projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'zonefs-6.5-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/dlemoal...
[linux-2.6-block.git] / tools / perf / builtin-trace.c
CommitLineData
a598bb5e
ACM		 1/*
2 * builtin-trace.c
3 *
4 * Builtin 'trace' command:
5 *
6 * Display a continuously updated trace of any workload, CPU, specific PID,
7 * system wide, etc. Default format is loosely strace like, but any other
8 * event may be specified using --event.
9 *
10 * Copyright (C) 2012, 2013, 2014, 2015 Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
11 *
12 * Initially based on the 'trace' prototype by Thomas Gleixner:
13 *
14 * http://lwn.net/Articles/415728/ ("Announcing a new utility: 'trace'")
a598bb5e
ACM		 15 */
16
aeb00b1a 17#include "util/record.h"
988bdb31 18#include <api/fs/tracing_path.h>
d891f2b7 19#ifdef HAVE_LIBBPF_SUPPORT
744fafc7 20#include <bpf/bpf.h>
d891f2b7 21#endif
ff7a4f98 22#include "util/bpf_map.h"
c3e78a34 23#include "util/rlimit.h"
514f1c67 24#include "builtin.h"
9ea42ba4 25#include "util/cgroup.h"
752fde44 26#include "util/color.h"
ac96287c 27#include "util/config.h"
7c304ee0 28#include "util/debug.h"
4a3cec84 29#include "util/dso.h"
092bd3cd 30#include "util/env.h"
5ab8c689 31#include "util/event.h"
ca125277
ACM		 32#include "util/evsel.h"
33#include "util/evsel_fprintf.h"
ea49e01c 34#include "util/synthetic-events.h"
514f1c67 35#include "util/evlist.h"
22ac4318 36#include "util/evswitch.h"
e0fcfb08 37#include "util/mmap.h"
8520a98d 38#include <subcmd/pager.h>
4b6ab94e 39#include <subcmd/exec-cmd.h>
752fde44 40#include "util/machine.h"
1101f69a 41#include "util/map.h"
daecf9e0 42#include "util/symbol.h"
9a3993d4 43#include "util/path.h"
6810fc91 44#include "util/session.h"
752fde44 45#include "util/thread.h"
4b6ab94e 46#include <subcmd/parse-options.h>
2ae3a312 47#include "util/strlist.h"
bdc89661 48#include "util/intlist.h"
514f1c67 49#include "util/thread_map.h"
bf2575c1 50#include "util/stat.h"
4a3cec84 51#include "util/tool.h"
2da39f1c 52#include "util/util.h"
fd5cead2 53#include "trace/beauty/beauty.h"
97978b3e 54#include "trace-event.h"
9aca7f17 55#include "util/parse-events.h"
ba504235 56#include "util/bpf-loader.h"
9b7c7728 57#include "util/tracepoint.h"
566a0885 58#include "callchain.h"
fea01392 59#include "print_binary.h"
a067558e 60#include "string2.h"
fd0db102 61#include "syscalltbl.h"
96c14451 62#include "rb_resort.h"
c1a604df 63#include "../perf.h"
514f1c67 64
a43783ae 65#include <errno.h>
fd20e811 66#include <inttypes.h>
4208735d 67#include <poll.h>
9607ad3a 68#include <signal.h>
514f1c67 69#include <stdlib.h>
017037ff 70#include <string.h>
8dd2a131 71#include <linux/err.h>
997bba8c 72#include <linux/filter.h>
877a7a11 73#include <linux/kernel.h>
39878d49 74#include <linux/random.h>
c6d4a494 75#include <linux/stringify.h>
bd48c63e 76#include <linux/time64.h>
7f7c536f 77#include <linux/zalloc.h>
bafae98e 78#include <fcntl.h>
4bcc4cff 79#include <sys/sysmacros.h>
514f1c67 80
3052ba56 81#include <linux/ctype.h>
7728fa0c 82#include <perf/mmap.h>
3d689ed6 83
378ef0f5
IR		 84#ifdef HAVE_LIBTRACEEVENT
85#include <traceevent/event-parse.h>
86#endif
87
c188e7ac
ACM		 88#ifndef O_CLOEXEC
89# define O_CLOEXEC 02000000
90#endif
91
83a51694
ACM		 92#ifndef F_LINUX_SPECIFIC_BASE
93# define F_LINUX_SPECIFIC_BASE 1024
94#endif
95
eadcab4c
LY		 96#define RAW_SYSCALL_ARGS_NUM 6
97
3f41b778
ACM		 98/*
99 * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100
100 */
888ca854
ACM		 101struct syscall_arg_fmt {
102 size_t (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
3f41b778 103 bool (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val);
888ca854
ACM		 104 unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
105 void *parm;
106 const char *name;
9597945d 107 u16 nr_entries; // for arrays
888ca854
ACM		 108 bool show_zero;
109};
110
111struct syscall_fmt {
112 const char *name;
113 const char *alias;
114 struct {
115 const char *sys_enter,
116 *sys_exit;
117 } bpf_prog_name;
eadcab4c 118 struct syscall_arg_fmt arg[RAW_SYSCALL_ARGS_NUM];
888ca854
ACM		 119 u8 nr_args;
120 bool errpid;
121 bool timeout;
122 bool hexret;
123};
124
d1d438a3
ACM		 125struct trace {
126 struct perf_tool tool;
fd0db102 127 struct syscalltbl *sctbl;
d1d438a3 128 struct {
d1d438a3 129 struct syscall *table;
3803a229
ACM		 130 struct { // per syscall BPF_MAP_TYPE_PROG_ARRAY
131 struct bpf_map *sys_enter,
132 *sys_exit;
133 } prog_array;
d1d438a3 134 struct {
32dcd021 135 struct evsel *sys_enter,
d3d1c4bd
ACM		 136 *sys_exit,
137 *augmented;
d1d438a3 138 } events;
5834da7f 139 struct bpf_program *unaugmented_prog;
d1d438a3 140 } syscalls;
ff7a4f98
ACM		 141 struct {
142 struct bpf_map *map;
143 } dump;
d1d438a3 144 struct record_opts opts;
63503dba 145 struct evlist *evlist;
d1d438a3
ACM		 146 struct machine *host;
147 struct thread *current;
c8c80570 148 struct bpf_object *bpf_obj;
9ea42ba4 149 struct cgroup *cgroup;
d1d438a3
ACM		 150 u64 base_time;
151 FILE *output;
152 unsigned long nr_events;
5067a8cd
ACM		 153 unsigned long nr_events_printed;
154 unsigned long max_events;
22ac4318 155 struct evswitch evswitch;
d1d438a3
ACM		 156 struct strlist *ev_qualifier;
157 struct {
158 size_t nr;
159 int *entries;
160 } ev_qualifier_ids;
d1d438a3
ACM		 161 struct {
162 size_t nr;
163 pid_t *entries;
744fafc7 164 struct bpf_map *map;
d1d438a3
ACM		 165 } filter_pids;
166 double duration_filter;
167 double runtime_ms;
168 struct {
169 u64 vfs_getname,
170 proc_getname;
171 } stats;
c6d4a494 172 unsigned int max_stack;
5cf9c84e 173 unsigned int min_stack;
b9b6a2ea 174 int raw_augmented_syscalls_args_size;
cd26ea6d 175 bool raw_augmented_syscalls;
79d725cd 176 bool fd_path_disabled;
b9b6a2ea 177 bool sort_events;
d1d438a3
ACM		 178 bool not_ev_qualifier;
179 bool live;
180 bool full_time;
181 bool sched;
182 bool multiple_threads;
183 bool summary;
184 bool summary_only;
b88b14db 185 bool errno_summary;
0a6545bd 186 bool failure_only;
d1d438a3 187 bool show_comm;
591421e1 188 bool print_sample;
d1d438a3
ACM		 189 bool show_tool_stats;
190 bool trace_syscalls;
f11b2803 191 bool libtraceevent_print;
44621819 192 bool kernel_syscallchains;
9ed45d59 193 s16 args_alignment;
b036146f 194 bool show_tstamp;
42e4a52d 195 bool show_duration;
e7c634fc 196 bool show_zeros;
9d6dc178 197 bool show_arg_names;
c65c83ff 198 bool show_string_prefix;
d1d438a3
ACM		 199 bool force;
200 bool vfs_getname;
201 int trace_pgfaults;
7e035929 202 char *perfconfig_events;
028713aa
JO		 203 struct {
204 struct ordered_events data;
205 u64 last;
206 } oe;
d1d438a3 207};
a1c2552d 208
77170988
ACM		 209struct tp_field {
210 int offset;
211 union {
212 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
213 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
214 };
215};
216
217#define TP_UINT_FIELD(bits) \
218static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
219{ \
55d43bca
DA		 220 u##bits value; \
221 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
222 return value; \
77170988
ACM		 223}
224
225TP_UINT_FIELD(8);
226TP_UINT_FIELD(16);
227TP_UINT_FIELD(32);
228TP_UINT_FIELD(64);
229
230#define TP_UINT_FIELD__SWAPPED(bits) \
231static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
232{ \
55d43bca
DA		 233 u##bits value; \
234 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
77170988
ACM		 235 return bswap_##bits(value);\
236}
237
238TP_UINT_FIELD__SWAPPED(16);
239TP_UINT_FIELD__SWAPPED(32);
240TP_UINT_FIELD__SWAPPED(64);
241
aa823f58 242static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
77170988 243{
aa823f58 244 field->offset = offset;
77170988 245
aa823f58 246 switch (size) {
77170988
ACM		 247 case 1:
248 field->integer = tp_field__u8;
249 break;
250 case 2:
251 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
252 break;
253 case 4:
254 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
255 break;
256 case 8:
257 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
258 break;
259 default:
260 return -1;
261 }
262
263 return 0;
264}
265
2c92f982 266static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
aa823f58
ACM		 267{
268 return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
269}
270
77170988
ACM		 271static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
272{
273 return sample->raw_data + field->offset;
274}
275
aa823f58 276static int __tp_field__init_ptr(struct tp_field *field, int offset)
77170988 277{
aa823f58 278 field->offset = offset;
77170988
ACM		 279 field->pointer = tp_field__ptr;
280 return 0;
281}
282
2c92f982 283static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
aa823f58
ACM		 284{
285 return __tp_field__init_ptr(field, format_field->offset);
286}
287
77170988
ACM		 288struct syscall_tp {
289 struct tp_field id;
290 union {
291 struct tp_field args, ret;
292 };
293};
294
2b00bb62
ACM		 295/*
296 * The evsel->priv as used by 'perf trace'
297 * sc: for raw_syscalls:sys_{enter,exit} and syscalls:sys_{enter,exit}_SYSCALLNAME
298 * fmt: for all the other tracepoints
299 */
300struct evsel_trace {
301 struct syscall_tp sc;
302 struct syscall_arg_fmt *fmt;
303};
304
305static struct evsel_trace *evsel_trace__new(void)
306{
307 return zalloc(sizeof(struct evsel_trace));
308}
309
310static void evsel_trace__delete(struct evsel_trace *et)
311{
312 if (et == NULL)
313 return;
314
315 zfree(&et->fmt);
316 free(et);
317}
318
fecd9907
ACM		 319/*
320 * Used with raw_syscalls:sys_{enter,exit} and with the
321 * syscalls:sys_{enter,exit}_SYSCALL tracepoints
322 */
323static inline struct syscall_tp *__evsel__syscall_tp(struct evsel *evsel)
324{
2b00bb62 325 struct evsel_trace *et = evsel->priv;
fecd9907 326
2b00bb62 327 return &et->sc;
fecd9907
ACM		 328}
329
8b913df5
ACM		 330static struct syscall_tp *evsel__syscall_tp(struct evsel *evsel)
331{
332 if (evsel->priv == NULL) {
2b00bb62
ACM		 333 evsel->priv = evsel_trace__new();
334 if (evsel->priv == NULL)
335 return NULL;
8b913df5
ACM		 336 }
337
338 return __evsel__syscall_tp(evsel);
339}
340
fecd9907
ACM		 341/*
342 * Used with all the other tracepoints.
343 */
344static inline struct syscall_arg_fmt *__evsel__syscall_arg_fmt(struct evsel *evsel)
345{
2b00bb62 346 struct evsel_trace *et = evsel->priv;
fecd9907 347
2b00bb62 348 return et->fmt;
fecd9907
ACM		 349}
350
8b913df5
ACM		 351static struct syscall_arg_fmt *evsel__syscall_arg_fmt(struct evsel *evsel)
352{
2b00bb62
ACM		 353 struct evsel_trace *et = evsel->priv;
354
8b913df5 355 if (evsel->priv == NULL) {
2b00bb62
ACM		 356 et = evsel->priv = evsel_trace__new();
357
358 if (et == NULL)
359 return NULL;
360 }
361
362 if (et->fmt == NULL) {
363 et->fmt = calloc(evsel->tp_format->format.nr_fields, sizeof(struct syscall_arg_fmt));
364 if (et->fmt == NULL)
365 goto out_delete;
8b913df5
ACM		 366 }
367
368 return __evsel__syscall_arg_fmt(evsel);
2b00bb62
ACM		 369
370out_delete:
371 evsel_trace__delete(evsel->priv);
372 evsel->priv = NULL;
373 return NULL;
8b913df5
ACM		 374}
375
794bca26 376static int evsel__init_tp_uint_field(struct evsel *evsel, struct tp_field *field, const char *name)
77170988 377{
efc0cdc9 378 struct tep_format_field *format_field = evsel__field(evsel, name);
77170988
ACM		 379
380 if (format_field == NULL)
381 return -1;
382
383 return tp_field__init_uint(field, format_field, evsel->needs_swap);
384}
385
386#define perf_evsel__init_sc_tp_uint_field(evsel, name) \
fecd9907 387 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
794bca26 388 evsel__init_tp_uint_field(evsel, &sc->name, #name); })
77170988 389
794bca26 390static int evsel__init_tp_ptr_field(struct evsel *evsel, struct tp_field *field, const char *name)
77170988 391{
efc0cdc9 392 struct tep_format_field *format_field = evsel__field(evsel, name);
77170988
ACM		 393
394 if (format_field == NULL)
395 return -1;
396
397 return tp_field__init_ptr(field, format_field);
398}
399
400#define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
fecd9907 401 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
794bca26 402 evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
77170988 403
5eb2dd2a 404static void evsel__delete_priv(struct evsel *evsel)
77170988 405{
04662523 406 zfree(&evsel->priv);
5eb2dd2a 407 evsel__delete(evsel);
77170988
ACM		 408}
409
794bca26 410static int evsel__init_syscall_tp(struct evsel *evsel)
d32855fa 411{
8b913df5 412 struct syscall_tp *sc = evsel__syscall_tp(evsel);
d32855fa 413
8b913df5 414 if (sc != NULL) {
794bca26
ACM		 415 if (evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
416 evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
8b913df5 417 return -ENOENT;
d32855fa
ACM		 418 return 0;
419 }
420
421 return -ENOMEM;
d32855fa
ACM		 422}
423
794bca26 424static int evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)
d3d1c4bd 425{
8b913df5 426 struct syscall_tp *sc = evsel__syscall_tp(evsel);
d3d1c4bd 427
8b913df5 428 if (sc != NULL) {
efc0cdc9 429 struct tep_format_field *syscall_id = evsel__field(tp, "id");
b9b6a2ea 430 if (syscall_id == NULL)
efc0cdc9 431 syscall_id = evsel__field(tp, "__syscall_nr");
8b913df5
ACM		 432 if (syscall_id == NULL ||
433 __tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
434 return -EINVAL;
d3d1c4bd
ACM		 435
436 return 0;
437 }
438
439 return -ENOMEM;
d3d1c4bd
ACM		 440}
441
794bca26 442static int evsel__init_augmented_syscall_tp_args(struct evsel *evsel)
d3d1c4bd 443{
fecd9907 444 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
d3d1c4bd
ACM		 445
446 return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
447}
448
794bca26 449static int evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)
b1a9e253 450{
fecd9907 451 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
b1a9e253
ACM		 452
453 return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
454}
455
794bca26 456static int evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)
96695d44 457{
8b913df5 458 if (evsel__syscall_tp(evsel) != NULL) {
96695d44 459 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
8b913df5 460 return -ENOENT;
96695d44
NK		 461
462 evsel->handler = handler;
463 return 0;
464 }
465
466 return -ENOMEM;
96695d44
NK		 467}
468
32dcd021 469static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
77170988 470{
8f6725a2 471 struct evsel *evsel = evsel__newtp("raw_syscalls", direction);
77170988 472
9aca7f17 473 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
8dd2a131 474 if (IS_ERR(evsel))
8f6725a2 475 evsel = evsel__newtp("syscalls", direction);
9aca7f17 476
8dd2a131
JO		 477 if (IS_ERR(evsel))
478 return NULL;
479
794bca26 480 if (evsel__init_raw_syscall_tp(evsel, handler))
8dd2a131 481 goto out_delete;
77170988
ACM		 482
483 return evsel;
484
485out_delete:
5eb2dd2a 486 evsel__delete_priv(evsel);
77170988
ACM		 487 return NULL;
488}
489
490#define perf_evsel__sc_tp_uint(evsel, name, sample) \
fecd9907 491 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
77170988
ACM		 492 fields->name.integer(&fields->name, sample); })
493
494#define perf_evsel__sc_tp_ptr(evsel, name, sample) \
fecd9907 495 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
77170988
ACM		 496 fields->name.pointer(&fields->name, sample); })
497
97c2a780
ACM		 498size_t strarray__scnprintf_suffix(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_suffix, int val)
499{
500 int idx = val - sa->offset;
501
502 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
503 size_t printed = scnprintf(bf, size, intfmt, val);
504 if (show_suffix)
505 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
506 return printed;
507 }
508
509 return scnprintf(bf, size, "%s%s", sa->entries[idx], show_suffix ? sa->prefix : "");
510}
511
c65c83ff 512size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
0ae79636
ACM		 513{
514 int idx = val - sa->offset;
1f115cb7 515
9614b8d6
ACM		 516 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
517 size_t printed = scnprintf(bf, size, intfmt, val);
518 if (show_prefix)
519 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
520 return printed;
521 }
1f115cb7 522
c65c83ff 523 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
03e3adc9
ACM		 524}
525
975b7c2f
ACM		 526static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
527 const char *intfmt,
528 struct syscall_arg *arg)
1f115cb7 529{
c65c83ff 530 return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
1f115cb7
ACM		 531}
532
975b7c2f
ACM		 533static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
534 struct syscall_arg *arg)
535{
536 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
537}
538
1f115cb7
ACM		 539#define SCA_STRARRAY syscall_arg__scnprintf_strarray
540
d066da97
ACM		 541bool syscall_arg__strtoul_strarray(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
542{
543 return strarray__strtoul(arg->parm, bf, size, ret);
544}
545
e0712baa
ACM		 546bool syscall_arg__strtoul_strarray_flags(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
547{
548 return strarray__strtoul_flags(arg->parm, bf, size, ret);
549}
550
1a8a90b8
ACM		 551bool syscall_arg__strtoul_strarrays(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
552{
553 return strarrays__strtoul(arg->parm, bf, size, ret);
554}
555
f5b91dbb
ACM		 556size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)
557{
558 return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);
559}
560
1f2d085e 561size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
83a51694 562{
9614b8d6 563 size_t printed;
83a51694
ACM		 564 int i;
565
566 for (i = 0; i < sas->nr_entries; ++i) {
567 struct strarray *sa = sas->entries[i];
1f2d085e 568 int idx = val - sa->offset;
83a51694
ACM		 569
570 if (idx >= 0 && idx < sa->nr_entries) {
571 if (sa->entries[idx] == NULL)
572 break;
1f2d085e 573 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
83a51694
ACM		 574 }
575 }
576
9614b8d6
ACM		 577 printed = scnprintf(bf, size, intfmt, val);
578 if (show_prefix)
579 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
580 return printed;
1f2d085e
ACM		 581}
582
d0a3a104
ACM		 583bool strarray__strtoul(struct strarray *sa, char *bf, size_t size, u64 *ret)
584{
585 int i;
586
587 for (i = 0; i < sa->nr_entries; ++i) {
588 if (sa->entries[i] && strncmp(sa->entries[i], bf, size) == 0 && sa->entries[i][size] == '\0') {
589 *ret = sa->offset + i;
590 return true;
591 }
592 }
593
594 return false;
595}
596
154c978d
ACM		 597bool strarray__strtoul_flags(struct strarray *sa, char *bf, size_t size, u64 *ret)
598{
599 u64 val = 0;
600 char *tok = bf, *sep, *end;
601
602 *ret = 0;
603
604 while (size != 0) {
605 int toklen = size;
606
607 sep = memchr(tok, '|', size);
608 if (sep != NULL) {
609 size -= sep - tok + 1;
610
611 end = sep - 1;
612 while (end > tok && isspace(*end))
613 --end;
614
615 toklen = end - tok + 1;
616 }
617
618 while (isspace(*tok))
619 ++tok;
620
621 if (isalpha(*tok) || *tok == '_') {
622 if (!strarray__strtoul(sa, tok, toklen, &val))
623 return false;
96b73141
CZ		 624 } else
625 val = strtoul(tok, NULL, 0);
154c978d
ACM		 626
627 *ret |= (1 << (val - 1));
628
629 if (sep == NULL)
630 break;
631 tok = sep + 1;
632 }
633
634 return true;
635}
636
d0a3a104
ACM		 637bool strarrays__strtoul(struct strarrays *sas, char *bf, size_t size, u64 *ret)
638{
639 int i;
640
641 for (i = 0; i < sas->nr_entries; ++i) {
642 struct strarray *sa = sas->entries[i];
643
644 if (strarray__strtoul(sa, bf, size, ret))
645 return true;
646 }
647
648 return false;
649}
650
1f2d085e
ACM		 651size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
652 struct syscall_arg *arg)
653{
654 return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
83a51694
ACM		 655}
656
48e1f91a
ACM		 657#ifndef AT_FDCWD
658#define AT_FDCWD -100
659#endif
660
75b757ca
ACM		 661static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
662 struct syscall_arg *arg)
663{
664 int fd = arg->val;
c65c83ff 665 const char *prefix = "AT_FD";
75b757ca
ACM		 666
667 if (fd == AT_FDCWD)
c65c83ff 668 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
75b757ca
ACM		 669
670 return syscall_arg__scnprintf_fd(bf, size, arg);
671}
672
673#define SCA_FDAT syscall_arg__scnprintf_fd_at
674
675static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
676 struct syscall_arg *arg);
677
678#define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
679
2c2b1623 680size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
13d4ff3e 681{
01533e97 682 return scnprintf(bf, size, "%#lx", arg->val);
13d4ff3e
ACM		 683}
684
ce05539f
ACM		 685size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
686{
687 if (arg->val == 0)
688 return scnprintf(bf, size, "NULL");
689 return syscall_arg__scnprintf_hex(bf, size, arg);
690}
691
2c2b1623 692size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
a1c2552d
ACM		 693{
694 return scnprintf(bf, size, "%d", arg->val);
695}
696
5dde91ed
ACM		 697size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
698{
699 return scnprintf(bf, size, "%ld", arg->val);
700}
701
9597945d
ACM		 702static size_t syscall_arg__scnprintf_char_array(char *bf, size_t size, struct syscall_arg *arg)
703{
704 // XXX Hey, maybe for sched:sched_switch prev/next comm fields we can
705 // fill missing comms using thread__set_comm()...
