~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/tools/perf/builtin-trace.c

Version: ~ [ linux-6.2-rc3 ] ~ [ linux-6.1.5 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.87 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.162 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.228 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.269 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.302 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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