1 /*
2 * builtin-trace.c
3 *
4 * Builtin 'trace' command:
5 *
6 * Display a continuously updated trace of any workload, CPU, specific PID,
7 * system wide, etc. Default format is loosely strace like, but any other
8 * event may be specified using --event.
9 *
10 * Copyright (C) 2012, 2013, 2014, 2015 Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
11 *
12 * Initially based on the 'trace' prototype by Thomas Gleixner:
13 *
14 * http://lwn.net/Articles/415728/ ("Announcing a new utility: 'trace'")
15 */
16
17 #include "util/record.h"
18 #include <api/fs/tracing_path.h>
19 #ifdef HAVE_LIBBPF_SUPPORT
20 #include <bpf/bpf.h>
21 #include <bpf/libbpf.h>
22 #ifdef HAVE_BPF_SKEL
23 #include "bpf_skel/augmented_raw_syscalls.skel.h"
24 #endif
25 #endif
26 #include "util/bpf_map.h"
27 #include "util/rlimit.h"
28 #include "builtin.h"
29 #include "util/cgroup.h"
30 #include "util/color.h"
31 #include "util/config.h"
32 #include "util/debug.h"
33 #include "util/dso.h"
34 #include "util/env.h"
35 #include "util/event.h"
36 #include "util/evsel.h"
37 #include "util/evsel_fprintf.h"
38 #include "util/synthetic-events.h"
39 #include "util/evlist.h"
40 #include "util/evswitch.h"
41 #include "util/mmap.h"
42 #include <subcmd/pager.h>
43 #include <subcmd/exec-cmd.h>
44 #include "util/machine.h"
45 #include "util/map.h"
46 #include "util/symbol.h"
47 #include "util/path.h"
48 #include "util/session.h"
49 #include "util/thread.h"
50 #include <subcmd/parse-options.h>
51 #include "util/strlist.h"
52 #include "util/intlist.h"
53 #include "util/thread_map.h"
54 #include "util/stat.h"
55 #include "util/tool.h"
56 #include "util/util.h"
57 #include "trace/beauty/beauty.h"
58 #include "trace-event.h"
59 #include "util/parse-events.h"
60 #include "util/tracepoint.h"
61 #include "callchain.h"
62 #include "print_binary.h"
63 #include "string2.h"
64 #include "syscalltbl.h"
65 #include "rb_resort.h"
66 #include "../perf.h"
67
68 #include <errno.h>
69 #include <inttypes.h>
70 #include <poll.h>
71 #include <signal.h>
72 #include <stdlib.h>
73 #include <string.h>
74 #include <linux/err.h>
75 #include <linux/filter.h>
76 #include <linux/kernel.h>
77 #include <linux/list_sort.h>
78 #include <linux/random.h>
79 #include <linux/stringify.h>
80 #include <linux/time64.h>
81 #include <linux/zalloc.h>
82 #include <fcntl.h>
83 #include <sys/sysmacros.h>
84
85 #include <linux/ctype.h>
86 #include <perf/mmap.h>
87
88 #ifdef HAVE_LIBTRACEEVENT
89 #include <traceevent/event-parse.h>
90 #endif
91
92 #ifndef O_CLOEXEC
93 # define O_CLOEXEC 02000000
94 #endif
95
96 #ifndef F_LINUX_SPECIFIC_BASE
97 # define F_LINUX_SPECIFIC_BASE 1024
98 #endif
99
100 #define RAW_SYSCALL_ARGS_NUM 6
101
102 /*
103 * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100
104 */
105 struct syscall_arg_fmt {
106 size_t (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
107 bool (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val);
108 unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
109 void *parm;
110 const char *name;
111 u16 nr_entries; // for arrays
112 bool show_zero;
113 };
114
115 struct syscall_fmt {
116 const char *name;
117 const char *alias;
118 struct {
119 const char *sys_enter,
120 *sys_exit;
121 } bpf_prog_name;
122 struct syscall_arg_fmt arg[RAW_SYSCALL_ARGS_NUM];
123 u8 nr_args;
124 bool errpid;
125 bool timeout;
126 bool hexret;
127 };
128
129 struct trace {
130 struct perf_tool tool;
131 struct syscalltbl *sctbl;
132 struct {
133 struct syscall *table;
134 struct {
135 struct evsel *sys_enter,
136 *sys_exit,
137 *bpf_output;
138 } events;
139 } syscalls;
140 #ifdef HAVE_BPF_SKEL
141 struct augmented_raw_syscalls_bpf *skel;
142 #endif
143 struct record_opts opts;
144 struct evlist *evlist;
145 struct machine *host;
146 struct thread *current;
147 struct cgroup *cgroup;
148 u64 base_time;
149 FILE *output;
150 unsigned long nr_events;
151 unsigned long nr_events_printed;
152 unsigned long max_events;
153 struct evswitch evswitch;
154 struct strlist *ev_qualifier;
155 struct {
156 size_t nr;
157 int *entries;
158 } ev_qualifier_ids;
159 struct {
160 size_t nr;
161 pid_t *entries;
162 struct bpf_map *map;
163 } filter_pids;
164 double duration_filter;
165 double runtime_ms;
166 struct {
167 u64 vfs_getname,
168 proc_getname;
169 } stats;
170 unsigned int max_stack;
171 unsigned int min_stack;
172 int raw_augmented_syscalls_args_size;
173 bool raw_augmented_syscalls;
174 bool fd_path_disabled;
175 bool sort_events;
176 bool not_ev_qualifier;
177 bool live;
178 bool full_time;
179 bool sched;
180 bool multiple_threads;
181 bool summary;
182 bool summary_only;
183 bool errno_summary;
184 bool failure_only;
185 bool show_comm;
186 bool print_sample;
187 bool show_tool_stats;
188 bool trace_syscalls;
189 bool libtraceevent_print;
190 bool kernel_syscallchains;
191 s16 args_alignment;
192 bool show_tstamp;
193 bool show_duration;
194 bool show_zeros;
195 bool show_arg_names;
196 bool show_string_prefix;
197 bool force;
198 bool vfs_getname;
199 int trace_pgfaults;
200 char *perfconfig_events;
201 struct {
202 struct ordered_events data;
203 u64 last;
204 } oe;
205 };
206
207 struct tp_field {
208 int offset;
209 union {
210 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
211 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
212 };
213 };
214
215 #define TP_UINT_FIELD(bits) \
216 static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
217 { \
218 u##bits value; \
219 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
220 return value; \
221 }
222
223 TP_UINT_FIELD(8);
224 TP_UINT_FIELD(16);
225 TP_UINT_FIELD(32);
226 TP_UINT_FIELD(64);
227
228 #define TP_UINT_FIELD__SWAPPED(bits) \
229 static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
230 { \
231 u##bits value; \
232 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
233 return bswap_##bits(value);\
234 }
235
236 TP_UINT_FIELD__SWAPPED(16);
237 TP_UINT_FIELD__SWAPPED(32);
238 TP_UINT_FIELD__SWAPPED(64);
239
240 static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
241 {
242 field->offset = offset;
243
244 switch (size) {
245 case 1:
246 field->integer = tp_field__u8;
247 break;
248 case 2:
249 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
250 break;
251 case 4:
252 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
253 break;
254 case 8:
255 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
256 break;
257 default:
258 return -1;
259 }
260
261 return 0;
262 }
263
264 static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
265 {
266 return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
267 }
268
269 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
270 {
271 return sample->raw_data + field->offset;
272 }
273
274 static int __tp_field__init_ptr(struct tp_field *field, int offset)
275 {
276 field->offset = offset;
277 field->pointer = tp_field__ptr;
278 return 0;
279 }
280
281 static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
282 {
283 return __tp_field__init_ptr(field, format_field->offset);
284 }
285
286 struct syscall_tp {
287 struct tp_field id;
288 union {
289 struct tp_field args, ret;
290 };
291 };
292
293 /*
294 * The evsel->priv as used by 'perf trace'
295 * sc: for raw_syscalls:sys_{enter,exit} and syscalls:sys_{enter,exit}_SYSCALLNAME
296 * fmt: for all the other tracepoints
297 */
298 struct evsel_trace {
299 struct syscall_tp sc;
300 struct syscall_arg_fmt *fmt;
301 };
302
303 static struct evsel_trace *evsel_trace__new(void)
304 {
305 return zalloc(sizeof(struct evsel_trace));
306 }
307
308 static void evsel_trace__delete(struct evsel_trace *et)
309 {
310 if (et == NULL)
311 return;
312
313 zfree(&et->fmt);
314 free(et);
315 }
316
317 /*
318 * Used with raw_syscalls:sys_{enter,exit} and with the
319 * syscalls:sys_{enter,exit}_SYSCALL tracepoints
320 */
321 static inline struct syscall_tp *__evsel__syscall_tp(struct evsel *evsel)
322 {
323 struct evsel_trace *et = evsel->priv;
324
325 return &et->sc;
326 }
327
328 static struct syscall_tp *evsel__syscall_tp(struct evsel *evsel)
329 {
330 if (evsel->priv == NULL) {
331 evsel->priv = evsel_trace__new();
332 if (evsel->priv == NULL)
333 return NULL;
334 }
335
336 return __evsel__syscall_tp(evsel);
337 }
338
339 /*
340 * Used with all the other tracepoints.
341 */
342 static inline struct syscall_arg_fmt *__evsel__syscall_arg_fmt(struct evsel *evsel)
343 {
344 struct evsel_trace *et = evsel->priv;
345
346 return et->fmt;
347 }
348
349 static struct syscall_arg_fmt *evsel__syscall_arg_fmt(struct evsel *evsel)
350 {
351 struct evsel_trace *et = evsel->priv;
352
353 if (evsel->priv == NULL) {
354 et = evsel->priv = evsel_trace__new();
355
356 if (et == NULL)
357 return NULL;
358 }
359
360 if (et->fmt == NULL) {
361 et->fmt = calloc(evsel->tp_format->format.nr_fields, sizeof(struct syscall_arg_fmt));
362 if (et->fmt == NULL)
363 goto out_delete;
364 }
365
366 return __evsel__syscall_arg_fmt(evsel);
367
368 out_delete:
369 evsel_trace__delete(evsel->priv);
370 evsel->priv = NULL;
371 return NULL;
372 }
373
374 static int evsel__init_tp_uint_field(struct evsel *evsel, struct tp_field *field, const char *name)
375 {
376 struct tep_format_field *format_field = evsel__field(evsel, name);
377
378 if (format_field == NULL)
379 return -1;
380
381 return tp_field__init_uint(field, format_field, evsel->needs_swap);
382 }
383
384 #define perf_evsel__init_sc_tp_uint_field(evsel, name) \
385 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
386 evsel__init_tp_uint_field(evsel, &sc->name, #name); })
387
388 static int evsel__init_tp_ptr_field(struct evsel *evsel, struct tp_field *field, const char *name)
389 {
390 struct tep_format_field *format_field = evsel__field(evsel, name);
391
392 if (format_field == NULL)
393 return -1;
394
395 return tp_field__init_ptr(field, format_field);
396 }
397
398 #define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
399 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
400 evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
401
402 static void evsel__delete_priv(struct evsel *evsel)
403 {
404 zfree(&evsel->priv);
405 evsel__delete(evsel);
406 }
407
408 static int evsel__init_syscall_tp(struct evsel *evsel)
409 {
410 struct syscall_tp *sc = evsel__syscall_tp(evsel);
411
412 if (sc != NULL) {
413 if (evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
414 evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
415 return -ENOENT;
416
417 return 0;
418 }
419
420 return -ENOMEM;
421 }
422
423 static int evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)
424 {
425 struct syscall_tp *sc = evsel__syscall_tp(evsel);
426
427 if (sc != NULL) {
428 struct tep_format_field *syscall_id = evsel__field(tp, "id");
429 if (syscall_id == NULL)
430 syscall_id = evsel__field(tp, "__syscall_nr");
431 if (syscall_id == NULL ||
432 __tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
433 return -EINVAL;
434
435 return 0;
436 }
437
438 return -ENOMEM;
439 }
440
441 static int evsel__init_augmented_syscall_tp_args(struct evsel *evsel)
442 {
443 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
444
445 return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
446 }
447
448 static int evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)
449 {
450 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
451
452 return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
453 }
454
455 static int evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)
456 {
457 if (evsel__syscall_tp(evsel) != NULL) {
458 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
459 return -ENOENT;
460
461 evsel->handler = handler;
462 return 0;
463 }
464
465 return -ENOMEM;
466 }
467
468 static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
469 {
470 struct evsel *evsel = evsel__newtp("raw_syscalls", direction);
471
472 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
473 if (IS_ERR(evsel))
474 evsel = evsel__newtp("syscalls", direction);
475
476 if (IS_ERR(evsel))
477 return NULL;
478
479 if (evsel__init_raw_syscall_tp(evsel, handler))
480 goto out_delete;
481
482 return evsel;
483
484 out_delete:
485 evsel__delete_priv(evsel);
486 return NULL;
487 }
488
489 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
490 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
491 fields->name.integer(&fields->name, sample); })
492
493 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
494 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
495 fields->name.pointer(&fields->name, sample); })
496
497 size_t strarray__scnprintf_suffix(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_suffix, int val)
498 {
499 int idx = val - sa->offset;
500
501 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
502 size_t printed = scnprintf(bf, size, intfmt, val);
503 if (show_suffix)
504 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
505 return printed;
506 }
507
508 return scnprintf(bf, size, "%s%s", sa->entries[idx], show_suffix ? sa->prefix : "");
509 }
510
511 size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
512 {
513 int idx = val - sa->offset;
514
515 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
516 size_t printed = scnprintf(bf, size, intfmt, val);
517 if (show_prefix)
518 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
519 return printed;
520 }
521
522 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
523 }
524
525 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
526 const char *intfmt,
527 struct syscall_arg *arg)
528 {
529 return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
530 }
531
532 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
533 struct syscall_arg *arg)
534 {
535 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
536 }
537
538 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
539
540 bool syscall_arg__strtoul_strarray(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
541 {
542 return strarray__strtoul(arg->parm, bf, size, ret);
543 }
544
545 bool syscall_arg__strtoul_strarray_flags(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
546 {
547 return strarray__strtoul_flags(arg->parm, bf, size, ret);
548 }
549
550 bool syscall_arg__strtoul_strarrays(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
551 {
552 return strarrays__strtoul(arg->parm, bf, size, ret);
553 }
554
555 size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)
556 {
557 return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);
558 }
559
560 size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
561 {
562 size_t printed;
563 int i;
564
565 for (i = 0; i < sas->nr_entries; ++i) {
566 struct strarray *sa = sas->entries[i];
567 int idx = val - sa->offset;
568
569 if (idx >= 0 && idx < sa->nr_entries) {
570 if (sa->entries[idx] == NULL)
571 break;
572 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
573 }
574 }
575
576 printed = scnprintf(bf, size, intfmt, val);
577 if (show_prefix)
578 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
579 return printed;
580 }
581
582 bool strarray__strtoul(struct strarray *sa, char *bf, size_t size, u64 *ret)
583 {
584 int i;
585
586 for (i = 0; i < sa->nr_entries; ++i) {
587 if (sa->entries[i] && strncmp(sa->entries[i], bf, size) == 0 && sa->entries[i][size] == '\0') {
588 *ret = sa->offset + i;
589 return true;
590 }
591 }
592
593 return false;
594 }
595
596 bool strarray__strtoul_flags(struct strarray *sa, char *bf, size_t size, u64 *ret)
597 {
598 u64 val = 0;
599 char *tok = bf, *sep, *end;
600
601 *ret = 0;
602
603 while (size != 0) {
604 int toklen = size;
605
606 sep = memchr(tok, '|', size);
607 if (sep != NULL) {
608 size -= sep - tok + 1;
609
610 end = sep - 1;
611 while (end > tok && isspace(*end))
612 --end;
613
614 toklen = end - tok + 1;
615 }
616
617 while (isspace(*tok))
618 ++tok;
619
620 if (isalpha(*tok) || *tok == '_') {
621 if (!strarray__strtoul(sa, tok, toklen, &val))
622 return false;
623 } else
624 val = strtoul(tok, NULL, 0);
625
626 *ret |= (1 << (val - 1));
627
628 if (sep == NULL)
629 break;
630 tok = sep + 1;
631 }
632
633 return true;
634 }
635
636 bool strarrays__strtoul(struct strarrays *sas, char *bf, size_t size, u64 *ret)
637 {
638 int i;
639
640 for (i = 0; i < sas->nr_entries; ++i) {
641 struct strarray *sa = sas->entries[i];
642
643 if (strarray__strtoul(sa, bf, size, ret))
644 return true;
645 }
646
647 return false;
648 }
649
650 size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
651 struct syscall_arg *arg)
652 {
653 return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
654 }
655
656 #ifndef AT_FDCWD
657 #define AT_FDCWD -100
658 #endif
659
660 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
661 struct syscall_arg *arg)
662 {
663 int fd = arg->val;
664 const char *prefix = "AT_FD";
665
666 if (fd == AT_FDCWD)
667 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
668
669 return syscall_arg__scnprintf_fd(bf, size, arg);
670 }
671
672 #define SCA_FDAT syscall_arg__scnprintf_fd_at
673
674 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
675 struct syscall_arg *arg);
676
677 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
678
679 size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
680 {
681 return scnprintf(bf, size, "%#lx", arg->val);
682 }
683
684 size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
685 {
686 if (arg->val == 0)
687 return scnprintf(bf, size, "NULL");
688 return syscall_arg__scnprintf_hex(bf, size, arg);
689 }
690
691 size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
692 {
693 return scnprintf(bf, size, "%d", arg->val);
694 }
695
696 size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
697 {
698 return scnprintf(bf, size, "%ld", arg->val);
699 }
700
701 static size_t syscall_arg__scnprintf_char_array(char *bf, size_t size, struct syscall_arg *arg)
702 {
703 // XXX Hey, maybe for sched:sched_switch prev/next comm fields we can
704 // fill missing comms using thread__set_comm()...
705 // here or in a special syscall_arg__scnprintf_pid_sched_tp...
