stack);
de9f498d
CD		 4795 signal(SIGINT, sighandler_interrupt);
4796
4797 memset(&sigchld_act, 0, sizeof(sigchld_act));
4798 sigchld_act.sa_flags = SA_SIGINFO;
4799 sigchld_act.sa_sigaction = sighandler_chld;
4800 sigaction(SIGCHLD, &sigchld_act, NULL);
4d08cb80 4801
0f98b11c 4802 trace.evlist = evlist__new();
fd0db102 4803 trace.sctbl = syscalltbl__new();
14a052df 4804
fd0db102 4805 if (trace.evlist == NULL || trace.sctbl == NULL) {
14a052df 4806 pr_err("Not enough memory to run!\n");
ff8f695c 4807 err = -ENOMEM;
14a052df
ACM		 4808 goto out;
4809 }
4810
c3e78a34
ACM		 4811 /*
4812 * Parsing .perfconfig may entail creating a BPF event, that may need
4813 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4814 * is too small. This affects just this process, not touching the
4815 * global setting. If it fails we'll get something in 'perf trace -v'
4816 * to help diagnose the problem.
4817 */
4818 rlimit__bump_memlock();
4819
ac96287c
ACM		 4820 err = perf_config(trace__config, &trace);
4821 if (err)
4822 goto out;
4823
6fdd9cb7
YS		 4824 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4825 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
8bd436b0
ACM		 4826
4827 /*
4828 * Here we already passed thru trace__parse_events_option() and it has
4829 * already figured out if -e syscall_name, if not but if --event
4830 * foo:bar was used, the user is interested _just_ in those, say,
4831 * tracepoint events, not in the strace-like syscall-name-based mode.
4832 *
4833 * This is important because we need to check if strace-like mode is
4834 * needed to decided if we should filter out the eBPF
4835 * __augmented_syscalls__ code, if it is in the mix, say, via
4836 * .perfconfig trace.add_events, and filter those out.
4837 */
4838 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4839 trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4840 trace.trace_syscalls = true;
4841 }
7e035929
ACM		 4842 /*
4843 * Now that we have --verbose figured out, lets see if we need to parse
4844 * events from .perfconfig, so that if those events fail parsing, say some
4845 * BPF program fails, then we'll be able to use --verbose to see what went
4846 * wrong in more detail.
4847 */
4848 if (trace.perfconfig_events != NULL) {
a910e466 4849 struct parse_events_error parse_err;
7e035929 4850
07eafd4e 4851 parse_events_error__init(&parse_err);
7e035929 4852 err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);
07eafd4e 4853 if (err)
6c191289 4854 parse_events_error__print(&parse_err, trace.perfconfig_events);
07eafd4e
IR		 4855 parse_events_error__exit(&parse_err);
4856 if (err)
7e035929 4857 goto out;
7e035929 4858 }
fd2eabaf 4859
9ea42ba4
ACM		 4860 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4861 usage_with_options_msg(trace_usage, trace_options,
4862 "cgroup monitoring only available in system-wide mode");
4863 }
4864
5e6da6be 4865#ifdef HAVE_BPF_SKEL
83a0943b
ACM		 4866 if (!trace.trace_syscalls)
4867 goto skip_augmentation;
4868
5e6da6be
IR		 4869 trace.skel = augmented_raw_syscalls_bpf__open();
4870 if (!trace.skel) {
4871 pr_debug("Failed to open augmented syscalls BPF skeleton");
4872 } else {
4873 /*
4874 * Disable attaching the BPF programs except for sys_enter and
4875 * sys_exit that tail call into this as necessary.
