projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'zstd-linus-v6.2' of https://github.com/terrelln/linux
[linux-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>
744fafc7 19#include <bpf/bpf.h>
ff7a4f98 20#include "util/bpf_map.h"
c3e78a34 21#include "util/rlimit.h"
514f1c67 22#include "builtin.h"
9ea42ba4 23#include "util/cgroup.h"
752fde44 24#include "util/color.h"
ac96287c 25#include "util/config.h"
7c304ee0 26#include "util/debug.h"
4a3cec84 27#include "util/dso.h"
092bd3cd 28#include "util/env.h"
5ab8c689 29#include "util/event.h"
ca125277
ACM		 30#include "util/evsel.h"
31#include "util/evsel_fprintf.h"
ea49e01c 32#include "util/synthetic-events.h"
514f1c67 33#include "util/evlist.h"
22ac4318 34#include "util/evswitch.h"
e0fcfb08 35#include "util/mmap.h"
8520a98d 36#include <subcmd/pager.h>
4b6ab94e 37#include <subcmd/exec-cmd.h>
752fde44 38#include "util/machine.h"
1101f69a 39#include "util/map.h"
daecf9e0 40#include "util/symbol.h"
9a3993d4 41#include "util/path.h"
6810fc91 42#include "util/session.h"
752fde44 43#include "util/thread.h"
4b6ab94e 44#include <subcmd/parse-options.h>
2ae3a312 45#include "util/strlist.h"
bdc89661 46#include "util/intlist.h"
514f1c67 47#include "util/thread_map.h"
bf2575c1 48#include "util/stat.h"
4a3cec84 49#include "util/tool.h"
2da39f1c 50#include "util/util.h"
fd5cead2 51#include "trace/beauty/beauty.h"
97978b3e 52#include "trace-event.h"
9aca7f17 53#include "util/parse-events.h"
ba504235 54#include "util/bpf-loader.h"
9b7c7728 55#include "util/tracepoint.h"
566a0885 56#include "callchain.h"
fea01392 57#include "print_binary.h"
a067558e 58#include "string2.h"
fd0db102 59#include "syscalltbl.h"
96c14451 60#include "rb_resort.h"
c1a604df 61#include "../perf.h"
514f1c67 62
a43783ae 63#include <errno.h>
fd20e811 64#include <inttypes.h>
4208735d 65#include <poll.h>
9607ad3a 66#include <signal.h>
514f1c67 67#include <stdlib.h>
017037ff 68#include <string.h>
8dd2a131 69#include <linux/err.h>
997bba8c 70#include <linux/filter.h>
877a7a11 71#include <linux/kernel.h>
39878d49 72#include <linux/random.h>
c6d4a494 73#include <linux/stringify.h>
bd48c63e 74#include <linux/time64.h>
7f7c536f 75#include <linux/zalloc.h>
bafae98e 76#include <fcntl.h>
4bcc4cff 77#include <sys/sysmacros.h>
514f1c67 78
3052ba56 79#include <linux/ctype.h>
7728fa0c 80#include <perf/mmap.h>
3d689ed6 81
378ef0f5
IR		 82#ifdef HAVE_LIBTRACEEVENT
83#include <traceevent/event-parse.h>
84#endif
85
c188e7ac
ACM		 86#ifndef O_CLOEXEC
87# define O_CLOEXEC 02000000
88#endif
89
83a51694
ACM		 90#ifndef F_LINUX_SPECIFIC_BASE
91# define F_LINUX_SPECIFIC_BASE 1024
92#endif
93
eadcab4c
LY		 94#define RAW_SYSCALL_ARGS_NUM 6
95
3f41b778
ACM		 96/*
97 * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100
98 */
888ca854
ACM		 99struct syscall_arg_fmt {
100 size_t (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
3f41b778 101 bool (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val);
888ca854
ACM		 102 unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
103 void *parm;
104 const char *name;
9597945d 105 u16 nr_entries; // for arrays
888ca854
ACM		 106 bool show_zero;
107};
108
109struct syscall_fmt {
110 const char *name;
111 const char *alias;
112 struct {
113 const char *sys_enter,
114 *sys_exit;
115 } bpf_prog_name;
eadcab4c 116 struct syscall_arg_fmt arg[RAW_SYSCALL_ARGS_NUM];
888ca854
ACM		 117 u8 nr_args;
118 bool errpid;
119 bool timeout;
120 bool hexret;
121};
122
d1d438a3
ACM		 123struct trace {
124 struct perf_tool tool;
fd0db102 125 struct syscalltbl *sctbl;
d1d438a3 126 struct {
d1d438a3 127 struct syscall *table;
3803a229
ACM		 128 struct { // per syscall BPF_MAP_TYPE_PROG_ARRAY
129 struct bpf_map *sys_enter,
130 *sys_exit;
131 } prog_array;
d1d438a3 132 struct {
32dcd021 133 struct evsel *sys_enter,
d3d1c4bd
ACM		 134 *sys_exit,
135 *augmented;
d1d438a3 136 } events;
5834da7f 137 struct bpf_program *unaugmented_prog;
d1d438a3 138 } syscalls;
ff7a4f98
ACM		 139 struct {
140 struct bpf_map *map;
141 } dump;
d1d438a3 142 struct record_opts opts;
63503dba 143 struct evlist *evlist;
d1d438a3
ACM		 144 struct machine *host;
145 struct thread *current;
c8c80570 146 struct bpf_object *bpf_obj;
9ea42ba4 147 struct cgroup *cgroup;
d1d438a3
ACM		 148 u64 base_time;
149 FILE *output;
150 unsigned long nr_events;
5067a8cd
ACM		 151 unsigned long nr_events_printed;
152 unsigned long max_events;
22ac4318 153 struct evswitch evswitch;
d1d438a3
ACM		 154 struct strlist *ev_qualifier;
155 struct {
156 size_t nr;
157 int *entries;
158 } ev_qualifier_ids;
d1d438a3
ACM		 159 struct {
160 size_t nr;
161 pid_t *entries;
744fafc7 162 struct bpf_map *map;
d1d438a3
ACM		 163 } filter_pids;
164 double duration_filter;
165 double runtime_ms;
166 struct {
167 u64 vfs_getname,
168 proc_getname;
169 } stats;
c6d4a494 170 unsigned int max_stack;
5cf9c84e 171 unsigned int min_stack;
b9b6a2ea 172 int raw_augmented_syscalls_args_size;
cd26ea6d 173 bool raw_augmented_syscalls;
79d725cd 174 bool fd_path_disabled;
b9b6a2ea 175 bool sort_events;
d1d438a3
ACM		 176 bool not_ev_qualifier;
177 bool live;
178 bool full_time;
179 bool sched;
180 bool multiple_threads;
181 bool summary;
182 bool summary_only;
b88b14db 183 bool errno_summary;
0a6545bd 184 bool failure_only;
d1d438a3 185 bool show_comm;
591421e1 186 bool print_sample;
d1d438a3
ACM		 187 bool show_tool_stats;
188 bool trace_syscalls;
f11b2803 189 bool libtraceevent_print;
44621819 190 bool kernel_syscallchains;
9ed45d59 191 s16 args_alignment;
b036146f 192 bool show_tstamp;
42e4a52d 193 bool show_duration;
e7c634fc 194 bool show_zeros;
9d6dc178 195 bool show_arg_names;
c65c83ff 196 bool show_string_prefix;
d1d438a3
ACM		 197 bool force;
198 bool vfs_getname;
199 int trace_pgfaults;
7e035929 200 char *perfconfig_events;
028713aa
JO		 201 struct {
202 struct ordered_events data;
203 u64 last;
204 } oe;
d1d438a3 205};
a1c2552d 206
77170988
ACM		 207struct tp_field {
208 int offset;
209 union {
210 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
211 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
212 };
213};
214
215#define TP_UINT_FIELD(bits) \
216static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
217{ \
55d43bca
DA		 218 u##bits value; \
219 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
220 return value; \
77170988
ACM		 221}
222
223TP_UINT_FIELD(8);
224TP_UINT_FIELD(16);
225TP_UINT_FIELD(32);
226TP_UINT_FIELD(64);
227
228#define TP_UINT_FIELD__SWAPPED(bits) \
229static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
230{ \
55d43bca
DA		 231 u##bits value; \
232 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
77170988
ACM		 233 return bswap_##bits(value);\
234}
235
236TP_UINT_FIELD__SWAPPED(16);
237TP_UINT_FIELD__SWAPPED(32);
238TP_UINT_FIELD__SWAPPED(64);
239
aa823f58 240static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
77170988 241{
aa823f58 242 field->offset = offset;
77170988 243
aa823f58 244 switch (size) {
77170988
ACM		 245 case 1:
246 field->integer = tp_field__u8;
247 break;
248 case 2:
249 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
250 break;
251 case 4:
252 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
253 break;
254 case 8:
255 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
256 break;
257 default:
258 return -1;
259 }
260
261 return 0;
262}
263
2c92f982 264static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
aa823f58
ACM		 265{
266 return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
267}
268
77170988
ACM		 269static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
270{
271 return sample->raw_data + field->offset;
272}
273
aa823f58 274static int __tp_field__init_ptr(struct tp_field *field, int offset)
77170988 275{
aa823f58 276 field->offset = offset;
77170988
ACM		 277 field->pointer = tp_field__ptr;
278 return 0;
279}
280
2c92f982 281static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
aa823f58
ACM		 282{
283 return __tp_field__init_ptr(field, format_field->offset);
284}
285
77170988
ACM		 286struct syscall_tp {
287 struct tp_field id;
288 union {
289 struct tp_field args, ret;
290 };
291};
292
2b00bb62
ACM		 293/*
294 * The evsel->priv as used by 'perf trace'
295 * sc: for raw_syscalls:sys_{enter,exit} and syscalls:sys_{enter,exit}_SYSCALLNAME
296 * fmt: for all the other tracepoints
297 */
298struct evsel_trace {
299 struct syscall_tp sc;
300 struct syscall_arg_fmt *fmt;
301};
302
303static struct evsel_trace *evsel_trace__new(void)
304{
305 return zalloc(sizeof(struct evsel_trace));
306}
307
308static void evsel_trace__delete(struct evsel_trace *et)
309{
310 if (et == NULL)
311 return;
312
313 zfree(&et->fmt);
314 free(et);
315}
316
fecd9907
ACM		 317/*
318 * Used with raw_syscalls:sys_{enter,exit} and with the
319 * syscalls:sys_{enter,exit}_SYSCALL tracepoints
320 */
321static inline struct syscall_tp *__evsel__syscall_tp(struct evsel *evsel)
322{
2b00bb62 323 struct evsel_trace *et = evsel->priv;
fecd9907 324
2b00bb62 325 return &et->sc;
fecd9907
ACM		 326}
327
8b913df5
ACM		 328static struct syscall_tp *evsel__syscall_tp(struct evsel *evsel)
329{
330 if (evsel->priv == NULL) {
2b00bb62
ACM		 331 evsel->priv = evsel_trace__new();
332 if (evsel->priv == NULL)
333 return NULL;
8b913df5
ACM		 334 }
335
336 return __evsel__syscall_tp(evsel);
337}
338
fecd9907
ACM		 339/*
340 * Used with all the other tracepoints.
341 */
342static inline struct syscall_arg_fmt *__evsel__syscall_arg_fmt(struct evsel *evsel)
343{
2b00bb62 344 struct evsel_trace *et = evsel->priv;
fecd9907 345
2b00bb62 346 return et->fmt;
fecd9907
ACM		 347}
348
8b913df5
ACM		 349static struct syscall_arg_fmt *evsel__syscall_arg_fmt(struct evsel *evsel)
350{
2b00bb62
ACM		 351 struct evsel_trace *et = evsel->priv;
352
8b913df5 353 if (evsel->priv == NULL) {
2b00bb62
ACM		 354 et = evsel->priv = evsel_trace__new();
355
356 if (et == NULL)
357 return NULL;
358 }
359
360 if (et->fmt == NULL) {
361 et->fmt = calloc(evsel->tp_format->format.nr_fields, sizeof(struct syscall_arg_fmt));
362 if (et->fmt == NULL)
363 goto out_delete;
8b913df5
ACM		 364 }
365
366 return __evsel__syscall_arg_fmt(evsel);
2b00bb62
ACM		 367
368out_delete:
369 evsel_trace__delete(evsel->priv);
370 evsel->priv = NULL;
371 return NULL;
8b913df5
ACM		 372}
373
794bca26 374static int evsel__init_tp_uint_field(struct evsel *evsel, struct tp_field *field, const char *name)
77170988 375{
efc0cdc9 376 struct tep_format_field *format_field = evsel__field(evsel, name);
77170988
ACM		 377
378 if (format_field == NULL)
379 return -1;
380
381 return tp_field__init_uint(field, format_field, evsel->needs_swap);
382}
383
384#define perf_evsel__init_sc_tp_uint_field(evsel, name) \
fecd9907 385 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
794bca26 386 evsel__init_tp_uint_field(evsel, &sc->name, #name); })
77170988 387
794bca26 388static int evsel__init_tp_ptr_field(struct evsel *evsel, struct tp_field *field, const char *name)
77170988 389{
efc0cdc9 390 struct tep_format_field *format_field = evsel__field(evsel, name);
77170988
ACM		 391
392 if (format_field == NULL)
393 return -1;
394
395 return tp_field__init_ptr(field, format_field);
396}
397
398#define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
fecd9907 399 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
794bca26 400 evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
77170988 401
5eb2dd2a 402static void evsel__delete_priv(struct evsel *evsel)
77170988 403{
04662523 404 zfree(&evsel->priv);
5eb2dd2a 405 evsel__delete(evsel);
77170988
ACM		 406}
407
794bca26 408static int evsel__init_syscall_tp(struct evsel *evsel)
d32855fa 409{
8b913df5 410 struct syscall_tp *sc = evsel__syscall_tp(evsel);
d32855fa 411
8b913df5 412 if (sc != NULL) {
794bca26
ACM		 413 if (evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
414 evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
8b913df5 415 return -ENOENT;
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
9b2036cd 915static 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
bcddbfc5
ACM		 1177static struct syscall_fmt *__syscall_fmt__find(struct syscall_fmt *fmts, const int nmemb, const char *name)
1178{
1179 return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1180}
1181
514f1c67
ACM		 1182static struct syscall_fmt *syscall_fmt__find(const char *name)
1183{
1184 const int nmemb = ARRAY_SIZE(syscall_fmts);
bcddbfc5 1185 return __syscall_fmt__find(syscall_fmts, nmemb, name);
514f1c67
ACM		 1186}
1187
bcddbfc5 1188static struct syscall_fmt *__syscall_fmt__find_by_alias(struct syscall_fmt *fmts, const int nmemb, const char *alias)
f932184e 1189{
bcddbfc5 1190 int i;
f932184e
ACM		 1191
1192 for (i = 0; i < nmemb; ++i) {
bcddbfc5
ACM		 1193 if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0)
1194 return &fmts[i];
f932184e
ACM		 1195 }
1196
1197 return NULL;
1198}
1199
bcddbfc5
ACM		 1200static struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
1201{
1202 const int nmemb = ARRAY_SIZE(syscall_fmts);
1203 return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias);
1204}
1205
6a648b53
ACM		 1206/*
1207 * is_exit: is this "exit" or "exit_group"?
1208 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
7a983a0f 1209 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
b8b1033f 1210 * nonexistent: Just a hole in the syscall table, syscall id not allocated
6a648b53 1211 */
514f1c67 1212struct syscall {
97fbf3f0 1213 struct tep_event *tp_format;
f208bd8d 1214 int nr_args;
7a983a0f 1215 int args_size;
6ff8fff4
ACM		 1216 struct {
1217 struct bpf_program *sys_enter,
1218 *sys_exit;
1219 } bpf_prog;
6a648b53
ACM		 1220 bool is_exit;
1221 bool is_open;
b8b1033f 1222 bool nonexistent;
2c92f982 1223 struct tep_format_field *args;
514f1c67
ACM		 1224 const char *name;
1225 struct syscall_fmt *fmt;
82d4a110 1226 struct syscall_arg_fmt *arg_fmt;
514f1c67
ACM		 1227};
1228
fd2b2975
ACM		 1229/*
1230 * We need to have this 'calculated' boolean because in some cases we really
1231 * don't know what is the duration of a syscall, for instance, when we start
1232 * a session and some threads are waiting for a syscall to finish, say 'poll',
1233 * in which case all we can do is to print "( ? ) for duration and for the
1234 * start timestamp.
1235 */
1236static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
60c907ab
ACM		 1237{
1238 double duration = (double)t / NSEC_PER_MSEC;
1239 size_t printed = fprintf(fp, "(");
1240
fd2b2975 1241 if (!calculated)
522283fe 1242 printed += fprintf(fp, " ");
fd2b2975 1243 else if (duration >= 1.0)
60c907ab
ACM		 1244 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1245 else if (duration >= 0.01)
1246 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1247 else
1248 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
c24ff998 1249 return printed + fprintf(fp, "): ");
60c907ab
ACM		 1250}
1251
f994592d
ACM		 1252/**
1253 * filename.ptr: The filename char pointer that will be vfs_getname'd
1254 * filename.entry_str_pos: Where to insert the string translated from
1255 * filename.ptr by the vfs_getname tracepoint/kprobe.
