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

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

Version: ~ [ linux-5.15-rc6 ] ~ [ linux-5.14.14 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.75 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.155 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.213 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.252 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.287 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.289 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ 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  * Released under the GPL v2. (and only v2, not any later version)
 17  */
 18 
 19 #include <traceevent/event-parse.h>
 20 #include <api/fs/tracing_path.h>
 21 #include <bpf/bpf.h>
 22 #include "builtin.h"
 23 #include "util/cgroup.h"
 24 #include "util/color.h"
 25 #include "util/config.h"
 26 #include "util/debug.h"
 27 #include "util/env.h"
 28 #include "util/event.h"
 29 #include "util/evlist.h"
 30 #include <subcmd/exec-cmd.h>
 31 #include "util/machine.h"
 32 #include "util/path.h"
 33 #include "util/session.h"
 34 #include "util/thread.h"
 35 #include <subcmd/parse-options.h>
 36 #include "util/strlist.h"
 37 #include "util/intlist.h"
 38 #include "util/thread_map.h"
 39 #include "util/stat.h"
 40 #include "trace/beauty/beauty.h"
 41 #include "trace-event.h"
 42 #include "util/parse-events.h"
 43 #include "util/bpf-loader.h"
 44 #include "callchain.h"
 45 #include "print_binary.h"
 46 #include "string2.h"
 47 #include "syscalltbl.h"
 48 #include "rb_resort.h"
 49 
 50 #include <errno.h>
 51 #include <inttypes.h>
 52 #include <poll.h>
 53 #include <signal.h>
 54 #include <stdlib.h>
 55 #include <string.h>
 56 #include <linux/err.h>
 57 #include <linux/filter.h>
 58 #include <linux/kernel.h>
 59 #include <linux/random.h>
 60 #include <linux/stringify.h>
 61 #include <linux/time64.h>
 62 #include <fcntl.h>
 63 #include <sys/sysmacros.h>
 64 
 65 #include "sane_ctype.h"
 66 
 67 #ifndef O_CLOEXEC
 68 # define O_CLOEXEC              02000000
 69 #endif
 70 
 71 #ifndef F_LINUX_SPECIFIC_BASE
 72 # define F_LINUX_SPECIFIC_BASE  1024
 73 #endif
 74 
 75 struct trace {
 76         struct perf_tool        tool;
 77         struct syscalltbl       *sctbl;
 78         struct {
 79                 int             max;
 80                 struct syscall  *table;
 81                 struct bpf_map  *map;
 82                 struct {
 83                         struct perf_evsel *sys_enter,
 84                                           *sys_exit,
 85                                           *augmented;
 86                 }               events;
 87         } syscalls;
 88         struct record_opts      opts;
 89         struct perf_evlist      *evlist;
 90         struct machine          *host;
 91         struct thread           *current;
 92         struct cgroup           *cgroup;
 93         u64                     base_time;
 94         FILE                    *output;
 95         unsigned long           nr_events;
 96         unsigned long           nr_events_printed;
 97         unsigned long           max_events;
 98         struct strlist          *ev_qualifier;
 99         struct {
100                 size_t          nr;
101                 int             *entries;
102         }                       ev_qualifier_ids;
103         struct {
104                 size_t          nr;
105                 pid_t           *entries;
106                 struct bpf_map  *map;
107         }                       filter_pids;
108         double                  duration_filter;
109         double                  runtime_ms;
110         struct {
111                 u64             vfs_getname,
112                                 proc_getname;
113         } stats;
114         unsigned int            max_stack;
115         unsigned int            min_stack;
116         int                     raw_augmented_syscalls_args_size;
117         bool                    raw_augmented_syscalls;
118         bool                    sort_events;
119         bool                    not_ev_qualifier;
120         bool                    live;
121         bool                    full_time;
122         bool                    sched;
123         bool                    multiple_threads;
124         bool                    summary;
125         bool                    summary_only;
126         bool                    failure_only;
127         bool                    show_comm;
128         bool                    print_sample;
129         bool                    show_tool_stats;
130         bool                    trace_syscalls;
131         bool                    kernel_syscallchains;
132         s16                     args_alignment;
133         bool                    show_tstamp;
134         bool                    show_duration;
135         bool                    show_zeros;
136         bool                    show_arg_names;
137         bool                    show_string_prefix;
138         bool                    force;
139         bool                    vfs_getname;
140         int                     trace_pgfaults;
141         struct {
142                 struct ordered_events   data;
143                 u64                     last;
144         } oe;
145 };
146 
147 struct tp_field {
148         int offset;
149         union {
150                 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
151                 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
152         };
153 };
154 
155 #define TP_UINT_FIELD(bits) \
156 static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
157 { \
158         u##bits value; \
159         memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
160         return value;  \
161 }
162 
163 TP_UINT_FIELD(8);
164 TP_UINT_FIELD(16);
165 TP_UINT_FIELD(32);
166 TP_UINT_FIELD(64);
167 
168 #define TP_UINT_FIELD__SWAPPED(bits) \
169 static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
170 { \
171         u##bits value; \
172         memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
173         return bswap_##bits(value);\
174 }
175 
176 TP_UINT_FIELD__SWAPPED(16);
177 TP_UINT_FIELD__SWAPPED(32);
178 TP_UINT_FIELD__SWAPPED(64);
179 
180 static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
181 {
182         field->offset = offset;
183 
184         switch (size) {
185         case 1:
186                 field->integer = tp_field__u8;
187                 break;
188         case 2:
189                 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
190                 break;
191         case 4:
192                 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
193                 break;
194         case 8:
195                 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
196                 break;
197         default:
198                 return -1;
199         }
200 
201         return 0;
202 }
203 
204 static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
205 {
206         return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
207 }
208 
209 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
210 {
211         return sample->raw_data + field->offset;
212 }
213 
214 static int __tp_field__init_ptr(struct tp_field *field, int offset)
215 {
216         field->offset = offset;
217         field->pointer = tp_field__ptr;
218         return 0;
219 }
220 
221 static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
222 {
223         return __tp_field__init_ptr(field, format_field->offset);
224 }
225 
226 struct syscall_tp {
227         struct tp_field id;
228         union {
229                 struct tp_field args, ret;
230         };
231 };
232 
233 static int perf_evsel__init_tp_uint_field(struct perf_evsel *evsel,
234                                           struct tp_field *field,
235                                           const char *name)
236 {
237         struct tep_format_field *format_field = perf_evsel__field(evsel, name);
238 
239         if (format_field == NULL)
240                 return -1;
241 
242         return tp_field__init_uint(field, format_field, evsel->needs_swap);
243 }
244 
245 #define perf_evsel__init_sc_tp_uint_field(evsel, name) \
246         ({ struct syscall_tp *sc = evsel->priv;\
247            perf_evsel__init_tp_uint_field(evsel, &sc->name, #name); })
248 
249 static int perf_evsel__init_tp_ptr_field(struct perf_evsel *evsel,
250                                          struct tp_field *field,
251                                          const char *name)
252 {
253         struct tep_format_field *format_field = perf_evsel__field(evsel, name);
254 
255         if (format_field == NULL)
256                 return -1;
257 
258         return tp_field__init_ptr(field, format_field);
259 }
260 
261 #define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
262         ({ struct syscall_tp *sc = evsel->priv;\
263            perf_evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
264 
265 static void perf_evsel__delete_priv(struct perf_evsel *evsel)
266 {
267         zfree(&evsel->priv);
268         perf_evsel__delete(evsel);
269 }
270 
271 static int perf_evsel__init_syscall_tp(struct perf_evsel *evsel)
272 {
273         struct syscall_tp *sc = evsel->priv = malloc(sizeof(struct syscall_tp));
274 
275         if (evsel->priv != NULL) {
276                 if (perf_evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
277                     perf_evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
278                         goto out_delete;
279                 return 0;
280         }
281 
282         return -ENOMEM;
283 out_delete:
284         zfree(&evsel->priv);
285         return -ENOENT;
286 }
287 
288 static int perf_evsel__init_augmented_syscall_tp(struct perf_evsel *evsel, struct perf_evsel *tp)
289 {
290         struct syscall_tp *sc = evsel->priv = malloc(sizeof(struct syscall_tp));
291 
292         if (evsel->priv != NULL) {
293                 struct tep_format_field *syscall_id = perf_evsel__field(tp, "id");
294                 if (syscall_id == NULL)
295                         syscall_id = perf_evsel__field(tp, "__syscall_nr");
296                 if (syscall_id == NULL)
297                         goto out_delete;
298                 if (__tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
299                         goto out_delete;
300 
301                 return 0;
302         }
303 
304         return -ENOMEM;
305 out_delete:
306         zfree(&evsel->priv);
307         return -EINVAL;
308 }
309 
310 static int perf_evsel__init_augmented_syscall_tp_args(struct perf_evsel *evsel)
311 {
312         struct syscall_tp *sc = evsel->priv;
313 
314         return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
315 }
316 
317 static int perf_evsel__init_augmented_syscall_tp_ret(struct perf_evsel *evsel)
318 {
319         struct syscall_tp *sc = evsel->priv;
320 
321         return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
322 }
323 
324 static int perf_evsel__init_raw_syscall_tp(struct perf_evsel *evsel, void *handler)
325 {
326         evsel->priv = malloc(sizeof(struct syscall_tp));
327         if (evsel->priv != NULL) {
328                 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
329                         goto out_delete;
330 
331                 evsel->handler = handler;
332                 return 0;
333         }
334 
335         return -ENOMEM;
336 
337 out_delete:
338         zfree(&evsel->priv);
339         return -ENOENT;
340 }
341 
342 static struct perf_evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
343 {
344         struct perf_evsel *evsel = perf_evsel__newtp("raw_syscalls", direction);
345 
346         /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
347         if (IS_ERR(evsel))
348                 evsel = perf_evsel__newtp("syscalls", direction);
349 
350         if (IS_ERR(evsel))
351                 return NULL;
352 
353         if (perf_evsel__init_raw_syscall_tp(evsel, handler))
354                 goto out_delete;
355 
356         return evsel;
357 
358 out_delete:
359         perf_evsel__delete_priv(evsel);
360         return NULL;
361 }
362 
363 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
364         ({ struct syscall_tp *fields = evsel->priv; \
365            fields->name.integer(&fields->name, sample); })
366 
367 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
368         ({ struct syscall_tp *fields = evsel->priv; \
369            fields->name.pointer(&fields->name, sample); })
370 
371 size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
372 {
373         int idx = val - sa->offset;
374 
375         if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
376                 size_t printed = scnprintf(bf, size, intfmt, val);
377                 if (show_prefix)
378                         printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
379                 return printed;
380         }
381 
382         return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
383 }
384 
385 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
386                                                 const char *intfmt,
387                                                 struct syscall_arg *arg)
388 {
389         return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
390 }
391 
392 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
393                                               struct syscall_arg *arg)
394 {
395         return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
396 }
397 
398 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
399 
400 size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
401 {
402         size_t printed;
403         int i;
404 
405         for (i = 0; i < sas->nr_entries; ++i) {
406                 struct strarray *sa = sas->entries[i];
407                 int idx = val - sa->offset;
408 
409                 if (idx >= 0 && idx < sa->nr_entries) {
410                         if (sa->entries[idx] == NULL)
411                                 break;
412                         return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
413                 }
414         }
415 
416         printed = scnprintf(bf, size, intfmt, val);
417         if (show_prefix)
418                 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
419         return printed;
420 }
421 
422 size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
423                                         struct syscall_arg *arg)
424 {
425         return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
426 }
427 
428 #ifndef AT_FDCWD
429 #define AT_FDCWD        -100
430 #endif
431 
432 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
433                                            struct syscall_arg *arg)
434 {
435         int fd = arg->val;
436         const char *prefix = "AT_FD";
437 
438         if (fd == AT_FDCWD)
439                 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
440 
441         return syscall_arg__scnprintf_fd(bf, size, arg);
442 }
443 
444 #define SCA_FDAT syscall_arg__scnprintf_fd_at
445 
446 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
447                                               struct syscall_arg *arg);
448 
449 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
450 
451 size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
452 {
453         return scnprintf(bf, size, "%#lx", arg->val);
454 }
455 
456 size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
457 {
458         if (arg->val == 0)
459                 return scnprintf(bf, size, "NULL");
460         return syscall_arg__scnprintf_hex(bf, size, arg);
461 }
462 
463 size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
464 {
465         return scnprintf(bf, size, "%d", arg->val);
466 }
467 
468 size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
469 {
470         return scnprintf(bf, size, "%ld", arg->val);
471 }
472 
473 static const char *bpf_cmd[] = {
474         "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
475         "MAP_GET_NEXT_KEY", "PROG_LOAD",
476 };
477 static DEFINE_STRARRAY(bpf_cmd, "BPF_");
478 
479 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
480 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
481 
482 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
483 static DEFINE_STRARRAY(itimers, "ITIMER_");
484 
485 static const char *keyctl_options[] = {
486         "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
487         "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
488         "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
489         "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
490         "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
491 };
492 static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
493 
494 static const char *whences[] = { "SET", "CUR", "END",
495 #ifdef SEEK_DATA
496 "DATA",
497 #endif
498 #ifdef SEEK_HOLE
499 "HOLE",
500 #endif
501 };
502 static DEFINE_STRARRAY(whences, "SEEK_");
503 
504 static const char *fcntl_cmds[] = {
505         "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
506         "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
507         "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
508         "GETOWNER_UIDS",
509 };
510 static DEFINE_STRARRAY(fcntl_cmds, "F_");
511 
512 static const char *fcntl_linux_specific_cmds[] = {
513         "SETLEASE", "GETLEASE", "NOTIFY", [5] = "CANCELLK", "DUPFD_CLOEXEC",
514         "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
515         "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
516 };
517 
518 static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
519 
520 static struct strarray *fcntl_cmds_arrays[] = {
521         &strarray__fcntl_cmds,
522         &strarray__fcntl_linux_specific_cmds,
523 };
524 
525 static DEFINE_STRARRAYS(fcntl_cmds_arrays);
526 
527 static const char *rlimit_resources[] = {
528         "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
529         "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
530         "RTTIME",
531 };
532 static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
533 
534 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
535 static DEFINE_STRARRAY(sighow, "SIG_");
536 
537 static const char *clockid[] = {
538         "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
539         "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
540         "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
541 };
542 static DEFINE_STRARRAY(clockid, "CLOCK_");
543 
544 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
545                                                  struct syscall_arg *arg)
546 {
547         bool show_prefix = arg->show_string_prefix;
548         const char *suffix = "_OK";
549         size_t printed = 0;
550         int mode = arg->val;
551 
552         if (mode == F_OK) /* 0 */
553                 return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
554 #define P_MODE(n) \
555         if (mode & n##_OK) { \
556                 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
557                 mode &= ~n##_OK; \
558         }
559 
560         P_MODE(R);
561         P_MODE(W);
562         P_MODE(X);
563 #undef P_MODE
564 
565         if (mode)
566                 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
567 
568         return printed;
569 }
570 
571 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
572 
573 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
574                                               struct syscall_arg *arg);
575 
576 #define SCA_FILENAME syscall_arg__scnprintf_filename
577 
578 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
579                                                 struct syscall_arg *arg)
580 {
581         bool show_prefix = arg->show_string_prefix;
582         const char *prefix = "O_";
583         int printed = 0, flags = arg->val;
584 
585 #define P_FLAG(n) \
586         if (flags & O_##n) { \
587                 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
588                 flags &= ~O_##n; \
589         }
590 
591         P_FLAG(CLOEXEC);
592         P_FLAG(NONBLOCK);
593 #undef P_FLAG
594 
595         if (flags)
596                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
597 
598         return printed;
599 }
600 
601 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
602 
603 #ifndef GRND_NONBLOCK
604 #define GRND_NONBLOCK   0x0001
605 #endif
606 #ifndef GRND_RANDOM
607 #define GRND_RANDOM     0x0002
608 #endif
609 
610 static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
611                                                    struct syscall_arg *arg)
612 {
613         bool show_prefix = arg->show_string_prefix;
614         const char *prefix = "GRND_";
615         int printed = 0, flags = arg->val;
616 
617 #define P_FLAG(n) \
618         if (flags & GRND_##n) { \
619                 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
620                 flags &= ~GRND_##n; \
621         }
622 
623         P_FLAG(RANDOM);
624         P_FLAG(NONBLOCK);
625 #undef P_FLAG
626 
627         if (flags)
628                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
629 
630         return printed;
631 }
632 
633 #define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
634 
635 #define STRARRAY(name, array) \
636           { .scnprintf  = SCA_STRARRAY, \
637             .parm       = &strarray__##array, }
638 
639 #include "trace/beauty/arch_errno_names.c"
640 #include "trace/beauty/eventfd.c"
641 #include "trace/beauty/futex_op.c"
642 #include "trace/beauty/futex_val3.c"
643 #include "trace/beauty/mmap.c"
644 #include "trace/beauty/mode_t.c"
645 #include "trace/beauty/msg_flags.c"
646 #include "trace/beauty/open_flags.c"
647 #include "trace/beauty/perf_event_open.c"
648 #include "trace/beauty/pid.c"
649 #include "trace/beauty/sched_policy.c"
650 #include "trace/beauty/seccomp.c"
651 #include "trace/beauty/signum.c"
652 #include "trace/beauty/socket_type.c"
653 #include "trace/beauty/waitid_options.c"
654 
655 struct syscall_arg_fmt {
656         size_t     (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
657         unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
658         void       *parm;
659         const char *name;
660         bool       show_zero;
661 };
662 
663 static struct syscall_fmt {
664         const char *name;
665         const char *alias;
666         struct syscall_arg_fmt arg[6];
667         u8         nr_args;
668         bool       errpid;
669         bool       timeout;
670         bool       hexret;
671 } syscall_fmts[] = {
672         { .name     = "access",
673           .arg = { [1] = { .scnprintf = SCA_ACCMODE,  /* mode */ }, }, },
674         { .name     = "arch_prctl",
675           .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
676                    [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
677         { .name     = "bind",
678           .arg = { [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ }, }, },
679         { .name     = "bpf",
680           .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
681         { .name     = "brk",        .hexret = true,
682           .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
683         { .name     = "clock_gettime",
684           .arg = { [0] = STRARRAY(clk_id, clockid), }, },
685         { .name     = "clone",      .errpid = true, .nr_args = 5,
686           .arg = { [0] = { .name = "flags",         .scnprintf = SCA_CLONE_FLAGS, },
687                    [1] = { .name = "child_stack",   .scnprintf = SCA_HEX, },
688                    [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
689                    [3] = { .name = "child_tidptr",  .scnprintf = SCA_HEX, },
690                    [4] = { .name = "tls",           .scnprintf = SCA_HEX, }, }, },
691         { .name     = "close",
692           .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
693         { .name     = "connect",
694           .arg = { [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ }, }, },
695         { .name     = "epoll_ctl",
696           .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
697         { .name     = "eventfd2",
698           .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
699         { .name     = "fchmodat",
700           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
701         { .name     = "fchownat",
702           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
703         { .name     = "fcntl",
704           .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
705                            .parm      = &strarrays__fcntl_cmds_arrays,
706                            .show_zero = true, },
707                    [2] = { .scnprintf =  SCA_FCNTL_ARG, /* arg */ }, }, },
708         { .name     = "flock",
709           .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
710         { .name     = "fstat", .alias = "newfstat", },
711         { .name     = "fstatat", .alias = "newfstatat", },
712         { .name     = "futex",
713           .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
714                    [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
715         { .name     = "futimesat",
716           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
717         { .name     = "getitimer",
718           .arg = { [0] = STRARRAY(which, itimers), }, },
719         { .name     = "getpid",     .errpid = true, },
720         { .name     = "getpgid",    .errpid = true, },
721         { .name     = "getppid",    .errpid = true, },
722         { .name     = "getrandom",
723           .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
724         { .name     = "getrlimit",
725           .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
726         { .name     = "gettid",     .errpid = true, },
727         { .name     = "ioctl",
728           .arg = {
729 #if defined(__i386__) || defined(__x86_64__)
730 /*
731  * FIXME: Make this available to all arches.