706 // here or in a special syscall_arg__scnprintf_pid_sched_tp...
c5e006cd 707 return scnprintf(bf, size, "\"%-.*s\"", arg->fmt->nr_entries ?: arg->len, arg->val);
9597945d
ACM		 708}
709
710#define SCA_CHAR_ARRAY syscall_arg__scnprintf_char_array
711
729a7841
ACM		 712static const char *bpf_cmd[] = {
713 "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
ea0056f0
WL		 714 "MAP_GET_NEXT_KEY", "PROG_LOAD", "OBJ_PIN", "OBJ_GET", "PROG_ATTACH",
715 "PROG_DETACH", "PROG_TEST_RUN", "PROG_GET_NEXT_ID", "MAP_GET_NEXT_ID",
716 "PROG_GET_FD_BY_ID", "MAP_GET_FD_BY_ID", "OBJ_GET_INFO_BY_FD",
717 "PROG_QUERY", "RAW_TRACEPOINT_OPEN", "BTF_LOAD", "BTF_GET_FD_BY_ID",
718 "TASK_FD_QUERY", "MAP_LOOKUP_AND_DELETE_ELEM", "MAP_FREEZE",
719 "BTF_GET_NEXT_ID", "MAP_LOOKUP_BATCH", "MAP_LOOKUP_AND_DELETE_BATCH",
720 "MAP_UPDATE_BATCH", "MAP_DELETE_BATCH", "LINK_CREATE", "LINK_UPDATE",
721 "LINK_GET_FD_BY_ID", "LINK_GET_NEXT_ID", "ENABLE_STATS", "ITER_CREATE",
722 "LINK_DETACH", "PROG_BIND_MAP",
729a7841 723};
2e3d7fac 724static DEFINE_STRARRAY(bpf_cmd, "BPF_");
729a7841 725
f6af0956
ACM		 726static const char *fsmount_flags[] = {
727 [1] = "CLOEXEC",
728};
729static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");
730
dcc6fd64
ACM		 731#include "trace/beauty/generated/fsconfig_arrays.c"
732
733static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");
734
03e3adc9 735static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
2e3d7fac 736static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
eac032c5 737
1f115cb7 738static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
2e3d7fac 739static DEFINE_STRARRAY(itimers, "ITIMER_");
1f115cb7 740
b62bee1b
ACM		 741static const char *keyctl_options[] = {
742 "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
743 "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
744 "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
745 "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
746 "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
747};
2e3d7fac 748static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
b62bee1b 749
efe6b882
ACM		 750static const char *whences[] = { "SET", "CUR", "END",
751#ifdef SEEK_DATA
752"DATA",
753#endif
754#ifdef SEEK_HOLE
755"HOLE",
756#endif
757};
2e3d7fac 758static DEFINE_STRARRAY(whences, "SEEK_");
f9da0b0c 759
80f587d5
ACM		 760static const char *fcntl_cmds[] = {
761 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
e000e5e3
ACM		 762 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
763 "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
764 "GETOWNER_UIDS",
80f587d5 765};
2e3d7fac 766static DEFINE_STRARRAY(fcntl_cmds, "F_");
80f587d5 767
83a51694
ACM		 768static const char *fcntl_linux_specific_cmds[] = {
769 "SETLEASE", "GETLEASE", "NOTIFY", [5] = "CANCELLK", "DUPFD_CLOEXEC",
770 "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
64e4561d 771 "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
83a51694
ACM		 772};
773
2e3d7fac 774static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
83a51694
ACM		 775
776static struct strarray *fcntl_cmds_arrays[] = {
777 &strarray__fcntl_cmds,
778 &strarray__fcntl_linux_specific_cmds,
779};
780
781static DEFINE_STRARRAYS(fcntl_cmds_arrays);
782
c045bf02
ACM		 783static const char *rlimit_resources[] = {
784 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
785 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
786 "RTTIME",
787};
2e3d7fac 788static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
c045bf02 789
eb5b1b14 790static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
2e3d7fac 791static DEFINE_STRARRAY(sighow, "SIG_");
eb5b1b14 792
4f8c1b74
DA		 793static const char *clockid[] = {
794 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
28ebb87c
ACM		 795 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
796 "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
4f8c1b74 797};
2e3d7fac 798static DEFINE_STRARRAY(clockid, "CLOCK_");
4f8c1b74 799
51108999
ACM		 800static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
801 struct syscall_arg *arg)
802{
c65c83ff
ACM		 803 bool show_prefix = arg->show_string_prefix;
804 const char *suffix = "_OK";
51108999
ACM		 805 size_t printed = 0;
806 int mode = arg->val;
807
808 if (mode == F_OK) /* 0 */
c65c83ff 809 return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
51108999
ACM		 810#define P_MODE(n) \
811 if (mode & n##_OK) { \
c65c83ff 812 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
51108999
ACM		 813 mode &= ~n##_OK; \
814 }
815
816 P_MODE(R);
817 P_MODE(W);
818 P_MODE(X);
819#undef P_MODE
820
821 if (mode)
822 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
823
824 return printed;
825}
826
827#define SCA_ACCMODE syscall_arg__scnprintf_access_mode
828
f994592d
ACM		 829static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
830 struct syscall_arg *arg);
831
832#define SCA_FILENAME syscall_arg__scnprintf_filename
833
46cce19b
ACM		 834static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
835 struct syscall_arg *arg)
836{
c65c83ff
ACM		 837 bool show_prefix = arg->show_string_prefix;
838 const char *prefix = "O_";
46cce19b
ACM		 839 int printed = 0, flags = arg->val;
840
841#define P_FLAG(n) \
842 if (flags & O_##n) { \
c65c83ff 843 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
46cce19b
ACM		 844 flags &= ~O_##n; \
845 }
846
847 P_FLAG(CLOEXEC);
848 P_FLAG(NONBLOCK);
849#undef P_FLAG
850
851 if (flags)
852 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
853
854 return printed;
855}
856
857#define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
858
a355a61e
ACM		 859#ifndef GRND_NONBLOCK
860#define GRND_NONBLOCK 0x0001
861#endif
862#ifndef GRND_RANDOM
863#define GRND_RANDOM 0x0002
864#endif
865
39878d49
ACM		 866static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
867 struct syscall_arg *arg)
868{
c65c83ff
ACM		 869 bool show_prefix = arg->show_string_prefix;
870 const char *prefix = "GRND_";
39878d49
ACM		 871 int printed = 0, flags = arg->val;
872
873#define P_FLAG(n) \
874 if (flags & GRND_##n) { \
c65c83ff 875 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
39878d49
ACM		 876 flags &= ~GRND_##n; \
877 }
878
879 P_FLAG(RANDOM);
880 P_FLAG(NONBLOCK);
881#undef P_FLAG
882
883 if (flags)
884 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
885
886 return printed;
887}
888
889#define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
890
82d4a110
ACM		 891#define STRARRAY(name, array) \
892 { .scnprintf = SCA_STRARRAY, \
d066da97 893 .strtoul = STUL_STRARRAY, \
82d4a110 894 .parm = &strarray__##array, }
453350dd 895
f6af0956
ACM		 896#define STRARRAY_FLAGS(name, array) \
897 { .scnprintf = SCA_STRARRAY_FLAGS, \
e0712baa 898 .strtoul = STUL_STRARRAY_FLAGS, \
f6af0956
ACM		 899 .parm = &strarray__##array, }
900
092bd3cd 901#include "trace/beauty/arch_errno_names.c"
ea8dc3ce 902#include "trace/beauty/eventfd.c"
d5d71e86 903#include "trace/beauty/futex_op.c"
3258abe0 904#include "trace/beauty/futex_val3.c"
df4cb167 905#include "trace/beauty/mmap.c"
ba2f22cf 906#include "trace/beauty/mode_t.c"
a30e6259 907#include "trace/beauty/msg_flags.c"
8f48df69 908#include "trace/beauty/open_flags.c"
62de344e 909#include "trace/beauty/perf_event_open.c"
d5d71e86 910#include "trace/beauty/pid.c"
a3bca91f 911#include "trace/beauty/sched_policy.c"
f5cd95ea 912#include "trace/beauty/seccomp.c"
12199d8e 913#include "trace/beauty/signum.c"
bbf86c43 914#include "trace/beauty/socket_type.c"
7206b900 915#include "trace/beauty/waitid_options.c"
a3bca91f 916
60995604 917static const struct syscall_fmt syscall_fmts[] = {
1f63139c 918 { .name = "access",
82d4a110 919 .arg = { [1] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, },
fb7068e7
ACM		 920 { .name = "arch_prctl",
921 .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
922 [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
02ef2884 923 { .name = "bind",
ef969ca6 924 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
247dd65b
ACM		 925 [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ },
926 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
1f63139c 927 { .name = "bpf",
82d4a110 928 .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
beccb2b5 929 { .name = "brk", .hexret = true,
ce05539f 930 .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
1f63139c 931 { .name = "clock_gettime",
82d4a110 932 .arg = { [0] = STRARRAY(clk_id, clockid), }, },
6ac73820
ACM		 933 { .name = "clock_nanosleep",
934 .arg = { [2] = { .scnprintf = SCA_TIMESPEC, /* rqtp */ }, }, },
33396a3a
ACM		 935 { .name = "clone", .errpid = true, .nr_args = 5,
936 .arg = { [0] = { .name = "flags", .scnprintf = SCA_CLONE_FLAGS, },
937 [1] = { .name = "child_stack", .scnprintf = SCA_HEX, },
938 [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
939 [3] = { .name = "child_tidptr", .scnprintf = SCA_HEX, },
940 [4] = { .name = "tls", .scnprintf = SCA_HEX, }, }, },
1f63139c 941 { .name = "close",
82d4a110 942 .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
d5a7e661 943 { .name = "connect",
ef969ca6
ACM		 944 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
945 [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ },
1d862752 946 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
1f63139c 947 { .name = "epoll_ctl",
82d4a110 948 .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
1f63139c 949 { .name = "eventfd2",
82d4a110 950 .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
1f63139c 951 { .name = "fchmodat",
82d4a110 952 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1f63139c 953 { .name = "fchownat",
82d4a110 954 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1f63139c 955 { .name = "fcntl",
82c38338
ACM		 956 .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
957 .strtoul = STUL_STRARRAYS,
39cc355b
ACM		 958 .parm = &strarrays__fcntl_cmds_arrays,
959 .show_zero = true, },
82d4a110 960 [2] = { .scnprintf = SCA_FCNTL_ARG, /* arg */ }, }, },
1f63139c 961 { .name = "flock",
82d4a110 962 .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
dcc6fd64
ACM		 963 { .name = "fsconfig",
964 .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },
f6af0956
ACM		 965 { .name = "fsmount",
966 .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),
967 [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },
693bd394
ACM		 968 { .name = "fspick",
969 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
970 [1] = { .scnprintf = SCA_FILENAME, /* path */ },
971 [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },
1f63139c
ACM		 972 { .name = "fstat", .alias = "newfstat", },
973 { .name = "fstatat", .alias = "newfstatat", },
974 { .name = "futex",
3258abe0
ACM		 975 .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
976 [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
1f63139c 977 { .name = "futimesat",
82d4a110 978 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1f63139c 979 { .name = "getitimer",
82d4a110 980 .arg = { [0] = STRARRAY(which, itimers), }, },
c65f1070 981 { .name = "getpid", .errpid = true, },
d1d438a3 982 { .name = "getpgid", .errpid = true, },
c65f1070 983 { .name = "getppid", .errpid = true, },
1f63139c 984 { .name = "getrandom",
82d4a110 985 .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
1f63139c 986 { .name = "getrlimit",
82d4a110 987 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
f1c1e45e
ACM		 988 { .name = "getsockopt",
989 .arg = { [1] = STRARRAY(level, socket_level), }, },
2d1073de 990 { .name = "gettid", .errpid = true, },
1f63139c 991 { .name = "ioctl",
82d4a110 992 .arg = {
844ae5b4
ACM		 993#if defined(__i386__) || defined(__x86_64__)
994/*
995 * FIXME: Make this available to all arches.
996 */
1cc47f2d 997 [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
82d4a110 998 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
844ae5b4 999#else
82d4a110 1000 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
844ae5b4 1001#endif
1de3038d
ACM		 1002 { .name = "kcmp", .nr_args = 5,
1003 .arg = { [0] = { .name = "pid1", .scnprintf = SCA_PID, },
1004 [1] = { .name = "pid2", .scnprintf = SCA_PID, },
1005 [2] = { .name = "type", .scnprintf = SCA_KCMP_TYPE, },
1006 [3] = { .name = "idx1", .scnprintf = SCA_KCMP_IDX, },
1007 [4] = { .name = "idx2", .scnprintf = SCA_KCMP_IDX, }, }, },
1f63139c 1008 { .name = "keyctl",
82d4a110 1009 .arg = { [0] = STRARRAY(option, keyctl_options), }, },
1f63139c 1010 { .name = "kill",
82d4a110 1011 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1f63139c 1012 { .name = "linkat",
82d4a110 1013 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1f63139c 1014 { .name = "lseek",
82d4a110 1015 .arg = { [2] = STRARRAY(whence, whences), }, },
1f63139c
ACM		 1016 { .name = "lstat", .alias = "newlstat", },
1017 { .name = "madvise",
82d4a110
ACM		 1018 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1019 [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
1f63139c 1020 { .name = "mkdirat",
82d4a110 1021 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1f63139c 1022 { .name = "mknodat",
82d4a110 1023 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
beccb2b5 1024 { .name = "mmap", .hexret = true,
54265664
JO		 1025/* The standard mmap maps to old_mmap on s390x */
1026#if defined(__s390x__)
1027 .alias = "old_mmap",
1028#endif
ce05539f 1029 .arg = { [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
27198a89
ACM		 1030 [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */
1031 .strtoul = STUL_STRARRAY_FLAGS,
1032 .parm = &strarray__mmap_flags, },
a6631340 1033 [5] = { .scnprintf = SCA_HEX, /* offset */ }, }, },
73d141ad 1034 { .name = "mount",
23c07a23
ACM		 1035 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
1036 [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
73d141ad 1037 .mask_val = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
566e3022
ACM		 1038 { .name = "move_mount",
1039 .arg = { [0] = { .scnprintf = SCA_FDAT, /* from_dfd */ },
1040 [1] = { .scnprintf = SCA_FILENAME, /* from_pathname */ },
1041 [2] = { .scnprintf = SCA_FDAT, /* to_dfd */ },
1042 [3] = { .scnprintf = SCA_FILENAME, /* to_pathname */ },
1043 [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },
1f63139c 1044 { .name = "mprotect",
82d4a110
ACM		 1045 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1046 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ }, }, },
1f63139c 1047 { .name = "mq_unlink",
82d4a110 1048 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
ae685380 1049 { .name = "mremap", .hexret = true,
ce05539f 1050 .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
1f63139c 1051 { .name = "name_to_handle_at",
82d4a110 1052 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1f63139c 1053 { .name = "newfstatat",
82d4a110 1054 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1f63139c 1055 { .name = "open",
82d4a110 1056 .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1f63139c 1057 { .name = "open_by_handle_at",
82d4a110
ACM		 1058 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1059 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1f63139c 1060 { .name = "openat",
82d4a110
ACM		 1061 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1062 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1f63139c 1063 { .name = "perf_event_open",
a9cd6c67
ACM		 1064 .arg = { [0] = { .scnprintf = SCA_PERF_ATTR, /* attr */ },
1065 [2] = { .scnprintf = SCA_INT, /* cpu */ },
82d4a110
ACM		 1066 [3] = { .scnprintf = SCA_FD, /* group_fd */ },
1067 [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
1f63139c 1068 { .name = "pipe2",
82d4a110 1069 .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
83bc9c37
ACM		 1070 { .name = "pkey_alloc",
1071 .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS, /* access_rights */ }, }, },
1072 { .name = "pkey_free",
1073 .arg = { [0] = { .scnprintf = SCA_INT, /* key */ }, }, },
1074 { .name = "pkey_mprotect",
1075 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1076 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
1077 [3] = { .scnprintf = SCA_INT, /* pkey */ }, }, },
1f63139c
ACM		 1078 { .name = "poll", .timeout = true, },
1079 { .name = "ppoll", .timeout = true, },
fb7068e7 1080 { .name = "prctl",
d7a07b29
ACM		 1081 .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */
1082 .strtoul = STUL_STRARRAY,
1083 .parm = &strarray__prctl_options, },
d688d037
ACM		 1084 [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
1085 [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
1f63139c
ACM		 1086 { .name = "pread", .alias = "pread64", },
1087 { .name = "preadv", .alias = "pread", },
1088 { .name = "prlimit64",
82d4a110 1089 .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
1f63139c
ACM		 1090 { .name = "pwrite", .alias = "pwrite64", },
1091 { .name = "readlinkat",
82d4a110 1092 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1f63139c 1093 { .name = "recvfrom",
82d4a110 1094 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1f63139c 1095 { .name = "recvmmsg",
82d4a110 1096 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1f63139c 1097 { .name = "recvmsg",
82d4a110 1098 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1f63139c 1099 { .name = "renameat",
436651ca
ACM		 1100 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1101 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
84a83541
ACM		 1102 { .name = "renameat2",
1103 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
ca7ff2c8
ACM		 1104 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
1105 [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
1f63139c 1106 { .name = "rt_sigaction",
82d4a110 1107 .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1f63139c 1108 { .name = "rt_sigprocmask",
82d4a110 1109 .arg = { [0] = STRARRAY(how, sighow), }, },
1f63139c 1110 { .name = "rt_sigqueueinfo",
82d4a110 1111 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1f63139c 1112 { .name = "rt_tgsigqueueinfo",
82d4a110 1113 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1f63139c 1114 { .name = "sched_setscheduler",
82d4a110 1115 .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
1f63139c 1116 { .name = "seccomp",
82d4a110
ACM		 1117 .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP, /* op */ },
1118 [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
1f63139c 1119 { .name = "select", .timeout = true, },
e4b00e93 1120 { .name = "sendfile", .alias = "sendfile64", },
1f63139c 1121 { .name = "sendmmsg",
82d4a110 1122 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1f63139c 1123 { .name = "sendmsg",
82d4a110 1124 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1f63139c 1125 { .name = "sendto",
6ebb6862
ACM		 1126 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
1127 [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
c65f1070 1128 { .name = "set_tid_address", .errpid = true, },
1f63139c 1129 { .name = "setitimer",
82d4a110 1130 .arg = { [0] = STRARRAY(which, itimers), }, },
1f63139c 1131 { .name = "setrlimit",
82d4a110 1132 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
0826b7fd
ACM		 1133 { .name = "setsockopt",
1134 .arg = { [1] = STRARRAY(level, socket_level), }, },
1f63139c 1135 { .name = "socket",
82d4a110 1136 .arg = { [0] = STRARRAY(family, socket_families),
162d3edb
ACM		 1137 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1138 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1f63139c 1139 { .name = "socketpair",
82d4a110 1140 .arg = { [0] = STRARRAY(family, socket_families),
162d3edb
ACM		 1141 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1142 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1f63139c
ACM		 1143 { .name = "stat", .alias = "newstat", },
1144 { .name = "statx",
82d4a110
ACM		 1145 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fdat */ },
1146 [2] = { .scnprintf = SCA_STATX_FLAGS, /* flags */ } ,
1147 [3] = { .scnprintf = SCA_STATX_MASK, /* mask */ }, }, },
1f63139c 1148 { .name = "swapoff",
82d4a110 1149 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1f63139c 1150 { .name = "swapon",
82d4a110 1151 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1f63139c 1152 { .name = "symlinkat",
82d4a110 1153 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
a9a187a7
ACM		 1154 { .name = "sync_file_range",
1155 .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },
1f63139c 1156 { .name = "tgkill",
82d4a110 1157 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1f63139c 1158 { .name = "tkill",
82d4a110 1159 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
476c92ca
ACM		 1160 { .name = "umount2", .alias = "umount",
1161 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
1f63139c
ACM		 1162 { .name = "uname", .alias = "newuname", },
1163 { .name = "unlinkat",
82d4a110 1164 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1f63139c 1165 { .name = "utimensat",
82d4a110 1166 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
11c8e39f 1167 { .name = "wait4", .errpid = true,
82d4a110 1168 .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
11c8e39f 1169 { .name = "waitid", .errpid = true,
82d4a110 1170 .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
514f1c67
ACM		 1171};
1172
1173static int syscall_fmt__cmp(const void *name, const void *fmtp)
1174{
1175 const struct syscall_fmt *fmt = fmtp;
1176 return strcmp(name, fmt->name);
1177}
1178
60995604
IR		 1179static const struct syscall_fmt *__syscall_fmt__find(const struct syscall_fmt *fmts,
1180 const int nmemb,
1181 const char *name)
bcddbfc5
ACM		 1182{
1183 return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1184}
1185
60995604 1186static const struct syscall_fmt *syscall_fmt__find(const char *name)
514f1c67
ACM		 1187{
1188 const int nmemb = ARRAY_SIZE(syscall_fmts);
bcddbfc5 1189 return __syscall_fmt__find(syscall_fmts, nmemb, name);
514f1c67
ACM		 1190}
1191
60995604
IR		 1192static const struct syscall_fmt *__syscall_fmt__find_by_alias(const struct syscall_fmt *fmts,
1193 const int nmemb, const char *alias)
f932184e 1194{
bcddbfc5 1195 int i;
f932184e
ACM		 1196
1197 for (i = 0; i < nmemb; ++i) {
bcddbfc5
ACM		 1198 if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0)
1199 return &fmts[i];
f932184e
ACM		 1200 }
1201
1202 return NULL;
1203}
1204
60995604 1205static const struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
bcddbfc5
ACM		 1206{
1207 const int nmemb = ARRAY_SIZE(syscall_fmts);
1208 return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias);
1209}
1210
6a648b53
ACM		 1211/*
1212 * is_exit: is this "exit" or "exit_group"?
1213 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
7a983a0f 1214 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
b8b1033f 1215 * nonexistent: Just a hole in the syscall table, syscall id not allocated
6a648b53 1216 */
514f1c67 1217struct syscall {
97fbf3f0 1218 struct tep_event *tp_format;
f208bd8d 1219 int nr_args;
7a983a0f 1220 int args_size;
6ff8fff4
ACM		 1221 struct {
1222 struct bpf_program *sys_enter,
1223 *sys_exit;
1224 } bpf_prog;
6a648b53
ACM		 1225 bool is_exit;
1226 bool is_open;
b8b1033f 1227 bool nonexistent;
2c92f982 1228 struct tep_format_field *args;
514f1c67 1229 const char *name;
60995604 1230 const struct syscall_fmt *fmt;
82d4a110 1231 struct syscall_arg_fmt *arg_fmt;
514f1c67
ACM		 1232};
1233
fd2b2975
ACM		 1234/*
1235 * We need to have this 'calculated' boolean because in some cases we really
1236 * don't know what is the duration of a syscall, for instance, when we start
1237 * a session and some threads are waiting for a syscall to finish, say 'poll',
1238 * in which case all we can do is to print "( ? ) for duration and for the
1239 * start timestamp.
1240 */
1241static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
60c907ab
ACM		 1242{
1243 double duration = (double)t / NSEC_PER_MSEC;
1244 size_t printed = fprintf(fp, "(");
1245
fd2b2975 1246 if (!calculated)
522283fe 1247 printed += fprintf(fp, " ");
fd2b2975 1248 else if (duration >= 1.0)
60c907ab
ACM		 1249 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1250 else if (duration >= 0.01)
1251 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1252 else
1253 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
c24ff998 1254 return printed + fprintf(fp, "): ");
60c907ab
ACM		 1255}
1256
f994592d
ACM		 1257/**
1258 * filename.ptr: The filename char pointer that will be vfs_getname'd
1259 * filename.entry_str_pos: Where to insert the string translated from
1260 * filename.ptr by the vfs_getname tracepoint/kprobe.
