?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
accel/ivpu: Enable HWS by default on all platforms
[drm/drm-misc.git] / tools / perf / builtin-trace.c
blob6a1a128fe645014d0347ad4ec3e0c9e77ec59aee
1 /*
2 * builtin-trace.c
3 *
4 * Builtin 'trace' command:
5 *
6 * Display a continuously updated trace of any workload, CPU, specific PID,
7 * system wide, etc. Default format is loosely strace like, but any other
8 * event may be specified using --event.
9 *
10 * Copyright (C) 2012, 2013, 2014, 2015 Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
11 *
12 * Initially based on the 'trace' prototype by Thomas Gleixner:
13 *
14 * http://lwn.net/Articles/415728/ ("Announcing a new utility: 'trace'")
15 */
16
17 #include "util/record.h"
18 #include <api/fs/tracing_path.h>
19 #ifdef HAVE_LIBBPF_SUPPORT
20 #include <bpf/bpf.h>
21 #include <bpf/libbpf.h>
22 #include <bpf/btf.h>
23 #ifdef HAVE_BPF_SKEL
24 #include "bpf_skel/augmented_raw_syscalls.skel.h"
25 #endif
26 #endif
27 #include "util/bpf_map.h"
28 #include "util/rlimit.h"
29 #include "builtin.h"
30 #include "util/cgroup.h"
31 #include "util/color.h"
32 #include "util/config.h"
33 #include "util/debug.h"
34 #include "util/dso.h"
35 #include "util/env.h"
36 #include "util/event.h"
37 #include "util/evsel.h"
38 #include "util/evsel_fprintf.h"
39 #include "util/synthetic-events.h"
40 #include "util/evlist.h"
41 #include "util/evswitch.h"
42 #include "util/mmap.h"
43 #include <subcmd/pager.h>
44 #include <subcmd/exec-cmd.h>
45 #include "util/machine.h"
46 #include "util/map.h"
47 #include "util/symbol.h"
48 #include "util/path.h"
49 #include "util/session.h"
50 #include "util/thread.h"
51 #include <subcmd/parse-options.h>
52 #include "util/strlist.h"
53 #include "util/intlist.h"
54 #include "util/thread_map.h"
55 #include "util/stat.h"
56 #include "util/tool.h"
57 #include "util/util.h"
58 #include "trace/beauty/beauty.h"
59 #include "trace-event.h"
60 #include "util/parse-events.h"
61 #include "util/tracepoint.h"
62 #include "callchain.h"
63 #include "print_binary.h"
64 #include "string2.h"
65 #include "syscalltbl.h"
66 #include "rb_resort.h"
67 #include "../perf.h"
68 #include "trace_augment.h"
69
70 #include <errno.h>
71 #include <inttypes.h>
72 #include <poll.h>
73 #include <signal.h>
74 #include <stdlib.h>
75 #include <string.h>
76 #include <linux/err.h>
77 #include <linux/filter.h>
78 #include <linux/kernel.h>
79 #include <linux/list_sort.h>
80 #include <linux/random.h>
81 #include <linux/stringify.h>
82 #include <linux/time64.h>
83 #include <linux/zalloc.h>
84 #include <fcntl.h>
85 #include <sys/sysmacros.h>
86
87 #include <linux/ctype.h>
88 #include <perf/mmap.h>
89
90 #ifdef HAVE_LIBTRACEEVENT
91 #include <event-parse.h>
92 #endif
93
94 #ifndef O_CLOEXEC
95 # define O_CLOEXEC 02000000
96 #endif
97
98 #ifndef F_LINUX_SPECIFIC_BASE
99 # define F_LINUX_SPECIFIC_BASE 1024
100 #endif
101
102 #define RAW_SYSCALL_ARGS_NUM 6
103
104 /*
105 * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100
106 *
107 * We have to explicitely mark the direction of the flow of data, if from the
108 * kernel to user space or the other way around, since the BPF collector we
109 * have so far copies only from user to kernel space, mark the arguments that
110 * go that direction, so that we don´t end up collecting the previous contents
111 * for syscall args that goes from kernel to user space.
112 */
113 struct syscall_arg_fmt {
114 size_t (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
115 bool (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val);
116 unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
117 void *parm;
118 const char *name;
119 u16 nr_entries; // for arrays
120 bool from_user;
121 bool show_zero;
122 #ifdef HAVE_LIBBPF_SUPPORT
123 const struct btf_type *type;
124 int type_id; /* used in btf_dump */
125 #endif
126 };
127
128 struct syscall_fmt {
129 const char *name;
130 const char *alias;
131 struct {
132 const char *sys_enter,
133 *sys_exit;
134 } bpf_prog_name;
135 struct syscall_arg_fmt arg[RAW_SYSCALL_ARGS_NUM];
136 u8 nr_args;
137 bool errpid;
138 bool timeout;
139 bool hexret;
140 };
141
142 struct trace {
143 struct perf_tool tool;
144 struct syscalltbl *sctbl;
145 struct {
146 struct syscall *table;
147 struct {
148 struct evsel *sys_enter,
149 *sys_exit,
150 *bpf_output;
151 } events;
152 } syscalls;
153 #ifdef HAVE_BPF_SKEL
154 struct augmented_raw_syscalls_bpf *skel;
155 #endif
156 #ifdef HAVE_LIBBPF_SUPPORT
157 struct btf *btf;
158 #endif
159 struct record_opts opts;
160 struct evlist *evlist;
161 struct machine *host;
162 struct thread *current;
163 struct cgroup *cgroup;
164 u64 base_time;
165 FILE *output;
166 unsigned long nr_events;
167 unsigned long nr_events_printed;
168 unsigned long max_events;
169 struct evswitch evswitch;
170 struct strlist *ev_qualifier;
171 struct {
172 size_t nr;
173 int *entries;
174 } ev_qualifier_ids;
175 struct {
176 size_t nr;
177 pid_t *entries;
178 struct bpf_map *map;
179 } filter_pids;
180 double duration_filter;
181 double runtime_ms;
182 struct {
183 u64 vfs_getname,
184 proc_getname;
185 } stats;
186 unsigned int max_stack;
187 unsigned int min_stack;
188 int raw_augmented_syscalls_args_size;
189 bool raw_augmented_syscalls;
190 bool fd_path_disabled;
191 bool sort_events;
192 bool not_ev_qualifier;
193 bool live;
194 bool full_time;
195 bool sched;
196 bool multiple_threads;
197 bool summary;
198 bool summary_only;
199 bool errno_summary;
200 bool failure_only;
201 bool show_comm;
202 bool print_sample;
203 bool show_tool_stats;
204 bool trace_syscalls;
205 bool libtraceevent_print;
206 bool kernel_syscallchains;
207 s16 args_alignment;
208 bool show_tstamp;
209 bool show_duration;
210 bool show_zeros;
211 bool show_arg_names;
212 bool show_string_prefix;
213 bool force;
214 bool vfs_getname;
215 bool force_btf;
216 int trace_pgfaults;
217 char *perfconfig_events;
218 struct {
219 struct ordered_events data;
220 u64 last;
221 } oe;
222 };
223
224 static void trace__load_vmlinux_btf(struct trace *trace __maybe_unused)
225 {
226 #ifdef HAVE_LIBBPF_SUPPORT
227 if (trace->btf != NULL)
228 return;
229
230 trace->btf = btf__load_vmlinux_btf();
231 if (verbose > 0) {
232 fprintf(trace->output, trace->btf ? "vmlinux BTF loaded\n" :
233 "Failed to load vmlinux BTF\n");
234 }
235 #endif
236 }
237
238 struct tp_field {
239 int offset;
240 union {
241 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
242 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
243 };
244 };
245
246 #define TP_UINT_FIELD(bits) \
247 static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
248 { \
249 u##bits value; \
250 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
251 return value; \
252 }
253
254 TP_UINT_FIELD(8);
255 TP_UINT_FIELD(16);
256 TP_UINT_FIELD(32);
257 TP_UINT_FIELD(64);
258
259 #define TP_UINT_FIELD__SWAPPED(bits) \
260 static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
261 { \
262 u##bits value; \
263 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
264 return bswap_##bits(value);\
265 }
266
267 TP_UINT_FIELD__SWAPPED(16);
268 TP_UINT_FIELD__SWAPPED(32);
269 TP_UINT_FIELD__SWAPPED(64);
270
271 static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
272 {
273 field->offset = offset;
274
275 switch (size) {
276 case 1:
277 field->integer = tp_field__u8;
278 break;
279 case 2:
280 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
281 break;
282 case 4:
283 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
284 break;
285 case 8:
286 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
287 break;
288 default:
289 return -1;
290 }
291
292 return 0;
293 }
294
295 static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
296 {
297 return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
298 }
299
300 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
301 {
302 return sample->raw_data + field->offset;
303 }
304
305 static int __tp_field__init_ptr(struct tp_field *field, int offset)
306 {
307 field->offset = offset;
308 field->pointer = tp_field__ptr;
309 return 0;
310 }
311
312 static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
313 {
314 return __tp_field__init_ptr(field, format_field->offset);
315 }
316
317 struct syscall_tp {
318 struct tp_field id;
319 union {
320 struct tp_field args, ret;
321 };
322 };
323
324 /*
325 * The evsel->priv as used by 'perf trace'
326 * sc: for raw_syscalls:sys_{enter,exit} and syscalls:sys_{enter,exit}_SYSCALLNAME
327 * fmt: for all the other tracepoints
328 */
329 struct evsel_trace {
330 struct syscall_tp sc;
331 struct syscall_arg_fmt *fmt;
332 };
333
334 static struct evsel_trace *evsel_trace__new(void)
335 {
336 return zalloc(sizeof(struct evsel_trace));
337 }
338
339 static void evsel_trace__delete(struct evsel_trace *et)
340 {
341 if (et == NULL)
342 return;
343
344 zfree(&et->fmt);
345 free(et);
346 }
347
348 /*
349 * Used with raw_syscalls:sys_{enter,exit} and with the
350 * syscalls:sys_{enter,exit}_SYSCALL tracepoints
351 */
352 static inline struct syscall_tp *__evsel__syscall_tp(struct evsel *evsel)
353 {
354 struct evsel_trace *et = evsel->priv;
355
356 return &et->sc;
357 }
358
359 static struct syscall_tp *evsel__syscall_tp(struct evsel *evsel)
360 {
361 if (evsel->priv == NULL) {
362 evsel->priv = evsel_trace__new();
363 if (evsel->priv == NULL)
364 return NULL;
365 }
366
367 return __evsel__syscall_tp(evsel);
368 }
369
370 /*
371 * Used with all the other tracepoints.
372 */
373 static inline struct syscall_arg_fmt *__evsel__syscall_arg_fmt(struct evsel *evsel)
374 {
375 struct evsel_trace *et = evsel->priv;
376
377 return et->fmt;
378 }
379
380 static struct syscall_arg_fmt *evsel__syscall_arg_fmt(struct evsel *evsel)
381 {
382 struct evsel_trace *et = evsel->priv;
383
384 if (evsel->priv == NULL) {
385 et = evsel->priv = evsel_trace__new();
386
387 if (et == NULL)
388 return NULL;
389 }
390
391 if (et->fmt == NULL) {
392 et->fmt = calloc(evsel->tp_format->format.nr_fields, sizeof(struct syscall_arg_fmt));
393 if (et->fmt == NULL)
394 goto out_delete;
395 }
396
397 return __evsel__syscall_arg_fmt(evsel);
398
399 out_delete:
400 evsel_trace__delete(evsel->priv);
401 evsel->priv = NULL;
402 return NULL;
403 }
404
405 static int evsel__init_tp_uint_field(struct evsel *evsel, struct tp_field *field, const char *name)
406 {
407 struct tep_format_field *format_field = evsel__field(evsel, name);
408
409 if (format_field == NULL)
410 return -1;
411
412 return tp_field__init_uint(field, format_field, evsel->needs_swap);
413 }
414
415 #define perf_evsel__init_sc_tp_uint_field(evsel, name) \
416 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
417 evsel__init_tp_uint_field(evsel, &sc->name, #name); })
418
419 static int evsel__init_tp_ptr_field(struct evsel *evsel, struct tp_field *field, const char *name)
420 {
421 struct tep_format_field *format_field = evsel__field(evsel, name);
422
423 if (format_field == NULL)
424 return -1;
425
426 return tp_field__init_ptr(field, format_field);
427 }
428
429 #define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
430 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
431 evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
432
433 static void evsel__delete_priv(struct evsel *evsel)
434 {
435 zfree(&evsel->priv);
436 evsel__delete(evsel);
437 }
438
439 static int evsel__init_syscall_tp(struct evsel *evsel)
440 {
441 struct syscall_tp *sc = evsel__syscall_tp(evsel);
442
443 if (sc != NULL) {
444 if (evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
445 evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
446 return -ENOENT;
447
448 return 0;
449 }
450
451 return -ENOMEM;
452 }
453
454 static int evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)
455 {
456 struct syscall_tp *sc = evsel__syscall_tp(evsel);
457
458 if (sc != NULL) {
459 struct tep_format_field *syscall_id = evsel__field(tp, "id");
460 if (syscall_id == NULL)
461 syscall_id = evsel__field(tp, "__syscall_nr");
462 if (syscall_id == NULL ||
463 __tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
464 return -EINVAL;
465
466 return 0;
467 }
468
469 return -ENOMEM;
470 }
471
472 static int evsel__init_augmented_syscall_tp_args(struct evsel *evsel)
473 {
474 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
475
476 return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
477 }
478
479 static int evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)
480 {
481 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
482
483 return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
484 }
485
486 static int evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)
487 {
488 if (evsel__syscall_tp(evsel) != NULL) {
489 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
490 return -ENOENT;
491
492 evsel->handler = handler;
493 return 0;
494 }
495
496 return -ENOMEM;
497 }
498
499 static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
500 {
501 struct evsel *evsel = evsel__newtp("raw_syscalls", direction);
502
503 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
504 if (IS_ERR(evsel))
505 evsel = evsel__newtp("syscalls", direction);
506
507 if (IS_ERR(evsel))
508 return NULL;
509
510 if (evsel__init_raw_syscall_tp(evsel, handler))
511 goto out_delete;
512
513 return evsel;
514
515 out_delete:
516 evsel__delete_priv(evsel);
517 return NULL;
518 }
519
520 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
521 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
522 fields->name.integer(&fields->name, sample); })
523
524 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
525 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
526 fields->name.pointer(&fields->name, sample); })
527
528 size_t strarray__scnprintf_suffix(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_suffix, int val)
529 {
530 int idx = val - sa->offset;
531
532 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
533 size_t printed = scnprintf(bf, size, intfmt, val);
534 if (show_suffix)
535 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
536 return printed;
537 }
538
539 return scnprintf(bf, size, "%s%s", sa->entries[idx], show_suffix ? sa->prefix : "");
540 }
541
542 size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
543 {
544 int idx = val - sa->offset;
545
546 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
547 size_t printed = scnprintf(bf, size, intfmt, val);
548 if (show_prefix)
549 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
550 return printed;
551 }
552
553 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
554 }
555
556 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
557 const char *intfmt,
558 struct syscall_arg *arg)
559 {
560 return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
561 }
562
563 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
564 struct syscall_arg *arg)
565 {
566 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
567 }
568
569 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
570
571 bool syscall_arg__strtoul_strarray(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
572 {
573 return strarray__strtoul(arg->parm, bf, size, ret);
574 }
575
576 bool syscall_arg__strtoul_strarray_flags(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
577 {
578 return strarray__strtoul_flags(arg->parm, bf, size, ret);
579 }
580
581 bool syscall_arg__strtoul_strarrays(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
582 {
583 return strarrays__strtoul(arg->parm, bf, size, ret);
584 }
585
586 size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)
587 {
588 return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);
589 }
590
591 size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
592 {
593 size_t printed;
594 int i;
595
596 for (i = 0; i < sas->nr_entries; ++i) {
597 struct strarray *sa = sas->entries[i];
598 int idx = val - sa->offset;
599
600 if (idx >= 0 && idx < sa->nr_entries) {
601 if (sa->entries[idx] == NULL)
602 break;
603 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
604 }
605 }
606
607 printed = scnprintf(bf, size, intfmt, val);
608 if (show_prefix)
609 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
610 return printed;
611 }
612
613 bool strarray__strtoul(struct strarray *sa, char *bf, size_t size, u64 *ret)
614 {
615 int i;
616
617 for (i = 0; i < sa->nr_entries; ++i) {
618 if (sa->entries[i] && strncmp(sa->entries[i], bf, size) == 0 && sa->entries[i][size] == '\0') {
619 *ret = sa->offset + i;
620 return true;
621 }
622 }
623
624 return false;
625 }
626
627 bool strarray__strtoul_flags(struct strarray *sa, char *bf, size_t size, u64 *ret)
628 {
629 u64 val = 0;
630 char *tok = bf, *sep, *end;
631
632 *ret = 0;
633
634 while (size != 0) {
635 int toklen = size;
636
637 sep = memchr(tok, '|', size);
638 if (sep != NULL) {
639 size -= sep - tok + 1;
640
641 end = sep - 1;
642 while (end > tok && isspace(*end))
643 --end;
644
645 toklen = end - tok + 1;
646 }
647
648 while (isspace(*tok))
649 ++tok;
650
651 if (isalpha(*tok) || *tok == '_') {
652 if (!strarray__strtoul(sa, tok, toklen, &val))
653 return false;
654 } else
655 val = strtoul(tok, NULL, 0);
656
657 *ret |= (1 << (val - 1));
658
659 if (sep == NULL)
660 break;
661 tok = sep + 1;
662 }
663
664 return true;
665 }
666
667 bool strarrays__strtoul(struct strarrays *sas, char *bf, size_t size, u64 *ret)
668 {
669 int i;
670
671 for (i = 0; i < sas->nr_entries; ++i) {
672 struct strarray *sa = sas->entries[i];
673
674 if (strarray__strtoul(sa, bf, size, ret))
675 return true;
676 }
677
678 return false;
679 }
680
681 size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
682 struct syscall_arg *arg)
683 {
684 return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
685 }
686
687 #ifndef AT_FDCWD
688 #define AT_FDCWD -100
689 #endif
690
691 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
692 struct syscall_arg *arg)
693 {
694 int fd = arg->val;
695 const char *prefix = "AT_FD";
696
697 if (fd == AT_FDCWD)
698 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
699
700 return syscall_arg__scnprintf_fd(bf, size, arg);
701 }
702
703 #define SCA_FDAT syscall_arg__scnprintf_fd_at
704
705 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
706 struct syscall_arg *arg);
707
708 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
709
710 size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
711 {
712 return scnprintf(bf, size, "%#lx", arg->val);
713 }
714
715 size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
716 {
717 if (arg->val == 0)
718 return scnprintf(bf, size, "NULL");
719 return syscall_arg__scnprintf_hex(bf, size, arg);
720 }
721
722 size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
723 {
724 return scnprintf(bf, size, "%d", arg->val);
725 }
726
727 size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
728 {
729 return scnprintf(bf, size, "%ld", arg->val);
730 }
731
732 static size_t syscall_arg__scnprintf_char_array(char *bf, size_t size, struct syscall_arg *arg)
733 {
734 // XXX Hey, maybe for sched:sched_switch prev/next comm fields we can
735 // fill missing comms using thread__set_comm()...
736 // here or in a special syscall_arg__scnprintf_pid_sched_tp...