732  */
733                    [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
734                    [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
735 #else
736                    [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
737 #endif
738         { .name     = "kcmp",       .nr_args = 5,
739           .arg = { [0] = { .name = "pid1",      .scnprintf = SCA_PID, },
740                    [1] = { .name = "pid2",      .scnprintf = SCA_PID, },
741                    [2] = { .name = "type",      .scnprintf = SCA_KCMP_TYPE, },
742                    [3] = { .name = "idx1",      .scnprintf = SCA_KCMP_IDX, },
743                    [4] = { .name = "idx2",      .scnprintf = SCA_KCMP_IDX, }, }, },
744         { .name     = "keyctl",
745           .arg = { [0] = STRARRAY(option, keyctl_options), }, },
746         { .name     = "kill",
747           .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
748         { .name     = "linkat",
749           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
750         { .name     = "lseek",
751           .arg = { [2] = STRARRAY(whence, whences), }, },
752         { .name     = "lstat", .alias = "newlstat", },
753         { .name     = "madvise",
754           .arg = { [0] = { .scnprintf = SCA_HEX,      /* start */ },
755                    [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
756         { .name     = "mkdirat",
757           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
758         { .name     = "mknodat",
759           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
760         { .name     = "mmap",       .hexret = true,
761 /* The standard mmap maps to old_mmap on s390x */
762 #if defined(__s390x__)
763         .alias = "old_mmap",
764 #endif
765           .arg = { [2] = { .scnprintf = SCA_MMAP_PROT,  /* prot */ },
766                    [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */ },
767                    [5] = { .scnprintf = SCA_HEX,        /* offset */ }, }, },
768         { .name     = "mount",
769           .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
770                    [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
771                            .mask_val  = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
772         { .name     = "mprotect",
773           .arg = { [0] = { .scnprintf = SCA_HEX,        /* start */ },
774                    [2] = { .scnprintf = SCA_MMAP_PROT,  /* prot */ }, }, },
775         { .name     = "mq_unlink",
776           .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
777         { .name     = "mremap",     .hexret = true,
778           .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
779         { .name     = "name_to_handle_at",
780           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
781         { .name     = "newfstatat",
782           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
783         { .name     = "open",
784           .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
785         { .name     = "open_by_handle_at",
786           .arg = { [0] = { .scnprintf = SCA_FDAT,       /* dfd */ },
787                    [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
788         { .name     = "openat",
789           .arg = { [0] = { .scnprintf = SCA_FDAT,       /* dfd */ },
790                    [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
791         { .name     = "perf_event_open",
792           .arg = { [2] = { .scnprintf = SCA_INT,        /* cpu */ },
793                    [3] = { .scnprintf = SCA_FD,         /* group_fd */ },
794                    [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
795         { .name     = "pipe2",
796           .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
797         { .name     = "pkey_alloc",
798           .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS,   /* access_rights */ }, }, },
799         { .name     = "pkey_free",
800           .arg = { [0] = { .scnprintf = SCA_INT,        /* key */ }, }, },
801         { .name     = "pkey_mprotect",
802           .arg = { [0] = { .scnprintf = SCA_HEX,        /* start */ },
803                    [2] = { .scnprintf = SCA_MMAP_PROT,  /* prot */ },
804                    [3] = { .scnprintf = SCA_INT,        /* pkey */ }, }, },
805         { .name     = "poll", .timeout = true, },
806         { .name     = "ppoll", .timeout = true, },
807         { .name     = "prctl",
808           .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */ },
809                    [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
810                    [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
811         { .name     = "pread", .alias = "pread64", },
812         { .name     = "preadv", .alias = "pread", },
813         { .name     = "prlimit64",
814           .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
815         { .name     = "pwrite", .alias = "pwrite64", },
816         { .name     = "readlinkat",
817           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
818         { .name     = "recvfrom",
819           .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
820         { .name     = "recvmmsg",
821           .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
822         { .name     = "recvmsg",
823           .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
824         { .name     = "renameat",
825           .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
826                    [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
827         { .name     = "renameat2",
828           .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
829                    [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
830                    [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
831         { .name     = "rt_sigaction",
832           .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
833         { .name     = "rt_sigprocmask",
834           .arg = { [0] = STRARRAY(how, sighow), }, },
835         { .name     = "rt_sigqueueinfo",
836           .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
837         { .name     = "rt_tgsigqueueinfo",
838           .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
839         { .name     = "sched_setscheduler",
840           .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
841         { .name     = "seccomp",
842           .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP,    /* op */ },
843                    [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
844         { .name     = "select", .timeout = true, },
845         { .name     = "sendmmsg",
846           .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
847         { .name     = "sendmsg",
848           .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
849         { .name     = "sendto",
850           .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
851                    [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
852         { .name     = "set_tid_address", .errpid = true, },
853         { .name     = "setitimer",
854           .arg = { [0] = STRARRAY(which, itimers), }, },
855         { .name     = "setrlimit",
856           .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
857         { .name     = "socket",
858           .arg = { [0] = STRARRAY(family, socket_families),
859                    [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
860                    [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
861         { .name     = "socketpair",
862           .arg = { [0] = STRARRAY(family, socket_families),
863                    [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
864                    [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
865         { .name     = "stat", .alias = "newstat", },
866         { .name     = "statx",
867           .arg = { [0] = { .scnprintf = SCA_FDAT,        /* fdat */ },
868                    [2] = { .scnprintf = SCA_STATX_FLAGS, /* flags */ } ,
869                    [3] = { .scnprintf = SCA_STATX_MASK,  /* mask */ }, }, },
870         { .name     = "swapoff",
871           .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
872         { .name     = "swapon",
873           .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
874         { .name     = "symlinkat",
875           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
876         { .name     = "tgkill",
877           .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
878         { .name     = "tkill",
879           .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
880         { .name     = "umount2", .alias = "umount",
881           .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
882         { .name     = "uname", .alias = "newuname", },
883         { .name     = "unlinkat",
884           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
885         { .name     = "utimensat",
886           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
887         { .name     = "wait4",      .errpid = true,
888           .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
889         { .name     = "waitid",     .errpid = true,
890           .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
891 };
892 
893 static int syscall_fmt__cmp(const void *name, const void *fmtp)
894 {
895         const struct syscall_fmt *fmt = fmtp;
896         return strcmp(name, fmt->name);
897 }
898 
899 static struct syscall_fmt *syscall_fmt__find(const char *name)
900 {
901         const int nmemb = ARRAY_SIZE(syscall_fmts);
902         return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
903 }
904 
905 static struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
906 {
907         int i, nmemb = ARRAY_SIZE(syscall_fmts);
908 
909         for (i = 0; i < nmemb; ++i) {
910                 if (syscall_fmts[i].alias && strcmp(syscall_fmts[i].alias, alias) == 0)
911                         return &syscall_fmts[i];
912         }
913 
914         return NULL;
915 }
916 
917 /*
918  * is_exit: is this "exit" or "exit_group"?
919  * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
920  * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
921  */
922 struct syscall {
923         struct tep_event    *tp_format;
924         int                 nr_args;
925         int                 args_size;
926         bool                is_exit;
927         bool                is_open;
928         struct tep_format_field *args;
929         const char          *name;
930         struct syscall_fmt  *fmt;
931         struct syscall_arg_fmt *arg_fmt;
932 };
933 
934 struct bpf_map_syscall_entry {
935         bool    enabled;
936 };
937 
938 /*
939  * We need to have this 'calculated' boolean because in some cases we really
940  * don't know what is the duration of a syscall, for instance, when we start
941  * a session and some threads are waiting for a syscall to finish, say 'poll',
942  * in which case all we can do is to print "( ? ) for duration and for the
943  * start timestamp.
944  */
945 static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
946 {
947         double duration = (double)t / NSEC_PER_MSEC;
948         size_t printed = fprintf(fp, "(");
949 
950         if (!calculated)
951                 printed += fprintf(fp, "         ");
952         else if (duration >= 1.0)
953                 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
954         else if (duration >= 0.01)
955                 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
956         else
957                 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
958         return printed + fprintf(fp, "): ");
959 }
960 
961 /**
962  * filename.ptr: The filename char pointer that will be vfs_getname'd
963  * filename.entry_str_pos: Where to insert the string translated from
964  *                         filename.ptr by the vfs_getname tracepoint/kprobe.
965  * ret_scnprintf: syscall args may set this to a different syscall return
966  *                formatter, for instance, fcntl may return fds, file flags, etc.