4876 */
4877 struct bpf_program *prog;
59247e33 4878
5e6da6be
IR		 4879 bpf_object__for_each_program(prog, trace.skel->obj) {
4880 if (prog != trace.skel->progs.sys_enter && prog != trace.skel->progs.sys_exit)
4881 bpf_program__set_autoattach(prog, /*autoattach=*/false);
4882 }
4883
4884 err = augmented_raw_syscalls_bpf__load(trace.skel);
4885
4886 if (err < 0) {
4887 libbpf_strerror(err, bf, sizeof(bf));
4888 pr_debug("Failed to load augmented syscalls BPF skeleton: %s\n", bf);
4889 } else {
4890 augmented_raw_syscalls_bpf__attach(trace.skel);
4891 trace__add_syscall_newtp(&trace);
ff7a4f98
ACM		 4892 }
4893 }
4894
5e6da6be
IR		 4895 err = bpf__setup_bpf_output(trace.evlist);
4896 if (err) {
4897 libbpf_strerror(err, bf, sizeof(bf));
4898 pr_err("ERROR: Setup BPF output event failed: %s\n", bf);
4899 goto out;
4900 }
4901 trace.syscalls.events.bpf_output = evlist__last(trace.evlist);
4902 assert(!strcmp(evsel__name(trace.syscalls.events.bpf_output), "__augmented_syscalls__"));
83a0943b 4903skip_augmentation:
5e6da6be
IR		 4904#endif
4905 err = -1;
4906
598d02c5
SF		 4907 if (trace.trace_pgfaults) {
4908 trace.opts.sample_address = true;
4909 trace.opts.sample_time = true;
4910 }
4911
f3e459d1
ACM		 4912 if (trace.opts.mmap_pages == UINT_MAX)
4913 mmap_pages_user_set = false;
4914
05614993 4915 if (trace.max_stack == UINT_MAX) {
029c75e5 4916 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
05614993
ACM		 4917 max_stack_user_set = false;
4918 }
4919
4920#ifdef HAVE_DWARF_UNWIND_SUPPORT
75d50117 4921 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
05614993 4922 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
75d50117 4923 }
05614993
ACM		 4924#endif
4925
2ddd5c04 4926 if (callchain_param.enabled) {
f3e459d1
ACM		 4927 if (!mmap_pages_user_set && geteuid() == 0)
4928 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
4929
566a0885 4930 symbol_conf.use_callchain = true;
f3e459d1 4931 }
566a0885 4932
6484d2f9 4933 if (trace.evlist->core.nr_entries > 0) {
206d635a 4934 evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);
d32855fa
ACM		 4935 if (evlist__set_syscall_tp_fields(trace.evlist)) {
4936 perror("failed to set syscalls:* tracepoint fields");
4937 goto out;
4938 }
4939 }
14a052df 4940
a761a8d1
ACM		 4941 if (trace.sort_events) {
4942 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
4943 ordered_events__set_copy_on_queue(&trace.oe.data, true);
4944 }
028713aa 4945
b1a9e253
ACM		 4946 /*
4947 * If we are augmenting syscalls, then combine what we put in the
4948 * __augmented_syscalls__ BPF map with what is in the
4949 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
4950 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
4951 *
4952 * We'll switch to look at two BPF maps, one for sys_enter and the
4953 * other for sys_exit when we start augmenting the sys_exit paths with
4954 * buffers that are being copied from kernel to userspace, think 'read'
4955 * syscall.
4956 */
5e6da6be 4957 if (trace.syscalls.events.bpf_output) {
b1a9e253 4958 evlist__for_each_entry(trace.evlist, evsel) {
8ab2e96d 4959 bool raw_syscalls_sys_exit = strcmp(evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
cd26ea6d
ACM		 4960
4961 if (raw_syscalls_sys_exit) {
4962 trace.raw_augmented_syscalls = true;
4963 goto init_augmented_syscall_tp;
4964 }
4965
5e6da6be 4966 if (trace.syscalls.events.bpf_output->priv == NULL &&
8ab2e96d 4967 strstr(evsel__name(evsel), "syscalls:sys_enter")) {
5e6da6be 4968 struct evsel *augmented = trace.syscalls.events.bpf_output;
794bca26
ACM		 4969 if (evsel__init_augmented_syscall_tp(augmented, evsel) ||
4970 evsel__init_augmented_syscall_tp_args(augmented))
b9b6a2ea 4971 goto out;
b119970a
ACM		 4972 /*
4973 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
4974 * Above we made sure we can get from the payload the tp fields
4975 * that we get from syscalls:sys_enter tracefs format file.