737 return scnprintf(bf, size, "\"%-.*s\"", arg->fmt->nr_entries ?: arg->len, arg->val);
738 }
739
740 #define SCA_CHAR_ARRAY syscall_arg__scnprintf_char_array
741
742 static const char *bpf_cmd[] = {
743 "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
744 "MAP_GET_NEXT_KEY", "PROG_LOAD", "OBJ_PIN", "OBJ_GET", "PROG_ATTACH",
745 "PROG_DETACH", "PROG_TEST_RUN", "PROG_GET_NEXT_ID", "MAP_GET_NEXT_ID",
746 "PROG_GET_FD_BY_ID", "MAP_GET_FD_BY_ID", "OBJ_GET_INFO_BY_FD",
747 "PROG_QUERY", "RAW_TRACEPOINT_OPEN", "BTF_LOAD", "BTF_GET_FD_BY_ID",
748 "TASK_FD_QUERY", "MAP_LOOKUP_AND_DELETE_ELEM", "MAP_FREEZE",
749 "BTF_GET_NEXT_ID", "MAP_LOOKUP_BATCH", "MAP_LOOKUP_AND_DELETE_BATCH",
750 "MAP_UPDATE_BATCH", "MAP_DELETE_BATCH", "LINK_CREATE", "LINK_UPDATE",
751 "LINK_GET_FD_BY_ID", "LINK_GET_NEXT_ID", "ENABLE_STATS", "ITER_CREATE",
752 "LINK_DETACH", "PROG_BIND_MAP",
753 };
754 static DEFINE_STRARRAY(bpf_cmd, "BPF_");
755
756 static const char *fsmount_flags[] = {
757 [1] = "CLOEXEC",
758 };
759 static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");
760
761 #include "trace/beauty/generated/fsconfig_arrays.c"
762
763 static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");
764
765 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
766 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
767
768 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
769 static DEFINE_STRARRAY(itimers, "ITIMER_");
770
771 static const char *keyctl_options[] = {
772 "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
773 "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
774 "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
775 "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
776 "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
777 };
778 static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
779
780 static const char *whences[] = { "SET", "CUR", "END",
781 #ifdef SEEK_DATA
782 "DATA",
783 #endif
784 #ifdef SEEK_HOLE
785 "HOLE",
786 #endif
787 };
788 static DEFINE_STRARRAY(whences, "SEEK_");
789
790 static const char *fcntl_cmds[] = {
791 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
792 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
793 "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
794 "GETOWNER_UIDS",
795 };
796 static DEFINE_STRARRAY(fcntl_cmds, "F_");
797
798 static const char *fcntl_linux_specific_cmds[] = {
799 "SETLEASE", "GETLEASE", "NOTIFY", "DUPFD_QUERY", [5] = "CANCELLK", "DUPFD_CLOEXEC",
800 "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
801 "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
802 };
803
804 static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
805
806 static struct strarray *fcntl_cmds_arrays[] = {
807 &strarray__fcntl_cmds,
808 &strarray__fcntl_linux_specific_cmds,
809 };
810
811 static DEFINE_STRARRAYS(fcntl_cmds_arrays);
812
813 static const char *rlimit_resources[] = {
814 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
815 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
816 "RTTIME",
817 };
818 static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
819
820 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
821 static DEFINE_STRARRAY(sighow, "SIG_");
822
823 static const char *clockid[] = {
824 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
825 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
826 "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
827 };
828 static DEFINE_STRARRAY(clockid, "CLOCK_");
829
830 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
831 struct syscall_arg *arg)
832 {
833 bool show_prefix = arg->show_string_prefix;
834 const char *suffix = "_OK";
835 size_t printed = 0;
836 int mode = arg->val;
837
838 if (mode == F_OK) /* 0 */
839 return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
840 #define P_MODE(n) \
841 if (mode & n##_OK) { \
842 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
843 mode &= ~n##_OK; \
844 }
845
846 P_MODE(R);
847 P_MODE(W);
848 P_MODE(X);
849 #undef P_MODE
850
851 if (mode)
852 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
853
854 return printed;
855 }
856
857 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
858
859 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
860 struct syscall_arg *arg);
861
862 #define SCA_FILENAME syscall_arg__scnprintf_filename
863
864 // 'argname' is just documentational at this point, to remove the previous comment with that info
865 #define SCA_FILENAME_FROM_USER(argname) \
866 { .scnprintf = SCA_FILENAME, \
867 .from_user = true, }
868
869 static size_t syscall_arg__scnprintf_buf(char *bf, size_t size, struct syscall_arg *arg);
870
871 #define SCA_BUF syscall_arg__scnprintf_buf
872
873 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
874 struct syscall_arg *arg)
875 {
876 bool show_prefix = arg->show_string_prefix;
877 const char *prefix = "O_";
878 int printed = 0, flags = arg->val;
879
880 #define P_FLAG(n) \
881 if (flags & O_##n) { \
882 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
883 flags &= ~O_##n; \
884 }
885
886 P_FLAG(CLOEXEC);
887 P_FLAG(NONBLOCK);
888 #undef P_FLAG
889
890 if (flags)
891 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
892
893 return printed;
894 }
895
896 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
897
898 #ifndef GRND_NONBLOCK
899 #define GRND_NONBLOCK 0x0001
900 #endif
901 #ifndef GRND_RANDOM
902 #define GRND_RANDOM 0x0002
903 #endif
904
905 static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
906 struct syscall_arg *arg)
907 {
908 bool show_prefix = arg->show_string_prefix;
909 const char *prefix = "GRND_";
910 int printed = 0, flags = arg->val;
911
912 #define P_FLAG(n) \
913 if (flags & GRND_##n) { \
914 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
915 flags &= ~GRND_##n; \
916 }
917
918 P_FLAG(RANDOM);
919 P_FLAG(NONBLOCK);
920 #undef P_FLAG
921
922 if (flags)
923 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
924
925 return printed;
926 }
927
928 #define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
929
930 #ifdef HAVE_LIBBPF_SUPPORT
931 static void syscall_arg_fmt__cache_btf_enum(struct syscall_arg_fmt *arg_fmt, struct btf *btf, char *type)
932 {
933 int id;
934
935 type = strstr(type, "enum ");
936 if (type == NULL)
937 return;
938
939 type += 5; // skip "enum " to get the enumeration name
940
941 id = btf__find_by_name(btf, type);
942 if (id < 0)
943 return;
944
945 arg_fmt->type = btf__type_by_id(btf, id);
946 }
947
948 static bool syscall_arg__strtoul_btf_enum(char *bf, size_t size, struct syscall_arg *arg, u64 *val)
949 {
950 const struct btf_type *bt = arg->fmt->type;
951 struct btf *btf = arg->trace->btf;
952 struct btf_enum *be = btf_enum(bt);
953
954 for (int i = 0; i < btf_vlen(bt); ++i, ++be) {
955 const char *name = btf__name_by_offset(btf, be->name_off);
956 int max_len = max(size, strlen(name));
957
958 if (strncmp(name, bf, max_len) == 0) {
959 *val = be->val;
960 return true;
961 }
962 }
963
964 return false;
965 }
966
967 static bool syscall_arg__strtoul_btf_type(char *bf, size_t size, struct syscall_arg *arg, u64 *val)
968 {
969 const struct btf_type *bt;
970 char *type = arg->type_name;
971 struct btf *btf;
972
973 trace__load_vmlinux_btf(arg->trace);
974
975 btf = arg->trace->btf;
976 if (btf == NULL)
977 return false;
978
979 if (arg->fmt->type == NULL) {
980 // See if this is an enum
981 syscall_arg_fmt__cache_btf_enum(arg->fmt, btf, type);
982 }
983
984 // Now let's see if we have a BTF type resolved
985 bt = arg->fmt->type;
986 if (bt == NULL)
987 return false;
988
989 // If it is an enum:
990 if (btf_is_enum(arg->fmt->type))
991 return syscall_arg__strtoul_btf_enum(bf, size, arg, val);
992
993 return false;
994 }
995
996 static size_t btf_enum_scnprintf(const struct btf_type *type, struct btf *btf, char *bf, size_t size, int val)
997 {
998 struct btf_enum *be = btf_enum(type);
999 const int nr_entries = btf_vlen(type);
1000
1001 for (int i = 0; i < nr_entries; ++i, ++be) {
1002 if (be->val == val) {
1003 return scnprintf(bf, size, "%s",
1004 btf__name_by_offset(btf, be->name_off));
1005 }
1006 }
1007
1008 return 0;
1009 }
1010
1011 struct trace_btf_dump_snprintf_ctx {
1012 char *bf;
1013 size_t printed, size;
1014 };
1015
1016 static void trace__btf_dump_snprintf(void *vctx, const char *fmt, va_list args)
1017 {
1018 struct trace_btf_dump_snprintf_ctx *ctx = vctx;
1019
1020 ctx->printed += vscnprintf(ctx->bf + ctx->printed, ctx->size - ctx->printed, fmt, args);
1021 }
1022
1023 static size_t btf_struct_scnprintf(const struct btf_type *type, struct btf *btf, char *bf, size_t size, struct syscall_arg *arg)
1024 {
1025 struct trace_btf_dump_snprintf_ctx ctx = {
1026 .bf = bf,
1027 .size = size,
1028 };
1029 struct augmented_arg *augmented_arg = arg->augmented.args;
1030 int type_id = arg->fmt->type_id, consumed;
1031 struct btf_dump *btf_dump;
1032
1033 LIBBPF_OPTS(btf_dump_opts, dump_opts);
1034 LIBBPF_OPTS(btf_dump_type_data_opts, dump_data_opts);
1035
1036 if (arg == NULL || arg->augmented.args == NULL)
1037 return 0;
1038
1039 dump_data_opts.compact = true;
1040 dump_data_opts.skip_names = !arg->trace->show_arg_names;
1041
1042 btf_dump = btf_dump__new(btf, trace__btf_dump_snprintf, &ctx, &dump_opts);
1043 if (btf_dump == NULL)
1044 return 0;
1045
1046 /* pretty print the struct data here */
1047 if (btf_dump__dump_type_data(btf_dump, type_id, arg->augmented.args->value, type->size, &dump_data_opts) == 0)
1048 return 0;
1049
1050 consumed = sizeof(*augmented_arg) + augmented_arg->size;
1051 arg->augmented.args = ((void *)arg->augmented.args) + consumed;
1052 arg->augmented.size -= consumed;
1053
1054 btf_dump__free(btf_dump);
1055
1056 return ctx.printed;
1057 }
1058
1059 static size_t trace__btf_scnprintf(struct trace *trace, struct syscall_arg *arg, char *bf,
1060 size_t size, int val, char *type)
1061 {
1062 struct syscall_arg_fmt *arg_fmt = arg->fmt;
1063
1064 if (trace->btf == NULL)
1065 return 0;
1066
1067 if (arg_fmt->type == NULL) {
1068 // Check if this is an enum and if we have the BTF type for it.
1069 syscall_arg_fmt__cache_btf_enum(arg_fmt, trace->btf, type);
1070 }
1071
1072 // Did we manage to find a BTF type for the syscall/tracepoint argument?
1073 if (arg_fmt->type == NULL)
1074 return 0;
1075
1076 if (btf_is_enum(arg_fmt->type))
1077 return btf_enum_scnprintf(arg_fmt->type, trace->btf, bf, size, val);
1078 else if (btf_is_struct(arg_fmt->type) || btf_is_union(arg_fmt->type))
1079 return btf_struct_scnprintf(arg_fmt->type, trace->btf, bf, size, arg);
1080
1081 return 0;
1082 }
1083
1084 #else // HAVE_LIBBPF_SUPPORT
1085 static size_t trace__btf_scnprintf(struct trace *trace __maybe_unused, struct syscall_arg *arg __maybe_unused,
1086 char *bf __maybe_unused, size_t size __maybe_unused, int val __maybe_unused,
1087 char *type __maybe_unused)
1088 {
1089 return 0;
1090 }
1091
1092 static bool syscall_arg__strtoul_btf_type(char *bf __maybe_unused, size_t size __maybe_unused,
1093 struct syscall_arg *arg __maybe_unused, u64 *val __maybe_unused)
1094 {
1095 return false;
1096 }
1097 #endif // HAVE_LIBBPF_SUPPORT
1098
1099 #define STUL_BTF_TYPE syscall_arg__strtoul_btf_type
1100
1101 #define STRARRAY(name, array) \
1102 { .scnprintf = SCA_STRARRAY, \
1103 .strtoul = STUL_STRARRAY, \
1104 .parm = &strarray__##array, }
1105
1106 #define STRARRAY_FLAGS(name, array) \
1107 { .scnprintf = SCA_STRARRAY_FLAGS, \
1108 .strtoul = STUL_STRARRAY_FLAGS, \
1109 .parm = &strarray__##array, }
1110
1111 #include "trace/beauty/arch_errno_names.c"
1112 #include "trace/beauty/eventfd.c"
1113 #include "trace/beauty/futex_op.c"
1114 #include "trace/beauty/futex_val3.c"
1115 #include "trace/beauty/mmap.c"
1116 #include "trace/beauty/mode_t.c"
1117 #include "trace/beauty/msg_flags.c"
1118 #include "trace/beauty/open_flags.c"
1119 #include "trace/beauty/perf_event_open.c"
1120 #include "trace/beauty/pid.c"
1121 #include "trace/beauty/sched_policy.c"
1122 #include "trace/beauty/seccomp.c"
1123 #include "trace/beauty/signum.c"
1124 #include "trace/beauty/socket_type.c"
1125 #include "trace/beauty/waitid_options.c"
1126
1127 static const struct syscall_fmt syscall_fmts[] = {
1128 { .name = "access",
1129 .arg = { [1] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, },
1130 { .name = "arch_prctl",
1131 .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
1132 [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
1133 { .name = "bind",
1134 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
1135 [1] = SCA_SOCKADDR_FROM_USER(umyaddr),
1136 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
1137 { .name = "bpf",
1138 .arg = { [0] = STRARRAY(cmd, bpf_cmd),
1139 [1] = { .from_user = true /* attr */, }, } },
1140 { .name = "brk", .hexret = true,
1141 .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
1142 { .name = "clock_gettime",
1143 .arg = { [0] = STRARRAY(clk_id, clockid), }, },
1144 { .name = "clock_nanosleep",
1145 .arg = { [2] = SCA_TIMESPEC_FROM_USER(req), }, },
1146 { .name = "clone", .errpid = true, .nr_args = 5,
1147 .arg = { [0] = { .name = "flags", .scnprintf = SCA_CLONE_FLAGS, },
1148 [1] = { .name = "child_stack", .scnprintf = SCA_HEX, },
1149 [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
1150 [3] = { .name = "child_tidptr", .scnprintf = SCA_HEX, },
1151 [4] = { .name = "tls", .scnprintf = SCA_HEX, }, }, },
1152 { .name = "close",
1153 .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
1154 { .name = "connect",
1155 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
1156 [1] = SCA_SOCKADDR_FROM_USER(servaddr),
1157 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
1158 { .name = "epoll_ctl",
1159 .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
1160 { .name = "eventfd2",
1161 .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
1162 { .name = "faccessat",
1163 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ },
1164 [1] = SCA_FILENAME_FROM_USER(pathname),
1165 [2] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, },
1166 { .name = "faccessat2",
1167 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ },
1168 [1] = SCA_FILENAME_FROM_USER(pathname),
1169 [2] = { .scnprintf = SCA_ACCMODE, /* mode */ },
1170 [3] = { .scnprintf = SCA_FACCESSAT2_FLAGS, /* flags */ }, }, },
1171 { .name = "fchmodat",
1172 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1173 { .name = "fchownat",
1174 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1175 { .name = "fcntl",
1176 .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
1177 .strtoul = STUL_STRARRAYS,
1178 .parm = &strarrays__fcntl_cmds_arrays,
1179 .show_zero = true, },
1180 [2] = { .scnprintf = SCA_FCNTL_ARG, /* arg */ }, }, },
1181 { .name = "flock",
1182 .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
1183 { .name = "fsconfig",
1184 .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },
1185 { .name = "fsmount",
1186 .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),
1187 [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },
1188 { .name = "fspick",
1189 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1190 [1] = SCA_FILENAME_FROM_USER(path),
1191 [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },
1192 { .name = "fstat", .alias = "newfstat", },
1193 { .name = "futex",
1194 .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
1195 [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
1196 { .name = "futimesat",
1197 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1198 { .name = "getitimer",
1199 .arg = { [0] = STRARRAY(which, itimers), }, },
1200 { .name = "getpid", .errpid = true, },
1201 { .name = "getpgid", .errpid = true, },
1202 { .name = "getppid", .errpid = true, },
1203 { .name = "getrandom",
1204 .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
1205 { .name = "getrlimit",
1206 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
1207 { .name = "getsockopt",
1208 .arg = { [1] = STRARRAY(level, socket_level), }, },
1209 { .name = "gettid", .errpid = true, },
1210 { .name = "ioctl",
1211 .arg = {
1212 #if defined(__i386__) || defined(__x86_64__)
1213 /*
1214 * FIXME: Make this available to all arches.
1215 */
1216 [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
1217 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
1218 #else
1219 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
1220 #endif
1221 { .name = "kcmp", .nr_args = 5,
1222 .arg = { [0] = { .name = "pid1", .scnprintf = SCA_PID, },
1223 [1] = { .name = "pid2", .scnprintf = SCA_PID, },
1224 [2] = { .name = "type", .scnprintf = SCA_KCMP_TYPE, },
1225 [3] = { .name = "idx1", .scnprintf = SCA_KCMP_IDX, },
1226 [4] = { .name = "idx2", .scnprintf = SCA_KCMP_IDX, }, }, },
1227 { .name = "keyctl",
1228 .arg = { [0] = STRARRAY(option, keyctl_options), }, },
1229 { .name = "kill",
1230 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1231 { .name = "linkat",
1232 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1233 { .name = "lseek",
1234 .arg = { [2] = STRARRAY(whence, whences), }, },
1235 { .name = "lstat", .alias = "newlstat", },
1236 { .name = "madvise",
1237 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1238 [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
1239 { .name = "mkdirat",
1240 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1241 { .name = "mknodat",
1242 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1243 { .name = "mmap", .hexret = true,
1244 /* The standard mmap maps to old_mmap on s390x */
1245 #if defined(__s390x__)
1246 .alias = "old_mmap",
1247 #endif
1248 .arg = { [2] = { .scnprintf = SCA_MMAP_PROT, .show_zero = true, /* prot */ },
1249 [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */
1250 .strtoul = STUL_STRARRAY_FLAGS,
1251 .parm = &strarray__mmap_flags, },
1252 [5] = { .scnprintf = SCA_HEX, /* offset */ }, }, },
1253 { .name = "mount",
1254 .arg = { [0] = SCA_FILENAME_FROM_USER(devname),
1255 [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
1256 .mask_val = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
1257 { .name = "move_mount",
1258 .arg = { [0] = { .scnprintf = SCA_FDAT, /* from_dfd */ },
1259 [1] = SCA_FILENAME_FROM_USER(pathname),
1260 [2] = { .scnprintf = SCA_FDAT, /* to_dfd */ },
1261 [3] = SCA_FILENAME_FROM_USER(pathname),
1262 [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },
1263 { .name = "mprotect",
1264 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1265 [2] = { .scnprintf = SCA_MMAP_PROT, .show_zero = true, /* prot */ }, }, },
1266 { .name = "mq_unlink",
1267 .arg = { [0] = SCA_FILENAME_FROM_USER(u_name), }, },
1268 { .name = "mremap", .hexret = true,
1269 .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
1270 { .name = "name_to_handle_at",
1271 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1272 { .name = "nanosleep",
1273 .arg = { [0] = SCA_TIMESPEC_FROM_USER(req), }, },
1274 { .name = "newfstatat", .alias = "fstatat",
1275 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ },
1276 [1] = SCA_FILENAME_FROM_USER(pathname),
1277 [3] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ }, }, },
1278 { .name = "open",
1279 .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1280 { .name = "open_by_handle_at",
1281 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1282 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1283 { .name = "openat",
1284 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1285 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1286 { .name = "perf_event_open",
1287 .arg = { [0] = SCA_PERF_ATTR_FROM_USER(attr),
1288 [2] = { .scnprintf = SCA_INT, /* cpu */ },
1289 [3] = { .scnprintf = SCA_FD, /* group_fd */ },
1290 [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
1291 { .name = "pipe2",
1292 .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
1293 { .name = "pkey_alloc",
1294 .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS, /* access_rights */ }, }, },
1295 { .name = "pkey_free",
1296 .arg = { [0] = { .scnprintf = SCA_INT, /* key */ }, }, },
1297 { .name = "pkey_mprotect",
1298 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1299 [2] = { .scnprintf = SCA_MMAP_PROT, .show_zero = true, /* prot */ },
1300 [3] = { .scnprintf = SCA_INT, /* pkey */ }, }, },
1301 { .name = "poll", .timeout = true, },
1302 { .name = "ppoll", .timeout = true, },
1303 { .name = "prctl",
1304 .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */
1305 .strtoul = STUL_STRARRAY,
1306 .parm = &strarray__prctl_options, },
1307 [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
1308 [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
1309 { .name = "pread", .alias = "pread64", },
1310 { .name = "preadv", .alias = "pread", },
1311 { .name = "prlimit64",
1312 .arg = { [1] = STRARRAY(resource, rlimit_resources),
1313 [2] = { .from_user = true /* new_rlim */, }, }, },
1314 { .name = "pwrite", .alias = "pwrite64", },
1315 { .name = "readlinkat",
1316 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1317 { .name = "recvfrom",
1318 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1319 { .name = "recvmmsg",
1320 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1321 { .name = "recvmsg",
1322 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1323 { .name = "renameat",
1324 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1325 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
1326 { .name = "renameat2",
1327 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1328 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
1329 [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
1330 { .name = "rseq", .errpid = true,
1331 .arg = { [0] = { .from_user = true /* rseq */, }, }, },
1332 { .name = "rt_sigaction",
1333 .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1334 { .name = "rt_sigprocmask",
1335 .arg = { [0] = STRARRAY(how, sighow), }, },
1336 { .name = "rt_sigqueueinfo",
1337 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1338 { .name = "rt_tgsigqueueinfo",
1339 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1340 { .name = "sched_setscheduler",
1341 .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
1342 { .name = "seccomp",
1343 .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP, /* op */ },
1344 [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
1345 { .name = "select", .timeout = true, },
1346 { .name = "sendfile", .alias = "sendfile64", },
1347 { .name = "sendmmsg",
1348 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1349 { .name = "sendmsg",
1350 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1351 { .name = "sendto",
1352 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
1353 [4] = SCA_SOCKADDR_FROM_USER(addr), }, },
1354 { .name = "set_robust_list", .errpid = true,
1355 .arg = { [0] = { .from_user = true /* head */, }, }, },
1356 { .name = "set_tid_address", .errpid = true, },
1357 { .name = "setitimer",
1358 .arg = { [0] = STRARRAY(which, itimers), }, },
1359 { .name = "setrlimit",
1360 .arg = { [0] = STRARRAY(resource, rlimit_resources),
1361 [1] = { .from_user = true /* rlim */, }, }, },
1362 { .name = "setsockopt",
1363 .arg = { [1] = STRARRAY(level, socket_level), }, },
1364 { .name = "socket",
1365 .arg = { [0] = STRARRAY(family, socket_families),
1366 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1367 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1368 { .name = "socketpair",
1369 .arg = { [0] = STRARRAY(family, socket_families),
1370 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1371 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1372 { .name = "stat", .alias = "newstat", },
1373 { .name = "statx",
1374 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fdat */ },
1375 [2] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ } ,
1376 [3] = { .scnprintf = SCA_STATX_MASK, /* mask */ }, }, },
1377 { .name = "swapoff",
1378 .arg = { [0] = SCA_FILENAME_FROM_USER(specialfile), }, },
1379 { .name = "swapon",
1380 .arg = { [0] = SCA_FILENAME_FROM_USER(specialfile), }, },
1381 { .name = "symlinkat",
1382 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1383 { .name = "sync_file_range",
1384 .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },
1385 { .name = "tgkill",
1386 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1387 { .name = "tkill",
1388 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1389 { .name = "umount2", .alias = "umount",
1390 .arg = { [0] = SCA_FILENAME_FROM_USER(name), }, },
1391 { .name = "uname", .alias = "newuname", },
1392 { .name = "unlinkat",
1393 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1394 [1] = SCA_FILENAME_FROM_USER(pathname),
1395 [2] = { .scnprintf = SCA_FS_AT_FLAGS, /* flags */ }, }, },
1396 { .name = "utimensat",
1397 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
1398 { .name = "wait4", .errpid = true,
1399 .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
1400 { .name = "waitid", .errpid = true,
1401 .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
1402 { .name = "write",
1403 .arg = { [1] = { .scnprintf = SCA_BUF /* buf */, .from_user = true, }, }, },
1404 };
1405
1406 static int syscall_fmt__cmp(const void *name, const void *fmtp)
1407 {
1408 const struct syscall_fmt *fmt = fmtp;
1409 return strcmp(name, fmt->name);
1410 }
1411
1412 static const struct syscall_fmt *__syscall_fmt__find(const struct syscall_fmt *fmts,
1413 const int nmemb,
1414 const char *name)
1415 {
1416 return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1417 }
1418
1419 static const struct syscall_fmt *syscall_fmt__find(const char *name)
1420 {
1421 const int nmemb = ARRAY_SIZE(syscall_fmts);
1422 return __syscall_fmt__find(syscall_fmts, nmemb, name);
1423 }
1424
1425 static const struct syscall_fmt *__syscall_fmt__find_by_alias(const struct syscall_fmt *fmts,
1426 const int nmemb, const char *alias)
1427 {
1428 int i;
1429
1430 for (i = 0; i < nmemb; ++i) {
1431 if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0)
1432 return &fmts[i];
1433 }
1434
1435 return NULL;
1436 }
1437
1438 static const struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
1439 {
1440 const int nmemb = ARRAY_SIZE(syscall_fmts);
1441 return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias);
1442 }
1443
1444 /*
1445 * is_exit: is this "exit" or "exit_group"?