967  */
968 struct thread_trace {
969         u64               entry_time;
970         bool              entry_pending;
971         unsigned long     nr_events;
972         unsigned long     pfmaj, pfmin;
973         char              *entry_str;
974         double            runtime_ms;
975         size_t            (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
976         struct {
977                 unsigned long ptr;
978                 short int     entry_str_pos;
979                 bool          pending_open;
980                 unsigned int  namelen;
981                 char          *name;
982         } filename;
983         struct {
984                 int           max;
985                 struct file   *table;
986         } files;
987 
988         struct intlist *syscall_stats;
989 };
990 
991 static struct thread_trace *thread_trace__new(void)
992 {
993         struct thread_trace *ttrace =  zalloc(sizeof(struct thread_trace));
994 
995         if (ttrace)
996                 ttrace->files.max = -1;
997 
998         ttrace->syscall_stats = intlist__new(NULL);
999 
1000         return ttrace;
1001 }
1002 
1003 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1004 {
1005         struct thread_trace *ttrace;
1006 
1007         if (thread == NULL)
1008                 goto fail;
1009 
1010         if (thread__priv(thread) == NULL)
1011                 thread__set_priv(thread, thread_trace__new());
1012 
1013         if (thread__priv(thread) == NULL)
1014                 goto fail;
1015 
1016         ttrace = thread__priv(thread);
1017         ++ttrace->nr_events;
1018 
1019         return ttrace;
1020 fail:
1021         color_fprintf(fp, PERF_COLOR_RED,
1022                       "WARNING: not enough memory, dropping samples!\n");
1023         return NULL;
1024 }
1025 
1026 
1027 void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
1028                                     size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
1029 {
1030         struct thread_trace *ttrace = thread__priv(arg->thread);
1031 
1032         ttrace->ret_scnprintf = ret_scnprintf;
1033 }
1034 
1035 #define TRACE_PFMAJ             (1 << 0)
1036 #define TRACE_PFMIN             (1 << 1)
1037 
1038 static const size_t trace__entry_str_size = 2048;
1039 
1040 static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
1041 {
1042         if (fd < 0)
1043                 return NULL;
1044 
1045         if (fd > ttrace->files.max) {
1046                 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
1047 
1048                 if (nfiles == NULL)
1049                         return NULL;
1050 
1051                 if (ttrace->files.max != -1) {
1052                         memset(nfiles + ttrace->files.max + 1, 0,
1053                                (fd - ttrace->files.max) * sizeof(struct file));
1054                 } else {
1055                         memset(nfiles, 0, (fd + 1) * sizeof(struct file));
1056                 }
1057 
1058                 ttrace->files.table = nfiles;
1059                 ttrace->files.max   = fd;
1060         }
1061 
1062         return ttrace->files.table + fd;
1063 }
1064 
1065 struct file *thread__files_entry(struct thread *thread, int fd)
1066 {
1067         return thread_trace__files_entry(thread__priv(thread), fd);
1068 }
1069 
1070 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1071 {
1072         struct thread_trace *ttrace = thread__priv(thread);
1073         struct file *file = thread_trace__files_entry(ttrace, fd);
1074 
1075         if (file != NULL) {
1076                 struct stat st;
1077                 if (stat(pathname, &st) == 0)
1078                         file->dev_maj = major(st.st_rdev);
1079                 file->pathname = strdup(pathname);
1080                 if (file->pathname)
1081                         return 0;
1082         }
1083 
1084         return -1;
1085 }
1086 
1087 static int thread__read_fd_path(struct thread *thread, int fd)
1088 {
1089         char linkname[PATH_MAX], pathname[PATH_MAX];
1090         struct stat st;
1091         int ret;
1092 
1093         if (thread->pid_ == thread->tid) {
1094                 scnprintf(linkname, sizeof(linkname),
1095                           "/proc/%d/fd/%d", thread->pid_, fd);
1096         } else {
1097                 scnprintf(linkname, sizeof(linkname),
1098                           "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1099         }
1100 
1101         if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1102                 return -1;
1103 
1104         ret = readlink(linkname, pathname, sizeof(pathname));
1105 
1106         if (ret < 0 || ret > st.st_size)
1107                 return -1;
1108 
1109         pathname[ret] = '\0';
1110         return trace__set_fd_pathname(thread, fd, pathname);
1111 }
1112 
1113 static const char *thread__fd_path(struct thread *thread, int fd,
1114                                    struct trace *trace)
1115 {
1116         struct thread_trace *ttrace = thread__priv(thread);
1117 
1118         if (ttrace == NULL)
1119                 return NULL;
1120 
1121         if (fd < 0)
1122                 return NULL;
1123 
1124         if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
1125                 if (!trace->live)
1126                         return NULL;
1127                 ++trace->stats.proc_getname;
1128                 if (thread__read_fd_path(thread, fd))
1129                         return NULL;
1130         }
1131 
1132         return ttrace->files.table[fd].pathname;
1133 }
1134 
1135 size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
1136 {
1137         int fd = arg->val;
1138         size_t printed = scnprintf(bf, size, "%d", fd);
1139         const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1140 
1141         if (path)
1142                 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1143 
1144         return printed;
1145 }
1146 
1147 size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1148 {
1149         size_t printed = scnprintf(bf, size, "%d", fd);
1150         struct thread *thread = machine__find_thread(trace->host, pid, pid);
1151 
1152         if (thread) {
1153                 const char *path = thread__fd_path(thread, fd, trace);
1154 
1155                 if (path)
1156                         printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1157 
1158                 thread__put(thread);
1159         }
1160 
1161         return printed;
1162 }
1163 
1164 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1165                                               struct syscall_arg *arg)
1166 {
1167         int fd = arg->val;
1168         size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1169         struct thread_trace *ttrace = thread__priv(arg->thread);
1170 
1171         if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1172                 zfree(&ttrace->files.table[fd].pathname);
1173 
1174         return printed;
1175 }
1176 
1177 static void thread__set_filename_pos(struct thread *thread, const char *bf,
1178                                      unsigned long ptr)
1179 {
1180         struct thread_trace *ttrace = thread__priv(thread);
1181 
1182         ttrace->filename.ptr = ptr;
1183         ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1184 }
1185 
1186 static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1187 {
1188         struct augmented_arg *augmented_arg = arg->augmented.args;
1189 
1190         return scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1191 }
1192 
1193 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1194                                               struct syscall_arg *arg)
1195 {
1196         unsigned long ptr = arg->val;
1197 
1198         if (arg->augmented.args)
1199                 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1200 
1201         if (!arg->trace->vfs_getname)
1202                 return scnprintf(bf, size, "%#x", ptr);
1203 
1204         thread__set_filename_pos(arg->thread, bf, ptr);
1205         return 0;
1206 }
1207 
1208 static bool trace__filter_duration(struct trace *trace, double t)
1209 {
1210         return t < (trace->duration_filter * NSEC_PER_MSEC);
1211 }
1212 
1213 static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1214 {
1215         double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1216 
1217         return fprintf(fp, "%10.3f ", ts);
1218 }
1219 
1220 /*
1221  * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1222  * using ttrace->entry_time for a thread that receives a sys_exit without
1223  * first having received a sys_enter ("poll" issued before tracing session
1224  * starts, lost sys_enter exit due to ring buffer overflow).
1225  */
1226 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1227 {
1228         if (tstamp > 0)
1229                 return __trace__fprintf_tstamp(trace, tstamp, fp);
1230 
1231         return fprintf(fp, "         ? ");
1232 }
1233 
1234 static bool done = false;
1235 static bool interrupted = false;
1236 
1237 static void sig_handler(int sig)
1238 {
1239         done = true;
1240         interrupted = sig == SIGINT;
1241 }
1242 
1243 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1244 {
1245         size_t printed = 0;
1246 
1247         if (trace->multiple_threads) {
1248                 if (trace->show_comm)
1249                         printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1250                 printed += fprintf(fp, "%d ", thread->tid);
1251         }
1252 
1253         return printed;
1254 }
1255 
1256 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1257                                         u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1258 {
1259         size_t printed = 0;
1260 
1261         if (trace->show_tstamp)
1262                 printed = trace__fprintf_tstamp(trace, tstamp, fp);
1263         if (trace->show_duration)
1264                 printed += fprintf_duration(duration, duration_calculated, fp);
1265         return printed + trace__fprintf_comm_tid(trace, thread, fp);
1266 }
1267 
1268 static int trace__process_event(struct trace *trace, struct machine *machine,
1269                                 union perf_event *event, struct perf_sample *sample)
1270 {
1271         int ret = 0;
1272 
1273         switch (event->header.type) {
1274         case PERF_RECORD_LOST:
1275                 color_fprintf(trace->output, PERF_COLOR_RED,
1276                               "LOST %" PRIu64 " events!\n", event->lost.lost);
1277                 ret = machine__process_lost_event(machine, event, sample);
1278                 break;
1279         default:
1280                 ret = machine__process_event(machine, event, sample);
1281                 break;
1282         }
1283 
1284         return ret;
1285 }
1286 
1287 static int trace__tool_process(struct perf_tool *tool,
1288                                union perf_event *event,
1289                                struct perf_sample *sample,
1290                                struct machine *machine)
1291 {
1292         struct trace *trace = container_of(tool, struct trace, tool);
1293         return trace__process_event(trace, machine, event, sample);
1294 }
1295 
1296 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1297 {
1298         struct machine *machine = vmachine;
1299 
1300         if (machine->kptr_restrict_warned)
1301                 return NULL;
1302 
1303         if (symbol_conf.kptr_restrict) {
1304                 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1305                            "Check /proc/sys/kernel/kptr_restrict.\n\n"
1306                            "Kernel samples will not be resolved.\n");
1307                 machine->kptr_restrict_warned = true;
1308                 return NULL;
1309         }
1310 
1311         return machine__resolve_kernel_addr(vmachine, addrp, modp);
1312 }
1313 
1314 static int trace__symbols_init(struct trace *trace, struct perf_evlist *evlist)
1315 {
1316         int err = symbol__init(NULL);
1317 
1318         if (err)
1319                 return err;
1320 
1321         trace->host = machine__new_host();
1322         if (trace->host == NULL)
1323                 return -ENOMEM;
1324 
1325         err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1326         if (err < 0)
1327                 goto out;
1328 
1329         err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1330                                             evlist->threads, trace__tool_process, false,
1331                                             1);
1332 out:
1333         if (err)
1334                 symbol__exit();
1335 
1336         return err;
1337 }
1338 
1339 static void trace__symbols__exit(struct trace *trace)
1340 {
1341         machine__exit(trace->host);
1342         trace->host = NULL;
1343 
1344         symbol__exit();
1345 }
1346 
1347 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1348 {
1349         int idx;
1350 
1351         if (nr_args == 6 && sc->fmt && sc->fmt->nr_args != 0)
1352                 nr_args = sc->fmt->nr_args;
1353 
1354         sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1355         if (sc->arg_fmt == NULL)
1356                 return -1;
1357 
1358         for (idx = 0; idx < nr_args; ++idx) {
1359                 if (sc->fmt)
1360                         sc->arg_fmt[idx] = sc->fmt->arg[idx];
1361         }
1362 
1363         sc->nr_args = nr_args;
1364         return 0;
1365 }
1366 
1367 static int syscall__set_arg_fmts(struct syscall *sc)
1368 {
1369         struct tep_format_field *field, *last_field = NULL;
1370         int idx = 0, len;
1371 
1372         for (field = sc->args; field; field = field->next, ++idx) {
1373                 last_field = field;
1374 
1375                 if (sc->fmt && sc->fmt->arg[idx].scnprintf)
1376                         continue;
1377 
1378                 if (strcmp(field->type, "const char *") == 0 &&
1379                          (strcmp(field->name, "filename") == 0 ||
1380                           strcmp(field->name, "path") == 0 ||
1381                           strcmp(field->name, "pathname") == 0))
1382                         sc->arg_fmt[idx].scnprintf = SCA_FILENAME;
1383                 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
1384                         sc->arg_fmt[idx].scnprintf = SCA_PTR;
1385                 else if (strcmp(field->type, "pid_t") == 0)
1386                         sc->arg_fmt[idx].scnprintf = SCA_PID;
1387                 else if (strcmp(field->type, "umode_t") == 0)
1388                         sc->arg_fmt[idx].scnprintf = SCA_MODE_T;
1389                 else if ((strcmp(field->type, "int") == 0 ||
1390                           strcmp(field->type, "unsigned int") == 0 ||
1391                           strcmp(field->type, "long") == 0) &&
1392                          (len = strlen(field->name)) >= 2 &&
1393                          strcmp(field->name + len - 2, "fd") == 0) {
1394                         /*
1395                          * /sys/kernel/tracing/events/syscalls/sys_enter*
1396                          * egrep 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
1397                          * 65 int
1398                          * 23 unsigned int
1399                          * 7 unsigned long
1400                          */
1401                         sc->arg_fmt[idx].scnprintf = SCA_FD;
1402                 }
1403         }
1404 
1405         if (last_field)
1406                 sc->args_size = last_field->offset + last_field->size;
1407 
1408         return 0;
1409 }
1410 
1411 static int trace__read_syscall_info(struct trace *trace, int id)
1412 {
1413         char tp_name[128];
1414         struct syscall *sc;
1415         const char *name = syscalltbl__name(trace->sctbl, id);
1416 
1417         if (name == NULL)
1418                 return -1;
1419 
1420         if (id > trace->syscalls.max) {
1421                 struct syscall *nsyscalls = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1422 
1423                 if (nsyscalls == NULL)
1424                         return -1;
1425 
1426                 if (trace->syscalls.max != -1) {
1427                         memset(nsyscalls + trace->syscalls.max + 1, 0,
1428                                (id - trace->syscalls.max) * sizeof(*sc));
1429                 } else {
1430                         memset(nsyscalls, 0, (id + 1) * sizeof(*sc));
1431                 }
1432 
1433                 trace->syscalls.table = nsyscalls;
1434                 trace->syscalls.max   = id;
1435         }
1436 
1437         sc = trace->syscalls.table + id;
1438         sc->name = name;
1439 
1440         sc->fmt  = syscall_fmt__find(sc->name);
1441 
1442         snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1443         sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1444 
1445         if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
1446                 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1447                 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1448         }
1449 
1450         if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ? 6 : sc->tp_format->format.nr_fields))
1451                 return -1;
1452 
1453         if (IS_ERR(sc->tp_format))
1454                 return -1;
1455 
1456         sc->args = sc->tp_format->format.fields;
1457         /*
1458          * We need to check and discard the first variable '__syscall_nr'
1459          * or 'nr' that mean the syscall number. It is needless here.
1460          * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1461          */
1462         if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
1463                 sc->args = sc->args->next;
1464                 --sc->nr_args;
1465         }
1466 
1467         sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1468         sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
1469 
1470         return syscall__set_arg_fmts(sc);
1471 }
1472 
1473 static int trace__validate_ev_qualifier(struct trace *trace)
1474 {
1475         int err = 0, i;
1476         size_t nr_allocated;
1477         struct str_node *pos;
1478 
1479         trace->ev_qualifier_ids.nr = strlist__nr_entries(trace->ev_qualifier);
1480         trace->ev_qualifier_ids.entries = malloc(trace->ev_qualifier_ids.nr *
1481                                                  sizeof(trace->ev_qualifier_ids.entries[0]));
1482 
1483         if (trace->ev_qualifier_ids.entries == NULL) {
1484                 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1485                        trace->output);
1486                 err = -EINVAL;
1487                 goto out;
1488         }
1489 
1490         nr_allocated = trace->ev_qualifier_ids.nr;
1491         i = 0;
1492 
1493         strlist__for_each_entry(pos, trace->ev_qualifier) {
1494                 const char *sc = pos->s;
1495                 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
1496 
1497                 if (id < 0) {
1498                         id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1499                         if (id >= 0)
1500                                 goto matches;
1501 
1502                         if (err == 0) {
1503                                 fputs("Error:\tInvalid syscall ", trace->output);
1504                                 err = -EINVAL;
1505                         } else {
1506                                 fputs(", ", trace->output);
1507                         }
1508 
1509                         fputs(sc, trace->output);
1510                 }
1511 matches:
1512                 trace->ev_qualifier_ids.entries[i++] = id;
1513                 if (match_next == -1)
1514                         continue;
1515 
1516                 while (1) {
1517                         id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1518                         if (id < 0)
1519                                 break;
1520                         if (nr_allocated == trace->ev_qualifier_ids.nr) {
1521                                 void *entries;
1522 
1523                                 nr_allocated += 8;
1524                                 entries = realloc(trace->ev_qualifier_ids.entries,
1525                                                   nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1526                                 if (entries == NULL) {
1527                                         err = -ENOMEM;
1528                                         fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1529                                         goto out_free;
1530                                 }
1531                                 trace->ev_qualifier_ids.entries = entries;
1532                         }
1533                         trace->ev_qualifier_ids.nr++;
1534                         trace->ev_qualifier_ids.entries[i++] = id;
1535                 }
1536         }
1537 
1538         if (err < 0) {
1539                 fputs("\nHint:\ttry 'perf list syscalls:sys_enter_*'"
1540                       "\nHint:\tand: 'man syscalls'\n", trace->output);
1541 out_free:
1542                 zfree(&trace->ev_qualifier_ids.entries);
1543                 trace->ev_qualifier_ids.nr = 0;
1544         }
1545 out:
1546         return err;
1547 }
1548 
1549 /*
1550  * args is to be interpreted as a series of longs but we need to handle
1551  * 8-byte unaligned accesses. args points to raw_data within the event
1552  * and raw_data is guaranteed to be 8-byte unaligned because it is
1553  * preceded by raw_size which is a u32. So we need to copy args to a temp
1554  * variable to read it. Most notably this avoids extended load instructions
1555  * on unaligned addresses
1556  */
1557 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
1558 {
1559         unsigned long val;
1560         unsigned char *p = arg->args + sizeof(unsigned long) * idx;
1561 
1562         memcpy(&val, p, sizeof(val));
1563         return val;
1564 }
1565 
1566 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1567                                       struct syscall_arg *arg)
1568 {
1569         if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1570                 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1571 
1572         return scnprintf(bf, size, "arg%d: ", arg->idx);
1573 }
1574 
1575 /*
1576  * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1577  * as mount 'flags' argument that needs ignoring some magic flag, see comment
1578  * in tools/perf/trace/beauty/mount_flags.c
1579  */
1580 static unsigned long syscall__mask_val(struct syscall *sc, struct syscall_arg *arg, unsigned long val)
1581 {
1582         if (sc->arg_fmt && sc->arg_fmt[arg->idx].mask_val)
1583                 return sc->arg_fmt[arg->idx].mask_val(arg, val);
1584 
1585         return val;
1586 }
1587 
1588 static size_t syscall__scnprintf_val(struct syscall *sc, char *bf, size_t size,
1589                                      struct syscall_arg *arg, unsigned long val)
1590 {
1591         if (sc->arg_fmt && sc->arg_fmt[arg->idx].scnprintf) {
1592                 arg->val = val;
1593                 if (sc->arg_fmt[arg->idx].parm)
1594                         arg->parm = sc->arg_fmt[arg->idx].parm;
1595                 return sc->arg_fmt[arg->idx].scnprintf(bf, size, arg);
1596         }
1597         return scnprintf(bf, size, "%ld", val);
1598 }
1599 
1600 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
1601                                       unsigned char *args, void *augmented_args, int augmented_args_size,
1602                                       struct trace *trace, struct thread *thread)
1603 {
1604         size_t printed = 0;
1605         unsigned long val;
1606         u8 bit = 1;
1607         struct syscall_arg arg = {
1608                 .args   = args,
1609                 .augmented = {
1610                         .size = augmented_args_size,
1611                         .args = augmented_args,
1612                 },
1613                 .idx    = 0,
1614                 .mask   = 0,
1615                 .trace  = trace,
1616                 .thread = thread,
1617                 .show_string_prefix = trace->show_string_prefix,
1618         };
1619         struct thread_trace *ttrace = thread__priv(thread);
1620 
1621         /*
1622          * Things like fcntl will set this in its 'cmd' formatter to pick the
1623          * right formatter for the return value (an fd? file flags?), which is
1624          * not needed for syscalls that always return a given type, say an fd.