84486caa
ACM		 1256 * ret_scnprintf: syscall args may set this to a different syscall return
1257 * formatter, for instance, fcntl may return fds, file flags, etc.
f994592d 1258 */
752fde44
ACM		 1259struct thread_trace {
1260 u64 entry_time;
752fde44 1261 bool entry_pending;
efd5745e 1262 unsigned long nr_events;
a2ea67d7 1263 unsigned long pfmaj, pfmin;
752fde44 1264 char *entry_str;
1302d88e 1265 double runtime_ms;
7ee57434 1266 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
f994592d
ACM		 1267 struct {
1268 unsigned long ptr;
7f4f8001
ACM		 1269 short int entry_str_pos;
1270 bool pending_open;
1271 unsigned int namelen;
1272 char *name;
f994592d 1273 } filename;
75b757ca 1274 struct {
f4a74fcb
ACM		 1275 int max;
1276 struct file *table;
1277 } files;
bf2575c1
DA		 1278
1279 struct intlist *syscall_stats;
752fde44
ACM		 1280};
1281
1282static struct thread_trace *thread_trace__new(void)
1283{
75b757ca
ACM		 1284 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
1285
7a6d49dc 1286 if (ttrace) {
f4a74fcb 1287 ttrace->files.max = -1;
7a6d49dc
LY		 1288 ttrace->syscall_stats = intlist__new(NULL);
1289 }
bf2575c1 1290
75b757ca 1291 return ttrace;
752fde44
ACM		 1292}
1293
c24ff998 1294static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
752fde44 1295{
efd5745e
ACM		 1296 struct thread_trace *ttrace;
1297
752fde44
ACM		 1298 if (thread == NULL)
1299 goto fail;
1300
89dceb22
NK		 1301 if (thread__priv(thread) == NULL)
1302 thread__set_priv(thread, thread_trace__new());
48000a1a 1303
89dceb22 1304 if (thread__priv(thread) == NULL)
752fde44
ACM		 1305 goto fail;
1306
89dceb22 1307 ttrace = thread__priv(thread);
efd5745e
ACM		 1308 ++ttrace->nr_events;
1309
1310 return ttrace;
752fde44 1311fail:
c24ff998 1312 color_fprintf(fp, PERF_COLOR_RED,
752fde44
ACM		 1313 "WARNING: not enough memory, dropping samples!\n");
1314 return NULL;
1315}
1316
84486caa
ACM		 1317
1318void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
7ee57434 1319 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
84486caa
ACM		 1320{
1321 struct thread_trace *ttrace = thread__priv(arg->thread);
1322
1323 ttrace->ret_scnprintf = ret_scnprintf;
1324}
1325
598d02c5
SF		 1326#define TRACE_PFMAJ (1 << 0)
1327#define TRACE_PFMIN (1 << 1)
1328
e4d44e83
ACM		 1329static const size_t trace__entry_str_size = 2048;
1330
d7e13484 1331static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
75b757ca 1332{
05107486
ACM		 1333 if (fd < 0)
1334 return NULL;
1335
f4a74fcb
ACM		 1336 if (fd > ttrace->files.max) {
1337 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
75b757ca 1338
f4a74fcb 1339 if (nfiles == NULL)
d7e13484 1340 return NULL;
75b757ca 1341
f4a74fcb
ACM		 1342 if (ttrace->files.max != -1) {
1343 memset(nfiles + ttrace->files.max + 1, 0,
1344 (fd - ttrace->files.max) * sizeof(struct file));
75b757ca 1345 } else {
f4a74fcb 1346 memset(nfiles, 0, (fd + 1) * sizeof(struct file));
75b757ca
ACM		 1347 }
1348
f4a74fcb
ACM		 1349 ttrace->files.table = nfiles;
1350 ttrace->files.max = fd;
75b757ca
ACM		 1351 }
1352
d7e13484
ACM		 1353 return ttrace->files.table + fd;
1354}
1355
2d473389
ACM		 1356struct file *thread__files_entry(struct thread *thread, int fd)
1357{
1358 return thread_trace__files_entry(thread__priv(thread), fd);
1359}
1360
d7e13484
ACM		 1361static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1362{
1363 struct thread_trace *ttrace = thread__priv(thread);
1364 struct file *file = thread_trace__files_entry(ttrace, fd);
1365
1366 if (file != NULL) {
4bcc4cff
ACM		 1367 struct stat st;
1368 if (stat(pathname, &st) == 0)
1369 file->dev_maj = major(st.st_rdev);
d7e13484
ACM		 1370 file->pathname = strdup(pathname);
1371 if (file->pathname)
1372 return 0;
1373 }
75b757ca 1374
d7e13484 1375 return -1;
75b757ca
ACM		 1376}
1377
97119f37
ACM		 1378static int thread__read_fd_path(struct thread *thread, int fd)
1379{
1380 char linkname[PATH_MAX], pathname[PATH_MAX];
1381 struct stat st;
1382 int ret;
1383
1384 if (thread->pid_ == thread->tid) {
1385 scnprintf(linkname, sizeof(linkname),
1386 "/proc/%d/fd/%d", thread->pid_, fd);
1387 } else {
1388 scnprintf(linkname, sizeof(linkname),
1389 "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1390 }
1391
1392 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1393 return -1;
1394
1395 ret = readlink(linkname, pathname, sizeof(pathname));
1396
1397 if (ret < 0 || ret > st.st_size)
1398 return -1;
1399
1400 pathname[ret] = '\0';
1401 return trace__set_fd_pathname(thread, fd, pathname);
1402}
1403
c522739d
ACM		 1404static const char *thread__fd_path(struct thread *thread, int fd,
1405 struct trace *trace)
75b757ca 1406{
89dceb22 1407 struct thread_trace *ttrace = thread__priv(thread);
75b757ca 1408
79d725cd 1409 if (ttrace == NULL || trace->fd_path_disabled)
75b757ca
ACM		 1410 return NULL;
1411
1412 if (fd < 0)
1413 return NULL;
1414
f4a74fcb 1415 if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
c522739d
ACM		 1416 if (!trace->live)
1417 return NULL;
1418 ++trace->stats.proc_getname;
cdcd1e6b 1419 if (thread__read_fd_path(thread, fd))
c522739d
ACM		 1420 return NULL;
1421 }
75b757ca 1422
f4a74fcb 1423 return ttrace->files.table[fd].pathname;
75b757ca
ACM		 1424}
1425
fc65eb82 1426size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
75b757ca
ACM		 1427{
1428 int fd = arg->val;
1429 size_t printed = scnprintf(bf, size, "%d", fd);
c522739d 1430 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
75b757ca
ACM		 1431
1432 if (path)
1433 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1434
1435 return printed;
1436}
1437
0a2f7540
ACM		 1438size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1439{
1440 size_t printed = scnprintf(bf, size, "%d", fd);
1441 struct thread *thread = machine__find_thread(trace->host, pid, pid);
1442
1443 if (thread) {
1444 const char *path = thread__fd_path(thread, fd, trace);
1445
1446 if (path)
1447 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1448
1449 thread__put(thread);
1450 }
1451
1452 return printed;
1453}
1454
75b757ca
ACM		 1455static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1456 struct syscall_arg *arg)
1457{
1458 int fd = arg->val;
1459 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
89dceb22 1460 struct thread_trace *ttrace = thread__priv(arg->thread);
75b757ca 1461
f4a74fcb
ACM		 1462 if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1463 zfree(&ttrace->files.table[fd].pathname);
75b757ca
ACM		 1464
1465 return printed;
1466}
1467
f994592d
ACM		 1468static void thread__set_filename_pos(struct thread *thread, const char *bf,
1469 unsigned long ptr)
1470{
1471 struct thread_trace *ttrace = thread__priv(thread);
1472
1473 ttrace->filename.ptr = ptr;
1474 ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1475}
1476
75d1e306
ACM		 1477static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1478{
1479 struct augmented_arg *augmented_arg = arg->augmented.args;
8195168e
ACM		 1480 size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1481 /*
1482 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1483 * we would have two strings, each prefixed by its size.
1484 */
1485 int consumed = sizeof(*augmented_arg) + augmented_arg->size;
75d1e306 1486
016f327c 1487 arg->augmented.args = ((void *)arg->augmented.args) + consumed;
8195168e 1488 arg->augmented.size -= consumed;
75d1e306 1489
8195168e 1490 return printed;
75d1e306
ACM		 1491}
1492
f994592d
ACM		 1493static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1494 struct syscall_arg *arg)
1495{
1496 unsigned long ptr = arg->val;
1497
75d1e306
ACM		 1498 if (arg->augmented.args)
1499 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1500
f994592d
ACM		 1501 if (!arg->trace->vfs_getname)
1502 return scnprintf(bf, size, "%#x", ptr);
1503
1504 thread__set_filename_pos(arg->thread, bf, ptr);
1505 return 0;
1506}
1507
ae9ed035
ACM		 1508static bool trace__filter_duration(struct trace *trace, double t)
1509{
1510 return t < (trace->duration_filter * NSEC_PER_MSEC);
1511}
1512
fd2b2975 1513static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
752fde44
ACM		 1514{
1515 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1516
60c907ab 1517 return fprintf(fp, "%10.3f ", ts);
752fde44
ACM		 1518}
1519
fd2b2975
ACM		 1520/*
1521 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1522 * using ttrace->entry_time for a thread that receives a sys_exit without
1523 * first having received a sys_enter ("poll" issued before tracing session
1524 * starts, lost sys_enter exit due to ring buffer overflow).
1525 */
1526static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1527{
1528 if (tstamp > 0)
1529 return __trace__fprintf_tstamp(trace, tstamp, fp);
1530
1531 return fprintf(fp, " ? ");
1532}
1533
de9f498d 1534static pid_t workload_pid = -1;
92ea0720
IR		 1535static volatile sig_atomic_t done = false;
1536static volatile sig_atomic_t interrupted = false;
f15eb531 1537
de9f498d 1538static void sighandler_interrupt(int sig __maybe_unused)
f15eb531 1539{
de9f498d
CD		 1540 done = interrupted = true;
1541}
1542
1543static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
1544 void *context __maybe_unused)
1545{
1546 if (info->si_pid == workload_pid)
1547 done = true;
f15eb531
NK		 1548}
1549
6dcbd212 1550static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
752fde44 1551{
6dcbd212 1552 size_t printed = 0;
752fde44 1553
50c95cbd
ACM		 1554 if (trace->multiple_threads) {
1555 if (trace->show_comm)
1902efe7 1556 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
38051234 1557 printed += fprintf(fp, "%d ", thread->tid);
50c95cbd 1558 }
752fde44
ACM		 1559
1560 return printed;
1561}
1562
6dcbd212
ACM		 1563static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1564 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1565{
b036146f
ACM		 1566 size_t printed = 0;
1567
1568 if (trace->show_tstamp)
1569 printed = trace__fprintf_tstamp(trace, tstamp, fp);
42e4a52d
ACM		 1570 if (trace->show_duration)
1571 printed += fprintf_duration(duration, duration_calculated, fp);
6dcbd212
ACM		 1572 return printed + trace__fprintf_comm_tid(trace, thread, fp);
1573}
1574
c24ff998 1575static int trace__process_event(struct trace *trace, struct machine *machine,
162f0bef 1576 union perf_event *event, struct perf_sample *sample)
752fde44
ACM		 1577{
1578 int ret = 0;
1579
1580 switch (event->header.type) {
1581 case PERF_RECORD_LOST:
c24ff998 1582 color_fprintf(trace->output, PERF_COLOR_RED,
752fde44 1583 "LOST %" PRIu64 " events!\n", event->lost.lost);
162f0bef 1584 ret = machine__process_lost_event(machine, event, sample);
3ed5ca2e 1585 break;
752fde44 1586 default:
162f0bef 1587 ret = machine__process_event(machine, event, sample);
752fde44
ACM		 1588 break;
1589 }
1590
1591 return ret;
1592}
1593
c24ff998 1594static int trace__tool_process(struct perf_tool *tool,
752fde44 1595 union perf_event *event,
162f0bef 1596 struct perf_sample *sample,
752fde44
ACM		 1597 struct machine *machine)
1598{
c24ff998 1599 struct trace *trace = container_of(tool, struct trace, tool);
162f0bef 1600 return trace__process_event(trace, machine, event, sample);
752fde44
ACM		 1601}
1602
caf8a0d0
ACM		 1603static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1604{
1605 struct machine *machine = vmachine;
1606
1607 if (machine->kptr_restrict_warned)
1608 return NULL;
1609
1610 if (symbol_conf.kptr_restrict) {
1611 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
d06e5fad 1612 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
caf8a0d0
ACM		 1613 "Kernel samples will not be resolved.\n");
1614 machine->kptr_restrict_warned = true;
1615 return NULL;
1616 }
1617
1618 return machine__resolve_kernel_addr(vmachine, addrp, modp);
1619}
1620
63503dba 1621static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
752fde44 1622{
0a7e6d1b 1623 int err = symbol__init(NULL);
752fde44
ACM		 1624
1625 if (err)
1626 return err;
1627
8fb598e5
DA		 1628 trace->host = machine__new_host();
1629 if (trace->host == NULL)
1630 return -ENOMEM;
752fde44 1631
cbd5c178
AV		 1632 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1633 if (err < 0)
1634 goto out;
706c3da4 1635
a33fbd56 1636 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
84111b9c
NK		 1637 evlist->core.threads, trace__tool_process,
1638 true, false, 1);
cbd5c178 1639out:
752fde44
ACM		 1640 if (err)
1641 symbol__exit();
1642
1643 return err;
1644}
1645
33974a41
AV		 1646static void trace__symbols__exit(struct trace *trace)
1647{
1648 machine__exit(trace->host);
1649 trace->host = NULL;
1650
1651 symbol__exit();
1652}
1653
5e58fcfa 1654static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
13d4ff3e 1655{
5e58fcfa 1656 int idx;
13d4ff3e 1657
eadcab4c 1658 if (nr_args == RAW_SYSCALL_ARGS_NUM && sc->fmt && sc->fmt->nr_args != 0)
332337da
ACM		 1659 nr_args = sc->fmt->nr_args;
1660
5e58fcfa 1661 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
82d4a110 1662 if (sc->arg_fmt == NULL)
13d4ff3e
ACM		 1663 return -1;
1664
5e58fcfa
ACM		 1665 for (idx = 0; idx < nr_args; ++idx) {
1666 if (sc->fmt)
82d4a110 1667 sc->arg_fmt[idx] = sc->fmt->arg[idx];
5e58fcfa 1668 }
82d4a110 1669
5e58fcfa
ACM		 1670 sc->nr_args = nr_args;
1671 return 0;
1672}
1673
5d88099b 1674static struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {
df604bfd
ACM		 1675 { .name = "msr", .scnprintf = SCA_X86_MSR, .strtoul = STUL_X86_MSR, },
1676 { .name = "vector", .scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, },
5d88099b
ACM		 1677};
1678
1679static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)
1680{
1681 const struct syscall_arg_fmt *fmt = fmtp;
1682 return strcmp(name, fmt->name);
1683}
1684
1685static struct syscall_arg_fmt *
1686__syscall_arg_fmt__find_by_name(struct syscall_arg_fmt *fmts, const int nmemb, const char *name)
1687{
1688 return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);
1689}
1690
1691static struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)
1692{
1693 const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);
1694 return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);
1695}
1696
8d1d4ff5
ACM		 1697static struct tep_format_field *
1698syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field)
5e58fcfa 1699{
8d1d4ff5
ACM		 1700 struct tep_format_field *last_field = NULL;
1701 int len;
5e58fcfa 1702
8d1d4ff5 1703 for (; field; field = field->next, ++arg) {
7a983a0f
ACM		 1704 last_field = field;
1705
8d1d4ff5 1706 if (arg->scnprintf)
5e58fcfa 1707 continue;
1f115cb7 1708
dea87bfb
ACM		 1709 len = strlen(field->name);
1710
82d4a110 1711 if (strcmp(field->type, "const char *") == 0 &&
dea87bfb
ACM		 1712 ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1713 strstr(field->name, "path") != NULL))
8d1d4ff5 1714 arg->scnprintf = SCA_FILENAME;
ce05539f 1715 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
8d1d4ff5 1716 arg->scnprintf = SCA_PTR;
d1d438a3 1717 else if (strcmp(field->type, "pid_t") == 0)
8d1d4ff5 1718 arg->scnprintf = SCA_PID;
ba2f22cf 1719 else if (strcmp(field->type, "umode_t") == 0)
8d1d4ff5 1720 arg->scnprintf = SCA_MODE_T;
c5e006cd 1721 else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) {
9597945d
ACM		 1722 arg->scnprintf = SCA_CHAR_ARRAY;
1723 arg->nr_entries = field->arraylen;
1724 } else if ((strcmp(field->type, "int") == 0 ||
b6565c90
ACM		 1725 strcmp(field->type, "unsigned int") == 0 ||
1726 strcmp(field->type, "long") == 0) &&
dea87bfb 1727 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
b6565c90
ACM		 1728 /*
1729 * /sys/kernel/tracing/events/syscalls/sys_enter*
818448e9 1730 * grep -E 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
b6565c90
ACM		 1731 * 65 int
1732 * 23 unsigned int
1733 * 7 unsigned long
1734 */
8d1d4ff5 1735 arg->scnprintf = SCA_FD;
5d88099b
ACM		 1736 } else {
1737 struct syscall_arg_fmt *fmt = syscall_arg_fmt__find_by_name(field->name);
1738
3f41b778 1739 if (fmt) {
5d88099b 1740 arg->scnprintf = fmt->scnprintf;
3f41b778
ACM		 1741 arg->strtoul = fmt->strtoul;
1742 }
b6565c90 1743 }
13d4ff3e
ACM		 1744 }
1745
8d1d4ff5
ACM		 1746 return last_field;
1747}
1748
1749static int syscall__set_arg_fmts(struct syscall *sc)
1750{
1751 struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args);
1752
7a983a0f
ACM		 1753 if (last_field)
1754 sc->args_size = last_field->offset + last_field->size;
1755
13d4ff3e
ACM		 1756 return 0;
1757}
1758
514f1c67
ACM		 1759static int trace__read_syscall_info(struct trace *trace, int id)
1760{
1761 char tp_name[128];
1762 struct syscall *sc;
fd0db102 1763 const char *name = syscalltbl__name(trace->sctbl, id);
3a531260 1764
d21cb73a 1765#ifdef HAVE_SYSCALL_TABLE_SUPPORT
30a910d7 1766 if (trace->syscalls.table == NULL) {
3e70008a 1767 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
30a910d7 1768 if (trace->syscalls.table == NULL)
5d2bd889 1769 return -ENOMEM;
514f1c67 1770 }
d21cb73a
ACM		 1771#else
1772 if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) {
1773 // When using libaudit we don't know beforehand what is the max syscall id
1774 struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1775
1776 if (table == NULL)
1777 return -ENOMEM;
514f1c67 1778
f3013f7e
JS		 1779 // Need to memset from offset 0 and +1 members if brand new
1780 if (trace->syscalls.table == NULL)
1781 memset(table, 0, (id + 1) * sizeof(*sc));
1782 else
1783 memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc));
d21cb73a
ACM		 1784
1785 trace->syscalls.table = table;
1786 trace->sctbl->syscalls.max_id = id;
1787 }
1788#endif
514f1c67 1789 sc = trace->syscalls.table + id;
b8b1033f 1790 if (sc->nonexistent)
d4223e17 1791 return -EEXIST;
2ae3a312 1792
b8b1033f
ACM		 1793 if (name == NULL) {
1794 sc->nonexistent = true;
d4223e17 1795 return -EEXIST;
b8b1033f
ACM		 1796 }
1797
1798 sc->name = name;
3a531260 1799 sc->fmt = syscall_fmt__find(sc->name);
514f1c67 1800
aec1930b 1801 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
97978b3e 1802 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
aec1930b 1803
8dd2a131 1804 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
aec1930b 1805 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
97978b3e 1806 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
aec1930b 1807 }
514f1c67 1808
03e9a5d8
LY		 1809 /*
1810 * Fails to read trace point format via sysfs node, so the trace point
1811 * doesn't exist. Set the 'nonexistent' flag as true.
1812 */
1813 if (IS_ERR(sc->tp_format)) {
1814 sc->nonexistent = true;
1815 return PTR_ERR(sc->tp_format);
1816 }
1817
eadcab4c
LY		 1818 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ?
1819 RAW_SYSCALL_ARGS_NUM : sc->tp_format->format.nr_fields))
5d2bd889 1820 return -ENOMEM;
5e58fcfa 1821
f208bd8d 1822 sc->args = sc->tp_format->format.fields;
c42de706
TS		 1823 /*
1824 * We need to check and discard the first variable '__syscall_nr'
1825 * or 'nr' that mean the syscall number. It is needless here.
