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 <traceevent/event-parse.h>
19 #include <api/fs/tracing_path.h>
20 #ifdef HAVE_LIBBPF_SUPPORT
21 #include <bpf/bpf.h>
22 #endif
23 #include "util/bpf_map.h"
24 #include "util/rlimit.h"
25 #include "builtin.h"
26 #include "util/cgroup.h"
27 #include "util/color.h"
28 #include "util/config.h"
29 #include "util/debug.h"
30 #include "util/dso.h"
31 #include "util/env.h"
32 #include "util/event.h"
33 #include "util/evsel.h"
34 #include "util/evsel_fprintf.h"
35 #include "util/synthetic-events.h"
36 #include "util/evlist.h"
37 #include "util/evswitch.h"
38 #include "util/mmap.h"
39 #include <subcmd/pager.h>
40 #include <subcmd/exec-cmd.h>
41 #include "util/machine.h"
42 #include "util/map.h"
43 #include "util/symbol.h"
44 #include "util/path.h"
45 #include "util/session.h"
46 #include "util/thread.h"
47 #include <subcmd/parse-options.h>
48 #include "util/strlist.h"
49 #include "util/intlist.h"
50 #include "util/thread_map.h"
51 #include "util/stat.h"
52 #include "util/tool.h"
53 #include "util/util.h"
54 #include "trace/beauty/beauty.h"
55 #include "trace-event.h"
56 #include "util/parse-events.h"
57 #include "util/bpf-loader.h"
58 #include "util/tracepoint.h"
59 #include "callchain.h"
60 #include "print_binary.h"
61 #include "string2.h"
62 #include "syscalltbl.h"
63 #include "rb_resort.h"
64 #include "../perf.h"
65
66 #include <errno.h>
67 #include <inttypes.h>
68 #include <poll.h>
69 #include <signal.h>
70 #include <stdlib.h>
71 #include <string.h>
72 #include <linux/err.h>
73 #include <linux/filter.h>
74 #include <linux/kernel.h>
75 #include <linux/random.h>
76 #include <linux/stringify.h>
77 #include <linux/time64.h>
78 #include <linux/zalloc.h>
79 #include <fcntl.h>
80 #include <sys/sysmacros.h>
81
82 #include <linux/ctype.h>
83 #include <perf/mmap.h>
84
85 #ifndef O_CLOEXEC
86 # define O_CLOEXEC 02000000
87 #endif
88
89 #ifndef F_LINUX_SPECIFIC_BASE
90 # define F_LINUX_SPECIFIC_BASE 1024
91 #endif
92
93 #define RAW_SYSCALL_ARGS_NUM 6
94
95 /*
96 * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100
97 */
98 struct syscall_arg_fmt {
99 size_t (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
100 bool (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val);
101 unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
102 void *parm;
103 const char *name;
104 u16 nr_entries; // for arrays
105 bool show_zero;
106 };
107
108 struct syscall_fmt {
109 const char *name;
110 const char *alias;
111 struct {
112 const char *sys_enter,
113 *sys_exit;
114 } bpf_prog_name;
115 struct syscall_arg_fmt arg[RAW_SYSCALL_ARGS_NUM];
116 u8 nr_args;
117 bool errpid;
118 bool timeout;
119 bool hexret;
120 };
121
122 struct trace {
123 struct perf_tool tool;
124 struct syscalltbl *sctbl;
125 struct {
126 struct syscall *table;
127 struct bpf_map *map;
128 struct { // per syscall BPF_MAP_TYPE_PROG_ARRAY
129 struct bpf_map *sys_enter,
130 *sys_exit;
131 } prog_array;
132 struct {
133 struct evsel *sys_enter,
134 *sys_exit,
135 *augmented;
136 } events;
137 struct bpf_program *unaugmented_prog;
138 } syscalls;
139 struct {
140 struct bpf_map *map;
141 } dump;
142 struct record_opts opts;
143 struct evlist *evlist;
144 struct machine *host;
145 struct thread *current;
146 struct bpf_object *bpf_obj;
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
__tp_field__init_uint(struct tp_field * field,int size,int offset,bool needs_swap)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
tp_field__init_uint(struct tp_field * field,struct tep_format_field * format_field,bool needs_swap)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
tp_field__ptr(struct tp_field * field,struct perf_sample * sample)269 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
270 {
271 return sample->raw_data + field->offset;
272 }
273
__tp_field__init_ptr(struct tp_field * field,int offset)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
tp_field__init_ptr(struct tp_field * field,struct tep_format_field * format_field)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
evsel_trace__new(void)303 static struct evsel_trace *evsel_trace__new(void)
304 {
305 return zalloc(sizeof(struct evsel_trace));
306 }
307
evsel_trace__delete(struct evsel_trace * et)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 */
__evsel__syscall_tp(struct evsel * evsel)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
evsel__syscall_tp(struct evsel * evsel)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 */
__evsel__syscall_arg_fmt(struct evsel * evsel)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
evsel__syscall_arg_fmt(struct evsel * evsel)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
evsel__init_tp_uint_field(struct evsel * evsel,struct tp_field * field,const char * name)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
evsel__init_tp_ptr_field(struct evsel * evsel,struct tp_field * field,const char * name)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
evsel__delete_priv(struct evsel * evsel)402 static void evsel__delete_priv(struct evsel *evsel)
403 {
404 zfree(&evsel->priv);
405 evsel__delete(evsel);
406 }
407
evsel__init_syscall_tp(struct evsel * evsel)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 return 0;
417 }
418
419 return -ENOMEM;
420 }
421
evsel__init_augmented_syscall_tp(struct evsel * evsel,struct evsel * tp)422 static int evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)
423 {
424 struct syscall_tp *sc = evsel__syscall_tp(evsel);
425
426 if (sc != NULL) {
427 struct tep_format_field *syscall_id = evsel__field(tp, "id");
428 if (syscall_id == NULL)
429 syscall_id = evsel__field(tp, "__syscall_nr");
430 if (syscall_id == NULL ||
431 __tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
432 return -EINVAL;
433
434 return 0;
435 }
436
437 return -ENOMEM;
438 }
439
evsel__init_augmented_syscall_tp_args(struct evsel * evsel)440 static int evsel__init_augmented_syscall_tp_args(struct evsel *evsel)
441 {
442 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
443
444 return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
445 }
446
evsel__init_augmented_syscall_tp_ret(struct evsel * evsel)447 static int evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)
448 {
449 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
450
451 return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
452 }
453
evsel__init_raw_syscall_tp(struct evsel * evsel,void * handler)454 static int evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)
455 {
456 if (evsel__syscall_tp(evsel) != NULL) {
457 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
458 return -ENOENT;
459
460 evsel->handler = handler;
461 return 0;
462 }
463
464 return -ENOMEM;
465 }
466
perf_evsel__raw_syscall_newtp(const char * direction,void * handler)467 static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
468 {
469 struct evsel *evsel = evsel__newtp("raw_syscalls", direction);
470
471 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
472 if (IS_ERR(evsel))
473 evsel = evsel__newtp("syscalls", direction);
474
475 if (IS_ERR(evsel))
476 return NULL;
477
478 if (evsel__init_raw_syscall_tp(evsel, handler))
479 goto out_delete;
480
481 return evsel;
482
483 out_delete:
484 evsel__delete_priv(evsel);
485 return NULL;
486 }
487
488 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
489 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
490 fields->name.integer(&fields->name, sample); })
491
492 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
493 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
494 fields->name.pointer(&fields->name, sample); })
495
strarray__scnprintf_suffix(struct strarray * sa,char * bf,size_t size,const char * intfmt,bool show_suffix,int val)496 size_t strarray__scnprintf_suffix(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_suffix, int val)
497 {
498 int idx = val - sa->offset;
499
500 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
501 size_t printed = scnprintf(bf, size, intfmt, val);
502 if (show_suffix)
503 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
504 return printed;
505 }
506
507 return scnprintf(bf, size, "%s%s", sa->entries[idx], show_suffix ? sa->prefix : "");
508 }
509
strarray__scnprintf(struct strarray * sa,char * bf,size_t size,const char * intfmt,bool show_prefix,int val)510 size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
511 {
512 int idx = val - sa->offset;
513
514 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
515 size_t printed = scnprintf(bf, size, intfmt, val);
516 if (show_prefix)
517 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
518 return printed;
519 }
520
521 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
522 }
523
__syscall_arg__scnprintf_strarray(char * bf,size_t size,const char * intfmt,struct syscall_arg * arg)524 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
525 const char *intfmt,
526 struct syscall_arg *arg)
527 {
528 return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
529 }
530
syscall_arg__scnprintf_strarray(char * bf,size_t size,struct syscall_arg * arg)531 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
532 struct syscall_arg *arg)
533 {
534 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
535 }
536
537 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
538
syscall_arg__strtoul_strarray(char * bf,size_t size,struct syscall_arg * arg,u64 * ret)539 bool syscall_arg__strtoul_strarray(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
540 {
541 return strarray__strtoul(arg->parm, bf, size, ret);
542 }
543
syscall_arg__strtoul_strarray_flags(char * bf,size_t size,struct syscall_arg * arg,u64 * ret)544 bool syscall_arg__strtoul_strarray_flags(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
545 {
546 return strarray__strtoul_flags(arg->parm, bf, size, ret);
547 }
548
syscall_arg__strtoul_strarrays(char * bf,size_t size,struct syscall_arg * arg,u64 * ret)549 bool syscall_arg__strtoul_strarrays(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
550 {
551 return strarrays__strtoul(arg->parm, bf, size, ret);
552 }
553
syscall_arg__scnprintf_strarray_flags(char * bf,size_t size,struct syscall_arg * arg)554 size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)
555 {
556 return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);
557 }
558
strarrays__scnprintf(struct strarrays * sas,char * bf,size_t size,const char * intfmt,bool show_prefix,int val)559 size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
560 {
561 size_t printed;
562 int i;
563
564 for (i = 0; i < sas->nr_entries; ++i) {
565 struct strarray *sa = sas->entries[i];
566 int idx = val - sa->offset;
567
568 if (idx >= 0 && idx < sa->nr_entries) {
569 if (sa->entries[idx] == NULL)
570 break;
571 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
572 }
573 }
574
575 printed = scnprintf(bf, size, intfmt, val);
576 if (show_prefix)
577 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
578 return printed;
579 }
580
strarray__strtoul(struct strarray * sa,char * bf,size_t size,u64 * ret)581 bool strarray__strtoul(struct strarray *sa, char *bf, size_t size, u64 *ret)
582 {
583 int i;
584
585 for (i = 0; i < sa->nr_entries; ++i) {
586 if (sa->entries[i] && strncmp(sa->entries[i], bf, size) == 0 && sa->entries[i][size] == '\0') {
587 *ret = sa->offset + i;
588 return true;
589 }
590 }
591
592 return false;
593 }
594
strarray__strtoul_flags(struct strarray * sa,char * bf,size_t size,u64 * ret)595 bool strarray__strtoul_flags(struct strarray *sa, char *bf, size_t size, u64 *ret)
596 {
597 u64 val = 0;
598 char *tok = bf, *sep, *end;
599
600 *ret = 0;
601
602 while (size != 0) {
603 int toklen = size;
604
605 sep = memchr(tok, '|', size);
606 if (sep != NULL) {
607 size -= sep - tok + 1;
608
609 end = sep - 1;
610 while (end > tok && isspace(*end))
611 --end;
612
613 toklen = end - tok + 1;
614 }
615
616 while (isspace(*tok))
617 ++tok;
618
619 if (isalpha(*tok) || *tok == '_') {
620 if (!strarray__strtoul(sa, tok, toklen, &val))
621 return false;
622 } else
623 val = strtoul(tok, NULL, 0);
624
625 *ret |= (1 << (val - 1));
626
627 if (sep == NULL)
628 break;
629 tok = sep + 1;
630 }
631
632 return true;
633 }
634
strarrays__strtoul(struct strarrays * sas,char * bf,size_t size,u64 * ret)635 bool strarrays__strtoul(struct strarrays *sas, char *bf, size_t size, u64 *ret)
636 {
637 int i;
638
639 for (i = 0; i < sas->nr_entries; ++i) {
640 struct strarray *sa = sas->entries[i];
641
642 if (strarray__strtoul(sa, bf, size, ret))
643 return true;
644 }
645
646 return false;
647 }
648
syscall_arg__scnprintf_strarrays(char * bf,size_t size,struct syscall_arg * arg)649 size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
650 struct syscall_arg *arg)
651 {
652 return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
653 }
654
655 #ifndef AT_FDCWD
656 #define AT_FDCWD -100
657 #endif
658
syscall_arg__scnprintf_fd_at(char * bf,size_t size,struct syscall_arg * arg)659 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
660 struct syscall_arg *arg)
661 {
662 int fd = arg->val;
663 const char *prefix = "AT_FD";
664
665 if (fd == AT_FDCWD)
666 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
667
668 return syscall_arg__scnprintf_fd(bf, size, arg);
669 }
670
671 #define SCA_FDAT syscall_arg__scnprintf_fd_at
672
673 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
674 struct syscall_arg *arg);
675
676 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
677
syscall_arg__scnprintf_hex(char * bf,size_t size,struct syscall_arg * arg)678 size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
679 {
680 return scnprintf(bf, size, "%#lx", arg->val);
681 }
682
syscall_arg__scnprintf_ptr(char * bf,size_t size,struct syscall_arg * arg)683 size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
684 {
685 if (arg->val == 0)
686 return scnprintf(bf, size, "NULL");
687 return syscall_arg__scnprintf_hex(bf, size, arg);
688 }
689
syscall_arg__scnprintf_int(char * bf,size_t size,struct syscall_arg * arg)690 size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
691 {
692 return scnprintf(bf, size, "%d", arg->val);
693 }
694
syscall_arg__scnprintf_long(char * bf,size_t size,struct syscall_arg * arg)695 size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
696 {
697 return scnprintf(bf, size, "%ld", arg->val);
698 }
699
syscall_arg__scnprintf_char_array(char * bf,size_t size,struct syscall_arg * arg)700 static size_t syscall_arg__scnprintf_char_array(char *bf, size_t size, struct syscall_arg *arg)
701 {
702 // XXX Hey, maybe for sched:sched_switch prev/next comm fields we can
703 // fill missing comms using thread__set_comm()...
704 // here or in a special syscall_arg__scnprintf_pid_sched_tp...
705 return scnprintf(bf, size, "\"%-.*s\"", arg->fmt->nr_entries ?: arg->len, arg->val);
706 }
707
708 #define SCA_CHAR_ARRAY syscall_arg__scnprintf_char_array
709
710 static const char *bpf_cmd[] = {
711 "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
712 "MAP_GET_NEXT_KEY", "PROG_LOAD", "OBJ_PIN", "OBJ_GET", "PROG_ATTACH",
713 "PROG_DETACH", "PROG_TEST_RUN", "PROG_GET_NEXT_ID", "MAP_GET_NEXT_ID",
714 "PROG_GET_FD_BY_ID", "MAP_GET_FD_BY_ID", "OBJ_GET_INFO_BY_FD",
715 "PROG_QUERY", "RAW_TRACEPOINT_OPEN", "BTF_LOAD", "BTF_GET_FD_BY_ID",
716 "TASK_FD_QUERY", "MAP_LOOKUP_AND_DELETE_ELEM", "MAP_FREEZE",
717 "BTF_GET_NEXT_ID", "MAP_LOOKUP_BATCH", "MAP_LOOKUP_AND_DELETE_BATCH",
718 "MAP_UPDATE_BATCH", "MAP_DELETE_BATCH", "LINK_CREATE", "LINK_UPDATE",
719 "LINK_GET_FD_BY_ID", "LINK_GET_NEXT_ID", "ENABLE_STATS", "ITER_CREATE",
720 "LINK_DETACH", "PROG_BIND_MAP",
721 };
722 static DEFINE_STRARRAY(bpf_cmd, "BPF_");
723
724 static const char *fsmount_flags[] = {
725 [1] = "CLOEXEC",
726 };
727 static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");
728
729 #include "trace/beauty/generated/fsconfig_arrays.c"
730
731 static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");
732
733 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
734 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
735
736 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
737 static DEFINE_STRARRAY(itimers, "ITIMER_");
738
739 static const char *keyctl_options[] = {
740 "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
741 "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
742 "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
743 "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
744 "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
745 };
746 static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
747
748 static const char *whences[] = { "SET", "CUR", "END",
749 #ifdef SEEK_DATA
750 "DATA",
751 #endif
752 #ifdef SEEK_HOLE
753 "HOLE",
754 #endif
755 };
756 static DEFINE_STRARRAY(whences, "SEEK_");
757
758 static const char *fcntl_cmds[] = {
759 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
760 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
761 "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
762 "GETOWNER_UIDS",
763 };
764 static DEFINE_STRARRAY(fcntl_cmds, "F_");
765
766 static const char *fcntl_linux_specific_cmds[] = {
767 "SETLEASE", "GETLEASE", "NOTIFY", [5] = "CANCELLK", "DUPFD_CLOEXEC",
768 "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
769 "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
770 };
771
772 static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
773
774 static struct strarray *fcntl_cmds_arrays[] = {
775 &strarray__fcntl_cmds,
776 &strarray__fcntl_linux_specific_cmds,
777 };
778
779 static DEFINE_STRARRAYS(fcntl_cmds_arrays);
780
781 static const char *rlimit_resources[] = {
782 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
783 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
784 "RTTIME",
785 };
786 static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
787
788 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
789 static DEFINE_STRARRAY(sighow, "SIG_");
790
791 static const char *clockid[] = {
792 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
793 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
794 "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
795 };
796 static DEFINE_STRARRAY(clockid, "CLOCK_");
797
syscall_arg__scnprintf_access_mode(char * bf,size_t size,struct syscall_arg * arg)798 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
799 struct syscall_arg *arg)
800 {
801 bool show_prefix = arg->show_string_prefix;
802 const char *suffix = "_OK";
803 size_t printed = 0;
804 int mode = arg->val;
805
806 if (mode == F_OK) /* 0 */
807 return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
808 #define P_MODE(n) \
809 if (mode & n##_OK) { \
810 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
811 mode &= ~n##_OK; \
812 }
813
814 P_MODE(R);
815 P_MODE(W);
816 P_MODE(X);
817 #undef P_MODE
818
819 if (mode)
820 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
821
822 return printed;
823 }
824
825 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
826
827 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
828 struct syscall_arg *arg);
829
830 #define SCA_FILENAME syscall_arg__scnprintf_filename
831
syscall_arg__scnprintf_pipe_flags(char * bf,size_t size,struct syscall_arg * arg)832 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
833 struct syscall_arg *arg)
834 {
835 bool show_prefix = arg->show_string_prefix;
836 const char *prefix = "O_";
837 int printed = 0, flags = arg->val;
838
839 #define P_FLAG(n) \
840 if (flags & O_##n) { \
841 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
842 flags &= ~O_##n; \
843 }
844
845 P_FLAG(CLOEXEC);
846 P_FLAG(NONBLOCK);
847 #undef P_FLAG
848
849 if (flags)
850 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
851
852 return printed;
853 }
854
855 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
856
857 #ifndef GRND_NONBLOCK
858 #define GRND_NONBLOCK 0x0001
859 #endif
860 #ifndef GRND_RANDOM
861 #define GRND_RANDOM 0x0002
862 #endif
863
syscall_arg__scnprintf_getrandom_flags(char * bf,size_t size,struct syscall_arg * arg)864 static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
865 struct syscall_arg *arg)
866 {
867 bool show_prefix = arg->show_string_prefix;
868 const char *prefix = "GRND_";
869 int printed = 0, flags = arg->val;
870
871 #define P_FLAG(n) \
872 if (flags & GRND_##n) { \
873 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
874 flags &= ~GRND_##n; \
875 }
876
877 P_FLAG(RANDOM);
878 P_FLAG(NONBLOCK);
879 #undef P_FLAG
880
881 if (flags)
882 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
883
884 return printed;
885 }
886
887 #define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
888
889 #define STRARRAY(name, array) \
890 { .scnprintf = SCA_STRARRAY, \
891 .strtoul = STUL_STRARRAY, \
892 .parm = &strarray__##array, }
893
894 #define STRARRAY_FLAGS(name, array) \
895 { .scnprintf = SCA_STRARRAY_FLAGS, \
896 .strtoul = STUL_STRARRAY_FLAGS, \
897 .parm = &strarray__##array, }
898
899 #include "trace/beauty/arch_errno_names.c"
900 #include "trace/beauty/eventfd.c"
901 #include "trace/beauty/futex_op.c"
902 #include "trace/beauty/futex_val3.c"
903 #include "trace/beauty/mmap.c"
904 #include "trace/beauty/mode_t.c"
905 #include "trace/beauty/msg_flags.c"
906 #include "trace/beauty/open_flags.c"
907 #include "trace/beauty/perf_event_open.c"
908 #include "trace/beauty/pid.c"
909 #include "trace/beauty/sched_policy.c"
910 #include "trace/beauty/seccomp.c"
911 #include "trace/beauty/signum.c"
912 #include "trace/beauty/socket_type.c"
913 #include "trace/beauty/waitid_options.c"
914
915 static struct syscall_fmt syscall_fmts[] = {
916 { .name = "access",
917 .arg = { [1] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, },
918 { .name = "arch_prctl",
919 .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
920 [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
921 { .name = "bind",
922 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
923 [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ },
924 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
925 { .name = "bpf",
926 .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
927 { .name = "brk", .hexret = true,
928 .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
929 { .name = "clock_gettime",
930 .arg = { [0] = STRARRAY(clk_id, clockid), }, },
931 { .name = "clone", .errpid = true, .nr_args = 5,
932 .arg = { [0] = { .name = "flags", .scnprintf = SCA_CLONE_FLAGS, },
933 [1] = { .name = "child_stack", .scnprintf = SCA_HEX, },
934 [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
935 [3] = { .name = "child_tidptr", .scnprintf = SCA_HEX, },
936 [4] = { .name = "tls", .scnprintf = SCA_HEX, }, }, },
937 { .name = "close",
938 .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
939 { .name = "connect",
940 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
941 [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ },
942 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
943 { .name = "epoll_ctl",
944 .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
945 { .name = "eventfd2",
946 .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
947 { .name = "fchmodat",
948 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
949 { .name = "fchownat",
950 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
951 { .name = "fcntl",
952 .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
953 .strtoul = STUL_STRARRAYS,
954 .parm = &strarrays__fcntl_cmds_arrays,
955 .show_zero = true, },
956 [2] = { .scnprintf = SCA_FCNTL_ARG, /* arg */ }, }, },
957 { .name = "flock",
958 .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
959 { .name = "fsconfig",
960 .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },
961 { .name = "fsmount",
962 .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),
963 [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },
964 { .name = "fspick",
965 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
966 [1] = { .scnprintf = SCA_FILENAME, /* path */ },
967 [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },
968 { .name = "fstat", .alias = "newfstat", },
969 { .name = "fstatat", .alias = "newfstatat", },
970 { .name = "futex",
971 .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
972 [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
973 { .name = "futimesat",
974 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
975 { .name = "getitimer",
976 .arg = { [0] = STRARRAY(which, itimers), }, },
977 { .name = "getpid", .errpid = true, },
978 { .name = "getpgid", .errpid = true, },
979 { .name = "getppid", .errpid = true, },
980 { .name = "getrandom",
981 .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
982 { .name = "getrlimit",
983 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
984 { .name = "getsockopt",
985 .arg = { [1] = STRARRAY(level, socket_level), }, },
986 { .name = "gettid", .errpid = true, },
987 { .name = "ioctl",
988 .arg = {
989 #if defined(__i386__) || defined(__x86_64__)
990 /*
991 * FIXME: Make this available to all arches.