1625          */
1626         ttrace->ret_scnprintf = NULL;
1627 
1628         if (sc->args != NULL) {
1629                 struct tep_format_field *field;
1630 
1631                 for (field = sc->args; field;
1632                      field = field->next, ++arg.idx, bit <<= 1) {
1633                         if (arg.mask & bit)
1634                                 continue;
1635 
1636                         val = syscall_arg__val(&arg, arg.idx);
1637                         /*
1638                          * Some syscall args need some mask, most don't and
1639                          * return val untouched.
1640                          */
1641                         val = syscall__mask_val(sc, &arg, val);
1642 
1643                         /*
1644                          * Suppress this argument if its value is zero and
1645                          * and we don't have a string associated in an
1646                          * strarray for it.
1647                          */
1648                         if (val == 0 &&
1649                             !trace->show_zeros &&
1650                             !(sc->arg_fmt &&
1651                               (sc->arg_fmt[arg.idx].show_zero ||
1652                                sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
1653                                sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
1654                               sc->arg_fmt[arg.idx].parm))
1655                                 continue;
1656 
1657                         printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
1658 
1659                         if (trace->show_arg_names)
1660                                 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
1661 
1662                         printed += syscall__scnprintf_val(sc, bf + printed, size - printed, &arg, val);
1663                 }
1664         } else if (IS_ERR(sc->tp_format)) {
1665                 /*
1666                  * If we managed to read the tracepoint /format file, then we
1667                  * may end up not having any args, like with gettid(), so only
1668                  * print the raw args when we didn't manage to read it.
1669                  */
1670                 while (arg.idx < sc->nr_args) {
1671                         if (arg.mask & bit)
1672                                 goto next_arg;
1673                         val = syscall_arg__val(&arg, arg.idx);
1674                         if (printed)
1675                                 printed += scnprintf(bf + printed, size - printed, ", ");
1676                         printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
1677                         printed += syscall__scnprintf_val(sc, bf + printed, size - printed, &arg, val);
1678 next_arg:
1679                         ++arg.idx;
1680                         bit <<= 1;
1681                 }
1682         }
1683 
1684         return printed;
1685 }
1686 
1687 typedef int (*tracepoint_handler)(struct trace *trace, struct perf_evsel *evsel,
1688                                   union perf_event *event,
1689                                   struct perf_sample *sample);
1690 
1691 static struct syscall *trace__syscall_info(struct trace *trace,
1692                                            struct perf_evsel *evsel, int id)
1693 {
1694 
1695         if (id < 0) {
1696 
1697                 /*
1698                  * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
1699                  * before that, leaving at a higher verbosity level till that is
1700                  * explained. Reproduced with plain ftrace with:
1701                  *
1702                  * echo 1 > /t/events/raw_syscalls/sys_exit/enable
1703                  * grep "NR -1 " /t/trace_pipe
1704                  *
1705                  * After generating some load on the machine.
1706                  */
1707                 if (verbose > 1) {
1708                         static u64 n;
1709                         fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
1710                                 id, perf_evsel__name(evsel), ++n);
1711                 }
1712                 return NULL;
1713         }
1714 
1715         if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL) &&
1716             trace__read_syscall_info(trace, id))
1717                 goto out_cant_read;
1718 
1719         if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL))
1720                 goto out_cant_read;
1721 
1722         return &trace->syscalls.table[id];
1723 
1724 out_cant_read:
1725         if (verbose > 0) {
1726                 fprintf(trace->output, "Problems reading syscall %d", id);
1727                 if (id <= trace->syscalls.max && trace->syscalls.table[id].name != NULL)
1728                         fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
1729                 fputs(" information\n", trace->output);
1730         }
1731         return NULL;
1732 }
1733 
1734 static void thread__update_stats(struct thread_trace *ttrace,
1735                                  int id, struct perf_sample *sample)
1736 {
1737         struct int_node *inode;
1738         struct stats *stats;
1739         u64 duration = 0;
1740 
1741         inode = intlist__findnew(ttrace->syscall_stats, id);
1742         if (inode == NULL)
1743                 return;
1744 
1745         stats = inode->priv;
1746         if (stats == NULL) {
1747                 stats = malloc(sizeof(struct stats));
1748                 if (stats == NULL)
1749                         return;
1750                 init_stats(stats);
1751                 inode->priv = stats;
1752         }
1753 
1754         if (ttrace->entry_time && sample->time > ttrace->entry_time)
1755                 duration = sample->time - ttrace->entry_time;
1756 
1757         update_stats(stats, duration);
1758 }
1759 
1760 static int trace__printf_interrupted_entry(struct trace *trace)
1761 {
1762         struct thread_trace *ttrace;
1763         size_t printed;
1764         int len;
1765 
1766         if (trace->failure_only || trace->current == NULL)
1767                 return 0;
1768 
1769         ttrace = thread__priv(trace->current);
1770 
1771         if (!ttrace->entry_pending)
1772                 return 0;
1773 
1774         printed  = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
1775         printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
1776 
1777         if (len < trace->args_alignment - 4)
1778                 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
1779 
1780         printed += fprintf(trace->output, " ...\n");
1781 
1782         ttrace->entry_pending = false;
1783         ++trace->nr_events_printed;
1784 
1785         return printed;
1786 }
1787 
1788 static int trace__fprintf_sample(struct trace *trace, struct perf_evsel *evsel,
1789                                  struct perf_sample *sample, struct thread *thread)
1790 {
1791         int printed = 0;
1792 
1793         if (trace->print_sample) {
1794                 double ts = (double)sample->time / NSEC_PER_MSEC;
1795 
1796                 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
1797                                    perf_evsel__name(evsel), ts,
1798                                    thread__comm_str(thread),
1799                                    sample->pid, sample->tid, sample->cpu);
1800         }
1801 
1802         return printed;
1803 }
1804 
1805 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
1806 {
1807         void *augmented_args = NULL;
1808         /*
1809          * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
1810          * and there we get all 6 syscall args plus the tracepoint common fields
1811          * that gets calculated at the start and the syscall_nr (another long).
1812          * So we check if that is the case and if so don't look after the
1813          * sc->args_size but always after the full raw_syscalls:sys_enter payload,
1814          * which is fixed.
1815          *
1816          * We'll revisit this later to pass s->args_size to the BPF augmenter
1817          * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
1818          * copies only what we need for each syscall, like what happens when we
1819          * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
1820          * traffic to just what is needed for each syscall.
1821          */
1822         int args_size = raw_augmented_args_size ?: sc->args_size;
1823 
1824         *augmented_args_size = sample->raw_size - args_size;
1825         if (*augmented_args_size > 0)
1826                 augmented_args = sample->raw_data + args_size;
1827 
1828         return augmented_args;
1829 }
1830 
1831 static int trace__sys_enter(struct trace *trace, struct perf_evsel *evsel,
1832                             union perf_event *event __maybe_unused,
1833                             struct perf_sample *sample)
1834 {
1835         char *msg;
1836         void *args;
1837         int printed = 0;
1838         struct thread *thread;
1839         int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
1840         int augmented_args_size = 0;
1841         void *augmented_args = NULL;
1842         struct syscall *sc = trace__syscall_info(trace, evsel, id);
1843         struct thread_trace *ttrace;
1844 
1845         if (sc == NULL)
1846                 return -1;
1847 
1848         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1849         ttrace = thread__trace(thread, trace->output);
1850         if (ttrace == NULL)
1851                 goto out_put;
1852 
1853         trace__fprintf_sample(trace, evsel, sample, thread);
1854 
1855         args = perf_evsel__sc_tp_ptr(evsel, args, sample);
1856 
1857         if (ttrace->entry_str == NULL) {
1858                 ttrace->entry_str = malloc(trace__entry_str_size);
1859                 if (!ttrace->entry_str)
1860                         goto out_put;
1861         }
1862 
1863         if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
1864                 trace__printf_interrupted_entry(trace);
1865         /*
1866          * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
1867          * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
1868          * this breaks syscall__augmented_args() check for augmented args, as we calculate
1869          * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
1870          * so when handling, say the openat syscall, we end up getting 6 args for the
1871          * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
1872          * thinking that the extra 2 u64 args are the augmented filename, so just check
1873          * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
1874          */
1875         if (evsel != trace->syscalls.events.sys_enter)
1876                 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
1877         ttrace->entry_time = sample->time;
1878         msg = ttrace->entry_str;
1879         printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
1880 
1881         printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
1882                                            args, augmented_args, augmented_args_size, trace, thread);
1883 
1884         if (sc->is_exit) {
1885                 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
1886                         int alignment = 0;
1887 
1888                         trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
1889                         printed = fprintf(trace->output, "%s)", ttrace->entry_str);
1890                         if (trace->args_alignment > printed)
1891                                 alignment = trace->args_alignment - printed;
1892                         fprintf(trace->output, "%*s= ?\n", alignment, " ");
1893                 }
1894         } else {
1895                 ttrace->entry_pending = true;
1896                 /* See trace__vfs_getname & trace__sys_exit */
1897                 ttrace->filename.pending_open = false;
1898         }
1899 
1900         if (trace->current != thread) {
1901                 thread__put(trace->current);
1902                 trace->current = thread__get(thread);
1903         }
1904         err = 0;
1905 out_put:
1906         thread__put(thread);
1907         return err;
1908 }
1909 
1910 static int trace__fprintf_sys_enter(struct trace *trace, struct perf_evsel *evsel,
1911                                     struct perf_sample *sample)
1912 {
1913         struct thread_trace *ttrace;
1914         struct thread *thread;
1915         int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
1916         struct syscall *sc = trace__syscall_info(trace, evsel, id);
1917         char msg[1024];
1918         void *args, *augmented_args = NULL;
1919         int augmented_args_size;
1920 
1921         if (sc == NULL)
1922                 return -1;
1923 
1924         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1925         ttrace = thread__trace(thread, trace->output);
1926         /*
1927          * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
1928          * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
1929          */
1930         if (ttrace == NULL)
1931                 goto out_put;
1932 
1933         args = perf_evsel__sc_tp_ptr(evsel, args, sample);
1934         augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
1935         syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
1936         fprintf(trace->output, "%s", msg);
1937         err = 0;
1938 out_put:
1939         thread__put(thread);
1940         return err;
1941 }
1942 
1943 static int trace__resolve_callchain(struct trace *trace, struct perf_evsel *evsel,
1944                                     struct perf_sample *sample,
1945                                     struct callchain_cursor *cursor)
1946 {
1947         struct addr_location al;
1948         int max_stack = evsel->attr.sample_max_stack ?
1949                         evsel->attr.sample_max_stack :
1950                         trace->max_stack;
1951         int err;
1952 
1953         if (machine__resolve(trace->host, &al, sample) < 0)
1954                 return -1;
1955 
1956         err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
1957         addr_location__put(&al);
1958         return err;
1959 }
1960 
1961 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
1962 {
1963         /* TODO: user-configurable print_opts */
1964         const unsigned int print_opts = EVSEL__PRINT_SYM |
1965                                         EVSEL__PRINT_DSO |
1966                                         EVSEL__PRINT_UNKNOWN_AS_ADDR;
1967 
1968         return sample__fprintf_callchain(sample, 38, print_opts, &callchain_cursor, trace->output);
1969 }
1970 
1971 static const char *errno_to_name(struct perf_evsel *evsel, int err)
1972 {
1973         struct perf_env *env = perf_evsel__env(evsel);
1974         const char *arch_name = perf_env__arch(env);
1975 
1976         return arch_syscalls__strerrno(arch_name, err);
1977 }
1978 
1979 static int trace__sys_exit(struct trace *trace, struct perf_evsel *evsel,
1980                            union perf_event *event __maybe_unused,
1981                            struct perf_sample *sample)
1982 {
1983         long ret;
1984         u64 duration = 0;
1985         bool duration_calculated = false;
1986         struct thread *thread;
1987         int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
1988         int alignment = trace->args_alignment;
1989         struct syscall *sc = trace__syscall_info(trace, evsel, id);
1990         struct thread_trace *ttrace;
1991 
1992         if (sc == NULL)
1993                 return -1;
1994 
1995         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1996         ttrace = thread__trace(thread, trace->output);
1997         if (ttrace == NULL)
1998                 goto out_put;
1999 
2000         trace__fprintf_sample(trace, evsel, sample, thread);
2001 
2002         if (trace->summary)
2003                 thread__update_stats(ttrace, id, sample);
2004 
2005         ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2006 
2007         if (sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2008                 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2009                 ttrace->filename.pending_open = false;
2010                 ++trace->stats.vfs_getname;
2011         }
2012 
2013         if (ttrace->entry_time) {
2014                 duration = sample->time - ttrace->entry_time;
2015                 if (trace__filter_duration(trace, duration))
2016                         goto out;
2017                 duration_calculated = true;
2018         } else if (trace->duration_filter)
2019                 goto out;
2020 
2021         if (sample->callchain) {
2022                 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2023                 if (callchain_ret == 0) {
2024                         if (callchain_cursor.nr < trace->min_stack)
2025                                 goto out;
2026                         callchain_ret = 1;
2027                 }
2028         }
2029 
2030         if (trace->summary_only || (ret >= 0 && trace->failure_only))
2031                 goto out;
2032 
2033         trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2034 
2035         if (ttrace->entry_pending) {
2036                 printed = fprintf(trace->output, "%s", ttrace->entry_str);
2037         } else {
2038                 printed += fprintf(trace->output, " ... [");
2039                 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2040                 printed += 9;
2041                 printed += fprintf(trace->output, "]: %s()", sc->name);
2042         }
2043 
2044         printed++; /* the closing ')' */
2045 
2046         if (alignment > printed)
2047                 alignment -= printed;
2048         else
2049                 alignment = 0;
2050 
2051         fprintf(trace->output, ")%*s= ", alignment, " ");
2052 
2053         if (sc->fmt == NULL) {
2054                 if (ret < 0)
2055                         goto errno_print;
2056 signed_print:
2057                 fprintf(trace->output, "%ld", ret);
2058         } else if (ret < 0) {
2059 errno_print: {
2060                 char bf[STRERR_BUFSIZE];
2061                 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2062                            *e = errno_to_name(evsel, -ret);
2063 
2064                 fprintf(trace->output, "-1 %s (%s)", e, emsg);
2065         }
2066         } else if (ret == 0 && sc->fmt->timeout)
2067                 fprintf(trace->output, "0 (Timeout)");
2068         else if (ttrace->ret_scnprintf) {
2069                 char bf[1024];
2070                 struct syscall_arg arg = {
2071                         .val    = ret,
2072                         .thread = thread,
2073                         .trace  = trace,
2074                 };
2075                 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2076                 ttrace->ret_scnprintf = NULL;
2077                 fprintf(trace->output, "%s", bf);
2078         } else if (sc->fmt->hexret)
2079                 fprintf(trace->output, "%#lx", ret);
2080         else if (sc->fmt->errpid) {
2081                 struct thread *child = machine__find_thread(trace->host, ret, ret);
2082 
2083                 if (child != NULL) {
2084                         fprintf(trace->output, "%ld", ret);
2085                         if (child->comm_set)
2086                                 fprintf(trace->output, " (%s)", thread__comm_str(child));
2087                         thread__put(child);
2088                 }
2089         } else
2090                 goto signed_print;
2091 
2092         fputc('\n', trace->output);
2093 
2094         /*
2095          * We only consider an 'event' for the sake of --max-events a non-filtered
2096          * sys_enter + sys_exit and other tracepoint events.