1826 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1827 */
1828 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
f208bd8d
ACM		 1829 sc->args = sc->args->next;
1830 --sc->nr_args;
1831 }
1832
5089f20e 1833 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
6a648b53 1834 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
5089f20e 1835
13d4ff3e 1836 return syscall__set_arg_fmts(sc);
514f1c67
ACM		 1837}
1838
794bca26 1839static int evsel__init_tp_arg_scnprintf(struct evsel *evsel)
947b843c 1840{
8b913df5 1841 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
947b843c 1842
8b913df5
ACM		 1843 if (fmt != NULL) {
1844 syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields);
947b843c
ACM		 1845 return 0;
1846 }
1847
1848 return -ENOMEM;
1849}
1850
83e69b92
ACM		 1851static int intcmp(const void *a, const void *b)
1852{
1853 const int *one = a, *another = b;
1854
1855 return *one - *another;
1856}
1857
d0cc439b
ACM		 1858static int trace__validate_ev_qualifier(struct trace *trace)
1859{
a4066d64 1860 int err = 0;
04c41bcb 1861 bool printed_invalid_prefix = false;
d0cc439b 1862 struct str_node *pos;
99f26f85 1863 size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
d0cc439b 1864
99f26f85 1865 trace->ev_qualifier_ids.entries = malloc(nr_allocated *
8b3ce757
ACM		 1866 sizeof(trace->ev_qualifier_ids.entries[0]));
1867
1868 if (trace->ev_qualifier_ids.entries == NULL) {
1869 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1870 trace->output);
1871 err = -EINVAL;
1872 goto out;
1873 }
1874
602a1f4d 1875 strlist__for_each_entry(pos, trace->ev_qualifier) {
d0cc439b 1876 const char *sc = pos->s;
27702bcf 1877 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
d0cc439b 1878
8b3ce757 1879 if (id < 0) {
27702bcf
ACM		 1880 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1881 if (id >= 0)
1882 goto matches;
1883
04c41bcb
ACM		 1884 if (!printed_invalid_prefix) {
1885 pr_debug("Skipping unknown syscalls: ");
1886 printed_invalid_prefix = true;
d0cc439b 1887 } else {
04c41bcb 1888 pr_debug(", ");
d0cc439b
ACM		 1889 }
1890
04c41bcb
ACM		 1891 pr_debug("%s", sc);
1892 continue;
d0cc439b 1893 }
27702bcf 1894matches:
99f26f85 1895 trace->ev_qualifier_ids.entries[nr_used++] = id;
27702bcf
ACM		 1896 if (match_next == -1)
1897 continue;
1898
1899 while (1) {
1900 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1901 if (id < 0)
1902 break;
99f26f85 1903 if (nr_allocated == nr_used) {
27702bcf
ACM		 1904 void *entries;
1905
1906 nr_allocated += 8;
1907 entries = realloc(trace->ev_qualifier_ids.entries,
1908 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1909 if (entries == NULL) {
1910 err = -ENOMEM;
1911 fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1912 goto out_free;
1913 }
1914 trace->ev_qualifier_ids.entries = entries;
1915 }
99f26f85 1916 trace->ev_qualifier_ids.entries[nr_used++] = id;
27702bcf 1917 }
d0cc439b
ACM		 1918 }
1919
99f26f85 1920 trace->ev_qualifier_ids.nr = nr_used;
83e69b92 1921 qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
8b3ce757 1922out:
04c41bcb
ACM		 1923 if (printed_invalid_prefix)
1924 pr_debug("\n");
d0cc439b 1925 return err;
04c41bcb
ACM		 1926out_free:
1927 zfree(&trace->ev_qualifier_ids.entries);
1928 trace->ev_qualifier_ids.nr = 0;
1929 goto out;
d0cc439b
ACM		 1930}
1931
3803a229
ACM		 1932static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1933{
1934 bool in_ev_qualifier;
1935
1936 if (trace->ev_qualifier_ids.nr == 0)
1937 return true;
1938
1939 in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1940 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1941
1942 if (in_ev_qualifier)
1943 return !trace->not_ev_qualifier;
1944
1945 return trace->not_ev_qualifier;
1946}
1947
55d43bca
DA		 1948/*
1949 * args is to be interpreted as a series of longs but we need to handle
1950 * 8-byte unaligned accesses. args points to raw_data within the event
1951 * and raw_data is guaranteed to be 8-byte unaligned because it is
1952 * preceded by raw_size which is a u32. So we need to copy args to a temp
1953 * variable to read it. Most notably this avoids extended load instructions
1954 * on unaligned addresses
1955 */
325f5091 1956unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
f9f83b33
ACM		 1957{
1958 unsigned long val;
325f5091 1959 unsigned char *p = arg->args + sizeof(unsigned long) * idx;
f9f83b33
ACM		 1960
1961 memcpy(&val, p, sizeof(val));
1962 return val;
1963}
1964
c51bdfec
ACM		 1965static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1966 struct syscall_arg *arg)
1967{
1968 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1969 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1970
1971 return scnprintf(bf, size, "arg%d: ", arg->idx);
1972}
1973
496fd346
ACM		 1974/*
1975 * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1976 * as mount 'flags' argument that needs ignoring some magic flag, see comment
1977 * in tools/perf/trace/beauty/mount_flags.c
1978 */
3e0c9b2c 1979static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)
496fd346 1980{
3e0c9b2c
ACM		 1981 if (fmt && fmt->mask_val)
1982 return fmt->mask_val(arg, val);
496fd346
ACM		 1983
1984 return val;
1985}
1986
3e0c9b2c
ACM		 1987static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,
1988 struct syscall_arg *arg, unsigned long val)
d032d79e 1989{
3e0c9b2c 1990 if (fmt && fmt->scnprintf) {
d032d79e 1991 arg->val = val;
3e0c9b2c
ACM		 1992 if (fmt->parm)
1993 arg->parm = fmt->parm;
1994 return fmt->scnprintf(bf, size, arg);
d032d79e
ACM		 1995 }
1996 return scnprintf(bf, size, "%ld", val);
1997}
1998
752fde44 1999static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
7a983a0f
ACM		 2000 unsigned char *args, void *augmented_args, int augmented_args_size,
2001 struct trace *trace, struct thread *thread)
514f1c67 2002{
514f1c67 2003 size_t printed = 0;
55d43bca 2004 unsigned long val;
d032d79e
ACM		 2005 u8 bit = 1;
2006 struct syscall_arg arg = {
2007 .args = args,
7a983a0f
ACM		 2008 .augmented = {
2009 .size = augmented_args_size,
2010 .args = augmented_args,
2011 },
d032d79e
ACM		 2012 .idx = 0,
2013 .mask = 0,
2014 .trace = trace,
2015 .thread = thread,
c65c83ff 2016 .show_string_prefix = trace->show_string_prefix,
d032d79e 2017 };
84486caa
ACM		 2018 struct thread_trace *ttrace = thread__priv(thread);
2019
2020 /*
2021 * Things like fcntl will set this in its 'cmd' formatter to pick the
2022 * right formatter for the return value (an fd? file flags?), which is
2023 * not needed for syscalls that always return a given type, say an fd.
2024 */
2025 ttrace->ret_scnprintf = NULL;
514f1c67 2026
f208bd8d 2027 if (sc->args != NULL) {
2c92f982 2028 struct tep_format_field *field;
6e7eeb51 2029
f208bd8d 2030 for (field = sc->args; field;
01533e97
ACM		 2031 field = field->next, ++arg.idx, bit <<= 1) {
2032 if (arg.mask & bit)
6e7eeb51 2033 continue;
55d43bca 2034
888ca854 2035 arg.fmt = &sc->arg_fmt[arg.idx];
f9f83b33 2036 val = syscall_arg__val(&arg, arg.idx);
496fd346
ACM		 2037 /*
2038 * Some syscall args need some mask, most don't and
2039 * return val untouched.
2040 */
3e0c9b2c 2041 val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);
55d43bca 2042
4aa58232
ACM		 2043 /*
2044 * Suppress this argument if its value is zero and
2045 * and we don't have a string associated in an
2046 * strarray for it.
2047 */
55d43bca 2048 if (val == 0 &&
e7c634fc 2049 !trace->show_zeros &&
82d4a110 2050 !(sc->arg_fmt &&
d47737d5
ACM		 2051 (sc->arg_fmt[arg.idx].show_zero ||
2052 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
82d4a110
ACM		 2053 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
2054 sc->arg_fmt[arg.idx].parm))
22ae5cf1
ACM		 2055 continue;
2056
9d6dc178
ACM		 2057 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2058
2059 if (trace->show_arg_names)
2060 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2061
3e0c9b2c
ACM		 2062 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],
2063 bf + printed, size - printed, &arg, val);
514f1c67 2064 }
4c4d6e51
ACM		 2065 } else if (IS_ERR(sc->tp_format)) {
2066 /*
2067 * If we managed to read the tracepoint /format file, then we
2068 * may end up not having any args, like with gettid(), so only
2069 * print the raw args when we didn't manage to read it.
2070 */
332337da 2071 while (arg.idx < sc->nr_args) {
d032d79e
ACM		 2072 if (arg.mask & bit)
2073 goto next_arg;
2074 val = syscall_arg__val(&arg, arg.idx);
c51bdfec
ACM		 2075 if (printed)
2076 printed += scnprintf(bf + printed, size - printed, ", ");
2077 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
3e0c9b2c 2078 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);
d032d79e
ACM		 2079next_arg:
2080 ++arg.idx;
2081 bit <<= 1;
514f1c67
ACM		 2082 }
2083 }
2084
2085 return printed;
2086}
2087
32dcd021 2088typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
0c82adcf 2089 union perf_event *event,
ba3d7dee
ACM		 2090 struct perf_sample *sample);
2091
2092static struct syscall *trace__syscall_info(struct trace *trace,
32dcd021 2093 struct evsel *evsel, int id)
ba3d7dee 2094{
5d2bd889 2095 int err = 0;
ba3d7dee
ACM		 2096
2097 if (id < 0) {
adaa18bf
ACM		 2098
2099 /*
2100 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
2101 * before that, leaving at a higher verbosity level till that is
2102 * explained. Reproduced with plain ftrace with:
2103 *
2104 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
2105 * grep "NR -1 " /t/trace_pipe
2106 *
2107 * After generating some load on the machine.
2108 */
2109 if (verbose > 1) {
2110 static u64 n;
2111 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
8ab2e96d 2112 id, evsel__name(evsel), ++n);
adaa18bf 2113 }
ba3d7dee
ACM		 2114 return NULL;
2115 }
2116
b8b1033f
ACM		 2117 err = -EINVAL;
2118
d21cb73a
ACM		 2119#ifdef HAVE_SYSCALL_TABLE_SUPPORT
2120 if (id > trace->sctbl->syscalls.max_id) {
2121#else
2122 if (id >= trace->sctbl->syscalls.max_id) {
2123 /*
2124 * With libaudit we don't know beforehand what is the max_id,
2125 * so we let trace__read_syscall_info() figure that out as we
2126 * go on reading syscalls.
2127 */
2128 err = trace__read_syscall_info(trace, id);
2129 if (err)
2130#endif
ba3d7dee 2131 goto out_cant_read;
d21cb73a 2132 }
ba3d7dee 2133
30a910d7
ACM		 2134 if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
2135 (err = trace__read_syscall_info(trace, id)) != 0)
ba3d7dee
ACM		 2136 goto out_cant_read;
2137
03e9a5d8 2138 if (trace->syscalls.table && trace->syscalls.table[id].nonexistent)
b8b1033f 2139 goto out_cant_read;
b8b1033f 2140
ba3d7dee
ACM		 2141 return &trace->syscalls.table[id];
2142
2143out_cant_read:
bb963e16 2144 if (verbose > 0) {
5d2bd889
ACM		 2145 char sbuf[STRERR_BUFSIZE];
2146 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
30a910d7 2147 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
7c304ee0
ACM		 2148 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
2149 fputs(" information\n", trace->output);
2150 }
ba3d7dee
ACM		 2151 return NULL;
2152}
2153
8eded45f
ACM		 2154struct syscall_stats {
2155 struct stats stats;
2156 u64 nr_failures;
b88b14db
ACM		 2157 int max_errno;
2158 u32 *errnos;
8eded45f
ACM		 2159};
2160
b88b14db
ACM		 2161static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace,
2162 int id, struct perf_sample *sample, long err, bool errno_summary)
bf2575c1
DA		 2163{
2164 struct int_node *inode;
8eded45f 2165 struct syscall_stats *stats;
bf2575c1
DA		 2166 u64 duration = 0;
2167
2168 inode = intlist__findnew(ttrace->syscall_stats, id);
2169 if (inode == NULL)
2170 return;
2171
2172 stats = inode->priv;
2173 if (stats == NULL) {
e3e7572f 2174 stats = zalloc(sizeof(*stats));
bf2575c1
DA		 2175 if (stats == NULL)
2176 return;
8eded45f 2177
8eded45f 2178 init_stats(&stats->stats);
bf2575c1
DA		 2179 inode->priv = stats;
2180 }
2181
2182 if (ttrace->entry_time && sample->time > ttrace->entry_time)
2183 duration = sample->time - ttrace->entry_time;
2184
8eded45f
ACM		 2185 update_stats(&stats->stats, duration);
2186
b88b14db 2187 if (err < 0) {
8eded45f 2188 ++stats->nr_failures;
b88b14db
ACM		 2189
2190 if (!errno_summary)
2191 return;
2192
2193 err = -err;
2194 if (err > stats->max_errno) {
2195 u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32));
2196
2197 if (new_errnos) {
2198 memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32));
2199 } else {
2200 pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n",
2201 thread__comm_str(thread), thread->pid_, thread->tid);
2202 return;
2203 }
2204
2205 stats->errnos = new_errnos;
2206 stats->max_errno = err;
2207 }
2208
2209 ++stats->errnos[err - 1];
2210 }
bf2575c1
DA		 2211}
2212
522283fe 2213static int trace__printf_interrupted_entry(struct trace *trace)
e596663e
ACM		 2214{
2215 struct thread_trace *ttrace;
e596663e 2216 size_t printed;
172bf02d 2217 int len;
e596663e 2218
0a6545bd 2219 if (trace->failure_only || trace->current == NULL)
e596663e
ACM		 2220 return 0;
2221
2222 ttrace = thread__priv(trace->current);
2223
2224 if (!ttrace->entry_pending)
2225 return 0;
2226
522283fe 2227 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
172bf02d
ACM		 2228 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
2229
2230 if (len < trace->args_alignment - 4)
2231 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
e596663e 2232
172bf02d
ACM		 2233 printed += fprintf(trace->output, " ...\n");
2234
2235 ttrace->entry_pending = false;
5067a8cd
ACM		 2236 ++trace->nr_events_printed;
2237
e596663e
ACM		 2238 return printed;
2239}
2240
32dcd021 2241static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
591421e1
ACM		 2242 struct perf_sample *sample, struct thread *thread)
2243{
2244 int printed = 0;
2245
2246 if (trace->print_sample) {
2247 double ts = (double)sample->time / NSEC_PER_MSEC;
2248
2249 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
8ab2e96d 2250 evsel__name(evsel), ts,
591421e1
ACM		 2251 thread__comm_str(thread),
2252 sample->pid, sample->tid, sample->cpu);
2253 }
2254
2255 return printed;
2256}
2257
b9b6a2ea 2258static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
8a041f86
ACM		 2259{
2260 void *augmented_args = NULL;
cd26ea6d
ACM		 2261 /*
2262 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
b9b6a2ea
ACM		 2263 * and there we get all 6 syscall args plus the tracepoint common fields
2264 * that gets calculated at the start and the syscall_nr (another long).
2265 * So we check if that is the case and if so don't look after the
2266 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
2267 * which is fixed.
cd26ea6d
ACM		 2268 *
2269 * We'll revisit this later to pass s->args_size to the BPF augmenter
2270 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
2271 * copies only what we need for each syscall, like what happens when we
2272 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
2273 * traffic to just what is needed for each syscall.
2274 */
b9b6a2ea 2275 int args_size = raw_augmented_args_size ?: sc->args_size;
8a041f86 2276
cd26ea6d 2277 *augmented_args_size = sample->raw_size - args_size;
8a041f86 2278 if (*augmented_args_size > 0)
cd26ea6d 2279 augmented_args = sample->raw_data + args_size;
8a041f86
ACM		 2280
2281 return augmented_args;
2282}
2283
f2ebf8ff
RM		 2284static void syscall__exit(struct syscall *sc)
2285{
2286 if (!sc)
2287 return;
2288
2289 free(sc->arg_fmt);
2290}
2291
32dcd021 2292static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
0c82adcf 2293 union perf_event *event __maybe_unused,
ba3d7dee
ACM		 2294 struct perf_sample *sample)
2295{
752fde44 2296 char *msg;
ba3d7dee 2297 void *args;
4b8a240e 2298 int printed = 0;
2ae3a312 2299 struct thread *thread;
b91fc39f 2300 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
db2da3f8
ACM		 2301 int augmented_args_size = 0;
2302 void *augmented_args = NULL;
bf2575c1 2303 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2ae3a312
ACM		 2304 struct thread_trace *ttrace;
2305
2306 if (sc == NULL)
2307 return -1;
ba3d7dee 2308
8fb598e5 2309 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
c24ff998 2310 ttrace = thread__trace(thread, trace->output);
2ae3a312 2311 if (ttrace == NULL)
b91fc39f 2312 goto out_put;
ba3d7dee 2313
591421e1
ACM		 2314 trace__fprintf_sample(trace, evsel, sample, thread);
2315
77170988 2316 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
752fde44
ACM		 2317
2318 if (ttrace->entry_str == NULL) {
e4d44e83 2319 ttrace->entry_str = malloc(trace__entry_str_size);
752fde44 2320 if (!ttrace->entry_str)
b91fc39f 2321 goto out_put;
752fde44
ACM		 2322 }
2323
5cf9c84e 2324 if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
522283fe 2325 trace__printf_interrupted_entry(trace);
db2da3f8
ACM		 2326 /*
2327 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2328 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2329 * this breaks syscall__augmented_args() check for augmented args, as we calculate
2330 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2331 * so when handling, say the openat syscall, we end up getting 6 args for the
2332 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2333 * thinking that the extra 2 u64 args are the augmented filename, so just check
2334 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2335 */
2336 if (evsel != trace->syscalls.events.sys_enter)
b9b6a2ea 2337 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
752fde44
ACM		 2338 ttrace->entry_time = sample->time;
2339 msg = ttrace->entry_str;
e4d44e83 2340 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
752fde44 2341
e4d44e83 2342 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
db2da3f8 2343 args, augmented_args, augmented_args_size, trace, thread);
752fde44 2344
5089f20e 2345 if (sc->is_exit) {
0a6545bd 2346 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
4b8a240e
ACM		 2347 int alignment = 0;
2348
fd2b2975 2349 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
4b8a240e
ACM		 2350 printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2351 if (trace->args_alignment > printed)
2352 alignment = trace->args_alignment - printed;
2353 fprintf(trace->output, "%*s= ?\n", alignment, " ");
ae9ed035 2354 }
7f4f8001 2355 } else {
752fde44 2356 ttrace->entry_pending = true;
7f4f8001
ACM		 2357 /* See trace__vfs_getname & trace__sys_exit */
2358 ttrace->filename.pending_open = false;
2359 }
ba3d7dee 2360
f3b623b8
ACM		 2361 if (trace->current != thread) {
2362 thread__put(trace->current);
2363 trace->current = thread__get(thread);
2364 }
b91fc39f
ACM		 2365 err = 0;
2366out_put:
2367 thread__put(thread);
2368 return err;
ba3d7dee
ACM		 2369}
2370
32dcd021 2371static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
a98392bb
ACM		 2372 struct perf_sample *sample)
2373{
a98392bb
ACM		 2374 struct thread_trace *ttrace;
2375 struct thread *thread;
f3acd886
ACM		 2376 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2377 struct syscall *sc = trace__syscall_info(trace, evsel, id);
a98392bb 2378 char msg[1024];
7a983a0f
ACM		 2379 void *args, *augmented_args = NULL;
2380 int augmented_args_size;
a98392bb 2381
a98392bb
ACM		 2382 if (sc == NULL)
2383 return -1;
2384
2385 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2386 ttrace = thread__trace(thread, trace->output);
2387 /*
2388 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2389 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2390 */
2391 if (ttrace == NULL)
2392 goto out_put;
2393
f3acd886 2394 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
b9b6a2ea 2395 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
7a983a0f 2396 syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
a98392bb
ACM		 2397 fprintf(trace->output, "%s", msg);
2398 err = 0;
2399out_put:
2400 thread__put(thread);
2401 return err;
2402}
2403
32dcd021 2404static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
5cf9c84e
ACM		 2405 struct perf_sample *sample,
2406 struct callchain_cursor *cursor)
202ff968
ACM		 2407{
2408 struct addr_location al;
1fc632ce
JO		 2409 int max_stack = evsel->core.attr.sample_max_stack ?