84486caa
ACM		 1261 * ret_scnprintf: syscall args may set this to a different syscall return
1262 * formatter, for instance, fcntl may return fds, file flags, etc.
f994592d 1263 */
752fde44
ACM		 1264struct thread_trace {
1265 u64 entry_time;
752fde44 1266 bool entry_pending;
efd5745e 1267 unsigned long nr_events;
a2ea67d7 1268 unsigned long pfmaj, pfmin;
752fde44 1269 char *entry_str;
1302d88e 1270 double runtime_ms;
7ee57434 1271 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
f994592d
ACM		 1272 struct {
1273 unsigned long ptr;
7f4f8001
ACM		 1274 short int entry_str_pos;
1275 bool pending_open;
1276 unsigned int namelen;
1277 char *name;
f994592d 1278 } filename;
75b757ca 1279 struct {
f4a74fcb
ACM		 1280 int max;
1281 struct file *table;
1282 } files;
bf2575c1
DA		 1283
1284 struct intlist *syscall_stats;
752fde44
ACM		 1285};
1286
1287static struct thread_trace *thread_trace__new(void)
1288{
75b757ca
ACM		 1289 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
1290
7a6d49dc 1291 if (ttrace) {
f4a74fcb 1292 ttrace->files.max = -1;
7a6d49dc
LY		 1293 ttrace->syscall_stats = intlist__new(NULL);
1294 }
bf2575c1 1295
75b757ca 1296 return ttrace;
752fde44
ACM		 1297}
1298
c24ff998 1299static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
752fde44 1300{
efd5745e
ACM		 1301 struct thread_trace *ttrace;
1302
752fde44
ACM		 1303 if (thread == NULL)
1304 goto fail;
1305
89dceb22
NK		 1306 if (thread__priv(thread) == NULL)
1307 thread__set_priv(thread, thread_trace__new());
48000a1a 1308
89dceb22 1309 if (thread__priv(thread) == NULL)
752fde44
ACM		 1310 goto fail;
1311
89dceb22 1312 ttrace = thread__priv(thread);
efd5745e
ACM		 1313 ++ttrace->nr_events;
1314
1315 return ttrace;
752fde44 1316fail:
c24ff998 1317 color_fprintf(fp, PERF_COLOR_RED,
752fde44
ACM		 1318 "WARNING: not enough memory, dropping samples!\n");
1319 return NULL;
1320}
1321
84486caa
ACM		 1322
1323void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
7ee57434 1324 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
84486caa
ACM		 1325{
1326 struct thread_trace *ttrace = thread__priv(arg->thread);
1327
1328 ttrace->ret_scnprintf = ret_scnprintf;
1329}
1330
598d02c5
SF		 1331#define TRACE_PFMAJ (1 << 0)
1332#define TRACE_PFMIN (1 << 1)
1333
e4d44e83
ACM		 1334static const size_t trace__entry_str_size = 2048;
1335
d7e13484 1336static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
75b757ca 1337{
05107486
ACM		 1338 if (fd < 0)
1339 return NULL;
1340
f4a74fcb
ACM		 1341 if (fd > ttrace->files.max) {
1342 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
75b757ca 1343
f4a74fcb 1344 if (nfiles == NULL)
d7e13484 1345 return NULL;
75b757ca 1346
f4a74fcb
ACM		 1347 if (ttrace->files.max != -1) {
1348 memset(nfiles + ttrace->files.max + 1, 0,
1349 (fd - ttrace->files.max) * sizeof(struct file));
75b757ca 1350 } else {
f4a74fcb 1351 memset(nfiles, 0, (fd + 1) * sizeof(struct file));
75b757ca
ACM		 1352 }
1353
f4a74fcb
ACM		 1354 ttrace->files.table = nfiles;
1355 ttrace->files.max = fd;
75b757ca
ACM		 1356 }
1357
d7e13484
ACM		 1358 return ttrace->files.table + fd;
1359}
1360
2d473389
ACM		 1361struct file *thread__files_entry(struct thread *thread, int fd)
1362{
1363 return thread_trace__files_entry(thread__priv(thread), fd);
1364}
1365
d7e13484
ACM		 1366static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1367{
1368 struct thread_trace *ttrace = thread__priv(thread);
1369 struct file *file = thread_trace__files_entry(ttrace, fd);
1370
1371 if (file != NULL) {
4bcc4cff
ACM		 1372 struct stat st;
1373 if (stat(pathname, &st) == 0)
1374 file->dev_maj = major(st.st_rdev);
d7e13484
ACM		 1375 file->pathname = strdup(pathname);
1376 if (file->pathname)
1377 return 0;
1378 }
75b757ca 1379
d7e13484 1380 return -1;
75b757ca
ACM		 1381}
1382
97119f37
ACM		 1383static int thread__read_fd_path(struct thread *thread, int fd)
1384{
1385 char linkname[PATH_MAX], pathname[PATH_MAX];
1386 struct stat st;
1387 int ret;
1388
ee84a303 1389 if (thread__pid(thread) == thread__tid(thread)) {
97119f37 1390 scnprintf(linkname, sizeof(linkname),
ee84a303 1391 "/proc/%d/fd/%d", thread__pid(thread), fd);
97119f37
ACM		 1392 } else {
1393 scnprintf(linkname, sizeof(linkname),
ee84a303
IR		 1394 "/proc/%d/task/%d/fd/%d",
1395 thread__pid(thread), thread__tid(thread), fd);
97119f37
ACM		 1396 }
1397
1398 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1399 return -1;
1400
1401 ret = readlink(linkname, pathname, sizeof(pathname));
1402
1403 if (ret < 0 || ret > st.st_size)
1404 return -1;
1405
1406 pathname[ret] = '\0';
1407 return trace__set_fd_pathname(thread, fd, pathname);
1408}
1409
c522739d
ACM		 1410static const char *thread__fd_path(struct thread *thread, int fd,
1411 struct trace *trace)
75b757ca 1412{
89dceb22 1413 struct thread_trace *ttrace = thread__priv(thread);
75b757ca 1414
79d725cd 1415 if (ttrace == NULL || trace->fd_path_disabled)
75b757ca
ACM		 1416 return NULL;
1417
1418 if (fd < 0)
1419 return NULL;
1420
f4a74fcb 1421 if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
c522739d
ACM		 1422 if (!trace->live)
1423 return NULL;
1424 ++trace->stats.proc_getname;
cdcd1e6b 1425 if (thread__read_fd_path(thread, fd))
c522739d
ACM		 1426 return NULL;
1427 }
75b757ca 1428
f4a74fcb 1429 return ttrace->files.table[fd].pathname;
75b757ca
ACM		 1430}
1431
fc65eb82 1432size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
75b757ca
ACM		 1433{
1434 int fd = arg->val;
1435 size_t printed = scnprintf(bf, size, "%d", fd);
c522739d 1436 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
75b757ca
ACM		 1437
1438 if (path)
1439 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1440
1441 return printed;
1442}
1443
0a2f7540
ACM		 1444size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1445{
1446 size_t printed = scnprintf(bf, size, "%d", fd);
1447 struct thread *thread = machine__find_thread(trace->host, pid, pid);
1448
1449 if (thread) {
1450 const char *path = thread__fd_path(thread, fd, trace);
1451
1452 if (path)
1453 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1454
1455 thread__put(thread);
1456 }
1457
1458 return printed;
1459}
1460
75b757ca
ACM		 1461static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1462 struct syscall_arg *arg)
1463{
1464 int fd = arg->val;
1465 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
89dceb22 1466 struct thread_trace *ttrace = thread__priv(arg->thread);
75b757ca 1467
f4a74fcb
ACM		 1468 if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1469 zfree(&ttrace->files.table[fd].pathname);
75b757ca
ACM		 1470
1471 return printed;
1472}
1473
f994592d
ACM		 1474static void thread__set_filename_pos(struct thread *thread, const char *bf,
1475 unsigned long ptr)
1476{
1477 struct thread_trace *ttrace = thread__priv(thread);
1478
1479 ttrace->filename.ptr = ptr;
1480 ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1481}
1482
75d1e306
ACM		 1483static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1484{
1485 struct augmented_arg *augmented_arg = arg->augmented.args;
8195168e
ACM		 1486 size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1487 /*
1488 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1489 * we would have two strings, each prefixed by its size.
1490 */
1491 int consumed = sizeof(*augmented_arg) + augmented_arg->size;
75d1e306 1492
016f327c 1493 arg->augmented.args = ((void *)arg->augmented.args) + consumed;
8195168e 1494 arg->augmented.size -= consumed;
75d1e306 1495
8195168e 1496 return printed;
75d1e306
ACM		 1497}
1498
f994592d
ACM		 1499static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1500 struct syscall_arg *arg)
1501{
1502 unsigned long ptr = arg->val;
1503
75d1e306
ACM		 1504 if (arg->augmented.args)
1505 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1506
f994592d
ACM		 1507 if (!arg->trace->vfs_getname)
1508 return scnprintf(bf, size, "%#x", ptr);
1509
1510 thread__set_filename_pos(arg->thread, bf, ptr);
1511 return 0;
1512}
1513
ae9ed035
ACM		 1514static bool trace__filter_duration(struct trace *trace, double t)
1515{
1516 return t < (trace->duration_filter * NSEC_PER_MSEC);
1517}
1518
fd2b2975 1519static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
752fde44
ACM		 1520{
1521 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1522
60c907ab 1523 return fprintf(fp, "%10.3f ", ts);
752fde44
ACM		 1524}
1525
fd2b2975
ACM		 1526/*
1527 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1528 * using ttrace->entry_time for a thread that receives a sys_exit without
1529 * first having received a sys_enter ("poll" issued before tracing session
1530 * starts, lost sys_enter exit due to ring buffer overflow).
1531 */
1532static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1533{
1534 if (tstamp > 0)
1535 return __trace__fprintf_tstamp(trace, tstamp, fp);
1536
1537 return fprintf(fp, " ? ");
1538}
1539
de9f498d 1540static pid_t workload_pid = -1;
92ea0720
IR		 1541static volatile sig_atomic_t done = false;
1542static volatile sig_atomic_t interrupted = false;
f15eb531 1543
de9f498d 1544static void sighandler_interrupt(int sig __maybe_unused)
f15eb531 1545{
de9f498d
CD		 1546 done = interrupted = true;
1547}
1548
1549static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
1550 void *context __maybe_unused)
1551{
1552 if (info->si_pid == workload_pid)
1553 done = true;
f15eb531
NK		 1554}
1555
6dcbd212 1556static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
752fde44 1557{
6dcbd212 1558 size_t printed = 0;
752fde44 1559
50c95cbd
ACM		 1560 if (trace->multiple_threads) {
1561 if (trace->show_comm)
1902efe7 1562 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
ee84a303 1563 printed += fprintf(fp, "%d ", thread__tid(thread));
50c95cbd 1564 }
752fde44
ACM		 1565
1566 return printed;
1567}
1568
6dcbd212
ACM		 1569static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1570 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1571{
b036146f
ACM		 1572 size_t printed = 0;
1573
1574 if (trace->show_tstamp)
1575 printed = trace__fprintf_tstamp(trace, tstamp, fp);
42e4a52d
ACM		 1576 if (trace->show_duration)
1577 printed += fprintf_duration(duration, duration_calculated, fp);
6dcbd212
ACM		 1578 return printed + trace__fprintf_comm_tid(trace, thread, fp);
1579}
1580
c24ff998 1581static int trace__process_event(struct trace *trace, struct machine *machine,
162f0bef 1582 union perf_event *event, struct perf_sample *sample)
752fde44
ACM		 1583{
1584 int ret = 0;
1585
1586 switch (event->header.type) {
1587 case PERF_RECORD_LOST:
c24ff998 1588 color_fprintf(trace->output, PERF_COLOR_RED,
752fde44 1589 "LOST %" PRIu64 " events!\n", event->lost.lost);
162f0bef 1590 ret = machine__process_lost_event(machine, event, sample);
3ed5ca2e 1591 break;
752fde44 1592 default:
162f0bef 1593 ret = machine__process_event(machine, event, sample);
752fde44
ACM		 1594 break;
1595 }
1596
1597 return ret;
1598}
1599
c24ff998 1600static int trace__tool_process(struct perf_tool *tool,
752fde44 1601 union perf_event *event,
162f0bef 1602 struct perf_sample *sample,
752fde44
ACM		 1603 struct machine *machine)
1604{
c24ff998 1605 struct trace *trace = container_of(tool, struct trace, tool);
162f0bef 1606 return trace__process_event(trace, machine, event, sample);
752fde44
ACM		 1607}
1608
caf8a0d0
ACM		 1609static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1610{
1611 struct machine *machine = vmachine;
1612
1613 if (machine->kptr_restrict_warned)
1614 return NULL;
1615
1616 if (symbol_conf.kptr_restrict) {
1617 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
d06e5fad 1618 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
caf8a0d0
ACM		 1619 "Kernel samples will not be resolved.\n");
1620 machine->kptr_restrict_warned = true;
1621 return NULL;
1622 }
1623
1624 return machine__resolve_kernel_addr(vmachine, addrp, modp);
1625}
1626
63503dba 1627static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
752fde44 1628{
0a7e6d1b 1629 int err = symbol__init(NULL);
752fde44
ACM		 1630
1631 if (err)
1632 return err;
1633
8fb598e5
DA		 1634 trace->host = machine__new_host();
1635 if (trace->host == NULL)
1636 return -ENOMEM;
752fde44 1637
cbd5c178
AV		 1638 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1639 if (err < 0)
1640 goto out;
706c3da4 1641
a33fbd56 1642 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
84111b9c
NK		 1643 evlist->core.threads, trace__tool_process,
1644 true, false, 1);
cbd5c178 1645out:
752fde44
ACM		 1646 if (err)
1647 symbol__exit();
1648
1649 return err;
1650}
1651
33974a41
AV		 1652static void trace__symbols__exit(struct trace *trace)
1653{
1654 machine__exit(trace->host);
1655 trace->host = NULL;
1656
1657 symbol__exit();
1658}
1659
5e58fcfa 1660static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
13d4ff3e 1661{
5e58fcfa 1662 int idx;
13d4ff3e 1663
eadcab4c 1664 if (nr_args == RAW_SYSCALL_ARGS_NUM && sc->fmt && sc->fmt->nr_args != 0)
332337da
ACM		 1665 nr_args = sc->fmt->nr_args;
1666
5e58fcfa 1667 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
82d4a110 1668 if (sc->arg_fmt == NULL)
13d4ff3e
ACM		 1669 return -1;
1670
5e58fcfa
ACM		 1671 for (idx = 0; idx < nr_args; ++idx) {
1672 if (sc->fmt)
82d4a110 1673 sc->arg_fmt[idx] = sc->fmt->arg[idx];
5e58fcfa 1674 }
82d4a110 1675
5e58fcfa
ACM		 1676 sc->nr_args = nr_args;
1677 return 0;
1678}
1679
60995604 1680static const struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {
df604bfd
ACM		 1681 { .name = "msr", .scnprintf = SCA_X86_MSR, .strtoul = STUL_X86_MSR, },
1682 { .name = "vector", .scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, },
5d88099b
ACM		 1683};
1684
1685static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)
1686{
1687 const struct syscall_arg_fmt *fmt = fmtp;
1688 return strcmp(name, fmt->name);
1689}
1690
60995604
IR		 1691static const struct syscall_arg_fmt *
1692__syscall_arg_fmt__find_by_name(const struct syscall_arg_fmt *fmts, const int nmemb,
1693 const char *name)
5d88099b
ACM		 1694{
1695 return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);
1696}
1697
60995604 1698static const struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)
5d88099b
ACM		 1699{
1700 const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);
1701 return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);
1702}
1703
8d1d4ff5
ACM		 1704static struct tep_format_field *
1705syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field)
5e58fcfa 1706{
8d1d4ff5
ACM		 1707 struct tep_format_field *last_field = NULL;
1708 int len;
5e58fcfa 1709
8d1d4ff5 1710 for (; field; field = field->next, ++arg) {
7a983a0f
ACM		 1711 last_field = field;
1712
8d1d4ff5 1713 if (arg->scnprintf)
5e58fcfa 1714 continue;
1f115cb7 1715
dea87bfb
ACM		 1716 len = strlen(field->name);
1717
82d4a110 1718 if (strcmp(field->type, "const char *") == 0 &&
dea87bfb
ACM		 1719 ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1720 strstr(field->name, "path") != NULL))
8d1d4ff5 1721 arg->scnprintf = SCA_FILENAME;
ce05539f 1722 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
8d1d4ff5 1723 arg->scnprintf = SCA_PTR;
d1d438a3 1724 else if (strcmp(field->type, "pid_t") == 0)
8d1d4ff5 1725 arg->scnprintf = SCA_PID;
ba2f22cf 1726 else if (strcmp(field->type, "umode_t") == 0)
8d1d4ff5 1727 arg->scnprintf = SCA_MODE_T;
c5e006cd 1728 else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) {
9597945d
ACM		 1729 arg->scnprintf = SCA_CHAR_ARRAY;
1730 arg->nr_entries = field->arraylen;
1731 } else if ((strcmp(field->type, "int") == 0 ||
b6565c90
ACM		 1732 strcmp(field->type, "unsigned int") == 0 ||
1733 strcmp(field->type, "long") == 0) &&
dea87bfb 1734 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
b6565c90
ACM		 1735 /*
1736 * /sys/kernel/tracing/events/syscalls/sys_enter*
818448e9 1737 * grep -E 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
b6565c90
ACM		 1738 * 65 int
1739 * 23 unsigned int
1740 * 7 unsigned long
1741 */
8d1d4ff5 1742 arg->scnprintf = SCA_FD;
60995604
IR		 1743 } else {
1744 const struct syscall_arg_fmt *fmt =
1745 syscall_arg_fmt__find_by_name(field->name);
5d88099b 1746
3f41b778 1747 if (fmt) {
5d88099b 1748 arg->scnprintf = fmt->scnprintf;
3f41b778
ACM		 1749 arg->strtoul = fmt->strtoul;
1750 }
b6565c90 1751 }
13d4ff3e
ACM		 1752 }
1753
8d1d4ff5
ACM		 1754 return last_field;
1755}
1756
1757static int syscall__set_arg_fmts(struct syscall *sc)
1758{
1759 struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args);
1760
7a983a0f
ACM		 1761 if (last_field)
1762 sc->args_size = last_field->offset + last_field->size;
1763
13d4ff3e
ACM		 1764 return 0;
1765}
1766
514f1c67
ACM		 1767static int trace__read_syscall_info(struct trace *trace, int id)
1768{
1769 char tp_name[128];
1770 struct syscall *sc;
fd0db102 1771 const char *name = syscalltbl__name(trace->sctbl, id);
3a531260 1772
d21cb73a 1773#ifdef HAVE_SYSCALL_TABLE_SUPPORT
30a910d7 1774 if (trace->syscalls.table == NULL) {
3e70008a 1775 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
30a910d7 1776 if (trace->syscalls.table == NULL)
5d2bd889 1777 return -ENOMEM;
514f1c67 1778 }
d21cb73a
ACM		 1779#else
1780 if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) {
1781 // When using libaudit we don't know beforehand what is the max syscall id
1782 struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1783
1784 if (table == NULL)
1785 return -ENOMEM;
514f1c67 1786
f3013f7e
JS		 1787 // Need to memset from offset 0 and +1 members if brand new
1788 if (trace->syscalls.table == NULL)
1789 memset(table, 0, (id + 1) * sizeof(*sc));
1790 else
1791 memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc));
d21cb73a
ACM		 1792
1793 trace->syscalls.table = table;
1794 trace->sctbl->syscalls.max_id = id;
1795 }
1796#endif
514f1c67 1797 sc = trace->syscalls.table + id;
b8b1033f 1798 if (sc->nonexistent)
d4223e17 1799 return -EEXIST;
2ae3a312 1800
b8b1033f
ACM		 1801 if (name == NULL) {
1802 sc->nonexistent = true;
d4223e17 1803 return -EEXIST;
b8b1033f
ACM		 1804 }
1805
1806 sc->name = name;
3a531260 1807 sc->fmt = syscall_fmt__find(sc->name);
514f1c67 1808
aec1930b 1809 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
97978b3e 1810 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
aec1930b 1811
8dd2a131 1812 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
aec1930b 1813 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
97978b3e 1814 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
aec1930b 1815 }
514f1c67 1816
03e9a5d8
LY		 1817 /*
1818 * Fails to read trace point format via sysfs node, so the trace point
1819 * doesn't exist. Set the 'nonexistent' flag as true.
1820 */
1821 if (IS_ERR(sc->tp_format)) {
1822 sc->nonexistent = true;
1823 return PTR_ERR(sc->tp_format);
1824 }
1825
eadcab4c
LY		 1826 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ?
1827 RAW_SYSCALL_ARGS_NUM : sc->tp_format->format.nr_fields))
5d2bd889 1828 return -ENOMEM;
5e58fcfa 1829
f208bd8d 1830 sc->args = sc->tp_format->format.fields;
c42de706
TS		 1831 /*
1832 * We need to check and discard the first variable '__syscall_nr'
1833 * or 'nr' that mean the syscall number. It is needless here.
1834 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1835 */
1836 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
f208bd8d
ACM		 1837 sc->args = sc->args->next;
1838 --sc->nr_args;
1839 }
1840
5089f20e 1841 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
6a648b53 1842 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
5089f20e 1843
13d4ff3e 1844 return syscall__set_arg_fmts(sc);
514f1c67
ACM		 1845}
1846
794bca26 1847static int evsel__init_tp_arg_scnprintf(struct evsel *evsel)
947b843c 1848{
8b913df5 1849 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
947b843c 1850
8b913df5
ACM		 1851 if (fmt != NULL) {
1852 syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields);
947b843c
ACM		 1853 return 0;
1854 }
1855
1856 return -ENOMEM;
1857}
1858
83e69b92
ACM		 1859static int intcmp(const void *a, const void *b)
1860{
1861 const int *one = a, *another = b;
1862
1863 return *one - *another;
1864}
1865
d0cc439b
ACM		 1866static int trace__validate_ev_qualifier(struct trace *trace)
1867{
a4066d64 1868 int err = 0;
04c41bcb 1869 bool printed_invalid_prefix = false;
d0cc439b 1870 struct str_node *pos;
99f26f85 1871 size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
d0cc439b 1872
99f26f85 1873 trace->ev_qualifier_ids.entries = malloc(nr_allocated *
8b3ce757
ACM		 1874 sizeof(trace->ev_qualifier_ids.entries[0]));
1875
1876 if (trace->ev_qualifier_ids.entries == NULL) {
1877 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1878 trace->output);
1879 err = -EINVAL;
1880 goto out;
1881 }
1882
602a1f4d 1883 strlist__for_each_entry(pos, trace->ev_qualifier) {
d0cc439b 1884 const char *sc = pos->s;
27702bcf 1885 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
d0cc439b 1886
8b3ce757 1887 if (id < 0) {
27702bcf
ACM		 1888 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1889 if (id >= 0)
1890 goto matches;
1891
04c41bcb
ACM		 1892 if (!printed_invalid_prefix) {
1893 pr_debug("Skipping unknown syscalls: ");
1894 printed_invalid_prefix = true;
d0cc439b 1895 } else {
04c41bcb 1896 pr_debug(", ");
d0cc439b
ACM		 1897 }
1898
04c41bcb
ACM		 1899 pr_debug("%s", sc);
1900 continue;
d0cc439b 1901 }
27702bcf 1902matches:
99f26f85 1903 trace->ev_qualifier_ids.entries[nr_used++] = id;
27702bcf
ACM		 1904 if (match_next == -1)
1905 continue;
1906
1907 while (1) {
1908 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1909 if (id < 0)
1910 break;
99f26f85 1911 if (nr_allocated == nr_used) {
27702bcf
ACM		 1912 void *entries;
1913
1914 nr_allocated += 8;
1915 entries = realloc(trace->ev_qualifier_ids.entries,
1916 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1917 if (entries == NULL) {
1918 err = -ENOMEM;
1919 fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1920 goto out_free;
1921 }
1922 trace->ev_qualifier_ids.entries = entries;
1923 }
99f26f85 1924 trace->ev_qualifier_ids.entries[nr_used++] = id;
27702bcf 1925 }
d0cc439b
ACM		 1926 }
1927
99f26f85 1928 trace->ev_qualifier_ids.nr = nr_used;
83e69b92 1929 qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
8b3ce757 1930out:
04c41bcb
ACM		 1931 if (printed_invalid_prefix)
1932 pr_debug("\n");
d0cc439b 1933 return err;
04c41bcb
ACM		 1934out_free:
1935 zfree(&trace->ev_qualifier_ids.entries);
1936 trace->ev_qualifier_ids.nr = 0;
1937 goto out;
d0cc439b
ACM		 1938}
1939
3803a229
ACM		 1940static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1941{
1942 bool in_ev_qualifier;
1943
1944 if (trace->ev_qualifier_ids.nr == 0)
1945 return true;
1946
1947 in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1948 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1949
1950 if (in_ev_qualifier)
1951 return !trace->not_ev_qualifier;
1952
1953 return trace->not_ev_qualifier;
1954}
1955
55d43bca
DA		 1956/*
1957 * args is to be interpreted as a series of longs but we need to handle
1958 * 8-byte unaligned accesses. args points to raw_data within the event
1959 * and raw_data is guaranteed to be 8-byte unaligned because it is
1960 * preceded by raw_size which is a u32. So we need to copy args to a temp
1961 * variable to read it. Most notably this avoids extended load instructions
1962 * on unaligned addresses
1963 */
325f5091 1964unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
f9f83b33
ACM		 1965{
1966 unsigned long val;
325f5091 1967 unsigned char *p = arg->args + sizeof(unsigned long) * idx;
f9f83b33
ACM		 1968
1969 memcpy(&val, p, sizeof(val));
1970 return val;
1971}
1972
c51bdfec
ACM		 1973static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1974 struct syscall_arg *arg)
1975{
1976 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1977 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1978
1979 return scnprintf(bf, size, "arg%d: ", arg->idx);
1980}
1981
496fd346
ACM		 1982/*
1983 * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1984 * as mount 'flags' argument that needs ignoring some magic flag, see comment
1985 * in tools/perf/trace/beauty/mount_flags.c
1986 */
3e0c9b2c 1987static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)
496fd346 1988{
3e0c9b2c
ACM		 1989 if (fmt && fmt->mask_val)
1990 return fmt->mask_val(arg, val);
496fd346
ACM		 1991
1992 return val;
1993}
1994
3e0c9b2c
ACM		 1995static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,
1996 struct syscall_arg *arg, unsigned long val)
d032d79e 1997{
3e0c9b2c 1998 if (fmt && fmt->scnprintf) {
d032d79e 1999 arg->val = val;
3e0c9b2c
ACM		 2000 if (fmt->parm)
2001 arg->parm = fmt->parm;
2002 return fmt->scnprintf(bf, size, arg);
d032d79e
ACM		 2003 }
2004 return scnprintf(bf, size, "%ld", val);
2005}
2006
752fde44 2007static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
7a983a0f
ACM		 2008 unsigned char *args, void *augmented_args, int augmented_args_size,
2009 struct trace *trace, struct thread *thread)
514f1c67 2010{
514f1c67 2011 size_t printed = 0;
55d43bca 2012 unsigned long val;
d032d79e
ACM		 2013 u8 bit = 1;
2014 struct syscall_arg arg = {
2015 .args = args,
7a983a0f
ACM		 2016 .augmented = {
2017 .size = augmented_args_size,
2018 .args = augmented_args,
2019 },
d032d79e
ACM		 2020 .idx = 0,
2021 .mask = 0,
2022 .trace = trace,
2023 .thread = thread,
c65c83ff 2024 .show_string_prefix = trace->show_string_prefix,
d032d79e 2025 };
84486caa
ACM		 2026 struct thread_trace *ttrace = thread__priv(thread);
2027
2028 /*
2029 * Things like fcntl will set this in its 'cmd' formatter to pick the
2030 * right formatter for the return value (an fd? file flags?), which is
2031 * not needed for syscalls that always return a given type, say an fd.