2097          */
2098         if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2099                 interrupted = true;
2100 
2101         if (callchain_ret > 0)
2102                 trace__fprintf_callchain(trace, sample);
2103         else if (callchain_ret < 0)
2104                 pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2105 out:
2106         ttrace->entry_pending = false;
2107         err = 0;
2108 out_put:
2109         thread__put(thread);
2110         return err;
2111 }
2112 
2113 static int trace__vfs_getname(struct trace *trace, struct perf_evsel *evsel,
2114                               union perf_event *event __maybe_unused,
2115                               struct perf_sample *sample)
2116 {
2117         struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2118         struct thread_trace *ttrace;
2119         size_t filename_len, entry_str_len, to_move;
2120         ssize_t remaining_space;
2121         char *pos;
2122         const char *filename = perf_evsel__rawptr(evsel, sample, "pathname");
2123 
2124         if (!thread)
2125                 goto out;
2126 
2127         ttrace = thread__priv(thread);
2128         if (!ttrace)
2129                 goto out_put;
2130 
2131         filename_len = strlen(filename);
2132         if (filename_len == 0)
2133                 goto out_put;
2134 
2135         if (ttrace->filename.namelen < filename_len) {
2136                 char *f = realloc(ttrace->filename.name, filename_len + 1);
2137 
2138                 if (f == NULL)
2139                         goto out_put;
2140 
2141                 ttrace->filename.namelen = filename_len;
2142                 ttrace->filename.name = f;
2143         }
2144 
2145         strcpy(ttrace->filename.name, filename);
2146         ttrace->filename.pending_open = true;
2147 
2148         if (!ttrace->filename.ptr)
2149                 goto out_put;
2150 
2151         entry_str_len = strlen(ttrace->entry_str);
2152         remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2153         if (remaining_space <= 0)
2154                 goto out_put;
2155 
2156         if (filename_len > (size_t)remaining_space) {
2157                 filename += filename_len - remaining_space;
2158                 filename_len = remaining_space;
2159         }
2160 
2161         to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2162         pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2163         memmove(pos + filename_len, pos, to_move);
2164         memcpy(pos, filename, filename_len);
2165 
2166         ttrace->filename.ptr = 0;
2167         ttrace->filename.entry_str_pos = 0;
2168 out_put:
2169         thread__put(thread);
2170 out:
2171         return 0;
2172 }
2173 
2174 static int trace__sched_stat_runtime(struct trace *trace, struct perf_evsel *evsel,
2175                                      union perf_event *event __maybe_unused,
2176                                      struct perf_sample *sample)
2177 {
2178         u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
2179         double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2180         struct thread *thread = machine__findnew_thread(trace->host,
2181                                                         sample->pid,
2182                                                         sample->tid);
2183         struct thread_trace *ttrace = thread__trace(thread, trace->output);
2184 
2185         if (ttrace == NULL)
2186                 goto out_dump;
2187 
2188         ttrace->runtime_ms += runtime_ms;
2189         trace->runtime_ms += runtime_ms;
2190 out_put:
2191         thread__put(thread);
2192         return 0;
2193 
2194 out_dump:
2195         fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2196                evsel->name,
2197                perf_evsel__strval(evsel, sample, "comm"),
2198                (pid_t)perf_evsel__intval(evsel, sample, "pid"),
2199                runtime,
2200                perf_evsel__intval(evsel, sample, "vruntime"));
2201         goto out_put;
2202 }
2203 
2204 static int bpf_output__printer(enum binary_printer_ops op,
2205                                unsigned int val, void *extra __maybe_unused, FILE *fp)
2206 {
2207         unsigned char ch = (unsigned char)val;
2208 
2209         switch (op) {
2210         case BINARY_PRINT_CHAR_DATA:
2211                 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
2212         case BINARY_PRINT_DATA_BEGIN:
2213         case BINARY_PRINT_LINE_BEGIN:
2214         case BINARY_PRINT_ADDR:
2215         case BINARY_PRINT_NUM_DATA:
2216         case BINARY_PRINT_NUM_PAD:
2217         case BINARY_PRINT_SEP:
2218         case BINARY_PRINT_CHAR_PAD:
2219         case BINARY_PRINT_LINE_END:
2220         case BINARY_PRINT_DATA_END:
2221         default:
2222                 break;
2223         }
2224 
2225         return 0;
2226 }
2227 
2228 static void bpf_output__fprintf(struct trace *trace,
2229                                 struct perf_sample *sample)
2230 {
2231         binary__fprintf(sample->raw_data, sample->raw_size, 8,
2232                         bpf_output__printer, NULL, trace->output);
2233         ++trace->nr_events_printed;
2234 }
2235 
2236 static int trace__event_handler(struct trace *trace, struct perf_evsel *evsel,
2237                                 union perf_event *event __maybe_unused,
2238                                 struct perf_sample *sample)
2239 {
2240         struct thread *thread;
2241         int callchain_ret = 0;
2242         /*
2243          * Check if we called perf_evsel__disable(evsel) due to, for instance,
2244          * this event's max_events having been hit and this is an entry coming
2245          * from the ring buffer that we should discard, since the max events
2246          * have already been considered/printed.
2247          */
2248         if (evsel->disabled)
2249                 return 0;
2250 
2251         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2252 
2253         if (sample->callchain) {
2254                 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2255                 if (callchain_ret == 0) {
2256                         if (callchain_cursor.nr < trace->min_stack)
2257                                 goto out;
2258                         callchain_ret = 1;
2259                 }
2260         }
2261 
2262         trace__printf_interrupted_entry(trace);
2263         trace__fprintf_tstamp(trace, sample->time, trace->output);
2264 
2265         if (trace->trace_syscalls && trace->show_duration)
2266                 fprintf(trace->output, "(         ): ");
2267 
2268         if (thread)
2269                 trace__fprintf_comm_tid(trace, thread, trace->output);
2270 
2271         if (evsel == trace->syscalls.events.augmented) {
2272                 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2273                 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2274 
2275                 if (sc) {
2276                         fprintf(trace->output, "%s(", sc->name);
2277                         trace__fprintf_sys_enter(trace, evsel, sample);
2278                         fputc(')', trace->output);
2279                         goto newline;
2280                 }
2281 
2282                 /*
2283                  * XXX: Not having the associated syscall info or not finding/adding
2284                  *      the thread should never happen, but if it does...
2285                  *      fall thru and print it as a bpf_output event.
2286                  */
2287         }
2288 
2289         fprintf(trace->output, "%s:", evsel->name);
2290 
2291         if (perf_evsel__is_bpf_output(evsel)) {
2292                 bpf_output__fprintf(trace, sample);
2293         } else if (evsel->tp_format) {
2294                 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2295                     trace__fprintf_sys_enter(trace, evsel, sample)) {
2296                         event_format__fprintf(evsel->tp_format, sample->cpu,
2297                                               sample->raw_data, sample->raw_size,
2298                                               trace->output);
2299                         ++trace->nr_events_printed;
2300 
2301                         if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2302                                 perf_evsel__disable(evsel);
2303                                 perf_evsel__close(evsel);
2304                         }
2305                 }
2306         }
2307 
2308 newline:
2309         fprintf(trace->output, "\n");
2310 
2311         if (callchain_ret > 0)
2312                 trace__fprintf_callchain(trace, sample);
2313         else if (callchain_ret < 0)
2314                 pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2315 out:
2316         thread__put(thread);
2317         return 0;
2318 }
2319 
2320 static void print_location(FILE *f, struct perf_sample *sample,
2321                            struct addr_location *al,
2322                            bool print_dso, bool print_sym)
2323 {
2324 
2325         if ((verbose > 0 || print_dso) && al->map)
2326                 fprintf(f, "%s@", al->map->dso->long_name);
2327 
2328         if ((verbose > 0 || print_sym) && al->sym)
2329                 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
2330                         al->addr - al->sym->start);
2331         else if (al->map)
2332                 fprintf(f, "0x%" PRIx64, al->addr);
2333         else
2334                 fprintf(f, "0x%" PRIx64, sample->addr);
2335 }
2336 
2337 static int trace__pgfault(struct trace *trace,
2338                           struct perf_evsel *evsel,
2339                           union perf_event *event __maybe_unused,
2340                           struct perf_sample *sample)
2341 {
2342         struct thread *thread;
2343         struct addr_location al;
2344         char map_type = 'd';
2345         struct thread_trace *ttrace;
2346         int err = -1;
2347         int callchain_ret = 0;
2348 
2349         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2350 
2351         if (sample->callchain) {
2352                 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2353                 if (callchain_ret == 0) {
2354                         if (callchain_cursor.nr < trace->min_stack)
2355                                 goto out_put;
2356                         callchain_ret = 1;
2357                 }
2358         }
2359 
2360         ttrace = thread__trace(thread, trace->output);
2361         if (ttrace == NULL)
2362                 goto out_put;
2363 
2364         if (evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2365                 ttrace->pfmaj++;
2366         else
2367                 ttrace->pfmin++;
2368 
2369         if (trace->summary_only)
2370                 goto out;
2371 
2372         thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
2373 
2374         trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
2375 
2376         fprintf(trace->output, "%sfault [",
2377                 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2378                 "maj" : "min");
2379 
2380         print_location(trace->output, sample, &al, false, true);
2381 
2382         fprintf(trace->output, "] => ");
2383 
2384         thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2385 
2386         if (!al.map) {
2387                 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2388 
2389                 if (al.map)
2390                         map_type = 'x';
2391                 else
2392                         map_type = '?';
2393         }
2394 
2395         print_location(trace->output, sample, &al, true, false);
2396 
2397         fprintf(trace->output, " (%c%c)\n", map_type, al.level);
2398 
2399         if (callchain_ret > 0)
2400                 trace__fprintf_callchain(trace, sample);
2401         else if (callchain_ret < 0)
2402                 pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2403 
2404         ++trace->nr_events_printed;
2405 out:
2406         err = 0;
2407 out_put:
2408         thread__put(thread);
2409         return err;
2410 }
2411 
2412 static void trace__set_base_time(struct trace *trace,
2413                                  struct perf_evsel *evsel,
2414                                  struct perf_sample *sample)
2415 {
2416         /*
2417          * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2418          * and don't use sample->time unconditionally, we may end up having
2419          * some other event in the future without PERF_SAMPLE_TIME for good
2420          * reason, i.e. we may not be interested in its timestamps, just in
2421          * it taking place, picking some piece of information when it
2422          * appears in our event stream (vfs_getname comes to mind).