2410 evsel->core.attr.sample_max_stack :
3a9e9a47 2411 trace->max_stack;
4291bf5c 2412 int err;
5cf9c84e 2413
4291bf5c 2414 if (machine__resolve(trace->host, &al, sample) < 0)
5cf9c84e
ACM		 2415 return -1;
2416
4291bf5c
ACM		 2417 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
2418 addr_location__put(&al);
2419 return err;
5cf9c84e
ACM		 2420}
2421
2422static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2423{
202ff968 2424 /* TODO: user-configurable print_opts */
e20ab86e
ACM		 2425 const unsigned int print_opts = EVSEL__PRINT_SYM |
2426 EVSEL__PRINT_DSO |
2427 EVSEL__PRINT_UNKNOWN_AS_ADDR;
202ff968 2428
9620bc36 2429 return sample__fprintf_callchain(sample, 38, print_opts, &callchain_cursor, symbol_conf.bt_stop_list, trace->output);
202ff968
ACM		 2430}
2431
32dcd021 2432static const char *errno_to_name(struct evsel *evsel, int err)
092bd3cd 2433{
6e6d1d65 2434 struct perf_env *env = evsel__env(evsel);
092bd3cd
HB		 2435 const char *arch_name = perf_env__arch(env);
2436
2437 return arch_syscalls__strerrno(arch_name, err);
2438}
2439
32dcd021 2440static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
0c82adcf 2441 union perf_event *event __maybe_unused,
ba3d7dee
ACM		 2442 struct perf_sample *sample)
2443{
2c82c3ad 2444 long ret;
60c907ab 2445 u64 duration = 0;
fd2b2975 2446 bool duration_calculated = false;
2ae3a312 2447 struct thread *thread;
4b8a240e
ACM		 2448 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2449 int alignment = trace->args_alignment;
bf2575c1 2450 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2ae3a312
ACM		 2451 struct thread_trace *ttrace;
2452
2453 if (sc == NULL)
2454 return -1;
ba3d7dee 2455
8fb598e5 2456 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
c24ff998 2457 ttrace = thread__trace(thread, trace->output);
2ae3a312 2458 if (ttrace == NULL)
b91fc39f 2459 goto out_put;
ba3d7dee 2460
591421e1
ACM		 2461 trace__fprintf_sample(trace, evsel, sample, thread);
2462
77170988 2463 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
ba3d7dee 2464
8eded45f 2465 if (trace->summary)
b88b14db 2466 thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary);
8eded45f 2467
79d725cd 2468 if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
7f4f8001
ACM		 2469 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2470 ttrace->filename.pending_open = false;
c522739d
ACM		 2471 ++trace->stats.vfs_getname;
2472 }
2473
ae9ed035 2474 if (ttrace->entry_time) {
60c907ab 2475 duration = sample->time - ttrace->entry_time;
ae9ed035
ACM		 2476 if (trace__filter_duration(trace, duration))
2477 goto out;
fd2b2975 2478 duration_calculated = true;
ae9ed035
ACM		 2479 } else if (trace->duration_filter)
2480 goto out;
60c907ab 2481
5cf9c84e
ACM		 2482 if (sample->callchain) {
2483 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2484 if (callchain_ret == 0) {
2485 if (callchain_cursor.nr < trace->min_stack)
2486 goto out;
2487 callchain_ret = 1;
2488 }
2489 }
2490
0a6545bd 2491 if (trace->summary_only || (ret >= 0 && trace->failure_only))
fd2eabaf
DA		 2492 goto out;
2493
fd2b2975 2494 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
752fde44
ACM		 2495
2496 if (ttrace->entry_pending) {
4b8a240e 2497 printed = fprintf(trace->output, "%s", ttrace->entry_str);
752fde44 2498 } else {
ac6e022c 2499 printed += fprintf(trace->output, " ... [");
c24ff998 2500 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
ac6e022c
ACM		 2501 printed += 9;
2502 printed += fprintf(trace->output, "]: %s()", sc->name);
752fde44
ACM		 2503 }
2504
4b8a240e
ACM		 2505 printed++; /* the closing ')' */
2506
2507 if (alignment > printed)
2508 alignment -= printed;
2509 else
2510 alignment = 0;
2511
2512 fprintf(trace->output, ")%*s= ", alignment, " ");
2513
da3c9a44 2514 if (sc->fmt == NULL) {
1f63139c
ACM		 2515 if (ret < 0)
2516 goto errno_print;
da3c9a44 2517signed_print:
4b8a240e 2518 fprintf(trace->output, "%ld", ret);
1f63139c
ACM		 2519 } else if (ret < 0) {
2520errno_print: {
942a91ed 2521 char bf[STRERR_BUFSIZE];
c8b5f2c9 2522 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
092bd3cd 2523 *e = errno_to_name(evsel, -ret);
ba3d7dee 2524
2c83dfae 2525 fprintf(trace->output, "-1 %s (%s)", e, emsg);
1f63139c 2526 }
da3c9a44 2527 } else if (ret == 0 && sc->fmt->timeout)
2c83dfae 2528 fprintf(trace->output, "0 (Timeout)");
84486caa
ACM		 2529 else if (ttrace->ret_scnprintf) {
2530 char bf[1024];
7ee57434
ACM		 2531 struct syscall_arg arg = {
2532 .val = ret,
2533 .thread = thread,
2534 .trace = trace,
2535 };
2536 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
84486caa 2537 ttrace->ret_scnprintf = NULL;
4b8a240e 2538 fprintf(trace->output, "%s", bf);
84486caa 2539 } else if (sc->fmt->hexret)
4b8a240e 2540 fprintf(trace->output, "%#lx", ret);
11c8e39f
ACM		 2541 else if (sc->fmt->errpid) {
2542 struct thread *child = machine__find_thread(trace->host, ret, ret);
2543
2544 if (child != NULL) {
4b8a240e 2545 fprintf(trace->output, "%ld", ret);
11c8e39f
ACM		 2546 if (child->comm_set)
2547 fprintf(trace->output, " (%s)", thread__comm_str(child));
2548 thread__put(child);
2549 }
2550 } else
da3c9a44 2551 goto signed_print;
ba3d7dee 2552
c24ff998 2553 fputc('\n', trace->output);
566a0885 2554
5067a8cd
ACM		 2555 /*
2556 * We only consider an 'event' for the sake of --max-events a non-filtered
2557 * sys_enter + sys_exit and other tracepoint events.
2558 */
2559 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2560 interrupted = true;
2561
5cf9c84e
ACM		 2562 if (callchain_ret > 0)
2563 trace__fprintf_callchain(trace, sample);
2564 else if (callchain_ret < 0)
8ab2e96d 2565 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
ae9ed035 2566out:
752fde44 2567 ttrace->entry_pending = false;
b91fc39f
ACM		 2568 err = 0;
2569out_put:
2570 thread__put(thread);
2571 return err;
ba3d7dee
ACM		 2572}
2573
32dcd021 2574static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
0c82adcf 2575 union perf_event *event __maybe_unused,
c522739d
ACM		 2576 struct perf_sample *sample)
2577{
f994592d
ACM		 2578 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2579 struct thread_trace *ttrace;
2580 size_t filename_len, entry_str_len, to_move;
2581 ssize_t remaining_space;
2582 char *pos;
efc0cdc9 2583 const char *filename = evsel__rawptr(evsel, sample, "pathname");
f994592d
ACM		 2584
2585 if (!thread)
2586 goto out;
2587
2588 ttrace = thread__priv(thread);
2589 if (!ttrace)
ef65e96e 2590 goto out_put;
f994592d 2591
7f4f8001 2592 filename_len = strlen(filename);
39f0e7a8 2593 if (filename_len == 0)
ef65e96e 2594 goto out_put;
7f4f8001
ACM		 2595
2596 if (ttrace->filename.namelen < filename_len) {
2597 char *f = realloc(ttrace->filename.name, filename_len + 1);
2598
2599 if (f == NULL)
ef65e96e 2600 goto out_put;
7f4f8001
ACM		 2601
2602 ttrace->filename.namelen = filename_len;
2603 ttrace->filename.name = f;
2604 }
2605
2606 strcpy(ttrace->filename.name, filename);
2607 ttrace->filename.pending_open = true;
2608
f994592d 2609 if (!ttrace->filename.ptr)
ef65e96e 2610 goto out_put;
f994592d
ACM		 2611
2612 entry_str_len = strlen(ttrace->entry_str);
2613 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2614 if (remaining_space <= 0)
ef65e96e 2615 goto out_put;
f994592d 2616
f994592d
ACM		 2617 if (filename_len > (size_t)remaining_space) {
2618 filename += filename_len - remaining_space;
2619 filename_len = remaining_space;
2620 }
2621
2622 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2623 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2624 memmove(pos + filename_len, pos, to_move);
2625 memcpy(pos, filename, filename_len);
2626
2627 ttrace->filename.ptr = 0;
2628 ttrace->filename.entry_str_pos = 0;
ef65e96e
ACM		 2629out_put:
2630 thread__put(thread);
f994592d 2631out:
c522739d
ACM		 2632 return 0;
2633}
2634
32dcd021 2635static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
0c82adcf 2636 union perf_event *event __maybe_unused,
1302d88e
ACM		 2637 struct perf_sample *sample)
2638{
efc0cdc9 2639 u64 runtime = evsel__intval(evsel, sample, "runtime");
1302d88e 2640 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
8fb598e5 2641 struct thread *thread = machine__findnew_thread(trace->host,
314add6b
AH		 2642 sample->pid,
2643 sample->tid);
c24ff998 2644 struct thread_trace *ttrace = thread__trace(thread, trace->output);
1302d88e
ACM		 2645
2646 if (ttrace == NULL)
2647 goto out_dump;
2648
2649 ttrace->runtime_ms += runtime_ms;
2650 trace->runtime_ms += runtime_ms;
ef65e96e 2651out_put:
b91fc39f 2652 thread__put(thread);
1302d88e
ACM		 2653 return 0;
2654
2655out_dump:
c24ff998 2656 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
1302d88e 2657 evsel->name,
efc0cdc9
ACM		 2658 evsel__strval(evsel, sample, "comm"),
2659 (pid_t)evsel__intval(evsel, sample, "pid"),
1302d88e 2660 runtime,
efc0cdc9 2661 evsel__intval(evsel, sample, "vruntime"));
ef65e96e 2662 goto out_put;
1302d88e
ACM		 2663}
2664
923d0c9a
ACM		 2665static int bpf_output__printer(enum binary_printer_ops op,
2666 unsigned int val, void *extra __maybe_unused, FILE *fp)
1d6c9407 2667{
1d6c9407
WN		 2668 unsigned char ch = (unsigned char)val;
2669
2670 switch (op) {
2671 case BINARY_PRINT_CHAR_DATA:
923d0c9a 2672 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
1d6c9407
WN		 2673 case BINARY_PRINT_DATA_BEGIN:
2674 case BINARY_PRINT_LINE_BEGIN:
2675 case BINARY_PRINT_ADDR:
2676 case BINARY_PRINT_NUM_DATA:
2677 case BINARY_PRINT_NUM_PAD:
2678 case BINARY_PRINT_SEP:
2679 case BINARY_PRINT_CHAR_PAD:
2680 case BINARY_PRINT_LINE_END:
2681 case BINARY_PRINT_DATA_END:
2682 default:
2683 break;
2684 }
923d0c9a
ACM		 2685
2686 return 0;
1d6c9407
WN		 2687}
2688
2689static void bpf_output__fprintf(struct trace *trace,
2690 struct perf_sample *sample)
2691{
923d0c9a
ACM		 2692 binary__fprintf(sample->raw_data, sample->raw_size, 8,
2693 bpf_output__printer, NULL, trace->output);
5067a8cd 2694 ++trace->nr_events_printed;
1d6c9407
WN		 2695}
2696
f11b2803
ACM		 2697static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,
2698 struct thread *thread, void *augmented_args, int augmented_args_size)
2699{
2700 char bf[2048];
2701 size_t size = sizeof(bf);
2702 struct tep_format_field *field = evsel->tp_format->format.fields;
fecd9907 2703 struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel);
f11b2803
ACM		 2704 size_t printed = 0;
2705 unsigned long val;
2706 u8 bit = 1;
2707 struct syscall_arg syscall_arg = {
2708 .augmented = {
2709 .size = augmented_args_size,
2710 .args = augmented_args,
2711 },
2712 .idx = 0,
2713 .mask = 0,
2714 .trace = trace,
2715 .thread = thread,
2716 .show_string_prefix = trace->show_string_prefix,
2717 };
2718
2719 for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {
2720 if (syscall_arg.mask & bit)
2721 continue;
2722
c5e006cd 2723 syscall_arg.len = 0;
f11b2803 2724 syscall_arg.fmt = arg;
c5e006cd
ACM		 2725 if (field->flags & TEP_FIELD_IS_ARRAY) {
2726 int offset = field->offset;
2727
2728 if (field->flags & TEP_FIELD_IS_DYNAMIC) {
2729 offset = format_field__intval(field, sample, evsel->needs_swap);
2730 syscall_arg.len = offset >> 16;
2731 offset &= 0xffff;
b9a49f8c 2732#ifdef HAVE_LIBTRACEEVENT_TEP_FIELD_IS_RELATIVE
7c689c83
MH		 2733 if (field->flags & TEP_FIELD_IS_RELATIVE)
2734 offset += field->offset + field->size;
b9a49f8c 2735#endif
c5e006cd
ACM		 2736 }
2737
2738 val = (uintptr_t)(sample->raw_data + offset);
2739 } else
f11b2803
ACM		 2740 val = format_field__intval(field, sample, evsel->needs_swap);
2741 /*
2742 * Some syscall args need some mask, most don't and
2743 * return val untouched.
2744 */
2745 val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);
2746
2747 /*
2748 * Suppress this argument if its value is zero and
632f5c22 2749 * we don't have a string associated in an
f11b2803
ACM		 2750 * strarray for it.
2751 */
2752 if (val == 0 &&
2753 !trace->show_zeros &&
2754 !((arg->show_zero ||
2755 arg->scnprintf == SCA_STRARRAY ||
2756 arg->scnprintf == SCA_STRARRAYS) &&
2757 arg->parm))
2758 continue;
2759
2760 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2761
888964a0 2762 if (trace->show_arg_names)
f11b2803
ACM		 2763 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2764
2765 printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);
2766 }
2767
2768 return printed + fprintf(trace->output, "%s", bf);
2769}
2770
32dcd021 2771static int trace__event_handler(struct trace *trace, struct evsel *evsel,
14a052df
ACM		 2772 union perf_event *event __maybe_unused,
2773 struct perf_sample *sample)
2774{
a9c5e6c1 2775 struct thread *thread;
7ad35615 2776 int callchain_ret = 0;
a9c5e6c1
ACM		 2777 /*
2778 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2779 * this event's max_events having been hit and this is an entry coming
2780 * from the ring buffer that we should discard, since the max events
2781 * have already been considered/printed.
2782 */
2783 if (evsel->disabled)
2784 return 0;
2785
2786 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
7ad35615
ACM		 2787
2788 if (sample->callchain) {
2789 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2790 if (callchain_ret == 0) {
2791 if (callchain_cursor.nr < trace->min_stack)
2792 goto out;
2793 callchain_ret = 1;
2794 }
2795 }
2796
522283fe 2797 trace__printf_interrupted_entry(trace);
14a052df 2798 trace__fprintf_tstamp(trace, sample->time, trace->output);
0808921a 2799
42e4a52d 2800 if (trace->trace_syscalls && trace->show_duration)
0808921a
ACM		 2801 fprintf(trace->output, "( ): ");
2802
c4191e55
ACM		 2803 if (thread)
2804 trace__fprintf_comm_tid(trace, thread, trace->output);
2805
1cdf618f
ACM		 2806 if (evsel == trace->syscalls.events.augmented) {
2807 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2808 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2809
2810 if (sc) {
c4191e55
ACM		 2811 fprintf(trace->output, "%s(", sc->name);
2812 trace__fprintf_sys_enter(trace, evsel, sample);
2813 fputc(')', trace->output);
2814 goto newline;
1cdf618f
ACM		 2815 }
2816
2817 /*
2818 * XXX: Not having the associated syscall info or not finding/adding
2819 * the thread should never happen, but if it does...
2820 * fall thru and print it as a bpf_output event.
2821 */
2822 }
2823
311baaf9 2824 fprintf(trace->output, "%s(", evsel->name);
14a052df 2825
c754c382 2826 if (evsel__is_bpf_output(evsel)) {
1cdf618f 2827 bpf_output__fprintf(trace, sample);
1d6c9407 2828 } else if (evsel->tp_format) {
a98392bb
ACM		 2829 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2830 trace__fprintf_sys_enter(trace, evsel, sample)) {
f11b2803
ACM		 2831 if (trace->libtraceevent_print) {
2832 event_format__fprintf(evsel->tp_format, sample->cpu,
2833 sample->raw_data, sample->raw_size,
2834 trace->output);
2835 } else {
2836 trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);
2837 }
a98392bb 2838 }
14a052df
ACM		 2839 }
2840
1cdf618f 2841newline:
311baaf9 2842 fprintf(trace->output, ")\n");
202ff968 2843
7ad35615
ACM		 2844 if (callchain_ret > 0)
2845 trace__fprintf_callchain(trace, sample);
2846 else if (callchain_ret < 0)
8ab2e96d 2847 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
db25bf98
ACM		 2848
2849 ++trace->nr_events_printed;
2850
2851 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2852 evsel__disable(evsel);
2853 evsel__close(evsel);
2854 }
7ad35615 2855out:
a937c665 2856 thread__put(thread);
14a052df
ACM		 2857 return 0;
2858}
2859
598d02c5
SF		 2860static void print_location(FILE *f, struct perf_sample *sample,
2861 struct addr_location *al,
2862 bool print_dso, bool print_sym)
2863{
2864
bb963e16 2865 if ((verbose > 0 || print_dso) && al->map)
598d02c5
SF		 2866 fprintf(f, "%s@", al->map->dso->long_name);
2867
bb963e16 2868 if ((verbose > 0 || print_sym) && al->sym)
4414a3c5 2869 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
598d02c5
SF		 2870 al->addr - al->sym->start);
2871 else if (al->map)
4414a3c5 2872 fprintf(f, "0x%" PRIx64, al->addr);
598d02c5 2873 else
4414a3c5 2874 fprintf(f, "0x%" PRIx64, sample->addr);
598d02c5
SF		 2875}
2876
2877static int trace__pgfault(struct trace *trace,
32dcd021 2878 struct evsel *evsel,
473398a2 2879 union perf_event *event __maybe_unused,
598d02c5
SF		 2880 struct perf_sample *sample)
2881{
2882 struct thread *thread;
598d02c5
SF		 2883 struct addr_location al;
2884 char map_type = 'd';
a2ea67d7 2885 struct thread_trace *ttrace;
b91fc39f 2886 int err = -1;
1df54290 2887 int callchain_ret = 0;
598d02c5
SF		 2888
2889 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1df54290
ACM		 2890
2891 if (sample->callchain) {
2892 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2893 if (callchain_ret == 0) {
2894 if (callchain_cursor.nr < trace->min_stack)
2895 goto out_put;
2896 callchain_ret = 1;
2897 }
2898 }
2899
a2ea67d7
SF		 2900 ttrace = thread__trace(thread, trace->output);
2901 if (ttrace == NULL)
b91fc39f 2902 goto out_put;
a2ea67d7 2903
1fc632ce 2904 if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
a2ea67d7
SF		 2905 ttrace->pfmaj++;
2906 else
2907 ttrace->pfmin++;
2908
2909 if (trace->summary_only)
b91fc39f 2910 goto out;
598d02c5 2911
4546263d 2912 thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
598d02c5 2913
fd2b2975 2914 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
598d02c5
SF		 2915
2916 fprintf(trace->output, "%sfault [",
1fc632ce 2917 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
598d02c5
SF		 2918 "maj" : "min");
2919
2920 print_location(trace->output, sample, &al, false, true);
2921
2922 fprintf(trace->output, "] => ");
2923
117d3c24 2924 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
598d02c5
SF		 2925
2926 if (!al.map) {
4546263d 2927 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
598d02c5
SF		 2928
2929 if (al.map)
2930 map_type = 'x';
2931 else
2932 map_type = '?';
2933 }
2934
2935 print_location(trace->output, sample, &al, true, false);
2936
2937 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
0c3a6ef4 2938
1df54290
ACM		 2939 if (callchain_ret > 0)
2940 trace__fprintf_callchain(trace, sample);
2941 else if (callchain_ret < 0)
8ab2e96d 2942 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
5067a8cd
ACM		 2943
2944 ++trace->nr_events_printed;
b91fc39f
ACM		 2945out:
2946 err = 0;
2947out_put:
2948 thread__put(thread);
2949 return err;
598d02c5
SF		 2950}
2951
e6001980 2952static void trace__set_base_time(struct trace *trace,
32dcd021 2953 struct evsel *evsel,
e6001980
ACM		 2954 struct perf_sample *sample)
2955{
8a07a809
ACM		 2956 /*
2957 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2958 * and don't use sample->time unconditionally, we may end up having
2959 * some other event in the future without PERF_SAMPLE_TIME for good
2960 * reason, i.e. we may not be interested in its timestamps, just in
2961 * it taking place, picking some piece of information when it
2962 * appears in our event stream (vfs_getname comes to mind).