2032 */
2033 ttrace->ret_scnprintf = NULL;
514f1c67 2034
f208bd8d 2035 if (sc->args != NULL) {
2c92f982 2036 struct tep_format_field *field;
6e7eeb51 2037
f208bd8d 2038 for (field = sc->args; field;
01533e97
ACM		 2039 field = field->next, ++arg.idx, bit <<= 1) {
2040 if (arg.mask & bit)
6e7eeb51 2041 continue;
55d43bca 2042
888ca854 2043 arg.fmt = &sc->arg_fmt[arg.idx];
f9f83b33 2044 val = syscall_arg__val(&arg, arg.idx);
496fd346
ACM		 2045 /*
2046 * Some syscall args need some mask, most don't and
2047 * return val untouched.
2048 */
3e0c9b2c 2049 val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);
55d43bca 2050
4aa58232
ACM		 2051 /*
2052 * Suppress this argument if its value is zero and
2053 * and we don't have a string associated in an
2054 * strarray for it.
2055 */
55d43bca 2056 if (val == 0 &&
e7c634fc 2057 !trace->show_zeros &&
82d4a110 2058 !(sc->arg_fmt &&
d47737d5
ACM		 2059 (sc->arg_fmt[arg.idx].show_zero ||
2060 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
82d4a110
ACM		 2061 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
2062 sc->arg_fmt[arg.idx].parm))
22ae5cf1
ACM		 2063 continue;
2064
9d6dc178
ACM		 2065 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2066
2067 if (trace->show_arg_names)
2068 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2069
3e0c9b2c
ACM		 2070 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],
2071 bf + printed, size - printed, &arg, val);
514f1c67 2072 }
4c4d6e51
ACM		 2073 } else if (IS_ERR(sc->tp_format)) {
2074 /*
2075 * If we managed to read the tracepoint /format file, then we
2076 * may end up not having any args, like with gettid(), so only
2077 * print the raw args when we didn't manage to read it.
2078 */
332337da 2079 while (arg.idx < sc->nr_args) {
d032d79e
ACM		 2080 if (arg.mask & bit)
2081 goto next_arg;
2082 val = syscall_arg__val(&arg, arg.idx);
c51bdfec
ACM		 2083 if (printed)
2084 printed += scnprintf(bf + printed, size - printed, ", ");
2085 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
3e0c9b2c 2086 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);
d032d79e
ACM		 2087next_arg:
2088 ++arg.idx;
2089 bit <<= 1;
514f1c67
ACM		 2090 }
2091 }
2092
2093 return printed;
2094}
2095
32dcd021 2096typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
0c82adcf 2097 union perf_event *event,
ba3d7dee
ACM		 2098 struct perf_sample *sample);
2099
2100static struct syscall *trace__syscall_info(struct trace *trace,
32dcd021 2101 struct evsel *evsel, int id)
ba3d7dee 2102{
5d2bd889 2103 int err = 0;
ba3d7dee
ACM		 2104
2105 if (id < 0) {
adaa18bf
ACM		 2106
2107 /*
2108 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
2109 * before that, leaving at a higher verbosity level till that is
2110 * explained. Reproduced with plain ftrace with:
2111 *
2112 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
2113 * grep "NR -1 " /t/trace_pipe
2114 *
2115 * After generating some load on the machine.
2116 */
2117 if (verbose > 1) {
2118 static u64 n;
2119 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
8ab2e96d 2120 id, evsel__name(evsel), ++n);
adaa18bf 2121 }
ba3d7dee
ACM		 2122 return NULL;
2123 }
2124
b8b1033f
ACM		 2125 err = -EINVAL;
2126
d21cb73a
ACM		 2127#ifdef HAVE_SYSCALL_TABLE_SUPPORT
2128 if (id > trace->sctbl->syscalls.max_id) {
2129#else
2130 if (id >= trace->sctbl->syscalls.max_id) {
2131 /*
2132 * With libaudit we don't know beforehand what is the max_id,
2133 * so we let trace__read_syscall_info() figure that out as we
2134 * go on reading syscalls.
2135 */
2136 err = trace__read_syscall_info(trace, id);
2137 if (err)
2138#endif
ba3d7dee 2139 goto out_cant_read;
d21cb73a 2140 }
ba3d7dee 2141
30a910d7
ACM		 2142 if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
2143 (err = trace__read_syscall_info(trace, id)) != 0)
ba3d7dee
ACM		 2144 goto out_cant_read;
2145
03e9a5d8 2146 if (trace->syscalls.table && trace->syscalls.table[id].nonexistent)
b8b1033f 2147 goto out_cant_read;
b8b1033f 2148
ba3d7dee
ACM		 2149 return &trace->syscalls.table[id];
2150
2151out_cant_read:
bb963e16 2152 if (verbose > 0) {
5d2bd889
ACM		 2153 char sbuf[STRERR_BUFSIZE];
2154 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
30a910d7 2155 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
7c304ee0
ACM		 2156 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
2157 fputs(" information\n", trace->output);
2158 }
ba3d7dee
ACM		 2159 return NULL;
2160}
2161
8eded45f
ACM		 2162struct syscall_stats {
2163 struct stats stats;
2164 u64 nr_failures;
b88b14db
ACM		 2165 int max_errno;
2166 u32 *errnos;
8eded45f
ACM		 2167};
2168
b88b14db
ACM		 2169static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace,
2170 int id, struct perf_sample *sample, long err, bool errno_summary)
bf2575c1
DA		 2171{
2172 struct int_node *inode;
8eded45f 2173 struct syscall_stats *stats;
bf2575c1
DA		 2174 u64 duration = 0;
2175
2176 inode = intlist__findnew(ttrace->syscall_stats, id);
2177 if (inode == NULL)
2178 return;
2179
2180 stats = inode->priv;
2181 if (stats == NULL) {
e3e7572f 2182 stats = zalloc(sizeof(*stats));
bf2575c1
DA		 2183 if (stats == NULL)
2184 return;
8eded45f 2185
8eded45f 2186 init_stats(&stats->stats);
bf2575c1
DA		 2187 inode->priv = stats;
2188 }
2189
2190 if (ttrace->entry_time && sample->time > ttrace->entry_time)
2191 duration = sample->time - ttrace->entry_time;
2192
8eded45f
ACM		 2193 update_stats(&stats->stats, duration);
2194
b88b14db 2195 if (err < 0) {
8eded45f 2196 ++stats->nr_failures;
b88b14db
ACM		 2197
2198 if (!errno_summary)
2199 return;
2200
2201 err = -err;
2202 if (err > stats->max_errno) {
2203 u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32));
2204
2205 if (new_errnos) {
2206 memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32));
2207 } else {
2208 pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n",
ee84a303
IR		 2209 thread__comm_str(thread), thread__pid(thread),
2210 thread__tid(thread));
b88b14db
ACM		 2211 return;
2212 }
2213
2214 stats->errnos = new_errnos;
2215 stats->max_errno = err;
2216 }
2217
2218 ++stats->errnos[err - 1];
2219 }
bf2575c1
DA		 2220}
2221
522283fe 2222static int trace__printf_interrupted_entry(struct trace *trace)
e596663e
ACM		 2223{
2224 struct thread_trace *ttrace;
e596663e 2225 size_t printed;
172bf02d 2226 int len;
e596663e 2227
0a6545bd 2228 if (trace->failure_only || trace->current == NULL)
e596663e
ACM		 2229 return 0;
2230
2231 ttrace = thread__priv(trace->current);
2232
2233 if (!ttrace->entry_pending)
2234 return 0;
2235
522283fe 2236 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
172bf02d
ACM		 2237 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
2238
2239 if (len < trace->args_alignment - 4)
2240 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
e596663e 2241
172bf02d
ACM		 2242 printed += fprintf(trace->output, " ...\n");
2243
2244 ttrace->entry_pending = false;
5067a8cd
ACM		 2245 ++trace->nr_events_printed;
2246
e596663e
ACM		 2247 return printed;
2248}
2249
32dcd021 2250static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
591421e1
ACM		 2251 struct perf_sample *sample, struct thread *thread)
2252{
2253 int printed = 0;
2254
2255 if (trace->print_sample) {
2256 double ts = (double)sample->time / NSEC_PER_MSEC;
2257
2258 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
8ab2e96d 2259 evsel__name(evsel), ts,
591421e1
ACM		 2260 thread__comm_str(thread),
2261 sample->pid, sample->tid, sample->cpu);
2262 }
2263
2264 return printed;
2265}
2266
b9b6a2ea 2267static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
8a041f86
ACM		 2268{
2269 void *augmented_args = NULL;
cd26ea6d
ACM		 2270 /*
2271 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
b9b6a2ea
ACM		 2272 * and there we get all 6 syscall args plus the tracepoint common fields
2273 * that gets calculated at the start and the syscall_nr (another long).
2274 * So we check if that is the case and if so don't look after the
2275 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
2276 * which is fixed.
cd26ea6d
ACM		 2277 *
2278 * We'll revisit this later to pass s->args_size to the BPF augmenter
2279 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
2280 * copies only what we need for each syscall, like what happens when we
2281 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
2282 * traffic to just what is needed for each syscall.
2283 */
b9b6a2ea 2284 int args_size = raw_augmented_args_size ?: sc->args_size;
8a041f86 2285
cd26ea6d 2286 *augmented_args_size = sample->raw_size - args_size;
8a041f86 2287 if (*augmented_args_size > 0)
cd26ea6d 2288 augmented_args = sample->raw_data + args_size;
8a041f86
ACM		 2289
2290 return augmented_args;
2291}
2292
f2ebf8ff
RM		 2293static void syscall__exit(struct syscall *sc)
2294{
2295 if (!sc)
2296 return;
2297
9997d5dd 2298 zfree(&sc->arg_fmt);
f2ebf8ff
RM		 2299}
2300
32dcd021 2301static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
0c82adcf 2302 union perf_event *event __maybe_unused,
ba3d7dee
ACM		 2303 struct perf_sample *sample)
2304{
752fde44 2305 char *msg;
ba3d7dee 2306 void *args;
4b8a240e 2307 int printed = 0;
2ae3a312 2308 struct thread *thread;
b91fc39f 2309 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
db2da3f8
ACM		 2310 int augmented_args_size = 0;
2311 void *augmented_args = NULL;
bf2575c1 2312 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2ae3a312
ACM		 2313 struct thread_trace *ttrace;
2314
2315 if (sc == NULL)
2316 return -1;
ba3d7dee 2317
8fb598e5 2318 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
c24ff998 2319 ttrace = thread__trace(thread, trace->output);
2ae3a312 2320 if (ttrace == NULL)
b91fc39f 2321 goto out_put;
ba3d7dee 2322
591421e1
ACM		 2323 trace__fprintf_sample(trace, evsel, sample, thread);
2324
77170988 2325 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
752fde44
ACM		 2326
2327 if (ttrace->entry_str == NULL) {
e4d44e83 2328 ttrace->entry_str = malloc(trace__entry_str_size);
752fde44 2329 if (!ttrace->entry_str)
b91fc39f 2330 goto out_put;
752fde44
ACM		 2331 }
2332
5cf9c84e 2333 if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
522283fe 2334 trace__printf_interrupted_entry(trace);
db2da3f8
ACM		 2335 /*
2336 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2337 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2338 * this breaks syscall__augmented_args() check for augmented args, as we calculate
2339 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2340 * so when handling, say the openat syscall, we end up getting 6 args for the
2341 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2342 * thinking that the extra 2 u64 args are the augmented filename, so just check
2343 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2344 */
2345 if (evsel != trace->syscalls.events.sys_enter)
b9b6a2ea 2346 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
752fde44
ACM		 2347 ttrace->entry_time = sample->time;
2348 msg = ttrace->entry_str;
e4d44e83 2349 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
752fde44 2350
e4d44e83 2351 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
db2da3f8 2352 args, augmented_args, augmented_args_size, trace, thread);
752fde44 2353
5089f20e 2354 if (sc->is_exit) {
0a6545bd 2355 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
4b8a240e
ACM		 2356 int alignment = 0;
2357
fd2b2975 2358 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
4b8a240e
ACM		 2359 printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2360 if (trace->args_alignment > printed)
2361 alignment = trace->args_alignment - printed;
2362 fprintf(trace->output, "%*s= ?\n", alignment, " ");
ae9ed035 2363 }
7f4f8001 2364 } else {
752fde44 2365 ttrace->entry_pending = true;
7f4f8001
ACM		 2366 /* See trace__vfs_getname & trace__sys_exit */
2367 ttrace->filename.pending_open = false;
2368 }
ba3d7dee 2369
f3b623b8
ACM		 2370 if (trace->current != thread) {
2371 thread__put(trace->current);
2372 trace->current = thread__get(thread);
2373 }
b91fc39f
ACM		 2374 err = 0;
2375out_put:
2376 thread__put(thread);
2377 return err;
ba3d7dee
ACM		 2378}
2379
32dcd021 2380static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
a98392bb
ACM		 2381 struct perf_sample *sample)
2382{
a98392bb
ACM		 2383 struct thread_trace *ttrace;
2384 struct thread *thread;
f3acd886
ACM		 2385 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2386 struct syscall *sc = trace__syscall_info(trace, evsel, id);
a98392bb 2387 char msg[1024];
7a983a0f
ACM		 2388 void *args, *augmented_args = NULL;
2389 int augmented_args_size;
a98392bb 2390
a98392bb
ACM		 2391 if (sc == NULL)
2392 return -1;
2393
2394 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2395 ttrace = thread__trace(thread, trace->output);
2396 /*
2397 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2398 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2399 */
2400 if (ttrace == NULL)
2401 goto out_put;
2402
f3acd886 2403 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
b9b6a2ea 2404 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
7a983a0f 2405 syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
a98392bb
ACM		 2406 fprintf(trace->output, "%s", msg);
2407 err = 0;
2408out_put:
2409 thread__put(thread);
2410 return err;
2411}
2412
32dcd021 2413static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
5cf9c84e
ACM		 2414 struct perf_sample *sample,
2415 struct callchain_cursor *cursor)
202ff968
ACM		 2416{
2417 struct addr_location al;
1fc632ce
JO		 2418 int max_stack = evsel->core.attr.sample_max_stack ?
2419 evsel->core.attr.sample_max_stack :
3a9e9a47 2420 trace->max_stack;
0dd5041c 2421 int err = -1;
5cf9c84e 2422
0dd5041c 2423 addr_location__init(&al);
4291bf5c 2424 if (machine__resolve(trace->host, &al, sample) < 0)
0dd5041c 2425 goto out;
5cf9c84e 2426
4291bf5c 2427 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
0dd5041c
IR		 2428out:
2429 addr_location__exit(&al);
4291bf5c 2430 return err;
5cf9c84e
ACM		 2431}
2432
2433static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2434{
202ff968 2435 /* TODO: user-configurable print_opts */
e20ab86e
ACM		 2436 const unsigned int print_opts = EVSEL__PRINT_SYM |
2437 EVSEL__PRINT_DSO |
2438 EVSEL__PRINT_UNKNOWN_AS_ADDR;
202ff968 2439
8ab12a20 2440 return sample__fprintf_callchain(sample, 38, print_opts, get_tls_callchain_cursor(), symbol_conf.bt_stop_list, trace->output);
202ff968
ACM		 2441}
2442
32dcd021 2443static const char *errno_to_name(struct evsel *evsel, int err)
092bd3cd 2444{
6e6d1d65 2445 struct perf_env *env = evsel__env(evsel);
092bd3cd
HB		 2446 const char *arch_name = perf_env__arch(env);
2447
2448 return arch_syscalls__strerrno(arch_name, err);
2449}
2450
32dcd021 2451static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
0c82adcf 2452 union perf_event *event __maybe_unused,
ba3d7dee
ACM		 2453 struct perf_sample *sample)
2454{
2c82c3ad 2455 long ret;
60c907ab 2456 u64 duration = 0;
fd2b2975 2457 bool duration_calculated = false;
2ae3a312 2458 struct thread *thread;
4b8a240e
ACM		 2459 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2460 int alignment = trace->args_alignment;
bf2575c1 2461 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2ae3a312
ACM		 2462 struct thread_trace *ttrace;
2463
2464 if (sc == NULL)
2465 return -1;
ba3d7dee 2466
8fb598e5 2467 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
c24ff998 2468 ttrace = thread__trace(thread, trace->output);
2ae3a312 2469 if (ttrace == NULL)
b91fc39f 2470 goto out_put;
ba3d7dee 2471
591421e1
ACM		 2472 trace__fprintf_sample(trace, evsel, sample, thread);
2473
77170988 2474 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
ba3d7dee 2475
8eded45f 2476 if (trace->summary)
b88b14db 2477 thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary);
8eded45f 2478
79d725cd 2479 if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
7f4f8001
ACM		 2480 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2481 ttrace->filename.pending_open = false;
c522739d
ACM		 2482 ++trace->stats.vfs_getname;
2483 }
2484
ae9ed035 2485 if (ttrace->entry_time) {
60c907ab 2486 duration = sample->time - ttrace->entry_time;
ae9ed035
ACM		 2487 if (trace__filter_duration(trace, duration))
2488 goto out;
fd2b2975 2489 duration_calculated = true;
ae9ed035
ACM		 2490 } else if (trace->duration_filter)
2491 goto out;
60c907ab 2492
5cf9c84e 2493 if (sample->callchain) {
8ab12a20
IR		 2494 struct callchain_cursor *cursor = get_tls_callchain_cursor();
2495
2496 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
5cf9c84e 2497 if (callchain_ret == 0) {
8ab12a20 2498 if (cursor->nr < trace->min_stack)
5cf9c84e
ACM		 2499 goto out;
2500 callchain_ret = 1;
2501 }
2502 }
2503
0a6545bd 2504 if (trace->summary_only || (ret >= 0 && trace->failure_only))
fd2eabaf
DA		 2505 goto out;
2506
fd2b2975 2507 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
752fde44
ACM		 2508
2509 if (ttrace->entry_pending) {
4b8a240e 2510 printed = fprintf(trace->output, "%s", ttrace->entry_str);
752fde44 2511 } else {
ac6e022c 2512 printed += fprintf(trace->output, " ... [");
c24ff998 2513 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
ac6e022c
ACM		 2514 printed += 9;
2515 printed += fprintf(trace->output, "]: %s()", sc->name);
752fde44
ACM		 2516 }
2517
4b8a240e
ACM		 2518 printed++; /* the closing ')' */
2519
2520 if (alignment > printed)
2521 alignment -= printed;
2522 else
2523 alignment = 0;
2524
2525 fprintf(trace->output, ")%*s= ", alignment, " ");
2526
da3c9a44 2527 if (sc->fmt == NULL) {
1f63139c
ACM		 2528 if (ret < 0)
2529 goto errno_print;
da3c9a44 2530signed_print:
4b8a240e 2531 fprintf(trace->output, "%ld", ret);
1f63139c
ACM		 2532 } else if (ret < 0) {
2533errno_print: {
942a91ed 2534 char bf[STRERR_BUFSIZE];
c8b5f2c9 2535 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
092bd3cd 2536 *e = errno_to_name(evsel, -ret);
ba3d7dee 2537
2c83dfae 2538 fprintf(trace->output, "-1 %s (%s)", e, emsg);
1f63139c 2539 }
da3c9a44 2540 } else if (ret == 0 && sc->fmt->timeout)
2c83dfae 2541 fprintf(trace->output, "0 (Timeout)");
84486caa
ACM		 2542 else if (ttrace->ret_scnprintf) {
2543 char bf[1024];
7ee57434
ACM		 2544 struct syscall_arg arg = {
2545 .val = ret,
2546 .thread = thread,
2547 .trace = trace,
2548 };
2549 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
84486caa 2550 ttrace->ret_scnprintf = NULL;
4b8a240e 2551 fprintf(trace->output, "%s", bf);
84486caa 2552 } else if (sc->fmt->hexret)
4b8a240e 2553 fprintf(trace->output, "%#lx", ret);
11c8e39f
ACM		 2554 else if (sc->fmt->errpid) {
2555 struct thread *child = machine__find_thread(trace->host, ret, ret);
2556
2557 if (child != NULL) {
4b8a240e 2558 fprintf(trace->output, "%ld", ret);
ee84a303 2559 if (thread__comm_set(child))
11c8e39f
ACM		 2560 fprintf(trace->output, " (%s)", thread__comm_str(child));
2561 thread__put(child);
2562 }
2563 } else
da3c9a44 2564 goto signed_print;
ba3d7dee 2565
c24ff998 2566 fputc('\n', trace->output);
566a0885 2567
5067a8cd
ACM		 2568 /*
2569 * We only consider an 'event' for the sake of --max-events a non-filtered
2570 * sys_enter + sys_exit and other tracepoint events.