1446 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
1447 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
1448 * nonexistent: Just a hole in the syscall table, syscall id not allocated
1449 */
1450 struct syscall {
1451 struct tep_event *tp_format;
1452 int nr_args;
1453 int args_size;
1454 struct {
1455 struct bpf_program *sys_enter,
1456 *sys_exit;
1457 } bpf_prog;
1458 bool is_exit;
1459 bool is_open;
1460 bool nonexistent;
1461 bool use_btf;
1462 struct tep_format_field *args;
1463 const char *name;
1464 const struct syscall_fmt *fmt;
1465 struct syscall_arg_fmt *arg_fmt;
1466 };
1467
1468 /*
1469 * We need to have this 'calculated' boolean because in some cases we really
1470 * don't know what is the duration of a syscall, for instance, when we start
1471 * a session and some threads are waiting for a syscall to finish, say 'poll',
1472 * in which case all we can do is to print "( ? ) for duration and for the
1473 * start timestamp.
1474 */
1475 static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
1476 {
1477 double duration = (double)t / NSEC_PER_MSEC;
1478 size_t printed = fprintf(fp, "(");
1479
1480 if (!calculated)
1481 printed += fprintf(fp, " ");
1482 else if (duration >= 1.0)
1483 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1484 else if (duration >= 0.01)
1485 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1486 else
1487 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
1488 return printed + fprintf(fp, "): ");
1489 }
1490
1491 /**
1492 * filename.ptr: The filename char pointer that will be vfs_getname'd
1493 * filename.entry_str_pos: Where to insert the string translated from
1494 * filename.ptr by the vfs_getname tracepoint/kprobe.
1495 * ret_scnprintf: syscall args may set this to a different syscall return
1496 * formatter, for instance, fcntl may return fds, file flags, etc.
1497 */
1498 struct thread_trace {
1499 u64 entry_time;
1500 bool entry_pending;
1501 unsigned long nr_events;
1502 unsigned long pfmaj, pfmin;
1503 char *entry_str;
1504 double runtime_ms;
1505 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1506 struct {
1507 unsigned long ptr;
1508 short int entry_str_pos;
1509 bool pending_open;
1510 unsigned int namelen;
1511 char *name;
1512 } filename;
1513 struct {
1514 int max;
1515 struct file *table;
1516 } files;
1517
1518 struct intlist *syscall_stats;
1519 };
1520
1521 static struct thread_trace *thread_trace__new(void)
1522 {
1523 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
1524
1525 if (ttrace) {
1526 ttrace->files.max = -1;
1527 ttrace->syscall_stats = intlist__new(NULL);
1528 }
1529
1530 return ttrace;
1531 }
1532
1533 static void thread_trace__free_files(struct thread_trace *ttrace);
1534
1535 static void thread_trace__delete(void *pttrace)
1536 {
1537 struct thread_trace *ttrace = pttrace;
1538
1539 if (!ttrace)
1540 return;
1541
1542 intlist__delete(ttrace->syscall_stats);
1543 ttrace->syscall_stats = NULL;
1544 thread_trace__free_files(ttrace);
1545 zfree(&ttrace->entry_str);
1546 free(ttrace);
1547 }
1548
1549 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1550 {
1551 struct thread_trace *ttrace;
1552
1553 if (thread == NULL)
1554 goto fail;
1555
1556 if (thread__priv(thread) == NULL)
1557 thread__set_priv(thread, thread_trace__new());
1558
1559 if (thread__priv(thread) == NULL)
1560 goto fail;
1561
1562 ttrace = thread__priv(thread);
1563 ++ttrace->nr_events;
1564
1565 return ttrace;
1566 fail:
1567 color_fprintf(fp, PERF_COLOR_RED,
1568 "WARNING: not enough memory, dropping samples!\n");
1569 return NULL;
1570 }
1571
1572
1573 void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
1574 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
1575 {
1576 struct thread_trace *ttrace = thread__priv(arg->thread);
1577
1578 ttrace->ret_scnprintf = ret_scnprintf;
1579 }
1580
1581 #define TRACE_PFMAJ (1 << 0)
1582 #define TRACE_PFMIN (1 << 1)
1583
1584 static const size_t trace__entry_str_size = 2048;
1585
1586 static void thread_trace__free_files(struct thread_trace *ttrace)
1587 {
1588 for (int i = 0; i < ttrace->files.max; ++i) {
1589 struct file *file = ttrace->files.table + i;
1590 zfree(&file->pathname);
1591 }
1592
1593 zfree(&ttrace->files.table);
1594 ttrace->files.max = -1;
1595 }
1596
1597 static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
1598 {
1599 if (fd < 0)
1600 return NULL;
1601
1602 if (fd > ttrace->files.max) {
1603 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
1604
1605 if (nfiles == NULL)
1606 return NULL;
1607
1608 if (ttrace->files.max != -1) {
1609 memset(nfiles + ttrace->files.max + 1, 0,
1610 (fd - ttrace->files.max) * sizeof(struct file));
1611 } else {
1612 memset(nfiles, 0, (fd + 1) * sizeof(struct file));
1613 }
1614
1615 ttrace->files.table = nfiles;
1616 ttrace->files.max = fd;
1617 }
1618
1619 return ttrace->files.table + fd;
1620 }
1621
1622 struct file *thread__files_entry(struct thread *thread, int fd)
1623 {
1624 return thread_trace__files_entry(thread__priv(thread), fd);
1625 }
1626
1627 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1628 {
1629 struct thread_trace *ttrace = thread__priv(thread);
1630 struct file *file = thread_trace__files_entry(ttrace, fd);
1631
1632 if (file != NULL) {
1633 struct stat st;
1634 if (stat(pathname, &st) == 0)
1635 file->dev_maj = major(st.st_rdev);
1636 file->pathname = strdup(pathname);
1637 if (file->pathname)
1638 return 0;
1639 }
1640
1641 return -1;
1642 }
1643
1644 static int thread__read_fd_path(struct thread *thread, int fd)
1645 {
1646 char linkname[PATH_MAX], pathname[PATH_MAX];
1647 struct stat st;
1648 int ret;
1649
1650 if (thread__pid(thread) == thread__tid(thread)) {
1651 scnprintf(linkname, sizeof(linkname),
1652 "/proc/%d/fd/%d", thread__pid(thread), fd);
1653 } else {
1654 scnprintf(linkname, sizeof(linkname),
1655 "/proc/%d/task/%d/fd/%d",
1656 thread__pid(thread), thread__tid(thread), fd);
1657 }
1658
1659 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1660 return -1;
1661
1662 ret = readlink(linkname, pathname, sizeof(pathname));
1663
1664 if (ret < 0 || ret > st.st_size)
1665 return -1;
1666
1667 pathname[ret] = '\0';
1668 return trace__set_fd_pathname(thread, fd, pathname);
1669 }
1670
1671 static const char *thread__fd_path(struct thread *thread, int fd,
1672 struct trace *trace)
1673 {
1674 struct thread_trace *ttrace = thread__priv(thread);
1675
1676 if (ttrace == NULL || trace->fd_path_disabled)
1677 return NULL;
1678
1679 if (fd < 0)
1680 return NULL;
1681
1682 if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
1683 if (!trace->live)
1684 return NULL;
1685 ++trace->stats.proc_getname;
1686 if (thread__read_fd_path(thread, fd))
1687 return NULL;
1688 }
1689
1690 return ttrace->files.table[fd].pathname;
1691 }
1692
1693 size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
1694 {
1695 int fd = arg->val;
1696 size_t printed = scnprintf(bf, size, "%d", fd);
1697 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1698
1699 if (path)
1700 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1701
1702 return printed;
1703 }
1704
1705 size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1706 {
1707 size_t printed = scnprintf(bf, size, "%d", fd);
1708 struct thread *thread = machine__find_thread(trace->host, pid, pid);
1709
1710 if (thread) {
1711 const char *path = thread__fd_path(thread, fd, trace);
1712
1713 if (path)
1714 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1715
1716 thread__put(thread);
1717 }
1718
1719 return printed;
1720 }
1721
1722 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1723 struct syscall_arg *arg)
1724 {
1725 int fd = arg->val;
1726 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1727 struct thread_trace *ttrace = thread__priv(arg->thread);
1728
1729 if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1730 zfree(&ttrace->files.table[fd].pathname);
1731
1732 return printed;
1733 }
1734
1735 static void thread__set_filename_pos(struct thread *thread, const char *bf,
1736 unsigned long ptr)
1737 {
1738 struct thread_trace *ttrace = thread__priv(thread);
1739
1740 ttrace->filename.ptr = ptr;
1741 ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1742 }
1743
1744 static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1745 {
1746 struct augmented_arg *augmented_arg = arg->augmented.args;
1747 size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1748 /*
1749 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1750 * we would have two strings, each prefixed by its size.
1751 */
1752 int consumed = sizeof(*augmented_arg) + augmented_arg->size;
1753
1754 arg->augmented.args = ((void *)arg->augmented.args) + consumed;
1755 arg->augmented.size -= consumed;
1756
1757 return printed;
1758 }
1759
1760 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1761 struct syscall_arg *arg)
1762 {
1763 unsigned long ptr = arg->val;
1764
1765 if (arg->augmented.args)
1766 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1767
1768 if (!arg->trace->vfs_getname)
1769 return scnprintf(bf, size, "%#x", ptr);
1770
1771 thread__set_filename_pos(arg->thread, bf, ptr);
1772 return 0;
1773 }
1774
1775 #define MAX_CONTROL_CHAR 31
1776 #define MAX_ASCII 127
1777
1778 static size_t syscall_arg__scnprintf_buf(char *bf, size_t size, struct syscall_arg *arg)
1779 {
1780 struct augmented_arg *augmented_arg = arg->augmented.args;
1781 unsigned char *orig = (unsigned char *)augmented_arg->value;
1782 size_t printed = 0;
1783 int consumed;
1784
1785 if (augmented_arg == NULL)
1786 return 0;
1787
1788 for (int j = 0; j < augmented_arg->size; ++j) {
1789 bool control_char = orig[j] <= MAX_CONTROL_CHAR || orig[j] >= MAX_ASCII;
1790 /* print control characters (0~31 and 127), and non-ascii characters in \(digits) */
1791 printed += scnprintf(bf + printed, size - printed, control_char ? "\\%d" : "%c", (int)orig[j]);
1792 }
1793
1794 consumed = sizeof(*augmented_arg) + augmented_arg->size;
1795 arg->augmented.args = ((void *)arg->augmented.args) + consumed;
1796 arg->augmented.size -= consumed;
1797
1798 return printed;
1799 }
1800
1801 static bool trace__filter_duration(struct trace *trace, double t)
1802 {
1803 return t < (trace->duration_filter * NSEC_PER_MSEC);
1804 }
1805
1806 static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1807 {
1808 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1809
1810 return fprintf(fp, "%10.3f ", ts);
1811 }
1812
1813 /*
1814 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1815 * using ttrace->entry_time for a thread that receives a sys_exit without
1816 * first having received a sys_enter ("poll" issued before tracing session
1817 * starts, lost sys_enter exit due to ring buffer overflow).
1818 */
1819 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1820 {
1821 if (tstamp > 0)
1822 return __trace__fprintf_tstamp(trace, tstamp, fp);
1823
1824 return fprintf(fp, " ? ");
1825 }
1826
1827 static pid_t workload_pid = -1;
1828 static volatile sig_atomic_t done = false;
1829 static volatile sig_atomic_t interrupted = false;
1830
1831 static void sighandler_interrupt(int sig __maybe_unused)
1832 {
1833 done = interrupted = true;
1834 }
1835
1836 static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
1837 void *context __maybe_unused)
1838 {
1839 if (info->si_pid == workload_pid)
1840 done = true;
1841 }
1842
1843 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1844 {
1845 size_t printed = 0;
1846
1847 if (trace->multiple_threads) {
1848 if (trace->show_comm)
1849 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1850 printed += fprintf(fp, "%d ", thread__tid(thread));
1851 }
1852
1853 return printed;
1854 }
1855
1856 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1857 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1858 {
1859 size_t printed = 0;
1860
1861 if (trace->show_tstamp)
1862 printed = trace__fprintf_tstamp(trace, tstamp, fp);
1863 if (trace->show_duration)
1864 printed += fprintf_duration(duration, duration_calculated, fp);
1865 return printed + trace__fprintf_comm_tid(trace, thread, fp);
1866 }
1867
1868 static int trace__process_event(struct trace *trace, struct machine *machine,
1869 union perf_event *event, struct perf_sample *sample)
1870 {
1871 int ret = 0;
1872
1873 switch (event->header.type) {
1874 case PERF_RECORD_LOST:
1875 color_fprintf(trace->output, PERF_COLOR_RED,
1876 "LOST %" PRIu64 " events!\n", (u64)event->lost.lost);
1877 ret = machine__process_lost_event(machine, event, sample);
1878 break;
1879 default:
1880 ret = machine__process_event(machine, event, sample);
1881 break;
1882 }
1883
1884 return ret;
1885 }
1886
1887 static int trace__tool_process(const struct perf_tool *tool,
1888 union perf_event *event,
1889 struct perf_sample *sample,
1890 struct machine *machine)
1891 {
1892 struct trace *trace = container_of(tool, struct trace, tool);
1893 return trace__process_event(trace, machine, event, sample);
1894 }
1895
1896 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1897 {
1898 struct machine *machine = vmachine;
1899
1900 if (machine->kptr_restrict_warned)
1901 return NULL;
1902
1903 if (symbol_conf.kptr_restrict) {
1904 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1905 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
1906 "Kernel samples will not be resolved.\n");
1907 machine->kptr_restrict_warned = true;
1908 return NULL;
1909 }
1910
1911 return machine__resolve_kernel_addr(vmachine, addrp, modp);
1912 }
1913
1914 static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
1915 {
1916 int err = symbol__init(NULL);
1917
1918 if (err)
1919 return err;
1920
1921 trace->host = machine__new_host();
1922 if (trace->host == NULL)
1923 return -ENOMEM;
1924
1925 thread__set_priv_destructor(thread_trace__delete);
1926
1927 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1928 if (err < 0)
1929 goto out;
1930
1931 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1932 evlist->core.threads, trace__tool_process,
1933 true, false, 1);
1934 out:
1935 if (err)
1936 symbol__exit();
1937
1938 return err;
1939 }
1940
1941 static void trace__symbols__exit(struct trace *trace)
1942 {
1943 machine__exit(trace->host);
1944 trace->host = NULL;
1945
1946 symbol__exit();
1947 }
1948
1949 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1950 {
1951 int idx;
1952
1953 if (nr_args == RAW_SYSCALL_ARGS_NUM && sc->fmt && sc->fmt->nr_args != 0)
1954 nr_args = sc->fmt->nr_args;
1955
1956 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1957 if (sc->arg_fmt == NULL)
1958 return -1;
1959
1960 for (idx = 0; idx < nr_args; ++idx) {
1961 if (sc->fmt)
1962 sc->arg_fmt[idx] = sc->fmt->arg[idx];
1963 }
1964
1965 sc->nr_args = nr_args;
1966 return 0;
1967 }
1968
1969 static const struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {
1970 { .name = "msr", .scnprintf = SCA_X86_MSR, .strtoul = STUL_X86_MSR, },
1971 { .name = "vector", .scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, },
1972 };
1973
1974 static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)
1975 {
1976 const struct syscall_arg_fmt *fmt = fmtp;
1977 return strcmp(name, fmt->name);
1978 }
1979
1980 static const struct syscall_arg_fmt *
1981 __syscall_arg_fmt__find_by_name(const struct syscall_arg_fmt *fmts, const int nmemb,
1982 const char *name)
1983 {
1984 return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);
1985 }
1986
1987 static const struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)
1988 {
1989 const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);
1990 return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);
1991 }
1992
1993 static struct tep_format_field *
1994 syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field,
1995 bool *use_btf)
1996 {
1997 struct tep_format_field *last_field = NULL;
1998 int len;
1999
2000 for (; field; field = field->next, ++arg) {
2001 last_field = field;
2002
2003 if (arg->scnprintf)
2004 continue;
2005
2006 len = strlen(field->name);
2007
2008 // As far as heuristics (or intention) goes this seems to hold true, and makes sense!
2009 if ((field->flags & TEP_FIELD_IS_POINTER) && strstarts(field->type, "const "))
2010 arg->from_user = true;
2011
2012 if (strcmp(field->type, "const char *") == 0 &&
2013 ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
2014 strstr(field->name, "path") != NULL)) {
2015 arg->scnprintf = SCA_FILENAME;
2016 } else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
2017 arg->scnprintf = SCA_PTR;
2018 else if (strcmp(field->type, "pid_t") == 0)
2019 arg->scnprintf = SCA_PID;
2020 else if (strcmp(field->type, "umode_t") == 0)
2021 arg->scnprintf = SCA_MODE_T;
2022 else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) {
2023 arg->scnprintf = SCA_CHAR_ARRAY;
2024 arg->nr_entries = field->arraylen;
2025 } else if ((strcmp(field->type, "int") == 0 ||
2026 strcmp(field->type, "unsigned int") == 0 ||
2027 strcmp(field->type, "long") == 0) &&
2028 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
2029 /*
2030 * /sys/kernel/tracing/events/syscalls/sys_enter*
2031 * grep -E 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
2032 * 65 int
2033 * 23 unsigned int
2034 * 7 unsigned long
2035 */
2036 arg->scnprintf = SCA_FD;
2037 } else if (strstr(field->type, "enum") && use_btf != NULL) {
2038 *use_btf = true;
2039 arg->strtoul = STUL_BTF_TYPE;
2040 } else {
2041 const struct syscall_arg_fmt *fmt =
2042 syscall_arg_fmt__find_by_name(field->name);
2043
2044 if (fmt) {
2045 arg->scnprintf = fmt->scnprintf;
2046 arg->strtoul = fmt->strtoul;
2047 }
2048 }
2049 }
2050
2051 return last_field;
2052 }
2053
2054 static int syscall__set_arg_fmts(struct syscall *sc)
2055 {
2056 struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args,
2057 &sc->use_btf);
2058
2059 if (last_field)
2060 sc->args_size = last_field->offset + last_field->size;
2061
2062 return 0;
2063 }
2064
2065 static int trace__read_syscall_info(struct trace *trace, int id)
2066 {
2067 char tp_name[128];
2068 struct syscall *sc;
2069 const char *name = syscalltbl__name(trace->sctbl, id);
2070 int err;
2071
2072 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
2073 if (trace->syscalls.table == NULL) {
2074 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
2075 if (trace->syscalls.table == NULL)
2076 return -ENOMEM;
2077 }
2078 #else
2079 if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) {
2080 // When using libaudit we don't know beforehand what is the max syscall id
2081 struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
2082
2083 if (table == NULL)
2084 return -ENOMEM;
2085
2086 // Need to memset from offset 0 and +1 members if brand new
2087 if (trace->syscalls.table == NULL)
2088 memset(table, 0, (id + 1) * sizeof(*sc));
2089 else
2090 memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc));
2091
2092 trace->syscalls.table = table;
2093 trace->sctbl->syscalls.max_id = id;
2094 }
2095 #endif
2096 sc = trace->syscalls.table + id;
2097 if (sc->nonexistent)
2098 return -EEXIST;
2099
2100 if (name == NULL) {
2101 sc->nonexistent = true;
2102 return -EEXIST;
2103 }
2104
2105 sc->name = name;
2106 sc->fmt = syscall_fmt__find(sc->name);
2107
2108 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
2109 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
2110
2111 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
2112 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
2113 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
2114 }
2115
2116 /*
2117 * Fails to read trace point format via sysfs node, so the trace point
2118 * doesn't exist. Set the 'nonexistent' flag as true.