2963 */
2964 if (trace->base_time == 0 && !trace->full_time &&
1fc632ce 2965 (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
e6001980
ACM		 2966 trace->base_time = sample->time;
2967}
2968
6810fc91 2969static int trace__process_sample(struct perf_tool *tool,
0c82adcf 2970 union perf_event *event,
6810fc91 2971 struct perf_sample *sample,
32dcd021 2972 struct evsel *evsel,
6810fc91
DA		 2973 struct machine *machine __maybe_unused)
2974{
2975 struct trace *trace = container_of(tool, struct trace, tool);
aa07df6e 2976 struct thread *thread;
6810fc91
DA		 2977 int err = 0;
2978
744a9719 2979 tracepoint_handler handler = evsel->handler;
6810fc91 2980
aa07df6e
DA		 2981 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2982 if (thread && thread__is_filtered(thread))
ef65e96e 2983 goto out;
bdc89661 2984
e6001980 2985 trace__set_base_time(trace, evsel, sample);
6810fc91 2986
3160565f
DA		 2987 if (handler) {
2988 ++trace->nr_events;
0c82adcf 2989 handler(trace, evsel, event, sample);
3160565f 2990 }
ef65e96e
ACM		 2991out:
2992 thread__put(thread);
6810fc91
DA		 2993 return err;
2994}
2995
1e28fe0a 2996static int trace__record(struct trace *trace, int argc, const char **argv)
5e2485b1
DA		 2997{
2998 unsigned int rec_argc, i, j;
2999 const char **rec_argv;
3000 const char * const record_args[] = {
3001 "record",
3002 "-R",
3003 "-m", "1024",
3004 "-c", "1",
5e2485b1 3005 };
7fbfe22c
ACM		 3006 pid_t pid = getpid();
3007 char *filter = asprintf__tp_filter_pids(1, &pid);
1e28fe0a
SF		 3008 const char * const sc_args[] = { "-e", };
3009 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
3010 const char * const majpf_args[] = { "-e", "major-faults" };
3011 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
3012 const char * const minpf_args[] = { "-e", "minor-faults" };
3013 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
7fbfe22c 3014 int err = -1;
1e28fe0a 3015
7fbfe22c
ACM		 3016 /* +3 is for the event string below and the pid filter */
3017 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 +
1e28fe0a 3018 majpf_args_nr + minpf_args_nr + argc;
5e2485b1
DA		 3019 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3020
7fbfe22c
ACM		 3021 if (rec_argv == NULL || filter == NULL)
3022 goto out_free;
5e2485b1 3023
1e28fe0a 3024 j = 0;
5e2485b1 3025 for (i = 0; i < ARRAY_SIZE(record_args); i++)
1e28fe0a
SF		 3026 rec_argv[j++] = record_args[i];
3027
e281a960
SF		 3028 if (trace->trace_syscalls) {
3029 for (i = 0; i < sc_args_nr; i++)
3030 rec_argv[j++] = sc_args[i];
3031
3032 /* event string may be different for older kernels - e.g., RHEL6 */
3033 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
3034 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
3035 else if (is_valid_tracepoint("syscalls:sys_enter"))
3036 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
3037 else {
3038 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
7fbfe22c 3039 goto out_free;
e281a960 3040 }
9aca7f17 3041 }
9aca7f17 3042
7fbfe22c
ACM		 3043 rec_argv[j++] = "--filter";
3044 rec_argv[j++] = filter;
3045
1e28fe0a
SF		 3046 if (trace->trace_pgfaults & TRACE_PFMAJ)
3047 for (i = 0; i < majpf_args_nr; i++)
3048 rec_argv[j++] = majpf_args[i];
3049
3050 if (trace->trace_pgfaults & TRACE_PFMIN)
3051 for (i = 0; i < minpf_args_nr; i++)
3052 rec_argv[j++] = minpf_args[i];
3053
3054 for (i = 0; i < (unsigned int)argc; i++)
3055 rec_argv[j++] = argv[i];
5e2485b1 3056
7fbfe22c
ACM		 3057 err = cmd_record(j, rec_argv);
3058out_free:
3059 free(filter);
3060 free(rec_argv);
3061 return err;
5e2485b1
DA		 3062}
3063
bf2575c1
DA		 3064static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
3065
a1cf3a75 3066static bool evlist__add_vfs_getname(struct evlist *evlist)
c522739d 3067{
6ab3bc24 3068 bool found = false;
32dcd021 3069 struct evsel *evsel, *tmp;
a910e466
IR		 3070 struct parse_events_error err;
3071 int ret;
8dd2a131 3072
07eafd4e 3073 parse_events_error__init(&err);
a910e466 3074 ret = parse_events(evlist, "probe:vfs_getname*", &err);
07eafd4e
IR		 3075 parse_events_error__exit(&err);
3076 if (ret)
08c98776 3077 return false;
c522739d 3078
6ab3bc24 3079 evlist__for_each_entry_safe(evlist, evsel, tmp) {
8ab2e96d 3080 if (!strstarts(evsel__name(evsel), "probe:vfs_getname"))
6ab3bc24
ACM		 3081 continue;
3082
efc0cdc9 3083 if (evsel__field(evsel, "pathname")) {
6ab3bc24
ACM		 3084 evsel->handler = trace__vfs_getname;
3085 found = true;
3086 continue;
3087 }
3088
b27c4ece 3089 list_del_init(&evsel->core.node);
6ab3bc24 3090 evsel->evlist = NULL;
5eb2dd2a 3091 evsel__delete(evsel);
c522739d
ACM		 3092 }
3093
6ab3bc24 3094 return found;
c522739d
ACM		 3095}
3096
8f6725a2 3097static struct evsel *evsel__new_pgfault(u64 config)
598d02c5 3098{
32dcd021 3099 struct evsel *evsel;
598d02c5
SF		 3100 struct perf_event_attr attr = {
3101 .type = PERF_TYPE_SOFTWARE,
3102 .mmap_data = 1,
598d02c5
SF		 3103 };
3104
3105 attr.config = config;
0524798c 3106 attr.sample_period = 1;
598d02c5
SF		 3107
3108 event_attr_init(&attr);
3109
365c3ae7 3110 evsel = evsel__new(&attr);
0ae537cb
ACM		 3111 if (evsel)
3112 evsel->handler = trace__pgfault;
598d02c5 3113
0ae537cb 3114 return evsel;
598d02c5
SF		 3115}
3116
3cb4d5e0
RM		 3117static void evlist__free_syscall_tp_fields(struct evlist *evlist)
3118{
3119 struct evsel *evsel;
3120
3121 evlist__for_each_entry(evlist, evsel) {
3122 struct evsel_trace *et = evsel->priv;
3123
3124 if (!et || !evsel->tp_format || strcmp(evsel->tp_format->system, "syscalls"))
3125 continue;
3126
3127 free(et->fmt);
3128 free(et);
3129 }
3130}
3131
ddbb1b13
ACM		 3132static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
3133{
3134 const u32 type = event->header.type;
32dcd021 3135 struct evsel *evsel;
ddbb1b13 3136
ddbb1b13
ACM		 3137 if (type != PERF_RECORD_SAMPLE) {
3138 trace__process_event(trace, trace->host, event, sample);
3139 return;
3140 }
3141
3ccf8a7b 3142 evsel = evlist__id2evsel(trace->evlist, sample->id);
ddbb1b13
ACM		 3143 if (evsel == NULL) {
3144 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
3145 return;
3146 }
3147
22ac4318
ACM		 3148 if (evswitch__discard(&trace->evswitch, evsel))
3149 return;
3150
e6001980
ACM		 3151 trace__set_base_time(trace, evsel, sample);
3152
1fc632ce 3153 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
ddbb1b13
ACM		 3154 sample->raw_data == NULL) {
3155 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
8ab2e96d 3156 evsel__name(evsel), sample->tid,
ddbb1b13
ACM		 3157 sample->cpu, sample->raw_size);
3158 } else {
3159 tracepoint_handler handler = evsel->handler;
3160 handler(trace, evsel, event, sample);
3161 }
5067a8cd
ACM		 3162
3163 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
3164 interrupted = true;
ddbb1b13
ACM		 3165}
3166
c27366f0
ACM		 3167static int trace__add_syscall_newtp(struct trace *trace)
3168{
3169 int ret = -1;
63503dba 3170 struct evlist *evlist = trace->evlist;
32dcd021 3171 struct evsel *sys_enter, *sys_exit;
c27366f0 3172
63f11c80 3173 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
c27366f0
ACM		 3174 if (sys_enter == NULL)
3175 goto out;
3176
3177 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
3178 goto out_delete_sys_enter;
3179
63f11c80 3180 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
c27366f0
ACM		 3181 if (sys_exit == NULL)
3182 goto out_delete_sys_enter;
3183
3184 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
3185 goto out_delete_sys_exit;
3186
6ec17b4e
ACM		 3187 evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
3188 evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
08e26396 3189
a1cf3a75
JO		 3190 evlist__add(evlist, sys_enter);
3191 evlist__add(evlist, sys_exit);
c27366f0 3192
2ddd5c04 3193 if (callchain_param.enabled && !trace->kernel_syscallchains) {
44621819
ACM		 3194 /*
3195 * We're interested only in the user space callchain
3196 * leading to the syscall, allow overriding that for
3197 * debugging reasons using --kernel_syscall_callchains
3198 */
1fc632ce 3199 sys_exit->core.attr.exclude_callchain_kernel = 1;
44621819
ACM		 3200 }
3201
8b3ce757
ACM		 3202 trace->syscalls.events.sys_enter = sys_enter;
3203 trace->syscalls.events.sys_exit = sys_exit;
c27366f0
ACM		 3204
3205 ret = 0;
3206out:
3207 return ret;
3208
3209out_delete_sys_exit:
5eb2dd2a 3210 evsel__delete_priv(sys_exit);
c27366f0 3211out_delete_sys_enter:
5eb2dd2a 3212 evsel__delete_priv(sys_enter);
c27366f0
ACM		 3213 goto out;
3214}
3215
246fbe03 3216static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
19867b61
ACM		 3217{
3218 int err = -1;
32dcd021 3219 struct evsel *sys_exit;
19867b61
ACM		 3220 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
3221 trace->ev_qualifier_ids.nr,
3222 trace->ev_qualifier_ids.entries);
3223
3224 if (filter == NULL)
3225 goto out_enomem;
3226
ad681adf 3227 if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) {
b15d0a4c 3228 sys_exit = trace->syscalls.events.sys_exit;
ad681adf 3229 err = evsel__append_tp_filter(sys_exit, filter);
b15d0a4c 3230 }
19867b61
ACM		 3231
3232 free(filter);
3233out:
3234 return err;
3235out_enomem:
3236 errno = ENOMEM;
3237 goto out;
3238}
c27366f0 3239
b27b38ed 3240#ifdef HAVE_LIBBPF_SUPPORT
beb64203
JO		 3241static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace, const char *name)
3242{
3243 if (trace->bpf_obj == NULL)
3244 return NULL;
3245
3246 return bpf_object__find_map_by_name(trace->bpf_obj, name);
3247}
3248
3249static void trace__set_bpf_map_filtered_pids(struct trace *trace)
3250{
3251 trace->filter_pids.map = trace__find_bpf_map_by_name(trace, "pids_filtered");
3252}
3253
3254static void trace__set_bpf_map_syscalls(struct trace *trace)
3255{
beb64203
JO		 3256 trace->syscalls.prog_array.sys_enter = trace__find_bpf_map_by_name(trace, "syscalls_sys_enter");
3257 trace->syscalls.prog_array.sys_exit = trace__find_bpf_map_by_name(trace, "syscalls_sys_exit");
3258}
3259
5834da7f
ACM		 3260static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
3261{
b098f336
KFL		 3262 struct bpf_program *pos, *prog = NULL;
3263 const char *sec_name;
3264
5834da7f
ACM		 3265 if (trace->bpf_obj == NULL)
3266 return NULL;
3267
b098f336
KFL		 3268 bpf_object__for_each_program(pos, trace->bpf_obj) {
3269 sec_name = bpf_program__section_name(pos);
3270 if (sec_name && !strcmp(sec_name, name)) {
3271 prog = pos;
3272 break;
3273 }
3274 }
3275
3276 return prog;
5834da7f
ACM		 3277}
3278
6ff8fff4
ACM		 3279static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
3280 const char *prog_name, const char *type)
3281{
3282 struct bpf_program *prog;
3283
8b8044e5
ACM		 3284 if (prog_name == NULL) {
3285 char default_prog_name[256];
3286 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->name);
3287 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3288 if (prog != NULL)
3289 goto out_found;
3290 if (sc->fmt && sc->fmt->alias) {
3291 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->fmt->alias);
3292 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3293 if (prog != NULL)
3294 goto out_found;
3295 }
6ff8fff4 3296 goto out_unaugmented;
8b8044e5 3297 }
6ff8fff4
ACM		 3298
3299 prog = trace__find_bpf_program_by_title(trace, prog_name);
8b8044e5
ACM		 3300
3301 if (prog != NULL) {
3302out_found:
6ff8fff4 3303 return prog;
8b8044e5 3304 }
6ff8fff4
ACM		 3305
3306 pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
3307 prog_name, type, sc->name);
3308out_unaugmented:
3309 return trace->syscalls.unaugmented_prog;
3310}
3311
3312static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
3313{
3314 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3315
3316 if (sc == NULL)
3317 return;
3318
8b8044e5
ACM		 3319 sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3320 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		 3321}
3322
3803a229
ACM		 3323static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
3324{
3325 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3326 return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3327}
3328
3329static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
3330{
3331 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3332 return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3333}
3334
ad4153f9
ACM		 3335static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
3336{
3337 struct tep_format_field *field, *candidate_field;
3338 int id;
3339
3340 /*
3341 * We're only interested in syscalls that have a pointer:
3342 */
3343 for (field = sc->args; field; field = field->next) {
3344 if (field->flags & TEP_FIELD_IS_POINTER)
3345 goto try_to_find_pair;
3346 }
3347
3348 return NULL;
3349
3350try_to_find_pair:
3351 for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) {
3352 struct syscall *pair = trace__syscall_info(trace, NULL, id);
3353 struct bpf_program *pair_prog;
3354 bool is_candidate = false;
3355
3356 if (pair == NULL || pair == sc ||
3357 pair->bpf_prog.sys_enter == trace->syscalls.unaugmented_prog)
3358 continue;
3359
3360 for (field = sc->args, candidate_field = pair->args;
3361 field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3362 bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3363 candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3364
3365 if (is_pointer) {
3366 if (!candidate_is_pointer) {
3367 // The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3368 continue;
3369 }
3370 } else {
3371 if (candidate_is_pointer) {
3372 // The candidate might copy a pointer we don't have, skip it.
3373 goto next_candidate;
3374 }
3375 continue;
3376 }
3377
3378 if (strcmp(field->type, candidate_field->type))
3379 goto next_candidate;
3380
3381 is_candidate = true;
3382 }
3383
3384 if (!is_candidate)
3385 goto next_candidate;
3386
3387 /*
3388 * Check if the tentative pair syscall augmenter has more pointers, if it has,
3389 * then it may be collecting that and we then can't use it, as it would collect
3390 * more than what is common to the two syscalls.
3391 */
3392 if (candidate_field) {
3393 for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3394 if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3395 goto next_candidate;
3396 }
3397
3398 pair_prog = pair->bpf_prog.sys_enter;
3399 /*
3400 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3401 * have been searched for, so search it here and if it returns the
3402 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3403 * program for a filtered syscall on a non-filtered one.
3404 *
3405 * For instance, we have "!syscalls:sys_enter_renameat" and that is
3406 * useful for "renameat2".
3407 */
3408 if (pair_prog == NULL) {
3409 pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3410 if (pair_prog == trace->syscalls.unaugmented_prog)
3411 goto next_candidate;
3412 }
3413
3414 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3415 return pair_prog;
3416 next_candidate:
3417 continue;
3418 }
3419
3420 return NULL;
3421}
3422
3803a229
ACM		 3423static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3424{
3425 int map_enter_fd = bpf_map__fd(trace->syscalls.prog_array.sys_enter),
3426 map_exit_fd = bpf_map__fd(trace->syscalls.prog_array.sys_exit);
3427 int err = 0, key;
3428
3429 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3430 int prog_fd;
3431
3432 if (!trace__syscall_enabled(trace, key))
3433 continue;
3434
3435 trace__init_syscall_bpf_progs(trace, key);
3436
3437 // It'll get at least the "!raw_syscalls:unaugmented"
3438 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3439 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3440 if (err)
3441 break;
3442 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3443 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3444 if (err)
3445 break;
3446 }
3447
ad4153f9
ACM		 3448 /*
3449 * Now lets do a second pass looking for enabled syscalls without
3450 * an augmenter that have a signature that is a superset of another
3451 * syscall with an augmenter so that we can auto-reuse it.
3452 *
3453 * I.e. if we have an augmenter for the "open" syscall that has
3454 * this signature:
3455 *
3456 * int open(const char *pathname, int flags, mode_t mode);
3457 *
3458 * I.e. that will collect just the first string argument, then we
3459 * can reuse it for the 'creat' syscall, that has this signature:
3460 *
3461 * int creat(const char *pathname, mode_t mode);
3462 *
3463 * and for:
3464 *
3465 * int stat(const char *pathname, struct stat *statbuf);
3466 * int lstat(const char *pathname, struct stat *statbuf);
3467 *
3468 * Because the 'open' augmenter will collect the first arg as a string,
3469 * and leave alone all the other args, which already helps with
3470 * beautifying 'stat' and 'lstat''s pathname arg.
3471 *
3472 * Then, in time, when 'stat' gets an augmenter that collects both
3473 * first and second arg (this one on the raw_syscalls:sys_exit prog
3474 * array tail call, then that one will be used.
3475 */
3476 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3477 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3478 struct bpf_program *pair_prog;
3479 int prog_fd;
3480
3481 if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3482 continue;
3483
3484 /*
3485 * For now we're just reusing the sys_enter prog, and if it
3486 * already has an augmenter, we don't need to find one.