2571 */
2572 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2573 interrupted = true;
2574
5cf9c84e
ACM		 2575 if (callchain_ret > 0)
2576 trace__fprintf_callchain(trace, sample);
2577 else if (callchain_ret < 0)
8ab2e96d 2578 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
ae9ed035 2579out:
752fde44 2580 ttrace->entry_pending = false;
b91fc39f
ACM		 2581 err = 0;
2582out_put:
2583 thread__put(thread);
2584 return err;
ba3d7dee
ACM		 2585}
2586
32dcd021 2587static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
0c82adcf 2588 union perf_event *event __maybe_unused,
c522739d
ACM		 2589 struct perf_sample *sample)
2590{
f994592d
ACM		 2591 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2592 struct thread_trace *ttrace;
2593 size_t filename_len, entry_str_len, to_move;
2594 ssize_t remaining_space;
2595 char *pos;
efc0cdc9 2596 const char *filename = evsel__rawptr(evsel, sample, "pathname");
f994592d
ACM		 2597
2598 if (!thread)
2599 goto out;
2600
2601 ttrace = thread__priv(thread);
2602 if (!ttrace)
ef65e96e 2603 goto out_put;
f994592d 2604
7f4f8001 2605 filename_len = strlen(filename);
39f0e7a8 2606 if (filename_len == 0)
ef65e96e 2607 goto out_put;
7f4f8001
ACM		 2608
2609 if (ttrace->filename.namelen < filename_len) {
2610 char *f = realloc(ttrace->filename.name, filename_len + 1);
2611
2612 if (f == NULL)
ef65e96e 2613 goto out_put;
7f4f8001
ACM		 2614
2615 ttrace->filename.namelen = filename_len;
2616 ttrace->filename.name = f;
2617 }
2618
2619 strcpy(ttrace->filename.name, filename);
2620 ttrace->filename.pending_open = true;
2621
f994592d 2622 if (!ttrace->filename.ptr)
ef65e96e 2623 goto out_put;
f994592d
ACM		 2624
2625 entry_str_len = strlen(ttrace->entry_str);
2626 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2627 if (remaining_space <= 0)
ef65e96e 2628 goto out_put;
f994592d 2629
f994592d
ACM		 2630 if (filename_len > (size_t)remaining_space) {
2631 filename += filename_len - remaining_space;
2632 filename_len = remaining_space;
2633 }
2634
2635 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2636 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2637 memmove(pos + filename_len, pos, to_move);
2638 memcpy(pos, filename, filename_len);
2639
2640 ttrace->filename.ptr = 0;
2641 ttrace->filename.entry_str_pos = 0;
ef65e96e
ACM		 2642out_put:
2643 thread__put(thread);
f994592d 2644out:
c522739d
ACM		 2645 return 0;
2646}
2647
32dcd021 2648static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
0c82adcf 2649 union perf_event *event __maybe_unused,
1302d88e
ACM		 2650 struct perf_sample *sample)
2651{
efc0cdc9 2652 u64 runtime = evsel__intval(evsel, sample, "runtime");
1302d88e 2653 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
8fb598e5 2654 struct thread *thread = machine__findnew_thread(trace->host,
314add6b
AH		 2655 sample->pid,
2656 sample->tid);
c24ff998 2657 struct thread_trace *ttrace = thread__trace(thread, trace->output);
1302d88e
ACM		 2658
2659 if (ttrace == NULL)
2660 goto out_dump;
2661
2662 ttrace->runtime_ms += runtime_ms;
2663 trace->runtime_ms += runtime_ms;
ef65e96e 2664out_put:
b91fc39f 2665 thread__put(thread);
1302d88e
ACM		 2666 return 0;
2667
2668out_dump:
c24ff998 2669 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
1302d88e 2670 evsel->name,
efc0cdc9
ACM		 2671 evsel__strval(evsel, sample, "comm"),
2672 (pid_t)evsel__intval(evsel, sample, "pid"),
1302d88e 2673 runtime,
efc0cdc9 2674 evsel__intval(evsel, sample, "vruntime"));
ef65e96e 2675 goto out_put;
1302d88e
ACM		 2676}
2677
923d0c9a
ACM		 2678static int bpf_output__printer(enum binary_printer_ops op,
2679 unsigned int val, void *extra __maybe_unused, FILE *fp)
1d6c9407 2680{
1d6c9407
WN		 2681 unsigned char ch = (unsigned char)val;
2682
2683 switch (op) {
2684 case BINARY_PRINT_CHAR_DATA:
923d0c9a 2685 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
1d6c9407
WN		 2686 case BINARY_PRINT_DATA_BEGIN:
2687 case BINARY_PRINT_LINE_BEGIN:
2688 case BINARY_PRINT_ADDR:
2689 case BINARY_PRINT_NUM_DATA:
2690 case BINARY_PRINT_NUM_PAD:
2691 case BINARY_PRINT_SEP:
2692 case BINARY_PRINT_CHAR_PAD:
2693 case BINARY_PRINT_LINE_END:
2694 case BINARY_PRINT_DATA_END:
2695 default:
2696 break;
2697 }
923d0c9a
ACM		 2698
2699 return 0;
1d6c9407
WN		 2700}
2701
2702static void bpf_output__fprintf(struct trace *trace,
2703 struct perf_sample *sample)
2704{
923d0c9a
ACM		 2705 binary__fprintf(sample->raw_data, sample->raw_size, 8,
2706 bpf_output__printer, NULL, trace->output);
5067a8cd 2707 ++trace->nr_events_printed;
1d6c9407
WN		 2708}
2709
f11b2803
ACM		 2710static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,
2711 struct thread *thread, void *augmented_args, int augmented_args_size)
2712{
2713 char bf[2048];
2714 size_t size = sizeof(bf);
2715 struct tep_format_field *field = evsel->tp_format->format.fields;
fecd9907 2716 struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel);
f11b2803
ACM		 2717 size_t printed = 0;
2718 unsigned long val;
2719 u8 bit = 1;
2720 struct syscall_arg syscall_arg = {
2721 .augmented = {
2722 .size = augmented_args_size,
2723 .args = augmented_args,
2724 },
2725 .idx = 0,
2726 .mask = 0,
2727 .trace = trace,
2728 .thread = thread,
2729 .show_string_prefix = trace->show_string_prefix,
2730 };
2731
2732 for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {
2733 if (syscall_arg.mask & bit)
2734 continue;
2735
c5e006cd 2736 syscall_arg.len = 0;
f11b2803 2737 syscall_arg.fmt = arg;
c5e006cd
ACM		 2738 if (field->flags & TEP_FIELD_IS_ARRAY) {
2739 int offset = field->offset;
2740
2741 if (field->flags & TEP_FIELD_IS_DYNAMIC) {
2742 offset = format_field__intval(field, sample, evsel->needs_swap);
2743 syscall_arg.len = offset >> 16;
2744 offset &= 0xffff;
1634bad3 2745 if (tep_field_is_relative(field->flags))
7c689c83 2746 offset += field->offset + field->size;
c5e006cd
ACM		 2747 }
2748
2749 val = (uintptr_t)(sample->raw_data + offset);
2750 } else
f11b2803
ACM		 2751 val = format_field__intval(field, sample, evsel->needs_swap);
2752 /*
2753 * Some syscall args need some mask, most don't and
2754 * return val untouched.
2755 */
2756 val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);
2757
2758 /*
2759 * Suppress this argument if its value is zero and
632f5c22 2760 * we don't have a string associated in an
f11b2803
ACM		 2761 * strarray for it.
2762 */
2763 if (val == 0 &&
2764 !trace->show_zeros &&
2765 !((arg->show_zero ||
2766 arg->scnprintf == SCA_STRARRAY ||
2767 arg->scnprintf == SCA_STRARRAYS) &&
2768 arg->parm))
2769 continue;
2770
2771 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2772
888964a0 2773 if (trace->show_arg_names)
f11b2803
ACM		 2774 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2775
2776 printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);
2777 }
2778
2779 return printed + fprintf(trace->output, "%s", bf);
2780}
2781
32dcd021 2782static int trace__event_handler(struct trace *trace, struct evsel *evsel,
14a052df
ACM		 2783 union perf_event *event __maybe_unused,
2784 struct perf_sample *sample)
2785{
a9c5e6c1 2786 struct thread *thread;
7ad35615 2787 int callchain_ret = 0;
a9c5e6c1
ACM		 2788 /*
2789 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2790 * this event's max_events having been hit and this is an entry coming
2791 * from the ring buffer that we should discard, since the max events
2792 * have already been considered/printed.
2793 */
2794 if (evsel->disabled)
2795 return 0;
2796
2797 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
7ad35615
ACM		 2798
2799 if (sample->callchain) {
8ab12a20
IR		 2800 struct callchain_cursor *cursor = get_tls_callchain_cursor();
2801
2802 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
7ad35615 2803 if (callchain_ret == 0) {
8ab12a20 2804 if (cursor->nr < trace->min_stack)
7ad35615
ACM		 2805 goto out;
2806 callchain_ret = 1;
2807 }
2808 }
2809
522283fe 2810 trace__printf_interrupted_entry(trace);
14a052df 2811 trace__fprintf_tstamp(trace, sample->time, trace->output);
0808921a 2812
42e4a52d 2813 if (trace->trace_syscalls && trace->show_duration)
0808921a
ACM		 2814 fprintf(trace->output, "( ): ");
2815
c4191e55
ACM		 2816 if (thread)
2817 trace__fprintf_comm_tid(trace, thread, trace->output);
2818
1cdf618f
ACM		 2819 if (evsel == trace->syscalls.events.augmented) {
2820 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2821 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2822
2823 if (sc) {
c4191e55
ACM		 2824 fprintf(trace->output, "%s(", sc->name);
2825 trace__fprintf_sys_enter(trace, evsel, sample);
2826 fputc(')', trace->output);
2827 goto newline;
1cdf618f
ACM		 2828 }
2829
2830 /*
2831 * XXX: Not having the associated syscall info or not finding/adding
2832 * the thread should never happen, but if it does...
2833 * fall thru and print it as a bpf_output event.
2834 */
2835 }
2836
311baaf9 2837 fprintf(trace->output, "%s(", evsel->name);
14a052df 2838
c754c382 2839 if (evsel__is_bpf_output(evsel)) {
1cdf618f 2840 bpf_output__fprintf(trace, sample);
1d6c9407 2841 } else if (evsel->tp_format) {
a98392bb
ACM		 2842 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2843 trace__fprintf_sys_enter(trace, evsel, sample)) {
f11b2803
ACM		 2844 if (trace->libtraceevent_print) {
2845 event_format__fprintf(evsel->tp_format, sample->cpu,
2846 sample->raw_data, sample->raw_size,
2847 trace->output);
2848 } else {
2849 trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);
2850 }
a98392bb 2851 }
14a052df
ACM		 2852 }
2853
1cdf618f 2854newline:
311baaf9 2855 fprintf(trace->output, ")\n");
202ff968 2856
7ad35615
ACM		 2857 if (callchain_ret > 0)
2858 trace__fprintf_callchain(trace, sample);
2859 else if (callchain_ret < 0)
8ab2e96d 2860 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
db25bf98
ACM		 2861
2862 ++trace->nr_events_printed;
2863
2864 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2865 evsel__disable(evsel);
2866 evsel__close(evsel);
2867 }
7ad35615 2868out:
a937c665 2869 thread__put(thread);
14a052df
ACM		 2870 return 0;
2871}
2872
598d02c5
SF		 2873static void print_location(FILE *f, struct perf_sample *sample,
2874 struct addr_location *al,
2875 bool print_dso, bool print_sym)
2876{
2877
bb963e16 2878 if ((verbose > 0 || print_dso) && al->map)
63df0e4b 2879 fprintf(f, "%s@", map__dso(al->map)->long_name);
598d02c5 2880
bb963e16 2881 if ((verbose > 0 || print_sym) && al->sym)
4414a3c5 2882 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
598d02c5
SF		 2883 al->addr - al->sym->start);
2884 else if (al->map)
4414a3c5 2885 fprintf(f, "0x%" PRIx64, al->addr);
598d02c5 2886 else
4414a3c5 2887 fprintf(f, "0x%" PRIx64, sample->addr);
598d02c5
SF		 2888}
2889
2890static int trace__pgfault(struct trace *trace,
32dcd021 2891 struct evsel *evsel,
473398a2 2892 union perf_event *event __maybe_unused,
598d02c5
SF		 2893 struct perf_sample *sample)
2894{
2895 struct thread *thread;
598d02c5
SF		 2896 struct addr_location al;
2897 char map_type = 'd';
a2ea67d7 2898 struct thread_trace *ttrace;
b91fc39f 2899 int err = -1;
1df54290 2900 int callchain_ret = 0;
598d02c5 2901
0dd5041c 2902 addr_location__init(&al);
598d02c5 2903 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1df54290
ACM		 2904
2905 if (sample->callchain) {
8ab12a20
IR		 2906 struct callchain_cursor *cursor = get_tls_callchain_cursor();
2907
2908 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
1df54290 2909 if (callchain_ret == 0) {
8ab12a20 2910 if (cursor->nr < trace->min_stack)
1df54290
ACM		 2911 goto out_put;
2912 callchain_ret = 1;
2913 }
2914 }
2915
a2ea67d7
SF		 2916 ttrace = thread__trace(thread, trace->output);
2917 if (ttrace == NULL)
b91fc39f 2918 goto out_put;
a2ea67d7 2919
1fc632ce 2920 if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
a2ea67d7
SF		 2921 ttrace->pfmaj++;
2922 else
2923 ttrace->pfmin++;
2924
2925 if (trace->summary_only)
b91fc39f 2926 goto out;
598d02c5 2927
4546263d 2928 thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
598d02c5 2929
fd2b2975 2930 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
598d02c5
SF		 2931
2932 fprintf(trace->output, "%sfault [",
1fc632ce 2933 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
598d02c5
SF		 2934 "maj" : "min");
2935
2936 print_location(trace->output, sample, &al, false, true);
2937
2938 fprintf(trace->output, "] => ");
2939
117d3c24 2940 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
598d02c5
SF		 2941
2942 if (!al.map) {
4546263d 2943 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
598d02c5
SF		 2944
2945 if (al.map)
2946 map_type = 'x';
2947 else
2948 map_type = '?';
2949 }
2950
2951 print_location(trace->output, sample, &al, true, false);
2952
2953 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
0c3a6ef4 2954
1df54290
ACM		 2955 if (callchain_ret > 0)
2956 trace__fprintf_callchain(trace, sample);
2957 else if (callchain_ret < 0)
8ab2e96d 2958 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
5067a8cd
ACM		 2959
2960 ++trace->nr_events_printed;
b91fc39f
ACM		 2961out:
2962 err = 0;
2963out_put:
2964 thread__put(thread);
0dd5041c 2965 addr_location__exit(&al);
b91fc39f 2966 return err;
598d02c5
SF		 2967}
2968
e6001980 2969static void trace__set_base_time(struct trace *trace,
32dcd021 2970 struct evsel *evsel,
e6001980
ACM		 2971 struct perf_sample *sample)
2972{
8a07a809
ACM		 2973 /*
2974 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2975 * and don't use sample->time unconditionally, we may end up having
2976 * some other event in the future without PERF_SAMPLE_TIME for good
2977 * reason, i.e. we may not be interested in its timestamps, just in
2978 * it taking place, picking some piece of information when it
2979 * appears in our event stream (vfs_getname comes to mind).
2980 */
2981 if (trace->base_time == 0 && !trace->full_time &&
1fc632ce 2982 (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
e6001980
ACM		 2983 trace->base_time = sample->time;
2984}
2985
6810fc91 2986static int trace__process_sample(struct perf_tool *tool,
0c82adcf 2987 union perf_event *event,
6810fc91 2988 struct perf_sample *sample,
32dcd021 2989 struct evsel *evsel,
6810fc91
DA		 2990 struct machine *machine __maybe_unused)
2991{
2992 struct trace *trace = container_of(tool, struct trace, tool);
aa07df6e 2993 struct thread *thread;
6810fc91
DA		 2994 int err = 0;
2995
744a9719 2996 tracepoint_handler handler = evsel->handler;
6810fc91 2997
aa07df6e
DA		 2998 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2999 if (thread && thread__is_filtered(thread))
ef65e96e 3000 goto out;
bdc89661 3001
e6001980 3002 trace__set_base_time(trace, evsel, sample);
6810fc91 3003
3160565f
DA		 3004 if (handler) {
3005 ++trace->nr_events;
0c82adcf 3006 handler(trace, evsel, event, sample);
3160565f 3007 }
ef65e96e
ACM		 3008out:
3009 thread__put(thread);
6810fc91
DA		 3010 return err;
3011}
3012
1e28fe0a 3013static int trace__record(struct trace *trace, int argc, const char **argv)
5e2485b1
DA		 3014{
3015 unsigned int rec_argc, i, j;
3016 const char **rec_argv;
3017 const char * const record_args[] = {
3018 "record",
3019 "-R",
3020 "-m", "1024",
3021 "-c", "1",
5e2485b1 3022 };
7fbfe22c
ACM		 3023 pid_t pid = getpid();
3024 char *filter = asprintf__tp_filter_pids(1, &pid);
1e28fe0a
SF		 3025 const char * const sc_args[] = { "-e", };
3026 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
3027 const char * const majpf_args[] = { "-e", "major-faults" };
3028 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
3029 const char * const minpf_args[] = { "-e", "minor-faults" };
3030 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
7fbfe22c 3031 int err = -1;
1e28fe0a 3032
7fbfe22c
ACM		 3033 /* +3 is for the event string below and the pid filter */
3034 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 +
1e28fe0a 3035 majpf_args_nr + minpf_args_nr + argc;
5e2485b1
DA		 3036 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3037
7fbfe22c
ACM		 3038 if (rec_argv == NULL || filter == NULL)
3039 goto out_free;
5e2485b1 3040
1e28fe0a 3041 j = 0;
5e2485b1 3042 for (i = 0; i < ARRAY_SIZE(record_args); i++)
1e28fe0a
SF		 3043 rec_argv[j++] = record_args[i];
3044
e281a960
SF		 3045 if (trace->trace_syscalls) {
3046 for (i = 0; i < sc_args_nr; i++)
3047 rec_argv[j++] = sc_args[i];
3048
3049 /* event string may be different for older kernels - e.g., RHEL6 */
3050 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
3051 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
3052 else if (is_valid_tracepoint("syscalls:sys_enter"))
3053 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
3054 else {
3055 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
7fbfe22c 3056 goto out_free;
e281a960 3057 }
9aca7f17 3058 }
9aca7f17 3059
7fbfe22c
ACM		 3060 rec_argv[j++] = "--filter";
3061 rec_argv[j++] = filter;
3062
1e28fe0a
SF		 3063 if (trace->trace_pgfaults & TRACE_PFMAJ)
3064 for (i = 0; i < majpf_args_nr; i++)
3065 rec_argv[j++] = majpf_args[i];
3066
3067 if (trace->trace_pgfaults & TRACE_PFMIN)
3068 for (i = 0; i < minpf_args_nr; i++)
3069 rec_argv[j++] = minpf_args[i];
3070
3071 for (i = 0; i < (unsigned int)argc; i++)
3072 rec_argv[j++] = argv[i];
5e2485b1 3073
7fbfe22c
ACM		 3074 err = cmd_record(j, rec_argv);
3075out_free:
3076 free(filter);
3077 free(rec_argv);
3078 return err;
5e2485b1
DA		 3079}
3080
bf2575c1
DA		 3081static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
3082
a1cf3a75 3083static bool evlist__add_vfs_getname(struct evlist *evlist)
c522739d 3084{
6ab3bc24 3085 bool found = false;
32dcd021 3086 struct evsel *evsel, *tmp;
a910e466
IR		 3087 struct parse_events_error err;
3088 int ret;
8dd2a131 3089
07eafd4e 3090 parse_events_error__init(&err);
a910e466 3091 ret = parse_events(evlist, "probe:vfs_getname*", &err);
07eafd4e
IR		 3092 parse_events_error__exit(&err);
3093 if (ret)
08c98776 3094 return false;
c522739d 3095
6ab3bc24 3096 evlist__for_each_entry_safe(evlist, evsel, tmp) {
8ab2e96d 3097 if (!strstarts(evsel__name(evsel), "probe:vfs_getname"))
6ab3bc24
ACM		 3098 continue;
3099
efc0cdc9 3100 if (evsel__field(evsel, "pathname")) {
6ab3bc24
ACM		 3101 evsel->handler = trace__vfs_getname;
3102 found = true;
3103 continue;
3104 }
3105
b27c4ece 3106 list_del_init(&evsel->core.node);
6ab3bc24 3107 evsel->evlist = NULL;
5eb2dd2a 3108 evsel__delete(evsel);
c522739d
ACM		 3109 }
3110
6ab3bc24 3111 return found;
c522739d
ACM		 3112}
3113
8f6725a2 3114static struct evsel *evsel__new_pgfault(u64 config)
598d02c5 3115{
32dcd021 3116 struct evsel *evsel;
598d02c5
SF		 3117 struct perf_event_attr attr = {
3118 .type = PERF_TYPE_SOFTWARE,
3119 .mmap_data = 1,
598d02c5
SF		 3120 };
3121
3122 attr.config = config;
0524798c 3123 attr.sample_period = 1;
598d02c5
SF		 3124
3125 event_attr_init(&attr);
3126
365c3ae7 3127 evsel = evsel__new(&attr);
0ae537cb
ACM		 3128 if (evsel)
3129 evsel->handler = trace__pgfault;
598d02c5 3130
0ae537cb 3131 return evsel;
598d02c5
SF		 3132}
3133
3cb4d5e0
RM		 3134static void evlist__free_syscall_tp_fields(struct evlist *evlist)
3135{
3136 struct evsel *evsel;
3137
3138 evlist__for_each_entry(evlist, evsel) {
3139 struct evsel_trace *et = evsel->priv;
3140
3141 if (!et || !evsel->tp_format || strcmp(evsel->tp_format->system, "syscalls"))
3142 continue;
3143
9997d5dd 3144 zfree(&et->fmt);
3cb4d5e0
RM		 3145 free(et);
3146 }
3147}
3148
ddbb1b13
ACM		 3149static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
3150{
3151 const u32 type = event->header.type;
32dcd021 3152 struct evsel *evsel;
ddbb1b13 3153
ddbb1b13
ACM		 3154 if (type != PERF_RECORD_SAMPLE) {
3155 trace__process_event(trace, trace->host, event, sample);
3156 return;
3157 }
3158
3ccf8a7b 3159 evsel = evlist__id2evsel(trace->evlist, sample->id);
ddbb1b13
ACM		 3160 if (evsel == NULL) {
3161 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
3162 return;
3163 }
3164
22ac4318
ACM		 3165 if (evswitch__discard(&trace->evswitch, evsel))
3166 return;
3167
e6001980
ACM		 3168 trace__set_base_time(trace, evsel, sample);
3169
1fc632ce 3170 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
ddbb1b13
ACM		 3171 sample->raw_data == NULL) {
3172 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
8ab2e96d 3173 evsel__name(evsel), sample->tid,
ddbb1b13
ACM		 3174 sample->cpu, sample->raw_size);
3175 } else {
3176 tracepoint_handler handler = evsel->handler;
3177 handler(trace, evsel, event, sample);
3178 }
5067a8cd
ACM		 3179
3180 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
3181 interrupted = true;
ddbb1b13
ACM		 3182}
3183
c27366f0
ACM		 3184static int trace__add_syscall_newtp(struct trace *trace)
3185{
3186 int ret = -1;
63503dba 3187 struct evlist *evlist = trace->evlist;
32dcd021 3188 struct evsel *sys_enter, *sys_exit;
c27366f0 3189
63f11c80 3190 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
c27366f0
ACM		 3191 if (sys_enter == NULL)
3192 goto out;
3193
3194 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
3195 goto out_delete_sys_enter;
3196
63f11c80 3197 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
c27366f0
ACM		 3198 if (sys_exit == NULL)
3199 goto out_delete_sys_enter;
3200
3201 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
3202 goto out_delete_sys_exit;
3203
6ec17b4e
ACM		 3204 evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
3205 evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
08e26396 3206
a1cf3a75
JO		 3207 evlist__add(evlist, sys_enter);
3208 evlist__add(evlist, sys_exit);
c27366f0 3209
2ddd5c04 3210 if (callchain_param.enabled && !trace->kernel_syscallchains) {
44621819
ACM		 3211 /*
3212 * We're interested only in the user space callchain
3213 * leading to the syscall, allow overriding that for
3214 * debugging reasons using --kernel_syscall_callchains
3215 */
1fc632ce 3216 sys_exit->core.attr.exclude_callchain_kernel = 1;
44621819
ACM		 3217 }
3218
8b3ce757
ACM		 3219 trace->syscalls.events.sys_enter = sys_enter;
3220 trace->syscalls.events.sys_exit = sys_exit;
c27366f0
ACM		 3221
3222 ret = 0;
3223out:
3224 return ret;
3225
3226out_delete_sys_exit:
5eb2dd2a 3227 evsel__delete_priv(sys_exit);
c27366f0 3228out_delete_sys_enter:
5eb2dd2a 3229 evsel__delete_priv(sys_enter);
c27366f0
ACM		 3230 goto out;
3231}
3232
246fbe03 3233static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
19867b61
ACM		 3234{
3235 int err = -1;
32dcd021 3236 struct evsel *sys_exit;
19867b61
ACM		 3237 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
3238 trace->ev_qualifier_ids.nr,
3239 trace->ev_qualifier_ids.entries);
3240
3241 if (filter == NULL)
3242 goto out_enomem;
3243
ad681adf 3244 if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) {
b15d0a4c 3245 sys_exit = trace->syscalls.events.sys_exit;
ad681adf 3246 err = evsel__append_tp_filter(sys_exit, filter);
b15d0a4c 3247 }
19867b61
ACM		 3248
3249 free(filter);
3250out:
3251 return err;
3252out_enomem:
3253 errno = ENOMEM;
3254 goto out;
3255}
c27366f0 3256
b27b38ed 3257#ifdef HAVE_LIBBPF_SUPPORT
beb64203
JO		 3258static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace, const char *name)
3259{
3260 if (trace->bpf_obj == NULL)
3261 return NULL;
3262
3263 return bpf_object__find_map_by_name(trace->bpf_obj, name);
3264}
3265
3266static void trace__set_bpf_map_filtered_pids(struct trace *trace)
3267{
3268 trace->filter_pids.map = trace__find_bpf_map_by_name(trace, "pids_filtered");
3269}
3270
3271static void trace__set_bpf_map_syscalls(struct trace *trace)
3272{
beb64203
JO		 3273 trace->syscalls.prog_array.sys_enter = trace__find_bpf_map_by_name(trace, "syscalls_sys_enter");
3274 trace->syscalls.prog_array.sys_exit = trace__find_bpf_map_by_name(trace, "syscalls_sys_exit");
3275}
3276
5834da7f
ACM		 3277static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
3278{
b098f336
KFL		 3279 struct bpf_program *pos, *prog = NULL;
3280 const char *sec_name;
3281
5834da7f
ACM		 3282 if (trace->bpf_obj == NULL)
3283 return NULL;
3284
b098f336
KFL		 3285 bpf_object__for_each_program(pos, trace->bpf_obj) {
3286 sec_name = bpf_program__section_name(pos);
3287 if (sec_name && !strcmp(sec_name, name)) {
3288 prog = pos;
3289 break;
3290 }
3291 }
3292
3293 return prog;
5834da7f
ACM		 3294}
3295
6ff8fff4
ACM		 3296static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
3297 const char *prog_name, const char *type)
3298{
3299 struct bpf_program *prog;
3300
8b8044e5
ACM		 3301 if (prog_name == NULL) {
3302 char default_prog_name[256];
3303 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->name);
3304 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3305 if (prog != NULL)
3306 goto out_found;
3307 if (sc->fmt && sc->fmt->alias) {
3308 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->fmt->alias);
3309 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3310 if (prog != NULL)
3311 goto out_found;
3312 }
6ff8fff4 3313 goto out_unaugmented;
8b8044e5 3314 }
6ff8fff4
ACM		 3315
3316 prog = trace__find_bpf_program_by_title(trace, prog_name);
8b8044e5
ACM		 3317
3318 if (prog != NULL) {
3319out_found:
6ff8fff4 3320 return prog;
8b8044e5 3321 }
6ff8fff4
ACM		 3322
3323 pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
3324 prog_name, type, sc->name);
3325out_unaugmented:
3326 return trace->syscalls.unaugmented_prog;
3327}
3328
3329static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
3330{
3331 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3332
3333 if (sc == NULL)
3334 return;
3335
8b8044e5
ACM		 3336 sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3337 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		 3338}
3339
3803a229
ACM		 3340static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
3341{
3342 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3343 return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3344}
3345
3346static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
3347{
3348 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3349 return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3350}
3351
ad4153f9
ACM		 3352static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
3353{
3354 struct tep_format_field *field, *candidate_field;
3355 int id;
3356
3357 /*
3358 * We're only interested in syscalls that have a pointer:
3359 */
3360 for (field = sc->args; field; field = field->next) {
3361 if (field->flags & TEP_FIELD_IS_POINTER)
3362 goto try_to_find_pair;
3363 }
3364
3365 return NULL;
3366
3367try_to_find_pair:
3368 for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) {
3369 struct syscall *pair = trace__syscall_info(trace, NULL, id);
3370 struct bpf_program *pair_prog;
3371 bool is_candidate = false;
3372
3373 if (pair == NULL || pair == sc ||
3374 pair->bpf_prog.sys_enter == trace->syscalls.unaugmented_prog)
3375 continue;
3376
3377 for (field = sc->args, candidate_field = pair->args;
3378 field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3379 bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3380 candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3381
3382 if (is_pointer) {
3383 if (!candidate_is_pointer) {
3384 // The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3385 continue;
3386 }
3387 } else {
3388 if (candidate_is_pointer) {
3389 // The candidate might copy a pointer we don't have, skip it.