2119 */
2120 if (IS_ERR(sc->tp_format)) {
2121 sc->nonexistent = true;
2122 return PTR_ERR(sc->tp_format);
2123 }
2124
2125 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ?
2126 RAW_SYSCALL_ARGS_NUM : sc->tp_format->format.nr_fields))
2127 return -ENOMEM;
2128
2129 sc->args = sc->tp_format->format.fields;
2130 /*
2131 * We need to check and discard the first variable '__syscall_nr'
2132 * or 'nr' that mean the syscall number. It is needless here.
2133 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
2134 */
2135 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
2136 sc->args = sc->args->next;
2137 --sc->nr_args;
2138 }
2139
2140 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
2141 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
2142
2143 err = syscall__set_arg_fmts(sc);
2144
2145 /* after calling syscall__set_arg_fmts() we'll know whether use_btf is true */
2146 if (sc->use_btf)
2147 trace__load_vmlinux_btf(trace);
2148
2149 return err;
2150 }
2151
2152 static int evsel__init_tp_arg_scnprintf(struct evsel *evsel, bool *use_btf)
2153 {
2154 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
2155
2156 if (fmt != NULL) {
2157 syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields, use_btf);
2158 return 0;
2159 }
2160
2161 return -ENOMEM;
2162 }
2163
2164 static int intcmp(const void *a, const void *b)
2165 {
2166 const int *one = a, *another = b;
2167
2168 return *one - *another;
2169 }
2170
2171 static int trace__validate_ev_qualifier(struct trace *trace)
2172 {
2173 int err = 0;
2174 bool printed_invalid_prefix = false;
2175 struct str_node *pos;
2176 size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
2177
2178 trace->ev_qualifier_ids.entries = malloc(nr_allocated *
2179 sizeof(trace->ev_qualifier_ids.entries[0]));
2180
2181 if (trace->ev_qualifier_ids.entries == NULL) {
2182 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
2183 trace->output);
2184 err = -EINVAL;
2185 goto out;
2186 }
2187
2188 strlist__for_each_entry(pos, trace->ev_qualifier) {
2189 const char *sc = pos->s;
2190 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
2191
2192 if (id < 0) {
2193 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
2194 if (id >= 0)
2195 goto matches;
2196
2197 if (!printed_invalid_prefix) {
2198 pr_debug("Skipping unknown syscalls: ");
2199 printed_invalid_prefix = true;
2200 } else {
2201 pr_debug(", ");
2202 }
2203
2204 pr_debug("%s", sc);
2205 continue;
2206 }
2207 matches:
2208 trace->ev_qualifier_ids.entries[nr_used++] = id;
2209 if (match_next == -1)
2210 continue;
2211
2212 while (1) {
2213 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
2214 if (id < 0)
2215 break;
2216 if (nr_allocated == nr_used) {
2217 void *entries;
2218
2219 nr_allocated += 8;
2220 entries = realloc(trace->ev_qualifier_ids.entries,
2221 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
2222 if (entries == NULL) {
2223 err = -ENOMEM;
2224 fputs("\nError:\t Not enough memory for parsing\n", trace->output);
2225 goto out_free;
2226 }
2227 trace->ev_qualifier_ids.entries = entries;
2228 }
2229 trace->ev_qualifier_ids.entries[nr_used++] = id;
2230 }
2231 }
2232
2233 trace->ev_qualifier_ids.nr = nr_used;
2234 qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
2235 out:
2236 if (printed_invalid_prefix)
2237 pr_debug("\n");
2238 return err;
2239 out_free:
2240 zfree(&trace->ev_qualifier_ids.entries);
2241 trace->ev_qualifier_ids.nr = 0;
2242 goto out;
2243 }
2244
2245 static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
2246 {
2247 bool in_ev_qualifier;
2248
2249 if (trace->ev_qualifier_ids.nr == 0)
2250 return true;
2251
2252 in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
2253 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
2254
2255 if (in_ev_qualifier)
2256 return !trace->not_ev_qualifier;
2257
2258 return trace->not_ev_qualifier;
2259 }
2260
2261 /*
2262 * args is to be interpreted as a series of longs but we need to handle
2263 * 8-byte unaligned accesses. args points to raw_data within the event
2264 * and raw_data is guaranteed to be 8-byte unaligned because it is
2265 * preceded by raw_size which is a u32. So we need to copy args to a temp
2266 * variable to read it. Most notably this avoids extended load instructions
2267 * on unaligned addresses
2268 */
2269 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
2270 {
2271 unsigned long val;
2272 unsigned char *p = arg->args + sizeof(unsigned long) * idx;
2273
2274 memcpy(&val, p, sizeof(val));
2275 return val;
2276 }
2277
2278 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
2279 struct syscall_arg *arg)
2280 {
2281 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
2282 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
2283
2284 return scnprintf(bf, size, "arg%d: ", arg->idx);
2285 }
2286
2287 /*
2288 * Check if the value is in fact zero, i.e. mask whatever needs masking, such
2289 * as mount 'flags' argument that needs ignoring some magic flag, see comment
2290 * in tools/perf/trace/beauty/mount_flags.c
2291 */
2292 static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)
2293 {
2294 if (fmt && fmt->mask_val)
2295 return fmt->mask_val(arg, val);
2296
2297 return val;
2298 }
2299
2300 static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,
2301 struct syscall_arg *arg, unsigned long val)
2302 {
2303 if (fmt && fmt->scnprintf) {
2304 arg->val = val;
2305 if (fmt->parm)
2306 arg->parm = fmt->parm;
2307 return fmt->scnprintf(bf, size, arg);
2308 }
2309 return scnprintf(bf, size, "%ld", val);
2310 }
2311
2312 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
2313 unsigned char *args, void *augmented_args, int augmented_args_size,
2314 struct trace *trace, struct thread *thread)
2315 {
2316 size_t printed = 0, btf_printed;
2317 unsigned long val;
2318 u8 bit = 1;
2319 struct syscall_arg arg = {
2320 .args = args,
2321 .augmented = {
2322 .size = augmented_args_size,
2323 .args = augmented_args,
2324 },
2325 .idx = 0,
2326 .mask = 0,
2327 .trace = trace,
2328 .thread = thread,
2329 .show_string_prefix = trace->show_string_prefix,
2330 };
2331 struct thread_trace *ttrace = thread__priv(thread);
2332 void *default_scnprintf;
2333
2334 /*
2335 * Things like fcntl will set this in its 'cmd' formatter to pick the
2336 * right formatter for the return value (an fd? file flags?), which is
2337 * not needed for syscalls that always return a given type, say an fd.
2338 */
2339 ttrace->ret_scnprintf = NULL;
2340
2341 if (sc->args != NULL) {
2342 struct tep_format_field *field;
2343
2344 for (field = sc->args; field;
2345 field = field->next, ++arg.idx, bit <<= 1) {
2346 if (arg.mask & bit)
2347 continue;
2348
2349 arg.fmt = &sc->arg_fmt[arg.idx];
2350 val = syscall_arg__val(&arg, arg.idx);
2351 /*
2352 * Some syscall args need some mask, most don't and
2353 * return val untouched.
2354 */
2355 val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);
2356
2357 /*
2358 * Suppress this argument if its value is zero and show_zero
2359 * property isn't set.
2360 *
2361 * If it has a BTF type, then override the zero suppression knob
2362 * as the common case is for zero in an enum to have an associated entry.
2363 */
2364 if (val == 0 && !trace->show_zeros &&
2365 !(sc->arg_fmt && sc->arg_fmt[arg.idx].show_zero) &&
2366 !(sc->arg_fmt && sc->arg_fmt[arg.idx].strtoul == STUL_BTF_TYPE))
2367 continue;
2368
2369 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2370
2371 if (trace->show_arg_names)
2372 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2373
2374 default_scnprintf = sc->arg_fmt[arg.idx].scnprintf;
2375
2376 if (trace->force_btf || default_scnprintf == NULL || default_scnprintf == SCA_PTR) {
2377 btf_printed = trace__btf_scnprintf(trace, &arg, bf + printed,
2378 size - printed, val, field->type);
2379 if (btf_printed) {
2380 printed += btf_printed;
2381 continue;
2382 }
2383 }
2384
2385 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],
2386 bf + printed, size - printed, &arg, val);
2387 }
2388 } else if (IS_ERR(sc->tp_format)) {
2389 /*
2390 * If we managed to read the tracepoint /format file, then we
2391 * may end up not having any args, like with gettid(), so only
2392 * print the raw args when we didn't manage to read it.
2393 */
2394 while (arg.idx < sc->nr_args) {
2395 if (arg.mask & bit)
2396 goto next_arg;
2397 val = syscall_arg__val(&arg, arg.idx);
2398 if (printed)
2399 printed += scnprintf(bf + printed, size - printed, ", ");
2400 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
2401 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);
2402 next_arg:
2403 ++arg.idx;
2404 bit <<= 1;
2405 }
2406 }
2407
2408 return printed;
2409 }
2410
2411 typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
2412 union perf_event *event,
2413 struct perf_sample *sample);
2414
2415 static struct syscall *trace__syscall_info(struct trace *trace,
2416 struct evsel *evsel, int id)
2417 {
2418 int err = 0;
2419
2420 if (id < 0) {
2421
2422 /*
2423 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
2424 * before that, leaving at a higher verbosity level till that is
2425 * explained. Reproduced with plain ftrace with:
2426 *
2427 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
2428 * grep "NR -1 " /t/trace_pipe
2429 *
2430 * After generating some load on the machine.
2431 */
2432 if (verbose > 1) {
2433 static u64 n;
2434 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
2435 id, evsel__name(evsel), ++n);
2436 }
2437 return NULL;
2438 }
2439
2440 err = -EINVAL;
2441
2442 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
2443 if (id > trace->sctbl->syscalls.max_id) {
2444 #else
2445 if (id >= trace->sctbl->syscalls.max_id) {
2446 /*
2447 * With libaudit we don't know beforehand what is the max_id,
2448 * so we let trace__read_syscall_info() figure that out as we
2449 * go on reading syscalls.
2450 */
2451 err = trace__read_syscall_info(trace, id);
2452 if (err)
2453 #endif
2454 goto out_cant_read;
2455 }
2456
2457 if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
2458 (err = trace__read_syscall_info(trace, id)) != 0)
2459 goto out_cant_read;
2460
2461 if (trace->syscalls.table && trace->syscalls.table[id].nonexistent)
2462 goto out_cant_read;
2463
2464 return &trace->syscalls.table[id];
2465
2466 out_cant_read:
2467 if (verbose > 0) {
2468 char sbuf[STRERR_BUFSIZE];
2469 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
2470 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
2471 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
2472 fputs(" information\n", trace->output);
2473 }
2474 return NULL;
2475 }
2476
2477 struct syscall_stats {
2478 struct stats stats;
2479 u64 nr_failures;
2480 int max_errno;
2481 u32 *errnos;
2482 };
2483
2484 static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace,
2485 int id, struct perf_sample *sample, long err, bool errno_summary)
2486 {
2487 struct int_node *inode;
2488 struct syscall_stats *stats;
2489 u64 duration = 0;
2490
2491 inode = intlist__findnew(ttrace->syscall_stats, id);
2492 if (inode == NULL)
2493 return;
2494
2495 stats = inode->priv;
2496 if (stats == NULL) {
2497 stats = zalloc(sizeof(*stats));
2498 if (stats == NULL)
2499 return;
2500
2501 init_stats(&stats->stats);
2502 inode->priv = stats;
2503 }
2504
2505 if (ttrace->entry_time && sample->time > ttrace->entry_time)
2506 duration = sample->time - ttrace->entry_time;
2507
2508 update_stats(&stats->stats, duration);
2509
2510 if (err < 0) {
2511 ++stats->nr_failures;
2512
2513 if (!errno_summary)
2514 return;
2515
2516 err = -err;
2517 if (err > stats->max_errno) {
2518 u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32));
2519
2520 if (new_errnos) {
2521 memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32));
2522 } else {
2523 pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n",
2524 thread__comm_str(thread), thread__pid(thread),
2525 thread__tid(thread));
2526 return;
2527 }
2528
2529 stats->errnos = new_errnos;
2530 stats->max_errno = err;
2531 }
2532
2533 ++stats->errnos[err - 1];
2534 }
2535 }
2536
2537 static int trace__printf_interrupted_entry(struct trace *trace)
2538 {
2539 struct thread_trace *ttrace;
2540 size_t printed;
2541 int len;
2542
2543 if (trace->failure_only || trace->current == NULL)
2544 return 0;
2545
2546 ttrace = thread__priv(trace->current);
2547
2548 if (!ttrace->entry_pending)
2549 return 0;
2550
2551 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
2552 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
2553
2554 if (len < trace->args_alignment - 4)
2555 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
2556
2557 printed += fprintf(trace->output, " ...\n");
2558
2559 ttrace->entry_pending = false;
2560 ++trace->nr_events_printed;
2561
2562 return printed;
2563 }
2564
2565 static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
2566 struct perf_sample *sample, struct thread *thread)
2567 {
2568 int printed = 0;
2569
2570 if (trace->print_sample) {
2571 double ts = (double)sample->time / NSEC_PER_MSEC;
2572
2573 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
2574 evsel__name(evsel), ts,
2575 thread__comm_str(thread),
2576 sample->pid, sample->tid, sample->cpu);
2577 }
2578
2579 return printed;
2580 }
2581
2582 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
2583 {
2584 void *augmented_args = NULL;
2585 /*
2586 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
2587 * and there we get all 6 syscall args plus the tracepoint common fields
2588 * that gets calculated at the start and the syscall_nr (another long).
2589 * So we check if that is the case and if so don't look after the
2590 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
2591 * which is fixed.
2592 *
2593 * We'll revisit this later to pass s->args_size to the BPF augmenter
2594 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
2595 * copies only what we need for each syscall, like what happens when we
2596 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
2597 * traffic to just what is needed for each syscall.
2598 */
2599 int args_size = raw_augmented_args_size ?: sc->args_size;
2600
2601 *augmented_args_size = sample->raw_size - args_size;
2602 if (*augmented_args_size > 0)
2603 augmented_args = sample->raw_data + args_size;
2604
2605 return augmented_args;
2606 }
2607
2608 static void syscall__exit(struct syscall *sc)
2609 {
2610 if (!sc)
2611 return;
2612
2613 zfree(&sc->arg_fmt);
2614 }
2615
2616 static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
2617 union perf_event *event __maybe_unused,
2618 struct perf_sample *sample)
2619 {
2620 char *msg;
2621 void *args;
2622 int printed = 0;
2623 struct thread *thread;
2624 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2625 int augmented_args_size = 0;
2626 void *augmented_args = NULL;
2627 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2628 struct thread_trace *ttrace;
2629
2630 if (sc == NULL)
2631 return -1;
2632
2633 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2634 ttrace = thread__trace(thread, trace->output);
2635 if (ttrace == NULL)
2636 goto out_put;
2637
2638 trace__fprintf_sample(trace, evsel, sample, thread);
2639
2640 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2641
2642 if (ttrace->entry_str == NULL) {
2643 ttrace->entry_str = malloc(trace__entry_str_size);
2644 if (!ttrace->entry_str)
2645 goto out_put;
2646 }
2647
2648 if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
2649 trace__printf_interrupted_entry(trace);
2650 /*
2651 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2652 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2653 * this breaks syscall__augmented_args() check for augmented args, as we calculate
2654 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2655 * so when handling, say the openat syscall, we end up getting 6 args for the
2656 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2657 * thinking that the extra 2 u64 args are the augmented filename, so just check
2658 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2659 */
2660 if (evsel != trace->syscalls.events.sys_enter)
2661 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2662 ttrace->entry_time = sample->time;
2663 msg = ttrace->entry_str;
2664 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
2665
2666 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
2667 args, augmented_args, augmented_args_size, trace, thread);
2668
2669 if (sc->is_exit) {
2670 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
2671 int alignment = 0;
2672
2673 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
2674 printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2675 if (trace->args_alignment > printed)
2676 alignment = trace->args_alignment - printed;
2677 fprintf(trace->output, "%*s= ?\n", alignment, " ");
2678 }
2679 } else {
2680 ttrace->entry_pending = true;
2681 /* See trace__vfs_getname & trace__sys_exit */
2682 ttrace->filename.pending_open = false;
2683 }
2684
2685 if (trace->current != thread) {
2686 thread__put(trace->current);
2687 trace->current = thread__get(thread);
2688 }
2689 err = 0;
2690 out_put:
2691 thread__put(thread);
2692 return err;
2693 }
2694
2695 static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
2696 struct perf_sample *sample)
2697 {
2698 struct thread_trace *ttrace;
2699 struct thread *thread;
2700 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2701 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2702 char msg[1024];
2703 void *args, *augmented_args = NULL;
2704 int augmented_args_size;
2705 size_t printed = 0;
2706
2707 if (sc == NULL)
2708 return -1;
2709
2710 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2711 ttrace = thread__trace(thread, trace->output);
2712 /*
2713 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2714 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2715 */
2716 if (ttrace == NULL)
2717 goto out_put;
2718
2719 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2720 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2721 printed += syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
2722 fprintf(trace->output, "%.*s", (int)printed, msg);
2723 err = 0;
2724 out_put:
2725 thread__put(thread);
2726 return err;
2727 }
2728
2729 static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
2730 struct perf_sample *sample,
2731 struct callchain_cursor *cursor)
2732 {
2733 struct addr_location al;
2734 int max_stack = evsel->core.attr.sample_max_stack ?