3487 */
3488 if (sc->bpf_prog.sys_enter != trace->syscalls.unaugmented_prog)
3489 continue;
3490
3491 /*
3492 * Look at all the other syscalls for one that has a signature
3493 * that is close enough that we can share:
3494 */
3495 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3496 if (pair_prog == NULL)
3497 continue;
3498
3499 sc->bpf_prog.sys_enter = pair_prog;
3500
3501 /*
3502 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3503 * with the fd for the program we're reusing:
3504 */
3505 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3506 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3507 if (err)
3508 break;
3509 }
3510
3511
3803a229
ACM		 3512 return err;
3513}
8bd436b0
ACM		 3514
3515static void trace__delete_augmented_syscalls(struct trace *trace)
3516{
3517 struct evsel *evsel, *tmp;
3518
3519 evlist__remove(trace->evlist, trace->syscalls.events.augmented);
3520 evsel__delete(trace->syscalls.events.augmented);
3521 trace->syscalls.events.augmented = NULL;
3522
3523 evlist__for_each_entry_safe(trace->evlist, tmp, evsel) {
3524 if (evsel->bpf_obj == trace->bpf_obj) {
3525 evlist__remove(trace->evlist, evsel);
3526 evsel__delete(evsel);
3527 }
3528
3529 }
3530
3531 bpf_object__close(trace->bpf_obj);
3532 trace->bpf_obj = NULL;
3533}
3534#else // HAVE_LIBBPF_SUPPORT
beb64203
JO		 3535static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace __maybe_unused,
3536 const char *name __maybe_unused)
3537{
3538 return NULL;
3539}
3540
3541static void trace__set_bpf_map_filtered_pids(struct trace *trace __maybe_unused)
3542{
3543}
3544
3545static void trace__set_bpf_map_syscalls(struct trace *trace __maybe_unused)
3546{
3547}
3548
5834da7f
ACM		 3549static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace __maybe_unused,
3550 const char *name __maybe_unused)
3551{
3552 return NULL;
3553}
3803a229
ACM		 3554
3555static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace __maybe_unused)
3556{
3557 return 0;
3558}
8bd436b0
ACM		 3559
3560static void trace__delete_augmented_syscalls(struct trace *trace __maybe_unused)
3561{
3562}
b27b38ed
ACM		 3563#endif // HAVE_LIBBPF_SUPPORT
3564
8bd436b0
ACM		 3565static bool trace__only_augmented_syscalls_evsels(struct trace *trace)
3566{
3567 struct evsel *evsel;
3568
3569 evlist__for_each_entry(trace->evlist, evsel) {
3570 if (evsel == trace->syscalls.events.augmented ||
3571 evsel->bpf_obj == trace->bpf_obj)
3572 continue;
3573
3574 return false;
3575 }
3576
3577 return true;
3578}
3579
246fbe03
ACM		 3580static int trace__set_ev_qualifier_filter(struct trace *trace)
3581{
f76214f9
ACM		 3582 if (trace->syscalls.events.sys_enter)
3583 return trace__set_ev_qualifier_tp_filter(trace);
3584 return 0;
246fbe03
ACM		 3585}
3586
a9964c43
ACM		 3587static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3588 size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3589{
3590 int err = 0;
3591#ifdef HAVE_LIBBPF_SUPPORT
3592 bool value = true;
3593 int map_fd = bpf_map__fd(map);
3594 size_t i;
3595
3596 for (i = 0; i < npids; ++i) {
3597 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3598 if (err)
3599 break;
3600 }
3601#endif
3602 return err;
3603}
3604
dd1a5037
ACM		 3605static int trace__set_filter_loop_pids(struct trace *trace)
3606{
a9964c43 3607 unsigned int nr = 1, err;
dd1a5037
ACM		 3608 pid_t pids[32] = {
3609 getpid(),
3610 };
082ab9a1
ACM		 3611 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3612
3613 while (thread && nr < ARRAY_SIZE(pids)) {
3614 struct thread *parent = machine__find_thread(trace->host, thread->ppid, thread->ppid);
3615
3616 if (parent == NULL)
3617 break;
3618
ca2da70c
ACM		 3619 if (!strcmp(thread__comm_str(parent), "sshd") ||
3620 strstarts(thread__comm_str(parent), "gnome-terminal")) {
082ab9a1
ACM		 3621 pids[nr++] = parent->tid;
3622 break;
3623 }
3624 thread = parent;
3625 }
dd1a5037 3626
24bf91a7 3627 err = evlist__append_tp_filter_pids(trace->evlist, nr, pids);
a9964c43
ACM		 3628 if (!err && trace->filter_pids.map)
3629 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3630
3631 return err;
3632}
3633
3634static int trace__set_filter_pids(struct trace *trace)
3635{
3636 int err = 0;
3637 /*
3638 * Better not use !target__has_task() here because we need to cover the
3639 * case where no threads were specified in the command line, but a
3640 * workload was, and in that case we will fill in the thread_map when
7b392ef0 3641 * we fork the workload in evlist__prepare_workload.
a9964c43
ACM		 3642 */
3643 if (trace->filter_pids.nr > 0) {
24bf91a7
ACM		 3644 err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3645 trace->filter_pids.entries);
a9964c43
ACM		 3646 if (!err && trace->filter_pids.map) {
3647 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3648 trace->filter_pids.entries);
3649 }
a2f354e3 3650 } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
a9964c43
ACM		 3651 err = trace__set_filter_loop_pids(trace);
3652 }
3653
3654 return err;
dd1a5037
ACM		 3655}
3656
7ba61524 3657static int __trace__deliver_event(struct trace *trace, union perf_event *event)
1f44b3e2 3658{
63503dba 3659 struct evlist *evlist = trace->evlist;
1f44b3e2 3660 struct perf_sample sample;
2a6599cd 3661 int err = evlist__parse_sample(evlist, event, &sample);
1f44b3e2 3662
1f44b3e2
JO		 3663 if (err)
3664 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3665 else
3666 trace__handle_event(trace, event, &sample);
3667
3668 return 0;
3669}
3670
a761a8d1 3671static int __trace__flush_events(struct trace *trace)
028713aa
JO		 3672{
3673 u64 first = ordered_events__first_time(&trace->oe.data);
3674 u64 flush = trace->oe.last - NSEC_PER_SEC;
3675
3676 /* Is there some thing to flush.. */
3677 if (first && first < flush)
3678 return ordered_events__flush_time(&trace->oe.data, flush);
3679
3680 return 0;
3681}
3682
a761a8d1
ACM		 3683static int trace__flush_events(struct trace *trace)
3684{
3685 return !trace->sort_events ? 0 : __trace__flush_events(trace);
3686}
3687
7ba61524 3688static int trace__deliver_event(struct trace *trace, union perf_event *event)
028713aa 3689{
028713aa
JO		 3690 int err;
3691
a761a8d1
ACM		 3692 if (!trace->sort_events)
3693 return __trace__deliver_event(trace, event);
3694
2a6599cd 3695 err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
028713aa
JO		 3696 if (err && err != -1)
3697 return err;
3698
2292083f 3699 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0, NULL);
028713aa
JO		 3700 if (err)
3701 return err;
3702
7ba61524 3703 return trace__flush_events(trace);
028713aa
JO		 3704}
3705
3706static int ordered_events__deliver_event(struct ordered_events *oe,
3707 struct ordered_event *event)
3708{
3709 struct trace *trace = container_of(oe, struct trace, oe.data);
3710
7ba61524 3711 return __trace__deliver_event(trace, event->event);
028713aa
JO		 3712}
3713
794bca26 3714static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg)
90df0249
ACM		 3715{
3716 struct tep_format_field *field;
fecd9907 3717 struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel);
90df0249
ACM		 3718
3719 if (evsel->tp_format == NULL || fmt == NULL)
3720 return NULL;
3721
3722 for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt)
3723 if (strcmp(field->name, arg) == 0)
3724 return fmt;
3725
3726 return NULL;
3727}
3728
3729static int trace__expand_filter(struct trace *trace __maybe_unused, struct evsel *evsel)
3730{
3731 char *tok, *left = evsel->filter, *new_filter = evsel->filter;
3732
3733 while ((tok = strpbrk(left, "=<>!")) != NULL) {
3734 char *right = tok + 1, *right_end;
3735
3736 if (*right == '=')
3737 ++right;
3738
3739 while (isspace(*right))
3740 ++right;
3741
3742 if (*right == '\0')
3743 break;
3744
3745 while (!isalpha(*left))
3746 if (++left == tok) {
3747 /*
3748 * Bail out, can't find the name of the argument that is being
3749 * used in the filter, let it try to set this filter, will fail later.
3750 */
3751 return 0;
3752 }
3753
3754 right_end = right + 1;
154c978d 3755 while (isalnum(*right_end) || *right_end == '_' || *right_end == '|')
90df0249
ACM		 3756 ++right_end;
3757
3758 if (isalpha(*right)) {
3759 struct syscall_arg_fmt *fmt;
3760 int left_size = tok - left,
3761 right_size = right_end - right;
3762 char arg[128];
3763
3764 while (isspace(left[left_size - 1]))
3765 --left_size;
3766
3767 scnprintf(arg, sizeof(arg), "%.*s", left_size, left);
3768
794bca26 3769 fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg);
90df0249 3770 if (fmt == NULL) {
3cdc8db9
ACM		 3771 pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n",
3772 arg, evsel->name, evsel->filter);
90df0249
ACM		 3773 return -1;
3774 }
3775
3776 pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ",
3777 arg, (int)(right - tok), tok, right_size, right);
3778
3779 if (fmt->strtoul) {
3780 u64 val;
9afec87e
ACM		 3781 struct syscall_arg syscall_arg = {
3782 .parm = fmt->parm,
3783 };
3784
3785 if (fmt->strtoul(right, right_size, &syscall_arg, &val)) {
90df0249
ACM		 3786 char *n, expansion[19];
3787 int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val);
3788 int expansion_offset = right - new_filter;
3789
3790 pr_debug("%s", expansion);
3791
3792 if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) {
3793 pr_debug(" out of memory!\n");
3794 free(new_filter);
3795 return -1;
3796 }
3797 if (new_filter != evsel->filter)
3798 free(new_filter);
3799 left = n + expansion_offset + expansion_lenght;
3800 new_filter = n;
3801 } else {
3802 pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3803 right_size, right, arg, evsel->name, evsel->filter);
3804 return -1;
3805 }
3806 } else {
3807 pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3808 arg, evsel->name, evsel->filter);
3809 return -1;
3810 }
3811
3812 pr_debug("\n");
3813 } else {
3814 left = right_end;
3815 }
3816 }
3817
3818 if (new_filter != evsel->filter) {
3819 pr_debug("New filter for %s: %s\n", evsel->name, new_filter);
ad681adf 3820 evsel__set_filter(evsel, new_filter);
90df0249
ACM		 3821 free(new_filter);
3822 }
3823
3824 return 0;
3825}
3826
3827static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel)
3828{
3829 struct evlist *evlist = trace->evlist;
3830 struct evsel *evsel;
3831
3832 evlist__for_each_entry(evlist, evsel) {
3833 if (evsel->filter == NULL)
3834 continue;
3835
3836 if (trace__expand_filter(trace, evsel)) {
3837 *err_evsel = evsel;
3838 return -1;
3839 }
3840 }
3841
3842 return 0;
3843}
3844
f15eb531 3845static int trace__run(struct trace *trace, int argc, const char **argv)
514f1c67 3846{
63503dba 3847 struct evlist *evlist = trace->evlist;
32dcd021 3848 struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
efd5745e
ACM		 3849 int err = -1, i;
3850 unsigned long before;
f15eb531 3851 const bool forks = argc > 0;
46fb3c21 3852 bool draining = false;
514f1c67 3853
75b757ca
ACM		 3854 trace->live = true;
3855
0df50e0b
ACM		 3856 if (!trace->raw_augmented_syscalls) {
3857 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3858 goto out_error_raw_syscalls;
514f1c67 3859
0df50e0b 3860 if (trace->trace_syscalls)
a1cf3a75 3861 trace->vfs_getname = evlist__add_vfs_getname(evlist);
0df50e0b 3862 }
c522739d 3863
0ae537cb 3864 if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
8f6725a2 3865 pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
0ae537cb
ACM		 3866 if (pgfault_maj == NULL)
3867 goto out_error_mem;
6ec17b4e 3868 evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
a1cf3a75 3869 evlist__add(evlist, pgfault_maj);
e2726d99 3870 }
598d02c5 3871
0ae537cb 3872 if ((trace->trace_pgfaults & TRACE_PFMIN)) {
8f6725a2 3873 pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
0ae537cb
ACM		 3874 if (pgfault_min == NULL)
3875 goto out_error_mem;
6ec17b4e 3876 evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
a1cf3a75 3877 evlist__add(evlist, pgfault_min);
0ae537cb 3878 }
598d02c5 3879
b4515ad6
GL		 3880 /* Enable ignoring missing threads when -u/-p option is defined. */
3881 trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid;
3882
1302d88e 3883 if (trace->sched &&
e251abee 3884 evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime))
2cc990ba 3885 goto out_error_sched_stat_runtime;
9ea42ba4
ACM		 3886 /*
3887 * If a global cgroup was set, apply it to all the events without an
3888 * explicit cgroup. I.e.:
3889 *
3890 * trace -G A -e sched:*switch
3891 *
3892 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3893 * _and_ sched:sched_switch to the 'A' cgroup, while:
3894 *
3895 * trace -e sched:*switch -G A
3896 *
3897 * will only set the sched:sched_switch event to the 'A' cgroup, all the
3898 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3899 * a cgroup (on the root cgroup, sys wide, etc).
3900 *
3901 * Multiple cgroups:
3902 *
3903 * trace -G A -e sched:*switch -G B
3904 *
3905 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
3906 * to the 'B' cgroup.
3907 *
3908 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
3909 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
3910 */
3911 if (trace->cgroup)
3912 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
3913
7748bb71 3914 err = evlist__create_maps(evlist, &trace->opts.target);
514f1c67 3915 if (err < 0) {
c24ff998 3916 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
514f1c67
ACM		 3917 goto out_delete_evlist;
3918 }
3919
752fde44
ACM		 3920 err = trace__symbols_init(trace, evlist);
3921 if (err < 0) {
c24ff998 3922 fprintf(trace->output, "Problems initializing symbol libraries!\n");
03ad9747 3923 goto out_delete_evlist;
752fde44
ACM		 3924 }
3925
78e1bc25 3926 evlist__config(evlist, &trace->opts, &callchain_param);
fde54b78 3927
f15eb531 3928 if (forks) {
7b392ef0 3929 err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL);
f15eb531 3930 if (err < 0) {
c24ff998 3931 fprintf(trace->output, "Couldn't run the workload!\n");
03ad9747 3932 goto out_delete_evlist;
f15eb531 3933 }
de9f498d 3934 workload_pid = evlist->workload.pid;
f15eb531
NK		 3935 }
3936
474ddc4c 3937 err = evlist__open(evlist);
a8f23d8f
ACM		 3938 if (err < 0)
3939 goto out_error_open;
514f1c67 3940
ba504235
WN		 3941 err = bpf__apply_obj_config();
3942 if (err) {
3943 char errbuf[BUFSIZ];
3944
3945 bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
3946 pr_err("ERROR: Apply config to BPF failed: %s\n",
3947 errbuf);
3948 goto out_error_open;
3949 }
3950
a9964c43 3951 err = trace__set_filter_pids(trace);
94ad89bc
ACM		 3952 if (err < 0)
3953 goto out_error_mem;
3954
3803a229
ACM		 3955 if (trace->syscalls.prog_array.sys_enter)
3956 trace__init_syscalls_bpf_prog_array_maps(trace);
3957
19867b61
ACM		 3958 if (trace->ev_qualifier_ids.nr > 0) {
3959 err = trace__set_ev_qualifier_filter(trace);
3960 if (err < 0)
3961 goto out_errno;
19867b61 3962
0df50e0b
ACM		 3963 if (trace->syscalls.events.sys_exit) {
3964 pr_debug("event qualifier tracepoint filter: %s\n",
3965 trace->syscalls.events.sys_exit->filter);
3966 }
2e5e5f87 3967 }
19867b61 3968
79d725cd
ACM		 3969 /*
3970 * If the "close" syscall is not traced, then we will not have the
3971 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
3972 * fd->pathname table and were ending up showing the last value set by
3973 * syscalls opening a pathname and associating it with a descriptor or
3974 * reading it from /proc/pid/fd/ in cases where that doesn't make
3975 * sense.
3976 *
3977 * So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
3978 * not in use.
3979 */
3980 trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
3981
90df0249
ACM		 3982 err = trace__expand_filters(trace, &evsel);
3983 if (err)
3984 goto out_delete_evlist;
24bf91a7 3985 err = evlist__apply_filters(evlist, &evsel);
94ad89bc
ACM		 3986 if (err < 0)
3987 goto out_error_apply_filters;
241b057c 3988
ff7a4f98
ACM		 3989 if (trace->dump.map)
3990 bpf_map__fprintf(trace->dump.map, trace->output);
3991
9521b5f2 3992 err = evlist__mmap(evlist, trace->opts.mmap_pages);
e09b18d4
ACM		 3993 if (err < 0)
3994 goto out_error_mmap;
514f1c67 3995
e36b7821 3996 if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
1c87f165 3997 evlist__enable(evlist);
cb24d01d 3998
f15eb531 3999 if (forks)
7b392ef0 4000 evlist__start_workload(evlist);
f15eb531 4001
e36b7821
AB		 4002 if (trace->opts.initial_delay) {
4003 usleep(trace->opts.initial_delay * 1000);
1c87f165 4004 evlist__enable(evlist);
e36b7821
AB		 4005 }
4006
a2f354e3 4007 trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
fd3f518f
IR		 4008 perf_thread_map__nr(evlist->core.threads) > 1 ||
4009 evlist__first(evlist)->core.attr.inherit;
bd3dda9a
ACM		 4010
4011 /*
1fc632ce
JO		 4012 * Now that we already used evsel->core.attr to ask the kernel to setup the
4013 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
bd3dda9a 4014 * trace__resolve_callchain(), allowing per-event max-stack settings
adba1634 4015 * to override an explicitly set --max-stack global setting.