3390 goto next_candidate;
3391 }
3392 continue;
3393 }
3394
3395 if (strcmp(field->type, candidate_field->type))
3396 goto next_candidate;
3397
3398 is_candidate = true;
3399 }
3400
3401 if (!is_candidate)
3402 goto next_candidate;
3403
3404 /*
3405 * Check if the tentative pair syscall augmenter has more pointers, if it has,
3406 * then it may be collecting that and we then can't use it, as it would collect
3407 * more than what is common to the two syscalls.
3408 */
3409 if (candidate_field) {
3410 for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3411 if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3412 goto next_candidate;
3413 }
3414
3415 pair_prog = pair->bpf_prog.sys_enter;
3416 /*
3417 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3418 * have been searched for, so search it here and if it returns the
3419 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3420 * program for a filtered syscall on a non-filtered one.
3421 *
3422 * For instance, we have "!syscalls:sys_enter_renameat" and that is
3423 * useful for "renameat2".
3424 */
3425 if (pair_prog == NULL) {
3426 pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3427 if (pair_prog == trace->syscalls.unaugmented_prog)
3428 goto next_candidate;
3429 }
3430
3431 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3432 return pair_prog;
3433 next_candidate:
3434 continue;
3435 }
3436
3437 return NULL;
3438}
3439
3803a229
ACM		 3440static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3441{
3442 int map_enter_fd = bpf_map__fd(trace->syscalls.prog_array.sys_enter),
3443 map_exit_fd = bpf_map__fd(trace->syscalls.prog_array.sys_exit);
3444 int err = 0, key;
3445
3446 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3447 int prog_fd;
3448
3449 if (!trace__syscall_enabled(trace, key))
3450 continue;
3451
3452 trace__init_syscall_bpf_progs(trace, key);
3453
3454 // It'll get at least the "!raw_syscalls:unaugmented"
3455 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3456 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3457 if (err)
3458 break;
3459 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3460 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3461 if (err)
3462 break;
3463 }
3464
ad4153f9
ACM		 3465 /*
3466 * Now lets do a second pass looking for enabled syscalls without
3467 * an augmenter that have a signature that is a superset of another
3468 * syscall with an augmenter so that we can auto-reuse it.
3469 *
3470 * I.e. if we have an augmenter for the "open" syscall that has
3471 * this signature:
3472 *
3473 * int open(const char *pathname, int flags, mode_t mode);
3474 *
3475 * I.e. that will collect just the first string argument, then we
3476 * can reuse it for the 'creat' syscall, that has this signature:
3477 *
3478 * int creat(const char *pathname, mode_t mode);
3479 *
3480 * and for:
3481 *
3482 * int stat(const char *pathname, struct stat *statbuf);
3483 * int lstat(const char *pathname, struct stat *statbuf);
3484 *
3485 * Because the 'open' augmenter will collect the first arg as a string,
3486 * and leave alone all the other args, which already helps with
3487 * beautifying 'stat' and 'lstat''s pathname arg.
3488 *
3489 * Then, in time, when 'stat' gets an augmenter that collects both
3490 * first and second arg (this one on the raw_syscalls:sys_exit prog
3491 * array tail call, then that one will be used.
3492 */
3493 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3494 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3495 struct bpf_program *pair_prog;
3496 int prog_fd;
3497
3498 if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3499 continue;
3500
3501 /*
3502 * For now we're just reusing the sys_enter prog, and if it
3503 * already has an augmenter, we don't need to find one.
3504 */
3505 if (sc->bpf_prog.sys_enter != trace->syscalls.unaugmented_prog)
3506 continue;
3507
3508 /*
3509 * Look at all the other syscalls for one that has a signature
3510 * that is close enough that we can share:
3511 */
3512 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3513 if (pair_prog == NULL)
3514 continue;
3515
3516 sc->bpf_prog.sys_enter = pair_prog;
3517
3518 /*
3519 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3520 * with the fd for the program we're reusing:
3521 */
3522 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3523 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3524 if (err)
3525 break;
3526 }
3527
3528
3803a229
ACM		 3529 return err;
3530}
8bd436b0
ACM		 3531
3532static void trace__delete_augmented_syscalls(struct trace *trace)
3533{
3534 struct evsel *evsel, *tmp;
3535
3536 evlist__remove(trace->evlist, trace->syscalls.events.augmented);
3537 evsel__delete(trace->syscalls.events.augmented);
3538 trace->syscalls.events.augmented = NULL;
3539
3540 evlist__for_each_entry_safe(trace->evlist, tmp, evsel) {
3541 if (evsel->bpf_obj == trace->bpf_obj) {
3542 evlist__remove(trace->evlist, evsel);
3543 evsel__delete(evsel);
3544 }
3545
3546 }
3547
3548 bpf_object__close(trace->bpf_obj);
3549 trace->bpf_obj = NULL;
3550}
3551#else // HAVE_LIBBPF_SUPPORT
beb64203
JO		 3552static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace __maybe_unused,
3553 const char *name __maybe_unused)
3554{
3555 return NULL;
3556}
3557
3558static void trace__set_bpf_map_filtered_pids(struct trace *trace __maybe_unused)
3559{
3560}
3561
3562static void trace__set_bpf_map_syscalls(struct trace *trace __maybe_unused)
3563{
3564}
3565
5834da7f
ACM		 3566static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace __maybe_unused,
3567 const char *name __maybe_unused)
3568{
3569 return NULL;
3570}
3803a229
ACM		 3571
3572static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace __maybe_unused)
3573{
3574 return 0;
3575}
8bd436b0
ACM		 3576
3577static void trace__delete_augmented_syscalls(struct trace *trace __maybe_unused)
3578{
3579}
b27b38ed
ACM		 3580#endif // HAVE_LIBBPF_SUPPORT
3581
8bd436b0
ACM		 3582static bool trace__only_augmented_syscalls_evsels(struct trace *trace)
3583{
3584 struct evsel *evsel;
3585
3586 evlist__for_each_entry(trace->evlist, evsel) {
3587 if (evsel == trace->syscalls.events.augmented ||
3588 evsel->bpf_obj == trace->bpf_obj)
3589 continue;
3590
3591 return false;
3592 }
3593
3594 return true;
3595}
3596
246fbe03
ACM		 3597static int trace__set_ev_qualifier_filter(struct trace *trace)
3598{
f76214f9
ACM		 3599 if (trace->syscalls.events.sys_enter)
3600 return trace__set_ev_qualifier_tp_filter(trace);
3601 return 0;
246fbe03
ACM		 3602}
3603
a9964c43
ACM		 3604static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3605 size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3606{
3607 int err = 0;
3608#ifdef HAVE_LIBBPF_SUPPORT
3609 bool value = true;
3610 int map_fd = bpf_map__fd(map);
3611 size_t i;
3612
3613 for (i = 0; i < npids; ++i) {
3614 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3615 if (err)
3616 break;
3617 }
3618#endif
3619 return err;
3620}
3621
dd1a5037
ACM		 3622static int trace__set_filter_loop_pids(struct trace *trace)
3623{
a9964c43 3624 unsigned int nr = 1, err;
dd1a5037
ACM		 3625 pid_t pids[32] = {
3626 getpid(),
3627 };
082ab9a1
ACM		 3628 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3629
3630 while (thread && nr < ARRAY_SIZE(pids)) {
ee84a303
IR		 3631 struct thread *parent = machine__find_thread(trace->host,
3632 thread__ppid(thread),
3633 thread__ppid(thread));
082ab9a1
ACM		 3634
3635 if (parent == NULL)
3636 break;
3637
ca2da70c
ACM		 3638 if (!strcmp(thread__comm_str(parent), "sshd") ||
3639 strstarts(thread__comm_str(parent), "gnome-terminal")) {
ee84a303 3640 pids[nr++] = thread__tid(parent);
082ab9a1
ACM		 3641 break;
3642 }
3643 thread = parent;
3644 }
dd1a5037 3645
24bf91a7 3646 err = evlist__append_tp_filter_pids(trace->evlist, nr, pids);
a9964c43
ACM		 3647 if (!err && trace->filter_pids.map)
3648 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3649
3650 return err;
3651}
3652
3653static int trace__set_filter_pids(struct trace *trace)
3654{
3655 int err = 0;
3656 /*
3657 * Better not use !target__has_task() here because we need to cover the
3658 * case where no threads were specified in the command line, but a
3659 * workload was, and in that case we will fill in the thread_map when
7b392ef0 3660 * we fork the workload in evlist__prepare_workload.
a9964c43
ACM		 3661 */
3662 if (trace->filter_pids.nr > 0) {
24bf91a7
ACM		 3663 err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3664 trace->filter_pids.entries);
a9964c43
ACM		 3665 if (!err && trace->filter_pids.map) {
3666 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3667 trace->filter_pids.entries);
3668 }
a2f354e3 3669 } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
a9964c43
ACM		 3670 err = trace__set_filter_loop_pids(trace);
3671 }
3672
3673 return err;
dd1a5037
ACM		 3674}
3675
7ba61524 3676static int __trace__deliver_event(struct trace *trace, union perf_event *event)
1f44b3e2 3677{
63503dba 3678 struct evlist *evlist = trace->evlist;
1f44b3e2 3679 struct perf_sample sample;
2a6599cd 3680 int err = evlist__parse_sample(evlist, event, &sample);
1f44b3e2 3681
1f44b3e2
JO		 3682 if (err)
3683 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3684 else
3685 trace__handle_event(trace, event, &sample);
3686
3687 return 0;
3688}
3689
a761a8d1 3690static int __trace__flush_events(struct trace *trace)
028713aa
JO		 3691{
3692 u64 first = ordered_events__first_time(&trace->oe.data);
3693 u64 flush = trace->oe.last - NSEC_PER_SEC;
3694
3695 /* Is there some thing to flush.. */
3696 if (first && first < flush)
3697 return ordered_events__flush_time(&trace->oe.data, flush);
3698
3699 return 0;
3700}
3701
a761a8d1
ACM		 3702static int trace__flush_events(struct trace *trace)
3703{
3704 return !trace->sort_events ? 0 : __trace__flush_events(trace);
3705}
3706
7ba61524 3707static int trace__deliver_event(struct trace *trace, union perf_event *event)
028713aa 3708{
028713aa
JO		 3709 int err;
3710
a761a8d1
ACM		 3711 if (!trace->sort_events)
3712 return __trace__deliver_event(trace, event);
3713
2a6599cd 3714 err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
028713aa
JO		 3715 if (err && err != -1)
3716 return err;
3717
2292083f 3718 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0, NULL);
028713aa
JO		 3719 if (err)
3720 return err;
3721
7ba61524 3722 return trace__flush_events(trace);
028713aa
JO		 3723}
3724
3725static int ordered_events__deliver_event(struct ordered_events *oe,
3726 struct ordered_event *event)
3727{
3728 struct trace *trace = container_of(oe, struct trace, oe.data);
3729
7ba61524 3730 return __trace__deliver_event(trace, event->event);
028713aa
JO		 3731}
3732
794bca26 3733static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg)
90df0249
ACM		 3734{
3735 struct tep_format_field *field;
fecd9907 3736 struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel);
90df0249
ACM		 3737
3738 if (evsel->tp_format == NULL || fmt == NULL)
3739 return NULL;
3740
3741 for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt)
3742 if (strcmp(field->name, arg) == 0)
3743 return fmt;
3744
3745 return NULL;
3746}
3747
3748static int trace__expand_filter(struct trace *trace __maybe_unused, struct evsel *evsel)
3749{
3750 char *tok, *left = evsel->filter, *new_filter = evsel->filter;
3751
3752 while ((tok = strpbrk(left, "=<>!")) != NULL) {
3753 char *right = tok + 1, *right_end;
3754
3755 if (*right == '=')
3756 ++right;
3757
3758 while (isspace(*right))
3759 ++right;
3760
3761 if (*right == '\0')
3762 break;
3763
3764 while (!isalpha(*left))
3765 if (++left == tok) {
3766 /*
3767 * Bail out, can't find the name of the argument that is being
3768 * used in the filter, let it try to set this filter, will fail later.
3769 */
3770 return 0;
3771 }
3772
3773 right_end = right + 1;
154c978d 3774 while (isalnum(*right_end) || *right_end == '_' || *right_end == '|')
90df0249
ACM		 3775 ++right_end;
3776
3777 if (isalpha(*right)) {
3778 struct syscall_arg_fmt *fmt;
3779 int left_size = tok - left,
3780 right_size = right_end - right;
3781 char arg[128];
3782
3783 while (isspace(left[left_size - 1]))
3784 --left_size;
3785
3786 scnprintf(arg, sizeof(arg), "%.*s", left_size, left);
3787
794bca26 3788 fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg);
90df0249 3789 if (fmt == NULL) {
3cdc8db9
ACM		 3790 pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n",
3791 arg, evsel->name, evsel->filter);
90df0249
ACM		 3792 return -1;
3793 }
3794
3795 pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ",
3796 arg, (int)(right - tok), tok, right_size, right);
3797
3798 if (fmt->strtoul) {
3799 u64 val;
9afec87e
ACM		 3800 struct syscall_arg syscall_arg = {
3801 .parm = fmt->parm,
3802 };
3803
3804 if (fmt->strtoul(right, right_size, &syscall_arg, &val)) {
90df0249
ACM		 3805 char *n, expansion[19];
3806 int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val);
3807 int expansion_offset = right - new_filter;
3808
3809 pr_debug("%s", expansion);
3810
3811 if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) {
3812 pr_debug(" out of memory!\n");
3813 free(new_filter);
3814 return -1;
3815 }
3816 if (new_filter != evsel->filter)
3817 free(new_filter);
3818 left = n + expansion_offset + expansion_lenght;
3819 new_filter = n;
3820 } else {
3821 pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3822 right_size, right, arg, evsel->name, evsel->filter);
3823 return -1;
3824 }
3825 } else {
3826 pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3827 arg, evsel->name, evsel->filter);
3828 return -1;
3829 }
3830
3831 pr_debug("\n");
3832 } else {
3833 left = right_end;
3834 }
3835 }
3836
3837 if (new_filter != evsel->filter) {
3838 pr_debug("New filter for %s: %s\n", evsel->name, new_filter);
ad681adf 3839 evsel__set_filter(evsel, new_filter);
90df0249
ACM		 3840 free(new_filter);
3841 }
3842
3843 return 0;
3844}
3845
3846static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel)
3847{
3848 struct evlist *evlist = trace->evlist;
3849 struct evsel *evsel;
3850
3851 evlist__for_each_entry(evlist, evsel) {
3852 if (evsel->filter == NULL)
3853 continue;
3854
3855 if (trace__expand_filter(trace, evsel)) {
3856 *err_evsel = evsel;
3857 return -1;
3858 }
3859 }
3860
3861 return 0;
3862}
3863
f15eb531 3864static int trace__run(struct trace *trace, int argc, const char **argv)
514f1c67 3865{
63503dba 3866 struct evlist *evlist = trace->evlist;
32dcd021 3867 struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
efd5745e
ACM		 3868 int err = -1, i;
3869 unsigned long before;
f15eb531 3870 const bool forks = argc > 0;
46fb3c21 3871 bool draining = false;
514f1c67 3872
75b757ca
ACM		 3873 trace->live = true;
3874
0df50e0b
ACM		 3875 if (!trace->raw_augmented_syscalls) {
3876 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3877 goto out_error_raw_syscalls;
514f1c67 3878
0df50e0b 3879 if (trace->trace_syscalls)
a1cf3a75 3880 trace->vfs_getname = evlist__add_vfs_getname(evlist);
0df50e0b 3881 }
c522739d 3882
0ae537cb 3883 if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
8f6725a2 3884 pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
0ae537cb
ACM		 3885 if (pgfault_maj == NULL)
3886 goto out_error_mem;
6ec17b4e 3887 evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
a1cf3a75 3888 evlist__add(evlist, pgfault_maj);
e2726d99 3889 }
598d02c5 3890
0ae537cb 3891 if ((trace->trace_pgfaults & TRACE_PFMIN)) {
8f6725a2 3892 pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
0ae537cb
ACM		 3893 if (pgfault_min == NULL)
3894 goto out_error_mem;
6ec17b4e 3895 evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
a1cf3a75 3896 evlist__add(evlist, pgfault_min);
0ae537cb 3897 }
598d02c5 3898
b4515ad6
GL		 3899 /* Enable ignoring missing threads when -u/-p option is defined. */
3900 trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid;
3901
1302d88e 3902 if (trace->sched &&
e251abee 3903 evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime))
2cc990ba 3904 goto out_error_sched_stat_runtime;
9ea42ba4
ACM		 3905 /*
3906 * If a global cgroup was set, apply it to all the events without an
3907 * explicit cgroup. I.e.:
3908 *
3909 * trace -G A -e sched:*switch
3910 *
3911 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3912 * _and_ sched:sched_switch to the 'A' cgroup, while:
3913 *
3914 * trace -e sched:*switch -G A
3915 *
3916 * will only set the sched:sched_switch event to the 'A' cgroup, all the
3917 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3918 * a cgroup (on the root cgroup, sys wide, etc).
3919 *
3920 * Multiple cgroups:
3921 *
3922 * trace -G A -e sched:*switch -G B
3923 *
3924 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
3925 * to the 'B' cgroup.