706 return scnprintf(bf, size, "\"%-.*s\"", arg->fmt->nr_entries ?: arg->len, arg->val);
707 }
708
709 #define SCA_CHAR_ARRAY syscall_arg__scnprintf_char_array
710
711 static const char *bpf_cmd[] = {
712 "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
713 "MAP_GET_NEXT_KEY", "PROG_LOAD", "OBJ_PIN", "OBJ_GET", "PROG_ATTACH",
714 "PROG_DETACH", "PROG_TEST_RUN", "PROG_GET_NEXT_ID", "MAP_GET_NEXT_ID",
715 "PROG_GET_FD_BY_ID", "MAP_GET_FD_BY_ID", "OBJ_GET_INFO_BY_FD",
716 "PROG_QUERY", "RAW_TRACEPOINT_OPEN", "BTF_LOAD", "BTF_GET_FD_BY_ID",
717 "TASK_FD_QUERY", "MAP_LOOKUP_AND_DELETE_ELEM", "MAP_FREEZE",
718 "BTF_GET_NEXT_ID", "MAP_LOOKUP_BATCH", "MAP_LOOKUP_AND_DELETE_BATCH",
719 "MAP_UPDATE_BATCH", "MAP_DELETE_BATCH", "LINK_CREATE", "LINK_UPDATE",
720 "LINK_GET_FD_BY_ID", "LINK_GET_NEXT_ID", "ENABLE_STATS", "ITER_CREATE",
721 "LINK_DETACH", "PROG_BIND_MAP",
722 };
723 static DEFINE_STRARRAY(bpf_cmd, "BPF_");
724
725 static const char *fsmount_flags[] = {
726 [1] = "CLOEXEC",
727 };
728 static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");
729
730 #include "trace/beauty/generated/fsconfig_arrays.c"
731
732 static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");
733
734 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
735 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
736
737 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
738 static DEFINE_STRARRAY(itimers, "ITIMER_");
739
740 static const char *keyctl_options[] = {
741 "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
742 "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
743 "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
744 "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
745 "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
746 };
747 static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
748
749 static const char *whences[] = { "SET", "CUR", "END",
750 #ifdef SEEK_DATA
751 "DATA",
752 #endif
753 #ifdef SEEK_HOLE
754 "HOLE",
755 #endif
756 };
757 static DEFINE_STRARRAY(whences, "SEEK_");
758
759 static const char *fcntl_cmds[] = {
760 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
761 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
762 "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
763 "GETOWNER_UIDS",
764 };
765 static DEFINE_STRARRAY(fcntl_cmds, "F_");
766
767 static const char *fcntl_linux_specific_cmds[] = {
768 "SETLEASE", "GETLEASE", "NOTIFY", "DUPFD_QUERY", [5] = "CANCELLK", "DUPFD_CLOEXEC",
769 "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
770 "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
771 };
772
773 static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
774
775 static struct strarray *fcntl_cmds_arrays[] = {
776 &strarray__fcntl_cmds,
777 &strarray__fcntl_linux_specific_cmds,
778 };
779
780 static DEFINE_STRARRAYS(fcntl_cmds_arrays);
781
782 static const char *rlimit_resources[] = {
783 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
784 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
785 "RTTIME",
786 };
787 static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
788
789 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
790 static DEFINE_STRARRAY(sighow, "SIG_");
791
792 static const char *clockid[] = {
793 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
794 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
795 "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
796 };
797 static DEFINE_STRARRAY(clockid, "CLOCK_");
798
799 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
800 struct syscall_arg *arg)
801 {
802 bool show_prefix = arg->show_string_prefix;
803 const char *suffix = "_OK";
804 size_t printed = 0;
805 int mode = arg->val;
806
807 if (mode == F_OK) /* 0 */
808 return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
809 #define P_MODE(n) \
810 if (mode & n##_OK) { \
811 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
812 mode &= ~n##_OK; \
813 }
814
815 P_MODE(R);
816 P_MODE(W);
817 P_MODE(X);
818 #undef P_MODE
819
820 if (mode)
821 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
822
823 return printed;
824 }
825
826 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
827
828 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
829 struct syscall_arg *arg);
830
831 #define SCA_FILENAME syscall_arg__scnprintf_filename
832
833 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
834 struct syscall_arg *arg)
835 {
836 bool show_prefix = arg->show_string_prefix;
837 const char *prefix = "O_";
838 int printed = 0, flags = arg->val;
839
840 #define P_FLAG(n) \
841 if (flags & O_##n) { \
842 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
843 flags &= ~O_##n; \
844 }
845
846 P_FLAG(CLOEXEC);
847 P_FLAG(NONBLOCK);
848 #undef P_FLAG
849
850 if (flags)
851 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
852
853 return printed;
854 }
855
856 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
857
858 #ifndef GRND_NONBLOCK
859 #define GRND_NONBLOCK 0x0001
860 #endif
861 #ifndef GRND_RANDOM
862 #define GRND_RANDOM 0x0002
863 #endif
864
865 static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
866 struct syscall_arg *arg)
867 {
868 bool show_prefix = arg->show_string_prefix;
869 const char *prefix = "GRND_";
870 int printed = 0, flags = arg->val;
871
872 #define P_FLAG(n) \
873 if (flags & GRND_##n) { \
874 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
875 flags &= ~GRND_##n; \
876 }
877
878 P_FLAG(RANDOM);
879 P_FLAG(NONBLOCK);
880 #undef P_FLAG
881
882 if (flags)
883 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
884
885 return printed;
886 }
887
888 #define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
889
890 #define STRARRAY(name, array) \
891 { .scnprintf = SCA_STRARRAY, \
892 .strtoul = STUL_STRARRAY, \
893 .parm = &strarray__##array, }
894
895 #define STRARRAY_FLAGS(name, array) \
896 { .scnprintf = SCA_STRARRAY_FLAGS, \
897 .strtoul = STUL_STRARRAY_FLAGS, \
898 .parm = &strarray__##array, }
899
900 #include "trace/beauty/arch_errno_names.c"
901 #include "trace/beauty/eventfd.c"
902 #include "trace/beauty/futex_op.c"
903 #include "trace/beauty/futex_val3.c"
904 #include "trace/beauty/mmap.c"
905 #include "trace/beauty/mode_t.c"
906 #include "trace/beauty/msg_flags.c"
907 #include "trace/beauty/open_flags.c"
908 #include "trace/beauty/perf_event_open.c"
909 #include "trace/beauty/pid.c"
910 #include "trace/beauty/sched_policy.c"
911 #include "trace/beauty/seccomp.c"
912 #include "trace/beauty/signum.c"
913 #include "trace/beauty/socket_type.c"
914 #include "trace/beauty/waitid_options.c"
915
916 static const struct syscall_fmt syscall_fmts[] = {
917 { .name = "access",
918 .arg = { [1] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, },
919 { .name = "arch_prctl",
920 .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
921 [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
922 { .name = "bind",
923 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
924 [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ },
925 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
926 { .name = "bpf",
927 .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
928 { .name = "brk", .hexret = true,
929 .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
930 { .name = "clock_gettime",
931 .arg = { [0] = STRARRAY(clk_id, clockid), }, },
932 { .name = "clock_nanosleep",
933 .arg = { [2] = { .scnprintf = SCA_TIMESPEC, /* rqtp */ }, }, },
934 { .name = "clone", .errpid = true, .nr_args = 5,
935 .arg = { [0] = { .name = "flags", .scnprintf = SCA_CLONE_FLAGS, },
936 [1] = { .name = "child_stack", .scnprintf = SCA_HEX, },
937 [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
938 [3] = { .name = "child_tidptr", .scnprintf = SCA_HEX, },
939 [4] = { .name = "tls", .scnprintf = SCA_HEX, }, }, },
940 { .name = "close",
941 .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
942 { .name = "connect",
943 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
944 [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ },
945 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
946 { .name = "epoll_ctl",
947 .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
948 { .name = "eventfd2",
949 .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
950 { .name = "faccessat",
951 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ },
952 [1] = { .scnprintf = SCA_FILENAME, /* pathname */ },
953 [2] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, },
954 { .name = "faccessat2",
955 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ },
956 [1] = { .scnprintf = SCA_FILENAME, /* pathname */ },
957 [2] = { .scnprintf = SCA_ACCMODE, /* mode */ },
958 [3] = { .scnprintf = SCA_FACCESSAT2_FLAGS, /* flags */ }, }, },
959 { .name = "fchmodat",
960 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
961 { .name = "fchownat",
962 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
963 { .name = "fcntl",
964 .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
965 .strtoul = STUL_STRARRAYS,
966 .parm = &strarrays__fcntl_cmds_arrays,
967 .show_zero = true, },
968 [2] = { .scnprintf = SCA_FCNTL_ARG, /* arg */ }, }, },
969 { .name = "flock",
970 .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
971 { .name = "fsconfig",
972 .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },
973 { .name = "fsmount",
974 .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),
975 [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },
976 { .name = "fspick",
977 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
978 [1] = { .scnprintf = SCA_FILENAME, /* path */ },
979 [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },
980 { .name = "fstat", .alias = "newfstat", },
981 { .name = "futex",
982 .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
983 [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
984 { .name = "futimesat",
985 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
986 { .name = "getitimer",
987 .arg = { [0] = STRARRAY(which, itimers), }, },
988 { .name = "getpid", .errpid = true, },
989 { .name = "getpgid", .errpid = true, },
990 { .name = "getppid", .errpid = true, },
991 { .name = "getrandom",
992 .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
993 { .name = "getrlimit",
994 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
995 { .name = "getsockopt",
996 .arg = { [1] = STRARRAY(level, socket_level), }, },
997 { .name = "gettid", .errpid = true, },
998 { .name = "ioctl",
999 .arg = {
1000 #if defined(__i386__) || defined(__x86_64__)
1001 /*
1002 * FIXME: Make this available to all arches.
1003 */
1004 [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
1005 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
1006 #else
1007 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
1008 #endif
1009 { .name = "kcmp", .nr_args = 5,
1010 .arg = { [0] = { .name = "pid1", .scnprintf = SCA_PID, },
1011 [1] = { .name = "pid2", .scnprintf = SCA_PID, },
1012 [2] = { .name = "type", .scnprintf = SCA_KCMP_TYPE, },
1013 [3] = { .name = "idx1", .scnprintf = SCA_KCMP_IDX, },
1014 [4] = { .name = "idx2", .scnprintf = SCA_KCMP_IDX, }, }, },
1015 { .name = "keyctl",
1016 .arg = { [0] = STRARRAY(option, keyctl_options), }, },
1017 { .name = "kill",
1018 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1019 { .name = "linkat",
1020 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1021 { .name = "lseek",
1022 .arg = { [2] = STRARRAY(whence, whences), }, },
1023 { .name = "lstat", .alias = "newlstat", },
1024 { .name = "madvise",
1025 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1026 [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
1027 { .name = "mkdirat",
1028 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1029 { .name = "mknodat",
1030 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1031 { .name = "mmap", .hexret = true,
1032 /* The standard mmap maps to old_mmap on s390x */
1033 #if defined(__s390x__)
1034 .alias = "old_mmap",
1035 #endif
1036 .arg = { [2] = { .scnprintf = SCA_MMAP_PROT, .show_zero = true, /* prot */ },
1037 [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */
1038 .strtoul = STUL_STRARRAY_FLAGS,
1039 .parm = &strarray__mmap_flags, },
1040 [5] = { .scnprintf = SCA_HEX, /* offset */ }, }, },
1041 { .name = "mount",
1042 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
1043 [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
1044 .mask_val = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
1045 { .name = "move_mount",
1046 .arg = { [0] = { .scnprintf = SCA_FDAT, /* from_dfd */ },
1047 [1] = { .scnprintf = SCA_FILENAME, /* from_pathname */ },
1048 [2] = { .scnprintf = SCA_FDAT, /* to_dfd */ },
1049 [3] = { .scnprintf = SCA_FILENAME, /* to_pathname */ },
1050 [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },
1051 { .name = "mprotect",
1052 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1053 [2] = { .scnprintf = SCA_MMAP_PROT, .show_zero = true, /* prot */ }, }, },
1054 { .name = "mq_unlink",
1055 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
1056 { .name = "mremap", .hexret = true,
1057 .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
1058 { .name = "name_to_handle_at",
1059 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1060 { .name = "nanosleep",
1061 .arg = { [0] = { .scnprintf = SCA_TIMESPEC, /* req */ }, }, },
1062 { .name = "newfstatat", .alias = "fstatat",
1063 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ },
1064 [1] = { .scnprintf = SCA_FILENAME, /* pathname */ },
1065 [3] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ }, }, },
1066 { .name = "open",
1067 .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1068 { .name = "open_by_handle_at",
1069 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1070 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1071 { .name = "openat",
1072 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1073 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1074 { .name = "perf_event_open",
1075 .arg = { [0] = { .scnprintf = SCA_PERF_ATTR, /* attr */ },
1076 [2] = { .scnprintf = SCA_INT, /* cpu */ },
1077 [3] = { .scnprintf = SCA_FD, /* group_fd */ },
1078 [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
1079 { .name = "pipe2",
1080 .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
1081 { .name = "pkey_alloc",
1082 .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS, /* access_rights */ }, }, },
1083 { .name = "pkey_free",
1084 .arg = { [0] = { .scnprintf = SCA_INT, /* key */ }, }, },
1085 { .name = "pkey_mprotect",
1086 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1087 [2] = { .scnprintf = SCA_MMAP_PROT, .show_zero = true, /* prot */ },
1088 [3] = { .scnprintf = SCA_INT, /* pkey */ }, }, },
1089 { .name = "poll", .timeout = true, },
1090 { .name = "ppoll", .timeout = true, },
1091 { .name = "prctl",
1092 .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */
1093 .strtoul = STUL_STRARRAY,
1094 .parm = &strarray__prctl_options, },
1095 [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
1096 [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
1097 { .name = "pread", .alias = "pread64", },
1098 { .name = "preadv", .alias = "pread", },
1099 { .name = "prlimit64",
1100 .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
1101 { .name = "pwrite", .alias = "pwrite64", },
1102 { .name = "readlinkat",
1103 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1104 { .name = "recvfrom",
1105 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1106 { .name = "recvmmsg",
1107 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1108 { .name = "recvmsg",
1109 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1110 { .name = "renameat",
1111 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1112 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
1113 { .name = "renameat2",
1114 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1115 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
1116 [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
1117 { .name = "rt_sigaction",
1118 .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1119 { .name = "rt_sigprocmask",
1120 .arg = { [0] = STRARRAY(how, sighow), }, },
1121 { .name = "rt_sigqueueinfo",
1122 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1123 { .name = "rt_tgsigqueueinfo",
1124 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1125 { .name = "sched_setscheduler",
1126 .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
1127 { .name = "seccomp",
1128 .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP, /* op */ },
1129 [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
1130 { .name = "select", .timeout = true, },
1131 { .name = "sendfile", .alias = "sendfile64", },
1132 { .name = "sendmmsg",
1133 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1134 { .name = "sendmsg",
1135 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1136 { .name = "sendto",
1137 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
1138 [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
1139 { .name = "set_tid_address", .errpid = true, },
1140 { .name = "setitimer",
1141 .arg = { [0] = STRARRAY(which, itimers), }, },
1142 { .name = "setrlimit",
1143 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
1144 { .name = "setsockopt",
1145 .arg = { [1] = STRARRAY(level, socket_level), }, },
1146 { .name = "socket",
1147 .arg = { [0] = STRARRAY(family, socket_families),
1148 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1149 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1150 { .name = "socketpair",
1151 .arg = { [0] = STRARRAY(family, socket_families),
1152 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1153 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1154 { .name = "stat", .alias = "newstat", },
1155 { .name = "statx",
1156 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fdat */ },
1157 [2] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ } ,
1158 [3] = { .scnprintf = SCA_STATX_MASK, /* mask */ }, }, },
1159 { .name = "swapoff",
1160 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1161 { .name = "swapon",
1162 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1163 { .name = "symlinkat",
1164 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1165 { .name = "sync_file_range",
1166 .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },
1167 { .name = "tgkill",
1168 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1169 { .name = "tkill",
1170 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1171 { .name = "umount2", .alias = "umount",
1172 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
1173 { .name = "uname", .alias = "newuname", },
1174 { .name = "unlinkat",
1175 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1176 [1] = { .scnprintf = SCA_FILENAME, /* pathname */ },
1177 [2] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ }, }, },
1178 { .name = "utimensat",
1179 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
1180 { .name = "wait4", .errpid = true,
1181 .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
1182 { .name = "waitid", .errpid = true,
1183 .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
1184 };
1185
1186 static int syscall_fmt__cmp(const void *name, const void *fmtp)
1187 {
1188 const struct syscall_fmt *fmt = fmtp;
1189 return strcmp(name, fmt->name);
1190 }
1191
1192 static const struct syscall_fmt *__syscall_fmt__find(const struct syscall_fmt *fmts,
1193 const int nmemb,
1194 const char *name)
1195 {
1196 return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1197 }
1198
1199 static const struct syscall_fmt *syscall_fmt__find(const char *name)
1200 {
1201 const int nmemb = ARRAY_SIZE(syscall_fmts);
1202 return __syscall_fmt__find(syscall_fmts, nmemb, name);
1203 }
1204
1205 static const struct syscall_fmt *__syscall_fmt__find_by_alias(const struct syscall_fmt *fmts,
1206 const int nmemb, const char *alias)
1207 {
1208 int i;
1209
1210 for (i = 0; i < nmemb; ++i) {
1211 if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0)
1212 return &fmts[i];
1213 }
1214
1215 return NULL;
1216 }
1217
1218 static const struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
1219 {
1220 const int nmemb = ARRAY_SIZE(syscall_fmts);
1221 return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias);
1222 }
1223
1224 /*
1225 * is_exit: is this "exit" or "exit_group"?
1226 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
1227 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
1228 * nonexistent: Just a hole in the syscall table, syscall id not allocated
1229 */
1230 struct syscall {
1231 struct tep_event *tp_format;
1232 int nr_args;
1233 int args_size;
1234 struct {
1235 struct bpf_program *sys_enter,
1236 *sys_exit;
1237 } bpf_prog;
1238 bool is_exit;
1239 bool is_open;
1240 bool nonexistent;
1241 struct tep_format_field *args;
1242 const char *name;
1243 const struct syscall_fmt *fmt;
1244 struct syscall_arg_fmt *arg_fmt;
1245 };
1246
1247 /*
1248 * We need to have this 'calculated' boolean because in some cases we really
1249 * don't know what is the duration of a syscall, for instance, when we start
1250 * a session and some threads are waiting for a syscall to finish, say 'poll',
1251 * in which case all we can do is to print "( ? ) for duration and for the
1252 * start timestamp.
1253 */
1254 static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
1255 {
1256 double duration = (double)t / NSEC_PER_MSEC;
1257 size_t printed = fprintf(fp, "(");
1258
1259 if (!calculated)
1260 printed += fprintf(fp, " ");
1261 else if (duration >= 1.0)
1262 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1263 else if (duration >= 0.01)
1264 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1265 else
1266 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
1267 return printed + fprintf(fp, "): ");
1268 }
1269
1270 /**
1271 * filename.ptr: The filename char pointer that will be vfs_getname'd
1272 * filename.entry_str_pos: Where to insert the string translated from
1273 * filename.ptr by the vfs_getname tracepoint/kprobe.
1274 * ret_scnprintf: syscall args may set this to a different syscall return
1275 * formatter, for instance, fcntl may return fds, file flags, etc.