992 */
993 [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
994 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
995 #else
996 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
997 #endif
998 { .name = "kcmp", .nr_args = 5,
999 .arg = { [0] = { .name = "pid1", .scnprintf = SCA_PID, },
1000 [1] = { .name = "pid2", .scnprintf = SCA_PID, },
1001 [2] = { .name = "type", .scnprintf = SCA_KCMP_TYPE, },
1002 [3] = { .name = "idx1", .scnprintf = SCA_KCMP_IDX, },
1003 [4] = { .name = "idx2", .scnprintf = SCA_KCMP_IDX, }, }, },
1004 { .name = "keyctl",
1005 .arg = { [0] = STRARRAY(option, keyctl_options), }, },
1006 { .name = "kill",
1007 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1008 { .name = "linkat",
1009 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1010 { .name = "lseek",
1011 .arg = { [2] = STRARRAY(whence, whences), }, },
1012 { .name = "lstat", .alias = "newlstat", },
1013 { .name = "madvise",
1014 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1015 [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
1016 { .name = "mkdirat",
1017 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1018 { .name = "mknodat",
1019 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1020 { .name = "mmap", .hexret = true,
1021 /* The standard mmap maps to old_mmap on s390x */
1022 #if defined(__s390x__)
1023 .alias = "old_mmap",
1024 #endif
1025 .arg = { [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
1026 [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */
1027 .strtoul = STUL_STRARRAY_FLAGS,
1028 .parm = &strarray__mmap_flags, },
1029 [5] = { .scnprintf = SCA_HEX, /* offset */ }, }, },
1030 { .name = "mount",
1031 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
1032 [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
1033 .mask_val = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
1034 { .name = "move_mount",
1035 .arg = { [0] = { .scnprintf = SCA_FDAT, /* from_dfd */ },
1036 [1] = { .scnprintf = SCA_FILENAME, /* from_pathname */ },
1037 [2] = { .scnprintf = SCA_FDAT, /* to_dfd */ },
1038 [3] = { .scnprintf = SCA_FILENAME, /* to_pathname */ },
1039 [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },
1040 { .name = "mprotect",
1041 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1042 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ }, }, },
1043 { .name = "mq_unlink",
1044 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
1045 { .name = "mremap", .hexret = true,
1046 .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
1047 { .name = "name_to_handle_at",
1048 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1049 { .name = "newfstatat",
1050 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1051 { .name = "open",
1052 .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1053 { .name = "open_by_handle_at",
1054 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1055 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1056 { .name = "openat",
1057 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1058 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1059 { .name = "perf_event_open",
1060 .arg = { [2] = { .scnprintf = SCA_INT, /* cpu */ },
1061 [3] = { .scnprintf = SCA_FD, /* group_fd */ },
1062 [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
1063 { .name = "pipe2",
1064 .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
1065 { .name = "pkey_alloc",
1066 .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS, /* access_rights */ }, }, },
1067 { .name = "pkey_free",
1068 .arg = { [0] = { .scnprintf = SCA_INT, /* key */ }, }, },
1069 { .name = "pkey_mprotect",
1070 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1071 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
1072 [3] = { .scnprintf = SCA_INT, /* pkey */ }, }, },
1073 { .name = "poll", .timeout = true, },
1074 { .name = "ppoll", .timeout = true, },
1075 { .name = "prctl",
1076 .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */
1077 .strtoul = STUL_STRARRAY,
1078 .parm = &strarray__prctl_options, },
1079 [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
1080 [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
1081 { .name = "pread", .alias = "pread64", },
1082 { .name = "preadv", .alias = "pread", },
1083 { .name = "prlimit64",
1084 .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
1085 { .name = "pwrite", .alias = "pwrite64", },
1086 { .name = "readlinkat",
1087 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1088 { .name = "recvfrom",
1089 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1090 { .name = "recvmmsg",
1091 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1092 { .name = "recvmsg",
1093 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1094 { .name = "renameat",
1095 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1096 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
1097 { .name = "renameat2",
1098 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1099 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
1100 [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
1101 { .name = "rt_sigaction",
1102 .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1103 { .name = "rt_sigprocmask",
1104 .arg = { [0] = STRARRAY(how, sighow), }, },
1105 { .name = "rt_sigqueueinfo",
1106 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1107 { .name = "rt_tgsigqueueinfo",
1108 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1109 { .name = "sched_setscheduler",
1110 .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
1111 { .name = "seccomp",
1112 .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP, /* op */ },
1113 [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
1114 { .name = "select", .timeout = true, },
1115 { .name = "sendfile", .alias = "sendfile64", },
1116 { .name = "sendmmsg",
1117 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1118 { .name = "sendmsg",
1119 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1120 { .name = "sendto",
1121 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
1122 [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
1123 { .name = "set_tid_address", .errpid = true, },
1124 { .name = "setitimer",
1125 .arg = { [0] = STRARRAY(which, itimers), }, },
1126 { .name = "setrlimit",
1127 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
1128 { .name = "setsockopt",
1129 .arg = { [1] = STRARRAY(level, socket_level), }, },
1130 { .name = "socket",
1131 .arg = { [0] = STRARRAY(family, socket_families),
1132 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1133 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1134 { .name = "socketpair",
1135 .arg = { [0] = STRARRAY(family, socket_families),
1136 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1137 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1138 { .name = "stat", .alias = "newstat", },
1139 { .name = "statx",
1140 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fdat */ },
1141 [2] = { .scnprintf = SCA_STATX_FLAGS, /* flags */ } ,
1142 [3] = { .scnprintf = SCA_STATX_MASK, /* mask */ }, }, },
1143 { .name = "swapoff",
1144 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1145 { .name = "swapon",
1146 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1147 { .name = "symlinkat",
1148 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1149 { .name = "sync_file_range",
1150 .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },
1151 { .name = "tgkill",
1152 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1153 { .name = "tkill",
1154 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1155 { .name = "umount2", .alias = "umount",
1156 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
1157 { .name = "uname", .alias = "newuname", },
1158 { .name = "unlinkat",
1159 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1160 { .name = "utimensat",
1161 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
1162 { .name = "wait4", .errpid = true,
1163 .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
1164 { .name = "waitid", .errpid = true,
1165 .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
1166 };
1167
syscall_fmt__cmp(const void * name,const void * fmtp)1168 static int syscall_fmt__cmp(const void *name, const void *fmtp)
1169 {
1170 const struct syscall_fmt *fmt = fmtp;
1171 return strcmp(name, fmt->name);
1172 }
1173
__syscall_fmt__find(struct syscall_fmt * fmts,const int nmemb,const char * name)1174 static struct syscall_fmt *__syscall_fmt__find(struct syscall_fmt *fmts, const int nmemb, const char *name)
1175 {
1176 return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1177 }
1178
syscall_fmt__find(const char * name)1179 static struct syscall_fmt *syscall_fmt__find(const char *name)
1180 {
1181 const int nmemb = ARRAY_SIZE(syscall_fmts);
1182 return __syscall_fmt__find(syscall_fmts, nmemb, name);
1183 }
1184
__syscall_fmt__find_by_alias(struct syscall_fmt * fmts,const int nmemb,const char * alias)1185 static struct syscall_fmt *__syscall_fmt__find_by_alias(struct syscall_fmt *fmts, const int nmemb, const char *alias)
1186 {
1187 int i;
1188
1189 for (i = 0; i < nmemb; ++i) {
1190 if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0)
1191 return &fmts[i];
1192 }
1193
1194 return NULL;
1195 }
1196
syscall_fmt__find_by_alias(const char * alias)1197 static struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
1198 {
1199 const int nmemb = ARRAY_SIZE(syscall_fmts);
1200 return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias);
1201 }
1202
1203 /*
1204 * is_exit: is this "exit" or "exit_group"?
1205 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
1206 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
1207 * nonexistent: Just a hole in the syscall table, syscall id not allocated
1208 */
1209 struct syscall {
1210 struct tep_event *tp_format;
1211 int nr_args;
1212 int args_size;
1213 struct {
1214 struct bpf_program *sys_enter,
1215 *sys_exit;
1216 } bpf_prog;
1217 bool is_exit;
1218 bool is_open;
1219 bool nonexistent;
1220 struct tep_format_field *args;
1221 const char *name;
1222 struct syscall_fmt *fmt;
1223 struct syscall_arg_fmt *arg_fmt;
1224 };
1225
1226 /*
1227 * Must match what is in the BPF program:
1228 *
1229 * tools/perf/examples/bpf/augmented_raw_syscalls.c
1230 */
1231 struct bpf_map_syscall_entry {
1232 bool enabled;
1233 u16 string_args_len[RAW_SYSCALL_ARGS_NUM];
1234 };
1235
1236 /*
1237 * We need to have this 'calculated' boolean because in some cases we really
1238 * don't know what is the duration of a syscall, for instance, when we start
1239 * a session and some threads are waiting for a syscall to finish, say 'poll',
1240 * in which case all we can do is to print "( ? ) for duration and for the
1241 * start timestamp.
1242 */
fprintf_duration(unsigned long t,bool calculated,FILE * fp)1243 static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
1244 {
1245 double duration = (double)t / NSEC_PER_MSEC;
1246 size_t printed = fprintf(fp, "(");
1247
1248 if (!calculated)
1249 printed += fprintf(fp, " ");
1250 else if (duration >= 1.0)
1251 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1252 else if (duration >= 0.01)
1253 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1254 else
1255 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
1256 return printed + fprintf(fp, "): ");
1257 }
1258
1259 /**
1260 * filename.ptr: The filename char pointer that will be vfs_getname'd
1261 * filename.entry_str_pos: Where to insert the string translated from
1262 * filename.ptr by the vfs_getname tracepoint/kprobe.
1263 * ret_scnprintf: syscall args may set this to a different syscall return
1264 * formatter, for instance, fcntl may return fds, file flags, etc.
1265 */
1266 struct thread_trace {
1267 u64 entry_time;
1268 bool entry_pending;
1269 unsigned long nr_events;
1270 unsigned long pfmaj, pfmin;
1271 char *entry_str;
1272 double runtime_ms;
1273 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1274 struct {
1275 unsigned long ptr;
1276 short int entry_str_pos;
1277 bool pending_open;
1278 unsigned int namelen;
1279 char *name;
1280 } filename;
1281 struct {
1282 int max;
1283 struct file *table;
1284 } files;
1285
1286 struct intlist *syscall_stats;
1287 };
1288
thread_trace__new(void)1289 static struct thread_trace *thread_trace__new(void)
1290 {
1291 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
1292
1293 if (ttrace) {
1294 ttrace->files.max = -1;
1295 ttrace->syscall_stats = intlist__new(NULL);
1296 }
1297
1298 return ttrace;
1299 }
1300
thread__trace(struct thread * thread,FILE * fp)1301 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1302 {
1303 struct thread_trace *ttrace;
1304
1305 if (thread == NULL)
1306 goto fail;
1307
1308 if (thread__priv(thread) == NULL)
1309 thread__set_priv(thread, thread_trace__new());
1310
1311 if (thread__priv(thread) == NULL)
1312 goto fail;
1313
1314 ttrace = thread__priv(thread);
1315 ++ttrace->nr_events;
1316
1317 return ttrace;
1318 fail:
1319 color_fprintf(fp, PERF_COLOR_RED,
1320 "WARNING: not enough memory, dropping samples!\n");
1321 return NULL;
1322 }
1323
1324
syscall_arg__set_ret_scnprintf(struct syscall_arg * arg,size_t (* ret_scnprintf)(char * bf,size_t size,struct syscall_arg * arg))1325 void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
1326 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
1327 {
1328 struct thread_trace *ttrace = thread__priv(arg->thread);
1329
1330 ttrace->ret_scnprintf = ret_scnprintf;
1331 }
1332
1333 #define TRACE_PFMAJ (1 << 0)
1334 #define TRACE_PFMIN (1 << 1)
1335
1336 static const size_t trace__entry_str_size = 2048;
1337
thread_trace__files_entry(struct thread_trace * ttrace,int fd)1338 static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
1339 {
1340 if (fd < 0)
1341 return NULL;
1342
1343 if (fd > ttrace->files.max) {
1344 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
1345
1346 if (nfiles == NULL)
1347 return NULL;
1348
1349 if (ttrace->files.max != -1) {
1350 memset(nfiles + ttrace->files.max + 1, 0,
1351 (fd - ttrace->files.max) * sizeof(struct file));
1352 } else {
1353 memset(nfiles, 0, (fd + 1) * sizeof(struct file));
1354 }
1355
1356 ttrace->files.table = nfiles;
1357 ttrace->files.max = fd;
1358 }
1359
1360 return ttrace->files.table + fd;
1361 }
1362
thread__files_entry(struct thread * thread,int fd)1363 struct file *thread__files_entry(struct thread *thread, int fd)
1364 {
1365 return thread_trace__files_entry(thread__priv(thread), fd);
1366 }
1367
trace__set_fd_pathname(struct thread * thread,int fd,const char * pathname)1368 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1369 {
1370 struct thread_trace *ttrace = thread__priv(thread);
1371 struct file *file = thread_trace__files_entry(ttrace, fd);
1372
1373 if (file != NULL) {
1374 struct stat st;
1375 if (stat(pathname, &st) == 0)
1376 file->dev_maj = major(st.st_rdev);
1377 file->pathname = strdup(pathname);
1378 if (file->pathname)
1379 return 0;
1380 }
1381
1382 return -1;
1383 }
1384
thread__read_fd_path(struct thread * thread,int fd)1385 static int thread__read_fd_path(struct thread *thread, int fd)
1386 {
1387 char linkname[PATH_MAX], pathname[PATH_MAX];
1388 struct stat st;
1389 int ret;
1390
1391 if (thread->pid_ == thread->tid) {
1392 scnprintf(linkname, sizeof(linkname),
1393 "/proc/%d/fd/%d", thread->pid_, fd);
1394 } else {
1395 scnprintf(linkname, sizeof(linkname),
1396 "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1397 }
1398
1399 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1400 return -1;
1401
1402 ret = readlink(linkname, pathname, sizeof(pathname));
1403
1404 if (ret < 0 || ret > st.st_size)
1405 return -1;
1406
1407 pathname[ret] = '\0';
1408 return trace__set_fd_pathname(thread, fd, pathname);
1409 }
1410
thread__fd_path(struct thread * thread,int fd,struct trace * trace)1411 static const char *thread__fd_path(struct thread *thread, int fd,
1412 struct trace *trace)
1413 {
1414 struct thread_trace *ttrace = thread__priv(thread);
1415
1416 if (ttrace == NULL || trace->fd_path_disabled)
1417 return NULL;
1418
1419 if (fd < 0)
1420 return NULL;
1421
1422 if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
1423 if (!trace->live)
1424 return NULL;
1425 ++trace->stats.proc_getname;
1426 if (thread__read_fd_path(thread, fd))
1427 return NULL;
1428 }
1429
1430 return ttrace->files.table[fd].pathname;
1431 }
1432
syscall_arg__scnprintf_fd(char * bf,size_t size,struct syscall_arg * arg)1433 size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
1434 {
1435 int fd = arg->val;
1436 size_t printed = scnprintf(bf, size, "%d", fd);
1437 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1438
1439 if (path)
1440 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1441
1442 return printed;
1443 }
1444
pid__scnprintf_fd(struct trace * trace,pid_t pid,int fd,char * bf,size_t size)1445 size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1446 {
1447 size_t printed = scnprintf(bf, size, "%d", fd);
1448 struct thread *thread = machine__find_thread(trace->host, pid, pid);
1449
1450 if (thread) {
1451 const char *path = thread__fd_path(thread, fd, trace);
1452
1453 if (path)
1454 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1455
1456 thread__put(thread);
1457 }
1458
1459 return printed;
1460 }
1461
syscall_arg__scnprintf_close_fd(char * bf,size_t size,struct syscall_arg * arg)1462 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1463 struct syscall_arg *arg)
1464 {
1465 int fd = arg->val;
1466 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1467 struct thread_trace *ttrace = thread__priv(arg->thread);
1468
1469 if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1470 zfree(&ttrace->files.table[fd].pathname);
1471
1472 return printed;
1473 }
1474
thread__set_filename_pos(struct thread * thread,const char * bf,unsigned long ptr)1475 static void thread__set_filename_pos(struct thread *thread, const char *bf,
1476 unsigned long ptr)
1477 {
1478 struct thread_trace *ttrace = thread__priv(thread);
1479
1480 ttrace->filename.ptr = ptr;
1481 ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1482 }
1483
syscall_arg__scnprintf_augmented_string(struct syscall_arg * arg,char * bf,size_t size)1484 static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1485 {
1486 struct augmented_arg *augmented_arg = arg->augmented.args;
1487 size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1488 /*
1489 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1490 * we would have two strings, each prefixed by its size.