bd3dda9a
ACM		 4016 */
4017 evlist__for_each_entry(evlist, evsel) {
27de9b2b 4018 if (evsel__has_callchain(evsel) &&
1fc632ce
JO		 4019 evsel->core.attr.sample_max_stack == 0)
4020 evsel->core.attr.sample_max_stack = trace->max_stack;
bd3dda9a 4021 }
514f1c67 4022again:
efd5745e 4023 before = trace->nr_events;
514f1c67 4024
c976ee11 4025 for (i = 0; i < evlist->core.nr_mmaps; i++) {
514f1c67 4026 union perf_event *event;
a5830532 4027 struct mmap *md;
514f1c67 4028
d7f55c62 4029 md = &evlist->mmap[i];
7c4d4182 4030 if (perf_mmap__read_init(&md->core) < 0)
d7f55c62
KL		 4031 continue;
4032
151ed5d7 4033 while ((event = perf_mmap__read_event(&md->core)) != NULL) {
efd5745e 4034 ++trace->nr_events;
514f1c67 4035
7ba61524 4036 err = trace__deliver_event(trace, event);
028713aa
JO		 4037 if (err)
4038 goto out_disable;
514f1c67 4039
7728fa0c 4040 perf_mmap__consume(&md->core);
20c5f10e 4041
ba209f85
ACM		 4042 if (interrupted)
4043 goto out_disable;
02ac5421
ACM		 4044
4045 if (done && !draining) {
e74676de 4046 evlist__disable(evlist);
02ac5421
ACM		 4047 draining = true;
4048 }
514f1c67 4049 }
32fdc2ca 4050 perf_mmap__read_done(&md->core);
514f1c67
ACM		 4051 }
4052
efd5745e 4053 if (trace->nr_events == before) {
ba209f85 4054 int timeout = done ? 100 : -1;
f15eb531 4055
80ab2987 4056 if (!draining && evlist__poll(evlist, timeout) > 0) {
f4009e7b 4057 if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
46fb3c21
ACM		 4058 draining = true;
4059
ba209f85 4060 goto again;
028713aa 4061 } else {
7ba61524 4062 if (trace__flush_events(trace))
028713aa 4063 goto out_disable;
46fb3c21 4064 }
ba209f85
ACM		 4065 } else {
4066 goto again;
f15eb531
NK		 4067 }
4068
ba209f85 4069out_disable:
f3b623b8
ACM		 4070 thread__zput(trace->current);
4071
e74676de 4072 evlist__disable(evlist);
514f1c67 4073
a761a8d1
ACM		 4074 if (trace->sort_events)
4075 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
028713aa 4076
c522739d
ACM		 4077 if (!err) {
4078 if (trace->summary)
4079 trace__fprintf_thread_summary(trace, trace->output);
4080
4081 if (trace->show_tool_stats) {
4082 fprintf(trace->output, "Stats:\n "
4083 " vfs_getname : %" PRIu64 "\n"
4084 " proc_getname: %" PRIu64 "\n",
4085 trace->stats.vfs_getname,
4086 trace->stats.proc_getname);
4087 }
4088 }
bf2575c1 4089
514f1c67 4090out_delete_evlist:
33974a41 4091 trace__symbols__exit(trace);
3cb4d5e0 4092 evlist__free_syscall_tp_fields(evlist);
c12995a5 4093 evlist__delete(evlist);
9ea42ba4 4094 cgroup__put(trace->cgroup);
14a052df 4095 trace->evlist = NULL;
75b757ca 4096 trace->live = false;
514f1c67 4097 return err;
6ef068cb
ACM		 4098{
4099 char errbuf[BUFSIZ];
a8f23d8f 4100
2cc990ba 4101out_error_sched_stat_runtime:
988bdb31 4102 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
2cc990ba
ACM		 4103 goto out_error;
4104
801c67b0 4105out_error_raw_syscalls:
988bdb31 4106 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
a8f23d8f
ACM		 4107 goto out_error;
4108
e09b18d4 4109out_error_mmap:
d1f249ec 4110 evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
e09b18d4
ACM		 4111 goto out_error;
4112
a8f23d8f 4113out_error_open:
d1f249ec 4114 evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
a8f23d8f
ACM		 4115
4116out_error:
6ef068cb 4117 fprintf(trace->output, "%s\n", errbuf);
87f91868 4118 goto out_delete_evlist;
94ad89bc
ACM		 4119
4120out_error_apply_filters:
4121 fprintf(trace->output,
4122 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
8ab2e96d 4123 evsel->filter, evsel__name(evsel), errno,
c8b5f2c9 4124 str_error_r(errno, errbuf, sizeof(errbuf)));
94ad89bc 4125 goto out_delete_evlist;
514f1c67 4126}
5ed08dae
ACM		 4127out_error_mem:
4128 fprintf(trace->output, "Not enough memory to run!\n");
4129 goto out_delete_evlist;
19867b61
ACM		 4130
4131out_errno:
4132 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
4133 goto out_delete_evlist;
a8f23d8f 4134}
514f1c67 4135
6810fc91
DA		 4136static int trace__replay(struct trace *trace)
4137{
32dcd021 4138 const struct evsel_str_handler handlers[] = {
c522739d 4139 { "probe:vfs_getname", trace__vfs_getname, },
6810fc91 4140 };
8ceb41d7 4141 struct perf_data data = {
2d4f2799
JO		 4142 .path = input_name,
4143 .mode = PERF_DATA_MODE_READ,
4144 .force = trace->force,
f5fc1412 4145 };
6810fc91 4146 struct perf_session *session;
32dcd021 4147 struct evsel *evsel;
6810fc91
DA		 4148 int err = -1;
4149
4150 trace->tool.sample = trace__process_sample;
4151 trace->tool.mmap = perf_event__process_mmap;
384c671e 4152 trace->tool.mmap2 = perf_event__process_mmap2;
6810fc91
DA		 4153 trace->tool.comm = perf_event__process_comm;
4154 trace->tool.exit = perf_event__process_exit;
4155 trace->tool.fork = perf_event__process_fork;
4156 trace->tool.attr = perf_event__process_attr;
f3b3614a 4157 trace->tool.tracing_data = perf_event__process_tracing_data;
6810fc91 4158 trace->tool.build_id = perf_event__process_build_id;
f3b3614a 4159 trace->tool.namespaces = perf_event__process_namespaces;
6810fc91 4160
0a8cb85c 4161 trace->tool.ordered_events = true;
6810fc91
DA		 4162 trace->tool.ordering_requires_timestamps = true;
4163
4164 /* add tid to output */
4165 trace->multiple_threads = true;
4166
2681bd85 4167 session = perf_session__new(&data, &trace->tool);
6ef81c55
MI		 4168 if (IS_ERR(session))
4169 return PTR_ERR(session);
6810fc91 4170
aa07df6e
DA		 4171 if (trace->opts.target.pid)
4172 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
4173
4174 if (trace->opts.target.tid)
4175 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
4176
0a7e6d1b 4177 if (symbol__init(&session->header.env) < 0)
cb2ffae2
NK		 4178 goto out;
4179
8fb598e5
DA		 4180 trace->host = &session->machines.host;
4181
6810fc91
DA		 4182 err = perf_session__set_tracepoints_handlers(session, handlers);
4183 if (err)
4184 goto out;
4185
b02736f7 4186 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter");
4b335e1e 4187 trace->syscalls.events.sys_enter = evsel;
9aca7f17
DA		 4188 /* older kernels have syscalls tp versus raw_syscalls */
4189 if (evsel == NULL)
b02736f7 4190 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter");
003824e8 4191
e281a960 4192 if (evsel &&
794bca26 4193 (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
e281a960 4194 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
003824e8
NK		 4195 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
4196 goto out;
4197 }
4198
b02736f7 4199 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit");
4b335e1e 4200 trace->syscalls.events.sys_exit = evsel;
9aca7f17 4201 if (evsel == NULL)
b02736f7 4202 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit");
e281a960 4203 if (evsel &&
794bca26 4204 (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
e281a960 4205 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
003824e8 4206 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
6810fc91
DA		 4207 goto out;
4208 }
4209
e5cadb93 4210 evlist__for_each_entry(session->evlist, evsel) {
1fc632ce
JO		 4211 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
4212 (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
4213 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
4214 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
1e28fe0a
SF		 4215 evsel->handler = trace__pgfault;
4216 }
4217
6810fc91
DA		 4218 setup_pager();
4219
b7b61cbe 4220 err = perf_session__process_events(session);
6810fc91
DA		 4221 if (err)
4222 pr_err("Failed to process events, error %d", err);
4223
bf2575c1
DA		 4224 else if (trace->summary)
4225 trace__fprintf_thread_summary(trace, trace->output);
4226
6810fc91
DA		 4227out:
4228 perf_session__delete(session);
4229
4230 return err;
4231}
4232
1302d88e
ACM		 4233static size_t trace__fprintf_threads_header(FILE *fp)
4234{
4235 size_t printed;
4236
99ff7150 4237 printed = fprintf(fp, "\n Summary of events:\n\n");
bf2575c1
DA		 4238
4239 return printed;
4240}
4241
b535d523 4242DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
8eded45f
ACM		 4243 struct syscall_stats *stats;
4244 double msecs;
4245 int syscall;
b535d523
ACM		 4246)
4247{
4248 struct int_node *source = rb_entry(nd, struct int_node, rb_node);
8eded45f 4249 struct syscall_stats *stats = source->priv;
b535d523
ACM		 4250
4251 entry->syscall = source->i;
4252 entry->stats = stats;
8eded45f 4253 entry->msecs = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0;
b535d523
ACM		 4254}
4255
bf2575c1
DA		 4256static size_t thread__dump_stats(struct thread_trace *ttrace,
4257 struct trace *trace, FILE *fp)
4258{
bf2575c1
DA		 4259 size_t printed = 0;
4260 struct syscall *sc;
b535d523
ACM		 4261 struct rb_node *nd;
4262 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
bf2575c1 4263
b535d523 4264 if (syscall_stats == NULL)
bf2575c1
DA		 4265 return 0;
4266
4267 printed += fprintf(fp, "\n");
4268
8eded45f
ACM		 4269 printed += fprintf(fp, " syscall calls errors total min avg max stddev\n");
4270 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
4271 printed += fprintf(fp, " --------------- -------- ------ -------- --------- --------- --------- ------\n");
99ff7150 4272
98a91837 4273 resort_rb__for_each_entry(nd, syscall_stats) {
8eded45f 4274 struct syscall_stats *stats = syscall_stats_entry->stats;
bf2575c1 4275 if (stats) {
8eded45f
ACM		 4276 double min = (double)(stats->stats.min) / NSEC_PER_MSEC;
4277 double max = (double)(stats->stats.max) / NSEC_PER_MSEC;
4278 double avg = avg_stats(&stats->stats);
bf2575c1 4279 double pct;
8eded45f 4280 u64 n = (u64)stats->stats.n;
bf2575c1 4281
8eded45f 4282 pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0;
bf2575c1
DA		 4283 avg /= NSEC_PER_MSEC;
4284
b535d523 4285 sc = &trace->syscalls.table[syscall_stats_entry->syscall];
99ff7150 4286 printed += fprintf(fp, " %-15s", sc->name);
8eded45f
ACM		 4287 printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f",
4288 n, stats->nr_failures, syscall_stats_entry->msecs, min, avg);
27a778b5 4289 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
b88b14db
ACM		 4290
4291 if (trace->errno_summary && stats->nr_failures) {
4292 const char *arch_name = perf_env__arch(trace->host->env);
4293 int e;
4294
4295 for (e = 0; e < stats->max_errno; ++e) {
4296 if (stats->errnos[e] != 0)
4297 fprintf(fp, "\t\t\t\t%s: %d\n", arch_syscalls__strerrno(arch_name, e + 1), stats->errnos[e]);
4298 }
4299 }
bf2575c1 4300 }
bf2575c1
DA		 4301 }
4302
b535d523 4303 resort_rb__delete(syscall_stats);
bf2575c1 4304 printed += fprintf(fp, "\n\n");
1302d88e
ACM		 4305
4306 return printed;
4307}
4308
96c14451 4309static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
896cbb56 4310{
96c14451 4311 size_t printed = 0;
89dceb22 4312 struct thread_trace *ttrace = thread__priv(thread);
896cbb56
DA		 4313 double ratio;
4314
4315 if (ttrace == NULL)
4316 return 0;
4317
4318 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
4319
15e65c69 4320 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
99ff7150 4321 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
15e65c69 4322 printed += fprintf(fp, "%.1f%%", ratio);
a2ea67d7
SF		 4323 if (ttrace->pfmaj)
4324 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
4325 if (ttrace->pfmin)
4326 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
03548ebf
ACM		 4327 if (trace->sched)
4328 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
4329 else if (fputc('\n', fp) != EOF)
4330 ++printed;
4331
bf2575c1 4332 printed += thread__dump_stats(ttrace, trace, fp);
896cbb56 4333
96c14451
ACM		 4334 return printed;
4335}
896cbb56 4336
96c14451
ACM		 4337static unsigned long thread__nr_events(struct thread_trace *ttrace)
4338{
4339 return ttrace ? ttrace->nr_events : 0;
4340}
4341
4342DEFINE_RESORT_RB(threads, (thread__nr_events(a->thread->priv) < thread__nr_events(b->thread->priv)),
4343 struct thread *thread;
4344)
4345{
4346 entry->thread = rb_entry(nd, struct thread, rb_node);
896cbb56
DA		 4347}
4348
1302d88e
ACM		 4349static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
4350{
96c14451
ACM		 4351 size_t printed = trace__fprintf_threads_header(fp);
4352 struct rb_node *nd;
91e467bc 4353 int i;
1302d88e 4354
91e467bc
KL		 4355 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
4356 DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
96c14451 4357
91e467bc
KL		 4358 if (threads == NULL) {
4359 fprintf(fp, "%s", "Error sorting output by nr_events!\n");
4360 return 0;
4361 }
896cbb56 4362
91e467bc
KL		 4363 resort_rb__for_each_entry(nd, threads)
4364 printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
96c14451 4365
91e467bc
KL		 4366 resort_rb__delete(threads);
4367 }
96c14451 4368 return printed;
1302d88e
ACM		 4369}
4370
ae9ed035
ACM		 4371static int trace__set_duration(const struct option *opt, const char *str,
4372 int unset __maybe_unused)
4373{
4374 struct trace *trace = opt->value;
4375
4376 trace->duration_filter = atof(str);
4377 return 0;
4378}
4379
6a0b3aba
ACM		 4380static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
4381 int unset __maybe_unused)
f078c385
ACM		 4382{
4383 int ret = -1;
4384 size_t i;
4385 struct trace *trace = opt->value;
4386 /*
4387 * FIXME: introduce a intarray class, plain parse csv and create a
4388 * { int nr, int entries[] } struct...
4389 */
4390 struct intlist *list = intlist__new(str);
4391
4392 if (list == NULL)
4393 return -1;
4394
4395 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
4396 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
4397
4398 if (trace->filter_pids.entries == NULL)
4399 goto out;
4400
4401 trace->filter_pids.entries[0] = getpid();
4402
4403 for (i = 1; i < trace->filter_pids.nr; ++i)
4404 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
4405
4406 intlist__delete(list);
4407 ret = 0;
4408out:
4409 return ret;
4410}
4411
c24ff998
ACM		 4412static int trace__open_output(struct trace *trace, const char *filename)
4413{
4414 struct stat st;
4415
4416 if (!stat(filename, &st) && st.st_size) {
4417 char oldname[PATH_MAX];
4418
4419 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
4420 unlink(oldname);
4421 rename(filename, oldname);
4422 }
4423
4424 trace->output = fopen(filename, "w");
4425
4426 return trace->output == NULL ? -errno : 0;
4427}
4428
598d02c5
SF		 4429static int parse_pagefaults(const struct option *opt, const char *str,
4430 int unset __maybe_unused)
4431{
4432 int *trace_pgfaults = opt->value;
4433
4434 if (strcmp(str, "all") == 0)
4435 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
4436 else if (strcmp(str, "maj") == 0)
4437 *trace_pgfaults |= TRACE_PFMAJ;
4438 else if (strcmp(str, "min") == 0)
4439 *trace_pgfaults |= TRACE_PFMIN;
4440 else
4441 return -1;
4442
4443 return 0;
4444}
4445
206d635a 4446static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)
14a052df 4447{
32dcd021 4448 struct evsel *evsel;
14a052df 4449
206d635a
ACM		 4450 evlist__for_each_entry(evlist, evsel) {
4451 if (evsel->handler == NULL)
4452 evsel->handler = handler;
4453 }
14a052df
ACM		 4454}
4455
362222f8
ACM		 4456static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name)
4457{
4458 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
4459
4460 if (fmt) {
4461 struct syscall_fmt *scfmt = syscall_fmt__find(name);
4462
4463 if (scfmt) {
4464 int skip = 0;
4465
4466 if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 ||
4467 strcmp(evsel->tp_format->format.fields->name, "nr") == 0)
4468 ++skip;
4469
4470 memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt));
4471 }
4472 }
4473}
4474
63503dba 4475static int evlist__set_syscall_tp_fields(struct evlist *evlist)
d32855fa 4476{
32dcd021 4477 struct evsel *evsel;
d32855fa
ACM		 4478
4479 evlist__for_each_entry(evlist, evsel) {
4480 if (evsel->priv || !evsel->tp_format)
4481 continue;
4482
947b843c 4483 if (strcmp(evsel->tp_format->system, "syscalls")) {
794bca26 4484 evsel__init_tp_arg_scnprintf(evsel);
d32855fa 4485 continue;
947b843c 4486 }
d32855fa 4487
794bca26 4488 if (evsel__init_syscall_tp(evsel))
d32855fa
ACM		 4489 return -1;
4490
4491 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
fecd9907 4492 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
d32855fa
ACM		 4493
4494 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
4495 return -1;
362222f8
ACM		 4496
4497 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1);
d32855fa 4498 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
fecd9907 4499 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
d32855fa
ACM		 4500
4501 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
4502 return -1;
362222f8
ACM		 4503
4504 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1);
d32855fa
ACM		 4505 }
4506 }
4507
4508 return 0;
4509}
4510
017037ff
ACM		 4511/*
4512 * XXX: Hackish, just splitting the combined -e+--event (syscalls
4513 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
4514 * existing facilities unchanged (trace->ev_qualifier + parse_options()).
4515 *
4516 * It'd be better to introduce a parse_options() variant that would return a
4517 * list with the terms it didn't match to an event...