3926 *
3927 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
3928 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
3929 */
3930 if (trace->cgroup)
3931 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
3932
7748bb71 3933 err = evlist__create_maps(evlist, &trace->opts.target);
514f1c67 3934 if (err < 0) {
c24ff998 3935 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
514f1c67
ACM		 3936 goto out_delete_evlist;
3937 }
3938
752fde44
ACM		 3939 err = trace__symbols_init(trace, evlist);
3940 if (err < 0) {
c24ff998 3941 fprintf(trace->output, "Problems initializing symbol libraries!\n");
03ad9747 3942 goto out_delete_evlist;
752fde44
ACM		 3943 }
3944
78e1bc25 3945 evlist__config(evlist, &trace->opts, &callchain_param);
fde54b78 3946
f15eb531 3947 if (forks) {
7b392ef0 3948 err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL);
f15eb531 3949 if (err < 0) {
c24ff998 3950 fprintf(trace->output, "Couldn't run the workload!\n");
03ad9747 3951 goto out_delete_evlist;
f15eb531 3952 }
de9f498d 3953 workload_pid = evlist->workload.pid;
f15eb531
NK		 3954 }
3955
474ddc4c 3956 err = evlist__open(evlist);
a8f23d8f
ACM		 3957 if (err < 0)
3958 goto out_error_open;
514f1c67 3959
ba504235
WN		 3960 err = bpf__apply_obj_config();
3961 if (err) {
3962 char errbuf[BUFSIZ];
3963
3964 bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
3965 pr_err("ERROR: Apply config to BPF failed: %s\n",
3966 errbuf);
3967 goto out_error_open;
3968 }
3969
a9964c43 3970 err = trace__set_filter_pids(trace);
94ad89bc
ACM		 3971 if (err < 0)
3972 goto out_error_mem;
3973
3803a229
ACM		 3974 if (trace->syscalls.prog_array.sys_enter)
3975 trace__init_syscalls_bpf_prog_array_maps(trace);
3976
19867b61
ACM		 3977 if (trace->ev_qualifier_ids.nr > 0) {
3978 err = trace__set_ev_qualifier_filter(trace);
3979 if (err < 0)
3980 goto out_errno;
19867b61 3981
0df50e0b
ACM		 3982 if (trace->syscalls.events.sys_exit) {
3983 pr_debug("event qualifier tracepoint filter: %s\n",
3984 trace->syscalls.events.sys_exit->filter);
3985 }
2e5e5f87 3986 }
19867b61 3987
79d725cd
ACM		 3988 /*
3989 * If the "close" syscall is not traced, then we will not have the
3990 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
3991 * fd->pathname table and were ending up showing the last value set by
3992 * syscalls opening a pathname and associating it with a descriptor or
3993 * reading it from /proc/pid/fd/ in cases where that doesn't make
3994 * sense.
3995 *
3996 * So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
3997 * not in use.
3998 */
3999 trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
4000
90df0249
ACM		 4001 err = trace__expand_filters(trace, &evsel);
4002 if (err)
4003 goto out_delete_evlist;
24bf91a7 4004 err = evlist__apply_filters(evlist, &evsel);
94ad89bc
ACM		 4005 if (err < 0)
4006 goto out_error_apply_filters;
241b057c 4007
ff7a4f98
ACM		 4008 if (trace->dump.map)
4009 bpf_map__fprintf(trace->dump.map, trace->output);
4010
9521b5f2 4011 err = evlist__mmap(evlist, trace->opts.mmap_pages);
e09b18d4
ACM		 4012 if (err < 0)
4013 goto out_error_mmap;
514f1c67 4014
cb4b9e68 4015 if (!target__none(&trace->opts.target) && !trace->opts.target.initial_delay)
1c87f165 4016 evlist__enable(evlist);
cb24d01d 4017
f15eb531 4018 if (forks)
7b392ef0 4019 evlist__start_workload(evlist);
f15eb531 4020
cb4b9e68
CD		 4021 if (trace->opts.target.initial_delay) {
4022 usleep(trace->opts.target.initial_delay * 1000);
1c87f165 4023 evlist__enable(evlist);
e36b7821
AB		 4024 }
4025
a2f354e3 4026 trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
fd3f518f
IR		 4027 perf_thread_map__nr(evlist->core.threads) > 1 ||
4028 evlist__first(evlist)->core.attr.inherit;
bd3dda9a
ACM		 4029
4030 /*
1fc632ce
JO		 4031 * Now that we already used evsel->core.attr to ask the kernel to setup the
4032 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
bd3dda9a 4033 * trace__resolve_callchain(), allowing per-event max-stack settings
adba1634 4034 * to override an explicitly set --max-stack global setting.
bd3dda9a
ACM		 4035 */
4036 evlist__for_each_entry(evlist, evsel) {
27de9b2b 4037 if (evsel__has_callchain(evsel) &&
1fc632ce
JO		 4038 evsel->core.attr.sample_max_stack == 0)
4039 evsel->core.attr.sample_max_stack = trace->max_stack;
bd3dda9a 4040 }
514f1c67 4041again:
efd5745e 4042 before = trace->nr_events;
514f1c67 4043
c976ee11 4044 for (i = 0; i < evlist->core.nr_mmaps; i++) {
514f1c67 4045 union perf_event *event;
a5830532 4046 struct mmap *md;
514f1c67 4047
d7f55c62 4048 md = &evlist->mmap[i];
7c4d4182 4049 if (perf_mmap__read_init(&md->core) < 0)
d7f55c62
KL		 4050 continue;
4051
151ed5d7 4052 while ((event = perf_mmap__read_event(&md->core)) != NULL) {
efd5745e 4053 ++trace->nr_events;
514f1c67 4054
7ba61524 4055 err = trace__deliver_event(trace, event);
028713aa
JO		 4056 if (err)
4057 goto out_disable;
514f1c67 4058
7728fa0c 4059 perf_mmap__consume(&md->core);
20c5f10e 4060
ba209f85
ACM		 4061 if (interrupted)
4062 goto out_disable;
02ac5421
ACM		 4063
4064 if (done && !draining) {
e74676de 4065 evlist__disable(evlist);
02ac5421
ACM		 4066 draining = true;
4067 }
514f1c67 4068 }
32fdc2ca 4069 perf_mmap__read_done(&md->core);
514f1c67
ACM		 4070 }
4071
efd5745e 4072 if (trace->nr_events == before) {
ba209f85 4073 int timeout = done ? 100 : -1;
f15eb531 4074
80ab2987 4075 if (!draining && evlist__poll(evlist, timeout) > 0) {
f4009e7b 4076 if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
46fb3c21
ACM		 4077 draining = true;
4078
ba209f85 4079 goto again;
028713aa 4080 } else {
7ba61524 4081 if (trace__flush_events(trace))
028713aa 4082 goto out_disable;
46fb3c21 4083 }
ba209f85
ACM		 4084 } else {
4085 goto again;
f15eb531
NK		 4086 }
4087
ba209f85 4088out_disable:
f3b623b8
ACM		 4089 thread__zput(trace->current);
4090
e74676de 4091 evlist__disable(evlist);
514f1c67 4092
a761a8d1
ACM		 4093 if (trace->sort_events)
4094 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
028713aa 4095
c522739d
ACM		 4096 if (!err) {
4097 if (trace->summary)
4098 trace__fprintf_thread_summary(trace, trace->output);
4099
4100 if (trace->show_tool_stats) {
4101 fprintf(trace->output, "Stats:\n "
4102 " vfs_getname : %" PRIu64 "\n"
4103 " proc_getname: %" PRIu64 "\n",
4104 trace->stats.vfs_getname,
4105 trace->stats.proc_getname);
4106 }
4107 }
bf2575c1 4108
514f1c67 4109out_delete_evlist:
33974a41 4110 trace__symbols__exit(trace);
3cb4d5e0 4111 evlist__free_syscall_tp_fields(evlist);
c12995a5 4112 evlist__delete(evlist);
9ea42ba4 4113 cgroup__put(trace->cgroup);
14a052df 4114 trace->evlist = NULL;
75b757ca 4115 trace->live = false;
514f1c67 4116 return err;
6ef068cb
ACM		 4117{
4118 char errbuf[BUFSIZ];
a8f23d8f 4119
2cc990ba 4120out_error_sched_stat_runtime:
988bdb31 4121 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
2cc990ba
ACM		 4122 goto out_error;
4123
801c67b0 4124out_error_raw_syscalls:
988bdb31 4125 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
a8f23d8f
ACM		 4126 goto out_error;
4127
e09b18d4 4128out_error_mmap:
d1f249ec 4129 evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
e09b18d4
ACM		 4130 goto out_error;
4131
a8f23d8f 4132out_error_open:
d1f249ec 4133 evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
a8f23d8f
ACM		 4134
4135out_error:
6ef068cb 4136 fprintf(trace->output, "%s\n", errbuf);
87f91868 4137 goto out_delete_evlist;
94ad89bc
ACM		 4138
4139out_error_apply_filters:
4140 fprintf(trace->output,
4141 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
8ab2e96d 4142 evsel->filter, evsel__name(evsel), errno,
c8b5f2c9 4143 str_error_r(errno, errbuf, sizeof(errbuf)));
94ad89bc 4144 goto out_delete_evlist;
514f1c67 4145}
5ed08dae
ACM		 4146out_error_mem:
4147 fprintf(trace->output, "Not enough memory to run!\n");
4148 goto out_delete_evlist;
19867b61
ACM		 4149
4150out_errno:
4151 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
4152 goto out_delete_evlist;
a8f23d8f 4153}
514f1c67 4154
6810fc91
DA		 4155static int trace__replay(struct trace *trace)
4156{
32dcd021 4157 const struct evsel_str_handler handlers[] = {
c522739d 4158 { "probe:vfs_getname", trace__vfs_getname, },
6810fc91 4159 };
8ceb41d7 4160 struct perf_data data = {
2d4f2799
JO		 4161 .path = input_name,
4162 .mode = PERF_DATA_MODE_READ,
4163 .force = trace->force,
f5fc1412 4164 };
6810fc91 4165 struct perf_session *session;
32dcd021 4166 struct evsel *evsel;
6810fc91
DA		 4167 int err = -1;
4168
4169 trace->tool.sample = trace__process_sample;
4170 trace->tool.mmap = perf_event__process_mmap;
384c671e 4171 trace->tool.mmap2 = perf_event__process_mmap2;
6810fc91
DA		 4172 trace->tool.comm = perf_event__process_comm;
4173 trace->tool.exit = perf_event__process_exit;
4174 trace->tool.fork = perf_event__process_fork;
4175 trace->tool.attr = perf_event__process_attr;
f3b3614a 4176 trace->tool.tracing_data = perf_event__process_tracing_data;
6810fc91 4177 trace->tool.build_id = perf_event__process_build_id;
f3b3614a 4178 trace->tool.namespaces = perf_event__process_namespaces;
6810fc91 4179
0a8cb85c 4180 trace->tool.ordered_events = true;
6810fc91
DA		 4181 trace->tool.ordering_requires_timestamps = true;
4182
4183 /* add tid to output */
4184 trace->multiple_threads = true;
4185
2681bd85 4186 session = perf_session__new(&data, &trace->tool);
6ef81c55
MI		 4187 if (IS_ERR(session))
4188 return PTR_ERR(session);
6810fc91 4189
aa07df6e
DA		 4190 if (trace->opts.target.pid)
4191 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
4192
4193 if (trace->opts.target.tid)
4194 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
4195
0a7e6d1b 4196 if (symbol__init(&session->header.env) < 0)
cb2ffae2
NK		 4197 goto out;
4198
8fb598e5
DA		 4199 trace->host = &session->machines.host;
4200
6810fc91
DA		 4201 err = perf_session__set_tracepoints_handlers(session, handlers);
4202 if (err)
4203 goto out;
4204
b02736f7 4205 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter");
4b335e1e 4206 trace->syscalls.events.sys_enter = evsel;
9aca7f17
DA		 4207 /* older kernels have syscalls tp versus raw_syscalls */
4208 if (evsel == NULL)
b02736f7 4209 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter");
003824e8 4210
e281a960 4211 if (evsel &&
794bca26 4212 (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
e281a960 4213 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
003824e8
NK		 4214 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
4215 goto out;
4216 }
4217
b02736f7 4218 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit");
4b335e1e 4219 trace->syscalls.events.sys_exit = evsel;
9aca7f17 4220 if (evsel == NULL)
b02736f7 4221 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit");
e281a960 4222 if (evsel &&
794bca26 4223 (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
e281a960 4224 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
003824e8 4225 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
6810fc91
DA		 4226 goto out;
4227 }
4228
e5cadb93 4229 evlist__for_each_entry(session->evlist, evsel) {
1fc632ce
JO		 4230 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
4231 (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
4232 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
4233 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
1e28fe0a
SF		 4234 evsel->handler = trace__pgfault;
4235 }
4236
6810fc91
DA		 4237 setup_pager();
4238
b7b61cbe 4239 err = perf_session__process_events(session);
6810fc91
DA		 4240 if (err)
4241 pr_err("Failed to process events, error %d", err);
4242
bf2575c1
DA		 4243 else if (trace->summary)
4244 trace__fprintf_thread_summary(trace, trace->output);
4245
6810fc91
DA		 4246out:
4247 perf_session__delete(session);
4248
4249 return err;
4250}
4251
1302d88e
ACM		 4252static size_t trace__fprintf_threads_header(FILE *fp)
4253{
4254 size_t printed;
4255
99ff7150 4256 printed = fprintf(fp, "\n Summary of events:\n\n");
bf2575c1
DA		 4257
4258 return printed;
4259}
4260
b535d523 4261DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
8eded45f
ACM		 4262 struct syscall_stats *stats;
4263 double msecs;
4264 int syscall;
b535d523
ACM		 4265)
4266{
4267 struct int_node *source = rb_entry(nd, struct int_node, rb_node);
8eded45f 4268 struct syscall_stats *stats = source->priv;
b535d523
ACM		 4269
4270 entry->syscall = source->i;
4271 entry->stats = stats;
8eded45f 4272 entry->msecs = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0;
b535d523
ACM		 4273}
4274
bf2575c1
DA		 4275static size_t thread__dump_stats(struct thread_trace *ttrace,
4276 struct trace *trace, FILE *fp)
4277{
bf2575c1
DA		 4278 size_t printed = 0;
4279 struct syscall *sc;
b535d523
ACM		 4280 struct rb_node *nd;
4281 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
bf2575c1 4282
b535d523 4283 if (syscall_stats == NULL)
bf2575c1
DA		 4284 return 0;
4285
4286 printed += fprintf(fp, "\n");
4287
8eded45f
ACM		 4288 printed += fprintf(fp, " syscall calls errors total min avg max stddev\n");
4289 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
4290 printed += fprintf(fp, " --------------- -------- ------ -------- --------- --------- --------- ------\n");
99ff7150 4291
98a91837 4292 resort_rb__for_each_entry(nd, syscall_stats) {
8eded45f 4293 struct syscall_stats *stats = syscall_stats_entry->stats;
bf2575c1 4294 if (stats) {
8eded45f
ACM		 4295 double min = (double)(stats->stats.min) / NSEC_PER_MSEC;
4296 double max = (double)(stats->stats.max) / NSEC_PER_MSEC;
4297 double avg = avg_stats(&stats->stats);
bf2575c1 4298 double pct;
8eded45f 4299 u64 n = (u64)stats->stats.n;
bf2575c1 4300
8eded45f 4301 pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0;
bf2575c1
DA		 4302 avg /= NSEC_PER_MSEC;
4303
b535d523 4304 sc = &trace->syscalls.table[syscall_stats_entry->syscall];
99ff7150 4305 printed += fprintf(fp, " %-15s", sc->name);
8eded45f
ACM		 4306 printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f",
4307 n, stats->nr_failures, syscall_stats_entry->msecs, min, avg);
27a778b5 4308 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
b88b14db
ACM		 4309
4310 if (trace->errno_summary && stats->nr_failures) {
4311 const char *arch_name = perf_env__arch(trace->host->env);
4312 int e;
4313
4314 for (e = 0; e < stats->max_errno; ++e) {
4315 if (stats->errnos[e] != 0)
4316 fprintf(fp, "\t\t\t\t%s: %d\n", arch_syscalls__strerrno(arch_name, e + 1), stats->errnos[e]);
4317 }
4318 }
bf2575c1 4319 }
bf2575c1
DA		 4320 }
4321
b535d523 4322 resort_rb__delete(syscall_stats);
bf2575c1 4323 printed += fprintf(fp, "\n\n");
1302d88e
ACM		 4324
4325 return printed;
4326}
4327
96c14451 4328static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
896cbb56 4329{
96c14451 4330 size_t printed = 0;
89dceb22 4331 struct thread_trace *ttrace = thread__priv(thread);
896cbb56
DA		 4332 double ratio;
4333
4334 if (ttrace == NULL)
4335 return 0;
4336
4337 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
4338
ee84a303 4339 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread__tid(thread));
99ff7150 4340 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
15e65c69 4341 printed += fprintf(fp, "%.1f%%", ratio);
a2ea67d7
SF		 4342 if (ttrace->pfmaj)
4343 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
4344 if (ttrace->pfmin)
4345 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
03548ebf
ACM		 4346 if (trace->sched)
4347 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
4348 else if (fputc('\n', fp) != EOF)
4349 ++printed;
4350
bf2575c1 4351 printed += thread__dump_stats(ttrace, trace, fp);
896cbb56 4352
96c14451
ACM		 4353 return printed;
4354}
896cbb56 4355
96c14451
ACM		 4356static unsigned long thread__nr_events(struct thread_trace *ttrace)
4357{
4358 return ttrace ? ttrace->nr_events : 0;
4359}
4360
ee84a303
IR		 4361DEFINE_RESORT_RB(threads,
4362 (thread__nr_events(thread__priv(a->thread)) <
4363 thread__nr_events(thread__priv(b->thread))),
96c14451
ACM		 4364 struct thread *thread;
4365)
4366{
7ee227f6 4367 entry->thread = rb_entry(nd, struct thread_rb_node, rb_node)->thread;
896cbb56
DA		 4368}
4369
1302d88e
ACM		 4370static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
4371{
96c14451
ACM		 4372 size_t printed = trace__fprintf_threads_header(fp);
4373 struct rb_node *nd;
91e467bc 4374 int i;
1302d88e 4375
91e467bc
KL		 4376 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
4377 DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
96c14451 4378
91e467bc
KL		 4379 if (threads == NULL) {
4380 fprintf(fp, "%s", "Error sorting output by nr_events!\n");
4381 return 0;
4382 }
896cbb56 4383
91e467bc
KL		 4384 resort_rb__for_each_entry(nd, threads)
4385 printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
96c14451 4386
91e467bc
KL		 4387 resort_rb__delete(threads);
4388 }
96c14451 4389 return printed;
1302d88e
ACM		 4390}
4391
ae9ed035
ACM		 4392static int trace__set_duration(const struct option *opt, const char *str,
4393 int unset __maybe_unused)
4394{
4395 struct trace *trace = opt->value;
4396
4397 trace->duration_filter = atof(str);
4398 return 0;
4399}
4400
6a0b3aba
ACM		 4401static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
4402 int unset __maybe_unused)
f078c385
ACM		 4403{
4404 int ret = -1;
4405 size_t i;
4406 struct trace *trace = opt->value;
4407 /*
4408 * FIXME: introduce a intarray class, plain parse csv and create a
4409 * { int nr, int entries[] } struct...
4410 */
4411 struct intlist *list = intlist__new(str);
4412
4413 if (list == NULL)
4414 return -1;
4415
4416 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
4417 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
4418
4419 if (trace->filter_pids.entries == NULL)
4420 goto out;
4421
4422 trace->filter_pids.entries[0] = getpid();
4423
4424 for (i = 1; i < trace->filter_pids.nr; ++i)
4425 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
4426
4427 intlist__delete(list);
4428 ret = 0;
4429out:
4430 return ret;
4431}
4432
c24ff998
ACM		 4433static int trace__open_output(struct trace *trace, const char *filename)
4434{
4435 struct stat st;
4436
4437 if (!stat(filename, &st) && st.st_size) {
4438 char oldname[PATH_MAX];
4439
4440 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
4441 unlink(oldname);
4442 rename(filename, oldname);
4443 }
4444
4445 trace->output = fopen(filename, "w");
4446
4447 return trace->output == NULL ? -errno : 0;
4448}
4449
598d02c5
SF		 4450static int parse_pagefaults(const struct option *opt, const char *str,
4451 int unset __maybe_unused)
4452{
4453 int *trace_pgfaults = opt->value;
4454
4455 if (strcmp(str, "all") == 0)
4456 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
4457 else if (strcmp(str, "maj") == 0)
4458 *trace_pgfaults |= TRACE_PFMAJ;
4459 else if (strcmp(str, "min") == 0)
4460 *trace_pgfaults |= TRACE_PFMIN;
4461 else
4462 return -1;
4463
4464 return 0;
4465}
4466
206d635a 4467static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)
14a052df 4468{
32dcd021 4469 struct evsel *evsel;
14a052df 4470
206d635a
ACM		 4471 evlist__for_each_entry(evlist, evsel) {
4472 if (evsel->handler == NULL)
4473 evsel->handler = handler;
4474 }
14a052df
ACM		 4475}
4476
362222f8
ACM		 4477static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name)
4478{
4479 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
4480
4481 if (fmt) {
60995604 4482 const struct syscall_fmt *scfmt = syscall_fmt__find(name);
362222f8
ACM		 4483
4484 if (scfmt) {
4485 int skip = 0;
4486
4487 if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 ||
4488 strcmp(evsel->tp_format->format.fields->name, "nr") == 0)
4489 ++skip;
4490
4491 memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt));
4492 }
4493 }
4494}
4495
63503dba 4496static int evlist__set_syscall_tp_fields(struct evlist *evlist)
d32855fa 4497{
32dcd021 4498 struct evsel *evsel;
d32855fa
ACM		 4499
4500 evlist__for_each_entry(evlist, evsel) {
4501 if (evsel->priv || !evsel->tp_format)
4502 continue;
4503
947b843c 4504 if (strcmp(evsel->tp_format->system, "syscalls")) {
794bca26 4505 evsel__init_tp_arg_scnprintf(evsel);
d32855fa 4506 continue;
947b843c 4507 }
d32855fa 4508
794bca26 4509 if (evsel__init_syscall_tp(evsel))
d32855fa
ACM		 4510 return -1;
4511
4512 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
fecd9907 4513 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
d32855fa
ACM		 4514
4515 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
4516 return -1;
362222f8
ACM		 4517
4518 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1);
d32855fa 4519 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
fecd9907 4520 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
d32855fa
ACM		 4521
4522 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
4523 return -1;
362222f8
ACM		 4524
4525 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1);
d32855fa
ACM		 4526 }
4527 }
4528
4529 return 0;
4530}
4531
017037ff
ACM		 4532/*
4533 * XXX: Hackish, just splitting the combined -e+--event (syscalls
4534 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
4535 * existing facilities unchanged (trace->ev_qualifier + parse_options()).
4536 *
4537 * It'd be better to introduce a parse_options() variant that would return a
4538 * list with the terms it didn't match to an event...