1491 */
1492 int consumed = sizeof(*augmented_arg) + augmented_arg->size;
1493
1494 arg->augmented.args = ((void *)arg->augmented.args) + consumed;
1495 arg->augmented.size -= consumed;
1496
1497 return printed;
1498 }
1499
syscall_arg__scnprintf_filename(char * bf,size_t size,struct syscall_arg * arg)1500 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1501 struct syscall_arg *arg)
1502 {
1503 unsigned long ptr = arg->val;
1504
1505 if (arg->augmented.args)
1506 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1507
1508 if (!arg->trace->vfs_getname)
1509 return scnprintf(bf, size, "%#x", ptr);
1510
1511 thread__set_filename_pos(arg->thread, bf, ptr);
1512 return 0;
1513 }
1514
trace__filter_duration(struct trace * trace,double t)1515 static bool trace__filter_duration(struct trace *trace, double t)
1516 {
1517 return t < (trace->duration_filter * NSEC_PER_MSEC);
1518 }
1519
__trace__fprintf_tstamp(struct trace * trace,u64 tstamp,FILE * fp)1520 static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1521 {
1522 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1523
1524 return fprintf(fp, "%10.3f ", ts);
1525 }
1526
1527 /*
1528 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1529 * using ttrace->entry_time for a thread that receives a sys_exit without
1530 * first having received a sys_enter ("poll" issued before tracing session
1531 * starts, lost sys_enter exit due to ring buffer overflow).
1532 */
trace__fprintf_tstamp(struct trace * trace,u64 tstamp,FILE * fp)1533 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1534 {
1535 if (tstamp > 0)
1536 return __trace__fprintf_tstamp(trace, tstamp, fp);
1537
1538 return fprintf(fp, " ? ");
1539 }
1540
1541 static pid_t workload_pid = -1;
1542 static bool done = false;
1543 static bool interrupted = false;
1544
sighandler_interrupt(int sig __maybe_unused)1545 static void sighandler_interrupt(int sig __maybe_unused)
1546 {
1547 done = interrupted = true;
1548 }
1549
sighandler_chld(int sig __maybe_unused,siginfo_t * info,void * context __maybe_unused)1550 static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
1551 void *context __maybe_unused)
1552 {
1553 if (info->si_pid == workload_pid)
1554 done = true;
1555 }
1556
trace__fprintf_comm_tid(struct trace * trace,struct thread * thread,FILE * fp)1557 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1558 {
1559 size_t printed = 0;
1560
1561 if (trace->multiple_threads) {
1562 if (trace->show_comm)
1563 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1564 printed += fprintf(fp, "%d ", thread->tid);
1565 }
1566
1567 return printed;
1568 }
1569
trace__fprintf_entry_head(struct trace * trace,struct thread * thread,u64 duration,bool duration_calculated,u64 tstamp,FILE * fp)1570 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1571 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1572 {
1573 size_t printed = 0;
1574
1575 if (trace->show_tstamp)
1576 printed = trace__fprintf_tstamp(trace, tstamp, fp);
1577 if (trace->show_duration)
1578 printed += fprintf_duration(duration, duration_calculated, fp);
1579 return printed + trace__fprintf_comm_tid(trace, thread, fp);
1580 }
1581
trace__process_event(struct trace * trace,struct machine * machine,union perf_event * event,struct perf_sample * sample)1582 static int trace__process_event(struct trace *trace, struct machine *machine,
1583 union perf_event *event, struct perf_sample *sample)
1584 {
1585 int ret = 0;
1586
1587 switch (event->header.type) {
1588 case PERF_RECORD_LOST:
1589 color_fprintf(trace->output, PERF_COLOR_RED,
1590 "LOST %" PRIu64 " events!\n", event->lost.lost);
1591 ret = machine__process_lost_event(machine, event, sample);
1592 break;
1593 default:
1594 ret = machine__process_event(machine, event, sample);
1595 break;
1596 }
1597
1598 return ret;
1599 }
1600
trace__tool_process(struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct machine * machine)1601 static int trace__tool_process(struct perf_tool *tool,
1602 union perf_event *event,
1603 struct perf_sample *sample,
1604 struct machine *machine)
1605 {
1606 struct trace *trace = container_of(tool, struct trace, tool);
1607 return trace__process_event(trace, machine, event, sample);
1608 }
1609
trace__machine__resolve_kernel_addr(void * vmachine,unsigned long long * addrp,char ** modp)1610 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1611 {
1612 struct machine *machine = vmachine;
1613
1614 if (machine->kptr_restrict_warned)
1615 return NULL;
1616
1617 if (symbol_conf.kptr_restrict) {
1618 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1619 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
1620 "Kernel samples will not be resolved.\n");
1621 machine->kptr_restrict_warned = true;
1622 return NULL;
1623 }
1624
1625 return machine__resolve_kernel_addr(vmachine, addrp, modp);
1626 }
1627
trace__symbols_init(struct trace * trace,struct evlist * evlist)1628 static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
1629 {
1630 int err = symbol__init(NULL);
1631
1632 if (err)
1633 return err;
1634
1635 trace->host = machine__new_host();
1636 if (trace->host == NULL)
1637 return -ENOMEM;
1638
1639 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1640 if (err < 0)
1641 goto out;
1642
1643 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1644 evlist->core.threads, trace__tool_process,
1645 true, false, 1);
1646 out:
1647 if (err)
1648 symbol__exit();
1649
1650 return err;
1651 }
1652
trace__symbols__exit(struct trace * trace)1653 static void trace__symbols__exit(struct trace *trace)
1654 {
1655 machine__exit(trace->host);
1656 trace->host = NULL;
1657
1658 symbol__exit();
1659 }
1660
syscall__alloc_arg_fmts(struct syscall * sc,int nr_args)1661 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1662 {
1663 int idx;
1664
1665 if (nr_args == RAW_SYSCALL_ARGS_NUM && sc->fmt && sc->fmt->nr_args != 0)
1666 nr_args = sc->fmt->nr_args;
1667
1668 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1669 if (sc->arg_fmt == NULL)
1670 return -1;
1671
1672 for (idx = 0; idx < nr_args; ++idx) {
1673 if (sc->fmt)
1674 sc->arg_fmt[idx] = sc->fmt->arg[idx];
1675 }
1676
1677 sc->nr_args = nr_args;
1678 return 0;
1679 }
1680
1681 static struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {
1682 { .name = "msr", .scnprintf = SCA_X86_MSR, .strtoul = STUL_X86_MSR, },
1683 { .name = "vector", .scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, },
1684 };
1685
syscall_arg_fmt__cmp(const void * name,const void * fmtp)1686 static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)
1687 {
1688 const struct syscall_arg_fmt *fmt = fmtp;
1689 return strcmp(name, fmt->name);
1690 }
1691
1692 static struct syscall_arg_fmt *
__syscall_arg_fmt__find_by_name(struct syscall_arg_fmt * fmts,const int nmemb,const char * name)1693 __syscall_arg_fmt__find_by_name(struct syscall_arg_fmt *fmts, const int nmemb, const char *name)
1694 {
1695 return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);
1696 }
1697
syscall_arg_fmt__find_by_name(const char * name)1698 static struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)
1699 {
1700 const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);
1701 return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);
1702 }
1703
1704 static struct tep_format_field *
syscall_arg_fmt__init_array(struct syscall_arg_fmt * arg,struct tep_format_field * field)1705 syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field)
1706 {
1707 struct tep_format_field *last_field = NULL;
1708 int len;
1709
1710 for (; field; field = field->next, ++arg) {
1711 last_field = field;
1712
1713 if (arg->scnprintf)
1714 continue;
1715
1716 len = strlen(field->name);
1717
1718 if (strcmp(field->type, "const char *") == 0 &&
1719 ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1720 strstr(field->name, "path") != NULL))
1721 arg->scnprintf = SCA_FILENAME;
1722 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
1723 arg->scnprintf = SCA_PTR;
1724 else if (strcmp(field->type, "pid_t") == 0)
1725 arg->scnprintf = SCA_PID;
1726 else if (strcmp(field->type, "umode_t") == 0)
1727 arg->scnprintf = SCA_MODE_T;
1728 else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) {
1729 arg->scnprintf = SCA_CHAR_ARRAY;
1730 arg->nr_entries = field->arraylen;
1731 } else if ((strcmp(field->type, "int") == 0 ||
1732 strcmp(field->type, "unsigned int") == 0 ||
1733 strcmp(field->type, "long") == 0) &&
1734 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
1735 /*
1736 * /sys/kernel/tracing/events/syscalls/sys_enter*
1737 * egrep 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
1738 * 65 int
1739 * 23 unsigned int
1740 * 7 unsigned long
1741 */
1742 arg->scnprintf = SCA_FD;
1743 } else {
1744 struct syscall_arg_fmt *fmt = syscall_arg_fmt__find_by_name(field->name);
1745
1746 if (fmt) {
1747 arg->scnprintf = fmt->scnprintf;
1748 arg->strtoul = fmt->strtoul;
1749 }
1750 }
1751 }
1752
1753 return last_field;
1754 }
1755
syscall__set_arg_fmts(struct syscall * sc)1756 static int syscall__set_arg_fmts(struct syscall *sc)
1757 {
1758 struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args);
1759
1760 if (last_field)
1761 sc->args_size = last_field->offset + last_field->size;
1762
1763 return 0;
1764 }
1765
trace__read_syscall_info(struct trace * trace,int id)1766 static int trace__read_syscall_info(struct trace *trace, int id)
1767 {
1768 char tp_name[128];
1769 struct syscall *sc;
1770 const char *name = syscalltbl__name(trace->sctbl, id);
1771
1772 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
1773 if (trace->syscalls.table == NULL) {
1774 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
1775 if (trace->syscalls.table == NULL)
1776 return -ENOMEM;
1777 }
1778 #else
1779 if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) {
1780 // When using libaudit we don't know beforehand what is the max syscall id
1781 struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1782
1783 if (table == NULL)
1784 return -ENOMEM;
1785
1786 // Need to memset from offset 0 and +1 members if brand new
1787 if (trace->syscalls.table == NULL)
1788 memset(table, 0, (id + 1) * sizeof(*sc));
1789 else
1790 memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc));
1791
1792 trace->syscalls.table = table;
1793 trace->sctbl->syscalls.max_id = id;
1794 }
1795 #endif
1796 sc = trace->syscalls.table + id;
1797 if (sc->nonexistent)
1798 return -EEXIST;
1799
1800 if (name == NULL) {
1801 sc->nonexistent = true;
1802 return -EEXIST;
1803 }
1804
1805 sc->name = name;
1806 sc->fmt = syscall_fmt__find(sc->name);
1807
1808 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1809 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1810
1811 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
1812 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1813 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1814 }
1815
1816 /*
1817 * Fails to read trace point format via sysfs node, so the trace point
1818 * doesn't exist. Set the 'nonexistent' flag as true.
1819 */
1820 if (IS_ERR(sc->tp_format)) {
1821 sc->nonexistent = true;
1822 return PTR_ERR(sc->tp_format);
1823 }
1824
1825 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ?
1826 RAW_SYSCALL_ARGS_NUM : sc->tp_format->format.nr_fields))
1827 return -ENOMEM;
1828
1829 sc->args = sc->tp_format->format.fields;
1830 /*
1831 * We need to check and discard the first variable '__syscall_nr'
1832 * or 'nr' that mean the syscall number. It is needless here.
1833 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1834 */
1835 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
1836 sc->args = sc->args->next;
1837 --sc->nr_args;
1838 }
1839
1840 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1841 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
1842
1843 return syscall__set_arg_fmts(sc);
1844 }
1845
evsel__init_tp_arg_scnprintf(struct evsel * evsel)1846 static int evsel__init_tp_arg_scnprintf(struct evsel *evsel)
1847 {
1848 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
1849
1850 if (fmt != NULL) {
1851 syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields);
1852 return 0;
1853 }
1854
1855 return -ENOMEM;
1856 }
1857
intcmp(const void * a,const void * b)1858 static int intcmp(const void *a, const void *b)
1859 {
1860 const int *one = a, *another = b;
1861
1862 return *one - *another;
1863 }
1864
trace__validate_ev_qualifier(struct trace * trace)1865 static int trace__validate_ev_qualifier(struct trace *trace)
1866 {
1867 int err = 0;
1868 bool printed_invalid_prefix = false;
1869 struct str_node *pos;
1870 size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
1871
1872 trace->ev_qualifier_ids.entries = malloc(nr_allocated *
1873 sizeof(trace->ev_qualifier_ids.entries[0]));
1874
1875 if (trace->ev_qualifier_ids.entries == NULL) {
1876 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1877 trace->output);
1878 err = -EINVAL;
1879 goto out;
1880 }
1881
1882 strlist__for_each_entry(pos, trace->ev_qualifier) {
1883 const char *sc = pos->s;
1884 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
1885
1886 if (id < 0) {
1887 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1888 if (id >= 0)
1889 goto matches;
1890
1891 if (!printed_invalid_prefix) {
1892 pr_debug("Skipping unknown syscalls: ");
1893 printed_invalid_prefix = true;
1894 } else {
1895 pr_debug(", ");
1896 }
1897
1898 pr_debug("%s", sc);
1899 continue;
1900 }
1901 matches:
1902 trace->ev_qualifier_ids.entries[nr_used++] = id;
1903 if (match_next == -1)
1904 continue;
1905
1906 while (1) {
1907 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1908 if (id < 0)
1909 break;
1910 if (nr_allocated == nr_used) {
1911 void *entries;
1912
1913 nr_allocated += 8;
1914 entries = realloc(trace->ev_qualifier_ids.entries,
1915 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1916 if (entries == NULL) {
1917 err = -ENOMEM;
1918 fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1919 goto out_free;
1920 }
1921 trace->ev_qualifier_ids.entries = entries;
1922 }
1923 trace->ev_qualifier_ids.entries[nr_used++] = id;
1924 }
1925 }
1926
1927 trace->ev_qualifier_ids.nr = nr_used;
1928 qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
1929 out:
1930 if (printed_invalid_prefix)
1931 pr_debug("\n");
1932 return err;
1933 out_free:
1934 zfree(&trace->ev_qualifier_ids.entries);
1935 trace->ev_qualifier_ids.nr = 0;
1936 goto out;
1937 }
1938
trace__syscall_enabled(struct trace * trace,int id)1939 static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1940 {
1941 bool in_ev_qualifier;
1942
1943 if (trace->ev_qualifier_ids.nr == 0)
1944 return true;
1945
1946 in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1947 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1948
1949 if (in_ev_qualifier)
1950 return !trace->not_ev_qualifier;
1951
1952 return trace->not_ev_qualifier;
1953 }
1954
1955 /*
1956 * args is to be interpreted as a series of longs but we need to handle
1957 * 8-byte unaligned accesses. args points to raw_data within the event
1958 * and raw_data is guaranteed to be 8-byte unaligned because it is
1959 * preceded by raw_size which is a u32. So we need to copy args to a temp
1960 * variable to read it. Most notably this avoids extended load instructions
1961 * on unaligned addresses
1962 */
syscall_arg__val(struct syscall_arg * arg,u8 idx)1963 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
1964 {
1965 unsigned long val;
1966 unsigned char *p = arg->args + sizeof(unsigned long) * idx;
1967
1968 memcpy(&val, p, sizeof(val));
1969 return val;
1970 }
1971
syscall__scnprintf_name(struct syscall * sc,char * bf,size_t size,struct syscall_arg * arg)1972 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1973 struct syscall_arg *arg)
1974 {
1975 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1976 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1977
1978 return scnprintf(bf, size, "arg%d: ", arg->idx);
1979 }
1980
1981 /*
1982 * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1983 * as mount 'flags' argument that needs ignoring some magic flag, see comment
1984 * in tools/perf/trace/beauty/mount_flags.c
1985 */
syscall_arg_fmt__mask_val(struct syscall_arg_fmt * fmt,struct syscall_arg * arg,unsigned long val)1986 static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)
1987 {
1988 if (fmt && fmt->mask_val)
1989 return fmt->mask_val(arg, val);
1990
1991 return val;
1992 }
1993
syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt * fmt,char * bf,size_t size,struct syscall_arg * arg,unsigned long val)1994 static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,
1995 struct syscall_arg *arg, unsigned long val)
1996 {
1997 if (fmt && fmt->scnprintf) {
1998 arg->val = val;
1999 if (fmt->parm)
2000 arg->parm = fmt->parm;
2001 return fmt->scnprintf(bf, size, arg);
2002 }
2003 return scnprintf(bf, size, "%ld", val);
2004 }
2005
syscall__scnprintf_args(struct syscall * sc,char * bf,size_t size,unsigned char * args,void * augmented_args,int augmented_args_size,struct trace * trace,struct thread * thread)2006 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
2007 unsigned char *args, void *augmented_args, int augmented_args_size,
2008 struct trace *trace, struct thread *thread)
2009 {
2010 size_t printed = 0;
2011 unsigned long val;
2012 u8 bit = 1;
2013 struct syscall_arg arg = {
2014 .args = args,
2015 .augmented = {
2016 .size = augmented_args_size,
2017 .args = augmented_args,
2018 },
2019 .idx = 0,
2020 .mask = 0,
2021 .trace = trace,
2022 .thread = thread,
2023 .show_string_prefix = trace->show_string_prefix,
2024 };
2025 struct thread_trace *ttrace = thread__priv(thread);
2026
2027 /*
2028 * Things like fcntl will set this in its 'cmd' formatter to pick the
2029 * right formatter for the return value (an fd? file flags?), which is
2030 * not needed for syscalls that always return a given type, say an fd.