2423          */
2424         if (trace->base_time == 0 && !trace->full_time &&
2425             (evsel->attr.sample_type & PERF_SAMPLE_TIME))
2426                 trace->base_time = sample->time;
2427 }
2428 
2429 static int trace__process_sample(struct perf_tool *tool,
2430                                  union perf_event *event,
2431                                  struct perf_sample *sample,
2432                                  struct perf_evsel *evsel,
2433                                  struct machine *machine __maybe_unused)
2434 {
2435         struct trace *trace = container_of(tool, struct trace, tool);
2436         struct thread *thread;
2437         int err = 0;
2438 
2439         tracepoint_handler handler = evsel->handler;
2440 
2441         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2442         if (thread && thread__is_filtered(thread))
2443                 goto out;
2444 
2445         trace__set_base_time(trace, evsel, sample);
2446 
2447         if (handler) {
2448                 ++trace->nr_events;
2449                 handler(trace, evsel, event, sample);
2450         }
2451 out:
2452         thread__put(thread);
2453         return err;
2454 }
2455 
2456 static int trace__record(struct trace *trace, int argc, const char **argv)
2457 {
2458         unsigned int rec_argc, i, j;
2459         const char **rec_argv;
2460         const char * const record_args[] = {
2461                 "record",
2462                 "-R",
2463                 "-m", "1024",
2464                 "-c", "1",
2465         };
2466 
2467         const char * const sc_args[] = { "-e", };
2468         unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
2469         const char * const majpf_args[] = { "-e", "major-faults" };
2470         unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
2471         const char * const minpf_args[] = { "-e", "minor-faults" };
2472         unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
2473 
2474         /* +1 is for the event string below */
2475         rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 1 +
2476                 majpf_args_nr + minpf_args_nr + argc;
2477         rec_argv = calloc(rec_argc + 1, sizeof(char *));
2478 
2479         if (rec_argv == NULL)
2480                 return -ENOMEM;
2481 
2482         j = 0;
2483         for (i = 0; i < ARRAY_SIZE(record_args); i++)
2484                 rec_argv[j++] = record_args[i];
2485 
2486         if (trace->trace_syscalls) {
2487                 for (i = 0; i < sc_args_nr; i++)
2488                         rec_argv[j++] = sc_args[i];
2489 
2490                 /* event string may be different for older kernels - e.g., RHEL6 */
2491                 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
2492                         rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
2493                 else if (is_valid_tracepoint("syscalls:sys_enter"))
2494                         rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
2495                 else {
2496                         pr_err("Neither raw_syscalls nor syscalls events exist.\n");
2497                         free(rec_argv);
2498                         return -1;
2499                 }
2500         }
2501 
2502         if (trace->trace_pgfaults & TRACE_PFMAJ)
2503                 for (i = 0; i < majpf_args_nr; i++)
2504                         rec_argv[j++] = majpf_args[i];
2505 
2506         if (trace->trace_pgfaults & TRACE_PFMIN)
2507                 for (i = 0; i < minpf_args_nr; i++)
2508                         rec_argv[j++] = minpf_args[i];
2509 
2510         for (i = 0; i < (unsigned int)argc; i++)
2511                 rec_argv[j++] = argv[i];
2512 
2513         return cmd_record(j, rec_argv);
2514 }
2515 
2516 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
2517 
2518 static bool perf_evlist__add_vfs_getname(struct perf_evlist *evlist)
2519 {
2520         bool found = false;
2521         struct perf_evsel *evsel, *tmp;
2522         struct parse_events_error err = { .idx = 0, };
2523         int ret = parse_events(evlist, "probe:vfs_getname*", &err);
2524 
2525         if (ret)
2526                 return false;
2527 
2528         evlist__for_each_entry_safe(evlist, evsel, tmp) {
2529                 if (!strstarts(perf_evsel__name(evsel), "probe:vfs_getname"))
2530                         continue;
2531 
2532                 if (perf_evsel__field(evsel, "pathname")) {
2533                         evsel->handler = trace__vfs_getname;
2534                         found = true;
2535                         continue;
2536                 }
2537 
2538                 list_del_init(&evsel->node);
2539                 evsel->evlist = NULL;
2540                 perf_evsel__delete(evsel);
2541         }
2542 
2543         return found;
2544 }
2545 
2546 static struct perf_evsel *perf_evsel__new_pgfault(u64 config)
2547 {
2548         struct perf_evsel *evsel;
2549         struct perf_event_attr attr = {
2550                 .type = PERF_TYPE_SOFTWARE,
2551                 .mmap_data = 1,
2552         };
2553 
2554         attr.config = config;
2555         attr.sample_period = 1;
2556 
2557         event_attr_init(&attr);
2558 
2559         evsel = perf_evsel__new(&attr);
2560         if (evsel)
2561                 evsel->handler = trace__pgfault;
2562 
2563         return evsel;
2564 }
2565 
2566 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
2567 {
2568         const u32 type = event->header.type;
2569         struct perf_evsel *evsel;
2570 
2571         if (type != PERF_RECORD_SAMPLE) {
2572                 trace__process_event(trace, trace->host, event, sample);
2573                 return;
2574         }
2575 
2576         evsel = perf_evlist__id2evsel(trace->evlist, sample->id);
2577         if (evsel == NULL) {
2578                 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
2579                 return;
2580         }
2581 
2582         trace__set_base_time(trace, evsel, sample);
2583 
2584         if (evsel->attr.type == PERF_TYPE_TRACEPOINT &&
2585             sample->raw_data == NULL) {
2586                 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
2587                        perf_evsel__name(evsel), sample->tid,
2588                        sample->cpu, sample->raw_size);
2589         } else {
2590                 tracepoint_handler handler = evsel->handler;
2591                 handler(trace, evsel, event, sample);
2592         }
2593 
2594         if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
2595                 interrupted = true;
2596 }
2597 
2598 static int trace__add_syscall_newtp(struct trace *trace)
2599 {
2600         int ret = -1;
2601         struct perf_evlist *evlist = trace->evlist;
2602         struct perf_evsel *sys_enter, *sys_exit;
2603 
2604         sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
2605         if (sys_enter == NULL)
2606                 goto out;
2607 
2608         if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
2609                 goto out_delete_sys_enter;
2610 
2611         sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
2612         if (sys_exit == NULL)
2613                 goto out_delete_sys_enter;
2614 
2615         if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
2616                 goto out_delete_sys_exit;
2617 
2618         perf_evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
2619         perf_evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
2620 
2621         perf_evlist__add(evlist, sys_enter);
2622         perf_evlist__add(evlist, sys_exit);
2623 
2624         if (callchain_param.enabled && !trace->kernel_syscallchains) {
2625                 /*
2626                  * We're interested only in the user space callchain
2627                  * leading to the syscall, allow overriding that for
2628                  * debugging reasons using --kernel_syscall_callchains
2629                  */
2630                 sys_exit->attr.exclude_callchain_kernel = 1;
2631         }
2632 
2633         trace->syscalls.events.sys_enter = sys_enter;
2634         trace->syscalls.events.sys_exit  = sys_exit;
2635 
2636         ret = 0;
2637 out:
2638         return ret;
2639 
2640 out_delete_sys_exit:
2641         perf_evsel__delete_priv(sys_exit);
2642 out_delete_sys_enter:
2643         perf_evsel__delete_priv(sys_enter);
2644         goto out;
2645 }
2646 
2647 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
2648 {
2649         int err = -1;
2650         struct perf_evsel *sys_exit;
2651         char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
2652                                                 trace->ev_qualifier_ids.nr,
2653                                                 trace->ev_qualifier_ids.entries);
2654 
2655         if (filter == NULL)
2656                 goto out_enomem;
2657 
2658         if (!perf_evsel__append_tp_filter(trace->syscalls.events.sys_enter,
2659                                           filter)) {
2660                 sys_exit = trace->syscalls.events.sys_exit;
2661                 err = perf_evsel__append_tp_filter(sys_exit, filter);
2662         }
2663 
2664         free(filter);
2665 out:
2666         return err;
2667 out_enomem:
2668         errno = ENOMEM;
2669         goto out;
2670 }
2671 
2672 #ifdef HAVE_LIBBPF_SUPPORT
2673 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace)
2674 {
2675         int fd = bpf_map__fd(trace->syscalls.map);
2676         struct bpf_map_syscall_entry value = {
2677                 .enabled = !trace->not_ev_qualifier,
2678         };
2679         int err = 0;
2680         size_t i;
2681 
2682         for (i = 0; i < trace->ev_qualifier_ids.nr; ++i) {
2683                 int key = trace->ev_qualifier_ids.entries[i];
2684 
2685                 err = bpf_map_update_elem(fd, &key, &value, BPF_EXIST);
2686                 if (err)
2687                         break;
2688         }
2689 
2690         return err;
2691 }
2692 
2693 static int __trace__init_syscalls_bpf_map(struct trace *trace, bool enabled)
2694 {
2695         int fd = bpf_map__fd(trace->syscalls.map);
2696         struct bpf_map_syscall_entry value = {
2697                 .enabled = enabled,
2698         };
2699         int err = 0, key;
2700 
2701         for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
2702                 err = bpf_map_update_elem(fd, &key, &value, BPF_ANY);
2703                 if (err)
2704                         break;
2705         }
2706 
2707         return err;
2708 }
2709 
2710 static int trace__init_syscalls_bpf_map(struct trace *trace)
2711 {
2712         bool enabled = true;
2713 
2714         if (trace->ev_qualifier_ids.nr)
2715                 enabled = trace->not_ev_qualifier;
2716 
2717         return __trace__init_syscalls_bpf_map(trace, enabled);
2718 }
2719 #else
2720 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace __maybe_unused)
2721 {
2722         return 0;
2723 }
2724 
2725 static int trace__init_syscalls_bpf_map(struct trace *trace __maybe_unused)
2726 {
2727         return 0;
2728 }
2729 #endif // HAVE_LIBBPF_SUPPORT
2730 
2731 static int trace__set_ev_qualifier_filter(struct trace *trace)
2732 {
2733         if (trace->syscalls.map)
2734                 return trace__set_ev_qualifier_bpf_filter(trace);
2735         if (trace->syscalls.events.sys_enter)
2736                 return trace__set_ev_qualifier_tp_filter(trace);
2737         return 0;
2738 }
2739 
2740 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
2741                                     size_t npids __maybe_unused, pid_t *pids __maybe_unused)
2742 {
2743         int err = 0;
2744 #ifdef HAVE_LIBBPF_SUPPORT
2745         bool value = true;
2746         int map_fd = bpf_map__fd(map);
2747         size_t i;
2748 
2749         for (i = 0; i < npids; ++i) {
2750                 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
2751                 if (err)
2752                         break;
2753         }
2754 #endif
2755         return err;
2756 }
2757 
2758 static int trace__set_filter_loop_pids(struct trace *trace)
2759 {
2760         unsigned int nr = 1, err;
2761         pid_t pids[32] = {
2762                 getpid(),
2763         };
2764         struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
2765 
2766         while (thread && nr < ARRAY_SIZE(pids)) {
2767                 struct thread *parent = machine__find_thread(trace->host, thread->ppid, thread->ppid);
2768 
2769                 if (parent == NULL)
2770                         break;
2771 
2772                 if (!strcmp(thread__comm_str(parent), "sshd")) {
2773                         pids[nr++] = parent->tid;
2774                         break;
2775                 }
2776                 thread = parent;
2777         }
2778 
2779         err = perf_evlist__set_tp_filter_pids(trace->evlist, nr, pids);
2780         if (!err && trace->filter_pids.map)
2781                 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
2782 
2783         return err;
2784 }
2785 
2786 static int trace__set_filter_pids(struct trace *trace)
2787 {
2788         int err = 0;
2789         /*
2790          * Better not use !target__has_task() here because we need to cover the
2791          * case where no threads were specified in the command line, but a
2792          * workload was, and in that case we will fill in the thread_map when
2793          * we fork the workload in perf_evlist__prepare_workload.
2794          */
2795         if (trace->filter_pids.nr > 0) {
2796                 err = perf_evlist__set_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
2797                                                       trace->filter_pids.entries);
2798                 if (!err && trace->filter_pids.map) {
2799                         err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
2800                                                        trace->filter_pids.entries);
2801                 }
2802         } else if (thread_map__pid(trace->evlist->threads, 0) == -1) {
2803                 err = trace__set_filter_loop_pids(trace);
2804         }
2805 
2806         return err;
2807 }
2808 
2809 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
2810 {
2811         struct perf_evlist *evlist = trace->evlist;
2812         struct perf_sample sample;
2813         int err;
2814 
2815         err = perf_evlist__parse_sample(evlist, event, &sample);
2816         if (err)
2817                 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
2818         else
2819                 trace__handle_event(trace, event, &sample);
2820 
2821         return 0;
2822 }
2823 
2824 static int __trace__flush_events(struct trace *trace)
2825 {
2826         u64 first = ordered_events__first_time(&trace->oe.data);
2827         u64 flush = trace->oe.last - NSEC_PER_SEC;
2828 
2829         /* Is there some thing to flush.. */
2830         if (first && first < flush)
2831                 return ordered_events__flush_time(&trace->oe.data, flush);
2832 
2833         return 0;
2834 }
2835 
2836 static int trace__flush_events(struct trace *trace)
2837 {
2838         return !trace->sort_events ? 0 : __trace__flush_events(trace);
2839 }
2840 
2841 static int trace__deliver_event(struct trace *trace, union perf_event *event)
2842 {
2843         int err;
2844 
2845         if (!trace->sort_events)
2846                 return __trace__deliver_event(trace, event);
2847 
2848         err = perf_evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
2849         if (err && err != -1)
2850                 return err;
2851 
2852         err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0);
2853         if (err)
2854                 return err;
2855 
2856         return trace__flush_events(trace);
2857 }
2858 
2859 static int ordered_events__deliver_event(struct ordered_events *oe,
2860                                          struct ordered_event *event)
2861 {
2862         struct trace *trace = container_of(oe, struct trace, oe.data);
2863 
2864         return __trace__deliver_event(trace, event->event);
2865 }
2866 
2867 static int trace__run(struct trace *trace, int argc, const char **argv)
2868 {
2869         struct perf_evlist *evlist = trace->evlist;
2870         struct perf_evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
2871         int err = -1, i;
2872         unsigned long before;
2873         const bool forks = argc > 0;
2874         bool draining = false;
2875 
2876         trace->live = true;
2877 
2878         if (!trace->raw_augmented_syscalls) {
2879                 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
2880                         goto out_error_raw_syscalls;
2881 
2882                 if (trace->trace_syscalls)
2883                         trace->vfs_getname = perf_evlist__add_vfs_getname(evlist);
2884         }
2885 
2886         if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
2887                 pgfault_maj = perf_evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
2888                 if (pgfault_maj == NULL)
2889                         goto out_error_mem;
2890                 perf_evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
2891                 perf_evlist__add(evlist, pgfault_maj);
2892         }
2893 
2894         if ((trace->trace_pgfaults & TRACE_PFMIN)) {
2895                 pgfault_min = perf_evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
2896                 if (pgfault_min == NULL)
2897                         goto out_error_mem;
2898                 perf_evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
2899                 perf_evlist__add(evlist, pgfault_min);
2900         }
2901 
2902         if (trace->sched &&
2903             perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
2904                                    trace__sched_stat_runtime))
2905                 goto out_error_sched_stat_runtime;
2906 
2907         /*
2908          * If a global cgroup was set, apply it to all the events without an
2909          * explicit cgroup. I.e.:
2910          *
2911          *      trace -G A -e sched:*switch
2912          *
2913          * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
2914          * _and_ sched:sched_switch to the 'A' cgroup, while:
2915          *
2916          * trace -e sched:*switch -G A
2917          *
2918          * will only set the sched:sched_switch event to the 'A' cgroup, all the
2919          * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
2920          * a cgroup (on the root cgroup, sys wide, etc).
2921          *
2922          * Multiple cgroups:
2923          *
2924          * trace -G A -e sched:*switch -G B
2925          *
2926          * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
2927          * to the 'B' cgroup.
2928          *
2929          * evlist__set_default_cgroup() grabs a reference of the passed cgroup
2930          * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
2931          */
2932         if (trace->cgroup)
2933                 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
2934 
2935         err = perf_evlist__create_maps(evlist, &trace->opts.target);
2936         if (err < 0) {
2937                 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
2938                 goto out_delete_evlist;
2939         }
2940 
2941         err = trace__symbols_init(trace, evlist);
2942         if (err < 0) {
2943                 fprintf(trace->output, "Problems initializing symbol libraries!\n");
2944                 goto out_delete_evlist;
2945         }
2946 
2947         perf_evlist__config(evlist, &trace->opts, &callchain_param);
2948 
2949         signal(SIGCHLD, sig_handler);
2950         signal(SIGINT, sig_handler);
2951 
2952         if (forks) {
2953                 err = perf_evlist__prepare_workload(evlist, &trace->opts.target,
2954                                                     argv, false, NULL);
2955                 if (err < 0) {
2956                         fprintf(trace->output, "Couldn't run the workload!\n");
2957                         goto out_delete_evlist;
2958                 }
2959         }
2960 
2961         err = perf_evlist__open(evlist);
2962         if (err < 0)
2963                 goto out_error_open;
2964 
2965         err = bpf__apply_obj_config();
2966         if (err) {
2967                 char errbuf[BUFSIZ];
2968 
2969                 bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
2970                 pr_err("ERROR: Apply config to BPF failed: %s\n",
2971                          errbuf);
2972                 goto out_error_open;
2973         }
2974 
2975         err = trace__set_filter_pids(trace);
2976         if (err < 0)
2977                 goto out_error_mem;
2978 
2979         if (trace->syscalls.map)
2980                 trace__init_syscalls_bpf_map(trace);
2981 
2982         if (trace->ev_qualifier_ids.nr > 0) {
2983                 err = trace__set_ev_qualifier_filter(trace);
2984                 if (err < 0)
2985                         goto out_errno;
2986 
2987                 if (trace->syscalls.events.sys_exit) {
2988                         pr_debug("event qualifier tracepoint filter: %s\n",
2989                                  trace->syscalls.events.sys_exit->filter);
2990                 }
2991         }
2992 
2993         err = perf_evlist__apply_filters(evlist, &evsel);
2994         if (err < 0)
2995                 goto out_error_apply_filters;
2996 
2997         err = perf_evlist__mmap(evlist, trace->opts.mmap_pages);
2998         if (err < 0)
2999                 goto out_error_mmap;
3000 
3001         if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
3002                 perf_evlist__enable(evlist);
3003 
3004         if (forks)
3005                 perf_evlist__start_workload(evlist);
3006 
3007         if (trace->opts.initial_delay) {
3008                 usleep(trace->opts.initial_delay * 1000);
3009                 perf_evlist__enable(evlist);
3010         }
3011 
3012         trace->multiple_threads = thread_map__pid(evlist->threads, 0) == -1 ||
3013                                   evlist->threads->nr > 1 ||
3014                                   perf_evlist__first(evlist)->attr.inherit;
3015 
3016         /*
3017          * Now that we already used evsel->attr to ask the kernel to setup the
3018          * events, lets reuse evsel->attr.sample_max_stack as the limit in
3019          * trace__resolve_callchain(), allowing per-event max-stack settings
3020          * to override an explicitly set --max-stack global setting.