1276 */
1277 struct thread_trace {
1278 u64 entry_time;
1279 bool entry_pending;
1280 unsigned long nr_events;
1281 unsigned long pfmaj, pfmin;
1282 char *entry_str;
1283 double runtime_ms;
1284 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1285 struct {
1286 unsigned long ptr;
1287 short int entry_str_pos;
1288 bool pending_open;
1289 unsigned int namelen;
1290 char *name;
1291 } filename;
1292 struct {
1293 int max;
1294 struct file *table;
1295 } files;
1296
1297 struct intlist *syscall_stats;
1298 };
1299
1300 static struct thread_trace *thread_trace__new(void)
1301 {
1302 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
1303
1304 if (ttrace) {
1305 ttrace->files.max = -1;
1306 ttrace->syscall_stats = intlist__new(NULL);
1307 }
1308
1309 return ttrace;
1310 }
1311
1312 static void thread_trace__free_files(struct thread_trace *ttrace);
1313
1314 static void thread_trace__delete(void *pttrace)
1315 {
1316 struct thread_trace *ttrace = pttrace;
1317
1318 if (!ttrace)
1319 return;
1320
1321 intlist__delete(ttrace->syscall_stats);
1322 ttrace->syscall_stats = NULL;
1323 thread_trace__free_files(ttrace);
1324 zfree(&ttrace->entry_str);
1325 free(ttrace);
1326 }
1327
1328 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1329 {
1330 struct thread_trace *ttrace;
1331
1332 if (thread == NULL)
1333 goto fail;
1334
1335 if (thread__priv(thread) == NULL)
1336 thread__set_priv(thread, thread_trace__new());
1337
1338 if (thread__priv(thread) == NULL)
1339 goto fail;
1340
1341 ttrace = thread__priv(thread);
1342 ++ttrace->nr_events;
1343
1344 return ttrace;
1345 fail:
1346 color_fprintf(fp, PERF_COLOR_RED,
1347 "WARNING: not enough memory, dropping samples!\n");
1348 return NULL;
1349 }
1350
1351
1352 void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
1353 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
1354 {
1355 struct thread_trace *ttrace = thread__priv(arg->thread);
1356
1357 ttrace->ret_scnprintf = ret_scnprintf;
1358 }
1359
1360 #define TRACE_PFMAJ (1 << 0)
1361 #define TRACE_PFMIN (1 << 1)
1362
1363 static const size_t trace__entry_str_size = 2048;
1364
1365 static void thread_trace__free_files(struct thread_trace *ttrace)
1366 {
1367 for (int i = 0; i < ttrace->files.max; ++i) {
1368 struct file *file = ttrace->files.table + i;
1369 zfree(&file->pathname);
1370 }
1371
1372 zfree(&ttrace->files.table);
1373 ttrace->files.max = -1;
1374 }
1375
1376 static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
1377 {
1378 if (fd < 0)
1379 return NULL;
1380
1381 if (fd > ttrace->files.max) {
1382 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
1383
1384 if (nfiles == NULL)
1385 return NULL;
1386
1387 if (ttrace->files.max != -1) {
1388 memset(nfiles + ttrace->files.max + 1, 0,
1389 (fd - ttrace->files.max) * sizeof(struct file));
1390 } else {
1391 memset(nfiles, 0, (fd + 1) * sizeof(struct file));
1392 }
1393
1394 ttrace->files.table = nfiles;
1395 ttrace->files.max = fd;
1396 }
1397
1398 return ttrace->files.table + fd;
1399 }
1400
1401 struct file *thread__files_entry(struct thread *thread, int fd)
1402 {
1403 return thread_trace__files_entry(thread__priv(thread), fd);
1404 }
1405
1406 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1407 {
1408 struct thread_trace *ttrace = thread__priv(thread);
1409 struct file *file = thread_trace__files_entry(ttrace, fd);
1410
1411 if (file != NULL) {
1412 struct stat st;
1413 if (stat(pathname, &st) == 0)
1414 file->dev_maj = major(st.st_rdev);
1415 file->pathname = strdup(pathname);
1416 if (file->pathname)
1417 return 0;
1418 }
1419
1420 return -1;
1421 }
1422
1423 static int thread__read_fd_path(struct thread *thread, int fd)
1424 {
1425 char linkname[PATH_MAX], pathname[PATH_MAX];
1426 struct stat st;
1427 int ret;
1428
1429 if (thread__pid(thread) == thread__tid(thread)) {
1430 scnprintf(linkname, sizeof(linkname),
1431 "/proc/%d/fd/%d", thread__pid(thread), fd);
1432 } else {
1433 scnprintf(linkname, sizeof(linkname),
1434 "/proc/%d/task/%d/fd/%d",
1435 thread__pid(thread), thread__tid(thread), fd);
1436 }
1437
1438 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1439 return -1;
1440
1441 ret = readlink(linkname, pathname, sizeof(pathname));
1442
1443 if (ret < 0 || ret > st.st_size)
1444 return -1;
1445
1446 pathname[ret] = '\0';
1447 return trace__set_fd_pathname(thread, fd, pathname);
1448 }
1449
1450 static const char *thread__fd_path(struct thread *thread, int fd,
1451 struct trace *trace)
1452 {
1453 struct thread_trace *ttrace = thread__priv(thread);
1454
1455 if (ttrace == NULL || trace->fd_path_disabled)
1456 return NULL;
1457
1458 if (fd < 0)
1459 return NULL;
1460
1461 if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
1462 if (!trace->live)
1463 return NULL;
1464 ++trace->stats.proc_getname;
1465 if (thread__read_fd_path(thread, fd))
1466 return NULL;
1467 }
1468
1469 return ttrace->files.table[fd].pathname;
1470 }
1471
1472 size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
1473 {
1474 int fd = arg->val;
1475 size_t printed = scnprintf(bf, size, "%d", fd);
1476 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1477
1478 if (path)
1479 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1480
1481 return printed;
1482 }
1483
1484 size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1485 {
1486 size_t printed = scnprintf(bf, size, "%d", fd);
1487 struct thread *thread = machine__find_thread(trace->host, pid, pid);
1488
1489 if (thread) {
1490 const char *path = thread__fd_path(thread, fd, trace);
1491
1492 if (path)
1493 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1494
1495 thread__put(thread);
1496 }
1497
1498 return printed;
1499 }
1500
1501 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1502 struct syscall_arg *arg)
1503 {
1504 int fd = arg->val;
1505 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1506 struct thread_trace *ttrace = thread__priv(arg->thread);
1507
1508 if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1509 zfree(&ttrace->files.table[fd].pathname);
1510
1511 return printed;
1512 }
1513
1514 static void thread__set_filename_pos(struct thread *thread, const char *bf,
1515 unsigned long ptr)
1516 {
1517 struct thread_trace *ttrace = thread__priv(thread);
1518
1519 ttrace->filename.ptr = ptr;
1520 ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1521 }
1522
1523 static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1524 {
1525 struct augmented_arg *augmented_arg = arg->augmented.args;
1526 size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1527 /*
1528 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1529 * we would have two strings, each prefixed by its size.
1530 */
1531 int consumed = sizeof(*augmented_arg) + augmented_arg->size;
1532
1533 arg->augmented.args = ((void *)arg->augmented.args) + consumed;
1534 arg->augmented.size -= consumed;
1535
1536 return printed;
1537 }
1538
1539 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1540 struct syscall_arg *arg)
1541 {
1542 unsigned long ptr = arg->val;
1543
1544 if (arg->augmented.args)
1545 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1546
1547 if (!arg->trace->vfs_getname)
1548 return scnprintf(bf, size, "%#x", ptr);
1549
1550 thread__set_filename_pos(arg->thread, bf, ptr);
1551 return 0;
1552 }
1553
1554 static bool trace__filter_duration(struct trace *trace, double t)
1555 {
1556 return t < (trace->duration_filter * NSEC_PER_MSEC);
1557 }
1558
1559 static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1560 {
1561 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1562
1563 return fprintf(fp, "%10.3f ", ts);
1564 }
1565
1566 /*
1567 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1568 * using ttrace->entry_time for a thread that receives a sys_exit without
1569 * first having received a sys_enter ("poll" issued before tracing session
1570 * starts, lost sys_enter exit due to ring buffer overflow).
1571 */
1572 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1573 {
1574 if (tstamp > 0)
1575 return __trace__fprintf_tstamp(trace, tstamp, fp);
1576
1577 return fprintf(fp, " ? ");
1578 }
1579
1580 static pid_t workload_pid = -1;
1581 static volatile sig_atomic_t done = false;
1582 static volatile sig_atomic_t interrupted = false;
1583
1584 static void sighandler_interrupt(int sig __maybe_unused)
1585 {
1586 done = interrupted = true;
1587 }
1588
1589 static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
1590 void *context __maybe_unused)
1591 {
1592 if (info->si_pid == workload_pid)
1593 done = true;
1594 }
1595
1596 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1597 {
1598 size_t printed = 0;
1599
1600 if (trace->multiple_threads) {
1601 if (trace->show_comm)
1602 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1603 printed += fprintf(fp, "%d ", thread__tid(thread));
1604 }
1605
1606 return printed;
1607 }
1608
1609 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1610 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1611 {
1612 size_t printed = 0;
1613
1614 if (trace->show_tstamp)
1615 printed = trace__fprintf_tstamp(trace, tstamp, fp);
1616 if (trace->show_duration)
1617 printed += fprintf_duration(duration, duration_calculated, fp);
1618 return printed + trace__fprintf_comm_tid(trace, thread, fp);
1619 }
1620
1621 static int trace__process_event(struct trace *trace, struct machine *machine,
1622 union perf_event *event, struct perf_sample *sample)
1623 {
1624 int ret = 0;
1625
1626 switch (event->header.type) {
1627 case PERF_RECORD_LOST:
1628 color_fprintf(trace->output, PERF_COLOR_RED,
1629 "LOST %" PRIu64 " events!\n", event->lost.lost);
1630 ret = machine__process_lost_event(machine, event, sample);
1631 break;
1632 default:
1633 ret = machine__process_event(machine, event, sample);
1634 break;
1635 }
1636
1637 return ret;
1638 }
1639
1640 static int trace__tool_process(struct perf_tool *tool,
1641 union perf_event *event,
1642 struct perf_sample *sample,
1643 struct machine *machine)
1644 {
1645 struct trace *trace = container_of(tool, struct trace, tool);
1646 return trace__process_event(trace, machine, event, sample);
1647 }
1648
1649 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1650 {
1651 struct machine *machine = vmachine;
1652
1653 if (machine->kptr_restrict_warned)
1654 return NULL;
1655
1656 if (symbol_conf.kptr_restrict) {
1657 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1658 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
1659 "Kernel samples will not be resolved.\n");
1660 machine->kptr_restrict_warned = true;
1661 return NULL;
1662 }
1663
1664 return machine__resolve_kernel_addr(vmachine, addrp, modp);
1665 }
1666
1667 static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
1668 {
1669 int err = symbol__init(NULL);
1670
1671 if (err)
1672 return err;
1673
1674 trace->host = machine__new_host();
1675 if (trace->host == NULL)
1676 return -ENOMEM;
1677
1678 thread__set_priv_destructor(thread_trace__delete);
1679
1680 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1681 if (err < 0)
1682 goto out;
1683
1684 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1685 evlist->core.threads, trace__tool_process,
1686 true, false, 1);
1687 out:
1688 if (err)
1689 symbol__exit();
1690
1691 return err;
1692 }
1693
1694 static void trace__symbols__exit(struct trace *trace)
1695 {
1696 machine__exit(trace->host);
1697 trace->host = NULL;
1698
1699 symbol__exit();
1700 }
1701
1702 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1703 {
1704 int idx;
1705
1706 if (nr_args == RAW_SYSCALL_ARGS_NUM && sc->fmt && sc->fmt->nr_args != 0)
1707 nr_args = sc->fmt->nr_args;
1708
1709 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1710 if (sc->arg_fmt == NULL)
1711 return -1;
1712
1713 for (idx = 0; idx < nr_args; ++idx) {
1714 if (sc->fmt)
1715 sc->arg_fmt[idx] = sc->fmt->arg[idx];
1716 }
1717
1718 sc->nr_args = nr_args;
1719 return 0;
1720 }
1721
1722 static const struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {
1723 { .name = "msr", .scnprintf = SCA_X86_MSR, .strtoul = STUL_X86_MSR, },
1724 { .name = "vector", .scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, },
1725 };
1726
1727 static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)
1728 {
1729 const struct syscall_arg_fmt *fmt = fmtp;
1730 return strcmp(name, fmt->name);
1731 }
1732
1733 static const struct syscall_arg_fmt *
1734 __syscall_arg_fmt__find_by_name(const struct syscall_arg_fmt *fmts, const int nmemb,
1735 const char *name)
1736 {
1737 return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);
1738 }
1739
1740 static const struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)
1741 {
1742 const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);
1743 return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);
1744 }
1745
1746 static struct tep_format_field *
1747 syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field)
1748 {
1749 struct tep_format_field *last_field = NULL;
1750 int len;
1751
1752 for (; field; field = field->next, ++arg) {
1753 last_field = field;
1754
1755 if (arg->scnprintf)
1756 continue;
1757
1758 len = strlen(field->name);
1759
1760 if (strcmp(field->type, "const char *") == 0 &&
1761 ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1762 strstr(field->name, "path") != NULL))
1763 arg->scnprintf = SCA_FILENAME;
1764 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
1765 arg->scnprintf = SCA_PTR;
1766 else if (strcmp(field->type, "pid_t") == 0)
1767 arg->scnprintf = SCA_PID;
1768 else if (strcmp(field->type, "umode_t") == 0)
1769 arg->scnprintf = SCA_MODE_T;
1770 else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) {
1771 arg->scnprintf = SCA_CHAR_ARRAY;
1772 arg->nr_entries = field->arraylen;
1773 } else if ((strcmp(field->type, "int") == 0 ||
1774 strcmp(field->type, "unsigned int") == 0 ||
1775 strcmp(field->type, "long") == 0) &&
1776 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
1777 /*
1778 * /sys/kernel/tracing/events/syscalls/sys_enter*
1779 * grep -E 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
1780 * 65 int
1781 * 23 unsigned int
1782 * 7 unsigned long
1783 */
1784 arg->scnprintf = SCA_FD;
1785 } else {
1786 const struct syscall_arg_fmt *fmt =
1787 syscall_arg_fmt__find_by_name(field->name);
1788
1789 if (fmt) {
1790 arg->scnprintf = fmt->scnprintf;
1791 arg->strtoul = fmt->strtoul;
1792 }
1793 }
1794 }
1795
1796 return last_field;
1797 }
1798
1799 static int syscall__set_arg_fmts(struct syscall *sc)
1800 {
1801 struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args);
1802
1803 if (last_field)
1804 sc->args_size = last_field->offset + last_field->size;
1805
1806 return 0;
1807 }
1808
1809 static int trace__read_syscall_info(struct trace *trace, int id)
1810 {
1811 char tp_name[128];
1812 struct syscall *sc;
1813 const char *name = syscalltbl__name(trace->sctbl, id);
1814
1815 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
1816 if (trace->syscalls.table == NULL) {
1817 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
1818 if (trace->syscalls.table == NULL)
1819 return -ENOMEM;
1820 }
1821 #else
1822 if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) {
1823 // When using libaudit we don't know beforehand what is the max syscall id
1824 struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1825
1826 if (table == NULL)
1827 return -ENOMEM;
1828
1829 // Need to memset from offset 0 and +1 members if brand new
1830 if (trace->syscalls.table == NULL)
1831 memset(table, 0, (id + 1) * sizeof(*sc));
1832 else
1833 memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc));
1834
1835 trace->syscalls.table = table;
1836 trace->sctbl->syscalls.max_id = id;
1837 }
1838 #endif
1839 sc = trace->syscalls.table + id;
1840 if (sc->nonexistent)
1841 return -EEXIST;
1842
1843 if (name == NULL) {
1844 sc->nonexistent = true;
1845 return -EEXIST;
1846 }
1847
1848 sc->name = name;
1849 sc->fmt = syscall_fmt__find(sc->name);
1850
1851 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1852 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1853
1854 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
1855 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1856 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1857 }
1858
1859 /*
1860 * Fails to read trace point format via sysfs node, so the trace point
1861 * doesn't exist. Set the 'nonexistent' flag as true.
1862 */
1863 if (IS_ERR(sc->tp_format)) {
1864 sc->nonexistent = true;
1865 return PTR_ERR(sc->tp_format);
1866 }
1867
1868 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ?
1869 RAW_SYSCALL_ARGS_NUM : sc->tp_format->format.nr_fields))
1870 return -ENOMEM;
1871
1872 sc->args = sc->tp_format->format.fields;
1873 /*
1874 * We need to check and discard the first variable '__syscall_nr'
1875 * or 'nr' that mean the syscall number. It is needless here.