4539 */
4540static int trace__parse_events_option(const struct option *opt, const char *str,
4541 int unset __maybe_unused)
4542{
4543 struct trace *trace = (struct trace *)opt->value;
4544 const char *s = str;
4545 char *sep = NULL, *lists[2] = { NULL, NULL, };
27702bcf 4546 int len = strlen(str) + 1, err = -1, list, idx;
017037ff
ACM		 4547 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
4548 char group_name[PATH_MAX];
60995604 4549 const struct syscall_fmt *fmt;
017037ff
ACM		 4550
4551 if (strace_groups_dir == NULL)
4552 return -1;
4553
4554 if (*s == '!') {
4555 ++s;
4556 trace->not_ev_qualifier = true;
4557 }
4558
4559 while (1) {
4560 if ((sep = strchr(s, ',')) != NULL)
4561 *sep = '\0';
4562
4563 list = 0;
27702bcf
ACM		 4564 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
4565 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
017037ff 4566 list = 1;
f932184e
ACM		 4567 goto do_concat;
4568 }
4569
4570 fmt = syscall_fmt__find_by_alias(s);
4571 if (fmt != NULL) {
4572 list = 1;
4573 s = fmt->name;
017037ff
ACM		 4574 } else {
4575 path__join(group_name, sizeof(group_name), strace_groups_dir, s);
4576 if (access(group_name, R_OK) == 0)
4577 list = 1;
4578 }
f932184e 4579do_concat:
017037ff
ACM		 4580 if (lists[list]) {
4581 sprintf(lists[list] + strlen(lists[list]), ",%s", s);
4582 } else {
4583 lists[list] = malloc(len);
4584 if (lists[list] == NULL)
4585 goto out;
4586 strcpy(lists[list], s);
4587 }
4588
4589 if (!sep)
4590 break;
4591
4592 *sep = ',';
4593 s = sep + 1;
4594 }
4595
4596 if (lists[1] != NULL) {
4597 struct strlist_config slist_config = {
4598 .dirname = strace_groups_dir,
4599 };
4600
4601 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4602 if (trace->ev_qualifier == NULL) {
4603 fputs("Not enough memory to parse event qualifier", trace->output);
4604 goto out;
4605 }
4606
4607 if (trace__validate_ev_qualifier(trace))
4608 goto out;
b912885a 4609 trace->trace_syscalls = true;
017037ff
ACM		 4610 }
4611
4612 err = 0;
4613
4614 if (lists[0]) {
411ad22e
IR		 4615 struct parse_events_option_args parse_events_option_args = {
4616 .evlistp = &trace->evlist,
4617 };
a6293f36 4618 struct option o = {
411ad22e 4619 .value = &parse_events_option_args,
a6293f36 4620 };
017037ff
ACM		 4621 err = parse_events_option(&o, lists[0], 0);
4622 }
4623out:
659ede7d
RM		 4624 free(strace_groups_dir);
4625 free(lists[0]);
4626 free(lists[1]);
017037ff
ACM		 4627 if (sep)
4628 *sep = ',';
4629
4630 return err;
4631}
4632
9ea42ba4
ACM		 4633static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4634{
4635 struct trace *trace = opt->value;
4636
a6293f36
SI		 4637 if (!list_empty(&trace->evlist->core.entries)) {
4638 struct option o = {
4639 .value = &trace->evlist,
4640 };
4641 return parse_cgroups(&o, str, unset);
4642 }
9ea42ba4
ACM		 4643 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
4644
4645 return 0;
4646}
4647
ac96287c
ACM		 4648static int trace__config(const char *var, const char *value, void *arg)
4649{
e7c634fc 4650 struct trace *trace = arg;
ac96287c
ACM		 4651 int err = 0;
4652
4653 if (!strcmp(var, "trace.add_events")) {
7e035929
ACM		 4654 trace->perfconfig_events = strdup(value);
4655 if (trace->perfconfig_events == NULL) {
4656 pr_err("Not enough memory for %s\n", "trace.add_events");
4657 return -1;
4658 }
b036146f
ACM		 4659 } else if (!strcmp(var, "trace.show_timestamp")) {
4660 trace->show_tstamp = perf_config_bool(var, value);
42e4a52d
ACM		 4661 } else if (!strcmp(var, "trace.show_duration")) {
4662 trace->show_duration = perf_config_bool(var, value);
9d6dc178
ACM		 4663 } else if (!strcmp(var, "trace.show_arg_names")) {
4664 trace->show_arg_names = perf_config_bool(var, value);
4665 if (!trace->show_arg_names)
4666 trace->show_zeros = true;
e7c634fc 4667 } else if (!strcmp(var, "trace.show_zeros")) {
9d6dc178
ACM		 4668 bool new_show_zeros = perf_config_bool(var, value);
4669 if (!trace->show_arg_names && !new_show_zeros) {
4670 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4671 goto out;
4672 }
4673 trace->show_zeros = new_show_zeros;
c65c83ff
ACM		 4674 } else if (!strcmp(var, "trace.show_prefix")) {
4675 trace->show_string_prefix = perf_config_bool(var, value);
d32de87e
ACM		 4676 } else if (!strcmp(var, "trace.no_inherit")) {
4677 trace->opts.no_inherit = perf_config_bool(var, value);
9ed45d59
ACM		 4678 } else if (!strcmp(var, "trace.args_alignment")) {
4679 int args_alignment = 0;
4680 if (perf_config_int(&args_alignment, var, value) == 0)
4681 trace->args_alignment = args_alignment;
f11b2803
ACM		 4682 } else if (!strcmp(var, "trace.tracepoint_beautifiers")) {
4683 if (strcasecmp(value, "libtraceevent") == 0)
4684 trace->libtraceevent_print = true;
4685 else if (strcasecmp(value, "libbeauty") == 0)
4686 trace->libtraceevent_print = false;
ac96287c 4687 }
9d6dc178 4688out:
ac96287c
ACM		 4689 return err;
4690}
4691
6c7f0ab0
RM		 4692static void trace__exit(struct trace *trace)
4693{
f2ebf8ff
RM		 4694 int i;
4695
6c7f0ab0 4696 strlist__delete(trace->ev_qualifier);
9997d5dd 4697 zfree(&trace->ev_qualifier_ids.entries);
f2ebf8ff
RM		 4698 if (trace->syscalls.table) {
4699 for (i = 0; i <= trace->sctbl->syscalls.max_id; i++)
4700 syscall__exit(&trace->syscalls.table[i]);
9997d5dd 4701 zfree(&trace->syscalls.table);
f2ebf8ff 4702 }
6c7f0ab0
RM		 4703 syscalltbl__delete(trace->sctbl);
4704 zfree(&trace->perfconfig_events);
4705}
4706
b0ad8ea6 4707int cmd_trace(int argc, const char **argv)
514f1c67 4708{
6fdd9cb7 4709 const char *trace_usage[] = {
f15eb531
NK		 4710 "perf trace [<options>] [<command>]",
4711 "perf trace [<options>] -- <command> [<options>]",
5e2485b1
DA		 4712 "perf trace record [<options>] [<command>]",
4713 "perf trace record [<options>] -- <command> [<options>]",
514f1c67
ACM		 4714 NULL
4715 };
4716 struct trace trace = {
514f1c67
ACM		 4717 .opts = {
4718 .target = {
4719 .uid = UINT_MAX,
4720 .uses_mmap = true,
4721 },
4722 .user_freq = UINT_MAX,
4723 .user_interval = ULLONG_MAX,
509051ea 4724 .no_buffering = true,
38d5447d 4725 .mmap_pages = UINT_MAX,
514f1c67 4726 },
007d66a0 4727 .output = stderr,
50c95cbd 4728 .show_comm = true,
b036146f 4729 .show_tstamp = true,
42e4a52d 4730 .show_duration = true,
9d6dc178 4731 .show_arg_names = true,
9ed45d59 4732 .args_alignment = 70,
b912885a 4733 .trace_syscalls = false,
44621819 4734 .kernel_syscallchains = false,
05614993 4735 .max_stack = UINT_MAX,
5067a8cd 4736 .max_events = ULONG_MAX,
514f1c67 4737 };
ff7a4f98 4738 const char *map_dump_str = NULL;
c24ff998 4739 const char *output_name = NULL;
514f1c67 4740 const struct option trace_options[] = {
017037ff
ACM		 4741 OPT_CALLBACK('e', "event", &trace, "event",
4742 "event/syscall selector. use 'perf list' to list available events",
4743 trace__parse_events_option),
d4097f19
ACM		 4744 OPT_CALLBACK(0, "filter", &trace.evlist, "filter",
4745 "event filter", parse_filter),
50c95cbd
ACM		 4746 OPT_BOOLEAN(0, "comm", &trace.show_comm,
4747 "show the thread COMM next to its id"),
c522739d 4748 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
017037ff
ACM		 4749 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4750 trace__parse_events_option),
c24ff998 4751 OPT_STRING('o', "output", &output_name, "file", "output file name"),
6810fc91 4752 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
514f1c67
ACM		 4753 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4754 "trace events on existing process id"),
ac9be8ee 4755 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
514f1c67 4756 "trace events on existing thread id"),
fa0e4ffe 4757 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
6a0b3aba 4758 "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
ac9be8ee 4759 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
514f1c67 4760 "system-wide collection from all CPUs"),
ac9be8ee 4761 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
514f1c67 4762 "list of cpus to monitor"),
6810fc91 4763 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
514f1c67 4764 "child tasks do not inherit counters"),
994a1f78 4765 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
25f84702 4766 "number of mmap data pages", evlist__parse_mmap_pages),
ac9be8ee 4767 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
514f1c67 4768 "user to profile"),
ae9ed035
ACM		 4769 OPT_CALLBACK(0, "duration", &trace, "float",
4770 "show only events with duration > N.M ms",
4771 trace__set_duration),
ff7a4f98
ACM		 4772#ifdef HAVE_LIBBPF_SUPPORT
4773 OPT_STRING(0, "map-dump", &map_dump_str, "BPF map", "BPF map to periodically dump"),
4774#endif
1302d88e 4775 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
7c304ee0 4776 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4bb09192
DA		 4777 OPT_BOOLEAN('T', "time", &trace.full_time,
4778 "Show full timestamp, not time relative to first start"),
0a6545bd
ACM		 4779 OPT_BOOLEAN(0, "failure", &trace.failure_only,
4780 "Show only syscalls that failed"),
fd2eabaf
DA		 4781 OPT_BOOLEAN('s', "summary", &trace.summary_only,
4782 "Show only syscall summary with statistics"),
4783 OPT_BOOLEAN('S', "with-summary", &trace.summary,
4784 "Show all syscalls and summary with statistics"),
b88b14db
ACM		 4785 OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary,
4786 "Show errno stats per syscall, use with -s or -S"),
598d02c5
SF		 4787 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4788 "Trace pagefaults", parse_pagefaults, "maj"),
e281a960 4789 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
e366a6d8 4790 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
566a0885
MW		 4791 OPT_CALLBACK(0, "call-graph", &trace.opts,
4792 "record_mode[,record_size]", record_callchain_help,
4793 &record_parse_callchain_opt),
f11b2803
ACM		 4794 OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,
4795 "Use libtraceevent to print the tracepoint arguments."),
44621819
ACM		 4796 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4797 "Show the kernel callchains on the syscall exit path"),
5067a8cd
ACM		 4798 OPT_ULONG(0, "max-events", &trace.max_events,
4799 "Set the maximum number of events to print, exit after that is reached. "),
5cf9c84e
ACM		 4800 OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4801 "Set the minimum stack depth when parsing the callchain, "
4802 "anything below the specified depth will be ignored."),
c6d4a494
ACM		 4803 OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4804 "Set the maximum stack depth when parsing the callchain, "
4805 "anything beyond the specified depth will be ignored. "
4cb93446 4806 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
a761a8d1
ACM		 4807 OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4808 "Sort batch of events before processing, use if getting out of order events"),
591421e1
ACM		 4809 OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4810 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
3fcb10e4 4811 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
9d9cad76 4812 "per thread proc mmap processing timeout in ms"),
9ea42ba4
ACM		 4813 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4814 trace__parse_cgroups),
cb4b9e68 4815 OPT_INTEGER('D', "delay", &trace.opts.target.initial_delay,
e36b7821
AB		 4816 "ms to wait before starting measurement after program "
4817 "start"),
22ac4318 4818 OPTS_EVSWITCH(&trace.evswitch),
514f1c67
ACM		 4819 OPT_END()
4820 };
ccd62a89 4821 bool __maybe_unused max_stack_user_set = true;
f3e459d1 4822 bool mmap_pages_user_set = true;
32dcd021 4823 struct evsel *evsel;
6fdd9cb7 4824 const char * const trace_subcommands[] = { "record", NULL };
78e890ea 4825 int err = -1;
32caf0d1 4826 char bf[BUFSIZ];
de9f498d 4827 struct sigaction sigchld_act;
514f1c67 4828
4d08cb80
ACM		 4829 signal(SIGSEGV, sighandler_dump_stack);
4830 signal(SIGFPE, sighandler_dump_stack);
de9f498d
CD		 4831 signal(SIGINT, sighandler_interrupt);
4832
4833 memset(&sigchld_act, 0, sizeof(sigchld_act));
4834 sigchld_act.sa_flags = SA_SIGINFO;
4835 sigchld_act.sa_sigaction = sighandler_chld;
4836 sigaction(SIGCHLD, &sigchld_act, NULL);
4d08cb80 4837
0f98b11c 4838 trace.evlist = evlist__new();
fd0db102 4839 trace.sctbl = syscalltbl__new();
14a052df 4840
fd0db102 4841 if (trace.evlist == NULL || trace.sctbl == NULL) {
14a052df 4842 pr_err("Not enough memory to run!\n");
ff8f695c 4843 err = -ENOMEM;
14a052df
ACM		 4844 goto out;
4845 }
4846
c3e78a34
ACM		 4847 /*
4848 * Parsing .perfconfig may entail creating a BPF event, that may need
4849 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4850 * is too small. This affects just this process, not touching the
4851 * global setting. If it fails we'll get something in 'perf trace -v'
4852 * to help diagnose the problem.
4853 */
4854 rlimit__bump_memlock();
4855
ac96287c
ACM		 4856 err = perf_config(trace__config, &trace);
4857 if (err)
4858 goto out;
4859
6fdd9cb7
YS		 4860 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4861 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
8bd436b0
ACM		 4862
4863 /*
4864 * Here we already passed thru trace__parse_events_option() and it has
4865 * already figured out if -e syscall_name, if not but if --event
4866 * foo:bar was used, the user is interested _just_ in those, say,
4867 * tracepoint events, not in the strace-like syscall-name-based mode.
4868 *
4869 * This is important because we need to check if strace-like mode is
4870 * needed to decided if we should filter out the eBPF
4871 * __augmented_syscalls__ code, if it is in the mix, say, via
4872 * .perfconfig trace.add_events, and filter those out.
4873 */
4874 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4875 trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4876 trace.trace_syscalls = true;
4877 }
7e035929
ACM		 4878 /*
4879 * Now that we have --verbose figured out, lets see if we need to parse
4880 * events from .perfconfig, so that if those events fail parsing, say some
4881 * BPF program fails, then we'll be able to use --verbose to see what went
4882 * wrong in more detail.
4883 */
4884 if (trace.perfconfig_events != NULL) {
a910e466 4885 struct parse_events_error parse_err;
7e035929 4886
07eafd4e 4887 parse_events_error__init(&parse_err);
7e035929 4888 err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);
07eafd4e 4889 if (err)
6c191289 4890 parse_events_error__print(&parse_err, trace.perfconfig_events);
07eafd4e
IR		 4891 parse_events_error__exit(&parse_err);
4892 if (err)
7e035929 4893 goto out;
7e035929 4894 }
fd2eabaf 4895
9ea42ba4
ACM		 4896 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4897 usage_with_options_msg(trace_usage, trace_options,
4898 "cgroup monitoring only available in system-wide mode");
4899 }
4900
78e890ea
ACM		 4901 evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
4902 if (IS_ERR(evsel)) {
4903 bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
e0b6d2ef
ACM		 4904 pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
4905 goto out;
4906 }
4907
744fafc7 4908 if (evsel) {
d3d1c4bd 4909 trace.syscalls.events.augmented = evsel;
c8c80570 4910
b02736f7 4911 evsel = evlist__find_tracepoint_by_name(trace.evlist, "raw_syscalls:sys_enter");
c8c80570
ACM		 4912 if (evsel == NULL) {
4913 pr_err("ERROR: raw_syscalls:sys_enter not found in the augmented BPF object\n");
4914 goto out;
4915 }
4916
4917 if (evsel->bpf_obj == NULL) {
4918 pr_err("ERROR: raw_syscalls:sys_enter not associated to a BPF object\n");
4919 goto out;
4920 }
4921
4922 trace.bpf_obj = evsel->bpf_obj;
4923
8bd436b0
ACM		 4924 /*
4925 * If we have _just_ the augmenter event but don't have a
4926 * explicit --syscalls, then assume we want all strace-like
4927 * syscalls:
4928 */
4929 if (!trace.trace_syscalls && trace__only_augmented_syscalls_evsels(&trace))
4930 trace.trace_syscalls = true;
4931 /*
4932 * So, if we have a syscall augmenter, but trace_syscalls, aka
4933 * strace-like syscall tracing is not set, then we need to trow
4934 * away the augmenter, i.e. all the events that were created
4935 * from that BPF object file.
4936 *
4937 * This is more to fix the current .perfconfig trace.add_events
4938 * style of setting up the strace-like eBPF based syscall point
4939 * payload augmenter.
4940 *
4941 * All this complexity will be avoided by adding an alternative
4942 * to trace.add_events in the form of
4943 * trace.bpf_augmented_syscalls, that will be only parsed if we
4944 * need it.
4945 *
4946 * .perfconfig trace.add_events is still useful if we want, for
4947 * instance, have msr_write.msr in some .perfconfig profile based
4948 * 'perf trace --config determinism.profile' mode, where for some
4949 * particular goal/workload type we want a set of events and
4950 * output mode (with timings, etc) instead of having to add
4951 * all via the command line.
4952 *
4953 * Also --config to specify an alternate .perfconfig file needs
4954 * to be implemented.
4955 */
4956 if (!trace.trace_syscalls) {
4957 trace__delete_augmented_syscalls(&trace);
4958 } else {
4959 trace__set_bpf_map_filtered_pids(&trace);
4960 trace__set_bpf_map_syscalls(&trace);
4961 trace.syscalls.unaugmented_prog = trace__find_bpf_program_by_title(&trace, "!raw_syscalls:unaugmented");
4962 }
744fafc7 4963 }
d3d1c4bd 4964
d7888573
WN		 4965 err = bpf__setup_stdout(trace.evlist);
4966 if (err) {
4967 bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
4968 pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
4969 goto out;
4970 }
4971
59247e33
ACM		 4972 err = -1;
4973
ff7a4f98 4974 if (map_dump_str) {
5ca0b7f5 4975 trace.dump.map = trace__find_bpf_map_by_name(&trace, map_dump_str);
ff7a4f98
ACM		 4976 if (trace.dump.map == NULL) {
4977 pr_err("ERROR: BPF map \"%s\" not found\n", map_dump_str);
4978 goto out;
4979 }
4980 }
4981
598d02c5
SF		 4982 if (trace.trace_pgfaults) {
4983 trace.opts.sample_address = true;
4984 trace.opts.sample_time = true;
4985 }
4986
f3e459d1
ACM		 4987 if (trace.opts.mmap_pages == UINT_MAX)
4988 mmap_pages_user_set = false;
4989
05614993 4990 if (trace.max_stack == UINT_MAX) {
029c75e5 4991 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
05614993
ACM		 4992 max_stack_user_set = false;
4993 }
4994
4995#ifdef HAVE_DWARF_UNWIND_SUPPORT
75d50117 4996 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
05614993 4997 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
75d50117 4998 }
05614993
ACM		 4999#endif
5000
2ddd5c04 5001 if (callchain_param.enabled) {
f3e459d1
ACM		 5002 if (!mmap_pages_user_set && geteuid() == 0)
5003 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
5004
566a0885 5005 symbol_conf.use_callchain = true;
f3e459d1 5006 }
566a0885 5007
6484d2f9 5008 if (trace.evlist->core.nr_entries > 0) {
206d635a 5009 evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);
d32855fa
ACM		 5010 if (evlist__set_syscall_tp_fields(trace.evlist)) {
5011 perror("failed to set syscalls:* tracepoint fields");
5012 goto out;
5013 }
5014 }
14a052df 5015
a761a8d1
ACM		 5016 if (trace.sort_events) {
5017 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
5018 ordered_events__set_copy_on_queue(&trace.oe.data, true);
5019 }
028713aa 5020
b1a9e253
ACM		 5021 /*
5022 * If we are augmenting syscalls, then combine what we put in the
5023 * __augmented_syscalls__ BPF map with what is in the
5024 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
5025 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
5026 *
5027 * We'll switch to look at two BPF maps, one for sys_enter and the
5028 * other for sys_exit when we start augmenting the sys_exit paths with
5029 * buffers that are being copied from kernel to userspace, think 'read'
5030 * syscall.
5031 */
5032 if (trace.syscalls.events.augmented) {
b1a9e253 5033 evlist__for_each_entry(trace.evlist, evsel) {
8ab2e96d 5034 bool raw_syscalls_sys_exit = strcmp(evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
cd26ea6d
ACM		 5035
5036 if (raw_syscalls_sys_exit) {
5037 trace.raw_augmented_syscalls = true;
5038 goto init_augmented_syscall_tp;
5039 }
5040
1d59cb1b 5041 if (trace.syscalls.events.augmented->priv == NULL &&
8ab2e96d 5042 strstr(evsel__name(evsel), "syscalls:sys_enter")) {
32dcd021 5043 struct evsel *augmented = trace.syscalls.events.augmented;
794bca26
ACM		 5044 if (evsel__init_augmented_syscall_tp(augmented, evsel) ||
5045 evsel__init_augmented_syscall_tp_args(augmented))
b9b6a2ea 5046 goto out;
b119970a
ACM		 5047 /*
5048 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
5049 * Above we made sure we can get from the payload the tp fields
5050 * that we get from syscalls:sys_enter tracefs format file.
5051 */
b9b6a2ea 5052 augmented->handler = trace__sys_enter;
b119970a
ACM		 5053 /*
5054 * Now we do the same for the *syscalls:sys_enter event so that
5055 * if we handle it directly, i.e. if the BPF prog returns 0 so
5056 * as not to filter it, then we'll handle it just like we would
5057 * for the BPF_OUTPUT one:
5058 */
794bca26
ACM		 5059 if (evsel__init_augmented_syscall_tp(evsel, evsel) ||
5060 evsel__init_augmented_syscall_tp_args(evsel))
b119970a
ACM		 5061 goto out;
5062 evsel->handler = trace__sys_enter;
b9b6a2ea
ACM		 5063 }
5064
8ab2e96d 5065 if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) {
b9b6a2ea 5066 struct syscall_tp *sc;
cd26ea6d 5067init_augmented_syscall_tp:
794bca26 5068 if (evsel__init_augmented_syscall_tp(evsel, evsel))
b9b6a2ea 5069 goto out;
fecd9907 5070 sc = __evsel__syscall_tp(evsel);
b9b6a2ea
ACM		 5071 /*
5072 * For now with BPF raw_augmented we hook into
5073 * raw_syscalls:sys_enter and there we get all
5074 * 6 syscall args plus the tracepoint common
5075 * fields and the syscall_nr (another long).
5076 * So we check if that is the case and if so
5077 * don't look after the sc->args_size but
5078 * always after the full raw_syscalls:sys_enter
5079 * payload, which is fixed.
5080 *
5081 * We'll revisit this later to pass
5082 * s->args_size to the BPF augmenter (now
5083 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
5084 * so that it copies only what we need for each
5085 * syscall, like what happens when we use
5086 * syscalls:sys_enter_NAME, so that we reduce
5087 * the kernel/userspace traffic to just what is
5088 * needed for each syscall.
5089 */
5090 if (trace.raw_augmented_syscalls)
5091 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
794bca26 5092 evsel__init_augmented_syscall_tp_ret(evsel);
b1a9e253
ACM		 5093 evsel->handler = trace__sys_exit;
5094 }
5095 }
5096 }
5097
1e28fe0a
SF		 5098 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
5099 return trace__record(&trace, argc-1, &argv[1]);
5100
b88b14db
ACM		 5101 /* Using just --errno-summary will trigger --summary */
5102 if (trace.errno_summary && !trace.summary && !trace.summary_only)
5103 trace.summary_only = true;
5104
1e28fe0a
SF		 5105 /* summary_only implies summary option, but don't overwrite summary if set */
5106 if (trace.summary_only)
5107 trace.summary = trace.summary_only;
5108
c24ff998
ACM		 5109 if (output_name != NULL) {
5110 err = trace__open_output(&trace, output_name);
5111 if (err < 0) {
5112 perror("failed to create output file");
5113 goto out;
5114 }
5115 }
5116
22ac4318
ACM		 5117 err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
5118 if (err)
5119 goto out_close;
5120
602ad878 5121 err = target__validate(&trace.opts.target);
32caf0d1 5122 if (err) {
602ad878 5123 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
c24ff998
ACM		 5124 fprintf(trace.output, "%s", bf);
5125 goto out_close;
32caf0d1
NK		 5126 }
5127
602ad878 5128 err = target__parse_uid(&trace.opts.target);
514f1c67 5129 if (err) {
602ad878 5130 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
c24ff998
ACM		 5131 fprintf(trace.output, "%s", bf);
5132 goto out_close;
514f1c67
ACM		 5133 }
5134
602ad878 5135 if (!argc && target__none(&trace.opts.target))
ee76120e
NK		 5136 trace.opts.target.system_wide = true;
5137
6810fc91
DA		 5138 if (input_name)
5139 err = trace__replay(&trace);
5140 else
5141 err = trace__run(&trace, argc, argv);
1302d88e 5142
c24ff998
ACM		 5143out_close:
5144 if (output_name != NULL)
5145 fclose(trace.output);
5146out:
6c7f0ab0 5147 trace__exit(&trace);
1302d88e 5148 return err;
514f1c67 5149}
Linux 6.x block layer and io_uring tree(s)