4518 */
4519static int trace__parse_events_option(const struct option *opt, const char *str,
4520 int unset __maybe_unused)
4521{
4522 struct trace *trace = (struct trace *)opt->value;
4523 const char *s = str;
4524 char *sep = NULL, *lists[2] = { NULL, NULL, };
27702bcf 4525 int len = strlen(str) + 1, err = -1, list, idx;
017037ff
ACM		 4526 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
4527 char group_name[PATH_MAX];
f932184e 4528 struct syscall_fmt *fmt;
017037ff
ACM		 4529
4530 if (strace_groups_dir == NULL)
4531 return -1;
4532
4533 if (*s == '!') {
4534 ++s;
4535 trace->not_ev_qualifier = true;
4536 }
4537
4538 while (1) {
4539 if ((sep = strchr(s, ',')) != NULL)
4540 *sep = '\0';
4541
4542 list = 0;
27702bcf
ACM		 4543 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
4544 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
017037ff 4545 list = 1;
f932184e
ACM		 4546 goto do_concat;
4547 }
4548
4549 fmt = syscall_fmt__find_by_alias(s);
4550 if (fmt != NULL) {
4551 list = 1;
4552 s = fmt->name;
017037ff
ACM		 4553 } else {
4554 path__join(group_name, sizeof(group_name), strace_groups_dir, s);
4555 if (access(group_name, R_OK) == 0)
4556 list = 1;
4557 }
f932184e 4558do_concat:
017037ff
ACM		 4559 if (lists[list]) {
4560 sprintf(lists[list] + strlen(lists[list]), ",%s", s);
4561 } else {
4562 lists[list] = malloc(len);
4563 if (lists[list] == NULL)
4564 goto out;
4565 strcpy(lists[list], s);
4566 }
4567
4568 if (!sep)
4569 break;
4570
4571 *sep = ',';
4572 s = sep + 1;
4573 }
4574
4575 if (lists[1] != NULL) {
4576 struct strlist_config slist_config = {
4577 .dirname = strace_groups_dir,
4578 };
4579
4580 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4581 if (trace->ev_qualifier == NULL) {
4582 fputs("Not enough memory to parse event qualifier", trace->output);
4583 goto out;
4584 }
4585
4586 if (trace__validate_ev_qualifier(trace))
4587 goto out;
b912885a 4588 trace->trace_syscalls = true;
017037ff
ACM		 4589 }
4590
4591 err = 0;
4592
4593 if (lists[0]) {
a6293f36
SI		 4594 struct option o = {
4595 .value = &trace->evlist,
4596 };
017037ff
ACM		 4597 err = parse_events_option(&o, lists[0], 0);
4598 }
4599out:
659ede7d
RM		 4600 free(strace_groups_dir);
4601 free(lists[0]);
4602 free(lists[1]);
017037ff
ACM		 4603 if (sep)
4604 *sep = ',';
4605
4606 return err;
4607}
4608
9ea42ba4
ACM		 4609static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4610{
4611 struct trace *trace = opt->value;
4612
a6293f36
SI		 4613 if (!list_empty(&trace->evlist->core.entries)) {
4614 struct option o = {
4615 .value = &trace->evlist,
4616 };
4617 return parse_cgroups(&o, str, unset);
4618 }
9ea42ba4
ACM		 4619 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
4620
4621 return 0;
4622}
4623
ac96287c
ACM		 4624static int trace__config(const char *var, const char *value, void *arg)
4625{
e7c634fc 4626 struct trace *trace = arg;
ac96287c
ACM		 4627 int err = 0;
4628
4629 if (!strcmp(var, "trace.add_events")) {
7e035929
ACM		 4630 trace->perfconfig_events = strdup(value);
4631 if (trace->perfconfig_events == NULL) {
4632 pr_err("Not enough memory for %s\n", "trace.add_events");
4633 return -1;
4634 }
b036146f
ACM		 4635 } else if (!strcmp(var, "trace.show_timestamp")) {
4636 trace->show_tstamp = perf_config_bool(var, value);
42e4a52d
ACM		 4637 } else if (!strcmp(var, "trace.show_duration")) {
4638 trace->show_duration = perf_config_bool(var, value);
9d6dc178
ACM		 4639 } else if (!strcmp(var, "trace.show_arg_names")) {
4640 trace->show_arg_names = perf_config_bool(var, value);
4641 if (!trace->show_arg_names)
4642 trace->show_zeros = true;
e7c634fc 4643 } else if (!strcmp(var, "trace.show_zeros")) {
9d6dc178
ACM		 4644 bool new_show_zeros = perf_config_bool(var, value);
4645 if (!trace->show_arg_names && !new_show_zeros) {
4646 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4647 goto out;
4648 }
4649 trace->show_zeros = new_show_zeros;
c65c83ff
ACM		 4650 } else if (!strcmp(var, "trace.show_prefix")) {
4651 trace->show_string_prefix = perf_config_bool(var, value);
d32de87e
ACM		 4652 } else if (!strcmp(var, "trace.no_inherit")) {
4653 trace->opts.no_inherit = perf_config_bool(var, value);
9ed45d59
ACM		 4654 } else if (!strcmp(var, "trace.args_alignment")) {
4655 int args_alignment = 0;
4656 if (perf_config_int(&args_alignment, var, value) == 0)
4657 trace->args_alignment = args_alignment;
f11b2803
ACM		 4658 } else if (!strcmp(var, "trace.tracepoint_beautifiers")) {
4659 if (strcasecmp(value, "libtraceevent") == 0)
4660 trace->libtraceevent_print = true;
4661 else if (strcasecmp(value, "libbeauty") == 0)
4662 trace->libtraceevent_print = false;
ac96287c 4663 }
9d6dc178 4664out:
ac96287c
ACM		 4665 return err;
4666}
4667
6c7f0ab0
RM		 4668static void trace__exit(struct trace *trace)
4669{
f2ebf8ff
RM		 4670 int i;
4671
6c7f0ab0
RM		 4672 strlist__delete(trace->ev_qualifier);
4673 free(trace->ev_qualifier_ids.entries);
f2ebf8ff
RM		 4674 if (trace->syscalls.table) {
4675 for (i = 0; i <= trace->sctbl->syscalls.max_id; i++)
4676 syscall__exit(&trace->syscalls.table[i]);
4677 free(trace->syscalls.table);
4678 }
6c7f0ab0
RM		 4679 syscalltbl__delete(trace->sctbl);
4680 zfree(&trace->perfconfig_events);
4681}
4682
b0ad8ea6 4683int cmd_trace(int argc, const char **argv)
514f1c67 4684{
6fdd9cb7 4685 const char *trace_usage[] = {
f15eb531
NK		 4686 "perf trace [<options>] [<command>]",
4687 "perf trace [<options>] -- <command> [<options>]",
5e2485b1
DA		 4688 "perf trace record [<options>] [<command>]",
4689 "perf trace record [<options>] -- <command> [<options>]",
514f1c67
ACM		 4690 NULL
4691 };
4692 struct trace trace = {
514f1c67
ACM		 4693 .opts = {
4694 .target = {
4695 .uid = UINT_MAX,
4696 .uses_mmap = true,
4697 },
4698 .user_freq = UINT_MAX,
4699 .user_interval = ULLONG_MAX,
509051ea 4700 .no_buffering = true,
38d5447d 4701 .mmap_pages = UINT_MAX,
514f1c67 4702 },
007d66a0 4703 .output = stderr,
50c95cbd 4704 .show_comm = true,
b036146f 4705 .show_tstamp = true,
42e4a52d 4706 .show_duration = true,
9d6dc178 4707 .show_arg_names = true,
9ed45d59 4708 .args_alignment = 70,
b912885a 4709 .trace_syscalls = false,
44621819 4710 .kernel_syscallchains = false,
05614993 4711 .max_stack = UINT_MAX,
5067a8cd 4712 .max_events = ULONG_MAX,
514f1c67 4713 };
ff7a4f98 4714 const char *map_dump_str = NULL;
c24ff998 4715 const char *output_name = NULL;
514f1c67 4716 const struct option trace_options[] = {
017037ff
ACM		 4717 OPT_CALLBACK('e', "event", &trace, "event",
4718 "event/syscall selector. use 'perf list' to list available events",
4719 trace__parse_events_option),
d4097f19
ACM		 4720 OPT_CALLBACK(0, "filter", &trace.evlist, "filter",
4721 "event filter", parse_filter),
50c95cbd
ACM		 4722 OPT_BOOLEAN(0, "comm", &trace.show_comm,
4723 "show the thread COMM next to its id"),
c522739d 4724 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
017037ff
ACM		 4725 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4726 trace__parse_events_option),
c24ff998 4727 OPT_STRING('o', "output", &output_name, "file", "output file name"),
6810fc91 4728 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
514f1c67
ACM		 4729 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4730 "trace events on existing process id"),
ac9be8ee 4731 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
514f1c67 4732 "trace events on existing thread id"),
fa0e4ffe 4733 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
6a0b3aba 4734 "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
ac9be8ee 4735 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
514f1c67 4736 "system-wide collection from all CPUs"),
ac9be8ee 4737 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
514f1c67 4738 "list of cpus to monitor"),
6810fc91 4739 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
514f1c67 4740 "child tasks do not inherit counters"),
994a1f78 4741 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
25f84702 4742 "number of mmap data pages", evlist__parse_mmap_pages),
ac9be8ee 4743 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
514f1c67 4744 "user to profile"),
ae9ed035
ACM		 4745 OPT_CALLBACK(0, "duration", &trace, "float",
4746 "show only events with duration > N.M ms",
4747 trace__set_duration),
ff7a4f98
ACM		 4748#ifdef HAVE_LIBBPF_SUPPORT
4749 OPT_STRING(0, "map-dump", &map_dump_str, "BPF map", "BPF map to periodically dump"),
4750#endif
1302d88e 4751 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
7c304ee0 4752 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4bb09192
DA		 4753 OPT_BOOLEAN('T', "time", &trace.full_time,
4754 "Show full timestamp, not time relative to first start"),
0a6545bd
ACM		 4755 OPT_BOOLEAN(0, "failure", &trace.failure_only,
4756 "Show only syscalls that failed"),
fd2eabaf
DA		 4757 OPT_BOOLEAN('s', "summary", &trace.summary_only,
4758 "Show only syscall summary with statistics"),
4759 OPT_BOOLEAN('S', "with-summary", &trace.summary,
4760 "Show all syscalls and summary with statistics"),
b88b14db
ACM		 4761 OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary,
4762 "Show errno stats per syscall, use with -s or -S"),
598d02c5
SF		 4763 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4764 "Trace pagefaults", parse_pagefaults, "maj"),
e281a960 4765 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
e366a6d8 4766 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
566a0885
MW		 4767 OPT_CALLBACK(0, "call-graph", &trace.opts,
4768 "record_mode[,record_size]", record_callchain_help,
4769 &record_parse_callchain_opt),
f11b2803
ACM		 4770 OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,
4771 "Use libtraceevent to print the tracepoint arguments."),
44621819
ACM		 4772 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4773 "Show the kernel callchains on the syscall exit path"),
5067a8cd
ACM		 4774 OPT_ULONG(0, "max-events", &trace.max_events,
4775 "Set the maximum number of events to print, exit after that is reached. "),
5cf9c84e
ACM		 4776 OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4777 "Set the minimum stack depth when parsing the callchain, "
4778 "anything below the specified depth will be ignored."),
c6d4a494
ACM		 4779 OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4780 "Set the maximum stack depth when parsing the callchain, "
4781 "anything beyond the specified depth will be ignored. "
4cb93446 4782 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
a761a8d1
ACM		 4783 OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4784 "Sort batch of events before processing, use if getting out of order events"),
591421e1
ACM		 4785 OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4786 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
3fcb10e4 4787 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
9d9cad76 4788 "per thread proc mmap processing timeout in ms"),
9ea42ba4
ACM		 4789 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4790 trace__parse_cgroups),
68cd3b45 4791 OPT_INTEGER('D', "delay", &trace.opts.initial_delay,
e36b7821
AB		 4792 "ms to wait before starting measurement after program "
4793 "start"),
22ac4318 4794 OPTS_EVSWITCH(&trace.evswitch),
514f1c67
ACM		 4795 OPT_END()
4796 };
ccd62a89 4797 bool __maybe_unused max_stack_user_set = true;
f3e459d1 4798 bool mmap_pages_user_set = true;
32dcd021 4799 struct evsel *evsel;
6fdd9cb7 4800 const char * const trace_subcommands[] = { "record", NULL };
78e890ea 4801 int err = -1;
32caf0d1 4802 char bf[BUFSIZ];
de9f498d 4803 struct sigaction sigchld_act;
514f1c67 4804
4d08cb80
ACM		 4805 signal(SIGSEGV, sighandler_dump_stack);
4806 signal(SIGFPE, sighandler_dump_stack);
de9f498d
CD		 4807 signal(SIGINT, sighandler_interrupt);
4808
4809 memset(&sigchld_act, 0, sizeof(sigchld_act));
4810 sigchld_act.sa_flags = SA_SIGINFO;
4811 sigchld_act.sa_sigaction = sighandler_chld;
4812 sigaction(SIGCHLD, &sigchld_act, NULL);
4d08cb80 4813
0f98b11c 4814 trace.evlist = evlist__new();
fd0db102 4815 trace.sctbl = syscalltbl__new();
14a052df 4816
fd0db102 4817 if (trace.evlist == NULL || trace.sctbl == NULL) {
14a052df 4818 pr_err("Not enough memory to run!\n");
ff8f695c 4819 err = -ENOMEM;
14a052df
ACM		 4820 goto out;
4821 }
4822
c3e78a34
ACM		 4823 /*
4824 * Parsing .perfconfig may entail creating a BPF event, that may need
4825 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4826 * is too small. This affects just this process, not touching the
4827 * global setting. If it fails we'll get something in 'perf trace -v'
4828 * to help diagnose the problem.
4829 */
4830 rlimit__bump_memlock();
4831
ac96287c
ACM		 4832 err = perf_config(trace__config, &trace);
4833 if (err)
4834 goto out;
4835
6fdd9cb7
YS		 4836 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4837 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
8bd436b0
ACM		 4838
4839 /*
4840 * Here we already passed thru trace__parse_events_option() and it has
4841 * already figured out if -e syscall_name, if not but if --event
4842 * foo:bar was used, the user is interested _just_ in those, say,
4843 * tracepoint events, not in the strace-like syscall-name-based mode.
4844 *
4845 * This is important because we need to check if strace-like mode is
4846 * needed to decided if we should filter out the eBPF
4847 * __augmented_syscalls__ code, if it is in the mix, say, via
4848 * .perfconfig trace.add_events, and filter those out.
4849 */
4850 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4851 trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4852 trace.trace_syscalls = true;
4853 }
7e035929
ACM		 4854 /*
4855 * Now that we have --verbose figured out, lets see if we need to parse
4856 * events from .perfconfig, so that if those events fail parsing, say some
4857 * BPF program fails, then we'll be able to use --verbose to see what went
4858 * wrong in more detail.
4859 */
4860 if (trace.perfconfig_events != NULL) {
a910e466 4861 struct parse_events_error parse_err;
7e035929 4862
07eafd4e 4863 parse_events_error__init(&parse_err);
7e035929 4864 err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);
07eafd4e 4865 if (err)
6c191289 4866 parse_events_error__print(&parse_err, trace.perfconfig_events);
07eafd4e
IR		 4867 parse_events_error__exit(&parse_err);
4868 if (err)
7e035929 4869 goto out;
7e035929 4870 }
fd2eabaf 4871
9ea42ba4
ACM		 4872 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4873 usage_with_options_msg(trace_usage, trace_options,
4874 "cgroup monitoring only available in system-wide mode");
4875 }
4876
78e890ea
ACM		 4877 evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
4878 if (IS_ERR(evsel)) {
4879 bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
e0b6d2ef
ACM		 4880 pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
4881 goto out;
4882 }
4883
744fafc7 4884 if (evsel) {
d3d1c4bd 4885 trace.syscalls.events.augmented = evsel;
c8c80570 4886
b02736f7 4887 evsel = evlist__find_tracepoint_by_name(trace.evlist, "raw_syscalls:sys_enter");
c8c80570
ACM		 4888 if (evsel == NULL) {
4889 pr_err("ERROR: raw_syscalls:sys_enter not found in the augmented BPF object\n");
4890 goto out;
4891 }
4892
4893 if (evsel->bpf_obj == NULL) {
4894 pr_err("ERROR: raw_syscalls:sys_enter not associated to a BPF object\n");
4895 goto out;
4896 }
4897
4898 trace.bpf_obj = evsel->bpf_obj;
4899
8bd436b0
ACM		 4900 /*
4901 * If we have _just_ the augmenter event but don't have a
4902 * explicit --syscalls, then assume we want all strace-like
4903 * syscalls:
4904 */
4905 if (!trace.trace_syscalls && trace__only_augmented_syscalls_evsels(&trace))
4906 trace.trace_syscalls = true;
4907 /*
4908 * So, if we have a syscall augmenter, but trace_syscalls, aka
4909 * strace-like syscall tracing is not set, then we need to trow
4910 * away the augmenter, i.e. all the events that were created
4911 * from that BPF object file.
4912 *
4913 * This is more to fix the current .perfconfig trace.add_events
4914 * style of setting up the strace-like eBPF based syscall point
4915 * payload augmenter.
4916 *
4917 * All this complexity will be avoided by adding an alternative
4918 * to trace.add_events in the form of
4919 * trace.bpf_augmented_syscalls, that will be only parsed if we
4920 * need it.
4921 *
4922 * .perfconfig trace.add_events is still useful if we want, for
4923 * instance, have msr_write.msr in some .perfconfig profile based
4924 * 'perf trace --config determinism.profile' mode, where for some
4925 * particular goal/workload type we want a set of events and
4926 * output mode (with timings, etc) instead of having to add
4927 * all via the command line.
4928 *
4929 * Also --config to specify an alternate .perfconfig file needs
4930 * to be implemented.
4931 */
4932 if (!trace.trace_syscalls) {
4933 trace__delete_augmented_syscalls(&trace);
4934 } else {
4935 trace__set_bpf_map_filtered_pids(&trace);
4936 trace__set_bpf_map_syscalls(&trace);
4937 trace.syscalls.unaugmented_prog = trace__find_bpf_program_by_title(&trace, "!raw_syscalls:unaugmented");
4938 }
744fafc7 4939 }
d3d1c4bd 4940
d7888573
WN		 4941 err = bpf__setup_stdout(trace.evlist);
4942 if (err) {
4943 bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
4944 pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
4945 goto out;
4946 }
4947
59247e33
ACM		 4948 err = -1;
4949
ff7a4f98 4950 if (map_dump_str) {
5ca0b7f5 4951 trace.dump.map = trace__find_bpf_map_by_name(&trace, map_dump_str);
ff7a4f98
ACM		 4952 if (trace.dump.map == NULL) {
4953 pr_err("ERROR: BPF map \"%s\" not found\n", map_dump_str);
4954 goto out;
4955 }
4956 }
4957
598d02c5
SF		 4958 if (trace.trace_pgfaults) {
4959 trace.opts.sample_address = true;
4960 trace.opts.sample_time = true;
4961 }
4962
f3e459d1
ACM		 4963 if (trace.opts.mmap_pages == UINT_MAX)
4964 mmap_pages_user_set = false;
4965
05614993 4966 if (trace.max_stack == UINT_MAX) {
029c75e5 4967 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
05614993
ACM		 4968 max_stack_user_set = false;
4969 }
4970
4971#ifdef HAVE_DWARF_UNWIND_SUPPORT
75d50117 4972 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
05614993 4973 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
75d50117 4974 }
05614993
ACM		 4975#endif
4976
2ddd5c04 4977 if (callchain_param.enabled) {
f3e459d1
ACM		 4978 if (!mmap_pages_user_set && geteuid() == 0)
4979 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
4980
566a0885 4981 symbol_conf.use_callchain = true;
f3e459d1 4982 }
566a0885 4983
6484d2f9 4984 if (trace.evlist->core.nr_entries > 0) {
206d635a 4985 evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);
d32855fa
ACM		 4986 if (evlist__set_syscall_tp_fields(trace.evlist)) {
4987 perror("failed to set syscalls:* tracepoint fields");
4988 goto out;
4989 }
4990 }
14a052df 4991
a761a8d1
ACM		 4992 if (trace.sort_events) {
4993 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
4994 ordered_events__set_copy_on_queue(&trace.oe.data, true);
4995 }
028713aa 4996
b1a9e253
ACM		 4997 /*
4998 * If we are augmenting syscalls, then combine what we put in the
4999 * __augmented_syscalls__ BPF map with what is in the
5000 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
5001 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
5002 *
5003 * We'll switch to look at two BPF maps, one for sys_enter and the
5004 * other for sys_exit when we start augmenting the sys_exit paths with
5005 * buffers that are being copied from kernel to userspace, think 'read'
5006 * syscall.
5007 */
5008 if (trace.syscalls.events.augmented) {
b1a9e253 5009 evlist__for_each_entry(trace.evlist, evsel) {
8ab2e96d 5010 bool raw_syscalls_sys_exit = strcmp(evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
cd26ea6d
ACM		 5011
5012 if (raw_syscalls_sys_exit) {
5013 trace.raw_augmented_syscalls = true;
5014 goto init_augmented_syscall_tp;
5015 }
5016
1d59cb1b 5017 if (trace.syscalls.events.augmented->priv == NULL &&
8ab2e96d 5018 strstr(evsel__name(evsel), "syscalls:sys_enter")) {
32dcd021 5019 struct evsel *augmented = trace.syscalls.events.augmented;
794bca26
ACM		 5020 if (evsel__init_augmented_syscall_tp(augmented, evsel) ||
5021 evsel__init_augmented_syscall_tp_args(augmented))
b9b6a2ea 5022 goto out;
b119970a
ACM		 5023 /*
5024 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
5025 * Above we made sure we can get from the payload the tp fields
5026 * that we get from syscalls:sys_enter tracefs format file.
5027 */
b9b6a2ea 5028 augmented->handler = trace__sys_enter;
b119970a
ACM		 5029 /*
5030 * Now we do the same for the *syscalls:sys_enter event so that
5031 * if we handle it directly, i.e. if the BPF prog returns 0 so
5032 * as not to filter it, then we'll handle it just like we would
5033 * for the BPF_OUTPUT one:
5034 */
794bca26
ACM		 5035 if (evsel__init_augmented_syscall_tp(evsel, evsel) ||
5036 evsel__init_augmented_syscall_tp_args(evsel))
b119970a
ACM		 5037 goto out;
5038 evsel->handler = trace__sys_enter;
b9b6a2ea
ACM		 5039 }
5040
8ab2e96d 5041 if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) {
b9b6a2ea 5042 struct syscall_tp *sc;
cd26ea6d 5043init_augmented_syscall_tp:
794bca26 5044 if (evsel__init_augmented_syscall_tp(evsel, evsel))
b9b6a2ea 5045 goto out;
fecd9907 5046 sc = __evsel__syscall_tp(evsel);
b9b6a2ea
ACM		 5047 /*
5048 * For now with BPF raw_augmented we hook into
5049 * raw_syscalls:sys_enter and there we get all
5050 * 6 syscall args plus the tracepoint common
5051 * fields and the syscall_nr (another long).
5052 * So we check if that is the case and if so
5053 * don't look after the sc->args_size but
5054 * always after the full raw_syscalls:sys_enter
5055 * payload, which is fixed.
5056 *
5057 * We'll revisit this later to pass
5058 * s->args_size to the BPF augmenter (now
5059 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
5060 * so that it copies only what we need for each
5061 * syscall, like what happens when we use
5062 * syscalls:sys_enter_NAME, so that we reduce
5063 * the kernel/userspace traffic to just what is
5064 * needed for each syscall.
5065 */
5066 if (trace.raw_augmented_syscalls)
5067 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
794bca26 5068 evsel__init_augmented_syscall_tp_ret(evsel);
b1a9e253
ACM		 5069 evsel->handler = trace__sys_exit;
5070 }
5071 }
5072 }
5073
1e28fe0a
SF		 5074 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
5075 return trace__record(&trace, argc-1, &argv[1]);
5076
b88b14db
ACM		 5077 /* Using just --errno-summary will trigger --summary */
5078 if (trace.errno_summary && !trace.summary && !trace.summary_only)
5079 trace.summary_only = true;
5080
1e28fe0a
SF		 5081 /* summary_only implies summary option, but don't overwrite summary if set */
5082 if (trace.summary_only)
5083 trace.summary = trace.summary_only;
5084
c24ff998
ACM		 5085 if (output_name != NULL) {
5086 err = trace__open_output(&trace, output_name);
5087 if (err < 0) {
5088 perror("failed to create output file");
5089 goto out;
5090 }
5091 }
5092
22ac4318
ACM		 5093 err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
5094 if (err)
5095 goto out_close;
5096
602ad878 5097 err = target__validate(&trace.opts.target);
32caf0d1 5098 if (err) {
602ad878 5099 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
c24ff998
ACM		 5100 fprintf(trace.output, "%s", bf);
5101 goto out_close;
32caf0d1
NK		 5102 }
5103
602ad878 5104 err = target__parse_uid(&trace.opts.target);
514f1c67 5105 if (err) {
602ad878 5106 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
c24ff998
ACM		 5107 fprintf(trace.output, "%s", bf);
5108 goto out_close;
514f1c67
ACM		 5109 }
5110
602ad878 5111 if (!argc && target__none(&trace.opts.target))
ee76120e
NK		 5112 trace.opts.target.system_wide = true;
5113
6810fc91
DA		 5114 if (input_name)
5115 err = trace__replay(&trace);
5116 else
5117 err = trace__run(&trace, argc, argv);
1302d88e 5118
c24ff998
ACM		 5119out_close:
5120 if (output_name != NULL)
5121 fclose(trace.output);
5122out:
6c7f0ab0 5123 trace__exit(&trace);
1302d88e 5124 return err;
514f1c67 5125}
Linux 6.x block layer and io_uring tree(s)