1876 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1877 */
1878 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
1879 sc->args = sc->args->next;
1880 --sc->nr_args;
1881 }
1882
1883 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1884 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
1885
1886 return syscall__set_arg_fmts(sc);
1887 }
1888
1889 static int evsel__init_tp_arg_scnprintf(struct evsel *evsel)
1890 {
1891 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
1892
1893 if (fmt != NULL) {
1894 syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields);
1895 return 0;
1896 }
1897
1898 return -ENOMEM;
1899 }
1900
1901 static int intcmp(const void *a, const void *b)
1902 {
1903 const int *one = a, *another = b;
1904
1905 return *one - *another;
1906 }
1907
1908 static int trace__validate_ev_qualifier(struct trace *trace)
1909 {
1910 int err = 0;
1911 bool printed_invalid_prefix = false;
1912 struct str_node *pos;
1913 size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
1914
1915 trace->ev_qualifier_ids.entries = malloc(nr_allocated *
1916 sizeof(trace->ev_qualifier_ids.entries[0]));
1917
1918 if (trace->ev_qualifier_ids.entries == NULL) {
1919 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1920 trace->output);
1921 err = -EINVAL;
1922 goto out;
1923 }
1924
1925 strlist__for_each_entry(pos, trace->ev_qualifier) {
1926 const char *sc = pos->s;
1927 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
1928
1929 if (id < 0) {
1930 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1931 if (id >= 0)
1932 goto matches;
1933
1934 if (!printed_invalid_prefix) {
1935 pr_debug("Skipping unknown syscalls: ");
1936 printed_invalid_prefix = true;
1937 } else {
1938 pr_debug(", ");
1939 }
1940
1941 pr_debug("%s", sc);
1942 continue;
1943 }
1944 matches:
1945 trace->ev_qualifier_ids.entries[nr_used++] = id;
1946 if (match_next == -1)
1947 continue;
1948
1949 while (1) {
1950 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1951 if (id < 0)
1952 break;
1953 if (nr_allocated == nr_used) {
1954 void *entries;
1955
1956 nr_allocated += 8;
1957 entries = realloc(trace->ev_qualifier_ids.entries,
1958 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1959 if (entries == NULL) {
1960 err = -ENOMEM;
1961 fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1962 goto out_free;
1963 }
1964 trace->ev_qualifier_ids.entries = entries;
1965 }
1966 trace->ev_qualifier_ids.entries[nr_used++] = id;
1967 }
1968 }
1969
1970 trace->ev_qualifier_ids.nr = nr_used;
1971 qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
1972 out:
1973 if (printed_invalid_prefix)
1974 pr_debug("\n");
1975 return err;
1976 out_free:
1977 zfree(&trace->ev_qualifier_ids.entries);
1978 trace->ev_qualifier_ids.nr = 0;
1979 goto out;
1980 }
1981
1982 static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1983 {
1984 bool in_ev_qualifier;
1985
1986 if (trace->ev_qualifier_ids.nr == 0)
1987 return true;
1988
1989 in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1990 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1991
1992 if (in_ev_qualifier)
1993 return !trace->not_ev_qualifier;
1994
1995 return trace->not_ev_qualifier;
1996 }
1997
1998 /*
1999 * args is to be interpreted as a series of longs but we need to handle
2000 * 8-byte unaligned accesses. args points to raw_data within the event
2001 * and raw_data is guaranteed to be 8-byte unaligned because it is
2002 * preceded by raw_size which is a u32. So we need to copy args to a temp
2003 * variable to read it. Most notably this avoids extended load instructions
2004 * on unaligned addresses
2005 */
2006 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
2007 {
2008 unsigned long val;
2009 unsigned char *p = arg->args + sizeof(unsigned long) * idx;
2010
2011 memcpy(&val, p, sizeof(val));
2012 return val;
2013 }
2014
2015 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
2016 struct syscall_arg *arg)
2017 {
2018 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
2019 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
2020
2021 return scnprintf(bf, size, "arg%d: ", arg->idx);
2022 }
2023
2024 /*
2025 * Check if the value is in fact zero, i.e. mask whatever needs masking, such
2026 * as mount 'flags' argument that needs ignoring some magic flag, see comment
2027 * in tools/perf/trace/beauty/mount_flags.c
2028 */
2029 static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)
2030 {
2031 if (fmt && fmt->mask_val)
2032 return fmt->mask_val(arg, val);
2033
2034 return val;
2035 }
2036
2037 static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,
2038 struct syscall_arg *arg, unsigned long val)
2039 {
2040 if (fmt && fmt->scnprintf) {
2041 arg->val = val;
2042 if (fmt->parm)
2043 arg->parm = fmt->parm;
2044 return fmt->scnprintf(bf, size, arg);
2045 }
2046 return scnprintf(bf, size, "%ld", val);
2047 }
2048
2049 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
2050 unsigned char *args, void *augmented_args, int augmented_args_size,
2051 struct trace *trace, struct thread *thread)
2052 {
2053 size_t printed = 0;
2054 unsigned long val;
2055 u8 bit = 1;
2056 struct syscall_arg arg = {
2057 .args = args,
2058 .augmented = {
2059 .size = augmented_args_size,
2060 .args = augmented_args,
2061 },
2062 .idx = 0,
2063 .mask = 0,
2064 .trace = trace,
2065 .thread = thread,
2066 .show_string_prefix = trace->show_string_prefix,
2067 };
2068 struct thread_trace *ttrace = thread__priv(thread);
2069
2070 /*
2071 * Things like fcntl will set this in its 'cmd' formatter to pick the
2072 * right formatter for the return value (an fd? file flags?), which is
2073 * not needed for syscalls that always return a given type, say an fd.
2074 */
2075 ttrace->ret_scnprintf = NULL;
2076
2077 if (sc->args != NULL) {
2078 struct tep_format_field *field;
2079
2080 for (field = sc->args; field;
2081 field = field->next, ++arg.idx, bit <<= 1) {
2082 if (arg.mask & bit)
2083 continue;
2084
2085 arg.fmt = &sc->arg_fmt[arg.idx];
2086 val = syscall_arg__val(&arg, arg.idx);
2087 /*
2088 * Some syscall args need some mask, most don't and
2089 * return val untouched.
2090 */
2091 val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);
2092
2093 /*
2094 * Suppress this argument if its value is zero and show_zero
2095 * property isn't set.
2096 */
2097 if (val == 0 && !trace->show_zeros &&
2098 !(sc->arg_fmt && sc->arg_fmt[arg.idx].show_zero))
2099 continue;
2100
2101 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2102
2103 if (trace->show_arg_names)
2104 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2105
2106 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],
2107 bf + printed, size - printed, &arg, val);
2108 }
2109 } else if (IS_ERR(sc->tp_format)) {
2110 /*
2111 * If we managed to read the tracepoint /format file, then we
2112 * may end up not having any args, like with gettid(), so only
2113 * print the raw args when we didn't manage to read it.
2114 */
2115 while (arg.idx < sc->nr_args) {
2116 if (arg.mask & bit)
2117 goto next_arg;
2118 val = syscall_arg__val(&arg, arg.idx);
2119 if (printed)
2120 printed += scnprintf(bf + printed, size - printed, ", ");
2121 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
2122 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);
2123 next_arg:
2124 ++arg.idx;
2125 bit <<= 1;
2126 }
2127 }
2128
2129 return printed;
2130 }
2131
2132 typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
2133 union perf_event *event,
2134 struct perf_sample *sample);
2135
2136 static struct syscall *trace__syscall_info(struct trace *trace,
2137 struct evsel *evsel, int id)
2138 {
2139 int err = 0;
2140
2141 if (id < 0) {
2142
2143 /*
2144 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
2145 * before that, leaving at a higher verbosity level till that is
2146 * explained. Reproduced with plain ftrace with:
2147 *
2148 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
2149 * grep "NR -1 " /t/trace_pipe
2150 *
2151 * After generating some load on the machine.
2152 */
2153 if (verbose > 1) {
2154 static u64 n;
2155 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
2156 id, evsel__name(evsel), ++n);
2157 }
2158 return NULL;
2159 }
2160
2161 err = -EINVAL;
2162
2163 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
2164 if (id > trace->sctbl->syscalls.max_id) {
2165 #else
2166 if (id >= trace->sctbl->syscalls.max_id) {
2167 /*
2168 * With libaudit we don't know beforehand what is the max_id,
2169 * so we let trace__read_syscall_info() figure that out as we
2170 * go on reading syscalls.
2171 */
2172 err = trace__read_syscall_info(trace, id);
2173 if (err)
2174 #endif
2175 goto out_cant_read;
2176 }
2177
2178 if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
2179 (err = trace__read_syscall_info(trace, id)) != 0)
2180 goto out_cant_read;
2181
2182 if (trace->syscalls.table && trace->syscalls.table[id].nonexistent)
2183 goto out_cant_read;
2184
2185 return &trace->syscalls.table[id];
2186
2187 out_cant_read:
2188 if (verbose > 0) {
2189 char sbuf[STRERR_BUFSIZE];
2190 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
2191 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
2192 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
2193 fputs(" information\n", trace->output);
2194 }
2195 return NULL;
2196 }
2197
2198 struct syscall_stats {
2199 struct stats stats;
2200 u64 nr_failures;
2201 int max_errno;
2202 u32 *errnos;
2203 };
2204
2205 static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace,
2206 int id, struct perf_sample *sample, long err, bool errno_summary)
2207 {
2208 struct int_node *inode;
2209 struct syscall_stats *stats;
2210 u64 duration = 0;
2211
2212 inode = intlist__findnew(ttrace->syscall_stats, id);
2213 if (inode == NULL)
2214 return;
2215
2216 stats = inode->priv;
2217 if (stats == NULL) {
2218 stats = zalloc(sizeof(*stats));
2219 if (stats == NULL)
2220 return;
2221
2222 init_stats(&stats->stats);
2223 inode->priv = stats;
2224 }
2225
2226 if (ttrace->entry_time && sample->time > ttrace->entry_time)
2227 duration = sample->time - ttrace->entry_time;
2228
2229 update_stats(&stats->stats, duration);
2230
2231 if (err < 0) {
2232 ++stats->nr_failures;
2233
2234 if (!errno_summary)
2235 return;
2236
2237 err = -err;
2238 if (err > stats->max_errno) {
2239 u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32));
2240
2241 if (new_errnos) {
2242 memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32));
2243 } else {
2244 pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n",
2245 thread__comm_str(thread), thread__pid(thread),
2246 thread__tid(thread));
2247 return;
2248 }
2249
2250 stats->errnos = new_errnos;
2251 stats->max_errno = err;
2252 }
2253
2254 ++stats->errnos[err - 1];
2255 }
2256 }
2257
2258 static int trace__printf_interrupted_entry(struct trace *trace)
2259 {
2260 struct thread_trace *ttrace;
2261 size_t printed;
2262 int len;
2263
2264 if (trace->failure_only || trace->current == NULL)
2265 return 0;
2266
2267 ttrace = thread__priv(trace->current);
2268
2269 if (!ttrace->entry_pending)
2270 return 0;
2271
2272 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
2273 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
2274
2275 if (len < trace->args_alignment - 4)
2276 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
2277
2278 printed += fprintf(trace->output, " ...\n");
2279
2280 ttrace->entry_pending = false;
2281 ++trace->nr_events_printed;
2282
2283 return printed;
2284 }
2285
2286 static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
2287 struct perf_sample *sample, struct thread *thread)
2288 {
2289 int printed = 0;
2290
2291 if (trace->print_sample) {
2292 double ts = (double)sample->time / NSEC_PER_MSEC;
2293
2294 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
2295 evsel__name(evsel), ts,
2296 thread__comm_str(thread),
2297 sample->pid, sample->tid, sample->cpu);
2298 }
2299
2300 return printed;
2301 }
2302
2303 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
2304 {
2305 void *augmented_args = NULL;
2306 /*
2307 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
2308 * and there we get all 6 syscall args plus the tracepoint common fields
2309 * that gets calculated at the start and the syscall_nr (another long).
2310 * So we check if that is the case and if so don't look after the
2311 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
2312 * which is fixed.
2313 *
2314 * We'll revisit this later to pass s->args_size to the BPF augmenter
2315 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
2316 * copies only what we need for each syscall, like what happens when we
2317 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
2318 * traffic to just what is needed for each syscall.
2319 */
2320 int args_size = raw_augmented_args_size ?: sc->args_size;
2321
2322 *augmented_args_size = sample->raw_size - args_size;
2323 if (*augmented_args_size > 0)
2324 augmented_args = sample->raw_data + args_size;
2325
2326 return augmented_args;
2327 }
2328
2329 static void syscall__exit(struct syscall *sc)
2330 {
2331 if (!sc)
2332 return;
2333
2334 zfree(&sc->arg_fmt);
2335 }
2336
2337 static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
2338 union perf_event *event __maybe_unused,
2339 struct perf_sample *sample)
2340 {
2341 char *msg;
2342 void *args;
2343 int printed = 0;
2344 struct thread *thread;
2345 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2346 int augmented_args_size = 0;
2347 void *augmented_args = NULL;
2348 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2349 struct thread_trace *ttrace;
2350
2351 if (sc == NULL)
2352 return -1;
2353
2354 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2355 ttrace = thread__trace(thread, trace->output);
2356 if (ttrace == NULL)
2357 goto out_put;
2358
2359 trace__fprintf_sample(trace, evsel, sample, thread);
2360
2361 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2362
2363 if (ttrace->entry_str == NULL) {
2364 ttrace->entry_str = malloc(trace__entry_str_size);
2365 if (!ttrace->entry_str)
2366 goto out_put;
2367 }
2368
2369 if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
2370 trace__printf_interrupted_entry(trace);
2371 /*
2372 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2373 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2374 * this breaks syscall__augmented_args() check for augmented args, as we calculate
2375 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2376 * so when handling, say the openat syscall, we end up getting 6 args for the
2377 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2378 * thinking that the extra 2 u64 args are the augmented filename, so just check
2379 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2380 */
2381 if (evsel != trace->syscalls.events.sys_enter)
2382 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2383 ttrace->entry_time = sample->time;
2384 msg = ttrace->entry_str;
2385 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
2386
2387 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
2388 args, augmented_args, augmented_args_size, trace, thread);
2389
2390 if (sc->is_exit) {
2391 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
2392 int alignment = 0;
2393
2394 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
2395 printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2396 if (trace->args_alignment > printed)
2397 alignment = trace->args_alignment - printed;
2398 fprintf(trace->output, "%*s= ?\n", alignment, " ");
2399 }
2400 } else {
2401 ttrace->entry_pending = true;
2402 /* See trace__vfs_getname & trace__sys_exit */
2403 ttrace->filename.pending_open = false;
2404 }
2405
2406 if (trace->current != thread) {
2407 thread__put(trace->current);
2408 trace->current = thread__get(thread);
2409 }
2410 err = 0;
2411 out_put:
2412 thread__put(thread);
2413 return err;
2414 }
2415
2416 static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
2417 struct perf_sample *sample)
2418 {
2419 struct thread_trace *ttrace;
2420 struct thread *thread;
2421 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2422 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2423 char msg[1024];
2424 void *args, *augmented_args = NULL;
2425 int augmented_args_size;
2426
2427 if (sc == NULL)
2428 return -1;
2429
2430 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2431 ttrace = thread__trace(thread, trace->output);
2432 /*
2433 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2434 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2435 */
2436 if (ttrace == NULL)
2437 goto out_put;
2438
2439 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2440 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2441 syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
2442 fprintf(trace->output, "%s", msg);
2443 err = 0;
2444 out_put:
2445 thread__put(thread);
2446 return err;
2447 }
2448
2449 static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
2450 struct perf_sample *sample,
2451 struct callchain_cursor *cursor)
2452 {
2453 struct addr_location al;
2454 int max_stack = evsel->core.attr.sample_max_stack ?
2455 evsel->core.attr.sample_max_stack :
2456 trace->max_stack;
2457 int err = -1;
2458
2459 addr_location__init(&al);
2460 if (machine__resolve(trace->host, &al, sample) < 0)
2461 goto out;
2462
2463 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
2464 out:
2465 addr_location__exit(&al);
2466 return err;
2467 }
2468
2469 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2470 {
2471 /* TODO: user-configurable print_opts */
2472 const unsigned int print_opts = EVSEL__PRINT_SYM |
2473 EVSEL__PRINT_DSO |
2474 EVSEL__PRINT_UNKNOWN_AS_ADDR;
2475
2476 return sample__fprintf_callchain(sample, 38, print_opts, get_tls_callchain_cursor(), symbol_conf.bt_stop_list, trace->output);
2477 }
2478
2479 static const char *errno_to_name(struct evsel *evsel, int err)
2480 {
2481 struct perf_env *env = evsel__env(evsel);
2482
2483 return perf_env__arch_strerrno(env, err);
2484 }
2485
2486 static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
2487 union perf_event *event __maybe_unused,
2488 struct perf_sample *sample)
2489 {
2490 long ret;
2491 u64 duration = 0;
2492 bool duration_calculated = false;
2493 struct thread *thread;
2494 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2495 int alignment = trace->args_alignment;
2496 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2497 struct thread_trace *ttrace;
2498
2499 if (sc == NULL)
2500 return -1;
2501
2502 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2503 ttrace = thread__trace(thread, trace->output);
2504 if (ttrace == NULL)
2505 goto out_put;
2506
2507 trace__fprintf_sample(trace, evsel, sample, thread);
2508
2509 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2510
2511 if (trace->summary)
2512 thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary);
2513
2514 if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2515 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2516 ttrace->filename.pending_open = false;
2517 ++trace->stats.vfs_getname;
2518 }
2519
2520 if (ttrace->entry_time) {
2521 duration = sample->time - ttrace->entry_time;
2522 if (trace__filter_duration(trace, duration))
2523 goto out;
2524 duration_calculated = true;
2525 } else if (trace->duration_filter)
2526 goto out;
2527
2528 if (sample->callchain) {
2529 struct callchain_cursor *cursor = get_tls_callchain_cursor();
2530
2531 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
2532 if (callchain_ret == 0) {
2533 if (cursor->nr < trace->min_stack)
2534 goto out;
2535 callchain_ret = 1;
2536 }
2537 }
2538
2539 if (trace->summary_only || (ret >= 0 && trace->failure_only))
2540 goto out;
2541
2542 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2543
2544 if (ttrace->entry_pending) {
2545 printed = fprintf(trace->output, "%s", ttrace->entry_str);
2546 } else {
2547 printed += fprintf(trace->output, " ... [");
2548 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2549 printed += 9;
2550 printed += fprintf(trace->output, "]: %s()", sc->name);
2551 }
2552
2553 printed++; /* the closing ')' */
2554
2555 if (alignment > printed)
2556 alignment -= printed;
2557 else
2558 alignment = 0;
2559
2560 fprintf(trace->output, ")%*s= ", alignment, " ");
2561
2562 if (sc->fmt == NULL) {
2563 if (ret < 0)
2564 goto errno_print;
2565 signed_print:
2566 fprintf(trace->output, "%ld", ret);
2567 } else if (ret < 0) {
2568 errno_print: {
2569 char bf[STRERR_BUFSIZE];
2570 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2571 *e = errno_to_name(evsel, -ret);
2572
2573 fprintf(trace->output, "-1 %s (%s)", e, emsg);
2574 }
2575 } else if (ret == 0 && sc->fmt->timeout)
2576 fprintf(trace->output, "0 (Timeout)");
2577 else if (ttrace->ret_scnprintf) {
2578 char bf[1024];
2579 struct syscall_arg arg = {
2580 .val = ret,
2581 .thread = thread,
2582 .trace = trace,
2583 };
2584 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2585 ttrace->ret_scnprintf = NULL;
2586 fprintf(trace->output, "%s", bf);
2587 } else if (sc->fmt->hexret)
2588 fprintf(trace->output, "%#lx", ret);
2589 else if (sc->fmt->errpid) {
2590 struct thread *child = machine__find_thread(trace->host, ret, ret);
2591
2592 if (child != NULL) {
2593 fprintf(trace->output, "%ld", ret);
2594 if (thread__comm_set(child))
2595 fprintf(trace->output, " (%s)", thread__comm_str(child));
2596 thread__put(child);
2597 }
2598 } else
2599 goto signed_print;
2600
2601 fputc('\n', trace->output);
2602
2603 /*
2604 * We only consider an 'event' for the sake of --max-events a non-filtered
2605 * sys_enter + sys_exit and other tracepoint events.