2031 */
2032 ttrace->ret_scnprintf = NULL;
2033
2034 if (sc->args != NULL) {
2035 struct tep_format_field *field;
2036
2037 for (field = sc->args; field;
2038 field = field->next, ++arg.idx, bit <<= 1) {
2039 if (arg.mask & bit)
2040 continue;
2041
2042 arg.fmt = &sc->arg_fmt[arg.idx];
2043 val = syscall_arg__val(&arg, arg.idx);
2044 /*
2045 * Some syscall args need some mask, most don't and
2046 * return val untouched.
2047 */
2048 val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);
2049
2050 /*
2051 * Suppress this argument if its value is zero and
2052 * and we don't have a string associated in an
2053 * strarray for it.
2054 */
2055 if (val == 0 &&
2056 !trace->show_zeros &&
2057 !(sc->arg_fmt &&
2058 (sc->arg_fmt[arg.idx].show_zero ||
2059 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
2060 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
2061 sc->arg_fmt[arg.idx].parm))
2062 continue;
2063
2064 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2065
2066 if (trace->show_arg_names)
2067 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2068
2069 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],
2070 bf + printed, size - printed, &arg, val);
2071 }
2072 } else if (IS_ERR(sc->tp_format)) {
2073 /*
2074 * If we managed to read the tracepoint /format file, then we
2075 * may end up not having any args, like with gettid(), so only
2076 * print the raw args when we didn't manage to read it.
2077 */
2078 while (arg.idx < sc->nr_args) {
2079 if (arg.mask & bit)
2080 goto next_arg;
2081 val = syscall_arg__val(&arg, arg.idx);
2082 if (printed)
2083 printed += scnprintf(bf + printed, size - printed, ", ");
2084 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
2085 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);
2086 next_arg:
2087 ++arg.idx;
2088 bit <<= 1;
2089 }
2090 }
2091
2092 return printed;
2093 }
2094
2095 typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
2096 union perf_event *event,
2097 struct perf_sample *sample);
2098
trace__syscall_info(struct trace * trace,struct evsel * evsel,int id)2099 static struct syscall *trace__syscall_info(struct trace *trace,
2100 struct evsel *evsel, int id)
2101 {
2102 int err = 0;
2103
2104 if (id < 0) {
2105
2106 /*
2107 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
2108 * before that, leaving at a higher verbosity level till that is
2109 * explained. Reproduced with plain ftrace with:
2110 *
2111 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
2112 * grep "NR -1 " /t/trace_pipe
2113 *
2114 * After generating some load on the machine.
2115 */
2116 if (verbose > 1) {
2117 static u64 n;
2118 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
2119 id, evsel__name(evsel), ++n);
2120 }
2121 return NULL;
2122 }
2123
2124 err = -EINVAL;
2125
2126 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
2127 if (id > trace->sctbl->syscalls.max_id) {
2128 #else
2129 if (id >= trace->sctbl->syscalls.max_id) {
2130 /*
2131 * With libaudit we don't know beforehand what is the max_id,
2132 * so we let trace__read_syscall_info() figure that out as we
2133 * go on reading syscalls.
2134 */
2135 err = trace__read_syscall_info(trace, id);
2136 if (err)
2137 #endif
2138 goto out_cant_read;
2139 }
2140
2141 if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
2142 (err = trace__read_syscall_info(trace, id)) != 0)
2143 goto out_cant_read;
2144
2145 if (trace->syscalls.table && trace->syscalls.table[id].nonexistent)
2146 goto out_cant_read;
2147
2148 return &trace->syscalls.table[id];
2149
2150 out_cant_read:
2151 if (verbose > 0) {
2152 char sbuf[STRERR_BUFSIZE];
2153 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
2154 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
2155 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
2156 fputs(" information\n", trace->output);
2157 }
2158 return NULL;
2159 }
2160
2161 struct syscall_stats {
2162 struct stats stats;
2163 u64 nr_failures;
2164 int max_errno;
2165 u32 *errnos;
2166 };
2167
2168 static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace,
2169 int id, struct perf_sample *sample, long err, bool errno_summary)
2170 {
2171 struct int_node *inode;
2172 struct syscall_stats *stats;
2173 u64 duration = 0;
2174
2175 inode = intlist__findnew(ttrace->syscall_stats, id);
2176 if (inode == NULL)
2177 return;
2178
2179 stats = inode->priv;
2180 if (stats == NULL) {
2181 stats = zalloc(sizeof(*stats));
2182 if (stats == NULL)
2183 return;
2184
2185 init_stats(&stats->stats);
2186 inode->priv = stats;
2187 }
2188
2189 if (ttrace->entry_time && sample->time > ttrace->entry_time)
2190 duration = sample->time - ttrace->entry_time;
2191
2192 update_stats(&stats->stats, duration);
2193
2194 if (err < 0) {
2195 ++stats->nr_failures;
2196
2197 if (!errno_summary)
2198 return;
2199
2200 err = -err;
2201 if (err > stats->max_errno) {
2202 u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32));
2203
2204 if (new_errnos) {
2205 memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32));
2206 } else {
2207 pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n",
2208 thread__comm_str(thread), thread->pid_, thread->tid);
2209 return;
2210 }
2211
2212 stats->errnos = new_errnos;
2213 stats->max_errno = err;
2214 }
2215
2216 ++stats->errnos[err - 1];
2217 }
2218 }
2219
2220 static int trace__printf_interrupted_entry(struct trace *trace)
2221 {
2222 struct thread_trace *ttrace;
2223 size_t printed;
2224 int len;
2225
2226 if (trace->failure_only || trace->current == NULL)
2227 return 0;
2228
2229 ttrace = thread__priv(trace->current);
2230
2231 if (!ttrace->entry_pending)
2232 return 0;
2233
2234 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
2235 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
2236
2237 if (len < trace->args_alignment - 4)
2238 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
2239
2240 printed += fprintf(trace->output, " ...\n");
2241
2242 ttrace->entry_pending = false;
2243 ++trace->nr_events_printed;
2244
2245 return printed;
2246 }
2247
2248 static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
2249 struct perf_sample *sample, struct thread *thread)
2250 {
2251 int printed = 0;
2252
2253 if (trace->print_sample) {
2254 double ts = (double)sample->time / NSEC_PER_MSEC;
2255
2256 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
2257 evsel__name(evsel), ts,
2258 thread__comm_str(thread),
2259 sample->pid, sample->tid, sample->cpu);
2260 }
2261
2262 return printed;
2263 }
2264
2265 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
2266 {
2267 void *augmented_args = NULL;
2268 /*
2269 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
2270 * and there we get all 6 syscall args plus the tracepoint common fields
2271 * that gets calculated at the start and the syscall_nr (another long).
2272 * So we check if that is the case and if so don't look after the
2273 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
2274 * which is fixed.
2275 *
2276 * We'll revisit this later to pass s->args_size to the BPF augmenter
2277 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
2278 * copies only what we need for each syscall, like what happens when we
2279 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
2280 * traffic to just what is needed for each syscall.
2281 */
2282 int args_size = raw_augmented_args_size ?: sc->args_size;
2283
2284 *augmented_args_size = sample->raw_size - args_size;
2285 if (*augmented_args_size > 0)
2286 augmented_args = sample->raw_data + args_size;
2287
2288 return augmented_args;
2289 }
2290
2291 static void syscall__exit(struct syscall *sc)
2292 {
2293 if (!sc)
2294 return;
2295
2296 free(sc->arg_fmt);
2297 }
2298
2299 static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
2300 union perf_event *event __maybe_unused,
2301 struct perf_sample *sample)
2302 {
2303 char *msg;
2304 void *args;
2305 int printed = 0;
2306 struct thread *thread;
2307 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2308 int augmented_args_size = 0;
2309 void *augmented_args = NULL;
2310 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2311 struct thread_trace *ttrace;
2312
2313 if (sc == NULL)
2314 return -1;
2315
2316 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2317 ttrace = thread__trace(thread, trace->output);
2318 if (ttrace == NULL)
2319 goto out_put;
2320
2321 trace__fprintf_sample(trace, evsel, sample, thread);
2322
2323 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2324
2325 if (ttrace->entry_str == NULL) {
2326 ttrace->entry_str = malloc(trace__entry_str_size);
2327 if (!ttrace->entry_str)
2328 goto out_put;
2329 }
2330
2331 if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
2332 trace__printf_interrupted_entry(trace);
2333 /*
2334 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2335 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2336 * this breaks syscall__augmented_args() check for augmented args, as we calculate
2337 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2338 * so when handling, say the openat syscall, we end up getting 6 args for the
2339 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2340 * thinking that the extra 2 u64 args are the augmented filename, so just check
2341 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2342 */
2343 if (evsel != trace->syscalls.events.sys_enter)
2344 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2345 ttrace->entry_time = sample->time;
2346 msg = ttrace->entry_str;
2347 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
2348
2349 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
2350 args, augmented_args, augmented_args_size, trace, thread);
2351
2352 if (sc->is_exit) {
2353 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
2354 int alignment = 0;
2355
2356 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
2357 printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2358 if (trace->args_alignment > printed)
2359 alignment = trace->args_alignment - printed;
2360 fprintf(trace->output, "%*s= ?\n", alignment, " ");
2361 }
2362 } else {
2363 ttrace->entry_pending = true;
2364 /* See trace__vfs_getname & trace__sys_exit */
2365 ttrace->filename.pending_open = false;
2366 }
2367
2368 if (trace->current != thread) {
2369 thread__put(trace->current);
2370 trace->current = thread__get(thread);
2371 }
2372 err = 0;
2373 out_put:
2374 thread__put(thread);
2375 return err;
2376 }
2377
2378 static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
2379 struct perf_sample *sample)
2380 {
2381 struct thread_trace *ttrace;
2382 struct thread *thread;
2383 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2384 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2385 char msg[1024];
2386 void *args, *augmented_args = NULL;
2387 int augmented_args_size;
2388
2389 if (sc == NULL)
2390 return -1;
2391
2392 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2393 ttrace = thread__trace(thread, trace->output);
2394 /*
2395 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2396 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2397 */
2398 if (ttrace == NULL)
2399 goto out_put;
2400
2401 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2402 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2403 syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
2404 fprintf(trace->output, "%s", msg);
2405 err = 0;
2406 out_put:
2407 thread__put(thread);
2408 return err;
2409 }
2410
2411 static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
2412 struct perf_sample *sample,
2413 struct callchain_cursor *cursor)
2414 {
2415 struct addr_location al;
2416 int max_stack = evsel->core.attr.sample_max_stack ?
2417 evsel->core.attr.sample_max_stack :
2418 trace->max_stack;
2419 int err;
2420
2421 if (machine__resolve(trace->host, &al, sample) < 0)
2422 return -1;
2423
2424 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
2425 addr_location__put(&al);
2426 return err;
2427 }
2428
2429 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2430 {
2431 /* TODO: user-configurable print_opts */
2432 const unsigned int print_opts = EVSEL__PRINT_SYM |
2433 EVSEL__PRINT_DSO |
2434 EVSEL__PRINT_UNKNOWN_AS_ADDR;
2435
2436 return sample__fprintf_callchain(sample, 38, print_opts, &callchain_cursor, symbol_conf.bt_stop_list, trace->output);
2437 }
2438
2439 static const char *errno_to_name(struct evsel *evsel, int err)
2440 {
2441 struct perf_env *env = evsel__env(evsel);
2442 const char *arch_name = perf_env__arch(env);
2443
2444 return arch_syscalls__strerrno(arch_name, err);
2445 }
2446
2447 static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
2448 union perf_event *event __maybe_unused,
2449 struct perf_sample *sample)
2450 {
2451 long ret;
2452 u64 duration = 0;
2453 bool duration_calculated = false;
2454 struct thread *thread;
2455 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2456 int alignment = trace->args_alignment;
2457 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2458 struct thread_trace *ttrace;
2459
2460 if (sc == NULL)
2461 return -1;
2462
2463 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2464 ttrace = thread__trace(thread, trace->output);
2465 if (ttrace == NULL)
2466 goto out_put;
2467
2468 trace__fprintf_sample(trace, evsel, sample, thread);
2469
2470 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2471
2472 if (trace->summary)
2473 thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary);
2474
2475 if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2476 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2477 ttrace->filename.pending_open = false;
2478 ++trace->stats.vfs_getname;
2479 }
2480
2481 if (ttrace->entry_time) {
2482 duration = sample->time - ttrace->entry_time;
2483 if (trace__filter_duration(trace, duration))
2484 goto out;
2485 duration_calculated = true;
2486 } else if (trace->duration_filter)
2487 goto out;
2488
2489 if (sample->callchain) {
2490 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2491 if (callchain_ret == 0) {
2492 if (callchain_cursor.nr < trace->min_stack)
2493 goto out;
2494 callchain_ret = 1;
2495 }
2496 }
2497
2498 if (trace->summary_only || (ret >= 0 && trace->failure_only))
2499 goto out;
2500
2501 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2502
2503 if (ttrace->entry_pending) {
2504 printed = fprintf(trace->output, "%s", ttrace->entry_str);
2505 } else {
2506 printed += fprintf(trace->output, " ... [");
2507 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2508 printed += 9;
2509 printed += fprintf(trace->output, "]: %s()", sc->name);
2510 }
2511
2512 printed++; /* the closing ')' */
2513
2514 if (alignment > printed)
2515 alignment -= printed;
2516 else
2517 alignment = 0;
2518
2519 fprintf(trace->output, ")%*s= ", alignment, " ");
2520
2521 if (sc->fmt == NULL) {
2522 if (ret < 0)
2523 goto errno_print;
2524 signed_print:
2525 fprintf(trace->output, "%ld", ret);
2526 } else if (ret < 0) {
2527 errno_print: {
2528 char bf[STRERR_BUFSIZE];
2529 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2530 *e = errno_to_name(evsel, -ret);
2531
2532 fprintf(trace->output, "-1 %s (%s)", e, emsg);
2533 }
2534 } else if (ret == 0 && sc->fmt->timeout)
2535 fprintf(trace->output, "0 (Timeout)");
2536 else if (ttrace->ret_scnprintf) {
2537 char bf[1024];
2538 struct syscall_arg arg = {
2539 .val = ret,
2540 .thread = thread,
2541 .trace = trace,
2542 };
2543 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2544 ttrace->ret_scnprintf = NULL;
2545 fprintf(trace->output, "%s", bf);
2546 } else if (sc->fmt->hexret)
2547 fprintf(trace->output, "%#lx", ret);
2548 else if (sc->fmt->errpid) {
2549 struct thread *child = machine__find_thread(trace->host, ret, ret);
2550
2551 if (child != NULL) {
2552 fprintf(trace->output, "%ld", ret);
2553 if (child->comm_set)
2554 fprintf(trace->output, " (%s)", thread__comm_str(child));
2555 thread__put(child);
2556 }
2557 } else
2558 goto signed_print;
2559
2560 fputc('\n', trace->output);
2561
2562 /*
2563 * We only consider an 'event' for the sake of --max-events a non-filtered
2564 * sys_enter + sys_exit and other tracepoint events.
2565 */
2566 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2567 interrupted = true;
2568
2569 if (callchain_ret > 0)
2570 trace__fprintf_callchain(trace, sample);
2571 else if (callchain_ret < 0)
2572 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2573 out:
2574 ttrace->entry_pending = false;
2575 err = 0;
2576 out_put:
2577 thread__put(thread);
2578 return err;
2579 }
2580
2581 static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
2582 union perf_event *event __maybe_unused,
2583 struct perf_sample *sample)
2584 {
2585 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2586 struct thread_trace *ttrace;
2587 size_t filename_len, entry_str_len, to_move;
2588 ssize_t remaining_space;
2589 char *pos;
2590 const char *filename = evsel__rawptr(evsel, sample, "pathname");
2591
2592 if (!thread)
2593 goto out;
2594
2595 ttrace = thread__priv(thread);
2596 if (!ttrace)
2597 goto out_put;
2598
2599 filename_len = strlen(filename);
2600 if (filename_len == 0)
2601 goto out_put;
2602
2603 if (ttrace->filename.namelen < filename_len) {
2604 char *f = realloc(ttrace->filename.name, filename_len + 1);
2605
2606 if (f == NULL)
2607 goto out_put;
2608
2609 ttrace->filename.namelen = filename_len;
2610 ttrace->filename.name = f;
2611 }
2612
2613 strcpy(ttrace->filename.name, filename);
2614 ttrace->filename.pending_open = true;
2615
2616 if (!ttrace->filename.ptr)
2617 goto out_put;
2618
2619 entry_str_len = strlen(ttrace->entry_str);
2620 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2621 if (remaining_space <= 0)
2622 goto out_put;
2623
2624 if (filename_len > (size_t)remaining_space) {
2625 filename += filename_len - remaining_space;
2626 filename_len = remaining_space;
2627 }
2628
2629 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2630 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2631 memmove(pos + filename_len, pos, to_move);
2632 memcpy(pos, filename, filename_len);
2633
2634 ttrace->filename.ptr = 0;
2635 ttrace->filename.entry_str_pos = 0;
2636 out_put:
2637 thread__put(thread);
2638 out:
2639 return 0;
2640 }
2641
2642 static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
2643 union perf_event *event __maybe_unused,
2644 struct perf_sample *sample)
2645 {
2646 u64 runtime = evsel__intval(evsel, sample, "runtime");
2647 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2648 struct thread *thread = machine__findnew_thread(trace->host,
2649 sample->pid,
2650 sample->tid);
2651 struct thread_trace *ttrace = thread__trace(thread, trace->output);
2652
2653 if (ttrace == NULL)
2654 goto out_dump;
2655
2656 ttrace->runtime_ms += runtime_ms;
2657 trace->runtime_ms += runtime_ms;
2658 out_put:
2659 thread__put(thread);
2660 return 0;
2661
2662 out_dump:
2663 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2664 evsel->name,
2665 evsel__strval(evsel, sample, "comm"),
2666 (pid_t)evsel__intval(evsel, sample, "pid"),
2667 runtime,
2668 evsel__intval(evsel, sample, "vruntime"));
2669 goto out_put;
2670 }
2671
2672 static int bpf_output__printer(enum binary_printer_ops op,
2673 unsigned int val, void *extra __maybe_unused, FILE *fp)
2674 {
2675 unsigned char ch = (unsigned char)val;
2676
2677 switch (op) {
2678 case BINARY_PRINT_CHAR_DATA:
2679 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
2680 case BINARY_PRINT_DATA_BEGIN:
2681 case BINARY_PRINT_LINE_BEGIN:
2682 case BINARY_PRINT_ADDR:
2683 case BINARY_PRINT_NUM_DATA:
2684 case BINARY_PRINT_NUM_PAD:
2685 case BINARY_PRINT_SEP:
2686 case BINARY_PRINT_CHAR_PAD:
2687 case BINARY_PRINT_LINE_END:
2688 case BINARY_PRINT_DATA_END:
2689 default:
2690 break;
2691 }
2692
2693 return 0;
2694 }
2695
2696 static void bpf_output__fprintf(struct trace *trace,
2697 struct perf_sample *sample)
2698 {
2699 binary__fprintf(sample->raw_data, sample->raw_size, 8,
2700 bpf_output__printer, NULL, trace->output);
2701 ++trace->nr_events_printed;
2702 }
2703
2704 static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,
2705 struct thread *thread, void *augmented_args, int augmented_args_size)
2706 {
2707 char bf[2048];
2708 size_t size = sizeof(bf);
2709 struct tep_format_field *field = evsel->tp_format->format.fields;
2710 struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel);
2711 size_t printed = 0;
2712 unsigned long val;
2713 u8 bit = 1;
2714 struct syscall_arg syscall_arg = {
2715 .augmented = {
2716 .size = augmented_args_size,
2717 .args = augmented_args,
2718 },
2719 .idx = 0,
2720 .mask = 0,
2721 .trace = trace,
2722 .thread = thread,
2723 .show_string_prefix = trace->show_string_prefix,
2724 };
2725
2726 for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {
2727 if (syscall_arg.mask & bit)
2728 continue;
2729
2730 syscall_arg.len = 0;
2731 syscall_arg.fmt = arg;
2732 if (field->flags & TEP_FIELD_IS_ARRAY) {
2733 int offset = field->offset;
2734
2735 if (field->flags & TEP_FIELD_IS_DYNAMIC) {
2736 offset = format_field__intval(field, sample, evsel->needs_swap);
2737 syscall_arg.len = offset >> 16;
2738 offset &= 0xffff;
2739 if (field->flags & TEP_FIELD_IS_RELATIVE)
2740 offset += field->offset + field->size;
2741 }
2742
2743 val = (uintptr_t)(sample->raw_data + offset);
2744 } else
2745 val = format_field__intval(field, sample, evsel->needs_swap);
2746 /*
2747 * Some syscall args need some mask, most don't and
2748 * return val untouched.