4976 */
b9b6a2ea 4977 augmented->handler = trace__sys_enter;
b119970a
ACM		 4978 /*
4979 * Now we do the same for the *syscalls:sys_enter event so that
4980 * if we handle it directly, i.e. if the BPF prog returns 0 so
4981 * as not to filter it, then we'll handle it just like we would
4982 * for the BPF_OUTPUT one:
4983 */
794bca26
ACM		 4984 if (evsel__init_augmented_syscall_tp(evsel, evsel) ||
4985 evsel__init_augmented_syscall_tp_args(evsel))
b119970a
ACM		 4986 goto out;
4987 evsel->handler = trace__sys_enter;
b9b6a2ea
ACM		 4988 }
4989
8ab2e96d 4990 if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) {
b9b6a2ea 4991 struct syscall_tp *sc;
cd26ea6d 4992init_augmented_syscall_tp:
794bca26 4993 if (evsel__init_augmented_syscall_tp(evsel, evsel))
b9b6a2ea 4994 goto out;
fecd9907 4995 sc = __evsel__syscall_tp(evsel);
b9b6a2ea
ACM		 4996 /*
4997 * For now with BPF raw_augmented we hook into
4998 * raw_syscalls:sys_enter and there we get all
4999 * 6 syscall args plus the tracepoint common
5000 * fields and the syscall_nr (another long).
5001 * So we check if that is the case and if so
5002 * don't look after the sc->args_size but
5003 * always after the full raw_syscalls:sys_enter
5004 * payload, which is fixed.
5005 *
5006 * We'll revisit this later to pass
5007 * s->args_size to the BPF augmenter (now
5008 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
5009 * so that it copies only what we need for each
5010 * syscall, like what happens when we use
5011 * syscalls:sys_enter_NAME, so that we reduce
5012 * the kernel/userspace traffic to just what is
5013 * needed for each syscall.
5014 */
5015 if (trace.raw_augmented_syscalls)
5016 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
794bca26 5017 evsel__init_augmented_syscall_tp_ret(evsel);
b1a9e253
ACM		 5018 evsel->handler = trace__sys_exit;
5019 }
5020 }
5021 }
5022
1e28fe0a
SF		 5023 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
5024 return trace__record(&trace, argc-1, &argv[1]);
5025
b88b14db
ACM		 5026 /* Using just --errno-summary will trigger --summary */
5027 if (trace.errno_summary && !trace.summary && !trace.summary_only)
5028 trace.summary_only = true;
5029
1e28fe0a
SF		 5030 /* summary_only implies summary option, but don't overwrite summary if set */
5031 if (trace.summary_only)
5032 trace.summary = trace.summary_only;
5033
c24ff998
ACM		 5034 if (output_name != NULL) {
5035 err = trace__open_output(&trace, output_name);
5036 if (err < 0) {
5037 perror("failed to create output file");
5038 goto out;
5039 }
5040 }
5041
22ac4318
ACM		 5042 err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
5043 if (err)
5044 goto out_close;
5045
602ad878 5046 err = target__validate(&trace.opts.target);
32caf0d1 5047 if (err) {
602ad878 5048 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
c24ff998
ACM		 5049 fprintf(trace.output, "%s", bf);
5050 goto out_close;
32caf0d1
NK		 5051 }
5052
602ad878 5053 err = target__parse_uid(&trace.opts.target);
514f1c67 5054 if (err) {
602ad878 5055 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
c24ff998
ACM		 5056 fprintf(trace.output, "%s", bf);
5057 goto out_close;
514f1c67
ACM		 5058 }
5059
602ad878 5060 if (!argc && target__none(&trace.opts.target))
ee76120e
NK		 5061 trace.opts.target.system_wide = true;
5062
6810fc91
DA		 5063 if (input_name)
5064 err = trace__replay(&trace);
5065 else
5066 err = trace__run(&trace, argc, argv);
1302d88e 5067
c24ff998
ACM		 5068out_close:
5069 if (output_name != NULL)
5070 fclose(trace.output);
5071out:
6c7f0ab0 5072 trace__exit(&trace);
5e6da6be
IR		 5073#ifdef HAVE_BPF_SKEL
5074 augmented_raw_syscalls_bpf__destroy(trace.skel);
5075#endif
1302d88e 5076 return err;
514f1c67 5077}
Linux 6.x block layer and io_uring tree(s)