2606 */
2607 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2608 interrupted = true;
2609
2610 if (callchain_ret > 0)
2611 trace__fprintf_callchain(trace, sample);
2612 else if (callchain_ret < 0)
2613 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2614 out:
2615 ttrace->entry_pending = false;
2616 err = 0;
2617 out_put:
2618 thread__put(thread);
2619 return err;
2620 }
2621
2622 static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
2623 union perf_event *event __maybe_unused,
2624 struct perf_sample *sample)
2625 {
2626 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2627 struct thread_trace *ttrace;
2628 size_t filename_len, entry_str_len, to_move;
2629 ssize_t remaining_space;
2630 char *pos;
2631 const char *filename = evsel__rawptr(evsel, sample, "pathname");
2632
2633 if (!thread)
2634 goto out;
2635
2636 ttrace = thread__priv(thread);
2637 if (!ttrace)
2638 goto out_put;
2639
2640 filename_len = strlen(filename);
2641 if (filename_len == 0)
2642 goto out_put;
2643
2644 if (ttrace->filename.namelen < filename_len) {
2645 char *f = realloc(ttrace->filename.name, filename_len + 1);
2646
2647 if (f == NULL)
2648 goto out_put;
2649
2650 ttrace->filename.namelen = filename_len;
2651 ttrace->filename.name = f;
2652 }
2653
2654 strcpy(ttrace->filename.name, filename);
2655 ttrace->filename.pending_open = true;
2656
2657 if (!ttrace->filename.ptr)
2658 goto out_put;
2659
2660 entry_str_len = strlen(ttrace->entry_str);
2661 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2662 if (remaining_space <= 0)
2663 goto out_put;
2664
2665 if (filename_len > (size_t)remaining_space) {
2666 filename += filename_len - remaining_space;
2667 filename_len = remaining_space;
2668 }
2669
2670 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2671 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2672 memmove(pos + filename_len, pos, to_move);
2673 memcpy(pos, filename, filename_len);
2674
2675 ttrace->filename.ptr = 0;
2676 ttrace->filename.entry_str_pos = 0;
2677 out_put:
2678 thread__put(thread);
2679 out:
2680 return 0;
2681 }
2682
2683 static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
2684 union perf_event *event __maybe_unused,
2685 struct perf_sample *sample)
2686 {
2687 u64 runtime = evsel__intval(evsel, sample, "runtime");
2688 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2689 struct thread *thread = machine__findnew_thread(trace->host,
2690 sample->pid,
2691 sample->tid);
2692 struct thread_trace *ttrace = thread__trace(thread, trace->output);
2693
2694 if (ttrace == NULL)
2695 goto out_dump;
2696
2697 ttrace->runtime_ms += runtime_ms;
2698 trace->runtime_ms += runtime_ms;
2699 out_put:
2700 thread__put(thread);
2701 return 0;
2702
2703 out_dump:
2704 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2705 evsel->name,
2706 evsel__strval(evsel, sample, "comm"),
2707 (pid_t)evsel__intval(evsel, sample, "pid"),
2708 runtime,
2709 evsel__intval(evsel, sample, "vruntime"));
2710 goto out_put;
2711 }
2712
2713 static int bpf_output__printer(enum binary_printer_ops op,
2714 unsigned int val, void *extra __maybe_unused, FILE *fp)
2715 {
2716 unsigned char ch = (unsigned char)val;
2717
2718 switch (op) {
2719 case BINARY_PRINT_CHAR_DATA:
2720 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
2721 case BINARY_PRINT_DATA_BEGIN:
2722 case BINARY_PRINT_LINE_BEGIN:
2723 case BINARY_PRINT_ADDR:
2724 case BINARY_PRINT_NUM_DATA:
2725 case BINARY_PRINT_NUM_PAD:
2726 case BINARY_PRINT_SEP:
2727 case BINARY_PRINT_CHAR_PAD:
2728 case BINARY_PRINT_LINE_END:
2729 case BINARY_PRINT_DATA_END:
2730 default:
2731 break;
2732 }
2733
2734 return 0;
2735 }
2736
2737 static void bpf_output__fprintf(struct trace *trace,
2738 struct perf_sample *sample)
2739 {
2740 binary__fprintf(sample->raw_data, sample->raw_size, 8,
2741 bpf_output__printer, NULL, trace->output);
2742 ++trace->nr_events_printed;
2743 }
2744
2745 static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,
2746 struct thread *thread, void *augmented_args, int augmented_args_size)
2747 {
2748 char bf[2048];
2749 size_t size = sizeof(bf);
2750 struct tep_format_field *field = evsel->tp_format->format.fields;
2751 struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel);
2752 size_t printed = 0;
2753 unsigned long val;
2754 u8 bit = 1;
2755 struct syscall_arg syscall_arg = {
2756 .augmented = {
2757 .size = augmented_args_size,
2758 .args = augmented_args,
2759 },
2760 .idx = 0,
2761 .mask = 0,
2762 .trace = trace,
2763 .thread = thread,
2764 .show_string_prefix = trace->show_string_prefix,
2765 };
2766
2767 for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {
2768 if (syscall_arg.mask & bit)
2769 continue;
2770
2771 syscall_arg.len = 0;
2772 syscall_arg.fmt = arg;
2773 if (field->flags & TEP_FIELD_IS_ARRAY) {
2774 int offset = field->offset;
2775
2776 if (field->flags & TEP_FIELD_IS_DYNAMIC) {
2777 offset = format_field__intval(field, sample, evsel->needs_swap);
2778 syscall_arg.len = offset >> 16;
2779 offset &= 0xffff;
2780 if (tep_field_is_relative(field->flags))
2781 offset += field->offset + field->size;
2782 }
2783
2784 val = (uintptr_t)(sample->raw_data + offset);
2785 } else
2786 val = format_field__intval(field, sample, evsel->needs_swap);
2787 /*
2788 * Some syscall args need some mask, most don't and
2789 * return val untouched.
2790 */
2791 val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);
2792
2793 /* Suppress this argument if its value is zero and show_zero property isn't set. */
2794 if (val == 0 && !trace->show_zeros && !arg->show_zero)
2795 continue;
2796
2797 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2798
2799 if (trace->show_arg_names)
2800 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2801
2802 printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);
2803 }
2804
2805 return printed + fprintf(trace->output, "%s", bf);
2806 }
2807
2808 static int trace__event_handler(struct trace *trace, struct evsel *evsel,
2809 union perf_event *event __maybe_unused,
2810 struct perf_sample *sample)
2811 {
2812 struct thread *thread;
2813 int callchain_ret = 0;
2814 /*
2815 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2816 * this event's max_events having been hit and this is an entry coming
2817 * from the ring buffer that we should discard, since the max events
2818 * have already been considered/printed.
2819 */
2820 if (evsel->disabled)
2821 return 0;
2822
2823 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2824
2825 if (sample->callchain) {
2826 struct callchain_cursor *cursor = get_tls_callchain_cursor();
2827
2828 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
2829 if (callchain_ret == 0) {
2830 if (cursor->nr < trace->min_stack)
2831 goto out;
2832 callchain_ret = 1;
2833 }
2834 }
2835
2836 trace__printf_interrupted_entry(trace);
2837 trace__fprintf_tstamp(trace, sample->time, trace->output);
2838
2839 if (trace->trace_syscalls && trace->show_duration)
2840 fprintf(trace->output, "( ): ");
2841
2842 if (thread)
2843 trace__fprintf_comm_tid(trace, thread, trace->output);
2844
2845 if (evsel == trace->syscalls.events.bpf_output) {
2846 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2847 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2848
2849 if (sc) {
2850 fprintf(trace->output, "%s(", sc->name);
2851 trace__fprintf_sys_enter(trace, evsel, sample);
2852 fputc(')', trace->output);
2853 goto newline;
2854 }
2855
2856 /*
2857 * XXX: Not having the associated syscall info or not finding/adding
2858 * the thread should never happen, but if it does...
2859 * fall thru and print it as a bpf_output event.
2860 */
2861 }
2862
2863 fprintf(trace->output, "%s(", evsel->name);
2864
2865 if (evsel__is_bpf_output(evsel)) {
2866 bpf_output__fprintf(trace, sample);
2867 } else if (evsel->tp_format) {
2868 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2869 trace__fprintf_sys_enter(trace, evsel, sample)) {
2870 if (trace->libtraceevent_print) {
2871 event_format__fprintf(evsel->tp_format, sample->cpu,
2872 sample->raw_data, sample->raw_size,
2873 trace->output);
2874 } else {
2875 trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);
2876 }
2877 }
2878 }
2879
2880 newline:
2881 fprintf(trace->output, ")\n");
2882
2883 if (callchain_ret > 0)
2884 trace__fprintf_callchain(trace, sample);
2885 else if (callchain_ret < 0)
2886 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2887
2888 ++trace->nr_events_printed;
2889
2890 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2891 evsel__disable(evsel);
2892 evsel__close(evsel);
2893 }
2894 out:
2895 thread__put(thread);
2896 return 0;
2897 }
2898
2899 static void print_location(FILE *f, struct perf_sample *sample,
2900 struct addr_location *al,
2901 bool print_dso, bool print_sym)
2902 {
2903
2904 if ((verbose > 0 || print_dso) && al->map)
2905 fprintf(f, "%s@", dso__long_name(map__dso(al->map)));
2906
2907 if ((verbose > 0 || print_sym) && al->sym)
2908 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
2909 al->addr - al->sym->start);
2910 else if (al->map)
2911 fprintf(f, "0x%" PRIx64, al->addr);
2912 else
2913 fprintf(f, "0x%" PRIx64, sample->addr);
2914 }
2915
2916 static int trace__pgfault(struct trace *trace,
2917 struct evsel *evsel,
2918 union perf_event *event __maybe_unused,
2919 struct perf_sample *sample)
2920 {
2921 struct thread *thread;
2922 struct addr_location al;
2923 char map_type = 'd';
2924 struct thread_trace *ttrace;
2925 int err = -1;
2926 int callchain_ret = 0;
2927
2928 addr_location__init(&al);
2929 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2930
2931 if (sample->callchain) {
2932 struct callchain_cursor *cursor = get_tls_callchain_cursor();
2933
2934 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
2935 if (callchain_ret == 0) {
2936 if (cursor->nr < trace->min_stack)
2937 goto out_put;
2938 callchain_ret = 1;
2939 }
2940 }
2941
2942 ttrace = thread__trace(thread, trace->output);
2943 if (ttrace == NULL)
2944 goto out_put;
2945
2946 if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2947 ttrace->pfmaj++;
2948 else
2949 ttrace->pfmin++;
2950
2951 if (trace->summary_only)
2952 goto out;
2953
2954 thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
2955
2956 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
2957
2958 fprintf(trace->output, "%sfault [",
2959 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2960 "maj" : "min");
2961
2962 print_location(trace->output, sample, &al, false, true);
2963
2964 fprintf(trace->output, "] => ");
2965
2966 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2967
2968 if (!al.map) {
2969 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2970
2971 if (al.map)
2972 map_type = 'x';
2973 else
2974 map_type = '?';
2975 }
2976
2977 print_location(trace->output, sample, &al, true, false);
2978
2979 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
2980
2981 if (callchain_ret > 0)
2982 trace__fprintf_callchain(trace, sample);
2983 else if (callchain_ret < 0)
2984 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2985
2986 ++trace->nr_events_printed;
2987 out:
2988 err = 0;
2989 out_put:
2990 thread__put(thread);
2991 addr_location__exit(&al);
2992 return err;
2993 }
2994
2995 static void trace__set_base_time(struct trace *trace,
2996 struct evsel *evsel,
2997 struct perf_sample *sample)
2998 {
2999 /*
3000 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
3001 * and don't use sample->time unconditionally, we may end up having
3002 * some other event in the future without PERF_SAMPLE_TIME for good
3003 * reason, i.e. we may not be interested in its timestamps, just in
3004 * it taking place, picking some piece of information when it
3005 * appears in our event stream (vfs_getname comes to mind).
3006 */
3007 if (trace->base_time == 0 && !trace->full_time &&
3008 (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
3009 trace->base_time = sample->time;
3010 }
3011
3012 static int trace__process_sample(struct perf_tool *tool,
3013 union perf_event *event,
3014 struct perf_sample *sample,
3015 struct evsel *evsel,
3016 struct machine *machine __maybe_unused)
3017 {
3018 struct trace *trace = container_of(tool, struct trace, tool);
3019 struct thread *thread;
3020 int err = 0;
3021
3022 tracepoint_handler handler = evsel->handler;
3023
3024 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
3025 if (thread && thread__is_filtered(thread))
3026 goto out;
3027
3028 trace__set_base_time(trace, evsel, sample);
3029
3030 if (handler) {
3031 ++trace->nr_events;
3032 handler(trace, evsel, event, sample);
3033 }
3034 out:
3035 thread__put(thread);
3036 return err;
3037 }
3038
3039 static int trace__record(struct trace *trace, int argc, const char **argv)
3040 {
3041 unsigned int rec_argc, i, j;
3042 const char **rec_argv;
3043 const char * const record_args[] = {
3044 "record",
3045 "-R",
3046 "-m", "1024",
3047 "-c", "1",
3048 };
3049 pid_t pid = getpid();
3050 char *filter = asprintf__tp_filter_pids(1, &pid);
3051 const char * const sc_args[] = { "-e", };
3052 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
3053 const char * const majpf_args[] = { "-e", "major-faults" };
3054 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
3055 const char * const minpf_args[] = { "-e", "minor-faults" };
3056 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
3057 int err = -1;
3058
3059 /* +3 is for the event string below and the pid filter */
3060 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 +
3061 majpf_args_nr + minpf_args_nr + argc;
3062 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3063
3064 if (rec_argv == NULL || filter == NULL)
3065 goto out_free;
3066
3067 j = 0;
3068 for (i = 0; i < ARRAY_SIZE(record_args); i++)
3069 rec_argv[j++] = record_args[i];
3070
3071 if (trace->trace_syscalls) {
3072 for (i = 0; i < sc_args_nr; i++)
3073 rec_argv[j++] = sc_args[i];
3074
3075 /* event string may be different for older kernels - e.g., RHEL6 */
3076 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
3077 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
3078 else if (is_valid_tracepoint("syscalls:sys_enter"))
3079 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
3080 else {
3081 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
3082 goto out_free;
3083 }
3084 }
3085
3086 rec_argv[j++] = "--filter";
3087 rec_argv[j++] = filter;
3088
3089 if (trace->trace_pgfaults & TRACE_PFMAJ)
3090 for (i = 0; i < majpf_args_nr; i++)
3091 rec_argv[j++] = majpf_args[i];
3092
3093 if (trace->trace_pgfaults & TRACE_PFMIN)
3094 for (i = 0; i < minpf_args_nr; i++)
3095 rec_argv[j++] = minpf_args[i];
3096
3097 for (i = 0; i < (unsigned int)argc; i++)
3098 rec_argv[j++] = argv[i];
3099
3100 err = cmd_record(j, rec_argv);
3101 out_free:
3102 free(filter);
3103 free(rec_argv);
3104 return err;
3105 }
3106
3107 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
3108
3109 static bool evlist__add_vfs_getname(struct evlist *evlist)
3110 {
3111 bool found = false;
3112 struct evsel *evsel, *tmp;
3113 struct parse_events_error err;
3114 int ret;
3115
3116 parse_events_error__init(&err);
3117 ret = parse_events(evlist, "probe:vfs_getname*", &err);
3118 parse_events_error__exit(&err);
3119 if (ret)
3120 return false;
3121
3122 evlist__for_each_entry_safe(evlist, evsel, tmp) {
3123 if (!strstarts(evsel__name(evsel), "probe:vfs_getname"))
3124 continue;
3125
3126 if (evsel__field(evsel, "pathname")) {
3127 evsel->handler = trace__vfs_getname;
3128 found = true;
3129 continue;
3130 }
3131
3132 list_del_init(&evsel->core.node);
3133 evsel->evlist = NULL;
3134 evsel__delete(evsel);
3135 }
3136
3137 return found;
3138 }
3139
3140 static struct evsel *evsel__new_pgfault(u64 config)
3141 {
3142 struct evsel *evsel;
3143 struct perf_event_attr attr = {
3144 .type = PERF_TYPE_SOFTWARE,
3145 .mmap_data = 1,
3146 };
3147
3148 attr.config = config;
3149 attr.sample_period = 1;
3150
3151 event_attr_init(&attr);
3152
3153 evsel = evsel__new(&attr);
3154 if (evsel)
3155 evsel->handler = trace__pgfault;
3156
3157 return evsel;
3158 }
3159
3160 static void evlist__free_syscall_tp_fields(struct evlist *evlist)
3161 {
3162 struct evsel *evsel;
3163
3164 evlist__for_each_entry(evlist, evsel) {
3165 evsel_trace__delete(evsel->priv);
3166 evsel->priv = NULL;
3167 }
3168 }
3169
3170 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
3171 {
3172 const u32 type = event->header.type;
3173 struct evsel *evsel;
3174
3175 if (type != PERF_RECORD_SAMPLE) {
3176 trace__process_event(trace, trace->host, event, sample);
3177 return;
3178 }
3179
3180 evsel = evlist__id2evsel(trace->evlist, sample->id);
3181 if (evsel == NULL) {
3182 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
3183 return;
3184 }
3185
3186 if (evswitch__discard(&trace->evswitch, evsel))
3187 return;
3188
3189 trace__set_base_time(trace, evsel, sample);
3190
3191 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
3192 sample->raw_data == NULL) {
3193 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
3194 evsel__name(evsel), sample->tid,
3195 sample->cpu, sample->raw_size);
3196 } else {
3197 tracepoint_handler handler = evsel->handler;
3198 handler(trace, evsel, event, sample);
3199 }
3200
3201 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
3202 interrupted = true;
3203 }
3204
3205 static int trace__add_syscall_newtp(struct trace *trace)
3206 {
3207 int ret = -1;
3208 struct evlist *evlist = trace->evlist;
3209 struct evsel *sys_enter, *sys_exit;
3210
3211 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
3212 if (sys_enter == NULL)
3213 goto out;
3214
3215 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
3216 goto out_delete_sys_enter;
3217
3218 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
3219 if (sys_exit == NULL)
3220 goto out_delete_sys_enter;
3221
3222 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
3223 goto out_delete_sys_exit;
3224
3225 evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
3226 evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
3227
3228 evlist__add(evlist, sys_enter);
3229 evlist__add(evlist, sys_exit);
3230
3231 if (callchain_param.enabled && !trace->kernel_syscallchains) {
3232 /*
3233 * We're interested only in the user space callchain
3234 * leading to the syscall, allow overriding that for
3235 * debugging reasons using --kernel_syscall_callchains
3236 */
3237 sys_exit->core.attr.exclude_callchain_kernel = 1;
3238 }
3239
3240 trace->syscalls.events.sys_enter = sys_enter;
3241 trace->syscalls.events.sys_exit = sys_exit;
3242
3243 ret = 0;
3244 out:
3245 return ret;
3246
3247 out_delete_sys_exit:
3248 evsel__delete_priv(sys_exit);
3249 out_delete_sys_enter:
3250 evsel__delete_priv(sys_enter);
3251 goto out;
3252 }
3253
3254 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
3255 {
3256 int err = -1;
3257 struct evsel *sys_exit;
3258 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
3259 trace->ev_qualifier_ids.nr,
3260 trace->ev_qualifier_ids.entries);
3261
3262 if (filter == NULL)
3263 goto out_enomem;
3264
3265 if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) {
3266 sys_exit = trace->syscalls.events.sys_exit;
3267 err = evsel__append_tp_filter(sys_exit, filter);
3268 }
3269
3270 free(filter);
3271 out:
3272 return err;
3273 out_enomem:
3274 errno = ENOMEM;
3275 goto out;
3276 }
3277
3278 #ifdef HAVE_BPF_SKEL
3279 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
3280 {
3281 struct bpf_program *pos, *prog = NULL;
3282 const char *sec_name;
3283
3284 if (trace->skel->obj == NULL)
3285 return NULL;
3286
3287 bpf_object__for_each_program(pos, trace->skel->obj) {
3288 sec_name = bpf_program__section_name(pos);
3289 if (sec_name && !strcmp(sec_name, name)) {
3290 prog = pos;
3291 break;
3292 }
3293 }
3294
3295 return prog;
3296 }
3297
3298 static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
3299 const char *prog_name, const char *type)
3300 {
3301 struct bpf_program *prog;
3302
3303 if (prog_name == NULL) {
3304 char default_prog_name[256];
3305 scnprintf(default_prog_name, sizeof(default_prog_name), "tp/syscalls/sys_%s_%s", type, sc->name);
3306 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3307 if (prog != NULL)
3308 goto out_found;
3309 if (sc->fmt && sc->fmt->alias) {
3310 scnprintf(default_prog_name, sizeof(default_prog_name), "tp/syscalls/sys_%s_%s", type, sc->fmt->alias);
3311 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3312 if (prog != NULL)
3313 goto out_found;
3314 }
3315 goto out_unaugmented;
3316 }
3317
3318 prog = trace__find_bpf_program_by_title(trace, prog_name);
3319
3320 if (prog != NULL) {
3321 out_found:
3322 return prog;
3323 }
3324
3325 pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
3326 prog_name, type, sc->name);
3327 out_unaugmented:
3328 return trace->skel->progs.syscall_unaugmented;
3329 }
3330
3331 static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
3332 {
3333 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3334
3335 if (sc == NULL)
3336 return;
3337
3338 sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3339 sc->bpf_prog.sys_exit = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit : NULL, "exit");
3340 }
3341
3342 static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
3343 {
3344 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3345 return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->skel->progs.syscall_unaugmented);
3346 }
3347
3348 static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
3349 {
3350 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3351 return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->skel->progs.syscall_unaugmented);
3352 }
3353
3354 static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
3355 {
3356 struct tep_format_field *field, *candidate_field;
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
3367 try_to_find_pair:
3368 for (int i = 0; i < trace->sctbl->syscalls.nr_entries; ++i) {
3369 int id = syscalltbl__id_at_idx(trace->sctbl, i);
3370 struct syscall *pair = trace__syscall_info(trace, NULL, id);
3371 struct bpf_program *pair_prog;
3372 bool is_candidate = false;
3373
3374 if (pair == NULL || pair == sc ||
3375 pair->bpf_prog.sys_enter == trace->skel->progs.syscall_unaugmented)
3376 continue;
3377
3378 for (field = sc->args, candidate_field = pair->args;
3379 field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3380 bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3381 candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3382
3383 if (is_pointer) {
3384 if (!candidate_is_pointer) {
3385 // The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3386 continue;
3387 }
3388 } else {
3389 if (candidate_is_pointer) {
3390 // The candidate might copy a pointer we don't have, skip it.