2749 */
2750 val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);
2751
2752 /*
2753 * Suppress this argument if its value is zero and
2754 * we don't have a string associated in an
2755 * strarray for it.
2756 */
2757 if (val == 0 &&
2758 !trace->show_zeros &&
2759 !((arg->show_zero ||
2760 arg->scnprintf == SCA_STRARRAY ||
2761 arg->scnprintf == SCA_STRARRAYS) &&
2762 arg->parm))
2763 continue;
2764
2765 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2766
2767 if (trace->show_arg_names)
2768 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2769
2770 printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);
2771 }
2772
2773 return printed + fprintf(trace->output, "%s", bf);
2774 }
2775
2776 static int trace__event_handler(struct trace *trace, struct evsel *evsel,
2777 union perf_event *event __maybe_unused,
2778 struct perf_sample *sample)
2779 {
2780 struct thread *thread;
2781 int callchain_ret = 0;
2782 /*
2783 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2784 * this event's max_events having been hit and this is an entry coming
2785 * from the ring buffer that we should discard, since the max events
2786 * have already been considered/printed.
2787 */
2788 if (evsel->disabled)
2789 return 0;
2790
2791 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2792
2793 if (sample->callchain) {
2794 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2795 if (callchain_ret == 0) {
2796 if (callchain_cursor.nr < trace->min_stack)
2797 goto out;
2798 callchain_ret = 1;
2799 }
2800 }
2801
2802 trace__printf_interrupted_entry(trace);
2803 trace__fprintf_tstamp(trace, sample->time, trace->output);
2804
2805 if (trace->trace_syscalls && trace->show_duration)
2806 fprintf(trace->output, "( ): ");
2807
2808 if (thread)
2809 trace__fprintf_comm_tid(trace, thread, trace->output);
2810
2811 if (evsel == trace->syscalls.events.augmented) {
2812 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2813 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2814
2815 if (sc) {
2816 fprintf(trace->output, "%s(", sc->name);
2817 trace__fprintf_sys_enter(trace, evsel, sample);
2818 fputc(')', trace->output);
2819 goto newline;
2820 }
2821
2822 /*
2823 * XXX: Not having the associated syscall info or not finding/adding
2824 * the thread should never happen, but if it does...
2825 * fall thru and print it as a bpf_output event.
2826 */
2827 }
2828
2829 fprintf(trace->output, "%s(", evsel->name);
2830
2831 if (evsel__is_bpf_output(evsel)) {
2832 bpf_output__fprintf(trace, sample);
2833 } else if (evsel->tp_format) {
2834 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2835 trace__fprintf_sys_enter(trace, evsel, sample)) {
2836 if (trace->libtraceevent_print) {
2837 event_format__fprintf(evsel->tp_format, sample->cpu,
2838 sample->raw_data, sample->raw_size,
2839 trace->output);
2840 } else {
2841 trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);
2842 }
2843 }
2844 }
2845
2846 newline:
2847 fprintf(trace->output, ")\n");
2848
2849 if (callchain_ret > 0)
2850 trace__fprintf_callchain(trace, sample);
2851 else if (callchain_ret < 0)
2852 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2853
2854 ++trace->nr_events_printed;
2855
2856 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2857 evsel__disable(evsel);
2858 evsel__close(evsel);
2859 }
2860 out:
2861 thread__put(thread);
2862 return 0;
2863 }
2864
2865 static void print_location(FILE *f, struct perf_sample *sample,
2866 struct addr_location *al,
2867 bool print_dso, bool print_sym)
2868 {
2869
2870 if ((verbose > 0 || print_dso) && al->map)
2871 fprintf(f, "%s@", al->map->dso->long_name);
2872
2873 if ((verbose > 0 || print_sym) && al->sym)
2874 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
2875 al->addr - al->sym->start);
2876 else if (al->map)
2877 fprintf(f, "0x%" PRIx64, al->addr);
2878 else
2879 fprintf(f, "0x%" PRIx64, sample->addr);
2880 }
2881
2882 static int trace__pgfault(struct trace *trace,
2883 struct evsel *evsel,
2884 union perf_event *event __maybe_unused,
2885 struct perf_sample *sample)
2886 {
2887 struct thread *thread;
2888 struct addr_location al;
2889 char map_type = 'd';
2890 struct thread_trace *ttrace;
2891 int err = -1;
2892 int callchain_ret = 0;
2893
2894 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2895
2896 if (sample->callchain) {
2897 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2898 if (callchain_ret == 0) {
2899 if (callchain_cursor.nr < trace->min_stack)
2900 goto out_put;
2901 callchain_ret = 1;
2902 }
2903 }
2904
2905 ttrace = thread__trace(thread, trace->output);
2906 if (ttrace == NULL)
2907 goto out_put;
2908
2909 if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2910 ttrace->pfmaj++;
2911 else
2912 ttrace->pfmin++;
2913
2914 if (trace->summary_only)
2915 goto out;
2916
2917 thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
2918
2919 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
2920
2921 fprintf(trace->output, "%sfault [",
2922 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2923 "maj" : "min");
2924
2925 print_location(trace->output, sample, &al, false, true);
2926
2927 fprintf(trace->output, "] => ");
2928
2929 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2930
2931 if (!al.map) {
2932 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2933
2934 if (al.map)
2935 map_type = 'x';
2936 else
2937 map_type = '?';
2938 }
2939
2940 print_location(trace->output, sample, &al, true, false);
2941
2942 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
2943
2944 if (callchain_ret > 0)
2945 trace__fprintf_callchain(trace, sample);
2946 else if (callchain_ret < 0)
2947 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2948
2949 ++trace->nr_events_printed;
2950 out:
2951 err = 0;
2952 out_put:
2953 thread__put(thread);
2954 return err;
2955 }
2956
2957 static void trace__set_base_time(struct trace *trace,
2958 struct evsel *evsel,
2959 struct perf_sample *sample)
2960 {
2961 /*
2962 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2963 * and don't use sample->time unconditionally, we may end up having
2964 * some other event in the future without PERF_SAMPLE_TIME for good
2965 * reason, i.e. we may not be interested in its timestamps, just in
2966 * it taking place, picking some piece of information when it
2967 * appears in our event stream (vfs_getname comes to mind).
2968 */
2969 if (trace->base_time == 0 && !trace->full_time &&
2970 (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
2971 trace->base_time = sample->time;
2972 }
2973
2974 static int trace__process_sample(struct perf_tool *tool,
2975 union perf_event *event,
2976 struct perf_sample *sample,
2977 struct evsel *evsel,
2978 struct machine *machine __maybe_unused)
2979 {
2980 struct trace *trace = container_of(tool, struct trace, tool);
2981 struct thread *thread;
2982 int err = 0;
2983
2984 tracepoint_handler handler = evsel->handler;
2985
2986 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2987 if (thread && thread__is_filtered(thread))
2988 goto out;
2989
2990 trace__set_base_time(trace, evsel, sample);
2991
2992 if (handler) {
2993 ++trace->nr_events;
2994 handler(trace, evsel, event, sample);
2995 }
2996 out:
2997 thread__put(thread);
2998 return err;
2999 }
3000
3001 static int trace__record(struct trace *trace, int argc, const char **argv)
3002 {
3003 unsigned int rec_argc, i, j;
3004 const char **rec_argv;
3005 const char * const record_args[] = {
3006 "record",
3007 "-R",
3008 "-m", "1024",
3009 "-c", "1",
3010 };
3011 pid_t pid = getpid();
3012 char *filter = asprintf__tp_filter_pids(1, &pid);
3013 const char * const sc_args[] = { "-e", };
3014 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
3015 const char * const majpf_args[] = { "-e", "major-faults" };
3016 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
3017 const char * const minpf_args[] = { "-e", "minor-faults" };
3018 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
3019 int err = -1;
3020
3021 /* +3 is for the event string below and the pid filter */
3022 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 +
3023 majpf_args_nr + minpf_args_nr + argc;
3024 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3025
3026 if (rec_argv == NULL || filter == NULL)
3027 goto out_free;
3028
3029 j = 0;
3030 for (i = 0; i < ARRAY_SIZE(record_args); i++)
3031 rec_argv[j++] = record_args[i];
3032
3033 if (trace->trace_syscalls) {
3034 for (i = 0; i < sc_args_nr; i++)
3035 rec_argv[j++] = sc_args[i];
3036
3037 /* event string may be different for older kernels - e.g., RHEL6 */
3038 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
3039 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
3040 else if (is_valid_tracepoint("syscalls:sys_enter"))
3041 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
3042 else {
3043 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
3044 goto out_free;
3045 }
3046 }
3047
3048 rec_argv[j++] = "--filter";
3049 rec_argv[j++] = filter;
3050
3051 if (trace->trace_pgfaults & TRACE_PFMAJ)
3052 for (i = 0; i < majpf_args_nr; i++)
3053 rec_argv[j++] = majpf_args[i];
3054
3055 if (trace->trace_pgfaults & TRACE_PFMIN)
3056 for (i = 0; i < minpf_args_nr; i++)
3057 rec_argv[j++] = minpf_args[i];
3058
3059 for (i = 0; i < (unsigned int)argc; i++)
3060 rec_argv[j++] = argv[i];
3061
3062 err = cmd_record(j, rec_argv);
3063 out_free:
3064 free(filter);
3065 free(rec_argv);
3066 return err;
3067 }
3068
3069 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
3070
3071 static bool evlist__add_vfs_getname(struct evlist *evlist)
3072 {
3073 bool found = false;
3074 struct evsel *evsel, *tmp;
3075 struct parse_events_error err;
3076 int ret;
3077
3078 parse_events_error__init(&err);
3079 ret = parse_events(evlist, "probe:vfs_getname*", &err);
3080 parse_events_error__exit(&err);
3081 if (ret)
3082 return false;
3083
3084 evlist__for_each_entry_safe(evlist, evsel, tmp) {
3085 if (!strstarts(evsel__name(evsel), "probe:vfs_getname"))
3086 continue;
3087
3088 if (evsel__field(evsel, "pathname")) {
3089 evsel->handler = trace__vfs_getname;
3090 found = true;
3091 continue;
3092 }
3093
3094 list_del_init(&evsel->core.node);
3095 evsel->evlist = NULL;
3096 evsel__delete(evsel);
3097 }
3098
3099 return found;
3100 }
3101
3102 static struct evsel *evsel__new_pgfault(u64 config)
3103 {
3104 struct evsel *evsel;
3105 struct perf_event_attr attr = {
3106 .type = PERF_TYPE_SOFTWARE,
3107 .mmap_data = 1,
3108 };
3109
3110 attr.config = config;
3111 attr.sample_period = 1;
3112
3113 event_attr_init(&attr);
3114
3115 evsel = evsel__new(&attr);
3116 if (evsel)
3117 evsel->handler = trace__pgfault;
3118
3119 return evsel;
3120 }
3121
3122 static void evlist__free_syscall_tp_fields(struct evlist *evlist)
3123 {
3124 struct evsel *evsel;
3125
3126 evlist__for_each_entry(evlist, evsel) {
3127 struct evsel_trace *et = evsel->priv;
3128
3129 if (!et || !evsel->tp_format || strcmp(evsel->tp_format->system, "syscalls"))
3130 continue;
3131
3132 free(et->fmt);
3133 free(et);
3134 }
3135 }
3136
3137 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
3138 {
3139 const u32 type = event->header.type;
3140 struct evsel *evsel;
3141
3142 if (type != PERF_RECORD_SAMPLE) {
3143 trace__process_event(trace, trace->host, event, sample);
3144 return;
3145 }
3146
3147 evsel = evlist__id2evsel(trace->evlist, sample->id);
3148 if (evsel == NULL) {
3149 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
3150 return;
3151 }
3152
3153 if (evswitch__discard(&trace->evswitch, evsel))
3154 return;
3155
3156 trace__set_base_time(trace, evsel, sample);
3157
3158 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
3159 sample->raw_data == NULL) {
3160 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
3161 evsel__name(evsel), sample->tid,
3162 sample->cpu, sample->raw_size);
3163 } else {
3164 tracepoint_handler handler = evsel->handler;
3165 handler(trace, evsel, event, sample);
3166 }
3167
3168 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
3169 interrupted = true;
3170 }
3171
3172 static int trace__add_syscall_newtp(struct trace *trace)
3173 {
3174 int ret = -1;
3175 struct evlist *evlist = trace->evlist;
3176 struct evsel *sys_enter, *sys_exit;
3177
3178 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
3179 if (sys_enter == NULL)
3180 goto out;
3181
3182 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
3183 goto out_delete_sys_enter;
3184
3185 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
3186 if (sys_exit == NULL)
3187 goto out_delete_sys_enter;
3188
3189 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
3190 goto out_delete_sys_exit;
3191
3192 evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
3193 evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
3194
3195 evlist__add(evlist, sys_enter);
3196 evlist__add(evlist, sys_exit);
3197
3198 if (callchain_param.enabled && !trace->kernel_syscallchains) {
3199 /*
3200 * We're interested only in the user space callchain
3201 * leading to the syscall, allow overriding that for
3202 * debugging reasons using --kernel_syscall_callchains
3203 */
3204 sys_exit->core.attr.exclude_callchain_kernel = 1;
3205 }
3206
3207 trace->syscalls.events.sys_enter = sys_enter;
3208 trace->syscalls.events.sys_exit = sys_exit;
3209
3210 ret = 0;
3211 out:
3212 return ret;
3213
3214 out_delete_sys_exit:
3215 evsel__delete_priv(sys_exit);
3216 out_delete_sys_enter:
3217 evsel__delete_priv(sys_enter);
3218 goto out;
3219 }
3220
3221 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
3222 {
3223 int err = -1;
3224 struct evsel *sys_exit;
3225 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
3226 trace->ev_qualifier_ids.nr,
3227 trace->ev_qualifier_ids.entries);
3228
3229 if (filter == NULL)
3230 goto out_enomem;
3231
3232 if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) {
3233 sys_exit = trace->syscalls.events.sys_exit;
3234 err = evsel__append_tp_filter(sys_exit, filter);
3235 }
3236
3237 free(filter);
3238 out:
3239 return err;
3240 out_enomem:
3241 errno = ENOMEM;
3242 goto out;
3243 }
3244
3245 #ifdef HAVE_LIBBPF_SUPPORT
3246 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace, const char *name)
3247 {
3248 if (trace->bpf_obj == NULL)
3249 return NULL;
3250
3251 return bpf_object__find_map_by_name(trace->bpf_obj, name);
3252 }
3253
3254 static void trace__set_bpf_map_filtered_pids(struct trace *trace)
3255 {
3256 trace->filter_pids.map = trace__find_bpf_map_by_name(trace, "pids_filtered");
3257 }
3258
3259 static void trace__set_bpf_map_syscalls(struct trace *trace)
3260 {
3261 trace->syscalls.map = trace__find_bpf_map_by_name(trace, "syscalls");
3262 trace->syscalls.prog_array.sys_enter = trace__find_bpf_map_by_name(trace, "syscalls_sys_enter");
3263 trace->syscalls.prog_array.sys_exit = trace__find_bpf_map_by_name(trace, "syscalls_sys_exit");
3264 }
3265
3266 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
3267 {
3268 struct bpf_program *pos, *prog = NULL;
3269 const char *sec_name;
3270
3271 if (trace->bpf_obj == NULL)
3272 return NULL;
3273
3274 bpf_object__for_each_program(pos, trace->bpf_obj) {
3275 sec_name = bpf_program__section_name(pos);
3276 if (sec_name && !strcmp(sec_name, name)) {
3277 prog = pos;
3278 break;
3279 }
3280 }
3281
3282 return prog;
3283 }
3284
3285 static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
3286 const char *prog_name, const char *type)
3287 {
3288 struct bpf_program *prog;
3289
3290 if (prog_name == NULL) {
3291 char default_prog_name[256];
3292 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->name);
3293 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3294 if (prog != NULL)
3295 goto out_found;
3296 if (sc->fmt && sc->fmt->alias) {
3297 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->fmt->alias);
3298 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3299 if (prog != NULL)
3300 goto out_found;
3301 }
3302 goto out_unaugmented;
3303 }
3304
3305 prog = trace__find_bpf_program_by_title(trace, prog_name);
3306
3307 if (prog != NULL) {
3308 out_found:
3309 return prog;
3310 }
3311
3312 pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
3313 prog_name, type, sc->name);
3314 out_unaugmented:
3315 return trace->syscalls.unaugmented_prog;
3316 }
3317
3318 static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
3319 {
3320 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3321
3322 if (sc == NULL)
3323 return;
3324
3325 sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3326 sc->bpf_prog.sys_exit = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit : NULL, "exit");
3327 }
3328
3329 static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
3330 {
3331 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3332 return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3333 }
3334
3335 static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
3336 {
3337 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3338 return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3339 }
3340
3341 static void trace__init_bpf_map_syscall_args(struct trace *trace, int id, struct bpf_map_syscall_entry *entry)
3342 {
3343 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3344 int arg = 0;
3345
3346 if (sc == NULL)
3347 goto out;
3348
3349 for (; arg < sc->nr_args; ++arg) {
3350 entry->string_args_len[arg] = 0;
3351 if (sc->arg_fmt[arg].scnprintf == SCA_FILENAME) {
3352 /* Should be set like strace -s strsize */
3353 entry->string_args_len[arg] = PATH_MAX;
3354 }
3355 }
3356 out:
3357 for (; arg < 6; ++arg)
3358 entry->string_args_len[arg] = 0;
3359 }
3360 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace)
3361 {
3362 int fd = bpf_map__fd(trace->syscalls.map);
3363 struct bpf_map_syscall_entry value = {
3364 .enabled = !trace->not_ev_qualifier,
3365 };
3366 int err = 0;
3367 size_t i;
3368
3369 for (i = 0; i < trace->ev_qualifier_ids.nr; ++i) {
3370 int key = trace->ev_qualifier_ids.entries[i];
3371
3372 if (value.enabled) {
3373 trace__init_bpf_map_syscall_args(trace, key, &value);
3374 trace__init_syscall_bpf_progs(trace, key);
3375 }
3376
3377 err = bpf_map_update_elem(fd, &key, &value, BPF_EXIST);
3378 if (err)
3379 break;
3380 }
3381
3382 return err;
3383 }
3384
3385 static int __trace__init_syscalls_bpf_map(struct trace *trace, bool enabled)
3386 {
3387 int fd = bpf_map__fd(trace->syscalls.map);
3388 struct bpf_map_syscall_entry value = {
3389 .enabled = enabled,
3390 };
3391 int err = 0, key;
3392
3393 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3394 if (enabled)
3395 trace__init_bpf_map_syscall_args(trace, key, &value);
3396
3397 err = bpf_map_update_elem(fd, &key, &value, BPF_ANY);
3398 if (err)
3399 break;
3400 }
3401
3402 return err;
3403 }
3404
3405 static int trace__init_syscalls_bpf_map(struct trace *trace)
3406 {
3407 bool enabled = true;
3408
3409 if (trace->ev_qualifier_ids.nr)
3410 enabled = trace->not_ev_qualifier;
3411
3412 return __trace__init_syscalls_bpf_map(trace, enabled);
3413 }
3414
3415 static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
3416 {
3417 struct tep_format_field *field, *candidate_field;
3418 int id;
3419
3420 /*
3421 * We're only interested in syscalls that have a pointer:
3422 */
3423 for (field = sc->args; field; field = field->next) {
3424 if (field->flags & TEP_FIELD_IS_POINTER)
3425 goto try_to_find_pair;
3426 }
3427
3428 return NULL;
3429
3430 try_to_find_pair:
3431 for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) {
3432 struct syscall *pair = trace__syscall_info(trace, NULL, id);
3433 struct bpf_program *pair_prog;
3434 bool is_candidate = false;
3435
3436 if (pair == NULL || pair == sc ||
3437 pair->bpf_prog.sys_enter == trace->syscalls.unaugmented_prog)
3438 continue;
3439
3440 for (field = sc->args, candidate_field = pair->args;
3441 field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3442 bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3443 candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3444
3445 if (is_pointer) {
3446 if (!candidate_is_pointer) {
3447 // The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3448 continue;
3449 }
3450 } else {
3451 if (candidate_is_pointer) {
3452 // The candidate might copy a pointer we don't have, skip it.