3021          */
3022         evlist__for_each_entry(evlist, evsel) {
3023                 if (evsel__has_callchain(evsel) &&
3024                     evsel->attr.sample_max_stack == 0)
3025                         evsel->attr.sample_max_stack = trace->max_stack;
3026         }
3027 again:
3028         before = trace->nr_events;
3029 
3030         for (i = 0; i < evlist->nr_mmaps; i++) {
3031                 union perf_event *event;
3032                 struct perf_mmap *md;
3033 
3034                 md = &evlist->mmap[i];
3035                 if (perf_mmap__read_init(md) < 0)
3036                         continue;
3037 
3038                 while ((event = perf_mmap__read_event(md)) != NULL) {
3039                         ++trace->nr_events;
3040 
3041                         err = trace__deliver_event(trace, event);
3042                         if (err)
3043                                 goto out_disable;
3044 
3045                         perf_mmap__consume(md);
3046 
3047                         if (interrupted)
3048                                 goto out_disable;
3049 
3050                         if (done && !draining) {
3051                                 perf_evlist__disable(evlist);
3052                                 draining = true;
3053                         }
3054                 }
3055                 perf_mmap__read_done(md);
3056         }
3057 
3058         if (trace->nr_events == before) {
3059                 int timeout = done ? 100 : -1;
3060 
3061                 if (!draining && perf_evlist__poll(evlist, timeout) > 0) {
3062                         if (perf_evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
3063                                 draining = true;
3064 
3065                         goto again;
3066                 } else {
3067                         if (trace__flush_events(trace))
3068                                 goto out_disable;
3069                 }
3070         } else {
3071                 goto again;
3072         }
3073 
3074 out_disable:
3075         thread__zput(trace->current);
3076 
3077         perf_evlist__disable(evlist);
3078 
3079         if (trace->sort_events)
3080                 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
3081 
3082         if (!err) {
3083                 if (trace->summary)
3084                         trace__fprintf_thread_summary(trace, trace->output);
3085 
3086                 if (trace->show_tool_stats) {
3087                         fprintf(trace->output, "Stats:\n "
3088                                                " vfs_getname : %" PRIu64 "\n"
3089                                                " proc_getname: %" PRIu64 "\n",
3090                                 trace->stats.vfs_getname,
3091                                 trace->stats.proc_getname);
3092                 }
3093         }
3094 
3095 out_delete_evlist:
3096         trace__symbols__exit(trace);
3097 
3098         perf_evlist__delete(evlist);
3099         cgroup__put(trace->cgroup);
3100         trace->evlist = NULL;
3101         trace->live = false;
3102         return err;
3103 {
3104         char errbuf[BUFSIZ];
3105 
3106 out_error_sched_stat_runtime:
3107         tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
3108         goto out_error;
3109 
3110 out_error_raw_syscalls:
3111         tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
3112         goto out_error;
3113 
3114 out_error_mmap:
3115         perf_evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
3116         goto out_error;
3117 
3118 out_error_open:
3119         perf_evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
3120 
3121 out_error:
3122         fprintf(trace->output, "%s\n", errbuf);
3123         goto out_delete_evlist;
3124 
3125 out_error_apply_filters:
3126         fprintf(trace->output,
3127                 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
3128                 evsel->filter, perf_evsel__name(evsel), errno,
3129                 str_error_r(errno, errbuf, sizeof(errbuf)));
3130         goto out_delete_evlist;
3131 }
3132 out_error_mem:
3133         fprintf(trace->output, "Not enough memory to run!\n");
3134         goto out_delete_evlist;
3135 
3136 out_errno:
3137         fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
3138         goto out_delete_evlist;
3139 }
3140 
3141 static int trace__replay(struct trace *trace)
3142 {
3143         const struct perf_evsel_str_handler handlers[] = {
3144                 { "probe:vfs_getname",       trace__vfs_getname, },
3145         };
3146         struct perf_data data = {
3147                 .file      = {
3148                         .path = input_name,
3149                 },
3150                 .mode      = PERF_DATA_MODE_READ,
3151                 .force     = trace->force,
3152         };
3153         struct perf_session *session;
3154         struct perf_evsel *evsel;
3155         int err = -1;
3156 
3157         trace->tool.sample        = trace__process_sample;
3158         trace->tool.mmap          = perf_event__process_mmap;
3159         trace->tool.mmap2         = perf_event__process_mmap2;
3160         trace->tool.comm          = perf_event__process_comm;
3161         trace->tool.exit          = perf_event__process_exit;
3162         trace->tool.fork          = perf_event__process_fork;
3163         trace->tool.attr          = perf_event__process_attr;
3164         trace->tool.tracing_data  = perf_event__process_tracing_data;
3165         trace->tool.build_id      = perf_event__process_build_id;
3166         trace->tool.namespaces    = perf_event__process_namespaces;
3167 
3168         trace->tool.ordered_events = true;
3169         trace->tool.ordering_requires_timestamps = true;
3170 
3171         /* add tid to output */
3172         trace->multiple_threads = true;
3173 
3174         session = perf_session__new(&data, false, &trace->tool);
3175         if (session == NULL)
3176                 return -1;
3177 
3178         if (trace->opts.target.pid)
3179                 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
3180 
3181         if (trace->opts.target.tid)
3182                 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
3183 
3184         if (symbol__init(&session->header.env) < 0)
3185                 goto out;
3186 
3187         trace->host = &session->machines.host;
3188 
3189         err = perf_session__set_tracepoints_handlers(session, handlers);
3190         if (err)
3191                 goto out;
3192 
3193         evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3194                                                      "raw_syscalls:sys_enter");
3195         /* older kernels have syscalls tp versus raw_syscalls */
3196         if (evsel == NULL)
3197                 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3198                                                              "syscalls:sys_enter");
3199 
3200         if (evsel &&
3201             (perf_evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
3202             perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
3203                 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
3204                 goto out;
3205         }
3206 
3207         evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3208                                                      "raw_syscalls:sys_exit");
3209         if (evsel == NULL)
3210                 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3211                                                              "syscalls:sys_exit");
3212         if (evsel &&
3213             (perf_evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
3214             perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
3215                 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
3216                 goto out;
3217         }
3218 
3219         evlist__for_each_entry(session->evlist, evsel) {
3220                 if (evsel->attr.type == PERF_TYPE_SOFTWARE &&
3221                     (evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
3222                      evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
3223                      evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS))
3224                         evsel->handler = trace__pgfault;
3225         }
3226 
3227         setup_pager();
3228 
3229         err = perf_session__process_events(session);
3230         if (err)
3231                 pr_err("Failed to process events, error %d", err);
3232 
3233         else if (trace->summary)
3234                 trace__fprintf_thread_summary(trace, trace->output);
3235 
3236 out:
3237         perf_session__delete(session);
3238 
3239         return err;
3240 }
3241 
3242 static size_t trace__fprintf_threads_header(FILE *fp)
3243 {
3244         size_t printed;
3245 
3246         printed  = fprintf(fp, "\n Summary of events:\n\n");
3247 
3248         return printed;
3249 }
3250 
3251 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
3252         struct stats    *stats;
3253         double          msecs;
3254         int             syscall;
3255 )
3256 {
3257         struct int_node *source = rb_entry(nd, struct int_node, rb_node);
3258         struct stats *stats = source->priv;
3259 
3260         entry->syscall = source->i;
3261         entry->stats   = stats;
3262         entry->msecs   = stats ? (u64)stats->n * (avg_stats(stats) / NSEC_PER_MSEC) : 0;
3263 }
3264 
3265 static size_t thread__dump_stats(struct thread_trace *ttrace,
3266                                  struct trace *trace, FILE *fp)
3267 {
3268         size_t printed = 0;
3269         struct syscall *sc;
3270         struct rb_node *nd;
3271         DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
3272 
3273         if (syscall_stats == NULL)
3274                 return 0;
3275 
3276         printed += fprintf(fp, "\n");
3277 
3278         printed += fprintf(fp, "   syscall            calls    total       min       avg       max      stddev\n");
3279         printed += fprintf(fp, "                               (msec)    (msec)    (msec)    (msec)        (%%)\n");
3280         printed += fprintf(fp, "   --------------- -------- --------- --------- --------- ---------     ------\n");
3281 
3282         resort_rb__for_each_entry(nd, syscall_stats) {
3283                 struct stats *stats = syscall_stats_entry->stats;
3284                 if (stats) {
3285                         double min = (double)(stats->min) / NSEC_PER_MSEC;
3286                         double max = (double)(stats->max) / NSEC_PER_MSEC;
3287                         double avg = avg_stats(stats);
3288                         double pct;
3289                         u64 n = (u64) stats->n;
3290 
3291                         pct = avg ? 100.0 * stddev_stats(stats)/avg : 0.0;
3292                         avg /= NSEC_PER_MSEC;
3293 
3294                         sc = &trace->syscalls.table[syscall_stats_entry->syscall];
3295                         printed += fprintf(fp, "   %-15s", sc->name);
3296                         printed += fprintf(fp, " %8" PRIu64 " %9.3f %9.3f %9.3f",
3297                                            n, syscall_stats_entry->msecs, min, avg);
3298                         printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
3299                 }
3300         }
3301 
3302         resort_rb__delete(syscall_stats);
3303         printed += fprintf(fp, "\n\n");
3304 
3305         return printed;
3306 }
3307 
3308 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
3309 {
3310         size_t printed = 0;
3311         struct thread_trace *ttrace = thread__priv(thread);
3312         double ratio;
3313 
3314         if (ttrace == NULL)
3315                 return 0;
3316 
3317         ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
3318 
3319         printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
3320         printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
3321         printed += fprintf(fp, "%.1f%%", ratio);
3322         if (ttrace->pfmaj)
3323                 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
3324         if (ttrace->pfmin)
3325                 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
3326         if (trace->sched)
3327                 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
3328         else if (fputc('\n', fp) != EOF)
3329                 ++printed;
3330 
3331         printed += thread__dump_stats(ttrace, trace, fp);
3332 
3333         return printed;
3334 }
3335 
3336 static unsigned long thread__nr_events(struct thread_trace *ttrace)
3337 {
3338         return ttrace ? ttrace->nr_events : 0;
3339 }
3340 
3341 DEFINE_RESORT_RB(threads, (thread__nr_events(a->thread->priv) < thread__nr_events(b->thread->priv)),
3342         struct thread *thread;
3343 )
3344 {
3345         entry->thread = rb_entry(nd, struct thread, rb_node);
3346 }
3347 
3348 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
3349 {
3350         size_t printed = trace__fprintf_threads_header(fp);
3351         struct rb_node *nd;
3352         int i;
3353 
3354         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
3355                 DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
3356 
3357                 if (threads == NULL) {
3358                         fprintf(fp, "%s", "Error sorting output by nr_events!\n");
3359                         return 0;
3360                 }
3361 
3362                 resort_rb__for_each_entry(nd, threads)
3363                         printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
3364 
3365                 resort_rb__delete(threads);
3366         }
3367         return printed;
3368 }
3369 
3370 static int trace__set_duration(const struct option *opt, const char *str,
3371                                int unset __maybe_unused)
3372 {
3373         struct trace *trace = opt->value;
3374 
3375         trace->duration_filter = atof(str);
3376         return 0;
3377 }
3378 
3379 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
3380                                               int unset __maybe_unused)
3381 {
3382         int ret = -1;
3383         size_t i;
3384         struct trace *trace = opt->value;
3385         /*
3386          * FIXME: introduce a intarray class, plain parse csv and create a
3387          * { int nr, int entries[] } struct...
3388          */
3389         struct intlist *list = intlist__new(str);
3390 
3391         if (list == NULL)
3392                 return -1;
3393 
3394         i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
3395         trace->filter_pids.entries = calloc(i, sizeof(pid_t));
3396 
3397         if (trace->filter_pids.entries == NULL)
3398                 goto out;
3399 
3400         trace->filter_pids.entries[0] = getpid();
3401 
3402         for (i = 1; i < trace->filter_pids.nr; ++i)
3403                 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
3404 
3405         intlist__delete(list);
3406         ret = 0;
3407 out:
3408         return ret;
3409 }
3410 
3411 static int trace__open_output(struct trace *trace, const char *filename)
3412 {
3413         struct stat st;
3414 
3415         if (!stat(filename, &st) && st.st_size) {
3416                 char oldname[PATH_MAX];
3417 
3418                 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
3419                 unlink(oldname);
3420                 rename(filename, oldname);
3421         }
3422 
3423         trace->output = fopen(filename, "w");
3424 
3425         return trace->output == NULL ? -errno : 0;
3426 }
3427 
3428 static int parse_pagefaults(const struct option *opt, const char *str,
3429                             int unset __maybe_unused)
3430 {
3431         int *trace_pgfaults = opt->value;
3432 
3433         if (strcmp(str, "all") == 0)
3434                 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
3435         else if (strcmp(str, "maj") == 0)
3436                 *trace_pgfaults |= TRACE_PFMAJ;
3437         else if (strcmp(str, "min") == 0)
3438                 *trace_pgfaults |= TRACE_PFMIN;
3439         else
3440                 return -1;
3441 
3442         return 0;
3443 }
3444 
3445 static void evlist__set_evsel_handler(struct perf_evlist *evlist, void *handler)
3446 {
3447         struct perf_evsel *evsel;
3448 
3449         evlist__for_each_entry(evlist, evsel)
3450                 evsel->handler = handler;
3451 }
3452 
3453 static int evlist__set_syscall_tp_fields(struct perf_evlist *evlist)
3454 {
3455         struct perf_evsel *evsel;
3456 
3457         evlist__for_each_entry(evlist, evsel) {
3458                 if (evsel->priv || !evsel->tp_format)
3459                         continue;
3460 
3461                 if (strcmp(evsel->tp_format->system, "syscalls"))
3462                         continue;
3463 
3464                 if (perf_evsel__init_syscall_tp(evsel))
3465                         return -1;
3466 
3467                 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
3468                         struct syscall_tp *sc = evsel->priv;
3469 
3470                         if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
3471                                 return -1;
3472                 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
3473                         struct syscall_tp *sc = evsel->priv;
3474 
3475                         if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
3476                                 return -1;
3477                 }
3478         }
3479 
3480         return 0;
3481 }
3482 
3483 /*
3484  * XXX: Hackish, just splitting the combined -e+--event (syscalls
3485  * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
3486  * existing facilities unchanged (trace->ev_qualifier + parse_options()).
3487  *
3488  * It'd be better to introduce a parse_options() variant that would return a
3489  * list with the terms it didn't match to an event...