3391 goto next_candidate;
3392 }
3393 continue;
3394 }
3395
3396 if (strcmp(field->type, candidate_field->type))
3397 goto next_candidate;
3398
3399 /*
3400 * This is limited in the BPF program but sys_write
3401 * uses "const char *" for its "buf" arg so we need to
3402 * use some heuristic that is kinda future proof...
3403 */
3404 if (strcmp(field->type, "const char *") == 0 &&
3405 !(strstr(field->name, "name") ||
3406 strstr(field->name, "path") ||
3407 strstr(field->name, "file") ||
3408 strstr(field->name, "root") ||
3409 strstr(field->name, "description")))
3410 goto next_candidate;
3411
3412 is_candidate = true;
3413 }
3414
3415 if (!is_candidate)
3416 goto next_candidate;
3417
3418 /*
3419 * Check if the tentative pair syscall augmenter has more pointers, if it has,
3420 * then it may be collecting that and we then can't use it, as it would collect
3421 * more than what is common to the two syscalls.
3422 */
3423 if (candidate_field) {
3424 for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3425 if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3426 goto next_candidate;
3427 }
3428
3429 pair_prog = pair->bpf_prog.sys_enter;
3430 /*
3431 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3432 * have been searched for, so search it here and if it returns the
3433 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3434 * program for a filtered syscall on a non-filtered one.
3435 *
3436 * For instance, we have "!syscalls:sys_enter_renameat" and that is
3437 * useful for "renameat2".
3438 */
3439 if (pair_prog == NULL) {
3440 pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3441 if (pair_prog == trace->skel->progs.syscall_unaugmented)
3442 goto next_candidate;
3443 }
3444
3445 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3446 return pair_prog;
3447 next_candidate:
3448 continue;
3449 }
3450
3451 return NULL;
3452 }
3453
3454 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3455 {
3456 int map_enter_fd = bpf_map__fd(trace->skel->maps.syscalls_sys_enter);
3457 int map_exit_fd = bpf_map__fd(trace->skel->maps.syscalls_sys_exit);
3458 int err = 0;
3459
3460 for (int i = 0; i < trace->sctbl->syscalls.nr_entries; ++i) {
3461 int prog_fd, key = syscalltbl__id_at_idx(trace->sctbl, i);
3462
3463 if (!trace__syscall_enabled(trace, key))
3464 continue;
3465
3466 trace__init_syscall_bpf_progs(trace, key);
3467
3468 // It'll get at least the "!raw_syscalls:unaugmented"
3469 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3470 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3471 if (err)
3472 break;
3473 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3474 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3475 if (err)
3476 break;
3477 }
3478
3479 /*
3480 * Now lets do a second pass looking for enabled syscalls without
3481 * an augmenter that have a signature that is a superset of another
3482 * syscall with an augmenter so that we can auto-reuse it.
3483 *
3484 * I.e. if we have an augmenter for the "open" syscall that has
3485 * this signature:
3486 *
3487 * int open(const char *pathname, int flags, mode_t mode);
3488 *
3489 * I.e. that will collect just the first string argument, then we
3490 * can reuse it for the 'creat' syscall, that has this signature:
3491 *
3492 * int creat(const char *pathname, mode_t mode);
3493 *
3494 * and for:
3495 *
3496 * int stat(const char *pathname, struct stat *statbuf);
3497 * int lstat(const char *pathname, struct stat *statbuf);
3498 *
3499 * Because the 'open' augmenter will collect the first arg as a string,
3500 * and leave alone all the other args, which already helps with
3501 * beautifying 'stat' and 'lstat''s pathname arg.
3502 *
3503 * Then, in time, when 'stat' gets an augmenter that collects both
3504 * first and second arg (this one on the raw_syscalls:sys_exit prog
3505 * array tail call, then that one will be used.
3506 */
3507 for (int i = 0; i < trace->sctbl->syscalls.nr_entries; ++i) {
3508 int key = syscalltbl__id_at_idx(trace->sctbl, i);
3509 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3510 struct bpf_program *pair_prog;
3511 int prog_fd;
3512
3513 if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3514 continue;
3515
3516 /*
3517 * For now we're just reusing the sys_enter prog, and if it
3518 * already has an augmenter, we don't need to find one.
3519 */
3520 if (sc->bpf_prog.sys_enter != trace->skel->progs.syscall_unaugmented)
3521 continue;
3522
3523 /*
3524 * Look at all the other syscalls for one that has a signature
3525 * that is close enough that we can share:
3526 */
3527 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3528 if (pair_prog == NULL)
3529 continue;
3530
3531 sc->bpf_prog.sys_enter = pair_prog;
3532
3533 /*
3534 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3535 * with the fd for the program we're reusing:
3536 */
3537 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3538 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3539 if (err)
3540 break;
3541 }
3542
3543 return err;
3544 }
3545 #endif // HAVE_BPF_SKEL
3546
3547 static int trace__set_ev_qualifier_filter(struct trace *trace)
3548 {
3549 if (trace->syscalls.events.sys_enter)
3550 return trace__set_ev_qualifier_tp_filter(trace);
3551 return 0;
3552 }
3553
3554 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3555 size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3556 {
3557 int err = 0;
3558 #ifdef HAVE_LIBBPF_SUPPORT
3559 bool value = true;
3560 int map_fd = bpf_map__fd(map);
3561 size_t i;
3562
3563 for (i = 0; i < npids; ++i) {
3564 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3565 if (err)
3566 break;
3567 }
3568 #endif
3569 return err;
3570 }
3571
3572 static int trace__set_filter_loop_pids(struct trace *trace)
3573 {
3574 unsigned int nr = 1, err;
3575 pid_t pids[32] = {
3576 getpid(),
3577 };
3578 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3579
3580 while (thread && nr < ARRAY_SIZE(pids)) {
3581 struct thread *parent = machine__find_thread(trace->host,
3582 thread__ppid(thread),
3583 thread__ppid(thread));
3584
3585 if (parent == NULL)
3586 break;
3587
3588 if (!strcmp(thread__comm_str(parent), "sshd") ||
3589 strstarts(thread__comm_str(parent), "gnome-terminal")) {
3590 pids[nr++] = thread__tid(parent);
3591 break;
3592 }
3593 thread = parent;
3594 }
3595
3596 err = evlist__append_tp_filter_pids(trace->evlist, nr, pids);
3597 if (!err && trace->filter_pids.map)
3598 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3599
3600 return err;
3601 }
3602
3603 static int trace__set_filter_pids(struct trace *trace)
3604 {
3605 int err = 0;
3606 /*
3607 * Better not use !target__has_task() here because we need to cover the
3608 * case where no threads were specified in the command line, but a
3609 * workload was, and in that case we will fill in the thread_map when
3610 * we fork the workload in evlist__prepare_workload.
3611 */
3612 if (trace->filter_pids.nr > 0) {
3613 err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3614 trace->filter_pids.entries);
3615 if (!err && trace->filter_pids.map) {
3616 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3617 trace->filter_pids.entries);
3618 }
3619 } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
3620 err = trace__set_filter_loop_pids(trace);
3621 }
3622
3623 return err;
3624 }
3625
3626 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
3627 {
3628 struct evlist *evlist = trace->evlist;
3629 struct perf_sample sample;
3630 int err = evlist__parse_sample(evlist, event, &sample);
3631
3632 if (err)
3633 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3634 else
3635 trace__handle_event(trace, event, &sample);
3636
3637 return 0;
3638 }
3639
3640 static int __trace__flush_events(struct trace *trace)
3641 {
3642 u64 first = ordered_events__first_time(&trace->oe.data);
3643 u64 flush = trace->oe.last - NSEC_PER_SEC;
3644
3645 /* Is there some thing to flush.. */
3646 if (first && first < flush)
3647 return ordered_events__flush_time(&trace->oe.data, flush);
3648
3649 return 0;
3650 }
3651
3652 static int trace__flush_events(struct trace *trace)
3653 {
3654 return !trace->sort_events ? 0 : __trace__flush_events(trace);
3655 }
3656
3657 static int trace__deliver_event(struct trace *trace, union perf_event *event)
3658 {
3659 int err;
3660
3661 if (!trace->sort_events)
3662 return __trace__deliver_event(trace, event);
3663
3664 err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
3665 if (err && err != -1)
3666 return err;
3667
3668 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0, NULL);
3669 if (err)
3670 return err;
3671
3672 return trace__flush_events(trace);
3673 }
3674
3675 static int ordered_events__deliver_event(struct ordered_events *oe,
3676 struct ordered_event *event)
3677 {
3678 struct trace *trace = container_of(oe, struct trace, oe.data);
3679
3680 return __trace__deliver_event(trace, event->event);
3681 }
3682
3683 static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg)
3684 {
3685 struct tep_format_field *field;
3686 struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel);
3687
3688 if (evsel->tp_format == NULL || fmt == NULL)
3689 return NULL;
3690
3691 for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt)
3692 if (strcmp(field->name, arg) == 0)
3693 return fmt;
3694
3695 return NULL;
3696 }
3697
3698 static int trace__expand_filter(struct trace *trace __maybe_unused, struct evsel *evsel)
3699 {
3700 char *tok, *left = evsel->filter, *new_filter = evsel->filter;
3701
3702 while ((tok = strpbrk(left, "=<>!")) != NULL) {
3703 char *right = tok + 1, *right_end;
3704
3705 if (*right == '=')
3706 ++right;
3707
3708 while (isspace(*right))
3709 ++right;
3710
3711 if (*right == '\0')
3712 break;
3713
3714 while (!isalpha(*left))
3715 if (++left == tok) {
3716 /*
3717 * Bail out, can't find the name of the argument that is being
3718 * used in the filter, let it try to set this filter, will fail later.
3719 */
3720 return 0;
3721 }
3722
3723 right_end = right + 1;
3724 while (isalnum(*right_end) || *right_end == '_' || *right_end == '|')
3725 ++right_end;
3726
3727 if (isalpha(*right)) {
3728 struct syscall_arg_fmt *fmt;
3729 int left_size = tok - left,
3730 right_size = right_end - right;
3731 char arg[128];
3732
3733 while (isspace(left[left_size - 1]))
3734 --left_size;
3735
3736 scnprintf(arg, sizeof(arg), "%.*s", left_size, left);
3737
3738 fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg);
3739 if (fmt == NULL) {
3740 pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n",
3741 arg, evsel->name, evsel->filter);
3742 return -1;
3743 }
3744
3745 pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ",
3746 arg, (int)(right - tok), tok, right_size, right);
3747
3748 if (fmt->strtoul) {
3749 u64 val;
3750 struct syscall_arg syscall_arg = {
3751 .parm = fmt->parm,
3752 };
3753
3754 if (fmt->strtoul(right, right_size, &syscall_arg, &val)) {
3755 char *n, expansion[19];
3756 int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val);
3757 int expansion_offset = right - new_filter;
3758
3759 pr_debug("%s", expansion);
3760
3761 if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) {
3762 pr_debug(" out of memory!\n");
3763 free(new_filter);
3764 return -1;
3765 }
3766 if (new_filter != evsel->filter)
3767 free(new_filter);
3768 left = n + expansion_offset + expansion_lenght;
3769 new_filter = n;
3770 } else {
3771 pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3772 right_size, right, arg, evsel->name, evsel->filter);
3773 return -1;
3774 }
3775 } else {
3776 pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3777 arg, evsel->name, evsel->filter);
3778 return -1;
3779 }
3780
3781 pr_debug("\n");
3782 } else {
3783 left = right_end;
3784 }
3785 }
3786
3787 if (new_filter != evsel->filter) {
3788 pr_debug("New filter for %s: %s\n", evsel->name, new_filter);
3789 evsel__set_filter(evsel, new_filter);
3790 free(new_filter);
3791 }
3792
3793 return 0;
3794 }
3795
3796 static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel)
3797 {
3798 struct evlist *evlist = trace->evlist;
3799 struct evsel *evsel;
3800
3801 evlist__for_each_entry(evlist, evsel) {
3802 if (evsel->filter == NULL)
3803 continue;
3804
3805 if (trace__expand_filter(trace, evsel)) {
3806 *err_evsel = evsel;
3807 return -1;
3808 }
3809 }
3810
3811 return 0;
3812 }
3813
3814 static int trace__run(struct trace *trace, int argc, const char **argv)
3815 {
3816 struct evlist *evlist = trace->evlist;
3817 struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
3818 int err = -1, i;
3819 unsigned long before;
3820 const bool forks = argc > 0;
3821 bool draining = false;
3822
3823 trace->live = true;
3824
3825 if (!trace->raw_augmented_syscalls) {
3826 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3827 goto out_error_raw_syscalls;
3828
3829 if (trace->trace_syscalls)
3830 trace->vfs_getname = evlist__add_vfs_getname(evlist);
3831 }
3832
3833 if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
3834 pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
3835 if (pgfault_maj == NULL)
3836 goto out_error_mem;
3837 evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
3838 evlist__add(evlist, pgfault_maj);
3839 }
3840
3841 if ((trace->trace_pgfaults & TRACE_PFMIN)) {
3842 pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
3843 if (pgfault_min == NULL)
3844 goto out_error_mem;
3845 evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
3846 evlist__add(evlist, pgfault_min);
3847 }
3848
3849 /* Enable ignoring missing threads when -u/-p option is defined. */
3850 trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid;
3851
3852 if (trace->sched &&
3853 evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime))
3854 goto out_error_sched_stat_runtime;
3855 /*
3856 * If a global cgroup was set, apply it to all the events without an
3857 * explicit cgroup. I.e.:
3858 *
3859 * trace -G A -e sched:*switch
3860 *
3861 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3862 * _and_ sched:sched_switch to the 'A' cgroup, while:
3863 *
3864 * trace -e sched:*switch -G A
3865 *
3866 * will only set the sched:sched_switch event to the 'A' cgroup, all the
3867 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3868 * a cgroup (on the root cgroup, sys wide, etc).
3869 *
3870 * Multiple cgroups:
3871 *
3872 * trace -G A -e sched:*switch -G B
3873 *
3874 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
3875 * to the 'B' cgroup.
3876 *
3877 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
3878 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
3879 */
3880 if (trace->cgroup)
3881 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
3882
3883 err = evlist__create_maps(evlist, &trace->opts.target);
3884 if (err < 0) {
3885 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
3886 goto out_delete_evlist;
3887 }
3888
3889 err = trace__symbols_init(trace, evlist);
3890 if (err < 0) {
3891 fprintf(trace->output, "Problems initializing symbol libraries!\n");
3892 goto out_delete_evlist;
3893 }
3894
3895 evlist__config(evlist, &trace->opts, &callchain_param);
3896
3897 if (forks) {
3898 err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL);
3899 if (err < 0) {
3900 fprintf(trace->output, "Couldn't run the workload!\n");
3901 goto out_delete_evlist;
3902 }
3903 workload_pid = evlist->workload.pid;
3904 }
3905
3906 err = evlist__open(evlist);
3907 if (err < 0)
3908 goto out_error_open;
3909 #ifdef HAVE_BPF_SKEL
3910 if (trace->syscalls.events.bpf_output) {
3911 struct perf_cpu cpu;
3912
3913 /*
3914 * Set up the __augmented_syscalls__ BPF map to hold for each
3915 * CPU the bpf-output event's file descriptor.
3916 */
3917 perf_cpu_map__for_each_cpu(cpu, i, trace->syscalls.events.bpf_output->core.cpus) {
3918 bpf_map__update_elem(trace->skel->maps.__augmented_syscalls__,
3919 &cpu.cpu, sizeof(int),
3920 xyarray__entry(trace->syscalls.events.bpf_output->core.fd,
3921 cpu.cpu, 0),
3922 sizeof(__u32), BPF_ANY);
3923 }
3924 }
3925 #endif
3926 err = trace__set_filter_pids(trace);
3927 if (err < 0)
3928 goto out_error_mem;
3929
3930 #ifdef HAVE_BPF_SKEL
3931 if (trace->skel && trace->skel->progs.sys_enter)
3932 trace__init_syscalls_bpf_prog_array_maps(trace);
3933 #endif
3934
3935 if (trace->ev_qualifier_ids.nr > 0) {
3936 err = trace__set_ev_qualifier_filter(trace);
3937 if (err < 0)
3938 goto out_errno;
3939
3940 if (trace->syscalls.events.sys_exit) {
3941 pr_debug("event qualifier tracepoint filter: %s\n",
3942 trace->syscalls.events.sys_exit->filter);
3943 }
3944 }
3945
3946 /*
3947 * If the "close" syscall is not traced, then we will not have the
3948 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
3949 * fd->pathname table and were ending up showing the last value set by
3950 * syscalls opening a pathname and associating it with a descriptor or
3951 * reading it from /proc/pid/fd/ in cases where that doesn't make
3952 * sense.
3953 *
3954 * So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
3955 * not in use.
3956 */
3957 trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
3958
3959 err = trace__expand_filters(trace, &evsel);
3960 if (err)
3961 goto out_delete_evlist;
3962 err = evlist__apply_filters(evlist, &evsel);
3963 if (err < 0)
3964 goto out_error_apply_filters;
3965
3966 err = evlist__mmap(evlist, trace->opts.mmap_pages);
3967 if (err < 0)
3968 goto out_error_mmap;
3969
3970 if (!target__none(&trace->opts.target) && !trace->opts.target.initial_delay)
3971 evlist__enable(evlist);
3972
3973 if (forks)
3974 evlist__start_workload(evlist);
3975
3976 if (trace->opts.target.initial_delay) {
3977 usleep(trace->opts.target.initial_delay * 1000);
3978 evlist__enable(evlist);
3979 }
3980
3981 trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
3982 perf_thread_map__nr(evlist->core.threads) > 1 ||
3983 evlist__first(evlist)->core.attr.inherit;
3984
3985 /*
3986 * Now that we already used evsel->core.attr to ask the kernel to setup the
3987 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
3988 * trace__resolve_callchain(), allowing per-event max-stack settings
3989 * to override an explicitly set --max-stack global setting.
3990 */
3991 evlist__for_each_entry(evlist, evsel) {
3992 if (evsel__has_callchain(evsel) &&
3993 evsel->core.attr.sample_max_stack == 0)
3994 evsel->core.attr.sample_max_stack = trace->max_stack;
3995 }
3996 again:
3997 before = trace->nr_events;
3998
3999 for (i = 0; i < evlist->core.nr_mmaps; i++) {
4000 union perf_event *event;
4001 struct mmap *md;
4002
4003 md = &evlist->mmap[i];
4004 if (perf_mmap__read_init(&md->core) < 0)
4005 continue;
4006
4007 while ((event = perf_mmap__read_event(&md->core)) != NULL) {
4008 ++trace->nr_events;
4009
4010 err = trace__deliver_event(trace, event);
4011 if (err)
4012 goto out_disable;
4013
4014 perf_mmap__consume(&md->core);
4015
4016 if (interrupted)
4017 goto out_disable;
4018
4019 if (done && !draining) {
4020 evlist__disable(evlist);
4021 draining = true;
4022 }
4023 }
4024 perf_mmap__read_done(&md->core);
4025 }
4026
4027 if (trace->nr_events == before) {
4028 int timeout = done ? 100 : -1;
4029
4030 if (!draining && evlist__poll(evlist, timeout) > 0) {
4031 if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
4032 draining = true;
4033
4034 goto again;
4035 } else {
4036 if (trace__flush_events(trace))
4037 goto out_disable;
4038 }
4039 } else {
4040 goto again;
4041 }
4042
4043 out_disable:
4044 thread__zput(trace->current);
4045
4046 evlist__disable(evlist);
4047
4048 if (trace->sort_events)
4049 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
4050
4051 if (!err) {
4052 if (trace->summary)
4053 trace__fprintf_thread_summary(trace, trace->output);
4054
4055 if (trace->show_tool_stats) {
4056 fprintf(trace->output, "Stats:\n "
4057 " vfs_getname : %" PRIu64 "\n"
4058 " proc_getname: %" PRIu64 "\n",
4059 trace->stats.vfs_getname,
4060 trace->stats.proc_getname);
4061 }
4062 }
4063
4064 out_delete_evlist:
4065 trace__symbols__exit(trace);
4066 evlist__free_syscall_tp_fields(evlist);
4067 evlist__delete(evlist);
4068 cgroup__put(trace->cgroup);
4069 trace->evlist = NULL;
4070 trace->live = false;
4071 return err;
4072 {
4073 char errbuf[BUFSIZ];
4074
4075 out_error_sched_stat_runtime:
4076 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
4077 goto out_error;
4078
4079 out_error_raw_syscalls:
4080 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
4081 goto out_error;
4082
4083 out_error_mmap:
4084 evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
4085 goto out_error;
4086
4087 out_error_open:
4088 evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
4089
4090 out_error:
4091 fprintf(trace->output, "%s\n", errbuf);
4092 goto out_delete_evlist;
4093
4094 out_error_apply_filters:
4095 fprintf(trace->output,
4096 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
4097 evsel->filter, evsel__name(evsel), errno,
4098 str_error_r(errno, errbuf, sizeof(errbuf)));
4099 goto out_delete_evlist;
4100 }
4101 out_error_mem:
4102 fprintf(trace->output, "Not enough memory to run!\n");
4103 goto out_delete_evlist;
4104
4105 out_errno:
4106 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
4107 goto out_delete_evlist;
4108 }
4109
4110 static int trace__replay(struct trace *trace)
4111 {
4112 const struct evsel_str_handler handlers[] = {
4113 { "probe:vfs_getname", trace__vfs_getname, },
4114 };
4115 struct perf_data data = {
4116 .path = input_name,
4117 .mode = PERF_DATA_MODE_READ,
4118 .force = trace->force,
4119 };
4120 struct perf_session *session;
4121 struct evsel *evsel;
4122 int err = -1;
4123
4124 trace->tool.sample = trace__process_sample;
4125 trace->tool.mmap = perf_event__process_mmap;
4126 trace->tool.mmap2 = perf_event__process_mmap2;
4127 trace->tool.comm = perf_event__process_comm;
4128 trace->tool.exit = perf_event__process_exit;
4129 trace->tool.fork = perf_event__process_fork;
4130 trace->tool.attr = perf_event__process_attr;
4131 trace->tool.tracing_data = perf_event__process_tracing_data;
4132 trace->tool.build_id = perf_event__process_build_id;
4133 trace->tool.namespaces = perf_event__process_namespaces;
4134
4135 trace->tool.ordered_events = true;
4136 trace->tool.ordering_requires_timestamps = true;
4137
4138 /* add tid to output */
4139 trace->multiple_threads = true;
4140
4141 session = perf_session__new(&data, &trace->tool);
4142 if (IS_ERR(session))
4143 return PTR_ERR(session);
4144
4145 if (trace->opts.target.pid)
4146 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
4147
4148 if (trace->opts.target.tid)
4149 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
4150
4151 if (symbol__init(&session->header.env) < 0)
4152 goto out;
4153
4154 trace->host = &session->machines.host;
4155
4156 err = perf_session__set_tracepoints_handlers(session, handlers);
4157 if (err)
4158 goto out;
4159
4160 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter");
4161 trace->syscalls.events.sys_enter = evsel;
4162 /* older kernels have syscalls tp versus raw_syscalls */
4163 if (evsel == NULL)
4164 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter");
4165
4166 if (evsel &&
4167 (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
4168 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
4169 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
4170 goto out;
4171 }
4172
4173 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit");
4174 trace->syscalls.events.sys_exit = evsel;
4175 if (evsel == NULL)
4176 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit");
4177 if (evsel &&
4178 (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
4179 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
4180 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
4181 goto out;
4182 }
4183
4184 evlist__for_each_entry(session->evlist, evsel) {
4185 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
4186 (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
4187 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
4188 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
4189 evsel->handler = trace__pgfault;
4190 }
4191
4192 setup_pager();
4193
4194 err = perf_session__process_events(session);
4195 if (err)
4196 pr_err("Failed to process events, error %d", err);
4197
4198 else if (trace->summary)
4199 trace__fprintf_thread_summary(trace, trace->output);
4200
4201 out:
4202 perf_session__delete(session);
4203
4204 return err;
4205 }
4206
4207 static size_t trace__fprintf_threads_header(FILE *fp)
4208 {
4209 size_t printed;
4210
4211 printed = fprintf(fp, "\n Summary of events:\n\n");
4212
4213 return printed;
4214 }
4215
4216 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
4217 struct syscall_stats *stats;
4218 double msecs;
4219 int syscall;
4220 )
4221 {
4222 struct int_node *source = rb_entry(nd, struct int_node, rb_node);
4223 struct syscall_stats *stats = source->priv;
4224
4225 entry->syscall = source->i;
4226 entry->stats = stats;
4227 entry->msecs = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0;
4228 }
4229
4230 static size_t thread__dump_stats(struct thread_trace *ttrace,
4231 struct trace *trace, FILE *fp)
4232 {
4233 size_t printed = 0;
4234 struct syscall *sc;
4235 struct rb_node *nd;
4236 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
4237
4238 if (syscall_stats == NULL)
4239 return 0;
4240
4241 printed += fprintf(fp, "\n");
4242
4243 printed += fprintf(fp, " syscall calls errors total min avg max stddev\n");
4244 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
4245 printed += fprintf(fp, " --------------- -------- ------ -------- --------- --------- --------- ------\n");
4246
4247 resort_rb__for_each_entry(nd, syscall_stats) {
4248 struct syscall_stats *stats = syscall_stats_entry->stats;
4249 if (stats) {
4250 double min = (double)(stats->stats.min) / NSEC_PER_MSEC;
4251 double max = (double)(stats->stats.max) / NSEC_PER_MSEC;
4252 double avg = avg_stats(&stats->stats);
4253 double pct;
4254 u64 n = (u64)stats->stats.n;
4255
4256 pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0;
4257 avg /= NSEC_PER_MSEC;
4258
4259 sc = &trace->syscalls.table[syscall_stats_entry->syscall];
4260 printed += fprintf(fp, " %-15s", sc->name);
4261 printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f",
4262 n, stats->nr_failures, syscall_stats_entry->msecs, min, avg);
4263 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
4264
4265 if (trace->errno_summary && stats->nr_failures) {
4266 int e;
4267
4268 for (e = 0; e < stats->max_errno; ++e) {
4269 if (stats->errnos[e] != 0)
4270 fprintf(fp, "\t\t\t\t%s: %d\n", perf_env__arch_strerrno(trace->host->env, e + 1), stats->errnos[e]);
4271 }
4272 }
4273 }
4274 }
4275
4276 resort_rb__delete(syscall_stats);
4277 printed += fprintf(fp, "\n\n");
4278
4279 return printed;
4280 }
4281
4282 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
4283 {
4284 size_t printed = 0;
4285 struct thread_trace *ttrace = thread__priv(thread);
4286 double ratio;
4287
4288 if (ttrace == NULL)
4289 return 0;
4290
4291 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
4292
4293 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread__tid(thread));
4294 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
4295 printed += fprintf(fp, "%.1f%%", ratio);
4296 if (ttrace->pfmaj)
4297 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
4298 if (ttrace->pfmin)
4299 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
4300 if (trace->sched)
4301 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
4302 else if (fputc('\n', fp) != EOF)
4303 ++printed;
4304
4305 printed += thread__dump_stats(ttrace, trace, fp);
4306
4307 return printed;
4308 }
4309
4310 static unsigned long thread__nr_events(struct thread_trace *ttrace)
4311 {
4312 return ttrace ? ttrace->nr_events : 0;
4313 }
4314
4315 static int trace_nr_events_cmp(void *priv __maybe_unused,
4316 const struct list_head *la,
4317 const struct list_head *lb)
4318 {
4319 struct thread_list *a = list_entry(la, struct thread_list, list);
4320 struct thread_list *b = list_entry(lb, struct thread_list, list);
4321 unsigned long a_nr_events = thread__nr_events(thread__priv(a->thread));
4322 unsigned long b_nr_events = thread__nr_events(thread__priv(b->thread));
4323
4324 if (a_nr_events != b_nr_events)
4325 return a_nr_events < b_nr_events ? -1 : 1;
4326
4327 /* Identical number of threads, place smaller tids first. */
4328 return thread__tid(a->thread) < thread__tid(b->thread)
4329 ? -1
4330 : (thread__tid(a->thread) > thread__tid(b->thread) ? 1 : 0);
4331 }
4332
4333 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
4334 {
4335 size_t printed = trace__fprintf_threads_header(fp);
4336 LIST_HEAD(threads);
4337
4338 if (machine__thread_list(trace->host, &threads) == 0) {
4339 struct thread_list *pos;
4340
4341 list_sort(NULL, &threads, trace_nr_events_cmp);
4342
4343 list_for_each_entry(pos, &threads, list)
4344 printed += trace__fprintf_thread(fp, pos->thread, trace);
4345 }
4346 thread_list__delete(&threads);
4347 return printed;
4348 }
4349
4350 static int trace__set_duration(const struct option *opt, const char *str,
4351 int unset __maybe_unused)
4352 {
4353 struct trace *trace = opt->value;
4354
4355 trace->duration_filter = atof(str);
4356 return 0;
4357 }
4358
4359 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
4360 int unset __maybe_unused)
4361 {
4362 int ret = -1;
4363 size_t i;
4364 struct trace *trace = opt->value;
4365 /*
4366 * FIXME: introduce a intarray class, plain parse csv and create a
4367 * { int nr, int entries[] } struct...
4368 */
4369 struct intlist *list = intlist__new(str);
4370
4371 if (list == NULL)
4372 return -1;
4373
4374 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
4375 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
4376
4377 if (trace->filter_pids.entries == NULL)
4378 goto out;
4379
4380 trace->filter_pids.entries[0] = getpid();
4381
4382 for (i = 1; i < trace->filter_pids.nr; ++i)
4383 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
4384
4385 intlist__delete(list);
4386 ret = 0;
4387 out:
4388 return ret;
4389 }
4390
4391 static int trace__open_output(struct trace *trace, const char *filename)
4392 {
4393 struct stat st;
4394
4395 if (!stat(filename, &st) && st.st_size) {
4396 char oldname[PATH_MAX];
4397
4398 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
4399 unlink(oldname);
4400 rename(filename, oldname);
4401 }
4402
4403 trace->output = fopen(filename, "w");
4404
4405 return trace->output == NULL ? -errno : 0;
4406 }
4407
4408 static int parse_pagefaults(const struct option *opt, const char *str,
4409 int unset __maybe_unused)
4410 {
4411 int *trace_pgfaults = opt->value;
4412
4413 if (strcmp(str, "all") == 0)
4414 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
4415 else if (strcmp(str, "maj") == 0)
4416 *trace_pgfaults |= TRACE_PFMAJ;
4417 else if (strcmp(str, "min") == 0)
4418 *trace_pgfaults |= TRACE_PFMIN;
4419 else
4420 return -1;
4421
4422 return 0;
4423 }
4424
4425 static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)
4426 {
4427 struct evsel *evsel;
4428
4429 evlist__for_each_entry(evlist, evsel) {
4430 if (evsel->handler == NULL)
4431 evsel->handler = handler;
4432 }
4433 }
4434
4435 static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name)
4436 {
4437 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
4438
4439 if (fmt) {
4440 const struct syscall_fmt *scfmt = syscall_fmt__find(name);
4441
4442 if (scfmt) {
4443 int skip = 0;
4444
4445 if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 ||
4446 strcmp(evsel->tp_format->format.fields->name, "nr") == 0)
4447 ++skip;
4448
4449 memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt));
4450 }
4451 }
4452 }
4453
4454 static int evlist__set_syscall_tp_fields(struct evlist *evlist)
4455 {
4456 struct evsel *evsel;
4457
4458 evlist__for_each_entry(evlist, evsel) {
4459 if (evsel->priv || !evsel->tp_format)
4460 continue;
4461
4462 if (strcmp(evsel->tp_format->system, "syscalls")) {
4463 evsel__init_tp_arg_scnprintf(evsel);
4464 continue;
4465 }
4466
4467 if (evsel__init_syscall_tp(evsel))
4468 return -1;
4469
4470 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
4471 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4472
4473 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
4474 return -1;
4475
4476 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1);
4477 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
4478 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4479
4480 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
4481 return -1;
4482
4483 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1);
4484 }
4485 }
4486
4487 return 0;
4488 }
4489
4490 /*
4491 * XXX: Hackish, just splitting the combined -e+--event (syscalls
4492 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
4493 * existing facilities unchanged (trace->ev_qualifier + parse_options()).
4494 *
4495 * It'd be better to introduce a parse_options() variant that would return a
4496 * list with the terms it didn't match to an event...
4497 */
4498 static int trace__parse_events_option(const struct option *opt, const char *str,
4499 int unset __maybe_unused)
4500 {
4501 struct trace *trace = (struct trace *)opt->value;
4502 const char *s = str;
4503 char *sep = NULL, *lists[2] = { NULL, NULL, };
4504 int len = strlen(str) + 1, err = -1, list, idx;
4505 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
4506 char group_name[PATH_MAX];
4507 const struct syscall_fmt *fmt;
4508
4509 if (strace_groups_dir == NULL)
4510 return -1;
4511
4512 if (*s == '!') {
4513 ++s;
4514 trace->not_ev_qualifier = true;
4515 }
4516
4517 while (1) {
4518 if ((sep = strchr(s, ',')) != NULL)
4519 *sep = '\0';
4520
4521 list = 0;
4522 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
4523 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
4524 list = 1;
4525 goto do_concat;
4526 }
4527
4528 fmt = syscall_fmt__find_by_alias(s);
4529 if (fmt != NULL) {
4530 list = 1;
4531 s = fmt->name;
4532 } else {
4533 path__join(group_name, sizeof(group_name), strace_groups_dir, s);
4534 if (access(group_name, R_OK) == 0)
4535 list = 1;
4536 }
4537 do_concat:
4538 if (lists[list]) {
4539 sprintf(lists[list] + strlen(lists[list]), ",%s", s);
4540 } else {
4541 lists[list] = malloc(len);
4542 if (lists[list] == NULL)
4543 goto out;
4544 strcpy(lists[list], s);
4545 }
4546
4547 if (!sep)
4548 break;
4549
4550 *sep = ',';
4551 s = sep + 1;
4552 }
4553
4554 if (lists[1] != NULL) {
4555 struct strlist_config slist_config = {
4556 .dirname = strace_groups_dir,
4557 };
4558
4559 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4560 if (trace->ev_qualifier == NULL) {
4561 fputs("Not enough memory to parse event qualifier", trace->output);
4562 goto out;
4563 }
4564
4565 if (trace__validate_ev_qualifier(trace))
4566 goto out;
4567 trace->trace_syscalls = true;
4568 }
4569
4570 err = 0;
4571
4572 if (lists[0]) {
4573 struct parse_events_option_args parse_events_option_args = {
4574 .evlistp = &trace->evlist,
4575 };
4576 struct option o = {
4577 .value = &parse_events_option_args,
4578 };
4579 err = parse_events_option(&o, lists[0], 0);
4580 }
4581 out:
4582 free(strace_groups_dir);
4583 free(lists[0]);
4584 free(lists[1]);
4585 if (sep)
4586 *sep = ',';
4587
4588 return err;
4589 }
4590
4591 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4592 {
4593 struct trace *trace = opt->value;
4594
4595 if (!list_empty(&trace->evlist->core.entries)) {
4596 struct option o = {
4597 .value = &trace->evlist,
4598 };
4599 return parse_cgroups(&o, str, unset);
4600 }
4601 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
4602
4603 return 0;
4604 }
4605
4606 static int trace__config(const char *var, const char *value, void *arg)
4607 {
4608 struct trace *trace = arg;
4609 int err = 0;
4610
4611 if (!strcmp(var, "trace.add_events")) {
4612 trace->perfconfig_events = strdup(value);
4613 if (trace->perfconfig_events == NULL) {
4614 pr_err("Not enough memory for %s\n", "trace.add_events");
4615 return -1;
4616 }
4617 } else if (!strcmp(var, "trace.show_timestamp")) {
4618 trace->show_tstamp = perf_config_bool(var, value);
4619 } else if (!strcmp(var, "trace.show_duration")) {
4620 trace->show_duration = perf_config_bool(var, value);
4621 } else if (!strcmp(var, "trace.show_arg_names")) {
4622 trace->show_arg_names = perf_config_bool(var, value);
4623 if (!trace->show_arg_names)
4624 trace->show_zeros = true;
4625 } else if (!strcmp(var, "trace.show_zeros")) {
4626 bool new_show_zeros = perf_config_bool(var, value);
4627 if (!trace->show_arg_names && !new_show_zeros) {
4628 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4629 goto out;
4630 }
4631 trace->show_zeros = new_show_zeros;
4632 } else if (!strcmp(var, "trace.show_prefix")) {
4633 trace->show_string_prefix = perf_config_bool(var, value);
4634 } else if (!strcmp(var, "trace.no_inherit")) {
4635 trace->opts.no_inherit = perf_config_bool(var, value);
4636 } else if (!strcmp(var, "trace.args_alignment")) {
4637 int args_alignment = 0;
4638 if (perf_config_int(&args_alignment, var, value) == 0)
4639 trace->args_alignment = args_alignment;
4640 } else if (!strcmp(var, "trace.tracepoint_beautifiers")) {
4641 if (strcasecmp(value, "libtraceevent") == 0)
4642 trace->libtraceevent_print = true;
4643 else if (strcasecmp(value, "libbeauty") == 0)
4644 trace->libtraceevent_print = false;
4645 }
4646 out:
4647 return err;
4648 }
4649
4650 static void trace__exit(struct trace *trace)
4651 {
4652 int i;
4653
4654 strlist__delete(trace->ev_qualifier);
4655 zfree(&trace->ev_qualifier_ids.entries);
4656 if (trace->syscalls.table) {
4657 for (i = 0; i <= trace->sctbl->syscalls.max_id; i++)
4658 syscall__exit(&trace->syscalls.table[i]);
4659 zfree(&trace->syscalls.table);
4660 }
4661 syscalltbl__delete(trace->sctbl);
4662 zfree(&trace->perfconfig_events);
4663 }
4664
4665 #ifdef HAVE_BPF_SKEL
4666 static int bpf__setup_bpf_output(struct evlist *evlist)
4667 {
4668 int err = parse_event(evlist, "bpf-output/no-inherit=1,name=__augmented_syscalls__/");
4669
4670 if (err)
4671 pr_debug("ERROR: failed to create the \"__augmented_syscalls__\" bpf-output event\n");
4672
4673 return err;
4674 }
4675 #endif
4676
4677 int cmd_trace(int argc, const char **argv)
4678 {
4679 const char *trace_usage[] = {
4680 "perf trace [<options>] [<command>]",
4681 "perf trace [<options>] -- <command> [<options>]",
4682 "perf trace record [<options>] [<command>]",
4683 "perf trace record [<options>] -- <command> [<options>]",
4684 NULL
4685 };
4686 struct trace trace = {
4687 .opts = {
4688 .target = {
4689 .uid = UINT_MAX,
4690 .uses_mmap = true,
4691 },
4692 .user_freq = UINT_MAX,
4693 .user_interval = ULLONG_MAX,
4694 .no_buffering = true,
4695 .mmap_pages = UINT_MAX,
4696 },
4697 .output = stderr,
4698 .show_comm = true,
4699 .show_tstamp = true,
4700 .show_duration = true,
4701 .show_arg_names = true,
4702 .args_alignment = 70,
4703 .trace_syscalls = false,
4704 .kernel_syscallchains = false,
4705 .max_stack = UINT_MAX,
4706 .max_events = ULONG_MAX,
4707 };
4708 const char *output_name = NULL;
4709 const struct option trace_options[] = {
4710 OPT_CALLBACK('e', "event", &trace, "event",
4711 "event/syscall selector. use 'perf list' to list available events",
4712 trace__parse_events_option),
4713 OPT_CALLBACK(0, "filter", &trace.evlist, "filter",
4714 "event filter", parse_filter),
4715 OPT_BOOLEAN(0, "comm", &trace.show_comm,
4716 "show the thread COMM next to its id"),
4717 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
4718 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4719 trace__parse_events_option),
4720 OPT_STRING('o', "output", &output_name, "file", "output file name"),
4721 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
4722 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4723 "trace events on existing process id"),
4724 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
4725 "trace events on existing thread id"),
4726 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
4727 "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
4728 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
4729 "system-wide collection from all CPUs"),
4730 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
4731 "list of cpus to monitor"),
4732 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
4733 "child tasks do not inherit counters"),
4734 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
4735 "number of mmap data pages", evlist__parse_mmap_pages),
4736 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
4737 "user to profile"),
4738 OPT_CALLBACK(0, "duration", &trace, "float",
4739 "show only events with duration > N.M ms",
4740 trace__set_duration),
4741 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
4742 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4743 OPT_BOOLEAN('T', "time", &trace.full_time,
4744 "Show full timestamp, not time relative to first start"),
4745 OPT_BOOLEAN(0, "failure", &trace.failure_only,
4746 "Show only syscalls that failed"),
4747 OPT_BOOLEAN('s', "summary", &trace.summary_only,
4748 "Show only syscall summary with statistics"),
4749 OPT_BOOLEAN('S', "with-summary", &trace.summary,
4750 "Show all syscalls and summary with statistics"),
4751 OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary,
4752 "Show errno stats per syscall, use with -s or -S"),
4753 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4754 "Trace pagefaults", parse_pagefaults, "maj"),
4755 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
4756 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
4757 OPT_CALLBACK(0, "call-graph", &trace.opts,
4758 "record_mode[,record_size]", record_callchain_help,
4759 &record_parse_callchain_opt),
4760 OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,
4761 "Use libtraceevent to print the tracepoint arguments."),
4762 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4763 "Show the kernel callchains on the syscall exit path"),
4764 OPT_ULONG(0, "max-events", &trace.max_events,
4765 "Set the maximum number of events to print, exit after that is reached. "),
4766 OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4767 "Set the minimum stack depth when parsing the callchain, "
4768 "anything below the specified depth will be ignored."),
4769 OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4770 "Set the maximum stack depth when parsing the callchain, "
4771 "anything beyond the specified depth will be ignored. "
4772 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
4773 OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4774 "Sort batch of events before processing, use if getting out of order events"),
4775 OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4776 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
4777 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
4778 "per thread proc mmap processing timeout in ms"),
4779 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4780 trace__parse_cgroups),
4781 OPT_INTEGER('D', "delay", &trace.opts.target.initial_delay,
4782 "ms to wait before starting measurement after program "
4783 "start"),
4784 OPTS_EVSWITCH(&trace.evswitch),
4785 OPT_END()
4786 };
4787 bool __maybe_unused max_stack_user_set = true;
4788 bool mmap_pages_user_set = true;
4789 struct evsel *evsel;
4790 const char * const trace_subcommands[] = { "record", NULL };
4791 int err = -1;
4792 char bf[BUFSIZ];
4793 struct sigaction sigchld_act;
4794
4795 signal(SIGSEGV, sighandler_dump_stack);
4796 signal(SIGFPE, sighandler_dump_stack);
4797 signal(SIGINT, sighandler_interrupt);
4798
4799 memset(&sigchld_act, 0, sizeof(sigchld_act));
4800 sigchld_act.sa_flags = SA_SIGINFO;
4801 sigchld_act.sa_sigaction = sighandler_chld;
4802 sigaction(SIGCHLD, &sigchld_act, NULL);
4803
4804 trace.evlist = evlist__new();
4805 trace.sctbl = syscalltbl__new();
4806
4807 if (trace.evlist == NULL || trace.sctbl == NULL) {
4808 pr_err("Not enough memory to run!\n");
4809 err = -ENOMEM;
4810 goto out;
4811 }
4812
4813 /*
4814 * Parsing .perfconfig may entail creating a BPF event, that may need
4815 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4816 * is too small. This affects just this process, not touching the
4817 * global setting. If it fails we'll get something in 'perf trace -v'
4818 * to help diagnose the problem.
4819 */
4820 rlimit__bump_memlock();
4821
4822 err = perf_config(trace__config, &trace);
4823 if (err)
4824 goto out;
4825
4826 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4827 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
4828
4829 /*
4830 * Here we already passed thru trace__parse_events_option() and it has
4831 * already figured out if -e syscall_name, if not but if --event
4832 * foo:bar was used, the user is interested _just_ in those, say,
4833 * tracepoint events, not in the strace-like syscall-name-based mode.
4834 *
4835 * This is important because we need to check if strace-like mode is
4836 * needed to decided if we should filter out the eBPF
4837 * __augmented_syscalls__ code, if it is in the mix, say, via
4838 * .perfconfig trace.add_events, and filter those out.
4839 */
4840 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4841 trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4842 trace.trace_syscalls = true;
4843 }
4844 /*
4845 * Now that we have --verbose figured out, lets see if we need to parse
4846 * events from .perfconfig, so that if those events fail parsing, say some
4847 * BPF program fails, then we'll be able to use --verbose to see what went
4848 * wrong in more detail.
4849 */
4850 if (trace.perfconfig_events != NULL) {
4851 struct parse_events_error parse_err;
4852
4853 parse_events_error__init(&parse_err);
4854 err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);
4855 if (err)
4856 parse_events_error__print(&parse_err, trace.perfconfig_events);
4857 parse_events_error__exit(&parse_err);
4858 if (err)
4859 goto out;
4860 }
4861
4862 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4863 usage_with_options_msg(trace_usage, trace_options,
4864 "cgroup monitoring only available in system-wide mode");
4865 }
4866
4867 #ifdef HAVE_BPF_SKEL
4868 if (!trace.trace_syscalls)
4869 goto skip_augmentation;
4870
4871 if ((argc >= 1) && (strcmp(argv[0], "record") == 0)) {
4872 pr_debug("Syscall augmentation fails with record, disabling augmentation");
4873 goto skip_augmentation;
4874 }
4875
4876 trace.skel = augmented_raw_syscalls_bpf__open();
4877 if (!trace.skel) {
4878 pr_debug("Failed to open augmented syscalls BPF skeleton");
4879 } else {
4880 /*
4881 * Disable attaching the BPF programs except for sys_enter and
4882 * sys_exit that tail call into this as necessary.
4883 */
4884 struct bpf_program *prog;
4885
4886 bpf_object__for_each_program(prog, trace.skel->obj) {
4887 if (prog != trace.skel->progs.sys_enter && prog != trace.skel->progs.sys_exit)
4888 bpf_program__set_autoattach(prog, /*autoattach=*/false);
4889 }
4890
4891 err = augmented_raw_syscalls_bpf__load(trace.skel);
4892
4893 if (err < 0) {
4894 libbpf_strerror(err, bf, sizeof(bf));
4895 pr_debug("Failed to load augmented syscalls BPF skeleton: %s\n", bf);
4896 } else {
4897 augmented_raw_syscalls_bpf__attach(trace.skel);
4898 trace__add_syscall_newtp(&trace);
4899 }
4900 }
4901
4902 err = bpf__setup_bpf_output(trace.evlist);
4903 if (err) {
4904 libbpf_strerror(err, bf, sizeof(bf));
4905 pr_err("ERROR: Setup BPF output event failed: %s\n", bf);
4906 goto out;
4907 }
4908 trace.syscalls.events.bpf_output = evlist__last(trace.evlist);
4909 assert(evsel__name_is(trace.syscalls.events.bpf_output, "__augmented_syscalls__"));
4910 skip_augmentation:
4911 #endif
4912 err = -1;
4913
4914 if (trace.trace_pgfaults) {
4915 trace.opts.sample_address = true;
4916 trace.opts.sample_time = true;
4917 }
4918
4919 if (trace.opts.mmap_pages == UINT_MAX)
4920 mmap_pages_user_set = false;
4921
4922 if (trace.max_stack == UINT_MAX) {
4923 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
4924 max_stack_user_set = false;
4925 }
4926
4927 #ifdef HAVE_DWARF_UNWIND_SUPPORT
4928 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
4929 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
4930 }
4931 #endif
4932
4933 if (callchain_param.enabled) {
4934 if (!mmap_pages_user_set && geteuid() == 0)
4935 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
4936
4937 symbol_conf.use_callchain = true;
4938 }
4939
4940 if (trace.evlist->core.nr_entries > 0) {
4941 evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);
4942 if (evlist__set_syscall_tp_fields(trace.evlist)) {
4943 perror("failed to set syscalls:* tracepoint fields");
4944 goto out;
4945 }
4946 }
4947
4948 if (trace.sort_events) {
4949 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
4950 ordered_events__set_copy_on_queue(&trace.oe.data, true);
4951 }
4952
4953 /*
4954 * If we are augmenting syscalls, then combine what we put in the
4955 * __augmented_syscalls__ BPF map with what is in the
4956 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
4957 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
4958 *
4959 * We'll switch to look at two BPF maps, one for sys_enter and the
4960 * other for sys_exit when we start augmenting the sys_exit paths with
4961 * buffers that are being copied from kernel to userspace, think 'read'
4962 * syscall.
4963 */
4964 if (trace.syscalls.events.bpf_output) {
4965 evlist__for_each_entry(trace.evlist, evsel) {
4966 bool raw_syscalls_sys_exit = evsel__name_is(evsel, "raw_syscalls:sys_exit");
4967
4968 if (raw_syscalls_sys_exit) {
4969 trace.raw_augmented_syscalls = true;
4970 goto init_augmented_syscall_tp;
4971 }
4972
4973 if (trace.syscalls.events.bpf_output->priv == NULL &&
4974 strstr(evsel__name(evsel), "syscalls:sys_enter")) {
4975 struct evsel *augmented = trace.syscalls.events.bpf_output;
4976 if (evsel__init_augmented_syscall_tp(augmented, evsel) ||
4977 evsel__init_augmented_syscall_tp_args(augmented))
4978 goto out;
4979 /*
4980 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
4981 * Above we made sure we can get from the payload the tp fields
4982 * that we get from syscalls:sys_enter tracefs format file.
4983 */
4984 augmented->handler = trace__sys_enter;
4985 /*
4986 * Now we do the same for the *syscalls:sys_enter event so that
4987 * if we handle it directly, i.e. if the BPF prog returns 0 so
4988 * as not to filter it, then we'll handle it just like we would
4989 * for the BPF_OUTPUT one:
4990 */
4991 if (evsel__init_augmented_syscall_tp(evsel, evsel) ||
4992 evsel__init_augmented_syscall_tp_args(evsel))
4993 goto out;
4994 evsel->handler = trace__sys_enter;
4995 }
4996
4997 if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) {
4998 struct syscall_tp *sc;
4999 init_augmented_syscall_tp:
5000 if (evsel__init_augmented_syscall_tp(evsel, evsel))
5001 goto out;
5002 sc = __evsel__syscall_tp(evsel);
5003 /*
5004 * For now with BPF raw_augmented we hook into
5005 * raw_syscalls:sys_enter and there we get all
5006 * 6 syscall args plus the tracepoint common
5007 * fields and the syscall_nr (another long).
5008 * So we check if that is the case and if so
5009 * don't look after the sc->args_size but
5010 * always after the full raw_syscalls:sys_enter
5011 * payload, which is fixed.
5012 *
5013 * We'll revisit this later to pass
5014 * s->args_size to the BPF augmenter (now
5015 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
5016 * so that it copies only what we need for each
5017 * syscall, like what happens when we use
5018 * syscalls:sys_enter_NAME, so that we reduce
5019 * the kernel/userspace traffic to just what is
5020 * needed for each syscall.
5021 */
5022 if (trace.raw_augmented_syscalls)
5023 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
5024 evsel__init_augmented_syscall_tp_ret(evsel);
5025 evsel->handler = trace__sys_exit;
5026 }
5027 }
5028 }
5029
5030 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
5031 return trace__record(&trace, argc-1, &argv[1]);
5032
5033 /* Using just --errno-summary will trigger --summary */
5034 if (trace.errno_summary && !trace.summary && !trace.summary_only)
5035 trace.summary_only = true;
5036
5037 /* summary_only implies summary option, but don't overwrite summary if set */
5038 if (trace.summary_only)
5039 trace.summary = trace.summary_only;
5040
5041 if (output_name != NULL) {
5042 err = trace__open_output(&trace, output_name);
5043 if (err < 0) {
5044 perror("failed to create output file");
5045 goto out;
5046 }
5047 }
5048
5049 err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
5050 if (err)
5051 goto out_close;
5052
5053 err = target__validate(&trace.opts.target);
5054 if (err) {
5055 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5056 fprintf(trace.output, "%s", bf);
5057 goto out_close;
5058 }
5059
5060 err = target__parse_uid(&trace.opts.target);
5061 if (err) {
5062 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5063 fprintf(trace.output, "%s", bf);
5064 goto out_close;
5065 }
5066
5067 if (!argc && target__none(&trace.opts.target))
5068 trace.opts.target.system_wide = true;
5069
5070 if (input_name)
5071 err = trace__replay(&trace);
5072 else
5073 err = trace__run(&trace, argc, argv);
5074
5075 out_close:
5076 if (output_name != NULL)
5077 fclose(trace.output);
5078 out:
5079 trace__exit(&trace);
5080 #ifdef HAVE_BPF_SKEL
5081 augmented_raw_syscalls_bpf__destroy(trace.skel);
5082 #endif
5083 return err;
5084 }
5085
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.