3453 goto next_candidate;
3454 }
3455 continue;
3456 }
3457
3458 if (strcmp(field->type, candidate_field->type))
3459 goto next_candidate;
3460
3461 is_candidate = true;
3462 }
3463
3464 if (!is_candidate)
3465 goto next_candidate;
3466
3467 /*
3468 * Check if the tentative pair syscall augmenter has more pointers, if it has,
3469 * then it may be collecting that and we then can't use it, as it would collect
3470 * more than what is common to the two syscalls.
3471 */
3472 if (candidate_field) {
3473 for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3474 if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3475 goto next_candidate;
3476 }
3477
3478 pair_prog = pair->bpf_prog.sys_enter;
3479 /*
3480 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3481 * have been searched for, so search it here and if it returns the
3482 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3483 * program for a filtered syscall on a non-filtered one.
3484 *
3485 * For instance, we have "!syscalls:sys_enter_renameat" and that is
3486 * useful for "renameat2".
3487 */
3488 if (pair_prog == NULL) {
3489 pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3490 if (pair_prog == trace->syscalls.unaugmented_prog)
3491 goto next_candidate;
3492 }
3493
3494 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3495 return pair_prog;
3496 next_candidate:
3497 continue;
3498 }
3499
3500 return NULL;
3501 }
3502
3503 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3504 {
3505 int map_enter_fd = bpf_map__fd(trace->syscalls.prog_array.sys_enter),
3506 map_exit_fd = bpf_map__fd(trace->syscalls.prog_array.sys_exit);
3507 int err = 0, key;
3508
3509 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3510 int prog_fd;
3511
3512 if (!trace__syscall_enabled(trace, key))
3513 continue;
3514
3515 trace__init_syscall_bpf_progs(trace, key);
3516
3517 // It'll get at least the "!raw_syscalls:unaugmented"
3518 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3519 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3520 if (err)
3521 break;
3522 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3523 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3524 if (err)
3525 break;
3526 }
3527
3528 /*
3529 * Now lets do a second pass looking for enabled syscalls without
3530 * an augmenter that have a signature that is a superset of another
3531 * syscall with an augmenter so that we can auto-reuse it.
3532 *
3533 * I.e. if we have an augmenter for the "open" syscall that has
3534 * this signature:
3535 *
3536 * int open(const char *pathname, int flags, mode_t mode);
3537 *
3538 * I.e. that will collect just the first string argument, then we
3539 * can reuse it for the 'creat' syscall, that has this signature:
3540 *
3541 * int creat(const char *pathname, mode_t mode);
3542 *
3543 * and for:
3544 *
3545 * int stat(const char *pathname, struct stat *statbuf);
3546 * int lstat(const char *pathname, struct stat *statbuf);
3547 *
3548 * Because the 'open' augmenter will collect the first arg as a string,
3549 * and leave alone all the other args, which already helps with
3550 * beautifying 'stat' and 'lstat''s pathname arg.
3551 *
3552 * Then, in time, when 'stat' gets an augmenter that collects both
3553 * first and second arg (this one on the raw_syscalls:sys_exit prog
3554 * array tail call, then that one will be used.
3555 */
3556 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3557 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3558 struct bpf_program *pair_prog;
3559 int prog_fd;
3560
3561 if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3562 continue;
3563
3564 /*
3565 * For now we're just reusing the sys_enter prog, and if it
3566 * already has an augmenter, we don't need to find one.
3567 */
3568 if (sc->bpf_prog.sys_enter != trace->syscalls.unaugmented_prog)
3569 continue;
3570
3571 /*
3572 * Look at all the other syscalls for one that has a signature
3573 * that is close enough that we can share:
3574 */
3575 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3576 if (pair_prog == NULL)
3577 continue;
3578
3579 sc->bpf_prog.sys_enter = pair_prog;
3580
3581 /*
3582 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3583 * with the fd for the program we're reusing:
3584 */
3585 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3586 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3587 if (err)
3588 break;
3589 }
3590
3591
3592 return err;
3593 }
3594
3595 static void trace__delete_augmented_syscalls(struct trace *trace)
3596 {
3597 struct evsel *evsel, *tmp;
3598
3599 evlist__remove(trace->evlist, trace->syscalls.events.augmented);
3600 evsel__delete(trace->syscalls.events.augmented);
3601 trace->syscalls.events.augmented = NULL;
3602
3603 evlist__for_each_entry_safe(trace->evlist, tmp, evsel) {
3604 if (evsel->bpf_obj == trace->bpf_obj) {
3605 evlist__remove(trace->evlist, evsel);
3606 evsel__delete(evsel);
3607 }
3608
3609 }
3610
3611 bpf_object__close(trace->bpf_obj);
3612 trace->bpf_obj = NULL;
3613 }
3614 #else // HAVE_LIBBPF_SUPPORT
3615 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace __maybe_unused,
3616 const char *name __maybe_unused)
3617 {
3618 return NULL;
3619 }
3620
3621 static void trace__set_bpf_map_filtered_pids(struct trace *trace __maybe_unused)
3622 {
3623 }
3624
3625 static void trace__set_bpf_map_syscalls(struct trace *trace __maybe_unused)
3626 {
3627 }
3628
3629 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace __maybe_unused)
3630 {
3631 return 0;
3632 }
3633
3634 static int trace__init_syscalls_bpf_map(struct trace *trace __maybe_unused)
3635 {
3636 return 0;
3637 }
3638
3639 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace __maybe_unused,
3640 const char *name __maybe_unused)
3641 {
3642 return NULL;
3643 }
3644
3645 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace __maybe_unused)
3646 {
3647 return 0;
3648 }
3649
3650 static void trace__delete_augmented_syscalls(struct trace *trace __maybe_unused)
3651 {
3652 }
3653 #endif // HAVE_LIBBPF_SUPPORT
3654
3655 static bool trace__only_augmented_syscalls_evsels(struct trace *trace)
3656 {
3657 struct evsel *evsel;
3658
3659 evlist__for_each_entry(trace->evlist, evsel) {
3660 if (evsel == trace->syscalls.events.augmented ||
3661 evsel->bpf_obj == trace->bpf_obj)
3662 continue;
3663
3664 return false;
3665 }
3666
3667 return true;
3668 }
3669
3670 static int trace__set_ev_qualifier_filter(struct trace *trace)
3671 {
3672 if (trace->syscalls.map)
3673 return trace__set_ev_qualifier_bpf_filter(trace);
3674 if (trace->syscalls.events.sys_enter)
3675 return trace__set_ev_qualifier_tp_filter(trace);
3676 return 0;
3677 }
3678
3679 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3680 size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3681 {
3682 int err = 0;
3683 #ifdef HAVE_LIBBPF_SUPPORT
3684 bool value = true;
3685 int map_fd = bpf_map__fd(map);
3686 size_t i;
3687
3688 for (i = 0; i < npids; ++i) {
3689 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3690 if (err)
3691 break;
3692 }
3693 #endif
3694 return err;
3695 }
3696
3697 static int trace__set_filter_loop_pids(struct trace *trace)
3698 {
3699 unsigned int nr = 1, err;
3700 pid_t pids[32] = {
3701 getpid(),
3702 };
3703 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3704
3705 while (thread && nr < ARRAY_SIZE(pids)) {
3706 struct thread *parent = machine__find_thread(trace->host, thread->ppid, thread->ppid);
3707
3708 if (parent == NULL)
3709 break;
3710
3711 if (!strcmp(thread__comm_str(parent), "sshd") ||
3712 strstarts(thread__comm_str(parent), "gnome-terminal")) {
3713 pids[nr++] = parent->tid;
3714 break;
3715 }
3716 thread = parent;
3717 }
3718
3719 err = evlist__append_tp_filter_pids(trace->evlist, nr, pids);
3720 if (!err && trace->filter_pids.map)
3721 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3722
3723 return err;
3724 }
3725
3726 static int trace__set_filter_pids(struct trace *trace)
3727 {
3728 int err = 0;
3729 /*
3730 * Better not use !target__has_task() here because we need to cover the
3731 * case where no threads were specified in the command line, but a
3732 * workload was, and in that case we will fill in the thread_map when
3733 * we fork the workload in evlist__prepare_workload.
3734 */
3735 if (trace->filter_pids.nr > 0) {
3736 err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3737 trace->filter_pids.entries);
3738 if (!err && trace->filter_pids.map) {
3739 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3740 trace->filter_pids.entries);
3741 }
3742 } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
3743 err = trace__set_filter_loop_pids(trace);
3744 }
3745
3746 return err;
3747 }
3748
3749 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
3750 {
3751 struct evlist *evlist = trace->evlist;
3752 struct perf_sample sample;
3753 int err = evlist__parse_sample(evlist, event, &sample);
3754
3755 if (err)
3756 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3757 else
3758 trace__handle_event(trace, event, &sample);
3759
3760 return 0;
3761 }
3762
3763 static int __trace__flush_events(struct trace *trace)
3764 {
3765 u64 first = ordered_events__first_time(&trace->oe.data);
3766 u64 flush = trace->oe.last - NSEC_PER_SEC;
3767
3768 /* Is there some thing to flush.. */
3769 if (first && first < flush)
3770 return ordered_events__flush_time(&trace->oe.data, flush);
3771
3772 return 0;
3773 }
3774
3775 static int trace__flush_events(struct trace *trace)
3776 {
3777 return !trace->sort_events ? 0 : __trace__flush_events(trace);
3778 }
3779
3780 static int trace__deliver_event(struct trace *trace, union perf_event *event)
3781 {
3782 int err;
3783
3784 if (!trace->sort_events)
3785 return __trace__deliver_event(trace, event);
3786
3787 err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
3788 if (err && err != -1)
3789 return err;
3790
3791 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0, NULL);
3792 if (err)
3793 return err;
3794
3795 return trace__flush_events(trace);
3796 }
3797
3798 static int ordered_events__deliver_event(struct ordered_events *oe,
3799 struct ordered_event *event)
3800 {
3801 struct trace *trace = container_of(oe, struct trace, oe.data);
3802
3803 return __trace__deliver_event(trace, event->event);
3804 }
3805
3806 static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg)
3807 {
3808 struct tep_format_field *field;
3809 struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel);
3810
3811 if (evsel->tp_format == NULL || fmt == NULL)
3812 return NULL;
3813
3814 for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt)
3815 if (strcmp(field->name, arg) == 0)
3816 return fmt;
3817
3818 return NULL;
3819 }
3820
3821 static int trace__expand_filter(struct trace *trace __maybe_unused, struct evsel *evsel)
3822 {
3823 char *tok, *left = evsel->filter, *new_filter = evsel->filter;
3824
3825 while ((tok = strpbrk(left, "=<>!")) != NULL) {
3826 char *right = tok + 1, *right_end;
3827
3828 if (*right == '=')
3829 ++right;
3830
3831 while (isspace(*right))
3832 ++right;
3833
3834 if (*right == '\0')
3835 break;
3836
3837 while (!isalpha(*left))
3838 if (++left == tok) {
3839 /*
3840 * Bail out, can't find the name of the argument that is being
3841 * used in the filter, let it try to set this filter, will fail later.
3842 */
3843 return 0;
3844 }
3845
3846 right_end = right + 1;
3847 while (isalnum(*right_end) || *right_end == '_' || *right_end == '|')
3848 ++right_end;
3849
3850 if (isalpha(*right)) {
3851 struct syscall_arg_fmt *fmt;
3852 int left_size = tok - left,
3853 right_size = right_end - right;
3854 char arg[128];
3855
3856 while (isspace(left[left_size - 1]))
3857 --left_size;
3858
3859 scnprintf(arg, sizeof(arg), "%.*s", left_size, left);
3860
3861 fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg);
3862 if (fmt == NULL) {
3863 pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n",
3864 arg, evsel->name, evsel->filter);
3865 return -1;
3866 }
3867
3868 pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ",
3869 arg, (int)(right - tok), tok, right_size, right);
3870
3871 if (fmt->strtoul) {
3872 u64 val;
3873 struct syscall_arg syscall_arg = {
3874 .parm = fmt->parm,
3875 };
3876
3877 if (fmt->strtoul(right, right_size, &syscall_arg, &val)) {
3878 char *n, expansion[19];
3879 int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val);
3880 int expansion_offset = right - new_filter;
3881
3882 pr_debug("%s", expansion);
3883
3884 if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) {
3885 pr_debug(" out of memory!\n");
3886 free(new_filter);
3887 return -1;
3888 }
3889 if (new_filter != evsel->filter)
3890 free(new_filter);
3891 left = n + expansion_offset + expansion_lenght;
3892 new_filter = n;
3893 } else {
3894 pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3895 right_size, right, arg, evsel->name, evsel->filter);
3896 return -1;
3897 }
3898 } else {
3899 pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3900 arg, evsel->name, evsel->filter);
3901 return -1;
3902 }
3903
3904 pr_debug("\n");
3905 } else {
3906 left = right_end;
3907 }
3908 }
3909
3910 if (new_filter != evsel->filter) {
3911 pr_debug("New filter for %s: %s\n", evsel->name, new_filter);
3912 evsel__set_filter(evsel, new_filter);
3913 free(new_filter);
3914 }
3915
3916 return 0;
3917 }
3918
3919 static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel)
3920 {
3921 struct evlist *evlist = trace->evlist;
3922 struct evsel *evsel;
3923
3924 evlist__for_each_entry(evlist, evsel) {
3925 if (evsel->filter == NULL)
3926 continue;
3927
3928 if (trace__expand_filter(trace, evsel)) {
3929 *err_evsel = evsel;
3930 return -1;
3931 }
3932 }
3933
3934 return 0;
3935 }
3936
3937 static int trace__run(struct trace *trace, int argc, const char **argv)
3938 {
3939 struct evlist *evlist = trace->evlist;
3940 struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
3941 int err = -1, i;
3942 unsigned long before;
3943 const bool forks = argc > 0;
3944 bool draining = false;
3945
3946 trace->live = true;
3947
3948 if (!trace->raw_augmented_syscalls) {
3949 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3950 goto out_error_raw_syscalls;
3951
3952 if (trace->trace_syscalls)
3953 trace->vfs_getname = evlist__add_vfs_getname(evlist);
3954 }
3955
3956 if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
3957 pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
3958 if (pgfault_maj == NULL)
3959 goto out_error_mem;
3960 evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
3961 evlist__add(evlist, pgfault_maj);
3962 }
3963
3964 if ((trace->trace_pgfaults & TRACE_PFMIN)) {
3965 pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
3966 if (pgfault_min == NULL)
3967 goto out_error_mem;
3968 evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
3969 evlist__add(evlist, pgfault_min);
3970 }
3971
3972 /* Enable ignoring missing threads when -u/-p option is defined. */
3973 trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid;
3974
3975 if (trace->sched &&
3976 evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime))
3977 goto out_error_sched_stat_runtime;
3978 /*
3979 * If a global cgroup was set, apply it to all the events without an
3980 * explicit cgroup. I.e.:
3981 *
3982 * trace -G A -e sched:*switch
3983 *
3984 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3985 * _and_ sched:sched_switch to the 'A' cgroup, while:
3986 *
3987 * trace -e sched:*switch -G A
3988 *
3989 * will only set the sched:sched_switch event to the 'A' cgroup, all the
3990 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3991 * a cgroup (on the root cgroup, sys wide, etc).
3992 *
3993 * Multiple cgroups:
3994 *
3995 * trace -G A -e sched:*switch -G B
3996 *
3997 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
3998 * to the 'B' cgroup.
3999 *
4000 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
4001 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
4002 */
4003 if (trace->cgroup)
4004 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
4005
4006 err = evlist__create_maps(evlist, &trace->opts.target);
4007 if (err < 0) {
4008 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
4009 goto out_delete_evlist;
4010 }
4011
4012 err = trace__symbols_init(trace, evlist);
4013 if (err < 0) {
4014 fprintf(trace->output, "Problems initializing symbol libraries!\n");
4015 goto out_delete_evlist;
4016 }
4017
4018 evlist__config(evlist, &trace->opts, &callchain_param);
4019
4020 if (forks) {
4021 err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL);
4022 if (err < 0) {
4023 fprintf(trace->output, "Couldn't run the workload!\n");
4024 goto out_delete_evlist;
4025 }
4026 workload_pid = evlist->workload.pid;
4027 }
4028
4029 err = evlist__open(evlist);
4030 if (err < 0)
4031 goto out_error_open;
4032
4033 err = bpf__apply_obj_config();
4034 if (err) {
4035 char errbuf[BUFSIZ];
4036
4037 bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
4038 pr_err("ERROR: Apply config to BPF failed: %s\n",
4039 errbuf);
4040 goto out_error_open;
4041 }
4042
4043 err = trace__set_filter_pids(trace);
4044 if (err < 0)
4045 goto out_error_mem;
4046
4047 if (trace->syscalls.map)
4048 trace__init_syscalls_bpf_map(trace);
4049
4050 if (trace->syscalls.prog_array.sys_enter)
4051 trace__init_syscalls_bpf_prog_array_maps(trace);
4052
4053 if (trace->ev_qualifier_ids.nr > 0) {
4054 err = trace__set_ev_qualifier_filter(trace);
4055 if (err < 0)
4056 goto out_errno;
4057
4058 if (trace->syscalls.events.sys_exit) {
4059 pr_debug("event qualifier tracepoint filter: %s\n",
4060 trace->syscalls.events.sys_exit->filter);
4061 }
4062 }
4063
4064 /*
4065 * If the "close" syscall is not traced, then we will not have the
4066 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
4067 * fd->pathname table and were ending up showing the last value set by
4068 * syscalls opening a pathname and associating it with a descriptor or
4069 * reading it from /proc/pid/fd/ in cases where that doesn't make
4070 * sense.
4071 *
4072 * So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
4073 * not in use.
4074 */
4075 trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
4076
4077 err = trace__expand_filters(trace, &evsel);
4078 if (err)
4079 goto out_delete_evlist;
4080 err = evlist__apply_filters(evlist, &evsel);
4081 if (err < 0)
4082 goto out_error_apply_filters;
4083
4084 if (trace->dump.map)
4085 bpf_map__fprintf(trace->dump.map, trace->output);
4086
4087 err = evlist__mmap(evlist, trace->opts.mmap_pages);
4088 if (err < 0)
4089 goto out_error_mmap;
4090
4091 if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
4092 evlist__enable(evlist);
4093
4094 if (forks)
4095 evlist__start_workload(evlist);
4096
4097 if (trace->opts.initial_delay) {
4098 usleep(trace->opts.initial_delay * 1000);
4099 evlist__enable(evlist);
4100 }
4101
4102 trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
4103 evlist->core.threads->nr > 1 ||
4104 evlist__first(evlist)->core.attr.inherit;
4105
4106 /*
4107 * Now that we already used evsel->core.attr to ask the kernel to setup the
4108 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
4109 * trace__resolve_callchain(), allowing per-event max-stack settings
4110 * to override an explicitly set --max-stack global setting.
4111 */
4112 evlist__for_each_entry(evlist, evsel) {
4113 if (evsel__has_callchain(evsel) &&
4114 evsel->core.attr.sample_max_stack == 0)
4115 evsel->core.attr.sample_max_stack = trace->max_stack;
4116 }
4117 again:
4118 before = trace->nr_events;
4119
4120 for (i = 0; i < evlist->core.nr_mmaps; i++) {
4121 union perf_event *event;
4122 struct mmap *md;
4123
4124 md = &evlist->mmap[i];
4125 if (perf_mmap__read_init(&md->core) < 0)
4126 continue;
4127
4128 while ((event = perf_mmap__read_event(&md->core)) != NULL) {
4129 ++trace->nr_events;
4130
4131 err = trace__deliver_event(trace, event);
4132 if (err)
4133 goto out_disable;
4134
4135 perf_mmap__consume(&md->core);
4136
4137 if (interrupted)
4138 goto out_disable;
4139
4140 if (done && !draining) {
4141 evlist__disable(evlist);
4142 draining = true;
4143 }
4144 }
4145 perf_mmap__read_done(&md->core);
4146 }
4147
4148 if (trace->nr_events == before) {
4149 int timeout = done ? 100 : -1;
4150
4151 if (!draining && evlist__poll(evlist, timeout) > 0) {
4152 if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
4153 draining = true;
4154
4155 goto again;
4156 } else {
4157 if (trace__flush_events(trace))
4158 goto out_disable;
4159 }
4160 } else {
4161 goto again;
4162 }
4163
4164 out_disable:
4165 thread__zput(trace->current);
4166
4167 evlist__disable(evlist);
4168
4169 if (trace->sort_events)
4170 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
4171
4172 if (!err) {
4173 if (trace->summary)
4174 trace__fprintf_thread_summary(trace, trace->output);
4175
4176 if (trace->show_tool_stats) {
4177 fprintf(trace->output, "Stats:\n "
4178 " vfs_getname : %" PRIu64 "\n"
4179 " proc_getname: %" PRIu64 "\n",
4180 trace->stats.vfs_getname,
4181 trace->stats.proc_getname);
4182 }
4183 }
4184
4185 out_delete_evlist:
4186 trace__symbols__exit(trace);
4187 evlist__free_syscall_tp_fields(evlist);
4188 evlist__delete(evlist);
4189 cgroup__put(trace->cgroup);
4190 trace->evlist = NULL;
4191 trace->live = false;
4192 return err;
4193 {
4194 char errbuf[BUFSIZ];
4195
4196 out_error_sched_stat_runtime:
4197 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
4198 goto out_error;
4199
4200 out_error_raw_syscalls:
4201 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
4202 goto out_error;
4203
4204 out_error_mmap:
4205 evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
4206 goto out_error;
4207
4208 out_error_open:
4209 evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
4210
4211 out_error:
4212 fprintf(trace->output, "%s\n", errbuf);
4213 goto out_delete_evlist;
4214
4215 out_error_apply_filters:
4216 fprintf(trace->output,
4217 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
4218 evsel->filter, evsel__name(evsel), errno,
4219 str_error_r(errno, errbuf, sizeof(errbuf)));
4220 goto out_delete_evlist;
4221 }
4222 out_error_mem:
4223 fprintf(trace->output, "Not enough memory to run!\n");
4224 goto out_delete_evlist;
4225
4226 out_errno:
4227 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
4228 goto out_delete_evlist;
4229 }
4230
4231 static int trace__replay(struct trace *trace)
4232 {
4233 const struct evsel_str_handler handlers[] = {
4234 { "probe:vfs_getname", trace__vfs_getname, },
4235 };
4236 struct perf_data data = {
4237 .path = input_name,
4238 .mode = PERF_DATA_MODE_READ,
4239 .force = trace->force,
4240 };
4241 struct perf_session *session;
4242 struct evsel *evsel;
4243 int err = -1;
4244
4245 trace->tool.sample = trace__process_sample;
4246 trace->tool.mmap = perf_event__process_mmap;
4247 trace->tool.mmap2 = perf_event__process_mmap2;
4248 trace->tool.comm = perf_event__process_comm;
4249 trace->tool.exit = perf_event__process_exit;
4250 trace->tool.fork = perf_event__process_fork;
4251 trace->tool.attr = perf_event__process_attr;
4252 trace->tool.tracing_data = perf_event__process_tracing_data;
4253 trace->tool.build_id = perf_event__process_build_id;
4254 trace->tool.namespaces = perf_event__process_namespaces;
4255
4256 trace->tool.ordered_events = true;
4257 trace->tool.ordering_requires_timestamps = true;
4258
4259 /* add tid to output */
4260 trace->multiple_threads = true;
4261
4262 session = perf_session__new(&data, &trace->tool);
4263 if (IS_ERR(session))
4264 return PTR_ERR(session);
4265
4266 if (trace->opts.target.pid)
4267 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
4268
4269 if (trace->opts.target.tid)
4270 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
4271
4272 if (symbol__init(&session->header.env) < 0)
4273 goto out;
4274
4275 trace->host = &session->machines.host;
4276
4277 err = perf_session__set_tracepoints_handlers(session, handlers);
4278 if (err)
4279 goto out;
4280
4281 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter");
4282 trace->syscalls.events.sys_enter = evsel;
4283 /* older kernels have syscalls tp versus raw_syscalls */
4284 if (evsel == NULL)
4285 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter");
4286
4287 if (evsel &&
4288 (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
4289 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
4290 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
4291 goto out;
4292 }
4293
4294 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit");
4295 trace->syscalls.events.sys_exit = evsel;
4296 if (evsel == NULL)
4297 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit");
4298 if (evsel &&
4299 (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
4300 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
4301 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
4302 goto out;
4303 }
4304
4305 evlist__for_each_entry(session->evlist, evsel) {
4306 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
4307 (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
4308 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
4309 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
4310 evsel->handler = trace__pgfault;
4311 }
4312
4313 setup_pager();
4314
4315 err = perf_session__process_events(session);
4316 if (err)
4317 pr_err("Failed to process events, error %d", err);
4318
4319 else if (trace->summary)
4320 trace__fprintf_thread_summary(trace, trace->output);
4321
4322 out:
4323 perf_session__delete(session);
4324
4325 return err;
4326 }
4327
4328 static size_t trace__fprintf_threads_header(FILE *fp)
4329 {
4330 size_t printed;
4331
4332 printed = fprintf(fp, "\n Summary of events:\n\n");
4333
4334 return printed;
4335 }
4336
4337 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
4338 struct syscall_stats *stats;
4339 double msecs;
4340 int syscall;
4341 )
4342 {
4343 struct int_node *source = rb_entry(nd, struct int_node, rb_node);
4344 struct syscall_stats *stats = source->priv;
4345
4346 entry->syscall = source->i;
4347 entry->stats = stats;
4348 entry->msecs = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0;
4349 }
4350
4351 static size_t thread__dump_stats(struct thread_trace *ttrace,
4352 struct trace *trace, FILE *fp)
4353 {
4354 size_t printed = 0;
4355 struct syscall *sc;
4356 struct rb_node *nd;
4357 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
4358
4359 if (syscall_stats == NULL)
4360 return 0;
4361
4362 printed += fprintf(fp, "\n");
4363
4364 printed += fprintf(fp, " syscall calls errors total min avg max stddev\n");
4365 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
4366 printed += fprintf(fp, " --------------- -------- ------ -------- --------- --------- --------- ------\n");
4367
4368 resort_rb__for_each_entry(nd, syscall_stats) {
4369 struct syscall_stats *stats = syscall_stats_entry->stats;
4370 if (stats) {
4371 double min = (double)(stats->stats.min) / NSEC_PER_MSEC;
4372 double max = (double)(stats->stats.max) / NSEC_PER_MSEC;
4373 double avg = avg_stats(&stats->stats);
4374 double pct;
4375 u64 n = (u64)stats->stats.n;
4376
4377 pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0;
4378 avg /= NSEC_PER_MSEC;
4379
4380 sc = &trace->syscalls.table[syscall_stats_entry->syscall];
4381 printed += fprintf(fp, " %-15s", sc->name);
4382 printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f",
4383 n, stats->nr_failures, syscall_stats_entry->msecs, min, avg);
4384 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
4385
4386 if (trace->errno_summary && stats->nr_failures) {
4387 const char *arch_name = perf_env__arch(trace->host->env);
4388 int e;
4389
4390 for (e = 0; e < stats->max_errno; ++e) {
4391 if (stats->errnos[e] != 0)
4392 fprintf(fp, "\t\t\t\t%s: %d\n", arch_syscalls__strerrno(arch_name, e + 1), stats->errnos[e]);
4393 }
4394 }
4395 }
4396 }
4397
4398 resort_rb__delete(syscall_stats);
4399 printed += fprintf(fp, "\n\n");
4400
4401 return printed;
4402 }
4403
4404 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
4405 {
4406 size_t printed = 0;
4407 struct thread_trace *ttrace = thread__priv(thread);
4408 double ratio;
4409
4410 if (ttrace == NULL)
4411 return 0;
4412
4413 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
4414
4415 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
4416 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
4417 printed += fprintf(fp, "%.1f%%", ratio);
4418 if (ttrace->pfmaj)
4419 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
4420 if (ttrace->pfmin)
4421 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
4422 if (trace->sched)
4423 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
4424 else if (fputc('\n', fp) != EOF)
4425 ++printed;
4426
4427 printed += thread__dump_stats(ttrace, trace, fp);
4428
4429 return printed;
4430 }
4431
4432 static unsigned long thread__nr_events(struct thread_trace *ttrace)
4433 {
4434 return ttrace ? ttrace->nr_events : 0;
4435 }
4436
4437 DEFINE_RESORT_RB(threads, (thread__nr_events(a->thread->priv) < thread__nr_events(b->thread->priv)),
4438 struct thread *thread;
4439 )
4440 {
4441 entry->thread = rb_entry(nd, struct thread, rb_node);
4442 }
4443
4444 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
4445 {
4446 size_t printed = trace__fprintf_threads_header(fp);
4447 struct rb_node *nd;
4448 int i;
4449
4450 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
4451 DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
4452
4453 if (threads == NULL) {
4454 fprintf(fp, "%s", "Error sorting output by nr_events!\n");
4455 return 0;
4456 }
4457
4458 resort_rb__for_each_entry(nd, threads)
4459 printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
4460
4461 resort_rb__delete(threads);
4462 }
4463 return printed;
4464 }
4465
4466 static int trace__set_duration(const struct option *opt, const char *str,
4467 int unset __maybe_unused)
4468 {
4469 struct trace *trace = opt->value;
4470
4471 trace->duration_filter = atof(str);
4472 return 0;
4473 }
4474
4475 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
4476 int unset __maybe_unused)
4477 {
4478 int ret = -1;
4479 size_t i;
4480 struct trace *trace = opt->value;
4481 /*
4482 * FIXME: introduce a intarray class, plain parse csv and create a
4483 * { int nr, int entries[] } struct...
4484 */
4485 struct intlist *list = intlist__new(str);
4486
4487 if (list == NULL)
4488 return -1;
4489
4490 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
4491 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
4492
4493 if (trace->filter_pids.entries == NULL)
4494 goto out;
4495
4496 trace->filter_pids.entries[0] = getpid();
4497
4498 for (i = 1; i < trace->filter_pids.nr; ++i)
4499 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
4500
4501 intlist__delete(list);
4502 ret = 0;
4503 out:
4504 return ret;
4505 }
4506
4507 static int trace__open_output(struct trace *trace, const char *filename)
4508 {
4509 struct stat st;
4510
4511 if (!stat(filename, &st) && st.st_size) {
4512 char oldname[PATH_MAX];
4513
4514 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
4515 unlink(oldname);
4516 rename(filename, oldname);
4517 }
4518
4519 trace->output = fopen(filename, "w");
4520
4521 return trace->output == NULL ? -errno : 0;
4522 }
4523
4524 static int parse_pagefaults(const struct option *opt, const char *str,
4525 int unset __maybe_unused)
4526 {
4527 int *trace_pgfaults = opt->value;
4528
4529 if (strcmp(str, "all") == 0)
4530 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
4531 else if (strcmp(str, "maj") == 0)
4532 *trace_pgfaults |= TRACE_PFMAJ;
4533 else if (strcmp(str, "min") == 0)
4534 *trace_pgfaults |= TRACE_PFMIN;
4535 else
4536 return -1;
4537
4538 return 0;
4539 }
4540
4541 static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)
4542 {
4543 struct evsel *evsel;
4544
4545 evlist__for_each_entry(evlist, evsel) {
4546 if (evsel->handler == NULL)
4547 evsel->handler = handler;
4548 }
4549 }
4550
4551 static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name)
4552 {
4553 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
4554
4555 if (fmt) {
4556 struct syscall_fmt *scfmt = syscall_fmt__find(name);
4557
4558 if (scfmt) {
4559 int skip = 0;
4560
4561 if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 ||
4562 strcmp(evsel->tp_format->format.fields->name, "nr") == 0)
4563 ++skip;
4564
4565 memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt));
4566 }
4567 }
4568 }
4569
4570 static int evlist__set_syscall_tp_fields(struct evlist *evlist)
4571 {
4572 struct evsel *evsel;
4573
4574 evlist__for_each_entry(evlist, evsel) {
4575 if (evsel->priv || !evsel->tp_format)
4576 continue;
4577
4578 if (strcmp(evsel->tp_format->system, "syscalls")) {
4579 evsel__init_tp_arg_scnprintf(evsel);
4580 continue;
4581 }
4582
4583 if (evsel__init_syscall_tp(evsel))
4584 return -1;
4585
4586 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
4587 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4588
4589 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
4590 return -1;
4591
4592 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1);
4593 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
4594 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4595
4596 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
4597 return -1;
4598
4599 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1);
4600 }
4601 }
4602
4603 return 0;
4604 }
4605
4606 /*
4607 * XXX: Hackish, just splitting the combined -e+--event (syscalls
4608 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
4609 * existing facilities unchanged (trace->ev_qualifier + parse_options()).
4610 *
4611 * It'd be better to introduce a parse_options() variant that would return a
4612 * list with the terms it didn't match to an event...
4613 */
4614 static int trace__parse_events_option(const struct option *opt, const char *str,
4615 int unset __maybe_unused)
4616 {
4617 struct trace *trace = (struct trace *)opt->value;
4618 const char *s = str;
4619 char *sep = NULL, *lists[2] = { NULL, NULL, };
4620 int len = strlen(str) + 1, err = -1, list, idx;
4621 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
4622 char group_name[PATH_MAX];
4623 struct syscall_fmt *fmt;
4624
4625 if (strace_groups_dir == NULL)
4626 return -1;
4627
4628 if (*s == '!') {
4629 ++s;
4630 trace->not_ev_qualifier = true;
4631 }
4632
4633 while (1) {
4634 if ((sep = strchr(s, ',')) != NULL)
4635 *sep = '\0';
4636
4637 list = 0;
4638 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
4639 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
4640 list = 1;
4641 goto do_concat;
4642 }
4643
4644 fmt = syscall_fmt__find_by_alias(s);
4645 if (fmt != NULL) {
4646 list = 1;
4647 s = fmt->name;
4648 } else {
4649 path__join(group_name, sizeof(group_name), strace_groups_dir, s);
4650 if (access(group_name, R_OK) == 0)
4651 list = 1;
4652 }
4653 do_concat:
4654 if (lists[list]) {
4655 sprintf(lists[list] + strlen(lists[list]), ",%s", s);
4656 } else {
4657 lists[list] = malloc(len);
4658 if (lists[list] == NULL)
4659 goto out;
4660 strcpy(lists[list], s);
4661 }
4662
4663 if (!sep)
4664 break;
4665
4666 *sep = ',';
4667 s = sep + 1;
4668 }
4669
4670 if (lists[1] != NULL) {
4671 struct strlist_config slist_config = {
4672 .dirname = strace_groups_dir,
4673 };
4674
4675 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4676 if (trace->ev_qualifier == NULL) {
4677 fputs("Not enough memory to parse event qualifier", trace->output);
4678 goto out;
4679 }
4680
4681 if (trace__validate_ev_qualifier(trace))
4682 goto out;
4683 trace->trace_syscalls = true;
4684 }
4685
4686 err = 0;
4687
4688 if (lists[0]) {
4689 struct option o = {
4690 .value = &trace->evlist,
4691 };
4692 err = parse_events_option(&o, lists[0], 0);
4693 }
4694 out:
4695 free(strace_groups_dir);
4696 free(lists[0]);
4697 free(lists[1]);
4698 if (sep)
4699 *sep = ',';
4700
4701 return err;
4702 }
4703
4704 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4705 {
4706 struct trace *trace = opt->value;
4707
4708 if (!list_empty(&trace->evlist->core.entries)) {
4709 struct option o = {
4710 .value = &trace->evlist,
4711 };
4712 return parse_cgroups(&o, str, unset);
4713 }
4714 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
4715
4716 return 0;
4717 }
4718
4719 static int trace__config(const char *var, const char *value, void *arg)
4720 {
4721 struct trace *trace = arg;
4722 int err = 0;
4723
4724 if (!strcmp(var, "trace.add_events")) {
4725 trace->perfconfig_events = strdup(value);
4726 if (trace->perfconfig_events == NULL) {
4727 pr_err("Not enough memory for %s\n", "trace.add_events");
4728 return -1;
4729 }
4730 } else if (!strcmp(var, "trace.show_timestamp")) {
4731 trace->show_tstamp = perf_config_bool(var, value);
4732 } else if (!strcmp(var, "trace.show_duration")) {
4733 trace->show_duration = perf_config_bool(var, value);
4734 } else if (!strcmp(var, "trace.show_arg_names")) {
4735 trace->show_arg_names = perf_config_bool(var, value);
4736 if (!trace->show_arg_names)
4737 trace->show_zeros = true;
4738 } else if (!strcmp(var, "trace.show_zeros")) {
4739 bool new_show_zeros = perf_config_bool(var, value);
4740 if (!trace->show_arg_names && !new_show_zeros) {
4741 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4742 goto out;
4743 }
4744 trace->show_zeros = new_show_zeros;
4745 } else if (!strcmp(var, "trace.show_prefix")) {
4746 trace->show_string_prefix = perf_config_bool(var, value);
4747 } else if (!strcmp(var, "trace.no_inherit")) {
4748 trace->opts.no_inherit = perf_config_bool(var, value);
4749 } else if (!strcmp(var, "trace.args_alignment")) {
4750 int args_alignment = 0;
4751 if (perf_config_int(&args_alignment, var, value) == 0)
4752 trace->args_alignment = args_alignment;
4753 } else if (!strcmp(var, "trace.tracepoint_beautifiers")) {
4754 if (strcasecmp(value, "libtraceevent") == 0)
4755 trace->libtraceevent_print = true;
4756 else if (strcasecmp(value, "libbeauty") == 0)
4757 trace->libtraceevent_print = false;
4758 }
4759 out:
4760 return err;
4761 }
4762
4763 static void trace__exit(struct trace *trace)
4764 {
4765 int i;
4766
4767 strlist__delete(trace->ev_qualifier);
4768 free(trace->ev_qualifier_ids.entries);
4769 if (trace->syscalls.table) {
4770 for (i = 0; i <= trace->sctbl->syscalls.max_id; i++)
4771 syscall__exit(&trace->syscalls.table[i]);
4772 free(trace->syscalls.table);
4773 }
4774 syscalltbl__delete(trace->sctbl);
4775 zfree(&trace->perfconfig_events);
4776 }
4777
4778 int cmd_trace(int argc, const char **argv)
4779 {
4780 const char *trace_usage[] = {
4781 "perf trace [<options>] [<command>]",
4782 "perf trace [<options>] -- <command> [<options>]",
4783 "perf trace record [<options>] [<command>]",
4784 "perf trace record [<options>] -- <command> [<options>]",
4785 NULL
4786 };
4787 struct trace trace = {
4788 .opts = {
4789 .target = {
4790 .uid = UINT_MAX,
4791 .uses_mmap = true,
4792 },
4793 .user_freq = UINT_MAX,
4794 .user_interval = ULLONG_MAX,
4795 .no_buffering = true,
4796 .mmap_pages = UINT_MAX,
4797 },
4798 .output = stderr,
4799 .show_comm = true,
4800 .show_tstamp = true,
4801 .show_duration = true,
4802 .show_arg_names = true,
4803 .args_alignment = 70,
4804 .trace_syscalls = false,
4805 .kernel_syscallchains = false,
4806 .max_stack = UINT_MAX,
4807 .max_events = ULONG_MAX,
4808 };
4809 const char *map_dump_str = NULL;
4810 const char *output_name = NULL;
4811 const struct option trace_options[] = {
4812 OPT_CALLBACK('e', "event", &trace, "event",
4813 "event/syscall selector. use 'perf list' to list available events",
4814 trace__parse_events_option),
4815 OPT_CALLBACK(0, "filter", &trace.evlist, "filter",
4816 "event filter", parse_filter),
4817 OPT_BOOLEAN(0, "comm", &trace.show_comm,
4818 "show the thread COMM next to its id"),
4819 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
4820 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4821 trace__parse_events_option),
4822 OPT_STRING('o', "output", &output_name, "file", "output file name"),
4823 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
4824 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4825 "trace events on existing process id"),
4826 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
4827 "trace events on existing thread id"),
4828 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
4829 "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
4830 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
4831 "system-wide collection from all CPUs"),
4832 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
4833 "list of cpus to monitor"),
4834 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
4835 "child tasks do not inherit counters"),
4836 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
4837 "number of mmap data pages", evlist__parse_mmap_pages),
4838 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
4839 "user to profile"),
4840 OPT_CALLBACK(0, "duration", &trace, "float",
4841 "show only events with duration > N.M ms",
4842 trace__set_duration),
4843 #ifdef HAVE_LIBBPF_SUPPORT
4844 OPT_STRING(0, "map-dump", &map_dump_str, "BPF map", "BPF map to periodically dump"),
4845 #endif
4846 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
4847 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4848 OPT_BOOLEAN('T', "time", &trace.full_time,
4849 "Show full timestamp, not time relative to first start"),
4850 OPT_BOOLEAN(0, "failure", &trace.failure_only,
4851 "Show only syscalls that failed"),
4852 OPT_BOOLEAN('s', "summary", &trace.summary_only,
4853 "Show only syscall summary with statistics"),
4854 OPT_BOOLEAN('S', "with-summary", &trace.summary,
4855 "Show all syscalls and summary with statistics"),
4856 OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary,
4857 "Show errno stats per syscall, use with -s or -S"),
4858 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4859 "Trace pagefaults", parse_pagefaults, "maj"),
4860 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
4861 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
4862 OPT_CALLBACK(0, "call-graph", &trace.opts,
4863 "record_mode[,record_size]", record_callchain_help,
4864 &record_parse_callchain_opt),
4865 OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,
4866 "Use libtraceevent to print the tracepoint arguments."),
4867 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4868 "Show the kernel callchains on the syscall exit path"),
4869 OPT_ULONG(0, "max-events", &trace.max_events,
4870 "Set the maximum number of events to print, exit after that is reached. "),
4871 OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4872 "Set the minimum stack depth when parsing the callchain, "
4873 "anything below the specified depth will be ignored."),
4874 OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4875 "Set the maximum stack depth when parsing the callchain, "
4876 "anything beyond the specified depth will be ignored. "
4877 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
4878 OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4879 "Sort batch of events before processing, use if getting out of order events"),
4880 OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4881 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
4882 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
4883 "per thread proc mmap processing timeout in ms"),
4884 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4885 trace__parse_cgroups),
4886 OPT_INTEGER('D', "delay", &trace.opts.initial_delay,
4887 "ms to wait before starting measurement after program "
4888 "start"),
4889 OPTS_EVSWITCH(&trace.evswitch),
4890 OPT_END()
4891 };
4892 bool __maybe_unused max_stack_user_set = true;
4893 bool mmap_pages_user_set = true;
4894 struct evsel *evsel;
4895 const char * const trace_subcommands[] = { "record", NULL };
4896 int err = -1;
4897 char bf[BUFSIZ];
4898 struct sigaction sigchld_act;
4899
4900 signal(SIGSEGV, sighandler_dump_stack);
4901 signal(SIGFPE, sighandler_dump_stack);
4902 signal(SIGINT, sighandler_interrupt);
4903
4904 memset(&sigchld_act, 0, sizeof(sigchld_act));
4905 sigchld_act.sa_flags = SA_SIGINFO;
4906 sigchld_act.sa_sigaction = sighandler_chld;
4907 sigaction(SIGCHLD, &sigchld_act, NULL);
4908
4909 trace.evlist = evlist__new();
4910 trace.sctbl = syscalltbl__new();
4911
4912 if (trace.evlist == NULL || trace.sctbl == NULL) {
4913 pr_err("Not enough memory to run!\n");
4914 err = -ENOMEM;
4915 goto out;
4916 }
4917
4918 /*
4919 * Parsing .perfconfig may entail creating a BPF event, that may need
4920 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4921 * is too small. This affects just this process, not touching the
4922 * global setting. If it fails we'll get something in 'perf trace -v'
4923 * to help diagnose the problem.
4924 */
4925 rlimit__bump_memlock();
4926
4927 err = perf_config(trace__config, &trace);
4928 if (err)
4929 goto out;
4930
4931 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4932 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
4933
4934 /*
4935 * Here we already passed thru trace__parse_events_option() and it has
4936 * already figured out if -e syscall_name, if not but if --event
4937 * foo:bar was used, the user is interested _just_ in those, say,
4938 * tracepoint events, not in the strace-like syscall-name-based mode.
4939 *
4940 * This is important because we need to check if strace-like mode is
4941 * needed to decided if we should filter out the eBPF
4942 * __augmented_syscalls__ code, if it is in the mix, say, via
4943 * .perfconfig trace.add_events, and filter those out.
4944 */
4945 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4946 trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4947 trace.trace_syscalls = true;
4948 }
4949 /*
4950 * Now that we have --verbose figured out, lets see if we need to parse
4951 * events from .perfconfig, so that if those events fail parsing, say some
4952 * BPF program fails, then we'll be able to use --verbose to see what went
4953 * wrong in more detail.
4954 */
4955 if (trace.perfconfig_events != NULL) {
4956 struct parse_events_error parse_err;
4957
4958 parse_events_error__init(&parse_err);
4959 err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);
4960 if (err)
4961 parse_events_error__print(&parse_err, trace.perfconfig_events);
4962 parse_events_error__exit(&parse_err);
4963 if (err)
4964 goto out;
4965 }
4966
4967 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4968 usage_with_options_msg(trace_usage, trace_options,
4969 "cgroup monitoring only available in system-wide mode");
4970 }
4971
4972 evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
4973 if (IS_ERR(evsel)) {
4974 bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
4975 pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
4976 goto out;
4977 }
4978
4979 if (evsel) {
4980 trace.syscalls.events.augmented = evsel;
4981
4982 evsel = evlist__find_tracepoint_by_name(trace.evlist, "raw_syscalls:sys_enter");
4983 if (evsel == NULL) {
4984 pr_err("ERROR: raw_syscalls:sys_enter not found in the augmented BPF object\n");
4985 goto out;
4986 }
4987
4988 if (evsel->bpf_obj == NULL) {
4989 pr_err("ERROR: raw_syscalls:sys_enter not associated to a BPF object\n");
4990 goto out;
4991 }
4992
4993 trace.bpf_obj = evsel->bpf_obj;
4994
4995 /*
4996 * If we have _just_ the augmenter event but don't have a
4997 * explicit --syscalls, then assume we want all strace-like
4998 * syscalls:
4999 */
5000 if (!trace.trace_syscalls && trace__only_augmented_syscalls_evsels(&trace))
5001 trace.trace_syscalls = true;
5002 /*
5003 * So, if we have a syscall augmenter, but trace_syscalls, aka
5004 * strace-like syscall tracing is not set, then we need to trow
5005 * away the augmenter, i.e. all the events that were created
5006 * from that BPF object file.
5007 *
5008 * This is more to fix the current .perfconfig trace.add_events
5009 * style of setting up the strace-like eBPF based syscall point
5010 * payload augmenter.
5011 *
5012 * All this complexity will be avoided by adding an alternative
5013 * to trace.add_events in the form of
5014 * trace.bpf_augmented_syscalls, that will be only parsed if we
5015 * need it.
5016 *
5017 * .perfconfig trace.add_events is still useful if we want, for
5018 * instance, have msr_write.msr in some .perfconfig profile based
5019 * 'perf trace --config determinism.profile' mode, where for some
5020 * particular goal/workload type we want a set of events and
5021 * output mode (with timings, etc) instead of having to add
5022 * all via the command line.
5023 *
5024 * Also --config to specify an alternate .perfconfig file needs
5025 * to be implemented.
5026 */
5027 if (!trace.trace_syscalls) {
5028 trace__delete_augmented_syscalls(&trace);
5029 } else {
5030 trace__set_bpf_map_filtered_pids(&trace);
5031 trace__set_bpf_map_syscalls(&trace);
5032 trace.syscalls.unaugmented_prog = trace__find_bpf_program_by_title(&trace, "!raw_syscalls:unaugmented");
5033 }
5034 }
5035
5036 err = bpf__setup_stdout(trace.evlist);
5037 if (err) {
5038 bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
5039 pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
5040 goto out;
5041 }
5042
5043 err = -1;
5044
5045 if (map_dump_str) {
5046 trace.dump.map = trace__find_bpf_map_by_name(&trace, map_dump_str);
5047 if (trace.dump.map == NULL) {
5048 pr_err("ERROR: BPF map \"%s\" not found\n", map_dump_str);
5049 goto out;
5050 }
5051 }
5052
5053 if (trace.trace_pgfaults) {
5054 trace.opts.sample_address = true;
5055 trace.opts.sample_time = true;
5056 }
5057
5058 if (trace.opts.mmap_pages == UINT_MAX)
5059 mmap_pages_user_set = false;
5060
5061 if (trace.max_stack == UINT_MAX) {
5062 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
5063 max_stack_user_set = false;
5064 }
5065
5066 #ifdef HAVE_DWARF_UNWIND_SUPPORT
5067 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
5068 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
5069 }
5070 #endif
5071
5072 if (callchain_param.enabled) {
5073 if (!mmap_pages_user_set && geteuid() == 0)
5074 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
5075
5076 symbol_conf.use_callchain = true;
5077 }
5078
5079 if (trace.evlist->core.nr_entries > 0) {
5080 evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);
5081 if (evlist__set_syscall_tp_fields(trace.evlist)) {
5082 perror("failed to set syscalls:* tracepoint fields");
5083 goto out;
5084 }
5085 }
5086
5087 if (trace.sort_events) {
5088 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
5089 ordered_events__set_copy_on_queue(&trace.oe.data, true);
5090 }
5091
5092 /*
5093 * If we are augmenting syscalls, then combine what we put in the
5094 * __augmented_syscalls__ BPF map with what is in the
5095 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
5096 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
5097 *
5098 * We'll switch to look at two BPF maps, one for sys_enter and the
5099 * other for sys_exit when we start augmenting the sys_exit paths with
5100 * buffers that are being copied from kernel to userspace, think 'read'
5101 * syscall.
5102 */
5103 if (trace.syscalls.events.augmented) {
5104 evlist__for_each_entry(trace.evlist, evsel) {
5105 bool raw_syscalls_sys_exit = strcmp(evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
5106
5107 if (raw_syscalls_sys_exit) {
5108 trace.raw_augmented_syscalls = true;
5109 goto init_augmented_syscall_tp;
5110 }
5111
5112 if (trace.syscalls.events.augmented->priv == NULL &&
5113 strstr(evsel__name(evsel), "syscalls:sys_enter")) {
5114 struct evsel *augmented = trace.syscalls.events.augmented;
5115 if (evsel__init_augmented_syscall_tp(augmented, evsel) ||
5116 evsel__init_augmented_syscall_tp_args(augmented))
5117 goto out;
5118 /*
5119 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
5120 * Above we made sure we can get from the payload the tp fields
5121 * that we get from syscalls:sys_enter tracefs format file.
5122 */
5123 augmented->handler = trace__sys_enter;
5124 /*
5125 * Now we do the same for the *syscalls:sys_enter event so that
5126 * if we handle it directly, i.e. if the BPF prog returns 0 so
5127 * as not to filter it, then we'll handle it just like we would
5128 * for the BPF_OUTPUT one:
5129 */
5130 if (evsel__init_augmented_syscall_tp(evsel, evsel) ||
5131 evsel__init_augmented_syscall_tp_args(evsel))
5132 goto out;
5133 evsel->handler = trace__sys_enter;
5134 }
5135
5136 if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) {
5137 struct syscall_tp *sc;
5138 init_augmented_syscall_tp:
5139 if (evsel__init_augmented_syscall_tp(evsel, evsel))
5140 goto out;
5141 sc = __evsel__syscall_tp(evsel);
5142 /*
5143 * For now with BPF raw_augmented we hook into
5144 * raw_syscalls:sys_enter and there we get all
5145 * 6 syscall args plus the tracepoint common
5146 * fields and the syscall_nr (another long).
5147 * So we check if that is the case and if so
5148 * don't look after the sc->args_size but
5149 * always after the full raw_syscalls:sys_enter
5150 * payload, which is fixed.
5151 *
5152 * We'll revisit this later to pass
5153 * s->args_size to the BPF augmenter (now
5154 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
5155 * so that it copies only what we need for each
5156 * syscall, like what happens when we use
5157 * syscalls:sys_enter_NAME, so that we reduce
5158 * the kernel/userspace traffic to just what is
5159 * needed for each syscall.
5160 */
5161 if (trace.raw_augmented_syscalls)
5162 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
5163 evsel__init_augmented_syscall_tp_ret(evsel);
5164 evsel->handler = trace__sys_exit;
5165 }
5166 }
5167 }
5168
5169 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
5170 return trace__record(&trace, argc-1, &argv[1]);
5171
5172 /* Using just --errno-summary will trigger --summary */
5173 if (trace.errno_summary && !trace.summary && !trace.summary_only)
5174 trace.summary_only = true;
5175
5176 /* summary_only implies summary option, but don't overwrite summary if set */
5177 if (trace.summary_only)
5178 trace.summary = trace.summary_only;
5179
5180 if (output_name != NULL) {
5181 err = trace__open_output(&trace, output_name);
5182 if (err < 0) {
5183 perror("failed to create output file");
5184 goto out;
5185 }
5186 }
5187
5188 err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
5189 if (err)
5190 goto out_close;
5191
5192 err = target__validate(&trace.opts.target);
5193 if (err) {
5194 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5195 fprintf(trace.output, "%s", bf);
5196 goto out_close;
5197 }
5198
5199 err = target__parse_uid(&trace.opts.target);
5200 if (err) {
5201 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5202 fprintf(trace.output, "%s", bf);
5203 goto out_close;
5204 }
5205
5206 if (!argc && target__none(&trace.opts.target))
5207 trace.opts.target.system_wide = true;
5208
5209 if (input_name)
5210 err = trace__replay(&trace);
5211 else
5212 err = trace__run(&trace, argc, argv);
5213
5214 out_close:
5215 if (output_name != NULL)
5216 fclose(trace.output);
5217 out:
5218 trace__exit(&trace);
5219 return err;
5220 }
5221