2735 evsel->core.attr.sample_max_stack :
2736 trace->max_stack;
2737 int err = -1;
2738
2739 addr_location__init(&al);
2740 if (machine__resolve(trace->host, &al, sample) < 0)
2741 goto out;
2742
2743 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
2744 out:
2745 addr_location__exit(&al);
2746 return err;
2747 }
2748
2749 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2750 {
2751 /* TODO: user-configurable print_opts */
2752 const unsigned int print_opts = EVSEL__PRINT_SYM |
2753 EVSEL__PRINT_DSO |
2754 EVSEL__PRINT_UNKNOWN_AS_ADDR;
2755
2756 return sample__fprintf_callchain(sample, 38, print_opts, get_tls_callchain_cursor(), symbol_conf.bt_stop_list, trace->output);
2757 }
2758
2759 static const char *errno_to_name(struct evsel *evsel, int err)
2760 {
2761 struct perf_env *env = evsel__env(evsel);
2762
2763 return perf_env__arch_strerrno(env, err);
2764 }
2765
2766 static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
2767 union perf_event *event __maybe_unused,
2768 struct perf_sample *sample)
2769 {
2770 long ret;
2771 u64 duration = 0;
2772 bool duration_calculated = false;
2773 struct thread *thread;
2774 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2775 int alignment = trace->args_alignment;
2776 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2777 struct thread_trace *ttrace;
2778
2779 if (sc == NULL)
2780 return -1;
2781
2782 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2783 ttrace = thread__trace(thread, trace->output);
2784 if (ttrace == NULL)
2785 goto out_put;
2786
2787 trace__fprintf_sample(trace, evsel, sample, thread);
2788
2789 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2790
2791 if (trace->summary)
2792 thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary);
2793
2794 if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2795 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2796 ttrace->filename.pending_open = false;
2797 ++trace->stats.vfs_getname;
2798 }
2799
2800 if (ttrace->entry_time) {
2801 duration = sample->time - ttrace->entry_time;
2802 if (trace__filter_duration(trace, duration))
2803 goto out;
2804 duration_calculated = true;
2805 } else if (trace->duration_filter)
2806 goto out;
2807
2808 if (sample->callchain) {
2809 struct callchain_cursor *cursor = get_tls_callchain_cursor();
2810
2811 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
2812 if (callchain_ret == 0) {
2813 if (cursor->nr < trace->min_stack)
2814 goto out;
2815 callchain_ret = 1;
2816 }
2817 }
2818
2819 if (trace->summary_only || (ret >= 0 && trace->failure_only))
2820 goto out;
2821
2822 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2823
2824 if (ttrace->entry_pending) {
2825 printed = fprintf(trace->output, "%s", ttrace->entry_str);
2826 } else {
2827 printed += fprintf(trace->output, " ... [");
2828 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2829 printed += 9;
2830 printed += fprintf(trace->output, "]: %s()", sc->name);
2831 }
2832
2833 printed++; /* the closing ')' */
2834
2835 if (alignment > printed)
2836 alignment -= printed;
2837 else
2838 alignment = 0;
2839
2840 fprintf(trace->output, ")%*s= ", alignment, " ");
2841
2842 if (sc->fmt == NULL) {
2843 if (ret < 0)
2844 goto errno_print;
2845 signed_print:
2846 fprintf(trace->output, "%ld", ret);
2847 } else if (ret < 0) {
2848 errno_print: {
2849 char bf[STRERR_BUFSIZE];
2850 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2851 *e = errno_to_name(evsel, -ret);
2852
2853 fprintf(trace->output, "-1 %s (%s)", e, emsg);
2854 }
2855 } else if (ret == 0 && sc->fmt->timeout)
2856 fprintf(trace->output, "0 (Timeout)");
2857 else if (ttrace->ret_scnprintf) {
2858 char bf[1024];
2859 struct syscall_arg arg = {
2860 .val = ret,
2861 .thread = thread,
2862 .trace = trace,
2863 };
2864 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2865 ttrace->ret_scnprintf = NULL;
2866 fprintf(trace->output, "%s", bf);
2867 } else if (sc->fmt->hexret)
2868 fprintf(trace->output, "%#lx", ret);
2869 else if (sc->fmt->errpid) {
2870 struct thread *child = machine__find_thread(trace->host, ret, ret);
2871
2872 if (child != NULL) {
2873 fprintf(trace->output, "%ld", ret);
2874 if (thread__comm_set(child))
2875 fprintf(trace->output, " (%s)", thread__comm_str(child));
2876 thread__put(child);
2877 }
2878 } else
2879 goto signed_print;
2880
2881 fputc('\n', trace->output);
2882
2883 /*
2884 * We only consider an 'event' for the sake of --max-events a non-filtered
2885 * sys_enter + sys_exit and other tracepoint events.
2886 */
2887 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2888 interrupted = true;
2889
2890 if (callchain_ret > 0)
2891 trace__fprintf_callchain(trace, sample);
2892 else if (callchain_ret < 0)
2893 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2894 out:
2895 ttrace->entry_pending = false;
2896 err = 0;
2897 out_put:
2898 thread__put(thread);
2899 return err;
2900 }
2901
2902 static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
2903 union perf_event *event __maybe_unused,
2904 struct perf_sample *sample)
2905 {
2906 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2907 struct thread_trace *ttrace;
2908 size_t filename_len, entry_str_len, to_move;
2909 ssize_t remaining_space;
2910 char *pos;
2911 const char *filename = evsel__rawptr(evsel, sample, "pathname");
2912
2913 if (!thread)
2914 goto out;
2915
2916 ttrace = thread__priv(thread);
2917 if (!ttrace)
2918 goto out_put;
2919
2920 filename_len = strlen(filename);
2921 if (filename_len == 0)
2922 goto out_put;
2923
2924 if (ttrace->filename.namelen < filename_len) {
2925 char *f = realloc(ttrace->filename.name, filename_len + 1);
2926
2927 if (f == NULL)
2928 goto out_put;
2929
2930 ttrace->filename.namelen = filename_len;
2931 ttrace->filename.name = f;
2932 }
2933
2934 strcpy(ttrace->filename.name, filename);
2935 ttrace->filename.pending_open = true;
2936
2937 if (!ttrace->filename.ptr)
2938 goto out_put;
2939
2940 entry_str_len = strlen(ttrace->entry_str);
2941 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2942 if (remaining_space <= 0)
2943 goto out_put;
2944
2945 if (filename_len > (size_t)remaining_space) {
2946 filename += filename_len - remaining_space;
2947 filename_len = remaining_space;
2948 }
2949
2950 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2951 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2952 memmove(pos + filename_len, pos, to_move);
2953 memcpy(pos, filename, filename_len);
2954
2955 ttrace->filename.ptr = 0;
2956 ttrace->filename.entry_str_pos = 0;
2957 out_put:
2958 thread__put(thread);
2959 out:
2960 return 0;
2961 }
2962
2963 static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
2964 union perf_event *event __maybe_unused,
2965 struct perf_sample *sample)
2966 {
2967 u64 runtime = evsel__intval(evsel, sample, "runtime");
2968 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2969 struct thread *thread = machine__findnew_thread(trace->host,
2970 sample->pid,
2971 sample->tid);
2972 struct thread_trace *ttrace = thread__trace(thread, trace->output);
2973
2974 if (ttrace == NULL)
2975 goto out_dump;
2976
2977 ttrace->runtime_ms += runtime_ms;
2978 trace->runtime_ms += runtime_ms;
2979 out_put:
2980 thread__put(thread);
2981 return 0;
2982
2983 out_dump:
2984 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2985 evsel->name,
2986 evsel__strval(evsel, sample, "comm"),
2987 (pid_t)evsel__intval(evsel, sample, "pid"),
2988 runtime,
2989 evsel__intval(evsel, sample, "vruntime"));
2990 goto out_put;
2991 }
2992
2993 static int bpf_output__printer(enum binary_printer_ops op,
2994 unsigned int val, void *extra __maybe_unused, FILE *fp)
2995 {
2996 unsigned char ch = (unsigned char)val;
2997
2998 switch (op) {
2999 case BINARY_PRINT_CHAR_DATA:
3000 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
3001 case BINARY_PRINT_DATA_BEGIN:
3002 case BINARY_PRINT_LINE_BEGIN:
3003 case BINARY_PRINT_ADDR:
3004 case BINARY_PRINT_NUM_DATA:
3005 case BINARY_PRINT_NUM_PAD:
3006 case BINARY_PRINT_SEP:
3007 case BINARY_PRINT_CHAR_PAD:
3008 case BINARY_PRINT_LINE_END:
3009 case BINARY_PRINT_DATA_END:
3010 default:
3011 break;
3012 }
3013
3014 return 0;
3015 }
3016
3017 static void bpf_output__fprintf(struct trace *trace,
3018 struct perf_sample *sample)
3019 {
3020 binary__fprintf(sample->raw_data, sample->raw_size, 8,
3021 bpf_output__printer, NULL, trace->output);
3022 ++trace->nr_events_printed;
3023 }
3024
3025 static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,
3026 struct thread *thread, void *augmented_args, int augmented_args_size)
3027 {
3028 char bf[2048];
3029 size_t size = sizeof(bf);
3030 struct tep_format_field *field = evsel->tp_format->format.fields;
3031 struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel);
3032 size_t printed = 0, btf_printed;
3033 unsigned long val;
3034 u8 bit = 1;
3035 struct syscall_arg syscall_arg = {
3036 .augmented = {
3037 .size = augmented_args_size,
3038 .args = augmented_args,
3039 },
3040 .idx = 0,
3041 .mask = 0,
3042 .trace = trace,
3043 .thread = thread,
3044 .show_string_prefix = trace->show_string_prefix,
3045 };
3046
3047 for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {
3048 if (syscall_arg.mask & bit)
3049 continue;
3050
3051 syscall_arg.len = 0;
3052 syscall_arg.fmt = arg;
3053 if (field->flags & TEP_FIELD_IS_ARRAY) {
3054 int offset = field->offset;
3055
3056 if (field->flags & TEP_FIELD_IS_DYNAMIC) {
3057 offset = format_field__intval(field, sample, evsel->needs_swap);
3058 syscall_arg.len = offset >> 16;
3059 offset &= 0xffff;
3060 if (tep_field_is_relative(field->flags))
3061 offset += field->offset + field->size;
3062 }
3063
3064 val = (uintptr_t)(sample->raw_data + offset);
3065 } else
3066 val = format_field__intval(field, sample, evsel->needs_swap);
3067 /*
3068 * Some syscall args need some mask, most don't and
3069 * return val untouched.
3070 */
3071 val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);
3072
3073 /* Suppress this argument if its value is zero and show_zero property isn't set. */
3074 if (val == 0 && !trace->show_zeros && !arg->show_zero && arg->strtoul != STUL_BTF_TYPE)
3075 continue;
3076
3077 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
3078
3079 if (trace->show_arg_names)
3080 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
3081
3082 btf_printed = trace__btf_scnprintf(trace, &syscall_arg, bf + printed, size - printed, val, field->type);
3083 if (btf_printed) {
3084 printed += btf_printed;
3085 continue;
3086 }
3087
3088 printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);
3089 }
3090
3091 return printed + fprintf(trace->output, "%.*s", (int)printed, bf);
3092 }
3093
3094 static int trace__event_handler(struct trace *trace, struct evsel *evsel,
3095 union perf_event *event __maybe_unused,
3096 struct perf_sample *sample)
3097 {
3098 struct thread *thread;
3099 int callchain_ret = 0;
3100
3101 if (evsel->nr_events_printed >= evsel->max_events)
3102 return 0;
3103
3104 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
3105
3106 if (sample->callchain) {
3107 struct callchain_cursor *cursor = get_tls_callchain_cursor();
3108
3109 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
3110 if (callchain_ret == 0) {
3111 if (cursor->nr < trace->min_stack)
3112 goto out;
3113 callchain_ret = 1;
3114 }
3115 }
3116
3117 trace__printf_interrupted_entry(trace);
3118 trace__fprintf_tstamp(trace, sample->time, trace->output);
3119
3120 if (trace->trace_syscalls && trace->show_duration)
3121 fprintf(trace->output, "( ): ");
3122
3123 if (thread)
3124 trace__fprintf_comm_tid(trace, thread, trace->output);
3125
3126 if (evsel == trace->syscalls.events.bpf_output) {
3127 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
3128 struct syscall *sc = trace__syscall_info(trace, evsel, id);
3129
3130 if (sc) {
3131 fprintf(trace->output, "%s(", sc->name);
3132 trace__fprintf_sys_enter(trace, evsel, sample);
3133 fputc(')', trace->output);
3134 goto newline;
3135 }
3136
3137 /*
3138 * XXX: Not having the associated syscall info or not finding/adding
3139 * the thread should never happen, but if it does...
3140 * fall thru and print it as a bpf_output event.
3141 */
3142 }
3143
3144 fprintf(trace->output, "%s(", evsel->name);
3145
3146 if (evsel__is_bpf_output(evsel)) {
3147 bpf_output__fprintf(trace, sample);
3148 } else if (evsel->tp_format) {
3149 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
3150 trace__fprintf_sys_enter(trace, evsel, sample)) {
3151 if (trace->libtraceevent_print) {
3152 event_format__fprintf(evsel->tp_format, sample->cpu,
3153 sample->raw_data, sample->raw_size,
3154 trace->output);
3155 } else {
3156 trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);
3157 }
3158 }
3159 }
3160
3161 newline:
3162 fprintf(trace->output, ")\n");
3163
3164 if (callchain_ret > 0)
3165 trace__fprintf_callchain(trace, sample);
3166 else if (callchain_ret < 0)
3167 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
3168
3169 ++trace->nr_events_printed;
3170
3171 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
3172 evsel__disable(evsel);
3173 evsel__close(evsel);
3174 }
3175 out:
3176 thread__put(thread);
3177 return 0;
3178 }
3179
3180 static void print_location(FILE *f, struct perf_sample *sample,
3181 struct addr_location *al,
3182 bool print_dso, bool print_sym)
3183 {
3184
3185 if ((verbose > 0 || print_dso) && al->map)
3186 fprintf(f, "%s@", dso__long_name(map__dso(al->map)));
3187
3188 if ((verbose > 0 || print_sym) && al->sym)
3189 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
3190 al->addr - al->sym->start);
3191 else if (al->map)
3192 fprintf(f, "0x%" PRIx64, al->addr);
3193 else
3194 fprintf(f, "0x%" PRIx64, sample->addr);
3195 }
3196
3197 static int trace__pgfault(struct trace *trace,
3198 struct evsel *evsel,
3199 union perf_event *event __maybe_unused,
3200 struct perf_sample *sample)
3201 {
3202 struct thread *thread;
3203 struct addr_location al;
3204 char map_type = 'd';
3205 struct thread_trace *ttrace;
3206 int err = -1;
3207 int callchain_ret = 0;
3208
3209 addr_location__init(&al);
3210 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
3211
3212 if (sample->callchain) {
3213 struct callchain_cursor *cursor = get_tls_callchain_cursor();
3214
3215 callchain_ret = trace__resolve_callchain(trace, evsel, sample, cursor);
3216 if (callchain_ret == 0) {
3217 if (cursor->nr < trace->min_stack)
3218 goto out_put;
3219 callchain_ret = 1;
3220 }
3221 }
3222
3223 ttrace = thread__trace(thread, trace->output);
3224 if (ttrace == NULL)
3225 goto out_put;
3226
3227 if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
3228 ttrace->pfmaj++;
3229 else
3230 ttrace->pfmin++;
3231
3232 if (trace->summary_only)
3233 goto out;
3234
3235 thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
3236
3237 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
3238
3239 fprintf(trace->output, "%sfault [",
3240 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
3241 "maj" : "min");
3242
3243 print_location(trace->output, sample, &al, false, true);
3244
3245 fprintf(trace->output, "] => ");
3246
3247 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
3248
3249 if (!al.map) {
3250 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
3251
3252 if (al.map)
3253 map_type = 'x';
3254 else
3255 map_type = '?';
3256 }
3257
3258 print_location(trace->output, sample, &al, true, false);
3259
3260 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
3261
3262 if (callchain_ret > 0)
3263 trace__fprintf_callchain(trace, sample);
3264 else if (callchain_ret < 0)
3265 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
3266
3267 ++trace->nr_events_printed;
3268 out:
3269 err = 0;
3270 out_put:
3271 thread__put(thread);
3272 addr_location__exit(&al);
3273 return err;
3274 }
3275
3276 static void trace__set_base_time(struct trace *trace,
3277 struct evsel *evsel,
3278 struct perf_sample *sample)
3279 {
3280 /*
3281 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
3282 * and don't use sample->time unconditionally, we may end up having
3283 * some other event in the future without PERF_SAMPLE_TIME for good
3284 * reason, i.e. we may not be interested in its timestamps, just in
3285 * it taking place, picking some piece of information when it
3286 * appears in our event stream (vfs_getname comes to mind).
3287 */
3288 if (trace->base_time == 0 && !trace->full_time &&
3289 (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
3290 trace->base_time = sample->time;
3291 }
3292
3293 static int trace__process_sample(const struct perf_tool *tool,
3294 union perf_event *event,
3295 struct perf_sample *sample,
3296 struct evsel *evsel,
3297 struct machine *machine __maybe_unused)
3298 {
3299 struct trace *trace = container_of(tool, struct trace, tool);
3300 struct thread *thread;
3301 int err = 0;
3302
3303 tracepoint_handler handler = evsel->handler;
3304
3305 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
3306 if (thread && thread__is_filtered(thread))
3307 goto out;
3308
3309 trace__set_base_time(trace, evsel, sample);
3310
3311 if (handler) {
3312 ++trace->nr_events;
3313 handler(trace, evsel, event, sample);
3314 }
3315 out:
3316 thread__put(thread);
3317 return err;
3318 }
3319
3320 static int trace__record(struct trace *trace, int argc, const char **argv)
3321 {
3322 unsigned int rec_argc, i, j;
3323 const char **rec_argv;
3324 const char * const record_args[] = {
3325 "record",
3326 "-R",
3327 "-m", "1024",
3328 "-c", "1",
3329 };
3330 pid_t pid = getpid();
3331 char *filter = asprintf__tp_filter_pids(1, &pid);
3332 const char * const sc_args[] = { "-e", };
3333 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
3334 const char * const majpf_args[] = { "-e", "major-faults" };
3335 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
3336 const char * const minpf_args[] = { "-e", "minor-faults" };
3337 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
3338 int err = -1;
3339
3340 /* +3 is for the event string below and the pid filter */
3341 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 +
3342 majpf_args_nr + minpf_args_nr + argc;
3343 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3344
3345 if (rec_argv == NULL || filter == NULL)
3346 goto out_free;
3347
3348 j = 0;
3349 for (i = 0; i < ARRAY_SIZE(record_args); i++)
3350 rec_argv[j++] = record_args[i];
3351
3352 if (trace->trace_syscalls) {
3353 for (i = 0; i < sc_args_nr; i++)
3354 rec_argv[j++] = sc_args[i];
3355
3356 /* event string may be different for older kernels - e.g., RHEL6 */
3357 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
3358 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
3359 else if (is_valid_tracepoint("syscalls:sys_enter"))
3360 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
3361 else {
3362 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
3363 goto out_free;
3364 }
3365 }
3366
3367 rec_argv[j++] = "--filter";
3368 rec_argv[j++] = filter;
3369
3370 if (trace->trace_pgfaults & TRACE_PFMAJ)
3371 for (i = 0; i < majpf_args_nr; i++)
3372 rec_argv[j++] = majpf_args[i];
3373
3374 if (trace->trace_pgfaults & TRACE_PFMIN)
3375 for (i = 0; i < minpf_args_nr; i++)
3376 rec_argv[j++] = minpf_args[i];
3377
3378 for (i = 0; i < (unsigned int)argc; i++)
3379 rec_argv[j++] = argv[i];
3380
3381 err = cmd_record(j, rec_argv);
3382 out_free:
3383 free(filter);
3384 free(rec_argv);
3385 return err;
3386 }
3387
3388 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
3389
3390 static bool evlist__add_vfs_getname(struct evlist *evlist)
3391 {
3392 bool found = false;
3393 struct evsel *evsel, *tmp;
3394 struct parse_events_error err;
3395 int ret;
3396
3397 parse_events_error__init(&err);
3398 ret = parse_events(evlist, "probe:vfs_getname*", &err);
3399 parse_events_error__exit(&err);
3400 if (ret)
3401 return false;
3402
3403 evlist__for_each_entry_safe(evlist, evsel, tmp) {
3404 if (!strstarts(evsel__name(evsel), "probe:vfs_getname"))
3405 continue;
3406
3407 if (evsel__field(evsel, "pathname")) {
3408 evsel->handler = trace__vfs_getname;
3409 found = true;
3410 continue;
3411 }
3412
3413 list_del_init(&evsel->core.node);
3414 evsel->evlist = NULL;
3415 evsel__delete(evsel);
3416 }
3417
3418 return found;
3419 }
3420
3421 static struct evsel *evsel__new_pgfault(u64 config)
3422 {
3423 struct evsel *evsel;
3424 struct perf_event_attr attr = {
3425 .type = PERF_TYPE_SOFTWARE,
3426 .mmap_data = 1,
3427 };
3428
3429 attr.config = config;
3430 attr.sample_period = 1;
3431
3432 event_attr_init(&attr);
3433
3434 evsel = evsel__new(&attr);
3435 if (evsel)
3436 evsel->handler = trace__pgfault;
3437
3438 return evsel;
3439 }
3440
3441 static void evlist__free_syscall_tp_fields(struct evlist *evlist)
3442 {
3443 struct evsel *evsel;
3444
3445 evlist__for_each_entry(evlist, evsel) {
3446 evsel_trace__delete(evsel->priv);
3447 evsel->priv = NULL;
3448 }
3449 }
3450
3451 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
3452 {
3453 const u32 type = event->header.type;
3454 struct evsel *evsel;
3455
3456 if (type != PERF_RECORD_SAMPLE) {
3457 trace__process_event(trace, trace->host, event, sample);
3458 return;
3459 }
3460
3461 evsel = evlist__id2evsel(trace->evlist, sample->id);
3462 if (evsel == NULL) {
3463 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
3464 return;
3465 }
3466
3467 if (evswitch__discard(&trace->evswitch, evsel))
3468 return;
3469
3470 trace__set_base_time(trace, evsel, sample);
3471
3472 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
3473 sample->raw_data == NULL) {
3474 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
3475 evsel__name(evsel), sample->tid,
3476 sample->cpu, sample->raw_size);
3477 } else {
3478 tracepoint_handler handler = evsel->handler;
3479 handler(trace, evsel, event, sample);
3480 }
3481
3482 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
3483 interrupted = true;
3484 }
3485
3486 static int trace__add_syscall_newtp(struct trace *trace)
3487 {
3488 int ret = -1;
3489 struct evlist *evlist = trace->evlist;
3490 struct evsel *sys_enter, *sys_exit;
3491
3492 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
3493 if (sys_enter == NULL)
3494 goto out;
3495
3496 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
3497 goto out_delete_sys_enter;
3498
3499 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
3500 if (sys_exit == NULL)
3501 goto out_delete_sys_enter;
3502
3503 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
3504 goto out_delete_sys_exit;
3505
3506 evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
3507 evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
3508
3509 evlist__add(evlist, sys_enter);
3510 evlist__add(evlist, sys_exit);
3511
3512 if (callchain_param.enabled && !trace->kernel_syscallchains) {
3513 /*
3514 * We're interested only in the user space callchain
3515 * leading to the syscall, allow overriding that for
3516 * debugging reasons using --kernel_syscall_callchains
3517 */
3518 sys_exit->core.attr.exclude_callchain_kernel = 1;
3519 }
3520
3521 trace->syscalls.events.sys_enter = sys_enter;
3522 trace->syscalls.events.sys_exit = sys_exit;
3523
3524 ret = 0;
3525 out:
3526 return ret;
3527
3528 out_delete_sys_exit:
3529 evsel__delete_priv(sys_exit);
3530 out_delete_sys_enter:
3531 evsel__delete_priv(sys_enter);
3532 goto out;
3533 }
3534
3535 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
3536 {
3537 int err = -1;
3538 struct evsel *sys_exit;
3539 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
3540 trace->ev_qualifier_ids.nr,
3541 trace->ev_qualifier_ids.entries);
3542
3543 if (filter == NULL)
3544 goto out_enomem;
3545
3546 if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) {
3547 sys_exit = trace->syscalls.events.sys_exit;
3548 err = evsel__append_tp_filter(sys_exit, filter);
3549 }
3550
3551 free(filter);
3552 out:
3553 return err;
3554 out_enomem:
3555 errno = ENOMEM;
3556 goto out;
3557 }
3558
3559 #ifdef HAVE_BPF_SKEL
3560 static int syscall_arg_fmt__cache_btf_struct(struct syscall_arg_fmt *arg_fmt, struct btf *btf, char *type)
3561 {
3562 int id;
3563
3564 if (arg_fmt->type != NULL)
3565 return -1;
3566
3567 id = btf__find_by_name(btf, type);
3568 if (id < 0)
3569 return -1;
3570
3571 arg_fmt->type = btf__type_by_id(btf, id);
3572 arg_fmt->type_id = id;
3573
3574 return 0;
3575 }
3576
3577 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
3578 {
3579 struct bpf_program *pos, *prog = NULL;
3580 const char *sec_name;
3581
3582 if (trace->skel->obj == NULL)
3583 return NULL;
3584
3585 bpf_object__for_each_program(pos, trace->skel->obj) {
3586 sec_name = bpf_program__section_name(pos);
3587 if (sec_name && !strcmp(sec_name, name)) {
3588 prog = pos;
3589 break;
3590 }
3591 }
3592
3593 return prog;
3594 }
3595
3596 static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
3597 const char *prog_name, const char *type)
3598 {
3599 struct bpf_program *prog;
3600
3601 if (prog_name == NULL) {
3602 char default_prog_name[256];
3603 scnprintf(default_prog_name, sizeof(default_prog_name), "tp/syscalls/sys_%s_%s", type, sc->name);
3604 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3605 if (prog != NULL)
3606 goto out_found;
3607 if (sc->fmt && sc->fmt->alias) {
3608 scnprintf(default_prog_name, sizeof(default_prog_name), "tp/syscalls/sys_%s_%s", type, sc->fmt->alias);
3609 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3610 if (prog != NULL)
3611 goto out_found;
3612 }
3613 goto out_unaugmented;
3614 }
3615
3616 prog = trace__find_bpf_program_by_title(trace, prog_name);
3617
3618 if (prog != NULL) {
3619 out_found:
3620 return prog;
3621 }
3622
3623 pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
3624 prog_name, type, sc->name);
3625 out_unaugmented:
3626 return trace->skel->progs.syscall_unaugmented;
3627 }
3628
3629 static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
3630 {
3631 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3632
3633 if (sc == NULL)
3634 return;
3635
3636 sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3637 sc->bpf_prog.sys_exit = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit : NULL, "exit");
3638 }
3639
3640 static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
3641 {
3642 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3643 return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->skel->progs.syscall_unaugmented);
3644 }
3645
3646 static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
3647 {
3648 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3649 return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->skel->progs.syscall_unaugmented);
3650 }
3651
3652 static int trace__bpf_sys_enter_beauty_map(struct trace *trace, int key, unsigned int *beauty_array)
3653 {
3654 struct tep_format_field *field;
3655 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3656 const struct btf_type *bt;
3657 char *struct_offset, *tmp, name[32];
3658 bool can_augment = false;
3659 int i, cnt;
3660
3661 if (sc == NULL)
3662 return -1;
3663
3664 trace__load_vmlinux_btf(trace);
3665 if (trace->btf == NULL)
3666 return -1;
3667
3668 for (i = 0, field = sc->args; field; ++i, field = field->next) {
3669 // XXX We're only collecting pointer payloads _from_ user space
3670 if (!sc->arg_fmt[i].from_user)
3671 continue;
3672
3673 struct_offset = strstr(field->type, "struct ");
3674 if (struct_offset == NULL)
3675 struct_offset = strstr(field->type, "union ");
3676 else
3677 struct_offset++; // "union" is shorter
3678
3679 if (field->flags & TEP_FIELD_IS_POINTER && struct_offset) { /* struct or union (think BPF's attr arg) */
3680 struct_offset += 6;
3681
3682 /* for 'struct foo *', we only want 'foo' */
3683 for (tmp = struct_offset, cnt = 0; *tmp != ' ' && *tmp != '\0'; ++tmp, ++cnt) {
3684 }
3685
3686 strncpy(name, struct_offset, cnt);
3687 name[cnt] = '\0';
3688
3689 /* cache struct's btf_type and type_id */
3690 if (syscall_arg_fmt__cache_btf_struct(&sc->arg_fmt[i], trace->btf, name))
3691 continue;
3692
3693 bt = sc->arg_fmt[i].type;
3694 beauty_array[i] = bt->size;
3695 can_augment = true;
3696 } else if (field->flags & TEP_FIELD_IS_POINTER && /* string */
3697 strcmp(field->type, "const char *") == 0 &&
3698 (strstr(field->name, "name") ||
3699 strstr(field->name, "path") ||
3700 strstr(field->name, "file") ||
3701 strstr(field->name, "root") ||
3702 strstr(field->name, "key") ||
3703 strstr(field->name, "special") ||
3704 strstr(field->name, "type") ||
3705 strstr(field->name, "description"))) {
3706 beauty_array[i] = 1;
3707 can_augment = true;
3708 } else if (field->flags & TEP_FIELD_IS_POINTER && /* buffer */
3709 strstr(field->type, "char *") &&
3710 (strstr(field->name, "buf") ||
3711 strstr(field->name, "val") ||
3712 strstr(field->name, "msg"))) {
3713 int j;
3714 struct tep_format_field *field_tmp;
3715
3716 /* find the size of the buffer that appears in pairs with buf */
3717 for (j = 0, field_tmp = sc->args; field_tmp; ++j, field_tmp = field_tmp->next) {
3718 if (!(field_tmp->flags & TEP_FIELD_IS_POINTER) && /* only integers */
3719 (strstr(field_tmp->name, "count") ||
3720 strstr(field_tmp->name, "siz") || /* size, bufsiz */
3721 (strstr(field_tmp->name, "len") && strcmp(field_tmp->name, "filename")))) {
3722 /* filename's got 'len' in it, we don't want that */
3723 beauty_array[i] = -(j + 1);
3724 can_augment = true;
3725 break;
3726 }
3727 }
3728 }
3729 }
3730
3731 if (can_augment)
3732 return 0;
3733
3734 return -1;
3735 }
3736
3737 static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
3738 {
3739 struct tep_format_field *field, *candidate_field;
3740 /*
3741 * We're only interested in syscalls that have a pointer:
3742 */
3743 for (field = sc->args; field; field = field->next) {
3744 if (field->flags & TEP_FIELD_IS_POINTER)
3745 goto try_to_find_pair;
3746 }
3747
3748 return NULL;
3749
3750 try_to_find_pair:
3751 for (int i = 0; i < trace->sctbl->syscalls.nr_entries; ++i) {
3752 int id = syscalltbl__id_at_idx(trace->sctbl, i);
3753 struct syscall *pair = trace__syscall_info(trace, NULL, id);
3754 struct bpf_program *pair_prog;
3755 bool is_candidate = false;
3756
3757 if (pair == NULL || pair == sc ||
3758 pair->bpf_prog.sys_enter == trace->skel->progs.syscall_unaugmented)
3759 continue;
3760
3761 for (field = sc->args, candidate_field = pair->args;
3762 field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3763 bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3764 candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3765
3766 if (is_pointer) {
3767 if (!candidate_is_pointer) {
3768 // The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3769 continue;
3770 }
3771 } else {
3772 if (candidate_is_pointer) {
3773 // The candidate might copy a pointer we don't have, skip it.
3774 goto next_candidate;
3775 }
3776 continue;
3777 }
3778
3779 if (strcmp(field->type, candidate_field->type))
3780 goto next_candidate;
3781
3782 /*
3783 * This is limited in the BPF program but sys_write
3784 * uses "const char *" for its "buf" arg so we need to
3785 * use some heuristic that is kinda future proof...
3786 */
3787 if (strcmp(field->type, "const char *") == 0 &&
3788 !(strstr(field->name, "name") ||
3789 strstr(field->name, "path") ||
3790 strstr(field->name, "file") ||
3791 strstr(field->name, "root") ||
3792 strstr(field->name, "description")))
3793 goto next_candidate;
3794
3795 is_candidate = true;
3796 }
3797
3798 if (!is_candidate)
3799 goto next_candidate;
3800
3801 /*
3802 * Check if the tentative pair syscall augmenter has more pointers, if it has,
3803 * then it may be collecting that and we then can't use it, as it would collect
3804 * more than what is common to the two syscalls.
3805 */
3806 if (candidate_field) {
3807 for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3808 if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3809 goto next_candidate;
3810 }
3811
3812 pair_prog = pair->bpf_prog.sys_enter;
3813 /*
3814 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3815 * have been searched for, so search it here and if it returns the
3816 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3817 * program for a filtered syscall on a non-filtered one.
3818 *
3819 * For instance, we have "!syscalls:sys_enter_renameat" and that is
3820 * useful for "renameat2".
3821 */
3822 if (pair_prog == NULL) {
3823 pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3824 if (pair_prog == trace->skel->progs.syscall_unaugmented)
3825 goto next_candidate;
3826 }
3827
3828 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3829 return pair_prog;
3830 next_candidate:
3831 continue;
3832 }
3833
3834 return NULL;
3835 }
3836
3837 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3838 {
3839 int map_enter_fd = bpf_map__fd(trace->skel->maps.syscalls_sys_enter);
3840 int map_exit_fd = bpf_map__fd(trace->skel->maps.syscalls_sys_exit);
3841 int beauty_map_fd = bpf_map__fd(trace->skel->maps.beauty_map_enter);
3842 int err = 0;
3843 unsigned int beauty_array[6];
3844
3845 for (int i = 0; i < trace->sctbl->syscalls.nr_entries; ++i) {
3846 int prog_fd, key = syscalltbl__id_at_idx(trace->sctbl, i);
3847
3848 if (!trace__syscall_enabled(trace, key))
3849 continue;
3850
3851 trace__init_syscall_bpf_progs(trace, key);
3852
3853 // It'll get at least the "!raw_syscalls:unaugmented"
3854 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3855 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3856 if (err)
3857 break;
3858 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3859 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3860 if (err)
3861 break;
3862
3863 /* use beauty_map to tell BPF how many bytes to collect, set beauty_map's value here */
3864 memset(beauty_array, 0, sizeof(beauty_array));
3865 err = trace__bpf_sys_enter_beauty_map(trace, key, (unsigned int *)beauty_array);
3866 if (err)
3867 continue;
3868 err = bpf_map_update_elem(beauty_map_fd, &key, beauty_array, BPF_ANY);
3869 if (err)
3870 break;
3871 }
3872
3873 /*
3874 * Now lets do a second pass looking for enabled syscalls without
3875 * an augmenter that have a signature that is a superset of another
3876 * syscall with an augmenter so that we can auto-reuse it.
3877 *
3878 * I.e. if we have an augmenter for the "open" syscall that has
3879 * this signature:
3880 *
3881 * int open(const char *pathname, int flags, mode_t mode);
3882 *
3883 * I.e. that will collect just the first string argument, then we
3884 * can reuse it for the 'creat' syscall, that has this signature:
3885 *
3886 * int creat(const char *pathname, mode_t mode);
3887 *
3888 * and for:
3889 *
3890 * int stat(const char *pathname, struct stat *statbuf);
3891 * int lstat(const char *pathname, struct stat *statbuf);
3892 *
3893 * Because the 'open' augmenter will collect the first arg as a string,
3894 * and leave alone all the other args, which already helps with
3895 * beautifying 'stat' and 'lstat''s pathname arg.
3896 *
3897 * Then, in time, when 'stat' gets an augmenter that collects both
3898 * first and second arg (this one on the raw_syscalls:sys_exit prog
3899 * array tail call, then that one will be used.
3900 */
3901 for (int i = 0; i < trace->sctbl->syscalls.nr_entries; ++i) {
3902 int key = syscalltbl__id_at_idx(trace->sctbl, i);
3903 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3904 struct bpf_program *pair_prog;
3905 int prog_fd;
3906
3907 if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3908 continue;
3909
3910 /*
3911 * For now we're just reusing the sys_enter prog, and if it
3912 * already has an augmenter, we don't need to find one.
3913 */
3914 if (sc->bpf_prog.sys_enter != trace->skel->progs.syscall_unaugmented)
3915 continue;
3916
3917 /*
3918 * Look at all the other syscalls for one that has a signature
3919 * that is close enough that we can share:
3920 */
3921 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3922 if (pair_prog == NULL)
3923 continue;
3924
3925 sc->bpf_prog.sys_enter = pair_prog;
3926
3927 /*
3928 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3929 * with the fd for the program we're reusing:
3930 */
3931 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3932 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3933 if (err)
3934 break;
3935 }
3936
3937 return err;
3938 }
3939 #endif // HAVE_BPF_SKEL
3940
3941 static int trace__set_ev_qualifier_filter(struct trace *trace)
3942 {
3943 if (trace->syscalls.events.sys_enter)
3944 return trace__set_ev_qualifier_tp_filter(trace);
3945 return 0;
3946 }
3947
3948 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3949 size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3950 {
3951 int err = 0;
3952 #ifdef HAVE_LIBBPF_SUPPORT
3953 bool value = true;
3954 int map_fd = bpf_map__fd(map);
3955 size_t i;
3956
3957 for (i = 0; i < npids; ++i) {
3958 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3959 if (err)
3960 break;
3961 }
3962 #endif
3963 return err;
3964 }
3965
3966 static int trace__set_filter_loop_pids(struct trace *trace)
3967 {
3968 unsigned int nr = 1, err;
3969 pid_t pids[32] = {
3970 getpid(),
3971 };
3972 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3973
3974 while (thread && nr < ARRAY_SIZE(pids)) {
3975 struct thread *parent = machine__find_thread(trace->host,
3976 thread__ppid(thread),
3977 thread__ppid(thread));
3978
3979 if (parent == NULL)
3980 break;
3981
3982 if (!strcmp(thread__comm_str(parent), "sshd") ||
3983 strstarts(thread__comm_str(parent), "gnome-terminal")) {
3984 pids[nr++] = thread__tid(parent);
3985 break;
3986 }
3987 thread = parent;
3988 }
3989
3990 err = evlist__append_tp_filter_pids(trace->evlist, nr, pids);
3991 if (!err && trace->filter_pids.map)
3992 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3993
3994 return err;
3995 }
3996
3997 static int trace__set_filter_pids(struct trace *trace)
3998 {
3999 int err = 0;
4000 /*
4001 * Better not use !target__has_task() here because we need to cover the
4002 * case where no threads were specified in the command line, but a
4003 * workload was, and in that case we will fill in the thread_map when
4004 * we fork the workload in evlist__prepare_workload.
4005 */
4006 if (trace->filter_pids.nr > 0) {
4007 err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
4008 trace->filter_pids.entries);
4009 if (!err && trace->filter_pids.map) {
4010 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
4011 trace->filter_pids.entries);
4012 }
4013 } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
4014 err = trace__set_filter_loop_pids(trace);
4015 }
4016
4017 return err;
4018 }
4019
4020 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
4021 {
4022 struct evlist *evlist = trace->evlist;
4023 struct perf_sample sample;
4024 int err = evlist__parse_sample(evlist, event, &sample);
4025
4026 if (err)
4027 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
4028 else
4029 trace__handle_event(trace, event, &sample);
4030
4031 return 0;
4032 }
4033
4034 static int __trace__flush_events(struct trace *trace)
4035 {
4036 u64 first = ordered_events__first_time(&trace->oe.data);
4037 u64 flush = trace->oe.last - NSEC_PER_SEC;
4038
4039 /* Is there some thing to flush.. */
4040 if (first && first < flush)
4041 return ordered_events__flush_time(&trace->oe.data, flush);
4042
4043 return 0;
4044 }
4045
4046 static int trace__flush_events(struct trace *trace)
4047 {
4048 return !trace->sort_events ? 0 : __trace__flush_events(trace);
4049 }
4050
4051 static int trace__deliver_event(struct trace *trace, union perf_event *event)
4052 {
4053 int err;
4054
4055 if (!trace->sort_events)
4056 return __trace__deliver_event(trace, event);
4057
4058 err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
4059 if (err && err != -1)
4060 return err;
4061
4062 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0, NULL);
4063 if (err)
4064 return err;
4065
4066 return trace__flush_events(trace);
4067 }
4068
4069 static int ordered_events__deliver_event(struct ordered_events *oe,
4070 struct ordered_event *event)
4071 {
4072 struct trace *trace = container_of(oe, struct trace, oe.data);
4073
4074 return __trace__deliver_event(trace, event->event);
4075 }
4076
4077 static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg,
4078 char **type)
4079 {
4080 struct tep_format_field *field;
4081 struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel);
4082
4083 if (evsel->tp_format == NULL || fmt == NULL)
4084 return NULL;
4085
4086 for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt)
4087 if (strcmp(field->name, arg) == 0) {
4088 *type = field->type;
4089 return fmt;
4090 }
4091
4092 return NULL;
4093 }
4094
4095 static int trace__expand_filter(struct trace *trace, struct evsel *evsel)
4096 {
4097 char *tok, *left = evsel->filter, *new_filter = evsel->filter;
4098
4099 while ((tok = strpbrk(left, "=<>!")) != NULL) {
4100 char *right = tok + 1, *right_end;
4101
4102 if (*right == '=')
4103 ++right;
4104
4105 while (isspace(*right))
4106 ++right;
4107
4108 if (*right == '\0')
4109 break;
4110
4111 while (!isalpha(*left))
4112 if (++left == tok) {
4113 /*
4114 * Bail out, can't find the name of the argument that is being
4115 * used in the filter, let it try to set this filter, will fail later.
4116 */
4117 return 0;
4118 }
4119
4120 right_end = right + 1;
4121 while (isalnum(*right_end) || *right_end == '_' || *right_end == '|')
4122 ++right_end;
4123
4124 if (isalpha(*right)) {
4125 struct syscall_arg_fmt *fmt;
4126 int left_size = tok - left,
4127 right_size = right_end - right;
4128 char arg[128], *type;
4129
4130 while (isspace(left[left_size - 1]))
4131 --left_size;
4132
4133 scnprintf(arg, sizeof(arg), "%.*s", left_size, left);
4134
4135 fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg, &type);
4136 if (fmt == NULL) {
4137 pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n",
4138 arg, evsel->name, evsel->filter);
4139 return -1;
4140 }
4141
4142 pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ",
4143 arg, (int)(right - tok), tok, right_size, right);
4144
4145 if (fmt->strtoul) {
4146 u64 val;
4147 struct syscall_arg syscall_arg = {
4148 .trace = trace,
4149 .fmt = fmt,
4150 .type_name = type,
4151 .parm = fmt->parm,
4152 };
4153
4154 if (fmt->strtoul(right, right_size, &syscall_arg, &val)) {
4155 char *n, expansion[19];
4156 int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val);
4157 int expansion_offset = right - new_filter;
4158
4159 pr_debug("%s", expansion);
4160
4161 if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) {
4162 pr_debug(" out of memory!\n");
4163 free(new_filter);
4164 return -1;
4165 }
4166 if (new_filter != evsel->filter)
4167 free(new_filter);
4168 left = n + expansion_offset + expansion_lenght;
4169 new_filter = n;
4170 } else {
4171 pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n",
4172 right_size, right, arg, evsel->name, evsel->filter);
4173 return -1;
4174 }
4175 } else {
4176 pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n",
4177 arg, evsel->name, evsel->filter);
4178 return -1;
4179 }
4180
4181 pr_debug("\n");
4182 } else {
4183 left = right_end;
4184 }
4185 }
4186
4187 if (new_filter != evsel->filter) {
4188 pr_debug("New filter for %s: %s\n", evsel->name, new_filter);
4189 evsel__set_filter(evsel, new_filter);
4190 free(new_filter);
4191 }
4192
4193 return 0;
4194 }
4195
4196 static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel)
4197 {
4198 struct evlist *evlist = trace->evlist;
4199 struct evsel *evsel;
4200
4201 evlist__for_each_entry(evlist, evsel) {
4202 if (evsel->filter == NULL)
4203 continue;
4204
4205 if (trace__expand_filter(trace, evsel)) {
4206 *err_evsel = evsel;
4207 return -1;
4208 }
4209 }
4210
4211 return 0;
4212 }
4213
4214 static int trace__run(struct trace *trace, int argc, const char **argv)
4215 {
4216 struct evlist *evlist = trace->evlist;
4217 struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
4218 int err = -1, i;
4219 unsigned long before;
4220 const bool forks = argc > 0;
4221 bool draining = false;
4222
4223 trace->live = true;
4224
4225 if (!trace->raw_augmented_syscalls) {
4226 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
4227 goto out_error_raw_syscalls;
4228
4229 if (trace->trace_syscalls)
4230 trace->vfs_getname = evlist__add_vfs_getname(evlist);
4231 }
4232
4233 if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
4234 pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
4235 if (pgfault_maj == NULL)
4236 goto out_error_mem;
4237 evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
4238 evlist__add(evlist, pgfault_maj);
4239 }
4240
4241 if ((trace->trace_pgfaults & TRACE_PFMIN)) {
4242 pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
4243 if (pgfault_min == NULL)
4244 goto out_error_mem;
4245 evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
4246 evlist__add(evlist, pgfault_min);
4247 }
4248
4249 /* Enable ignoring missing threads when -u/-p option is defined. */
4250 trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid;
4251
4252 if (trace->sched &&
4253 evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime))
4254 goto out_error_sched_stat_runtime;
4255 /*
4256 * If a global cgroup was set, apply it to all the events without an
4257 * explicit cgroup. I.e.:
4258 *
4259 * trace -G A -e sched:*switch
4260 *
4261 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
4262 * _and_ sched:sched_switch to the 'A' cgroup, while:
4263 *
4264 * trace -e sched:*switch -G A
4265 *
4266 * will only set the sched:sched_switch event to the 'A' cgroup, all the
4267 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
4268 * a cgroup (on the root cgroup, sys wide, etc).
4269 *
4270 * Multiple cgroups:
4271 *
4272 * trace -G A -e sched:*switch -G B
4273 *
4274 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
4275 * to the 'B' cgroup.
4276 *
4277 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
4278 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
4279 */
4280 if (trace->cgroup)
4281 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
4282
4283 err = evlist__create_maps(evlist, &trace->opts.target);
4284 if (err < 0) {
4285 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
4286 goto out_delete_evlist;
4287 }
4288
4289 err = trace__symbols_init(trace, evlist);
4290 if (err < 0) {
4291 fprintf(trace->output, "Problems initializing symbol libraries!\n");
4292 goto out_delete_evlist;
4293 }
4294
4295 evlist__config(evlist, &trace->opts, &callchain_param);
4296
4297 if (forks) {
4298 err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL);
4299 if (err < 0) {
4300 fprintf(trace->output, "Couldn't run the workload!\n");
4301 goto out_delete_evlist;
4302 }
4303 workload_pid = evlist->workload.pid;
4304 }
4305
4306 err = evlist__open(evlist);
4307 if (err < 0)
4308 goto out_error_open;
4309 #ifdef HAVE_BPF_SKEL
4310 if (trace->syscalls.events.bpf_output) {
4311 struct perf_cpu cpu;
4312
4313 /*
4314 * Set up the __augmented_syscalls__ BPF map to hold for each
4315 * CPU the bpf-output event's file descriptor.
4316 */
4317 perf_cpu_map__for_each_cpu(cpu, i, trace->syscalls.events.bpf_output->core.cpus) {
4318 bpf_map__update_elem(trace->skel->maps.__augmented_syscalls__,
4319 &cpu.cpu, sizeof(int),
4320 xyarray__entry(trace->syscalls.events.bpf_output->core.fd,
4321 cpu.cpu, 0),
4322 sizeof(__u32), BPF_ANY);
4323 }
4324 }
4325
4326 if (trace->skel)
4327 trace->filter_pids.map = trace->skel->maps.pids_filtered;
4328 #endif
4329 err = trace__set_filter_pids(trace);
4330 if (err < 0)
4331 goto out_error_mem;
4332
4333 #ifdef HAVE_BPF_SKEL
4334 if (trace->skel && trace->skel->progs.sys_enter)
4335 trace__init_syscalls_bpf_prog_array_maps(trace);
4336 #endif
4337
4338 if (trace->ev_qualifier_ids.nr > 0) {
4339 err = trace__set_ev_qualifier_filter(trace);
4340 if (err < 0)
4341 goto out_errno;
4342
4343 if (trace->syscalls.events.sys_exit) {
4344 pr_debug("event qualifier tracepoint filter: %s\n",
4345 trace->syscalls.events.sys_exit->filter);
4346 }
4347 }
4348
4349 /*
4350 * If the "close" syscall is not traced, then we will not have the
4351 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
4352 * fd->pathname table and were ending up showing the last value set by
4353 * syscalls opening a pathname and associating it with a descriptor or
4354 * reading it from /proc/pid/fd/ in cases where that doesn't make
4355 * sense.
4356 *
4357 * So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
4358 * not in use.
4359 */
4360 trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
4361
4362 err = trace__expand_filters(trace, &evsel);
4363 if (err)
4364 goto out_delete_evlist;
4365 err = evlist__apply_filters(evlist, &evsel, &trace->opts.target);
4366 if (err < 0)
4367 goto out_error_apply_filters;
4368
4369 err = evlist__mmap(evlist, trace->opts.mmap_pages);
4370 if (err < 0)
4371 goto out_error_mmap;
4372
4373 if (!target__none(&trace->opts.target) && !trace->opts.target.initial_delay)
4374 evlist__enable(evlist);
4375
4376 if (forks)
4377 evlist__start_workload(evlist);
4378
4379 if (trace->opts.target.initial_delay) {
4380 usleep(trace->opts.target.initial_delay * 1000);
4381 evlist__enable(evlist);
4382 }
4383
4384 trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
4385 perf_thread_map__nr(evlist->core.threads) > 1 ||
4386 evlist__first(evlist)->core.attr.inherit;
4387
4388 /*
4389 * Now that we already used evsel->core.attr to ask the kernel to setup the
4390 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
4391 * trace__resolve_callchain(), allowing per-event max-stack settings
4392 * to override an explicitly set --max-stack global setting.
4393 */
4394 evlist__for_each_entry(evlist, evsel) {
4395 if (evsel__has_callchain(evsel) &&
4396 evsel->core.attr.sample_max_stack == 0)
4397 evsel->core.attr.sample_max_stack = trace->max_stack;
4398 }
4399 again:
4400 before = trace->nr_events;
4401
4402 for (i = 0; i < evlist->core.nr_mmaps; i++) {
4403 union perf_event *event;
4404 struct mmap *md;
4405
4406 md = &evlist->mmap[i];
4407 if (perf_mmap__read_init(&md->core) < 0)
4408 continue;
4409
4410 while ((event = perf_mmap__read_event(&md->core)) != NULL) {
4411 ++trace->nr_events;
4412
4413 err = trace__deliver_event(trace, event);
4414 if (err)
4415 goto out_disable;
4416
4417 perf_mmap__consume(&md->core);
4418
4419 if (interrupted)
4420 goto out_disable;
4421
4422 if (done && !draining) {
4423 evlist__disable(evlist);
4424 draining = true;
4425 }
4426 }
4427 perf_mmap__read_done(&md->core);
4428 }
4429
4430 if (trace->nr_events == before) {
4431 int timeout = done ? 100 : -1;
4432
4433 if (!draining && evlist__poll(evlist, timeout) > 0) {
4434 if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
4435 draining = true;
4436
4437 goto again;
4438 } else {
4439 if (trace__flush_events(trace))
4440 goto out_disable;
4441 }
4442 } else {
4443 goto again;
4444 }
4445
4446 out_disable:
4447 thread__zput(trace->current);
4448
4449 evlist__disable(evlist);
4450
4451 if (trace->sort_events)
4452 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
4453
4454 if (!err) {
4455 if (trace->summary)
4456 trace__fprintf_thread_summary(trace, trace->output);
4457
4458 if (trace->show_tool_stats) {
4459 fprintf(trace->output, "Stats:\n "
4460 " vfs_getname : %" PRIu64 "\n"
4461 " proc_getname: %" PRIu64 "\n",
4462 trace->stats.vfs_getname,
4463 trace->stats.proc_getname);
4464 }
4465 }
4466
4467 out_delete_evlist:
4468 trace__symbols__exit(trace);
4469 evlist__free_syscall_tp_fields(evlist);
4470 evlist__delete(evlist);
4471 cgroup__put(trace->cgroup);
4472 trace->evlist = NULL;
4473 trace->live = false;
4474 return err;
4475 {
4476 char errbuf[BUFSIZ];
4477
4478 out_error_sched_stat_runtime:
4479 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
4480 goto out_error;
4481
4482 out_error_raw_syscalls:
4483 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
4484 goto out_error;
4485
4486 out_error_mmap:
4487 evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
4488 goto out_error;
4489
4490 out_error_open:
4491 evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
4492
4493 out_error:
4494 fprintf(trace->output, "%s\n", errbuf);
4495 goto out_delete_evlist;
4496
4497 out_error_apply_filters:
4498 fprintf(trace->output,
4499 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
4500 evsel->filter, evsel__name(evsel), errno,
4501 str_error_r(errno, errbuf, sizeof(errbuf)));
4502 goto out_delete_evlist;
4503 }
4504 out_error_mem:
4505 fprintf(trace->output, "Not enough memory to run!\n");
4506 goto out_delete_evlist;
4507
4508 out_errno:
4509 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
4510 goto out_delete_evlist;
4511 }
4512
4513 static int trace__replay(struct trace *trace)
4514 {
4515 const struct evsel_str_handler handlers[] = {
4516 { "probe:vfs_getname", trace__vfs_getname, },
4517 };
4518 struct perf_data data = {
4519 .path = input_name,
4520 .mode = PERF_DATA_MODE_READ,
4521 .force = trace->force,
4522 };
4523 struct perf_session *session;
4524 struct evsel *evsel;
4525 int err = -1;
4526
4527 trace->tool.sample = trace__process_sample;
4528 trace->tool.mmap = perf_event__process_mmap;
4529 trace->tool.mmap2 = perf_event__process_mmap2;
4530 trace->tool.comm = perf_event__process_comm;
4531 trace->tool.exit = perf_event__process_exit;
4532 trace->tool.fork = perf_event__process_fork;
4533 trace->tool.attr = perf_event__process_attr;
4534 trace->tool.tracing_data = perf_event__process_tracing_data;
4535 trace->tool.build_id = perf_event__process_build_id;
4536 trace->tool.namespaces = perf_event__process_namespaces;
4537
4538 trace->tool.ordered_events = true;
4539 trace->tool.ordering_requires_timestamps = true;
4540
4541 /* add tid to output */
4542 trace->multiple_threads = true;
4543
4544 session = perf_session__new(&data, &trace->tool);
4545 if (IS_ERR(session))
4546 return PTR_ERR(session);
4547
4548 if (trace->opts.target.pid)
4549 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
4550
4551 if (trace->opts.target.tid)
4552 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
4553
4554 if (symbol__init(&session->header.env) < 0)
4555 goto out;
4556
4557 trace->host = &session->machines.host;
4558
4559 err = perf_session__set_tracepoints_handlers(session, handlers);
4560 if (err)
4561 goto out;
4562
4563 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter");
4564 trace->syscalls.events.sys_enter = evsel;
4565 /* older kernels have syscalls tp versus raw_syscalls */
4566 if (evsel == NULL)
4567 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter");
4568
4569 if (evsel &&
4570 (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
4571 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
4572 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
4573 goto out;
4574 }
4575
4576 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit");
4577 trace->syscalls.events.sys_exit = evsel;
4578 if (evsel == NULL)
4579 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit");
4580 if (evsel &&
4581 (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
4582 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
4583 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
4584 goto out;
4585 }
4586
4587 evlist__for_each_entry(session->evlist, evsel) {
4588 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
4589 (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
4590 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
4591 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
4592 evsel->handler = trace__pgfault;
4593 }
4594
4595 setup_pager();
4596
4597 err = perf_session__process_events(session);
4598 if (err)
4599 pr_err("Failed to process events, error %d", err);
4600
4601 else if (trace->summary)
4602 trace__fprintf_thread_summary(trace, trace->output);
4603
4604 out:
4605 perf_session__delete(session);
4606
4607 return err;
4608 }
4609
4610 static size_t trace__fprintf_threads_header(FILE *fp)
4611 {
4612 size_t printed;
4613
4614 printed = fprintf(fp, "\n Summary of events:\n\n");
4615
4616 return printed;
4617 }
4618
4619 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
4620 struct syscall_stats *stats;
4621 double msecs;
4622 int syscall;
4623 )
4624 {
4625 struct int_node *source = rb_entry(nd, struct int_node, rb_node);
4626 struct syscall_stats *stats = source->priv;
4627
4628 entry->syscall = source->i;
4629 entry->stats = stats;
4630 entry->msecs = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0;
4631 }
4632
4633 static size_t thread__dump_stats(struct thread_trace *ttrace,
4634 struct trace *trace, FILE *fp)
4635 {
4636 size_t printed = 0;
4637 struct syscall *sc;
4638 struct rb_node *nd;
4639 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
4640
4641 if (syscall_stats == NULL)
4642 return 0;
4643
4644 printed += fprintf(fp, "\n");
4645
4646 printed += fprintf(fp, " syscall calls errors total min avg max stddev\n");
4647 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
4648 printed += fprintf(fp, " --------------- -------- ------ -------- --------- --------- --------- ------\n");
4649
4650 resort_rb__for_each_entry(nd, syscall_stats) {
4651 struct syscall_stats *stats = syscall_stats_entry->stats;
4652 if (stats) {
4653 double min = (double)(stats->stats.min) / NSEC_PER_MSEC;
4654 double max = (double)(stats->stats.max) / NSEC_PER_MSEC;
4655 double avg = avg_stats(&stats->stats);
4656 double pct;
4657 u64 n = (u64)stats->stats.n;
4658
4659 pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0;
4660 avg /= NSEC_PER_MSEC;
4661
4662 sc = &trace->syscalls.table[syscall_stats_entry->syscall];
4663 printed += fprintf(fp, " %-15s", sc->name);
4664 printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f",
4665 n, stats->nr_failures, syscall_stats_entry->msecs, min, avg);
4666 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
4667
4668 if (trace->errno_summary && stats->nr_failures) {
4669 int e;
4670
4671 for (e = 0; e < stats->max_errno; ++e) {
4672 if (stats->errnos[e] != 0)
4673 fprintf(fp, "\t\t\t\t%s: %d\n", perf_env__arch_strerrno(trace->host->env, e + 1), stats->errnos[e]);
4674 }
4675 }
4676 }
4677 }
4678
4679 resort_rb__delete(syscall_stats);
4680 printed += fprintf(fp, "\n\n");
4681
4682 return printed;
4683 }
4684
4685 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
4686 {
4687 size_t printed = 0;
4688 struct thread_trace *ttrace = thread__priv(thread);
4689 double ratio;
4690
4691 if (ttrace == NULL)
4692 return 0;
4693
4694 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
4695
4696 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread__tid(thread));
4697 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
4698 printed += fprintf(fp, "%.1f%%", ratio);
4699 if (ttrace->pfmaj)
4700 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
4701 if (ttrace->pfmin)
4702 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
4703 if (trace->sched)
4704 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
4705 else if (fputc('\n', fp) != EOF)
4706 ++printed;
4707
4708 printed += thread__dump_stats(ttrace, trace, fp);
4709
4710 return printed;
4711 }
4712
4713 static unsigned long thread__nr_events(struct thread_trace *ttrace)
4714 {
4715 return ttrace ? ttrace->nr_events : 0;
4716 }
4717
4718 static int trace_nr_events_cmp(void *priv __maybe_unused,
4719 const struct list_head *la,
4720 const struct list_head *lb)
4721 {
4722 struct thread_list *a = list_entry(la, struct thread_list, list);
4723 struct thread_list *b = list_entry(lb, struct thread_list, list);
4724 unsigned long a_nr_events = thread__nr_events(thread__priv(a->thread));
4725 unsigned long b_nr_events = thread__nr_events(thread__priv(b->thread));
4726
4727 if (a_nr_events != b_nr_events)
4728 return a_nr_events < b_nr_events ? -1 : 1;
4729
4730 /* Identical number of threads, place smaller tids first. */
4731 return thread__tid(a->thread) < thread__tid(b->thread)
4732 ? -1
4733 : (thread__tid(a->thread) > thread__tid(b->thread) ? 1 : 0);
4734 }
4735
4736 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
4737 {
4738 size_t printed = trace__fprintf_threads_header(fp);
4739 LIST_HEAD(threads);
4740
4741 if (machine__thread_list(trace->host, &threads) == 0) {
4742 struct thread_list *pos;
4743
4744 list_sort(NULL, &threads, trace_nr_events_cmp);
4745
4746 list_for_each_entry(pos, &threads, list)
4747 printed += trace__fprintf_thread(fp, pos->thread, trace);
4748 }
4749 thread_list__delete(&threads);
4750 return printed;
4751 }
4752
4753 static int trace__set_duration(const struct option *opt, const char *str,
4754 int unset __maybe_unused)
4755 {
4756 struct trace *trace = opt->value;
4757
4758 trace->duration_filter = atof(str);
4759 return 0;
4760 }
4761
4762 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
4763 int unset __maybe_unused)
4764 {
4765 int ret = -1;
4766 size_t i;
4767 struct trace *trace = opt->value;
4768 /*
4769 * FIXME: introduce a intarray class, plain parse csv and create a
4770 * { int nr, int entries[] } struct...
4771 */
4772 struct intlist *list = intlist__new(str);
4773
4774 if (list == NULL)
4775 return -1;
4776
4777 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
4778 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
4779
4780 if (trace->filter_pids.entries == NULL)
4781 goto out;
4782
4783 trace->filter_pids.entries[0] = getpid();
4784
4785 for (i = 1; i < trace->filter_pids.nr; ++i)
4786 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
4787
4788 intlist__delete(list);
4789 ret = 0;
4790 out:
4791 return ret;
4792 }
4793
4794 static int trace__open_output(struct trace *trace, const char *filename)
4795 {
4796 struct stat st;
4797
4798 if (!stat(filename, &st) && st.st_size) {
4799 char oldname[PATH_MAX];
4800
4801 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
4802 unlink(oldname);
4803 rename(filename, oldname);
4804 }
4805
4806 trace->output = fopen(filename, "w");
4807
4808 return trace->output == NULL ? -errno : 0;
4809 }
4810
4811 static int parse_pagefaults(const struct option *opt, const char *str,
4812 int unset __maybe_unused)
4813 {
4814 int *trace_pgfaults = opt->value;
4815
4816 if (strcmp(str, "all") == 0)
4817 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
4818 else if (strcmp(str, "maj") == 0)
4819 *trace_pgfaults |= TRACE_PFMAJ;
4820 else if (strcmp(str, "min") == 0)
4821 *trace_pgfaults |= TRACE_PFMIN;
4822 else
4823 return -1;
4824
4825 return 0;
4826 }
4827
4828 static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)
4829 {
4830 struct evsel *evsel;
4831
4832 evlist__for_each_entry(evlist, evsel) {
4833 if (evsel->handler == NULL)
4834 evsel->handler = handler;
4835 }
4836 }
4837
4838 static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name)
4839 {
4840 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
4841
4842 if (fmt) {
4843 const struct syscall_fmt *scfmt = syscall_fmt__find(name);
4844
4845 if (scfmt) {
4846 int skip = 0;
4847
4848 if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 ||
4849 strcmp(evsel->tp_format->format.fields->name, "nr") == 0)
4850 ++skip;
4851
4852 memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt));
4853 }
4854 }
4855 }
4856
4857 static int evlist__set_syscall_tp_fields(struct evlist *evlist, bool *use_btf)
4858 {
4859 struct evsel *evsel;
4860
4861 evlist__for_each_entry(evlist, evsel) {
4862 if (evsel->priv || !evsel->tp_format)
4863 continue;
4864
4865 if (strcmp(evsel->tp_format->system, "syscalls")) {
4866 evsel__init_tp_arg_scnprintf(evsel, use_btf);
4867 continue;
4868 }
4869
4870 if (evsel__init_syscall_tp(evsel))
4871 return -1;
4872
4873 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
4874 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4875
4876 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
4877 return -1;
4878
4879 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1);
4880 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
4881 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4882
4883 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
4884 return -1;
4885
4886 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1);
4887 }
4888 }
4889
4890 return 0;
4891 }
4892
4893 /*
4894 * XXX: Hackish, just splitting the combined -e+--event (syscalls
4895 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
4896 * existing facilities unchanged (trace->ev_qualifier + parse_options()).
4897 *
4898 * It'd be better to introduce a parse_options() variant that would return a
4899 * list with the terms it didn't match to an event...
4900 */
4901 static int trace__parse_events_option(const struct option *opt, const char *str,
4902 int unset __maybe_unused)
4903 {
4904 struct trace *trace = (struct trace *)opt->value;
4905 const char *s = str;
4906 char *sep = NULL, *lists[2] = { NULL, NULL, };
4907 int len = strlen(str) + 1, err = -1, list, idx;
4908 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
4909 char group_name[PATH_MAX];
4910 const struct syscall_fmt *fmt;
4911
4912 if (strace_groups_dir == NULL)
4913 return -1;
4914
4915 if (*s == '!') {
4916 ++s;
4917 trace->not_ev_qualifier = true;
4918 }
4919
4920 while (1) {
4921 if ((sep = strchr(s, ',')) != NULL)
4922 *sep = '\0';
4923
4924 list = 0;
4925 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
4926 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
4927 list = 1;
4928 goto do_concat;
4929 }
4930
4931 fmt = syscall_fmt__find_by_alias(s);
4932 if (fmt != NULL) {
4933 list = 1;
4934 s = fmt->name;
4935 } else {
4936 path__join(group_name, sizeof(group_name), strace_groups_dir, s);
4937 if (access(group_name, R_OK) == 0)
4938 list = 1;
4939 }
4940 do_concat:
4941 if (lists[list]) {
4942 sprintf(lists[list] + strlen(lists[list]), ",%s", s);
4943 } else {
4944 lists[list] = malloc(len);
4945 if (lists[list] == NULL)
4946 goto out;
4947 strcpy(lists[list], s);
4948 }
4949
4950 if (!sep)
4951 break;
4952
4953 *sep = ',';
4954 s = sep + 1;
4955 }
4956
4957 if (lists[1] != NULL) {
4958 struct strlist_config slist_config = {
4959 .dirname = strace_groups_dir,
4960 };
4961
4962 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4963 if (trace->ev_qualifier == NULL) {
4964 fputs("Not enough memory to parse event qualifier", trace->output);
4965 goto out;
4966 }
4967
4968 if (trace__validate_ev_qualifier(trace))
4969 goto out;
4970 trace->trace_syscalls = true;
4971 }
4972
4973 err = 0;
4974
4975 if (lists[0]) {
4976 struct parse_events_option_args parse_events_option_args = {
4977 .evlistp = &trace->evlist,
4978 };
4979 struct option o = {
4980 .value = &parse_events_option_args,
4981 };
4982 err = parse_events_option(&o, lists[0], 0);
4983 }
4984 out:
4985 free(strace_groups_dir);
4986 free(lists[0]);
4987 free(lists[1]);
4988 if (sep)
4989 *sep = ',';
4990
4991 return err;
4992 }
4993
4994 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4995 {
4996 struct trace *trace = opt->value;
4997
4998 if (!list_empty(&trace->evlist->core.entries)) {
4999 struct option o = {
5000 .value = &trace->evlist,
5001 };
5002 return parse_cgroups(&o, str, unset);
5003 }
5004 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
5005
5006 return 0;
5007 }
5008
5009 static int trace__config(const char *var, const char *value, void *arg)
5010 {
5011 struct trace *trace = arg;
5012 int err = 0;
5013
5014 if (!strcmp(var, "trace.add_events")) {
5015 trace->perfconfig_events = strdup(value);
5016 if (trace->perfconfig_events == NULL) {
5017 pr_err("Not enough memory for %s\n", "trace.add_events");
5018 return -1;
5019 }
5020 } else if (!strcmp(var, "trace.show_timestamp")) {
5021 trace->show_tstamp = perf_config_bool(var, value);
5022 } else if (!strcmp(var, "trace.show_duration")) {
5023 trace->show_duration = perf_config_bool(var, value);
5024 } else if (!strcmp(var, "trace.show_arg_names")) {
5025 trace->show_arg_names = perf_config_bool(var, value);
5026 if (!trace->show_arg_names)
5027 trace->show_zeros = true;
5028 } else if (!strcmp(var, "trace.show_zeros")) {
5029 bool new_show_zeros = perf_config_bool(var, value);
5030 if (!trace->show_arg_names && !new_show_zeros) {
5031 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
5032 goto out;
5033 }
5034 trace->show_zeros = new_show_zeros;
5035 } else if (!strcmp(var, "trace.show_prefix")) {
5036 trace->show_string_prefix = perf_config_bool(var, value);
5037 } else if (!strcmp(var, "trace.no_inherit")) {
5038 trace->opts.no_inherit = perf_config_bool(var, value);
5039 } else if (!strcmp(var, "trace.args_alignment")) {
5040 int args_alignment = 0;
5041 if (perf_config_int(&args_alignment, var, value) == 0)
5042 trace->args_alignment = args_alignment;
5043 } else if (!strcmp(var, "trace.tracepoint_beautifiers")) {
5044 if (strcasecmp(value, "libtraceevent") == 0)
5045 trace->libtraceevent_print = true;
5046 else if (strcasecmp(value, "libbeauty") == 0)
5047 trace->libtraceevent_print = false;
5048 }
5049 out:
5050 return err;
5051 }
5052
5053 static void trace__exit(struct trace *trace)
5054 {
5055 int i;
5056
5057 strlist__delete(trace->ev_qualifier);
5058 zfree(&trace->ev_qualifier_ids.entries);
5059 if (trace->syscalls.table) {
5060 for (i = 0; i <= trace->sctbl->syscalls.max_id; i++)
5061 syscall__exit(&trace->syscalls.table[i]);
5062 zfree(&trace->syscalls.table);
5063 }
5064 syscalltbl__delete(trace->sctbl);
5065 zfree(&trace->perfconfig_events);
5066 }
5067
5068 #ifdef HAVE_BPF_SKEL
5069 static int bpf__setup_bpf_output(struct evlist *evlist)
5070 {
5071 int err = parse_event(evlist, "bpf-output/no-inherit=1,name=__augmented_syscalls__/");
5072
5073 if (err)
5074 pr_debug("ERROR: failed to create the \"__augmented_syscalls__\" bpf-output event\n");
5075
5076 return err;
5077 }
5078 #endif
5079
5080 int cmd_trace(int argc, const char **argv)
5081 {
5082 const char *trace_usage[] = {
5083 "perf trace [<options>] [<command>]",
5084 "perf trace [<options>] -- <command> [<options>]",
5085 "perf trace record [<options>] [<command>]",
5086 "perf trace record [<options>] -- <command> [<options>]",
5087 NULL
5088 };
5089 struct trace trace = {
5090 .opts = {
5091 .target = {
5092 .uid = UINT_MAX,
5093 .uses_mmap = true,
5094 },
5095 .user_freq = UINT_MAX,
5096 .user_interval = ULLONG_MAX,
5097 .no_buffering = true,
5098 .mmap_pages = UINT_MAX,
5099 },
5100 .output = stderr,
5101 .show_comm = true,
5102 .show_tstamp = true,
5103 .show_duration = true,
5104 .show_arg_names = true,
5105 .args_alignment = 70,
5106 .trace_syscalls = false,
5107 .kernel_syscallchains = false,
5108 .max_stack = UINT_MAX,
5109 .max_events = ULONG_MAX,
5110 };
5111 const char *output_name = NULL;
5112 const struct option trace_options[] = {
5113 OPT_CALLBACK('e', "event", &trace, "event",
5114 "event/syscall selector. use 'perf list' to list available events",
5115 trace__parse_events_option),
5116 OPT_CALLBACK(0, "filter", &trace.evlist, "filter",
5117 "event filter", parse_filter),
5118 OPT_BOOLEAN(0, "comm", &trace.show_comm,
5119 "show the thread COMM next to its id"),
5120 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
5121 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
5122 trace__parse_events_option),
5123 OPT_STRING('o', "output", &output_name, "file", "output file name"),
5124 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
5125 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
5126 "trace events on existing process id"),
5127 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
5128 "trace events on existing thread id"),
5129 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
5130 "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
5131 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
5132 "system-wide collection from all CPUs"),
5133 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
5134 "list of cpus to monitor"),
5135 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
5136 "child tasks do not inherit counters"),
5137 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
5138 "number of mmap data pages", evlist__parse_mmap_pages),
5139 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
5140 "user to profile"),
5141 OPT_CALLBACK(0, "duration", &trace, "float",
5142 "show only events with duration > N.M ms",
5143 trace__set_duration),
5144 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
5145 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
5146 OPT_BOOLEAN('T', "time", &trace.full_time,
5147 "Show full timestamp, not time relative to first start"),
5148 OPT_BOOLEAN(0, "failure", &trace.failure_only,
5149 "Show only syscalls that failed"),
5150 OPT_BOOLEAN('s', "summary", &trace.summary_only,
5151 "Show only syscall summary with statistics"),
5152 OPT_BOOLEAN('S', "with-summary", &trace.summary,
5153 "Show all syscalls and summary with statistics"),
5154 OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary,
5155 "Show errno stats per syscall, use with -s or -S"),
5156 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
5157 "Trace pagefaults", parse_pagefaults, "maj"),
5158 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
5159 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
5160 OPT_CALLBACK(0, "call-graph", &trace.opts,
5161 "record_mode[,record_size]", record_callchain_help,
5162 &record_parse_callchain_opt),
5163 OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,
5164 "Use libtraceevent to print the tracepoint arguments."),
5165 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
5166 "Show the kernel callchains on the syscall exit path"),
5167 OPT_ULONG(0, "max-events", &trace.max_events,
5168 "Set the maximum number of events to print, exit after that is reached. "),
5169 OPT_UINTEGER(0, "min-stack", &trace.min_stack,
5170 "Set the minimum stack depth when parsing the callchain, "
5171 "anything below the specified depth will be ignored."),
5172 OPT_UINTEGER(0, "max-stack", &trace.max_stack,
5173 "Set the maximum stack depth when parsing the callchain, "
5174 "anything beyond the specified depth will be ignored. "
5175 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
5176 OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
5177 "Sort batch of events before processing, use if getting out of order events"),
5178 OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
5179 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
5180 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
5181 "per thread proc mmap processing timeout in ms"),
5182 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
5183 trace__parse_cgroups),
5184 OPT_INTEGER('D', "delay", &trace.opts.target.initial_delay,
5185 "ms to wait before starting measurement after program "
5186 "start"),
5187 OPT_BOOLEAN(0, "force-btf", &trace.force_btf, "Prefer btf_dump general pretty printer"
5188 "to customized ones"),
5189 OPTS_EVSWITCH(&trace.evswitch),
5190 OPT_END()
5191 };
5192 bool __maybe_unused max_stack_user_set = true;
5193 bool mmap_pages_user_set = true;
5194 struct evsel *evsel;
5195 const char * const trace_subcommands[] = { "record", NULL };
5196 int err = -1;
5197 char bf[BUFSIZ];
5198 struct sigaction sigchld_act;
5199
5200 signal(SIGSEGV, sighandler_dump_stack);
5201 signal(SIGFPE, sighandler_dump_stack);
5202 signal(SIGINT, sighandler_interrupt);
5203
5204 memset(&sigchld_act, 0, sizeof(sigchld_act));
5205 sigchld_act.sa_flags = SA_SIGINFO;
5206 sigchld_act.sa_sigaction = sighandler_chld;
5207 sigaction(SIGCHLD, &sigchld_act, NULL);
5208
5209 trace.evlist = evlist__new();
5210 trace.sctbl = syscalltbl__new();
5211
5212 if (trace.evlist == NULL || trace.sctbl == NULL) {
5213 pr_err("Not enough memory to run!\n");
5214 err = -ENOMEM;
5215 goto out;
5216 }
5217
5218 /*
5219 * Parsing .perfconfig may entail creating a BPF event, that may need
5220 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
5221 * is too small. This affects just this process, not touching the
5222 * global setting. If it fails we'll get something in 'perf trace -v'
5223 * to help diagnose the problem.
5224 */
5225 rlimit__bump_memlock();
5226
5227 err = perf_config(trace__config, &trace);
5228 if (err)
5229 goto out;
5230
5231 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
5232 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
5233
5234 /*
5235 * Here we already passed thru trace__parse_events_option() and it has
5236 * already figured out if -e syscall_name, if not but if --event
5237 * foo:bar was used, the user is interested _just_ in those, say,
5238 * tracepoint events, not in the strace-like syscall-name-based mode.
5239 *
5240 * This is important because we need to check if strace-like mode is
5241 * needed to decided if we should filter out the eBPF
5242 * __augmented_syscalls__ code, if it is in the mix, say, via
5243 * .perfconfig trace.add_events, and filter those out.
5244 */
5245 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
5246 trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
5247 trace.trace_syscalls = true;
5248 }
5249 /*
5250 * Now that we have --verbose figured out, lets see if we need to parse
5251 * events from .perfconfig, so that if those events fail parsing, say some
5252 * BPF program fails, then we'll be able to use --verbose to see what went
5253 * wrong in more detail.
5254 */
5255 if (trace.perfconfig_events != NULL) {
5256 struct parse_events_error parse_err;
5257
5258 parse_events_error__init(&parse_err);
5259 err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);
5260 if (err)
5261 parse_events_error__print(&parse_err, trace.perfconfig_events);
5262 parse_events_error__exit(&parse_err);
5263 if (err)
5264 goto out;
5265 }
5266
5267 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
5268 usage_with_options_msg(trace_usage, trace_options,
5269 "cgroup monitoring only available in system-wide mode");
5270 }
5271
5272 #ifdef HAVE_BPF_SKEL
5273 if (!trace.trace_syscalls)
5274 goto skip_augmentation;
5275
5276 if ((argc >= 1) && (strcmp(argv[0], "record") == 0)) {
5277 pr_debug("Syscall augmentation fails with record, disabling augmentation");
5278 goto skip_augmentation;
5279 }
5280
5281 trace.skel = augmented_raw_syscalls_bpf__open();
5282 if (!trace.skel) {
5283 pr_debug("Failed to open augmented syscalls BPF skeleton");
5284 } else {
5285 /*
5286 * Disable attaching the BPF programs except for sys_enter and
5287 * sys_exit that tail call into this as necessary.
5288 */
5289 struct bpf_program *prog;
5290
5291 bpf_object__for_each_program(prog, trace.skel->obj) {
5292 if (prog != trace.skel->progs.sys_enter && prog != trace.skel->progs.sys_exit)
5293 bpf_program__set_autoattach(prog, /*autoattach=*/false);
5294 }
5295
5296 err = augmented_raw_syscalls_bpf__load(trace.skel);
5297
5298 if (err < 0) {
5299 libbpf_strerror(err, bf, sizeof(bf));
5300 pr_debug("Failed to load augmented syscalls BPF skeleton: %s\n", bf);
5301 } else {
5302 augmented_raw_syscalls_bpf__attach(trace.skel);
5303 trace__add_syscall_newtp(&trace);
5304 }
5305 }
5306
5307 err = bpf__setup_bpf_output(trace.evlist);
5308 if (err) {
5309 libbpf_strerror(err, bf, sizeof(bf));
5310 pr_err("ERROR: Setup BPF output event failed: %s\n", bf);
5311 goto out;
5312 }
5313 trace.syscalls.events.bpf_output = evlist__last(trace.evlist);
5314 assert(evsel__name_is(trace.syscalls.events.bpf_output, "__augmented_syscalls__"));
5315 skip_augmentation:
5316 #endif
5317 err = -1;
5318
5319 if (trace.trace_pgfaults) {
5320 trace.opts.sample_address = true;
5321 trace.opts.sample_time = true;
5322 }
5323
5324 if (trace.opts.mmap_pages == UINT_MAX)
5325 mmap_pages_user_set = false;
5326
5327 if (trace.max_stack == UINT_MAX) {
5328 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
5329 max_stack_user_set = false;
5330 }
5331
5332 #ifdef HAVE_DWARF_UNWIND_SUPPORT
5333 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
5334 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
5335 }
5336 #endif
5337
5338 if (callchain_param.enabled) {
5339 if (!mmap_pages_user_set && geteuid() == 0)
5340 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
5341
5342 symbol_conf.use_callchain = true;
5343 }
5344
5345 if (trace.evlist->core.nr_entries > 0) {
5346 bool use_btf = false;
5347
5348 evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);
5349 if (evlist__set_syscall_tp_fields(trace.evlist, &use_btf)) {
5350 perror("failed to set syscalls:* tracepoint fields");
5351 goto out;
5352 }
5353
5354 if (use_btf)
5355 trace__load_vmlinux_btf(&trace);
5356 }
5357
5358 if (trace.sort_events) {
5359 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
5360 ordered_events__set_copy_on_queue(&trace.oe.data, true);
5361 }
5362
5363 /*
5364 * If we are augmenting syscalls, then combine what we put in the
5365 * __augmented_syscalls__ BPF map with what is in the
5366 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
5367 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
5368 *
5369 * We'll switch to look at two BPF maps, one for sys_enter and the
5370 * other for sys_exit when we start augmenting the sys_exit paths with
5371 * buffers that are being copied from kernel to userspace, think 'read'
5372 * syscall.
5373 */
5374 if (trace.syscalls.events.bpf_output) {
5375 evlist__for_each_entry(trace.evlist, evsel) {
5376 bool raw_syscalls_sys_exit = evsel__name_is(evsel, "raw_syscalls:sys_exit");
5377
5378 if (raw_syscalls_sys_exit) {
5379 trace.raw_augmented_syscalls = true;
5380 goto init_augmented_syscall_tp;
5381 }
5382
5383 if (trace.syscalls.events.bpf_output->priv == NULL &&
5384 strstr(evsel__name(evsel), "syscalls:sys_enter")) {
5385 struct evsel *augmented = trace.syscalls.events.bpf_output;
5386 if (evsel__init_augmented_syscall_tp(augmented, evsel) ||
5387 evsel__init_augmented_syscall_tp_args(augmented))
5388 goto out;
5389 /*
5390 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
5391 * Above we made sure we can get from the payload the tp fields
5392 * that we get from syscalls:sys_enter tracefs format file.
5393 */
5394 augmented->handler = trace__sys_enter;
5395 /*
5396 * Now we do the same for the *syscalls:sys_enter event so that
5397 * if we handle it directly, i.e. if the BPF prog returns 0 so
5398 * as not to filter it, then we'll handle it just like we would
5399 * for the BPF_OUTPUT one:
5400 */
5401 if (evsel__init_augmented_syscall_tp(evsel, evsel) ||
5402 evsel__init_augmented_syscall_tp_args(evsel))
5403 goto out;
5404 evsel->handler = trace__sys_enter;
5405 }
5406
5407 if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) {
5408 struct syscall_tp *sc;
5409 init_augmented_syscall_tp:
5410 if (evsel__init_augmented_syscall_tp(evsel, evsel))
5411 goto out;
5412 sc = __evsel__syscall_tp(evsel);
5413 /*
5414 * For now with BPF raw_augmented we hook into
5415 * raw_syscalls:sys_enter and there we get all
5416 * 6 syscall args plus the tracepoint common
5417 * fields and the syscall_nr (another long).
5418 * So we check if that is the case and if so
5419 * don't look after the sc->args_size but
5420 * always after the full raw_syscalls:sys_enter
5421 * payload, which is fixed.
5422 *
5423 * We'll revisit this later to pass
5424 * s->args_size to the BPF augmenter (now
5425 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
5426 * so that it copies only what we need for each
5427 * syscall, like what happens when we use
5428 * syscalls:sys_enter_NAME, so that we reduce
5429 * the kernel/userspace traffic to just what is
5430 * needed for each syscall.
5431 */
5432 if (trace.raw_augmented_syscalls)
5433 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
5434 evsel__init_augmented_syscall_tp_ret(evsel);
5435 evsel->handler = trace__sys_exit;
5436 }
5437 }
5438 }
5439
5440 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
5441 return trace__record(&trace, argc-1, &argv[1]);
5442
5443 /* Using just --errno-summary will trigger --summary */
5444 if (trace.errno_summary && !trace.summary && !trace.summary_only)
5445 trace.summary_only = true;
5446
5447 /* summary_only implies summary option, but don't overwrite summary if set */
5448 if (trace.summary_only)
5449 trace.summary = trace.summary_only;
5450
5451 /* Keep exited threads, otherwise information might be lost for summary */
5452 if (trace.summary)
5453 symbol_conf.keep_exited_threads = true;
5454
5455 if (output_name != NULL) {
5456 err = trace__open_output(&trace, output_name);
5457 if (err < 0) {
5458 perror("failed to create output file");
5459 goto out;
5460 }
5461 }
5462
5463 err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
5464 if (err)
5465 goto out_close;
5466
5467 err = target__validate(&trace.opts.target);
5468 if (err) {
5469 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5470 fprintf(trace.output, "%s", bf);
5471 goto out_close;
5472 }
5473
5474 err = target__parse_uid(&trace.opts.target);
5475 if (err) {
5476 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5477 fprintf(trace.output, "%s", bf);
5478 goto out_close;
5479 }
5480
5481 if (!argc && target__none(&trace.opts.target))
5482 trace.opts.target.system_wide = true;
5483
5484 if (input_name)
5485 err = trace__replay(&trace);
5486 else
5487 err = trace__run(&trace, argc, argv);
5488
5489 out_close:
5490 if (output_name != NULL)
5491 fclose(trace.output);
5492 out:
5493 trace__exit(&trace);
5494 #ifdef HAVE_BPF_SKEL
5495 augmented_raw_syscalls_bpf__destroy(trace.skel);
5496 #endif
5497 return err;
5498 }
Kernel DRM miscellaneous fixes and cross-tree changes