3490  */
3491 static int trace__parse_events_option(const struct option *opt, const char *str,
3492                                       int unset __maybe_unused)
3493 {
3494         struct trace *trace = (struct trace *)opt->value;
3495         const char *s = str;
3496         char *sep = NULL, *lists[2] = { NULL, NULL, };
3497         int len = strlen(str) + 1, err = -1, list, idx;
3498         char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
3499         char group_name[PATH_MAX];
3500         struct syscall_fmt *fmt;
3501 
3502         if (strace_groups_dir == NULL)
3503                 return -1;
3504 
3505         if (*s == '!') {
3506                 ++s;
3507                 trace->not_ev_qualifier = true;
3508         }
3509 
3510         while (1) {
3511                 if ((sep = strchr(s, ',')) != NULL)
3512                         *sep = '\0';
3513 
3514                 list = 0;
3515                 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
3516                     syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
3517                         list = 1;
3518                         goto do_concat;
3519                 }
3520 
3521                 fmt = syscall_fmt__find_by_alias(s);
3522                 if (fmt != NULL) {
3523                         list = 1;
3524                         s = fmt->name;
3525                 } else {
3526                         path__join(group_name, sizeof(group_name), strace_groups_dir, s);
3527                         if (access(group_name, R_OK) == 0)
3528                                 list = 1;
3529                 }
3530 do_concat:
3531                 if (lists[list]) {
3532                         sprintf(lists[list] + strlen(lists[list]), ",%s", s);
3533                 } else {
3534                         lists[list] = malloc(len);
3535                         if (lists[list] == NULL)
3536                                 goto out;
3537                         strcpy(lists[list], s);
3538                 }
3539 
3540                 if (!sep)
3541                         break;
3542 
3543                 *sep = ',';
3544                 s = sep + 1;
3545         }
3546 
3547         if (lists[1] != NULL) {
3548                 struct strlist_config slist_config = {
3549                         .dirname = strace_groups_dir,
3550                 };
3551 
3552                 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
3553                 if (trace->ev_qualifier == NULL) {
3554                         fputs("Not enough memory to parse event qualifier", trace->output);
3555                         goto out;
3556                 }
3557 
3558                 if (trace__validate_ev_qualifier(trace))
3559                         goto out;
3560                 trace->trace_syscalls = true;
3561         }
3562 
3563         err = 0;
3564 
3565         if (lists[0]) {
3566                 struct option o = OPT_CALLBACK('e', "event", &trace->evlist, "event",
3567                                                "event selector. use 'perf list' to list available events",
3568                                                parse_events_option);
3569                 err = parse_events_option(&o, lists[0], 0);
3570         }
3571 out:
3572         if (sep)
3573                 *sep = ',';
3574 
3575         return err;
3576 }
3577 
3578 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
3579 {
3580         struct trace *trace = opt->value;
3581 
3582         if (!list_empty(&trace->evlist->entries))
3583                 return parse_cgroups(opt, str, unset);
3584 
3585         trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
3586 
3587         return 0;
3588 }
3589 
3590 static struct bpf_map *bpf__find_map_by_name(const char *name)
3591 {
3592         struct bpf_object *obj, *tmp;
3593 
3594         bpf_object__for_each_safe(obj, tmp) {
3595                 struct bpf_map *map = bpf_object__find_map_by_name(obj, name);
3596                 if (map)
3597                         return map;
3598 
3599         }
3600 
3601         return NULL;
3602 }
3603 
3604 static void trace__set_bpf_map_filtered_pids(struct trace *trace)
3605 {
3606         trace->filter_pids.map = bpf__find_map_by_name("pids_filtered");
3607 }
3608 
3609 static void trace__set_bpf_map_syscalls(struct trace *trace)
3610 {
3611         trace->syscalls.map = bpf__find_map_by_name("syscalls");
3612 }
3613 
3614 static int trace__config(const char *var, const char *value, void *arg)
3615 {
3616         struct trace *trace = arg;
3617         int err = 0;
3618 
3619         if (!strcmp(var, "trace.add_events")) {
3620                 struct option o = OPT_CALLBACK('e', "event", &trace->evlist, "event",
3621                                                "event selector. use 'perf list' to list available events",
3622                                                parse_events_option);
3623                 err = parse_events_option(&o, value, 0);
3624         } else if (!strcmp(var, "trace.show_timestamp")) {
3625                 trace->show_tstamp = perf_config_bool(var, value);
3626         } else if (!strcmp(var, "trace.show_duration")) {
3627                 trace->show_duration = perf_config_bool(var, value);
3628         } else if (!strcmp(var, "trace.show_arg_names")) {
3629                 trace->show_arg_names = perf_config_bool(var, value);
3630                 if (!trace->show_arg_names)
3631                         trace->show_zeros = true;
3632         } else if (!strcmp(var, "trace.show_zeros")) {
3633                 bool new_show_zeros = perf_config_bool(var, value);
3634                 if (!trace->show_arg_names && !new_show_zeros) {
3635                         pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
3636                         goto out;
3637                 }
3638                 trace->show_zeros = new_show_zeros;
3639         } else if (!strcmp(var, "trace.show_prefix")) {
3640                 trace->show_string_prefix = perf_config_bool(var, value);
3641         } else if (!strcmp(var, "trace.no_inherit")) {
3642                 trace->opts.no_inherit = perf_config_bool(var, value);
3643         } else if (!strcmp(var, "trace.args_alignment")) {
3644                 int args_alignment = 0;
3645                 if (perf_config_int(&args_alignment, var, value) == 0)
3646                         trace->args_alignment = args_alignment;
3647         }
3648 out:
3649         return err;
3650 }
3651 
3652 int cmd_trace(int argc, const char **argv)
3653 {
3654         const char *trace_usage[] = {
3655                 "perf trace [<options>] [<command>]",
3656                 "perf trace [<options>] -- <command> [<options>]",
3657                 "perf trace record [<options>] [<command>]",
3658                 "perf trace record [<options>] -- <command> [<options>]",
3659                 NULL
3660         };
3661         struct trace trace = {
3662                 .syscalls = {
3663                         . max = -1,
3664                 },
3665                 .opts = {
3666                         .target = {
3667                                 .uid       = UINT_MAX,
3668                                 .uses_mmap = true,
3669                         },
3670                         .user_freq     = UINT_MAX,
3671                         .user_interval = ULLONG_MAX,
3672                         .no_buffering  = true,
3673                         .mmap_pages    = UINT_MAX,
3674                 },
3675                 .output = stderr,
3676                 .show_comm = true,
3677                 .show_tstamp = true,
3678                 .show_duration = true,
3679                 .show_arg_names = true,
3680                 .args_alignment = 70,
3681                 .trace_syscalls = false,
3682                 .kernel_syscallchains = false,
3683                 .max_stack = UINT_MAX,
3684                 .max_events = ULONG_MAX,
3685         };
3686         const char *output_name = NULL;
3687         const struct option trace_options[] = {
3688         OPT_CALLBACK('e', "event", &trace, "event",
3689                      "event/syscall selector. use 'perf list' to list available events",
3690                      trace__parse_events_option),
3691         OPT_BOOLEAN(0, "comm", &trace.show_comm,
3692                     "show the thread COMM next to its id"),
3693         OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
3694         OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
3695                      trace__parse_events_option),
3696         OPT_STRING('o', "output", &output_name, "file", "output file name"),
3697         OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
3698         OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
3699                     "trace events on existing process id"),
3700         OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
3701                     "trace events on existing thread id"),
3702         OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
3703                      "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
3704         OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
3705                     "system-wide collection from all CPUs"),
3706         OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
3707                     "list of cpus to monitor"),
3708         OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
3709                     "child tasks do not inherit counters"),
3710         OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
3711                      "number of mmap data pages",
3712                      perf_evlist__parse_mmap_pages),
3713         OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
3714                    "user to profile"),
3715         OPT_CALLBACK(0, "duration", &trace, "float",
3716                      "show only events with duration > N.M ms",
3717                      trace__set_duration),
3718         OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
3719         OPT_INCR('v', "verbose", &verbose, "be more verbose"),
3720         OPT_BOOLEAN('T', "time", &trace.full_time,
3721                     "Show full timestamp, not time relative to first start"),
3722         OPT_BOOLEAN(0, "failure", &trace.failure_only,
3723                     "Show only syscalls that failed"),
3724         OPT_BOOLEAN('s', "summary", &trace.summary_only,
3725                     "Show only syscall summary with statistics"),
3726         OPT_BOOLEAN('S', "with-summary", &trace.summary,
3727                     "Show all syscalls and summary with statistics"),
3728         OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
3729                      "Trace pagefaults", parse_pagefaults, "maj"),
3730         OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
3731         OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
3732         OPT_CALLBACK(0, "call-graph", &trace.opts,
3733                      "record_mode[,record_size]", record_callchain_help,
3734                      &record_parse_callchain_opt),
3735         OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
3736                     "Show the kernel callchains on the syscall exit path"),
3737         OPT_ULONG(0, "max-events", &trace.max_events,
3738                 "Set the maximum number of events to print, exit after that is reached. "),
3739         OPT_UINTEGER(0, "min-stack", &trace.min_stack,
3740                      "Set the minimum stack depth when parsing the callchain, "
3741                      "anything below the specified depth will be ignored."),
3742         OPT_UINTEGER(0, "max-stack", &trace.max_stack,
3743                      "Set the maximum stack depth when parsing the callchain, "
3744                      "anything beyond the specified depth will be ignored. "
3745                      "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
3746         OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
3747                         "Sort batch of events before processing, use if getting out of order events"),
3748         OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
3749                         "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
3750         OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
3751                         "per thread proc mmap processing timeout in ms"),
3752         OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
3753                      trace__parse_cgroups),
3754         OPT_UINTEGER('D', "delay", &trace.opts.initial_delay,
3755                      "ms to wait before starting measurement after program "
3756                      "start"),
3757         OPT_END()
3758         };
3759         bool __maybe_unused max_stack_user_set = true;
3760         bool mmap_pages_user_set = true;
3761         struct perf_evsel *evsel;
3762         const char * const trace_subcommands[] = { "record", NULL };
3763         int err = -1;
3764         char bf[BUFSIZ];
3765 
3766         signal(SIGSEGV, sighandler_dump_stack);
3767         signal(SIGFPE, sighandler_dump_stack);
3768 
3769         trace.evlist = perf_evlist__new();
3770         trace.sctbl = syscalltbl__new();
3771 
3772         if (trace.evlist == NULL || trace.sctbl == NULL) {
3773                 pr_err("Not enough memory to run!\n");
3774                 err = -ENOMEM;
3775                 goto out;
3776         }
3777 
3778         err = perf_config(trace__config, &trace);
3779         if (err)
3780                 goto out;
3781 
3782         argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
3783                                  trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
3784 
3785         if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
3786                 usage_with_options_msg(trace_usage, trace_options,
3787                                        "cgroup monitoring only available in system-wide mode");
3788         }
3789 
3790         evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
3791         if (IS_ERR(evsel)) {
3792                 bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
3793                 pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
3794                 goto out;
3795         }
3796 
3797         if (evsel) {
3798                 trace.syscalls.events.augmented = evsel;
3799                 trace__set_bpf_map_filtered_pids(&trace);
3800                 trace__set_bpf_map_syscalls(&trace);
3801         }
3802 
3803         err = bpf__setup_stdout(trace.evlist);
3804         if (err) {
3805                 bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
3806                 pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
3807                 goto out;
3808         }
3809 
3810         err = -1;
3811 
3812         if (trace.trace_pgfaults) {
3813                 trace.opts.sample_address = true;
3814                 trace.opts.sample_time = true;
3815         }
3816 
3817         if (trace.opts.mmap_pages == UINT_MAX)
3818                 mmap_pages_user_set = false;
3819 
3820         if (trace.max_stack == UINT_MAX) {
3821                 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
3822                 max_stack_user_set = false;
3823         }
3824 
3825 #ifdef HAVE_DWARF_UNWIND_SUPPORT
3826         if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
3827                 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
3828         }
3829 #endif
3830 
3831         if (callchain_param.enabled) {
3832                 if (!mmap_pages_user_set && geteuid() == 0)
3833                         trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
3834 
3835                 symbol_conf.use_callchain = true;
3836         }
3837 
3838         if (trace.evlist->nr_entries > 0) {
3839                 evlist__set_evsel_handler(trace.evlist, trace__event_handler);
3840                 if (evlist__set_syscall_tp_fields(trace.evlist)) {
3841                         perror("failed to set syscalls:* tracepoint fields");
3842                         goto out;
3843                 }
3844         }
3845 
3846         if (trace.sort_events) {
3847                 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
3848                 ordered_events__set_copy_on_queue(&trace.oe.data, true);
3849         }
3850 
3851         /*
3852          * If we are augmenting syscalls, then combine what we put in the
3853          * __augmented_syscalls__ BPF map with what is in the
3854          * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
3855          * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
3856          *
3857          * We'll switch to look at two BPF maps, one for sys_enter and the
3858          * other for sys_exit when we start augmenting the sys_exit paths with
3859          * buffers that are being copied from kernel to userspace, think 'read'
3860          * syscall.
3861          */
3862         if (trace.syscalls.events.augmented) {
3863                 evlist__for_each_entry(trace.evlist, evsel) {
3864                         bool raw_syscalls_sys_exit = strcmp(perf_evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
3865 
3866                         if (raw_syscalls_sys_exit) {
3867                                 trace.raw_augmented_syscalls = true;
3868                                 goto init_augmented_syscall_tp;
3869                         }
3870 
3871                         if (trace.syscalls.events.augmented->priv == NULL &&
3872                             strstr(perf_evsel__name(evsel), "syscalls:sys_enter")) {
3873                                 struct perf_evsel *augmented = trace.syscalls.events.augmented;
3874                                 if (perf_evsel__init_augmented_syscall_tp(augmented, evsel) ||
3875                                     perf_evsel__init_augmented_syscall_tp_args(augmented))
3876                                         goto out;
3877                                 augmented->handler = trace__sys_enter;
3878                         }
3879 
3880                         if (strstarts(perf_evsel__name(evsel), "syscalls:sys_exit_")) {
3881                                 struct syscall_tp *sc;
3882 init_augmented_syscall_tp:
3883                                 if (perf_evsel__init_augmented_syscall_tp(evsel, evsel))
3884                                         goto out;
3885                                 sc = evsel->priv;
3886                                 /*
3887                                  * For now with BPF raw_augmented we hook into
3888                                  * raw_syscalls:sys_enter and there we get all
3889                                  * 6 syscall args plus the tracepoint common
3890                                  * fields and the syscall_nr (another long).
3891                                  * So we check if that is the case and if so
3892                                  * don't look after the sc->args_size but
3893                                  * always after the full raw_syscalls:sys_enter
3894                                  * payload, which is fixed.
3895                                  *
3896                                  * We'll revisit this later to pass
3897                                  * s->args_size to the BPF augmenter (now
3898                                  * tools/perf/examples/bpf/augmented_raw_syscalls.c,
3899                                  * so that it copies only what we need for each
3900                                  * syscall, like what happens when we use
3901                                  * syscalls:sys_enter_NAME, so that we reduce
3902                                  * the kernel/userspace traffic to just what is
3903                                  * needed for each syscall.
3904                                  */
3905                                 if (trace.raw_augmented_syscalls)
3906                                         trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
3907                                 perf_evsel__init_augmented_syscall_tp_ret(evsel);
3908                                 evsel->handler = trace__sys_exit;
3909                         }
3910                 }
3911         }
3912 
3913         if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
3914                 return trace__record(&trace, argc-1, &argv[1]);
3915 
3916         /* summary_only implies summary option, but don't overwrite summary if set */
3917         if (trace.summary_only)
3918                 trace.summary = trace.summary_only;
3919 
3920         if (!trace.trace_syscalls && !trace.trace_pgfaults &&
3921             trace.evlist->nr_entries == 0 /* Was --events used? */) {
3922                 trace.trace_syscalls = true;
3923         }
3924 
3925         if (output_name != NULL) {
3926                 err = trace__open_output(&trace, output_name);
3927                 if (err < 0) {
3928                         perror("failed to create output file");
3929                         goto out;
3930                 }
3931         }
3932 
3933         err = target__validate(&trace.opts.target);
3934         if (err) {
3935                 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
3936                 fprintf(trace.output, "%s", bf);
3937                 goto out_close;
3938         }
3939 
3940         err = target__parse_uid(&trace.opts.target);
3941         if (err) {
3942                 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
3943                 fprintf(trace.output, "%s", bf);
3944                 goto out_close;
3945         }
3946 
3947         if (!argc && target__none(&trace.opts.target))
3948                 trace.opts.target.system_wide = true;
3949 
3950         if (input_name)
3951                 err = trace__replay(&trace);
3952         else
3953                 err = trace__run(&trace, argc, argv);
3954 
3955 out_close:
3956         if (output_name != NULL)
3957                 fclose(trace.output);
3958 out:
3959         return err;
3960 }
3961 

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

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp