1 /*
2 * builtin-trace.c
3 *
4 * Builtin 'trace' command:
5 *
6 * Display a continuously updated trace of any workload, CPU, specific PID,
7 * system wide, etc. Default format is loosely strace like, but any other
8 * event may be specified using --event.
9 *
10 * Copyright (C) 2012, 2013, 2014, 2015 Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
11 *
12 * Initially based on the 'trace' prototype by Thomas Gleixner:
13 *
14 * http://lwn.net/Articles/415728/ ("Announcing a new utility: 'trace'")
15 */
16
17 #include "util/record.h"
18 #include <traceevent/event-parse.h>
19 #include <api/fs/tracing_path.h>
20 #include <bpf/bpf.h>
21 #include "util/bpf_map.h"
22 #include "util/rlimit.h"
23 #include "builtin.h"
24 #include "util/cgroup.h"
25 #include "util/color.h"
26 #include "util/config.h"
27 #include "util/debug.h"
28 #include "util/dso.h"
29 #include "util/env.h"
30 #include "util/event.h"
31 #include "util/evsel.h"
32 #include "util/evsel_fprintf.h"
33 #include "util/synthetic-events.h"
34 #include "util/evlist.h"
35 #include "util/evswitch.h"
36 #include "util/mmap.h"
37 #include <subcmd/pager.h>
38 #include <subcmd/exec-cmd.h>
39 #include "util/machine.h"
40 #include "util/map.h"
41 #include "util/symbol.h"
42 #include "util/path.h"
43 #include "util/session.h"
44 #include "util/thread.h"
45 #include <subcmd/parse-options.h>
46 #include "util/strlist.h"
47 #include "util/intlist.h"
48 #include "util/thread_map.h"
49 #include "util/stat.h"
50 #include "util/tool.h"
51 #include "util/util.h"
52 #include "trace/beauty/beauty.h"
53 #include "trace-event.h"
54 #include "util/parse-events.h"
55 #include "util/bpf-loader.h"
56 #include "callchain.h"
57 #include "print_binary.h"
58 #include "string2.h"
59 #include "syscalltbl.h"
60 #include "rb_resort.h"
61 #include "../perf.h"
62
63 #include <errno.h>
64 #include <inttypes.h>
65 #include <poll.h>
66 #include <signal.h>
67 #include <stdlib.h>
68 #include <string.h>
69 #include <linux/err.h>
70 #include <linux/filter.h>
71 #include <linux/kernel.h>
72 #include <linux/random.h>
73 #include <linux/stringify.h>
74 #include <linux/time64.h>
75 #include <linux/zalloc.h>
76 #include <fcntl.h>
77 #include <sys/sysmacros.h>
78
79 #include <linux/ctype.h>
80
81 #ifndef O_CLOEXEC
82 # define O_CLOEXEC 02000000
83 #endif
84
85 #ifndef F_LINUX_SPECIFIC_BASE
86 # define F_LINUX_SPECIFIC_BASE 1024
87 #endif
88
89 #define RAW_SYSCALL_ARGS_NUM 6
90
91 /*
92 * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100
93 */
94 struct syscall_arg_fmt {
95 size_t (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
96 bool (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val);
97 unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
98 void *parm;
99 const char *name;
100 bool show_zero;
101 };
102
103 struct syscall_fmt {
104 const char *name;
105 const char *alias;
106 struct {
107 const char *sys_enter,
108 *sys_exit;
109 } bpf_prog_name;
110 struct syscall_arg_fmt arg[RAW_SYSCALL_ARGS_NUM];
111 u8 nr_args;
112 bool errpid;
113 bool timeout;
114 bool hexret;
115 };
116
117 struct trace {
118 struct perf_tool tool;
119 struct syscalltbl *sctbl;
120 struct {
121 struct syscall *table;
122 struct bpf_map *map;
123 struct { // per syscall BPF_MAP_TYPE_PROG_ARRAY
124 struct bpf_map *sys_enter,
125 *sys_exit;
126 } prog_array;
127 struct {
128 struct evsel *sys_enter,
129 *sys_exit,
130 *augmented;
131 } events;
132 struct bpf_program *unaugmented_prog;
133 } syscalls;
134 struct {
135 struct bpf_map *map;
136 } dump;
137 struct record_opts opts;
138 struct evlist *evlist;
139 struct machine *host;
140 struct thread *current;
141 struct bpf_object *bpf_obj;
142 struct cgroup *cgroup;
143 u64 base_time;
144 FILE *output;
145 unsigned long nr_events;
146 unsigned long nr_events_printed;
147 unsigned long max_events;
148 struct evswitch evswitch;
149 struct strlist *ev_qualifier;
150 struct {
151 size_t nr;
152 int *entries;
153 } ev_qualifier_ids;
154 struct {
155 size_t nr;
156 pid_t *entries;
157 struct bpf_map *map;
158 } filter_pids;
159 double duration_filter;
160 double runtime_ms;
161 struct {
162 u64 vfs_getname,
163 proc_getname;
164 } stats;
165 unsigned int max_stack;
166 unsigned int min_stack;
167 int raw_augmented_syscalls_args_size;
168 bool raw_augmented_syscalls;
169 bool fd_path_disabled;
170 bool sort_events;
171 bool not_ev_qualifier;
172 bool live;
173 bool full_time;
174 bool sched;
175 bool multiple_threads;
176 bool summary;
177 bool summary_only;
178 bool failure_only;
179 bool show_comm;
180 bool print_sample;
181 bool show_tool_stats;
182 bool trace_syscalls;
183 bool kernel_syscallchains;
184 s16 args_alignment;
185 bool show_tstamp;
186 bool show_duration;
187 bool show_zeros;
188 bool show_arg_names;
189 bool show_string_prefix;
190 bool force;
191 bool vfs_getname;
192 int trace_pgfaults;
193 struct {
194 struct ordered_events data;
195 u64 last;
196 } oe;
197 };
198
199 struct tp_field {
200 int offset;
201 union {
202 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
203 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
204 };
205 };
206
207 #define TP_UINT_FIELD(bits) \
208 static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
209 { \
210 u##bits value; \
211 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
212 return value; \
213 }
214
215 TP_UINT_FIELD(8);
216 TP_UINT_FIELD(16);
217 TP_UINT_FIELD(32);
218 TP_UINT_FIELD(64);
219
220 #define TP_UINT_FIELD__SWAPPED(bits) \
221 static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
222 { \
223 u##bits value; \
224 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
225 return bswap_##bits(value);\
226 }
227
228 TP_UINT_FIELD__SWAPPED(16);
229 TP_UINT_FIELD__SWAPPED(32);
230 TP_UINT_FIELD__SWAPPED(64);
231
__tp_field__init_uint(struct tp_field * field,int size,int offset,bool needs_swap)232 static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
233 {
234 field->offset = offset;
235
236 switch (size) {
237 case 1:
238 field->integer = tp_field__u8;
239 break;
240 case 2:
241 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
242 break;
243 case 4:
244 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
245 break;
246 case 8:
247 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
248 break;
249 default:
250 return -1;
251 }
252
253 return 0;
254 }
255
tp_field__init_uint(struct tp_field * field,struct tep_format_field * format_field,bool needs_swap)256 static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
257 {
258 return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
259 }
260
tp_field__ptr(struct tp_field * field,struct perf_sample * sample)261 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
262 {
263 return sample->raw_data + field->offset;
264 }
265
__tp_field__init_ptr(struct tp_field * field,int offset)266 static int __tp_field__init_ptr(struct tp_field *field, int offset)
267 {
268 field->offset = offset;
269 field->pointer = tp_field__ptr;
270 return 0;
271 }
272
tp_field__init_ptr(struct tp_field * field,struct tep_format_field * format_field)273 static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
274 {
275 return __tp_field__init_ptr(field, format_field->offset);
276 }
277
278 struct syscall_tp {
279 struct tp_field id;
280 union {
281 struct tp_field args, ret;
282 };
283 };
284
perf_evsel__init_tp_uint_field(struct evsel * evsel,struct tp_field * field,const char * name)285 static int perf_evsel__init_tp_uint_field(struct evsel *evsel,
286 struct tp_field *field,
287 const char *name)
288 {
289 struct tep_format_field *format_field = perf_evsel__field(evsel, name);
290
291 if (format_field == NULL)
292 return -1;
293
294 return tp_field__init_uint(field, format_field, evsel->needs_swap);
295 }
296
297 #define perf_evsel__init_sc_tp_uint_field(evsel, name) \
298 ({ struct syscall_tp *sc = evsel->priv;\
299 perf_evsel__init_tp_uint_field(evsel, &sc->name, #name); })
300
perf_evsel__init_tp_ptr_field(struct evsel * evsel,struct tp_field * field,const char * name)301 static int perf_evsel__init_tp_ptr_field(struct evsel *evsel,
302 struct tp_field *field,
303 const char *name)
304 {
305 struct tep_format_field *format_field = perf_evsel__field(evsel, name);
306
307 if (format_field == NULL)
308 return -1;
309
310 return tp_field__init_ptr(field, format_field);
311 }
312
313 #define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
314 ({ struct syscall_tp *sc = evsel->priv;\
315 perf_evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
316
evsel__delete_priv(struct evsel * evsel)317 static void evsel__delete_priv(struct evsel *evsel)
318 {
319 zfree(&evsel->priv);
320 evsel__delete(evsel);
321 }
322
perf_evsel__init_syscall_tp(struct evsel * evsel)323 static int perf_evsel__init_syscall_tp(struct evsel *evsel)
324 {
325 struct syscall_tp *sc = evsel->priv = malloc(sizeof(struct syscall_tp));
326
327 if (evsel->priv != NULL) {
328 if (perf_evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
329 perf_evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
330 goto out_delete;
331 return 0;
332 }
333
334 return -ENOMEM;
335 out_delete:
336 zfree(&evsel->priv);
337 return -ENOENT;
338 }
339
perf_evsel__init_augmented_syscall_tp(struct evsel * evsel,struct evsel * tp)340 static int perf_evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)
341 {
342 struct syscall_tp *sc = evsel->priv = malloc(sizeof(struct syscall_tp));
343
344 if (evsel->priv != NULL) {
345 struct tep_format_field *syscall_id = perf_evsel__field(tp, "id");
346 if (syscall_id == NULL)
347 syscall_id = perf_evsel__field(tp, "__syscall_nr");
348 if (syscall_id == NULL)
349 goto out_delete;
350 if (__tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
351 goto out_delete;
352
353 return 0;
354 }
355
356 return -ENOMEM;
357 out_delete:
358 zfree(&evsel->priv);
359 return -EINVAL;
360 }
361
perf_evsel__init_augmented_syscall_tp_args(struct evsel * evsel)362 static int perf_evsel__init_augmented_syscall_tp_args(struct evsel *evsel)
363 {
364 struct syscall_tp *sc = evsel->priv;
365
366 return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
367 }
368
perf_evsel__init_augmented_syscall_tp_ret(struct evsel * evsel)369 static int perf_evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)
370 {
371 struct syscall_tp *sc = evsel->priv;
372
373 return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
374 }
375
perf_evsel__init_raw_syscall_tp(struct evsel * evsel,void * handler)376 static int perf_evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)
377 {
378 evsel->priv = malloc(sizeof(struct syscall_tp));
379 if (evsel->priv != NULL) {
380 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
381 goto out_delete;
382
383 evsel->handler = handler;
384 return 0;
385 }
386
387 return -ENOMEM;
388
389 out_delete:
390 zfree(&evsel->priv);
391 return -ENOENT;
392 }
393
perf_evsel__raw_syscall_newtp(const char * direction,void * handler)394 static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
395 {
396 struct evsel *evsel = perf_evsel__newtp("raw_syscalls", direction);
397
398 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
399 if (IS_ERR(evsel))
400 evsel = perf_evsel__newtp("syscalls", direction);
401
402 if (IS_ERR(evsel))
403 return NULL;
404
405 if (perf_evsel__init_raw_syscall_tp(evsel, handler))
406 goto out_delete;
407
408 return evsel;
409
410 out_delete:
411 evsel__delete_priv(evsel);
412 return NULL;
413 }
414
415 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
416 ({ struct syscall_tp *fields = evsel->priv; \
417 fields->name.integer(&fields->name, sample); })
418
419 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
420 ({ struct syscall_tp *fields = evsel->priv; \
421 fields->name.pointer(&fields->name, sample); })
422
strarray__scnprintf(struct strarray * sa,char * bf,size_t size,const char * intfmt,bool show_prefix,int val)423 size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
424 {
425 int idx = val - sa->offset;
426
427 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
428 size_t printed = scnprintf(bf, size, intfmt, val);
429 if (show_prefix)
430 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
431 return printed;
432 }
433
434 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
435 }
436
__syscall_arg__scnprintf_strarray(char * bf,size_t size,const char * intfmt,struct syscall_arg * arg)437 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
438 const char *intfmt,
439 struct syscall_arg *arg)
440 {
441 return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
442 }
443
syscall_arg__scnprintf_strarray(char * bf,size_t size,struct syscall_arg * arg)444 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
445 struct syscall_arg *arg)
446 {
447 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
448 }
449
450 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
451
syscall_arg__scnprintf_strarray_flags(char * bf,size_t size,struct syscall_arg * arg)452 size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)
453 {
454 return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);
455 }
456
strarrays__scnprintf(struct strarrays * sas,char * bf,size_t size,const char * intfmt,bool show_prefix,int val)457 size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
458 {
459 size_t printed;
460 int i;
461
462 for (i = 0; i < sas->nr_entries; ++i) {
463 struct strarray *sa = sas->entries[i];
464 int idx = val - sa->offset;
465
466 if (idx >= 0 && idx < sa->nr_entries) {
467 if (sa->entries[idx] == NULL)
468 break;
469 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
470 }
471 }
472
473 printed = scnprintf(bf, size, intfmt, val);
474 if (show_prefix)
475 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
476 return printed;
477 }
478
syscall_arg__scnprintf_strarrays(char * bf,size_t size,struct syscall_arg * arg)479 size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
480 struct syscall_arg *arg)
481 {
482 return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
483 }
484
485 #ifndef AT_FDCWD
486 #define AT_FDCWD -100
487 #endif
488
syscall_arg__scnprintf_fd_at(char * bf,size_t size,struct syscall_arg * arg)489 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
490 struct syscall_arg *arg)
491 {
492 int fd = arg->val;
493 const char *prefix = "AT_FD";
494
495 if (fd == AT_FDCWD)
496 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
497
498 return syscall_arg__scnprintf_fd(bf, size, arg);
499 }
500
501 #define SCA_FDAT syscall_arg__scnprintf_fd_at
502
503 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
504 struct syscall_arg *arg);
505
506 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
507
syscall_arg__scnprintf_hex(char * bf,size_t size,struct syscall_arg * arg)508 size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
509 {
510 return scnprintf(bf, size, "%#lx", arg->val);
511 }
512
syscall_arg__scnprintf_ptr(char * bf,size_t size,struct syscall_arg * arg)513 size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
514 {
515 if (arg->val == 0)
516 return scnprintf(bf, size, "NULL");
517 return syscall_arg__scnprintf_hex(bf, size, arg);
518 }
519
syscall_arg__scnprintf_int(char * bf,size_t size,struct syscall_arg * arg)520 size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
521 {
522 return scnprintf(bf, size, "%d", arg->val);
523 }
524
syscall_arg__scnprintf_long(char * bf,size_t size,struct syscall_arg * arg)525 size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
526 {
527 return scnprintf(bf, size, "%ld", arg->val);
528 }
529
530 static const char *bpf_cmd[] = {
531 "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
532 "MAP_GET_NEXT_KEY", "PROG_LOAD",
533 };
534 static DEFINE_STRARRAY(bpf_cmd, "BPF_");
535
536 static const char *fsmount_flags[] = {
537 [1] = "CLOEXEC",
538 };
539 static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");
540
541 #include "trace/beauty/generated/fsconfig_arrays.c"
542
543 static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");
544
545 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
546 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
547
548 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
549 static DEFINE_STRARRAY(itimers, "ITIMER_");
550
551 static const char *keyctl_options[] = {
552 "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
553 "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
554 "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
555 "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
556 "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
557 };
558 static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
559
560 static const char *whences[] = { "SET", "CUR", "END",
561 #ifdef SEEK_DATA
562 "DATA",
563 #endif
564 #ifdef SEEK_HOLE
565 "HOLE",
566 #endif
567 };
568 static DEFINE_STRARRAY(whences, "SEEK_");
569
570 static const char *fcntl_cmds[] = {
571 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
572 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
573 "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
574 "GETOWNER_UIDS",
575 };
576 static DEFINE_STRARRAY(fcntl_cmds, "F_");
577
578 static const char *fcntl_linux_specific_cmds[] = {
579 "SETLEASE", "GETLEASE", "NOTIFY", [5] = "CANCELLK", "DUPFD_CLOEXEC",
580 "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
581 "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
582 };
583
584 static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
585
586 static struct strarray *fcntl_cmds_arrays[] = {
587 &strarray__fcntl_cmds,
588 &strarray__fcntl_linux_specific_cmds,
589 };
590
591 static DEFINE_STRARRAYS(fcntl_cmds_arrays);
592
593 static const char *rlimit_resources[] = {
594 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
595 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
596 "RTTIME",
597 };
598 static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
599
600 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
601 static DEFINE_STRARRAY(sighow, "SIG_");
602
603 static const char *clockid[] = {
604 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
605 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
606 "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
607 };
608 static DEFINE_STRARRAY(clockid, "CLOCK_");
609
syscall_arg__scnprintf_access_mode(char * bf,size_t size,struct syscall_arg * arg)610 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
611 struct syscall_arg *arg)
612 {
613 bool show_prefix = arg->show_string_prefix;
614 const char *suffix = "_OK";
615 size_t printed = 0;
616 int mode = arg->val;
617
618 if (mode == F_OK) /* 0 */
619 return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
620 #define P_MODE(n) \
621 if (mode & n##_OK) { \
622 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
623 mode &= ~n##_OK; \
624 }
625
626 P_MODE(R);
627 P_MODE(W);
628 P_MODE(X);
629 #undef P_MODE
630
631 if (mode)
632 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
633
634 return printed;
635 }
636
637 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
638
639 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
640 struct syscall_arg *arg);
641
642 #define SCA_FILENAME syscall_arg__scnprintf_filename
643
syscall_arg__scnprintf_pipe_flags(char * bf,size_t size,struct syscall_arg * arg)644 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
645 struct syscall_arg *arg)
646 {
647 bool show_prefix = arg->show_string_prefix;
648 const char *prefix = "O_";
649 int printed = 0, flags = arg->val;
650
651 #define P_FLAG(n) \
652 if (flags & O_##n) { \
653 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
654 flags &= ~O_##n; \
655 }
656
657 P_FLAG(CLOEXEC);
658 P_FLAG(NONBLOCK);
659 #undef P_FLAG
660
661 if (flags)
662 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
663
664 return printed;
665 }
666
667 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
668
669 #ifndef GRND_NONBLOCK
670 #define GRND_NONBLOCK 0x0001
671 #endif
672 #ifndef GRND_RANDOM
673 #define GRND_RANDOM 0x0002
674 #endif
675
syscall_arg__scnprintf_getrandom_flags(char * bf,size_t size,struct syscall_arg * arg)676 static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
677 struct syscall_arg *arg)
678 {
679 bool show_prefix = arg->show_string_prefix;
680 const char *prefix = "GRND_";
681 int printed = 0, flags = arg->val;
682
683 #define P_FLAG(n) \
684 if (flags & GRND_##n) { \
685 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
686 flags &= ~GRND_##n; \
687 }
688
689 P_FLAG(RANDOM);
690 P_FLAG(NONBLOCK);
691 #undef P_FLAG
692
693 if (flags)
694 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
695
696 return printed;
697 }
698
699 #define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
700
701 #define STRARRAY(name, array) \
702 { .scnprintf = SCA_STRARRAY, \
703 .parm = &strarray__##array, }
704
705 #define STRARRAY_FLAGS(name, array) \
706 { .scnprintf = SCA_STRARRAY_FLAGS, \
707 .parm = &strarray__##array, }
708
709 #include "trace/beauty/arch_errno_names.c"
710 #include "trace/beauty/eventfd.c"
711 #include "trace/beauty/futex_op.c"
712 #include "trace/beauty/futex_val3.c"
713 #include "trace/beauty/mmap.c"
714 #include "trace/beauty/mode_t.c"
715 #include "trace/beauty/msg_flags.c"
716 #include "trace/beauty/open_flags.c"
717 #include "trace/beauty/perf_event_open.c"
718 #include "trace/beauty/pid.c"
719 #include "trace/beauty/sched_policy.c"
720 #include "trace/beauty/seccomp.c"
721 #include "trace/beauty/signum.c"
722 #include "trace/beauty/socket_type.c"
723 #include "trace/beauty/waitid_options.c"
724
725 static struct syscall_fmt syscall_fmts[] = {
726 { .name = "access",
727 .arg = { [1] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, },
728 { .name = "arch_prctl",
729 .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
730 [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
731 { .name = "bind",
732 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
733 [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ },
734 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
735 { .name = "bpf",
736 .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
737 { .name = "brk", .hexret = true,
738 .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
739 { .name = "clock_gettime",
740 .arg = { [0] = STRARRAY(clk_id, clockid), }, },
741 { .name = "clone", .errpid = true, .nr_args = 5,
742 .arg = { [0] = { .name = "flags", .scnprintf = SCA_CLONE_FLAGS, },
743 [1] = { .name = "child_stack", .scnprintf = SCA_HEX, },
744 [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
745 [3] = { .name = "child_tidptr", .scnprintf = SCA_HEX, },
746 [4] = { .name = "tls", .scnprintf = SCA_HEX, }, }, },
747 { .name = "close",
748 .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
749 { .name = "connect",
750 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
751 [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ },
752 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
753 { .name = "epoll_ctl",
754 .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
755 { .name = "eventfd2",
756 .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
757 { .name = "fchmodat",
758 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
759 { .name = "fchownat",
760 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
761 { .name = "fcntl",
762 .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
763 .parm = &strarrays__fcntl_cmds_arrays,
764 .show_zero = true, },
765 [2] = { .scnprintf = SCA_FCNTL_ARG, /* arg */ }, }, },
766 { .name = "flock",
767 .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
768 { .name = "fsconfig",
769 .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },
770 { .name = "fsmount",
771 .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),
772 [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },
773 { .name = "fspick",
774 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
775 [1] = { .scnprintf = SCA_FILENAME, /* path */ },
776 [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },
777 { .name = "fstat", .alias = "newfstat", },
778 { .name = "fstatat", .alias = "newfstatat", },
779 { .name = "futex",
780 .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
781 [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
782 { .name = "futimesat",
783 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
784 { .name = "getitimer",
785 .arg = { [0] = STRARRAY(which, itimers), }, },
786 { .name = "getpid", .errpid = true, },
787 { .name = "getpgid", .errpid = true, },
788 { .name = "getppid", .errpid = true, },
789 { .name = "getrandom",
790 .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
791 { .name = "getrlimit",
792 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
793 { .name = "gettid", .errpid = true, },
794 { .name = "ioctl",
795 .arg = {
796 #if defined(__i386__) || defined(__x86_64__)
797 /*
798 * FIXME: Make this available to all arches.
799 */
800 [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
801 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
802 #else
803 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
804 #endif
805 { .name = "kcmp", .nr_args = 5,
806 .arg = { [0] = { .name = "pid1", .scnprintf = SCA_PID, },
807 [1] = { .name = "pid2", .scnprintf = SCA_PID, },
808 [2] = { .name = "type", .scnprintf = SCA_KCMP_TYPE, },
809 [3] = { .name = "idx1", .scnprintf = SCA_KCMP_IDX, },
810 [4] = { .name = "idx2", .scnprintf = SCA_KCMP_IDX, }, }, },
811 { .name = "keyctl",
812 .arg = { [0] = STRARRAY(option, keyctl_options), }, },
813 { .name = "kill",
814 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
815 { .name = "linkat",
816 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
817 { .name = "lseek",
818 .arg = { [2] = STRARRAY(whence, whences), }, },
819 { .name = "lstat", .alias = "newlstat", },
820 { .name = "madvise",
821 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
822 [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
823 { .name = "mkdirat",
824 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
825 { .name = "mknodat",
826 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
827 { .name = "mmap", .hexret = true,
828 /* The standard mmap maps to old_mmap on s390x */
829 #if defined(__s390x__)
830 .alias = "old_mmap",
831 #endif
832 .arg = { [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
833 [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */ },
834 [5] = { .scnprintf = SCA_HEX, /* offset */ }, }, },
835 { .name = "mount",
836 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
837 [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
838 .mask_val = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
839 { .name = "move_mount",
840 .arg = { [0] = { .scnprintf = SCA_FDAT, /* from_dfd */ },
841 [1] = { .scnprintf = SCA_FILENAME, /* from_pathname */ },
842 [2] = { .scnprintf = SCA_FDAT, /* to_dfd */ },
843 [3] = { .scnprintf = SCA_FILENAME, /* to_pathname */ },
844 [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },
845 { .name = "mprotect",
846 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
847 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ }, }, },
848 { .name = "mq_unlink",
849 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
850 { .name = "mremap", .hexret = true,
851 .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
852 { .name = "name_to_handle_at",
853 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
854 { .name = "newfstatat",
855 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
856 { .name = "open",
857 .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
858 { .name = "open_by_handle_at",
859 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
860 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
861 { .name = "openat",
862 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
863 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
864 { .name = "perf_event_open",
865 .arg = { [2] = { .scnprintf = SCA_INT, /* cpu */ },
866 [3] = { .scnprintf = SCA_FD, /* group_fd */ },
867 [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
868 { .name = "pipe2",
869 .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
870 { .name = "pkey_alloc",
871 .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS, /* access_rights */ }, }, },
872 { .name = "pkey_free",
873 .arg = { [0] = { .scnprintf = SCA_INT, /* key */ }, }, },
874 { .name = "pkey_mprotect",
875 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
876 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
877 [3] = { .scnprintf = SCA_INT, /* pkey */ }, }, },
878 { .name = "poll", .timeout = true, },
879 { .name = "ppoll", .timeout = true, },
880 { .name = "prctl",
881 .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */ },
882 [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
883 [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
884 { .name = "pread", .alias = "pread64", },
885 { .name = "preadv", .alias = "pread", },
886 { .name = "prlimit64",
887 .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
888 { .name = "pwrite", .alias = "pwrite64", },
889 { .name = "readlinkat",
890 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
891 { .name = "recvfrom",
892 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
893 { .name = "recvmmsg",
894 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
895 { .name = "recvmsg",
896 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
897 { .name = "renameat",
898 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
899 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
900 { .name = "renameat2",
901 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
902 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
903 [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
904 { .name = "rt_sigaction",
905 .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
906 { .name = "rt_sigprocmask",
907 .arg = { [0] = STRARRAY(how, sighow), }, },
908 { .name = "rt_sigqueueinfo",
909 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
910 { .name = "rt_tgsigqueueinfo",
911 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
912 { .name = "sched_setscheduler",
913 .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
914 { .name = "seccomp",
915 .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP, /* op */ },
916 [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
917 { .name = "select", .timeout = true, },
918 { .name = "sendfile", .alias = "sendfile64", },
919 { .name = "sendmmsg",
920 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
921 { .name = "sendmsg",
922 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
923 { .name = "sendto",
924 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
925 [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
926 { .name = "set_tid_address", .errpid = true, },
927 { .name = "setitimer",
928 .arg = { [0] = STRARRAY(which, itimers), }, },
929 { .name = "setrlimit",
930 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
931 { .name = "socket",
932 .arg = { [0] = STRARRAY(family, socket_families),
933 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
934 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
935 { .name = "socketpair",
936 .arg = { [0] = STRARRAY(family, socket_families),
937 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
938 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
939 { .name = "stat", .alias = "newstat", },
940 { .name = "statx",
941 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fdat */ },
942 [2] = { .scnprintf = SCA_STATX_FLAGS, /* flags */ } ,
943 [3] = { .scnprintf = SCA_STATX_MASK, /* mask */ }, }, },
944 { .name = "swapoff",
945 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
946 { .name = "swapon",
947 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
948 { .name = "symlinkat",
949 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
950 { .name = "sync_file_range",
951 .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },
952 { .name = "tgkill",
953 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
954 { .name = "tkill",
955 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
956 { .name = "umount2", .alias = "umount",
957 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
958 { .name = "uname", .alias = "newuname", },
959 { .name = "unlinkat",
960 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
961 { .name = "utimensat",
962 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
963 { .name = "wait4", .errpid = true,
964 .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
965 { .name = "waitid", .errpid = true,
966 .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
967 };
968
syscall_fmt__cmp(const void * name,const void * fmtp)969 static int syscall_fmt__cmp(const void *name, const void *fmtp)
970 {
971 const struct syscall_fmt *fmt = fmtp;
972 return strcmp(name, fmt->name);
973 }
974
syscall_fmt__find(const char * name)975 static struct syscall_fmt *syscall_fmt__find(const char *name)
976 {
977 const int nmemb = ARRAY_SIZE(syscall_fmts);
978 return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
979 }
980
syscall_fmt__find_by_alias(const char * alias)981 static struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
982 {
983 int i, nmemb = ARRAY_SIZE(syscall_fmts);
984
985 for (i = 0; i < nmemb; ++i) {
986 if (syscall_fmts[i].alias && strcmp(syscall_fmts[i].alias, alias) == 0)
987 return &syscall_fmts[i];
988 }
989
990 return NULL;
991 }
992
993 /*
994 * is_exit: is this "exit" or "exit_group"?
995 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
996 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
997 * nonexistent: Just a hole in the syscall table, syscall id not allocated
998 */
999 struct syscall {
1000 struct tep_event *tp_format;
1001 int nr_args;
1002 int args_size;
1003 struct {
1004 struct bpf_program *sys_enter,
1005 *sys_exit;
1006 } bpf_prog;
1007 bool is_exit;
1008 bool is_open;
1009 bool nonexistent;
1010 struct tep_format_field *args;
1011 const char *name;
1012 struct syscall_fmt *fmt;
1013 struct syscall_arg_fmt *arg_fmt;
1014 };
1015
1016 /*
1017 * Must match what is in the BPF program:
1018 *
1019 * tools/perf/examples/bpf/augmented_raw_syscalls.c
1020 */
1021 struct bpf_map_syscall_entry {
1022 bool enabled;
1023 u16 string_args_len[RAW_SYSCALL_ARGS_NUM];
1024 };
1025
1026 /*
1027 * We need to have this 'calculated' boolean because in some cases we really
1028 * don't know what is the duration of a syscall, for instance, when we start
1029 * a session and some threads are waiting for a syscall to finish, say 'poll',
1030 * in which case all we can do is to print "( ? ) for duration and for the
1031 * start timestamp.
1032 */
fprintf_duration(unsigned long t,bool calculated,FILE * fp)1033 static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
1034 {
1035 double duration = (double)t / NSEC_PER_MSEC;
1036 size_t printed = fprintf(fp, "(");
1037
1038 if (!calculated)
1039 printed += fprintf(fp, " ");
1040 else if (duration >= 1.0)
1041 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1042 else if (duration >= 0.01)
1043 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1044 else
1045 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
1046 return printed + fprintf(fp, "): ");
1047 }
1048
1049 /**
1050 * filename.ptr: The filename char pointer that will be vfs_getname'd
1051 * filename.entry_str_pos: Where to insert the string translated from
1052 * filename.ptr by the vfs_getname tracepoint/kprobe.
1053 * ret_scnprintf: syscall args may set this to a different syscall return
1054 * formatter, for instance, fcntl may return fds, file flags, etc.
1055 */
1056 struct thread_trace {
1057 u64 entry_time;
1058 bool entry_pending;
1059 unsigned long nr_events;
1060 unsigned long pfmaj, pfmin;
1061 char *entry_str;
1062 double runtime_ms;
1063 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1064 struct {
1065 unsigned long ptr;
1066 short int entry_str_pos;
1067 bool pending_open;
1068 unsigned int namelen;
1069 char *name;
1070 } filename;
1071 struct {
1072 int max;
1073 struct file *table;
1074 } files;
1075
1076 struct intlist *syscall_stats;
1077 };
1078
thread_trace__new(void)1079 static struct thread_trace *thread_trace__new(void)
1080 {
1081 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
1082
1083 if (ttrace) {
1084 ttrace->files.max = -1;
1085 ttrace->syscall_stats = intlist__new(NULL);
1086 }
1087
1088 return ttrace;
1089 }
1090
thread__trace(struct thread * thread,FILE * fp)1091 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1092 {
1093 struct thread_trace *ttrace;
1094
1095 if (thread == NULL)
1096 goto fail;
1097
1098 if (thread__priv(thread) == NULL)
1099 thread__set_priv(thread, thread_trace__new());
1100
1101 if (thread__priv(thread) == NULL)
1102 goto fail;
1103
1104 ttrace = thread__priv(thread);
1105 ++ttrace->nr_events;
1106
1107 return ttrace;
1108 fail:
1109 color_fprintf(fp, PERF_COLOR_RED,
1110 "WARNING: not enough memory, dropping samples!\n");
1111 return NULL;
1112 }
1113
1114
syscall_arg__set_ret_scnprintf(struct syscall_arg * arg,size_t (* ret_scnprintf)(char * bf,size_t size,struct syscall_arg * arg))1115 void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
1116 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
1117 {
1118 struct thread_trace *ttrace = thread__priv(arg->thread);
1119
1120 ttrace->ret_scnprintf = ret_scnprintf;
1121 }
1122
1123 #define TRACE_PFMAJ (1 << 0)
1124 #define TRACE_PFMIN (1 << 1)
1125
1126 static const size_t trace__entry_str_size = 2048;
1127
thread_trace__files_entry(struct thread_trace * ttrace,int fd)1128 static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
1129 {
1130 if (fd < 0)
1131 return NULL;
1132
1133 if (fd > ttrace->files.max) {
1134 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
1135
1136 if (nfiles == NULL)
1137 return NULL;
1138
1139 if (ttrace->files.max != -1) {
1140 memset(nfiles + ttrace->files.max + 1, 0,
1141 (fd - ttrace->files.max) * sizeof(struct file));
1142 } else {
1143 memset(nfiles, 0, (fd + 1) * sizeof(struct file));
1144 }
1145
1146 ttrace->files.table = nfiles;
1147 ttrace->files.max = fd;
1148 }
1149
1150 return ttrace->files.table + fd;
1151 }
1152
thread__files_entry(struct thread * thread,int fd)1153 struct file *thread__files_entry(struct thread *thread, int fd)
1154 {
1155 return thread_trace__files_entry(thread__priv(thread), fd);
1156 }
1157
trace__set_fd_pathname(struct thread * thread,int fd,const char * pathname)1158 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1159 {
1160 struct thread_trace *ttrace = thread__priv(thread);
1161 struct file *file = thread_trace__files_entry(ttrace, fd);
1162
1163 if (file != NULL) {
1164 struct stat st;
1165 if (stat(pathname, &st) == 0)
1166 file->dev_maj = major(st.st_rdev);
1167 file->pathname = strdup(pathname);
1168 if (file->pathname)
1169 return 0;
1170 }
1171
1172 return -1;
1173 }
1174
thread__read_fd_path(struct thread * thread,int fd)1175 static int thread__read_fd_path(struct thread *thread, int fd)
1176 {
1177 char linkname[PATH_MAX], pathname[PATH_MAX];
1178 struct stat st;
1179 int ret;
1180
1181 if (thread->pid_ == thread->tid) {
1182 scnprintf(linkname, sizeof(linkname),
1183 "/proc/%d/fd/%d", thread->pid_, fd);
1184 } else {
1185 scnprintf(linkname, sizeof(linkname),
1186 "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1187 }
1188
1189 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1190 return -1;
1191
1192 ret = readlink(linkname, pathname, sizeof(pathname));
1193
1194 if (ret < 0 || ret > st.st_size)
1195 return -1;
1196
1197 pathname[ret] = '\0';
1198 return trace__set_fd_pathname(thread, fd, pathname);
1199 }
1200
thread__fd_path(struct thread * thread,int fd,struct trace * trace)1201 static const char *thread__fd_path(struct thread *thread, int fd,
1202 struct trace *trace)
1203 {
1204 struct thread_trace *ttrace = thread__priv(thread);
1205
1206 if (ttrace == NULL || trace->fd_path_disabled)
1207 return NULL;
1208
1209 if (fd < 0)
1210 return NULL;
1211
1212 if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
1213 if (!trace->live)
1214 return NULL;
1215 ++trace->stats.proc_getname;
1216 if (thread__read_fd_path(thread, fd))
1217 return NULL;
1218 }
1219
1220 return ttrace->files.table[fd].pathname;
1221 }
1222
syscall_arg__scnprintf_fd(char * bf,size_t size,struct syscall_arg * arg)1223 size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
1224 {
1225 int fd = arg->val;
1226 size_t printed = scnprintf(bf, size, "%d", fd);
1227 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1228
1229 if (path)
1230 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1231
1232 return printed;
1233 }
1234
pid__scnprintf_fd(struct trace * trace,pid_t pid,int fd,char * bf,size_t size)1235 size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1236 {
1237 size_t printed = scnprintf(bf, size, "%d", fd);
1238 struct thread *thread = machine__find_thread(trace->host, pid, pid);
1239
1240 if (thread) {
1241 const char *path = thread__fd_path(thread, fd, trace);
1242
1243 if (path)
1244 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1245
1246 thread__put(thread);
1247 }
1248
1249 return printed;
1250 }
1251
syscall_arg__scnprintf_close_fd(char * bf,size_t size,struct syscall_arg * arg)1252 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1253 struct syscall_arg *arg)
1254 {
1255 int fd = arg->val;
1256 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1257 struct thread_trace *ttrace = thread__priv(arg->thread);
1258
1259 if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1260 zfree(&ttrace->files.table[fd].pathname);
1261
1262 return printed;
1263 }
1264
thread__set_filename_pos(struct thread * thread,const char * bf,unsigned long ptr)1265 static void thread__set_filename_pos(struct thread *thread, const char *bf,
1266 unsigned long ptr)
1267 {
1268 struct thread_trace *ttrace = thread__priv(thread);
1269
1270 ttrace->filename.ptr = ptr;
1271 ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1272 }
1273
syscall_arg__scnprintf_augmented_string(struct syscall_arg * arg,char * bf,size_t size)1274 static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1275 {
1276 struct augmented_arg *augmented_arg = arg->augmented.args;
1277 size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1278 /*
1279 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1280 * we would have two strings, each prefixed by its size.
1281 */
1282 int consumed = sizeof(*augmented_arg) + augmented_arg->size;
1283
1284 arg->augmented.args = ((void *)arg->augmented.args) + consumed;
1285 arg->augmented.size -= consumed;
1286
1287 return printed;
1288 }
1289
syscall_arg__scnprintf_filename(char * bf,size_t size,struct syscall_arg * arg)1290 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1291 struct syscall_arg *arg)
1292 {
1293 unsigned long ptr = arg->val;
1294
1295 if (arg->augmented.args)
1296 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1297
1298 if (!arg->trace->vfs_getname)
1299 return scnprintf(bf, size, "%#x", ptr);
1300
1301 thread__set_filename_pos(arg->thread, bf, ptr);
1302 return 0;
1303 }
1304
trace__filter_duration(struct trace * trace,double t)1305 static bool trace__filter_duration(struct trace *trace, double t)
1306 {
1307 return t < (trace->duration_filter * NSEC_PER_MSEC);
1308 }
1309
__trace__fprintf_tstamp(struct trace * trace,u64 tstamp,FILE * fp)1310 static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1311 {
1312 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1313
1314 return fprintf(fp, "%10.3f ", ts);
1315 }
1316
1317 /*
1318 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1319 * using ttrace->entry_time for a thread that receives a sys_exit without
1320 * first having received a sys_enter ("poll" issued before tracing session
1321 * starts, lost sys_enter exit due to ring buffer overflow).
1322 */
trace__fprintf_tstamp(struct trace * trace,u64 tstamp,FILE * fp)1323 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1324 {
1325 if (tstamp > 0)
1326 return __trace__fprintf_tstamp(trace, tstamp, fp);
1327
1328 return fprintf(fp, " ? ");
1329 }
1330
1331 static bool done = false;
1332 static bool interrupted = false;
1333
sig_handler(int sig)1334 static void sig_handler(int sig)
1335 {
1336 done = true;
1337 interrupted = sig == SIGINT;
1338 }
1339
trace__fprintf_comm_tid(struct trace * trace,struct thread * thread,FILE * fp)1340 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1341 {
1342 size_t printed = 0;
1343
1344 if (trace->multiple_threads) {
1345 if (trace->show_comm)
1346 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1347 printed += fprintf(fp, "%d ", thread->tid);
1348 }
1349
1350 return printed;
1351 }
1352
trace__fprintf_entry_head(struct trace * trace,struct thread * thread,u64 duration,bool duration_calculated,u64 tstamp,FILE * fp)1353 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1354 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1355 {
1356 size_t printed = 0;
1357
1358 if (trace->show_tstamp)
1359 printed = trace__fprintf_tstamp(trace, tstamp, fp);
1360 if (trace->show_duration)
1361 printed += fprintf_duration(duration, duration_calculated, fp);
1362 return printed + trace__fprintf_comm_tid(trace, thread, fp);
1363 }
1364
trace__process_event(struct trace * trace,struct machine * machine,union perf_event * event,struct perf_sample * sample)1365 static int trace__process_event(struct trace *trace, struct machine *machine,
1366 union perf_event *event, struct perf_sample *sample)
1367 {
1368 int ret = 0;
1369
1370 switch (event->header.type) {
1371 case PERF_RECORD_LOST:
1372 color_fprintf(trace->output, PERF_COLOR_RED,
1373 "LOST %" PRIu64 " events!\n", event->lost.lost);
1374 ret = machine__process_lost_event(machine, event, sample);
1375 break;
1376 default:
1377 ret = machine__process_event(machine, event, sample);
1378 break;
1379 }
1380
1381 return ret;
1382 }
1383
trace__tool_process(struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct machine * machine)1384 static int trace__tool_process(struct perf_tool *tool,
1385 union perf_event *event,
1386 struct perf_sample *sample,
1387 struct machine *machine)
1388 {
1389 struct trace *trace = container_of(tool, struct trace, tool);
1390 return trace__process_event(trace, machine, event, sample);
1391 }
1392
trace__machine__resolve_kernel_addr(void * vmachine,unsigned long long * addrp,char ** modp)1393 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1394 {
1395 struct machine *machine = vmachine;
1396
1397 if (machine->kptr_restrict_warned)
1398 return NULL;
1399
1400 if (symbol_conf.kptr_restrict) {
1401 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1402 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
1403 "Kernel samples will not be resolved.\n");
1404 machine->kptr_restrict_warned = true;
1405 return NULL;
1406 }
1407
1408 return machine__resolve_kernel_addr(vmachine, addrp, modp);
1409 }
1410
trace__symbols_init(struct trace * trace,struct evlist * evlist)1411 static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
1412 {
1413 int err = symbol__init(NULL);
1414
1415 if (err)
1416 return err;
1417
1418 trace->host = machine__new_host();
1419 if (trace->host == NULL)
1420 return -ENOMEM;
1421
1422 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1423 if (err < 0)
1424 goto out;
1425
1426 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1427 evlist->core.threads, trace__tool_process, false,
1428 1);
1429 out:
1430 if (err)
1431 symbol__exit();
1432
1433 return err;
1434 }
1435
trace__symbols__exit(struct trace * trace)1436 static void trace__symbols__exit(struct trace *trace)
1437 {
1438 machine__exit(trace->host);
1439 trace->host = NULL;
1440
1441 symbol__exit();
1442 }
1443
syscall__alloc_arg_fmts(struct syscall * sc,int nr_args)1444 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1445 {
1446 int idx;
1447
1448 if (nr_args == RAW_SYSCALL_ARGS_NUM && sc->fmt && sc->fmt->nr_args != 0)
1449 nr_args = sc->fmt->nr_args;
1450
1451 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1452 if (sc->arg_fmt == NULL)
1453 return -1;
1454
1455 for (idx = 0; idx < nr_args; ++idx) {
1456 if (sc->fmt)
1457 sc->arg_fmt[idx] = sc->fmt->arg[idx];
1458 }
1459
1460 sc->nr_args = nr_args;
1461 return 0;
1462 }
1463
1464 static struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {
1465 };
1466
syscall_arg_fmt__cmp(const void * name,const void * fmtp)1467 static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)
1468 {
1469 const struct syscall_arg_fmt *fmt = fmtp;
1470 return strcmp(name, fmt->name);
1471 }
1472
1473 static struct syscall_arg_fmt *
__syscall_arg_fmt__find_by_name(struct syscall_arg_fmt * fmts,const int nmemb,const char * name)1474 __syscall_arg_fmt__find_by_name(struct syscall_arg_fmt *fmts, const int nmemb, const char *name)
1475 {
1476 return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);
1477 }
1478
syscall_arg_fmt__find_by_name(const char * name)1479 static struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)
1480 {
1481 const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);
1482 return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);
1483 }
1484
1485 static struct tep_format_field *
syscall_arg_fmt__init_array(struct syscall_arg_fmt * arg,struct tep_format_field * field)1486 syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field)
1487 {
1488 struct tep_format_field *last_field = NULL;
1489 int len;
1490
1491 for (; field; field = field->next, ++arg) {
1492 last_field = field;
1493
1494 if (arg->scnprintf)
1495 continue;
1496
1497 len = strlen(field->name);
1498
1499 if (strcmp(field->type, "const char *") == 0 &&
1500 ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1501 strstr(field->name, "path") != NULL))
1502 arg->scnprintf = SCA_FILENAME;
1503 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
1504 arg->scnprintf = SCA_PTR;
1505 else if (strcmp(field->type, "pid_t") == 0)
1506 arg->scnprintf = SCA_PID;
1507 else if (strcmp(field->type, "umode_t") == 0)
1508 arg->scnprintf = SCA_MODE_T;
1509 else if ((strcmp(field->type, "int") == 0 ||
1510 strcmp(field->type, "unsigned int") == 0 ||
1511 strcmp(field->type, "long") == 0) &&
1512 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
1513 /*
1514 * /sys/kernel/tracing/events/syscalls/sys_enter*
1515 * egrep 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
1516 * 65 int
1517 * 23 unsigned int
1518 * 7 unsigned long
1519 */
1520 arg->scnprintf = SCA_FD;
1521 } else {
1522 struct syscall_arg_fmt *fmt = syscall_arg_fmt__find_by_name(field->name);
1523
1524 if (fmt) {
1525 arg->scnprintf = fmt->scnprintf;
1526 arg->strtoul = fmt->strtoul;
1527 }
1528 }
1529 }
1530
1531 return last_field;
1532 }
1533
syscall__set_arg_fmts(struct syscall * sc)1534 static int syscall__set_arg_fmts(struct syscall *sc)
1535 {
1536 struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args);
1537
1538 if (last_field)
1539 sc->args_size = last_field->offset + last_field->size;
1540
1541 return 0;
1542 }
1543
trace__read_syscall_info(struct trace * trace,int id)1544 static int trace__read_syscall_info(struct trace *trace, int id)
1545 {
1546 char tp_name[128];
1547 struct syscall *sc;
1548 const char *name = syscalltbl__name(trace->sctbl, id);
1549
1550 if (trace->syscalls.table == NULL) {
1551 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
1552 if (trace->syscalls.table == NULL)
1553 return -ENOMEM;
1554 }
1555
1556 sc = trace->syscalls.table + id;
1557 if (sc->nonexistent)
1558 return -EEXIST;
1559
1560 if (name == NULL) {
1561 sc->nonexistent = true;
1562 return -EEXIST;
1563 }
1564
1565 sc->name = name;
1566 sc->fmt = syscall_fmt__find(sc->name);
1567
1568 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1569 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1570
1571 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
1572 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1573 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1574 }
1575
1576 /*
1577 * Fails to read trace point format via sysfs node, so the trace point
1578 * doesn't exist. Set the 'nonexistent' flag as true.
1579 */
1580 if (IS_ERR(sc->tp_format)) {
1581 sc->nonexistent = true;
1582 return PTR_ERR(sc->tp_format);
1583 }
1584
1585 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ?
1586 RAW_SYSCALL_ARGS_NUM : sc->tp_format->format.nr_fields))
1587 return -ENOMEM;
1588
1589 sc->args = sc->tp_format->format.fields;
1590 /*
1591 * We need to check and discard the first variable '__syscall_nr'
1592 * or 'nr' that mean the syscall number. It is needless here.
1593 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1594 */
1595 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
1596 sc->args = sc->args->next;
1597 --sc->nr_args;
1598 }
1599
1600 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1601 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
1602
1603 return syscall__set_arg_fmts(sc);
1604 }
1605
intcmp(const void * a,const void * b)1606 static int intcmp(const void *a, const void *b)
1607 {
1608 const int *one = a, *another = b;
1609
1610 return *one - *another;
1611 }
1612
trace__validate_ev_qualifier(struct trace * trace)1613 static int trace__validate_ev_qualifier(struct trace *trace)
1614 {
1615 int err = 0;
1616 bool printed_invalid_prefix = false;
1617 struct str_node *pos;
1618 size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
1619
1620 trace->ev_qualifier_ids.entries = malloc(nr_allocated *
1621 sizeof(trace->ev_qualifier_ids.entries[0]));
1622
1623 if (trace->ev_qualifier_ids.entries == NULL) {
1624 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1625 trace->output);
1626 err = -EINVAL;
1627 goto out;
1628 }
1629
1630 strlist__for_each_entry(pos, trace->ev_qualifier) {
1631 const char *sc = pos->s;
1632 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
1633
1634 if (id < 0) {
1635 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1636 if (id >= 0)
1637 goto matches;
1638
1639 if (!printed_invalid_prefix) {
1640 pr_debug("Skipping unknown syscalls: ");
1641 printed_invalid_prefix = true;
1642 } else {
1643 pr_debug(", ");
1644 }
1645
1646 pr_debug("%s", sc);
1647 continue;
1648 }
1649 matches:
1650 trace->ev_qualifier_ids.entries[nr_used++] = id;
1651 if (match_next == -1)
1652 continue;
1653
1654 while (1) {
1655 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1656 if (id < 0)
1657 break;
1658 if (nr_allocated == nr_used) {
1659 void *entries;
1660
1661 nr_allocated += 8;
1662 entries = realloc(trace->ev_qualifier_ids.entries,
1663 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1664 if (entries == NULL) {
1665 err = -ENOMEM;
1666 fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1667 goto out_free;
1668 }
1669 trace->ev_qualifier_ids.entries = entries;
1670 }
1671 trace->ev_qualifier_ids.entries[nr_used++] = id;
1672 }
1673 }
1674
1675 trace->ev_qualifier_ids.nr = nr_used;
1676 qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
1677 out:
1678 if (printed_invalid_prefix)
1679 pr_debug("\n");
1680 return err;
1681 out_free:
1682 zfree(&trace->ev_qualifier_ids.entries);
1683 trace->ev_qualifier_ids.nr = 0;
1684 goto out;
1685 }
1686
trace__syscall_enabled(struct trace * trace,int id)1687 static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1688 {
1689 bool in_ev_qualifier;
1690
1691 if (trace->ev_qualifier_ids.nr == 0)
1692 return true;
1693
1694 in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1695 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1696
1697 if (in_ev_qualifier)
1698 return !trace->not_ev_qualifier;
1699
1700 return trace->not_ev_qualifier;
1701 }
1702
1703 /*
1704 * args is to be interpreted as a series of longs but we need to handle
1705 * 8-byte unaligned accesses. args points to raw_data within the event
1706 * and raw_data is guaranteed to be 8-byte unaligned because it is
1707 * preceded by raw_size which is a u32. So we need to copy args to a temp
1708 * variable to read it. Most notably this avoids extended load instructions
1709 * on unaligned addresses
1710 */
syscall_arg__val(struct syscall_arg * arg,u8 idx)1711 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
1712 {
1713 unsigned long val;
1714 unsigned char *p = arg->args + sizeof(unsigned long) * idx;
1715
1716 memcpy(&val, p, sizeof(val));
1717 return val;
1718 }
1719
syscall__scnprintf_name(struct syscall * sc,char * bf,size_t size,struct syscall_arg * arg)1720 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1721 struct syscall_arg *arg)
1722 {
1723 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1724 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1725
1726 return scnprintf(bf, size, "arg%d: ", arg->idx);
1727 }
1728
1729 /*
1730 * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1731 * as mount 'flags' argument that needs ignoring some magic flag, see comment
1732 * in tools/perf/trace/beauty/mount_flags.c
1733 */
syscall__mask_val(struct syscall * sc,struct syscall_arg * arg,unsigned long val)1734 static unsigned long syscall__mask_val(struct syscall *sc, struct syscall_arg *arg, unsigned long val)
1735 {
1736 if (sc->arg_fmt && sc->arg_fmt[arg->idx].mask_val)
1737 return sc->arg_fmt[arg->idx].mask_val(arg, val);
1738
1739 return val;
1740 }
1741
syscall__scnprintf_val(struct syscall * sc,char * bf,size_t size,struct syscall_arg * arg,unsigned long val)1742 static size_t syscall__scnprintf_val(struct syscall *sc, char *bf, size_t size,
1743 struct syscall_arg *arg, unsigned long val)
1744 {
1745 if (sc->arg_fmt && sc->arg_fmt[arg->idx].scnprintf) {
1746 arg->val = val;
1747 if (sc->arg_fmt[arg->idx].parm)
1748 arg->parm = sc->arg_fmt[arg->idx].parm;
1749 return sc->arg_fmt[arg->idx].scnprintf(bf, size, arg);
1750 }
1751 return scnprintf(bf, size, "%ld", val);
1752 }
1753
syscall__scnprintf_args(struct syscall * sc,char * bf,size_t size,unsigned char * args,void * augmented_args,int augmented_args_size,struct trace * trace,struct thread * thread)1754 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
1755 unsigned char *args, void *augmented_args, int augmented_args_size,
1756 struct trace *trace, struct thread *thread)
1757 {
1758 size_t printed = 0;
1759 unsigned long val;
1760 u8 bit = 1;
1761 struct syscall_arg arg = {
1762 .args = args,
1763 .augmented = {
1764 .size = augmented_args_size,
1765 .args = augmented_args,
1766 },
1767 .idx = 0,
1768 .mask = 0,
1769 .trace = trace,
1770 .thread = thread,
1771 .show_string_prefix = trace->show_string_prefix,
1772 };
1773 struct thread_trace *ttrace = thread__priv(thread);
1774
1775 /*
1776 * Things like fcntl will set this in its 'cmd' formatter to pick the
1777 * right formatter for the return value (an fd? file flags?), which is
1778 * not needed for syscalls that always return a given type, say an fd.
1779 */
1780 ttrace->ret_scnprintf = NULL;
1781
1782 if (sc->args != NULL) {
1783 struct tep_format_field *field;
1784
1785 for (field = sc->args; field;
1786 field = field->next, ++arg.idx, bit <<= 1) {
1787 if (arg.mask & bit)
1788 continue;
1789
1790 arg.fmt = &sc->arg_fmt[arg.idx];
1791 val = syscall_arg__val(&arg, arg.idx);
1792 /*
1793 * Some syscall args need some mask, most don't and
1794 * return val untouched.
1795 */
1796 val = syscall__mask_val(sc, &arg, val);
1797
1798 /*
1799 * Suppress this argument if its value is zero and
1800 * and we don't have a string associated in an
1801 * strarray for it.
1802 */
1803 if (val == 0 &&
1804 !trace->show_zeros &&
1805 !(sc->arg_fmt &&
1806 (sc->arg_fmt[arg.idx].show_zero ||
1807 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
1808 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
1809 sc->arg_fmt[arg.idx].parm))
1810 continue;
1811
1812 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
1813
1814 if (trace->show_arg_names)
1815 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
1816
1817 printed += syscall__scnprintf_val(sc, bf + printed, size - printed, &arg, val);
1818 }
1819 } else if (IS_ERR(sc->tp_format)) {
1820 /*
1821 * If we managed to read the tracepoint /format file, then we
1822 * may end up not having any args, like with gettid(), so only
1823 * print the raw args when we didn't manage to read it.
1824 */
1825 while (arg.idx < sc->nr_args) {
1826 if (arg.mask & bit)
1827 goto next_arg;
1828 val = syscall_arg__val(&arg, arg.idx);
1829 if (printed)
1830 printed += scnprintf(bf + printed, size - printed, ", ");
1831 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
1832 printed += syscall__scnprintf_val(sc, bf + printed, size - printed, &arg, val);
1833 next_arg:
1834 ++arg.idx;
1835 bit <<= 1;
1836 }
1837 }
1838
1839 return printed;
1840 }
1841
1842 typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
1843 union perf_event *event,
1844 struct perf_sample *sample);
1845
trace__syscall_info(struct trace * trace,struct evsel * evsel,int id)1846 static struct syscall *trace__syscall_info(struct trace *trace,
1847 struct evsel *evsel, int id)
1848 {
1849 int err = 0;
1850
1851 if (id < 0) {
1852
1853 /*
1854 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
1855 * before that, leaving at a higher verbosity level till that is
1856 * explained. Reproduced with plain ftrace with:
1857 *
1858 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
1859 * grep "NR -1 " /t/trace_pipe
1860 *
1861 * After generating some load on the machine.
1862 */
1863 if (verbose > 1) {
1864 static u64 n;
1865 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
1866 id, perf_evsel__name(evsel), ++n);
1867 }
1868 return NULL;
1869 }
1870
1871 err = -EINVAL;
1872
1873 if (id > trace->sctbl->syscalls.max_id)
1874 goto out_cant_read;
1875
1876 if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
1877 (err = trace__read_syscall_info(trace, id)) != 0)
1878 goto out_cant_read;
1879
1880 if (trace->syscalls.table && trace->syscalls.table[id].nonexistent)
1881 goto out_cant_read;
1882
1883 return &trace->syscalls.table[id];
1884
1885 out_cant_read:
1886 if (verbose > 0) {
1887 char sbuf[STRERR_BUFSIZE];
1888 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
1889 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
1890 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
1891 fputs(" information\n", trace->output);
1892 }
1893 return NULL;
1894 }
1895
thread__update_stats(struct thread_trace * ttrace,int id,struct perf_sample * sample)1896 static void thread__update_stats(struct thread_trace *ttrace,
1897 int id, struct perf_sample *sample)
1898 {
1899 struct int_node *inode;
1900 struct stats *stats;
1901 u64 duration = 0;
1902
1903 inode = intlist__findnew(ttrace->syscall_stats, id);
1904 if (inode == NULL)
1905 return;
1906
1907 stats = inode->priv;
1908 if (stats == NULL) {
1909 stats = malloc(sizeof(struct stats));
1910 if (stats == NULL)
1911 return;
1912 init_stats(stats);
1913 inode->priv = stats;
1914 }
1915
1916 if (ttrace->entry_time && sample->time > ttrace->entry_time)
1917 duration = sample->time - ttrace->entry_time;
1918
1919 update_stats(stats, duration);
1920 }
1921
trace__printf_interrupted_entry(struct trace * trace)1922 static int trace__printf_interrupted_entry(struct trace *trace)
1923 {
1924 struct thread_trace *ttrace;
1925 size_t printed;
1926 int len;
1927
1928 if (trace->failure_only || trace->current == NULL)
1929 return 0;
1930
1931 ttrace = thread__priv(trace->current);
1932
1933 if (!ttrace->entry_pending)
1934 return 0;
1935
1936 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
1937 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
1938
1939 if (len < trace->args_alignment - 4)
1940 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
1941
1942 printed += fprintf(trace->output, " ...\n");
1943
1944 ttrace->entry_pending = false;
1945 ++trace->nr_events_printed;
1946
1947 return printed;
1948 }
1949
trace__fprintf_sample(struct trace * trace,struct evsel * evsel,struct perf_sample * sample,struct thread * thread)1950 static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
1951 struct perf_sample *sample, struct thread *thread)
1952 {
1953 int printed = 0;
1954
1955 if (trace->print_sample) {
1956 double ts = (double)sample->time / NSEC_PER_MSEC;
1957
1958 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
1959 perf_evsel__name(evsel), ts,
1960 thread__comm_str(thread),
1961 sample->pid, sample->tid, sample->cpu);
1962 }
1963
1964 return printed;
1965 }
1966
syscall__augmented_args(struct syscall * sc,struct perf_sample * sample,int * augmented_args_size,int raw_augmented_args_size)1967 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
1968 {
1969 void *augmented_args = NULL;
1970 /*
1971 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
1972 * and there we get all 6 syscall args plus the tracepoint common fields
1973 * that gets calculated at the start and the syscall_nr (another long).
1974 * So we check if that is the case and if so don't look after the
1975 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
1976 * which is fixed.
1977 *
1978 * We'll revisit this later to pass s->args_size to the BPF augmenter
1979 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
1980 * copies only what we need for each syscall, like what happens when we
1981 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
1982 * traffic to just what is needed for each syscall.
1983 */
1984 int args_size = raw_augmented_args_size ?: sc->args_size;
1985
1986 *augmented_args_size = sample->raw_size - args_size;
1987 if (*augmented_args_size > 0)
1988 augmented_args = sample->raw_data + args_size;
1989
1990 return augmented_args;
1991 }
1992
trace__sys_enter(struct trace * trace,struct evsel * evsel,union perf_event * event __maybe_unused,struct perf_sample * sample)1993 static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
1994 union perf_event *event __maybe_unused,
1995 struct perf_sample *sample)
1996 {
1997 char *msg;
1998 void *args;
1999 int printed = 0;
2000 struct thread *thread;
2001 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2002 int augmented_args_size = 0;
2003 void *augmented_args = NULL;
2004 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2005 struct thread_trace *ttrace;
2006
2007 if (sc == NULL)
2008 return -1;
2009
2010 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2011 ttrace = thread__trace(thread, trace->output);
2012 if (ttrace == NULL)
2013 goto out_put;
2014
2015 trace__fprintf_sample(trace, evsel, sample, thread);
2016
2017 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2018
2019 if (ttrace->entry_str == NULL) {
2020 ttrace->entry_str = malloc(trace__entry_str_size);
2021 if (!ttrace->entry_str)
2022 goto out_put;
2023 }
2024
2025 if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
2026 trace__printf_interrupted_entry(trace);
2027 /*
2028 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2029 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2030 * this breaks syscall__augmented_args() check for augmented args, as we calculate
2031 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2032 * so when handling, say the openat syscall, we end up getting 6 args for the
2033 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2034 * thinking that the extra 2 u64 args are the augmented filename, so just check
2035 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2036 */
2037 if (evsel != trace->syscalls.events.sys_enter)
2038 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2039 ttrace->entry_time = sample->time;
2040 msg = ttrace->entry_str;
2041 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
2042
2043 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
2044 args, augmented_args, augmented_args_size, trace, thread);
2045
2046 if (sc->is_exit) {
2047 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
2048 int alignment = 0;
2049
2050 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
2051 printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2052 if (trace->args_alignment > printed)
2053 alignment = trace->args_alignment - printed;
2054 fprintf(trace->output, "%*s= ?\n", alignment, " ");
2055 }
2056 } else {
2057 ttrace->entry_pending = true;
2058 /* See trace__vfs_getname & trace__sys_exit */
2059 ttrace->filename.pending_open = false;
2060 }
2061
2062 if (trace->current != thread) {
2063 thread__put(trace->current);
2064 trace->current = thread__get(thread);
2065 }
2066 err = 0;
2067 out_put:
2068 thread__put(thread);
2069 return err;
2070 }
2071
trace__fprintf_sys_enter(struct trace * trace,struct evsel * evsel,struct perf_sample * sample)2072 static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
2073 struct perf_sample *sample)
2074 {
2075 struct thread_trace *ttrace;
2076 struct thread *thread;
2077 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2078 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2079 char msg[1024];
2080 void *args, *augmented_args = NULL;
2081 int augmented_args_size;
2082
2083 if (sc == NULL)
2084 return -1;
2085
2086 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2087 ttrace = thread__trace(thread, trace->output);
2088 /*
2089 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2090 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2091 */
2092 if (ttrace == NULL)
2093 goto out_put;
2094
2095 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2096 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2097 syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
2098 fprintf(trace->output, "%s", msg);
2099 err = 0;
2100 out_put:
2101 thread__put(thread);
2102 return err;
2103 }
2104
trace__resolve_callchain(struct trace * trace,struct evsel * evsel,struct perf_sample * sample,struct callchain_cursor * cursor)2105 static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
2106 struct perf_sample *sample,
2107 struct callchain_cursor *cursor)
2108 {
2109 struct addr_location al;
2110 int max_stack = evsel->core.attr.sample_max_stack ?
2111 evsel->core.attr.sample_max_stack :
2112 trace->max_stack;
2113 int err;
2114
2115 if (machine__resolve(trace->host, &al, sample) < 0)
2116 return -1;
2117
2118 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
2119 addr_location__put(&al);
2120 return err;
2121 }
2122
trace__fprintf_callchain(struct trace * trace,struct perf_sample * sample)2123 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2124 {
2125 /* TODO: user-configurable print_opts */
2126 const unsigned int print_opts = EVSEL__PRINT_SYM |
2127 EVSEL__PRINT_DSO |
2128 EVSEL__PRINT_UNKNOWN_AS_ADDR;
2129
2130 return sample__fprintf_callchain(sample, 38, print_opts, &callchain_cursor, symbol_conf.bt_stop_list, trace->output);
2131 }
2132
errno_to_name(struct evsel * evsel,int err)2133 static const char *errno_to_name(struct evsel *evsel, int err)
2134 {
2135 struct perf_env *env = perf_evsel__env(evsel);
2136 const char *arch_name = perf_env__arch(env);
2137
2138 return arch_syscalls__strerrno(arch_name, err);
2139 }
2140
trace__sys_exit(struct trace * trace,struct evsel * evsel,union perf_event * event __maybe_unused,struct perf_sample * sample)2141 static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
2142 union perf_event *event __maybe_unused,
2143 struct perf_sample *sample)
2144 {
2145 long ret;
2146 u64 duration = 0;
2147 bool duration_calculated = false;
2148 struct thread *thread;
2149 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2150 int alignment = trace->args_alignment;
2151 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2152 struct thread_trace *ttrace;
2153
2154 if (sc == NULL)
2155 return -1;
2156
2157 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2158 ttrace = thread__trace(thread, trace->output);
2159 if (ttrace == NULL)
2160 goto out_put;
2161
2162 trace__fprintf_sample(trace, evsel, sample, thread);
2163
2164 if (trace->summary)
2165 thread__update_stats(ttrace, id, sample);
2166
2167 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2168
2169 if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2170 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2171 ttrace->filename.pending_open = false;
2172 ++trace->stats.vfs_getname;
2173 }
2174
2175 if (ttrace->entry_time) {
2176 duration = sample->time - ttrace->entry_time;
2177 if (trace__filter_duration(trace, duration))
2178 goto out;
2179 duration_calculated = true;
2180 } else if (trace->duration_filter)
2181 goto out;
2182
2183 if (sample->callchain) {
2184 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2185 if (callchain_ret == 0) {
2186 if (callchain_cursor.nr < trace->min_stack)
2187 goto out;
2188 callchain_ret = 1;
2189 }
2190 }
2191
2192 if (trace->summary_only || (ret >= 0 && trace->failure_only))
2193 goto out;
2194
2195 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2196
2197 if (ttrace->entry_pending) {
2198 printed = fprintf(trace->output, "%s", ttrace->entry_str);
2199 } else {
2200 printed += fprintf(trace->output, " ... [");
2201 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2202 printed += 9;
2203 printed += fprintf(trace->output, "]: %s()", sc->name);
2204 }
2205
2206 printed++; /* the closing ')' */
2207
2208 if (alignment > printed)
2209 alignment -= printed;
2210 else
2211 alignment = 0;
2212
2213 fprintf(trace->output, ")%*s= ", alignment, " ");
2214
2215 if (sc->fmt == NULL) {
2216 if (ret < 0)
2217 goto errno_print;
2218 signed_print:
2219 fprintf(trace->output, "%ld", ret);
2220 } else if (ret < 0) {
2221 errno_print: {
2222 char bf[STRERR_BUFSIZE];
2223 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2224 *e = errno_to_name(evsel, -ret);
2225
2226 fprintf(trace->output, "-1 %s (%s)", e, emsg);
2227 }
2228 } else if (ret == 0 && sc->fmt->timeout)
2229 fprintf(trace->output, "0 (Timeout)");
2230 else if (ttrace->ret_scnprintf) {
2231 char bf[1024];
2232 struct syscall_arg arg = {
2233 .val = ret,
2234 .thread = thread,
2235 .trace = trace,
2236 };
2237 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2238 ttrace->ret_scnprintf = NULL;
2239 fprintf(trace->output, "%s", bf);
2240 } else if (sc->fmt->hexret)
2241 fprintf(trace->output, "%#lx", ret);
2242 else if (sc->fmt->errpid) {
2243 struct thread *child = machine__find_thread(trace->host, ret, ret);
2244
2245 if (child != NULL) {
2246 fprintf(trace->output, "%ld", ret);
2247 if (child->comm_set)
2248 fprintf(trace->output, " (%s)", thread__comm_str(child));
2249 thread__put(child);
2250 }
2251 } else
2252 goto signed_print;
2253
2254 fputc('\n', trace->output);
2255
2256 /*
2257 * We only consider an 'event' for the sake of --max-events a non-filtered
2258 * sys_enter + sys_exit and other tracepoint events.
2259 */
2260 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2261 interrupted = true;
2262
2263 if (callchain_ret > 0)
2264 trace__fprintf_callchain(trace, sample);
2265 else if (callchain_ret < 0)
2266 pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2267 out:
2268 ttrace->entry_pending = false;
2269 err = 0;
2270 out_put:
2271 thread__put(thread);
2272 return err;
2273 }
2274
trace__vfs_getname(struct trace * trace,struct evsel * evsel,union perf_event * event __maybe_unused,struct perf_sample * sample)2275 static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
2276 union perf_event *event __maybe_unused,
2277 struct perf_sample *sample)
2278 {
2279 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2280 struct thread_trace *ttrace;
2281 size_t filename_len, entry_str_len, to_move;
2282 ssize_t remaining_space;
2283 char *pos;
2284 const char *filename = perf_evsel__rawptr(evsel, sample, "pathname");
2285
2286 if (!thread)
2287 goto out;
2288
2289 ttrace = thread__priv(thread);
2290 if (!ttrace)
2291 goto out_put;
2292
2293 filename_len = strlen(filename);
2294 if (filename_len == 0)
2295 goto out_put;
2296
2297 if (ttrace->filename.namelen < filename_len) {
2298 char *f = realloc(ttrace->filename.name, filename_len + 1);
2299
2300 if (f == NULL)
2301 goto out_put;
2302
2303 ttrace->filename.namelen = filename_len;
2304 ttrace->filename.name = f;
2305 }
2306
2307 strcpy(ttrace->filename.name, filename);
2308 ttrace->filename.pending_open = true;
2309
2310 if (!ttrace->filename.ptr)
2311 goto out_put;
2312
2313 entry_str_len = strlen(ttrace->entry_str);
2314 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2315 if (remaining_space <= 0)
2316 goto out_put;
2317
2318 if (filename_len > (size_t)remaining_space) {
2319 filename += filename_len - remaining_space;
2320 filename_len = remaining_space;
2321 }
2322
2323 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2324 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2325 memmove(pos + filename_len, pos, to_move);
2326 memcpy(pos, filename, filename_len);
2327
2328 ttrace->filename.ptr = 0;
2329 ttrace->filename.entry_str_pos = 0;
2330 out_put:
2331 thread__put(thread);
2332 out:
2333 return 0;
2334 }
2335
trace__sched_stat_runtime(struct trace * trace,struct evsel * evsel,union perf_event * event __maybe_unused,struct perf_sample * sample)2336 static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
2337 union perf_event *event __maybe_unused,
2338 struct perf_sample *sample)
2339 {
2340 u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
2341 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2342 struct thread *thread = machine__findnew_thread(trace->host,
2343 sample->pid,
2344 sample->tid);
2345 struct thread_trace *ttrace = thread__trace(thread, trace->output);
2346
2347 if (ttrace == NULL)
2348 goto out_dump;
2349
2350 ttrace->runtime_ms += runtime_ms;
2351 trace->runtime_ms += runtime_ms;
2352 out_put:
2353 thread__put(thread);
2354 return 0;
2355
2356 out_dump:
2357 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2358 evsel->name,
2359 perf_evsel__strval(evsel, sample, "comm"),
2360 (pid_t)perf_evsel__intval(evsel, sample, "pid"),
2361 runtime,
2362 perf_evsel__intval(evsel, sample, "vruntime"));
2363 goto out_put;
2364 }
2365
bpf_output__printer(enum binary_printer_ops op,unsigned int val,void * extra __maybe_unused,FILE * fp)2366 static int bpf_output__printer(enum binary_printer_ops op,
2367 unsigned int val, void *extra __maybe_unused, FILE *fp)
2368 {
2369 unsigned char ch = (unsigned char)val;
2370
2371 switch (op) {
2372 case BINARY_PRINT_CHAR_DATA:
2373 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
2374 case BINARY_PRINT_DATA_BEGIN:
2375 case BINARY_PRINT_LINE_BEGIN:
2376 case BINARY_PRINT_ADDR:
2377 case BINARY_PRINT_NUM_DATA:
2378 case BINARY_PRINT_NUM_PAD:
2379 case BINARY_PRINT_SEP:
2380 case BINARY_PRINT_CHAR_PAD:
2381 case BINARY_PRINT_LINE_END:
2382 case BINARY_PRINT_DATA_END:
2383 default:
2384 break;
2385 }
2386
2387 return 0;
2388 }
2389
bpf_output__fprintf(struct trace * trace,struct perf_sample * sample)2390 static void bpf_output__fprintf(struct trace *trace,
2391 struct perf_sample *sample)
2392 {
2393 binary__fprintf(sample->raw_data, sample->raw_size, 8,
2394 bpf_output__printer, NULL, trace->output);
2395 ++trace->nr_events_printed;
2396 }
2397
trace__event_handler(struct trace * trace,struct evsel * evsel,union perf_event * event __maybe_unused,struct perf_sample * sample)2398 static int trace__event_handler(struct trace *trace, struct evsel *evsel,
2399 union perf_event *event __maybe_unused,
2400 struct perf_sample *sample)
2401 {
2402 struct thread *thread;
2403 int callchain_ret = 0;
2404 /*
2405 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2406 * this event's max_events having been hit and this is an entry coming
2407 * from the ring buffer that we should discard, since the max events
2408 * have already been considered/printed.
2409 */
2410 if (evsel->disabled)
2411 return 0;
2412
2413 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2414
2415 if (sample->callchain) {
2416 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2417 if (callchain_ret == 0) {
2418 if (callchain_cursor.nr < trace->min_stack)
2419 goto out;
2420 callchain_ret = 1;
2421 }
2422 }
2423
2424 trace__printf_interrupted_entry(trace);
2425 trace__fprintf_tstamp(trace, sample->time, trace->output);
2426
2427 if (trace->trace_syscalls && trace->show_duration)
2428 fprintf(trace->output, "( ): ");
2429
2430 if (thread)
2431 trace__fprintf_comm_tid(trace, thread, trace->output);
2432
2433 if (evsel == trace->syscalls.events.augmented) {
2434 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2435 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2436
2437 if (sc) {
2438 fprintf(trace->output, "%s(", sc->name);
2439 trace__fprintf_sys_enter(trace, evsel, sample);
2440 fputc(')', trace->output);
2441 goto newline;
2442 }
2443
2444 /*
2445 * XXX: Not having the associated syscall info or not finding/adding
2446 * the thread should never happen, but if it does...
2447 * fall thru and print it as a bpf_output event.
2448 */
2449 }
2450
2451 fprintf(trace->output, "%s:", evsel->name);
2452
2453 if (perf_evsel__is_bpf_output(evsel)) {
2454 bpf_output__fprintf(trace, sample);
2455 } else if (evsel->tp_format) {
2456 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2457 trace__fprintf_sys_enter(trace, evsel, sample)) {
2458 event_format__fprintf(evsel->tp_format, sample->cpu,
2459 sample->raw_data, sample->raw_size,
2460 trace->output);
2461 ++trace->nr_events_printed;
2462
2463 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2464 evsel__disable(evsel);
2465 evsel__close(evsel);
2466 }
2467 }
2468 }
2469
2470 newline:
2471 fprintf(trace->output, "\n");
2472
2473 if (callchain_ret > 0)
2474 trace__fprintf_callchain(trace, sample);
2475 else if (callchain_ret < 0)
2476 pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2477 out:
2478 thread__put(thread);
2479 return 0;
2480 }
2481
print_location(FILE * f,struct perf_sample * sample,struct addr_location * al,bool print_dso,bool print_sym)2482 static void print_location(FILE *f, struct perf_sample *sample,
2483 struct addr_location *al,
2484 bool print_dso, bool print_sym)
2485 {
2486
2487 if ((verbose > 0 || print_dso) && al->map)
2488 fprintf(f, "%s@", al->map->dso->long_name);
2489
2490 if ((verbose > 0 || print_sym) && al->sym)
2491 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
2492 al->addr - al->sym->start);
2493 else if (al->map)
2494 fprintf(f, "0x%" PRIx64, al->addr);
2495 else
2496 fprintf(f, "0x%" PRIx64, sample->addr);
2497 }
2498
trace__pgfault(struct trace * trace,struct evsel * evsel,union perf_event * event __maybe_unused,struct perf_sample * sample)2499 static int trace__pgfault(struct trace *trace,
2500 struct evsel *evsel,
2501 union perf_event *event __maybe_unused,
2502 struct perf_sample *sample)
2503 {
2504 struct thread *thread;
2505 struct addr_location al;
2506 char map_type = 'd';
2507 struct thread_trace *ttrace;
2508 int err = -1;
2509 int callchain_ret = 0;
2510
2511 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2512
2513 if (sample->callchain) {
2514 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2515 if (callchain_ret == 0) {
2516 if (callchain_cursor.nr < trace->min_stack)
2517 goto out_put;
2518 callchain_ret = 1;
2519 }
2520 }
2521
2522 ttrace = thread__trace(thread, trace->output);
2523 if (ttrace == NULL)
2524 goto out_put;
2525
2526 if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2527 ttrace->pfmaj++;
2528 else
2529 ttrace->pfmin++;
2530
2531 if (trace->summary_only)
2532 goto out;
2533
2534 thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
2535
2536 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
2537
2538 fprintf(trace->output, "%sfault [",
2539 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2540 "maj" : "min");
2541
2542 print_location(trace->output, sample, &al, false, true);
2543
2544 fprintf(trace->output, "] => ");
2545
2546 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2547
2548 if (!al.map) {
2549 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2550
2551 if (al.map)
2552 map_type = 'x';
2553 else
2554 map_type = '?';
2555 }
2556
2557 print_location(trace->output, sample, &al, true, false);
2558
2559 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
2560
2561 if (callchain_ret > 0)
2562 trace__fprintf_callchain(trace, sample);
2563 else if (callchain_ret < 0)
2564 pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2565
2566 ++trace->nr_events_printed;
2567 out:
2568 err = 0;
2569 out_put:
2570 thread__put(thread);
2571 return err;
2572 }
2573
trace__set_base_time(struct trace * trace,struct evsel * evsel,struct perf_sample * sample)2574 static void trace__set_base_time(struct trace *trace,
2575 struct evsel *evsel,
2576 struct perf_sample *sample)
2577 {
2578 /*
2579 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2580 * and don't use sample->time unconditionally, we may end up having
2581 * some other event in the future without PERF_SAMPLE_TIME for good
2582 * reason, i.e. we may not be interested in its timestamps, just in
2583 * it taking place, picking some piece of information when it
2584 * appears in our event stream (vfs_getname comes to mind).
2585 */
2586 if (trace->base_time == 0 && !trace->full_time &&
2587 (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
2588 trace->base_time = sample->time;
2589 }
2590
trace__process_sample(struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct evsel * evsel,struct machine * machine __maybe_unused)2591 static int trace__process_sample(struct perf_tool *tool,
2592 union perf_event *event,
2593 struct perf_sample *sample,
2594 struct evsel *evsel,
2595 struct machine *machine __maybe_unused)
2596 {
2597 struct trace *trace = container_of(tool, struct trace, tool);
2598 struct thread *thread;
2599 int err = 0;
2600
2601 tracepoint_handler handler = evsel->handler;
2602
2603 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2604 if (thread && thread__is_filtered(thread))
2605 goto out;
2606
2607 trace__set_base_time(trace, evsel, sample);
2608
2609 if (handler) {
2610 ++trace->nr_events;
2611 handler(trace, evsel, event, sample);
2612 }
2613 out:
2614 thread__put(thread);
2615 return err;
2616 }
2617
trace__record(struct trace * trace,int argc,const char ** argv)2618 static int trace__record(struct trace *trace, int argc, const char **argv)
2619 {
2620 unsigned int rec_argc, i, j;
2621 const char **rec_argv;
2622 const char * const record_args[] = {
2623 "record",
2624 "-R",
2625 "-m", "1024",
2626 "-c", "1",
2627 };
2628
2629 const char * const sc_args[] = { "-e", };
2630 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
2631 const char * const majpf_args[] = { "-e", "major-faults" };
2632 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
2633 const char * const minpf_args[] = { "-e", "minor-faults" };
2634 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
2635
2636 /* +1 is for the event string below */
2637 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 1 +
2638 majpf_args_nr + minpf_args_nr + argc;
2639 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2640
2641 if (rec_argv == NULL)
2642 return -ENOMEM;
2643
2644 j = 0;
2645 for (i = 0; i < ARRAY_SIZE(record_args); i++)
2646 rec_argv[j++] = record_args[i];
2647
2648 if (trace->trace_syscalls) {
2649 for (i = 0; i < sc_args_nr; i++)
2650 rec_argv[j++] = sc_args[i];
2651
2652 /* event string may be different for older kernels - e.g., RHEL6 */
2653 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
2654 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
2655 else if (is_valid_tracepoint("syscalls:sys_enter"))
2656 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
2657 else {
2658 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
2659 free(rec_argv);
2660 return -1;
2661 }
2662 }
2663
2664 if (trace->trace_pgfaults & TRACE_PFMAJ)
2665 for (i = 0; i < majpf_args_nr; i++)
2666 rec_argv[j++] = majpf_args[i];
2667
2668 if (trace->trace_pgfaults & TRACE_PFMIN)
2669 for (i = 0; i < minpf_args_nr; i++)
2670 rec_argv[j++] = minpf_args[i];
2671
2672 for (i = 0; i < (unsigned int)argc; i++)
2673 rec_argv[j++] = argv[i];
2674
2675 return cmd_record(j, rec_argv);
2676 }
2677
2678 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
2679
evlist__add_vfs_getname(struct evlist * evlist)2680 static bool evlist__add_vfs_getname(struct evlist *evlist)
2681 {
2682 bool found = false;
2683 struct evsel *evsel, *tmp;
2684 struct parse_events_error err = { .idx = 0, };
2685 int ret = parse_events(evlist, "probe:vfs_getname*", &err);
2686
2687 if (ret)
2688 return false;
2689
2690 evlist__for_each_entry_safe(evlist, evsel, tmp) {
2691 if (!strstarts(perf_evsel__name(evsel), "probe:vfs_getname"))
2692 continue;
2693
2694 if (perf_evsel__field(evsel, "pathname")) {
2695 evsel->handler = trace__vfs_getname;
2696 found = true;
2697 continue;
2698 }
2699
2700 list_del_init(&evsel->core.node);
2701 evsel->evlist = NULL;
2702 evsel__delete(evsel);
2703 }
2704
2705 return found;
2706 }
2707
perf_evsel__new_pgfault(u64 config)2708 static struct evsel *perf_evsel__new_pgfault(u64 config)
2709 {
2710 struct evsel *evsel;
2711 struct perf_event_attr attr = {
2712 .type = PERF_TYPE_SOFTWARE,
2713 .mmap_data = 1,
2714 };
2715
2716 attr.config = config;
2717 attr.sample_period = 1;
2718
2719 event_attr_init(&attr);
2720
2721 evsel = evsel__new(&attr);
2722 if (evsel)
2723 evsel->handler = trace__pgfault;
2724
2725 return evsel;
2726 }
2727
trace__handle_event(struct trace * trace,union perf_event * event,struct perf_sample * sample)2728 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
2729 {
2730 const u32 type = event->header.type;
2731 struct evsel *evsel;
2732
2733 if (type != PERF_RECORD_SAMPLE) {
2734 trace__process_event(trace, trace->host, event, sample);
2735 return;
2736 }
2737
2738 evsel = perf_evlist__id2evsel(trace->evlist, sample->id);
2739 if (evsel == NULL) {
2740 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
2741 return;
2742 }
2743
2744 if (evswitch__discard(&trace->evswitch, evsel))
2745 return;
2746
2747 trace__set_base_time(trace, evsel, sample);
2748
2749 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
2750 sample->raw_data == NULL) {
2751 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
2752 perf_evsel__name(evsel), sample->tid,
2753 sample->cpu, sample->raw_size);
2754 } else {
2755 tracepoint_handler handler = evsel->handler;
2756 handler(trace, evsel, event, sample);
2757 }
2758
2759 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
2760 interrupted = true;
2761 }
2762
trace__add_syscall_newtp(struct trace * trace)2763 static int trace__add_syscall_newtp(struct trace *trace)
2764 {
2765 int ret = -1;
2766 struct evlist *evlist = trace->evlist;
2767 struct evsel *sys_enter, *sys_exit;
2768
2769 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
2770 if (sys_enter == NULL)
2771 goto out;
2772
2773 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
2774 goto out_delete_sys_enter;
2775
2776 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
2777 if (sys_exit == NULL)
2778 goto out_delete_sys_enter;
2779
2780 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
2781 goto out_delete_sys_exit;
2782
2783 perf_evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
2784 perf_evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
2785
2786 evlist__add(evlist, sys_enter);
2787 evlist__add(evlist, sys_exit);
2788
2789 if (callchain_param.enabled && !trace->kernel_syscallchains) {
2790 /*
2791 * We're interested only in the user space callchain
2792 * leading to the syscall, allow overriding that for
2793 * debugging reasons using --kernel_syscall_callchains
2794 */
2795 sys_exit->core.attr.exclude_callchain_kernel = 1;
2796 }
2797
2798 trace->syscalls.events.sys_enter = sys_enter;
2799 trace->syscalls.events.sys_exit = sys_exit;
2800
2801 ret = 0;
2802 out:
2803 return ret;
2804
2805 out_delete_sys_exit:
2806 evsel__delete_priv(sys_exit);
2807 out_delete_sys_enter:
2808 evsel__delete_priv(sys_enter);
2809 goto out;
2810 }
2811
trace__set_ev_qualifier_tp_filter(struct trace * trace)2812 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
2813 {
2814 int err = -1;
2815 struct evsel *sys_exit;
2816 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
2817 trace->ev_qualifier_ids.nr,
2818 trace->ev_qualifier_ids.entries);
2819
2820 if (filter == NULL)
2821 goto out_enomem;
2822
2823 if (!perf_evsel__append_tp_filter(trace->syscalls.events.sys_enter,
2824 filter)) {
2825 sys_exit = trace->syscalls.events.sys_exit;
2826 err = perf_evsel__append_tp_filter(sys_exit, filter);
2827 }
2828
2829 free(filter);
2830 out:
2831 return err;
2832 out_enomem:
2833 errno = ENOMEM;
2834 goto out;
2835 }
2836
2837 #ifdef HAVE_LIBBPF_SUPPORT
trace__find_bpf_program_by_title(struct trace * trace,const char * name)2838 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
2839 {
2840 if (trace->bpf_obj == NULL)
2841 return NULL;
2842
2843 return bpf_object__find_program_by_title(trace->bpf_obj, name);
2844 }
2845
trace__find_syscall_bpf_prog(struct trace * trace,struct syscall * sc,const char * prog_name,const char * type)2846 static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
2847 const char *prog_name, const char *type)
2848 {
2849 struct bpf_program *prog;
2850
2851 if (prog_name == NULL) {
2852 char default_prog_name[256];
2853 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->name);
2854 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
2855 if (prog != NULL)
2856 goto out_found;
2857 if (sc->fmt && sc->fmt->alias) {
2858 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->fmt->alias);
2859 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
2860 if (prog != NULL)
2861 goto out_found;
2862 }
2863 goto out_unaugmented;
2864 }
2865
2866 prog = trace__find_bpf_program_by_title(trace, prog_name);
2867
2868 if (prog != NULL) {
2869 out_found:
2870 return prog;
2871 }
2872
2873 pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
2874 prog_name, type, sc->name);
2875 out_unaugmented:
2876 return trace->syscalls.unaugmented_prog;
2877 }
2878
trace__init_syscall_bpf_progs(struct trace * trace,int id)2879 static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
2880 {
2881 struct syscall *sc = trace__syscall_info(trace, NULL, id);
2882
2883 if (sc == NULL)
2884 return;
2885
2886 sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
2887 sc->bpf_prog.sys_exit = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit : NULL, "exit");
2888 }
2889
trace__bpf_prog_sys_enter_fd(struct trace * trace,int id)2890 static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
2891 {
2892 struct syscall *sc = trace__syscall_info(trace, NULL, id);
2893 return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->syscalls.unaugmented_prog);
2894 }
2895
trace__bpf_prog_sys_exit_fd(struct trace * trace,int id)2896 static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
2897 {
2898 struct syscall *sc = trace__syscall_info(trace, NULL, id);
2899 return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->syscalls.unaugmented_prog);
2900 }
2901
trace__init_bpf_map_syscall_args(struct trace * trace,int id,struct bpf_map_syscall_entry * entry)2902 static void trace__init_bpf_map_syscall_args(struct trace *trace, int id, struct bpf_map_syscall_entry *entry)
2903 {
2904 struct syscall *sc = trace__syscall_info(trace, NULL, id);
2905 int arg = 0;
2906
2907 if (sc == NULL)
2908 goto out;
2909
2910 for (; arg < sc->nr_args; ++arg) {
2911 entry->string_args_len[arg] = 0;
2912 if (sc->arg_fmt[arg].scnprintf == SCA_FILENAME) {
2913 /* Should be set like strace -s strsize */
2914 entry->string_args_len[arg] = PATH_MAX;
2915 }
2916 }
2917 out:
2918 for (; arg < 6; ++arg)
2919 entry->string_args_len[arg] = 0;
2920 }
trace__set_ev_qualifier_bpf_filter(struct trace * trace)2921 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace)
2922 {
2923 int fd = bpf_map__fd(trace->syscalls.map);
2924 struct bpf_map_syscall_entry value = {
2925 .enabled = !trace->not_ev_qualifier,
2926 };
2927 int err = 0;
2928 size_t i;
2929
2930 for (i = 0; i < trace->ev_qualifier_ids.nr; ++i) {
2931 int key = trace->ev_qualifier_ids.entries[i];
2932
2933 if (value.enabled) {
2934 trace__init_bpf_map_syscall_args(trace, key, &value);
2935 trace__init_syscall_bpf_progs(trace, key);
2936 }
2937
2938 err = bpf_map_update_elem(fd, &key, &value, BPF_EXIST);
2939 if (err)
2940 break;
2941 }
2942
2943 return err;
2944 }
2945
__trace__init_syscalls_bpf_map(struct trace * trace,bool enabled)2946 static int __trace__init_syscalls_bpf_map(struct trace *trace, bool enabled)
2947 {
2948 int fd = bpf_map__fd(trace->syscalls.map);
2949 struct bpf_map_syscall_entry value = {
2950 .enabled = enabled,
2951 };
2952 int err = 0, key;
2953
2954 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
2955 if (enabled)
2956 trace__init_bpf_map_syscall_args(trace, key, &value);
2957
2958 err = bpf_map_update_elem(fd, &key, &value, BPF_ANY);
2959 if (err)
2960 break;
2961 }
2962
2963 return err;
2964 }
2965
trace__init_syscalls_bpf_map(struct trace * trace)2966 static int trace__init_syscalls_bpf_map(struct trace *trace)
2967 {
2968 bool enabled = true;
2969
2970 if (trace->ev_qualifier_ids.nr)
2971 enabled = trace->not_ev_qualifier;
2972
2973 return __trace__init_syscalls_bpf_map(trace, enabled);
2974 }
2975
trace__find_usable_bpf_prog_entry(struct trace * trace,struct syscall * sc)2976 static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
2977 {
2978 struct tep_format_field *field, *candidate_field;
2979 int id;
2980
2981 /*
2982 * We're only interested in syscalls that have a pointer:
2983 */
2984 for (field = sc->args; field; field = field->next) {
2985 if (field->flags & TEP_FIELD_IS_POINTER)
2986 goto try_to_find_pair;
2987 }
2988
2989 return NULL;
2990
2991 try_to_find_pair:
2992 for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) {
2993 struct syscall *pair = trace__syscall_info(trace, NULL, id);
2994 struct bpf_program *pair_prog;
2995 bool is_candidate = false;
2996
2997 if (pair == NULL || pair == sc ||
2998 pair->bpf_prog.sys_enter == trace->syscalls.unaugmented_prog)
2999 continue;
3000
3001 for (field = sc->args, candidate_field = pair->args;
3002 field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3003 bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3004 candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3005
3006 if (is_pointer) {
3007 if (!candidate_is_pointer) {
3008 // The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3009 continue;
3010 }
3011 } else {
3012 if (candidate_is_pointer) {
3013 // The candidate might copy a pointer we don't have, skip it.
3014 goto next_candidate;
3015 }
3016 continue;
3017 }
3018
3019 if (strcmp(field->type, candidate_field->type))
3020 goto next_candidate;
3021
3022 is_candidate = true;
3023 }
3024
3025 if (!is_candidate)
3026 goto next_candidate;
3027
3028 /*
3029 * Check if the tentative pair syscall augmenter has more pointers, if it has,
3030 * then it may be collecting that and we then can't use it, as it would collect
3031 * more than what is common to the two syscalls.
3032 */
3033 if (candidate_field) {
3034 for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3035 if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3036 goto next_candidate;
3037 }
3038
3039 pair_prog = pair->bpf_prog.sys_enter;
3040 /*
3041 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3042 * have been searched for, so search it here and if it returns the
3043 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3044 * program for a filtered syscall on a non-filtered one.
3045 *
3046 * For instance, we have "!syscalls:sys_enter_renameat" and that is
3047 * useful for "renameat2".
3048 */
3049 if (pair_prog == NULL) {
3050 pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3051 if (pair_prog == trace->syscalls.unaugmented_prog)
3052 goto next_candidate;
3053 }
3054
3055 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3056 return pair_prog;
3057 next_candidate:
3058 continue;
3059 }
3060
3061 return NULL;
3062 }
3063
trace__init_syscalls_bpf_prog_array_maps(struct trace * trace)3064 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3065 {
3066 int map_enter_fd = bpf_map__fd(trace->syscalls.prog_array.sys_enter),
3067 map_exit_fd = bpf_map__fd(trace->syscalls.prog_array.sys_exit);
3068 int err = 0, key;
3069
3070 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3071 int prog_fd;
3072
3073 if (!trace__syscall_enabled(trace, key))
3074 continue;
3075
3076 trace__init_syscall_bpf_progs(trace, key);
3077
3078 // It'll get at least the "!raw_syscalls:unaugmented"
3079 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3080 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3081 if (err)
3082 break;
3083 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3084 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3085 if (err)
3086 break;
3087 }
3088
3089 /*
3090 * Now lets do a second pass looking for enabled syscalls without
3091 * an augmenter that have a signature that is a superset of another
3092 * syscall with an augmenter so that we can auto-reuse it.
3093 *
3094 * I.e. if we have an augmenter for the "open" syscall that has
3095 * this signature:
3096 *
3097 * int open(const char *pathname, int flags, mode_t mode);
3098 *
3099 * I.e. that will collect just the first string argument, then we
3100 * can reuse it for the 'creat' syscall, that has this signature:
3101 *
3102 * int creat(const char *pathname, mode_t mode);
3103 *
3104 * and for:
3105 *
3106 * int stat(const char *pathname, struct stat *statbuf);
3107 * int lstat(const char *pathname, struct stat *statbuf);
3108 *
3109 * Because the 'open' augmenter will collect the first arg as a string,
3110 * and leave alone all the other args, which already helps with
3111 * beautifying 'stat' and 'lstat''s pathname arg.
3112 *
3113 * Then, in time, when 'stat' gets an augmenter that collects both
3114 * first and second arg (this one on the raw_syscalls:sys_exit prog
3115 * array tail call, then that one will be used.
3116 */
3117 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3118 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3119 struct bpf_program *pair_prog;
3120 int prog_fd;
3121
3122 if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3123 continue;
3124
3125 /*
3126 * For now we're just reusing the sys_enter prog, and if it
3127 * already has an augmenter, we don't need to find one.
3128 */
3129 if (sc->bpf_prog.sys_enter != trace->syscalls.unaugmented_prog)
3130 continue;
3131
3132 /*
3133 * Look at all the other syscalls for one that has a signature
3134 * that is close enough that we can share:
3135 */
3136 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3137 if (pair_prog == NULL)
3138 continue;
3139
3140 sc->bpf_prog.sys_enter = pair_prog;
3141
3142 /*
3143 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3144 * with the fd for the program we're reusing:
3145 */
3146 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3147 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3148 if (err)
3149 break;
3150 }
3151
3152
3153 return err;
3154 }
3155 #else
trace__set_ev_qualifier_bpf_filter(struct trace * trace __maybe_unused)3156 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace __maybe_unused)
3157 {
3158 return 0;
3159 }
3160
trace__init_syscalls_bpf_map(struct trace * trace __maybe_unused)3161 static int trace__init_syscalls_bpf_map(struct trace *trace __maybe_unused)
3162 {
3163 return 0;
3164 }
3165
trace__find_bpf_program_by_title(struct trace * trace __maybe_unused,const char * name __maybe_unused)3166 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace __maybe_unused,
3167 const char *name __maybe_unused)
3168 {
3169 return NULL;
3170 }
3171
trace__init_syscalls_bpf_prog_array_maps(struct trace * trace __maybe_unused)3172 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace __maybe_unused)
3173 {
3174 return 0;
3175 }
3176 #endif // HAVE_LIBBPF_SUPPORT
3177
trace__set_ev_qualifier_filter(struct trace * trace)3178 static int trace__set_ev_qualifier_filter(struct trace *trace)
3179 {
3180 if (trace->syscalls.map)
3181 return trace__set_ev_qualifier_bpf_filter(trace);
3182 if (trace->syscalls.events.sys_enter)
3183 return trace__set_ev_qualifier_tp_filter(trace);
3184 return 0;
3185 }
3186
bpf_map__set_filter_pids(struct bpf_map * map __maybe_unused,size_t npids __maybe_unused,pid_t * pids __maybe_unused)3187 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3188 size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3189 {
3190 int err = 0;
3191 #ifdef HAVE_LIBBPF_SUPPORT
3192 bool value = true;
3193 int map_fd = bpf_map__fd(map);
3194 size_t i;
3195
3196 for (i = 0; i < npids; ++i) {
3197 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3198 if (err)
3199 break;
3200 }
3201 #endif
3202 return err;
3203 }
3204
trace__set_filter_loop_pids(struct trace * trace)3205 static int trace__set_filter_loop_pids(struct trace *trace)
3206 {
3207 unsigned int nr = 1, err;
3208 pid_t pids[32] = {
3209 getpid(),
3210 };
3211 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3212
3213 while (thread && nr < ARRAY_SIZE(pids)) {
3214 struct thread *parent = machine__find_thread(trace->host, thread->ppid, thread->ppid);
3215
3216 if (parent == NULL)
3217 break;
3218
3219 if (!strcmp(thread__comm_str(parent), "sshd") ||
3220 strstarts(thread__comm_str(parent), "gnome-terminal")) {
3221 pids[nr++] = parent->tid;
3222 break;
3223 }
3224 thread = parent;
3225 }
3226
3227 err = perf_evlist__set_tp_filter_pids(trace->evlist, nr, pids);
3228 if (!err && trace->filter_pids.map)
3229 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3230
3231 return err;
3232 }
3233
trace__set_filter_pids(struct trace * trace)3234 static int trace__set_filter_pids(struct trace *trace)
3235 {
3236 int err = 0;
3237 /*
3238 * Better not use !target__has_task() here because we need to cover the
3239 * case where no threads were specified in the command line, but a
3240 * workload was, and in that case we will fill in the thread_map when
3241 * we fork the workload in perf_evlist__prepare_workload.
3242 */
3243 if (trace->filter_pids.nr > 0) {
3244 err = perf_evlist__set_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3245 trace->filter_pids.entries);
3246 if (!err && trace->filter_pids.map) {
3247 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3248 trace->filter_pids.entries);
3249 }
3250 } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
3251 err = trace__set_filter_loop_pids(trace);
3252 }
3253
3254 return err;
3255 }
3256
__trace__deliver_event(struct trace * trace,union perf_event * event)3257 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
3258 {
3259 struct evlist *evlist = trace->evlist;
3260 struct perf_sample sample;
3261 int err;
3262
3263 err = perf_evlist__parse_sample(evlist, event, &sample);
3264 if (err)
3265 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3266 else
3267 trace__handle_event(trace, event, &sample);
3268
3269 return 0;
3270 }
3271
__trace__flush_events(struct trace * trace)3272 static int __trace__flush_events(struct trace *trace)
3273 {
3274 u64 first = ordered_events__first_time(&trace->oe.data);
3275 u64 flush = trace->oe.last - NSEC_PER_SEC;
3276
3277 /* Is there some thing to flush.. */
3278 if (first && first < flush)
3279 return ordered_events__flush_time(&trace->oe.data, flush);
3280
3281 return 0;
3282 }
3283
trace__flush_events(struct trace * trace)3284 static int trace__flush_events(struct trace *trace)
3285 {
3286 return !trace->sort_events ? 0 : __trace__flush_events(trace);
3287 }
3288
trace__deliver_event(struct trace * trace,union perf_event * event)3289 static int trace__deliver_event(struct trace *trace, union perf_event *event)
3290 {
3291 int err;
3292
3293 if (!trace->sort_events)
3294 return __trace__deliver_event(trace, event);
3295
3296 err = perf_evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
3297 if (err && err != -1)
3298 return err;
3299
3300 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0);
3301 if (err)
3302 return err;
3303
3304 return trace__flush_events(trace);
3305 }
3306
ordered_events__deliver_event(struct ordered_events * oe,struct ordered_event * event)3307 static int ordered_events__deliver_event(struct ordered_events *oe,
3308 struct ordered_event *event)
3309 {
3310 struct trace *trace = container_of(oe, struct trace, oe.data);
3311
3312 return __trace__deliver_event(trace, event->event);
3313 }
3314
trace__run(struct trace * trace,int argc,const char ** argv)3315 static int trace__run(struct trace *trace, int argc, const char **argv)
3316 {
3317 struct evlist *evlist = trace->evlist;
3318 struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
3319 int err = -1, i;
3320 unsigned long before;
3321 const bool forks = argc > 0;
3322 bool draining = false;
3323
3324 trace->live = true;
3325
3326 if (!trace->raw_augmented_syscalls) {
3327 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3328 goto out_error_raw_syscalls;
3329
3330 if (trace->trace_syscalls)
3331 trace->vfs_getname = evlist__add_vfs_getname(evlist);
3332 }
3333
3334 if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
3335 pgfault_maj = perf_evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
3336 if (pgfault_maj == NULL)
3337 goto out_error_mem;
3338 perf_evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
3339 evlist__add(evlist, pgfault_maj);
3340 }
3341
3342 if ((trace->trace_pgfaults & TRACE_PFMIN)) {
3343 pgfault_min = perf_evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
3344 if (pgfault_min == NULL)
3345 goto out_error_mem;
3346 perf_evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
3347 evlist__add(evlist, pgfault_min);
3348 }
3349
3350 if (trace->sched &&
3351 perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
3352 trace__sched_stat_runtime))
3353 goto out_error_sched_stat_runtime;
3354
3355 /*
3356 * If a global cgroup was set, apply it to all the events without an
3357 * explicit cgroup. I.e.:
3358 *
3359 * trace -G A -e sched:*switch
3360 *
3361 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3362 * _and_ sched:sched_switch to the 'A' cgroup, while:
3363 *
3364 * trace -e sched:*switch -G A
3365 *
3366 * will only set the sched:sched_switch event to the 'A' cgroup, all the
3367 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3368 * a cgroup (on the root cgroup, sys wide, etc).
3369 *
3370 * Multiple cgroups:
3371 *
3372 * trace -G A -e sched:*switch -G B
3373 *
3374 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
3375 * to the 'B' cgroup.
3376 *
3377 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
3378 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
3379 */
3380 if (trace->cgroup)
3381 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
3382
3383 err = perf_evlist__create_maps(evlist, &trace->opts.target);
3384 if (err < 0) {
3385 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
3386 goto out_delete_evlist;
3387 }
3388
3389 err = trace__symbols_init(trace, evlist);
3390 if (err < 0) {
3391 fprintf(trace->output, "Problems initializing symbol libraries!\n");
3392 goto out_delete_evlist;
3393 }
3394
3395 perf_evlist__config(evlist, &trace->opts, &callchain_param);
3396
3397 signal(SIGCHLD, sig_handler);
3398 signal(SIGINT, sig_handler);
3399
3400 if (forks) {
3401 err = perf_evlist__prepare_workload(evlist, &trace->opts.target,
3402 argv, false, NULL);
3403 if (err < 0) {
3404 fprintf(trace->output, "Couldn't run the workload!\n");
3405 goto out_delete_evlist;
3406 }
3407 }
3408
3409 err = evlist__open(evlist);
3410 if (err < 0)
3411 goto out_error_open;
3412
3413 err = bpf__apply_obj_config();
3414 if (err) {
3415 char errbuf[BUFSIZ];
3416
3417 bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
3418 pr_err("ERROR: Apply config to BPF failed: %s\n",
3419 errbuf);
3420 goto out_error_open;
3421 }
3422
3423 err = trace__set_filter_pids(trace);
3424 if (err < 0)
3425 goto out_error_mem;
3426
3427 if (trace->syscalls.map)
3428 trace__init_syscalls_bpf_map(trace);
3429
3430 if (trace->syscalls.prog_array.sys_enter)
3431 trace__init_syscalls_bpf_prog_array_maps(trace);
3432
3433 if (trace->ev_qualifier_ids.nr > 0) {
3434 err = trace__set_ev_qualifier_filter(trace);
3435 if (err < 0)
3436 goto out_errno;
3437
3438 if (trace->syscalls.events.sys_exit) {
3439 pr_debug("event qualifier tracepoint filter: %s\n",
3440 trace->syscalls.events.sys_exit->filter);
3441 }
3442 }
3443
3444 /*
3445 * If the "close" syscall is not traced, then we will not have the
3446 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
3447 * fd->pathname table and were ending up showing the last value set by
3448 * syscalls opening a pathname and associating it with a descriptor or
3449 * reading it from /proc/pid/fd/ in cases where that doesn't make
3450 * sense.
3451 *
3452 * So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
3453 * not in use.
3454 */
3455 trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
3456
3457 err = perf_evlist__apply_filters(evlist, &evsel);
3458 if (err < 0)
3459 goto out_error_apply_filters;
3460
3461 if (trace->dump.map)
3462 bpf_map__fprintf(trace->dump.map, trace->output);
3463
3464 err = evlist__mmap(evlist, trace->opts.mmap_pages);
3465 if (err < 0)
3466 goto out_error_mmap;
3467
3468 if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
3469 evlist__enable(evlist);
3470
3471 if (forks)
3472 perf_evlist__start_workload(evlist);
3473
3474 if (trace->opts.initial_delay) {
3475 usleep(trace->opts.initial_delay * 1000);
3476 evlist__enable(evlist);
3477 }
3478
3479 trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
3480 evlist->core.threads->nr > 1 ||
3481 evlist__first(evlist)->core.attr.inherit;
3482
3483 /*
3484 * Now that we already used evsel->core.attr to ask the kernel to setup the
3485 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
3486 * trace__resolve_callchain(), allowing per-event max-stack settings
3487 * to override an explicitly set --max-stack global setting.
3488 */
3489 evlist__for_each_entry(evlist, evsel) {
3490 if (evsel__has_callchain(evsel) &&
3491 evsel->core.attr.sample_max_stack == 0)
3492 evsel->core.attr.sample_max_stack = trace->max_stack;
3493 }
3494 again:
3495 before = trace->nr_events;
3496
3497 for (i = 0; i < evlist->core.nr_mmaps; i++) {
3498 union perf_event *event;
3499 struct mmap *md;
3500
3501 md = &evlist->mmap[i];
3502 if (perf_mmap__read_init(md) < 0)
3503 continue;
3504
3505 while ((event = perf_mmap__read_event(md)) != NULL) {
3506 ++trace->nr_events;
3507
3508 err = trace__deliver_event(trace, event);
3509 if (err)
3510 goto out_disable;
3511
3512 perf_mmap__consume(md);
3513
3514 if (interrupted)
3515 goto out_disable;
3516
3517 if (done && !draining) {
3518 evlist__disable(evlist);
3519 draining = true;
3520 }
3521 }
3522 perf_mmap__read_done(md);
3523 }
3524
3525 if (trace->nr_events == before) {
3526 int timeout = done ? 100 : -1;
3527
3528 if (!draining && evlist__poll(evlist, timeout) > 0) {
3529 if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
3530 draining = true;
3531
3532 goto again;
3533 } else {
3534 if (trace__flush_events(trace))
3535 goto out_disable;
3536 }
3537 } else {
3538 goto again;
3539 }
3540
3541 out_disable:
3542 thread__zput(trace->current);
3543
3544 evlist__disable(evlist);
3545
3546 if (trace->sort_events)
3547 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
3548
3549 if (!err) {
3550 if (trace->summary)
3551 trace__fprintf_thread_summary(trace, trace->output);
3552
3553 if (trace->show_tool_stats) {
3554 fprintf(trace->output, "Stats:\n "
3555 " vfs_getname : %" PRIu64 "\n"
3556 " proc_getname: %" PRIu64 "\n",
3557 trace->stats.vfs_getname,
3558 trace->stats.proc_getname);
3559 }
3560 }
3561
3562 out_delete_evlist:
3563 trace__symbols__exit(trace);
3564
3565 evlist__delete(evlist);
3566 cgroup__put(trace->cgroup);
3567 trace->evlist = NULL;
3568 trace->live = false;
3569 return err;
3570 {
3571 char errbuf[BUFSIZ];
3572
3573 out_error_sched_stat_runtime:
3574 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
3575 goto out_error;
3576
3577 out_error_raw_syscalls:
3578 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
3579 goto out_error;
3580
3581 out_error_mmap:
3582 perf_evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
3583 goto out_error;
3584
3585 out_error_open:
3586 perf_evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
3587
3588 out_error:
3589 fprintf(trace->output, "%s\n", errbuf);
3590 goto out_delete_evlist;
3591
3592 out_error_apply_filters:
3593 fprintf(trace->output,
3594 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
3595 evsel->filter, perf_evsel__name(evsel), errno,
3596 str_error_r(errno, errbuf, sizeof(errbuf)));
3597 goto out_delete_evlist;
3598 }
3599 out_error_mem:
3600 fprintf(trace->output, "Not enough memory to run!\n");
3601 goto out_delete_evlist;
3602
3603 out_errno:
3604 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
3605 goto out_delete_evlist;
3606 }
3607
trace__replay(struct trace * trace)3608 static int trace__replay(struct trace *trace)
3609 {
3610 const struct evsel_str_handler handlers[] = {
3611 { "probe:vfs_getname", trace__vfs_getname, },
3612 };
3613 struct perf_data data = {
3614 .path = input_name,
3615 .mode = PERF_DATA_MODE_READ,
3616 .force = trace->force,
3617 };
3618 struct perf_session *session;
3619 struct evsel *evsel;
3620 int err = -1;
3621
3622 trace->tool.sample = trace__process_sample;
3623 trace->tool.mmap = perf_event__process_mmap;
3624 trace->tool.mmap2 = perf_event__process_mmap2;
3625 trace->tool.comm = perf_event__process_comm;
3626 trace->tool.exit = perf_event__process_exit;
3627 trace->tool.fork = perf_event__process_fork;
3628 trace->tool.attr = perf_event__process_attr;
3629 trace->tool.tracing_data = perf_event__process_tracing_data;
3630 trace->tool.build_id = perf_event__process_build_id;
3631 trace->tool.namespaces = perf_event__process_namespaces;
3632
3633 trace->tool.ordered_events = true;
3634 trace->tool.ordering_requires_timestamps = true;
3635
3636 /* add tid to output */
3637 trace->multiple_threads = true;
3638
3639 session = perf_session__new(&data, false, &trace->tool);
3640 if (IS_ERR(session))
3641 return PTR_ERR(session);
3642
3643 if (trace->opts.target.pid)
3644 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
3645
3646 if (trace->opts.target.tid)
3647 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
3648
3649 if (symbol__init(&session->header.env) < 0)
3650 goto out;
3651
3652 trace->host = &session->machines.host;
3653
3654 err = perf_session__set_tracepoints_handlers(session, handlers);
3655 if (err)
3656 goto out;
3657
3658 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3659 "raw_syscalls:sys_enter");
3660 /* older kernels have syscalls tp versus raw_syscalls */
3661 if (evsel == NULL)
3662 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3663 "syscalls:sys_enter");
3664
3665 if (evsel &&
3666 (perf_evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
3667 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
3668 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
3669 goto out;
3670 }
3671
3672 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3673 "raw_syscalls:sys_exit");
3674 if (evsel == NULL)
3675 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3676 "syscalls:sys_exit");
3677 if (evsel &&
3678 (perf_evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
3679 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
3680 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
3681 goto out;
3682 }
3683
3684 evlist__for_each_entry(session->evlist, evsel) {
3685 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
3686 (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
3687 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
3688 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
3689 evsel->handler = trace__pgfault;
3690 }
3691
3692 setup_pager();
3693
3694 err = perf_session__process_events(session);
3695 if (err)
3696 pr_err("Failed to process events, error %d", err);
3697
3698 else if (trace->summary)
3699 trace__fprintf_thread_summary(trace, trace->output);
3700
3701 out:
3702 perf_session__delete(session);
3703
3704 return err;
3705 }
3706
trace__fprintf_threads_header(FILE * fp)3707 static size_t trace__fprintf_threads_header(FILE *fp)
3708 {
3709 size_t printed;
3710
3711 printed = fprintf(fp, "\n Summary of events:\n\n");
3712
3713 return printed;
3714 }
3715
3716 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
3717 struct stats *stats;
3718 double msecs;
3719 int syscall;
3720 )
3721 {
3722 struct int_node *source = rb_entry(nd, struct int_node, rb_node);
3723 struct stats *stats = source->priv;
3724
3725 entry->syscall = source->i;
3726 entry->stats = stats;
3727 entry->msecs = stats ? (u64)stats->n * (avg_stats(stats) / NSEC_PER_MSEC) : 0;
3728 }
3729
thread__dump_stats(struct thread_trace * ttrace,struct trace * trace,FILE * fp)3730 static size_t thread__dump_stats(struct thread_trace *ttrace,
3731 struct trace *trace, FILE *fp)
3732 {
3733 size_t printed = 0;
3734 struct syscall *sc;
3735 struct rb_node *nd;
3736 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
3737
3738 if (syscall_stats == NULL)
3739 return 0;
3740
3741 printed += fprintf(fp, "\n");
3742
3743 printed += fprintf(fp, " syscall calls total min avg max stddev\n");
3744 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
3745 printed += fprintf(fp, " --------------- -------- --------- --------- --------- --------- ------\n");
3746
3747 resort_rb__for_each_entry(nd, syscall_stats) {
3748 struct stats *stats = syscall_stats_entry->stats;
3749 if (stats) {
3750 double min = (double)(stats->min) / NSEC_PER_MSEC;
3751 double max = (double)(stats->max) / NSEC_PER_MSEC;
3752 double avg = avg_stats(stats);
3753 double pct;
3754 u64 n = (u64) stats->n;
3755
3756 pct = avg ? 100.0 * stddev_stats(stats)/avg : 0.0;
3757 avg /= NSEC_PER_MSEC;
3758
3759 sc = &trace->syscalls.table[syscall_stats_entry->syscall];
3760 printed += fprintf(fp, " %-15s", sc->name);
3761 printed += fprintf(fp, " %8" PRIu64 " %9.3f %9.3f %9.3f",
3762 n, syscall_stats_entry->msecs, min, avg);
3763 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
3764 }
3765 }
3766
3767 resort_rb__delete(syscall_stats);
3768 printed += fprintf(fp, "\n\n");
3769
3770 return printed;
3771 }
3772
trace__fprintf_thread(FILE * fp,struct thread * thread,struct trace * trace)3773 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
3774 {
3775 size_t printed = 0;
3776 struct thread_trace *ttrace = thread__priv(thread);
3777 double ratio;
3778
3779 if (ttrace == NULL)
3780 return 0;
3781
3782 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
3783
3784 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
3785 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
3786 printed += fprintf(fp, "%.1f%%", ratio);
3787 if (ttrace->pfmaj)
3788 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
3789 if (ttrace->pfmin)
3790 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
3791 if (trace->sched)
3792 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
3793 else if (fputc('\n', fp) != EOF)
3794 ++printed;
3795
3796 printed += thread__dump_stats(ttrace, trace, fp);
3797
3798 return printed;
3799 }
3800
thread__nr_events(struct thread_trace * ttrace)3801 static unsigned long thread__nr_events(struct thread_trace *ttrace)
3802 {
3803 return ttrace ? ttrace->nr_events : 0;
3804 }
3805
3806 DEFINE_RESORT_RB(threads, (thread__nr_events(a->thread->priv) < thread__nr_events(b->thread->priv)),
3807 struct thread *thread;
3808 )
3809 {
3810 entry->thread = rb_entry(nd, struct thread, rb_node);
3811 }
3812
trace__fprintf_thread_summary(struct trace * trace,FILE * fp)3813 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
3814 {
3815 size_t printed = trace__fprintf_threads_header(fp);
3816 struct rb_node *nd;
3817 int i;
3818
3819 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
3820 DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
3821
3822 if (threads == NULL) {
3823 fprintf(fp, "%s", "Error sorting output by nr_events!\n");
3824 return 0;
3825 }
3826
3827 resort_rb__for_each_entry(nd, threads)
3828 printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
3829
3830 resort_rb__delete(threads);
3831 }
3832 return printed;
3833 }
3834
trace__set_duration(const struct option * opt,const char * str,int unset __maybe_unused)3835 static int trace__set_duration(const struct option *opt, const char *str,
3836 int unset __maybe_unused)
3837 {
3838 struct trace *trace = opt->value;
3839
3840 trace->duration_filter = atof(str);
3841 return 0;
3842 }
3843
trace__set_filter_pids_from_option(const struct option * opt,const char * str,int unset __maybe_unused)3844 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
3845 int unset __maybe_unused)
3846 {
3847 int ret = -1;
3848 size_t i;
3849 struct trace *trace = opt->value;
3850 /*
3851 * FIXME: introduce a intarray class, plain parse csv and create a
3852 * { int nr, int entries[] } struct...
3853 */
3854 struct intlist *list = intlist__new(str);
3855
3856 if (list == NULL)
3857 return -1;
3858
3859 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
3860 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
3861
3862 if (trace->filter_pids.entries == NULL)
3863 goto out;
3864
3865 trace->filter_pids.entries[0] = getpid();
3866
3867 for (i = 1; i < trace->filter_pids.nr; ++i)
3868 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
3869
3870 intlist__delete(list);
3871 ret = 0;
3872 out:
3873 return ret;
3874 }
3875
trace__open_output(struct trace * trace,const char * filename)3876 static int trace__open_output(struct trace *trace, const char *filename)
3877 {
3878 struct stat st;
3879
3880 if (!stat(filename, &st) && st.st_size) {
3881 char oldname[PATH_MAX];
3882
3883 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
3884 unlink(oldname);
3885 rename(filename, oldname);
3886 }
3887
3888 trace->output = fopen(filename, "w");
3889
3890 return trace->output == NULL ? -errno : 0;
3891 }
3892
parse_pagefaults(const struct option * opt,const char * str,int unset __maybe_unused)3893 static int parse_pagefaults(const struct option *opt, const char *str,
3894 int unset __maybe_unused)
3895 {
3896 int *trace_pgfaults = opt->value;
3897
3898 if (strcmp(str, "all") == 0)
3899 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
3900 else if (strcmp(str, "maj") == 0)
3901 *trace_pgfaults |= TRACE_PFMAJ;
3902 else if (strcmp(str, "min") == 0)
3903 *trace_pgfaults |= TRACE_PFMIN;
3904 else
3905 return -1;
3906
3907 return 0;
3908 }
3909
evlist__set_evsel_handler(struct evlist * evlist,void * handler)3910 static void evlist__set_evsel_handler(struct evlist *evlist, void *handler)
3911 {
3912 struct evsel *evsel;
3913
3914 evlist__for_each_entry(evlist, evsel)
3915 evsel->handler = handler;
3916 }
3917
evlist__set_syscall_tp_fields(struct evlist * evlist)3918 static int evlist__set_syscall_tp_fields(struct evlist *evlist)
3919 {
3920 struct evsel *evsel;
3921
3922 evlist__for_each_entry(evlist, evsel) {
3923 if (evsel->priv || !evsel->tp_format)
3924 continue;
3925
3926 if (strcmp(evsel->tp_format->system, "syscalls"))
3927 continue;
3928
3929 if (perf_evsel__init_syscall_tp(evsel))
3930 return -1;
3931
3932 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
3933 struct syscall_tp *sc = evsel->priv;
3934
3935 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
3936 return -1;
3937 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
3938 struct syscall_tp *sc = evsel->priv;
3939
3940 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
3941 return -1;
3942 }
3943 }
3944
3945 return 0;
3946 }
3947
3948 /*
3949 * XXX: Hackish, just splitting the combined -e+--event (syscalls
3950 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
3951 * existing facilities unchanged (trace->ev_qualifier + parse_options()).
3952 *
3953 * It'd be better to introduce a parse_options() variant that would return a
3954 * list with the terms it didn't match to an event...
3955 */
trace__parse_events_option(const struct option * opt,const char * str,int unset __maybe_unused)3956 static int trace__parse_events_option(const struct option *opt, const char *str,
3957 int unset __maybe_unused)
3958 {
3959 struct trace *trace = (struct trace *)opt->value;
3960 const char *s = str;
3961 char *sep = NULL, *lists[2] = { NULL, NULL, };
3962 int len = strlen(str) + 1, err = -1, list, idx;
3963 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
3964 char group_name[PATH_MAX];
3965 struct syscall_fmt *fmt;
3966
3967 if (strace_groups_dir == NULL)
3968 return -1;
3969
3970 if (*s == '!') {
3971 ++s;
3972 trace->not_ev_qualifier = true;
3973 }
3974
3975 while (1) {
3976 if ((sep = strchr(s, ',')) != NULL)
3977 *sep = '\0';
3978
3979 list = 0;
3980 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
3981 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
3982 list = 1;
3983 goto do_concat;
3984 }
3985
3986 fmt = syscall_fmt__find_by_alias(s);
3987 if (fmt != NULL) {
3988 list = 1;
3989 s = fmt->name;
3990 } else {
3991 path__join(group_name, sizeof(group_name), strace_groups_dir, s);
3992 if (access(group_name, R_OK) == 0)
3993 list = 1;
3994 }
3995 do_concat:
3996 if (lists[list]) {
3997 sprintf(lists[list] + strlen(lists[list]), ",%s", s);
3998 } else {
3999 lists[list] = malloc(len);
4000 if (lists[list] == NULL)
4001 goto out;
4002 strcpy(lists[list], s);
4003 }
4004
4005 if (!sep)
4006 break;
4007
4008 *sep = ',';
4009 s = sep + 1;
4010 }
4011
4012 if (lists[1] != NULL) {
4013 struct strlist_config slist_config = {
4014 .dirname = strace_groups_dir,
4015 };
4016
4017 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4018 if (trace->ev_qualifier == NULL) {
4019 fputs("Not enough memory to parse event qualifier", trace->output);
4020 goto out;
4021 }
4022
4023 if (trace__validate_ev_qualifier(trace))
4024 goto out;
4025 trace->trace_syscalls = true;
4026 }
4027
4028 err = 0;
4029
4030 if (lists[0]) {
4031 struct option o = {
4032 .value = &trace->evlist,
4033 };
4034 err = parse_events_option(&o, lists[0], 0);
4035 }
4036 out:
4037 if (sep)
4038 *sep = ',';
4039
4040 return err;
4041 }
4042
trace__parse_cgroups(const struct option * opt,const char * str,int unset)4043 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4044 {
4045 struct trace *trace = opt->value;
4046
4047 if (!list_empty(&trace->evlist->core.entries)) {
4048 struct option o = {
4049 .value = &trace->evlist,
4050 };
4051 return parse_cgroups(&o, str, unset);
4052 }
4053 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
4054
4055 return 0;
4056 }
4057
trace__find_bpf_map_by_name(struct trace * trace,const char * name)4058 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace, const char *name)
4059 {
4060 if (trace->bpf_obj == NULL)
4061 return NULL;
4062
4063 return bpf_object__find_map_by_name(trace->bpf_obj, name);
4064 }
4065
trace__set_bpf_map_filtered_pids(struct trace * trace)4066 static void trace__set_bpf_map_filtered_pids(struct trace *trace)
4067 {
4068 trace->filter_pids.map = trace__find_bpf_map_by_name(trace, "pids_filtered");
4069 }
4070
trace__set_bpf_map_syscalls(struct trace * trace)4071 static void trace__set_bpf_map_syscalls(struct trace *trace)
4072 {
4073 trace->syscalls.map = trace__find_bpf_map_by_name(trace, "syscalls");
4074 trace->syscalls.prog_array.sys_enter = trace__find_bpf_map_by_name(trace, "syscalls_sys_enter");
4075 trace->syscalls.prog_array.sys_exit = trace__find_bpf_map_by_name(trace, "syscalls_sys_exit");
4076 }
4077
trace__config(const char * var,const char * value,void * arg)4078 static int trace__config(const char *var, const char *value, void *arg)
4079 {
4080 struct trace *trace = arg;
4081 int err = 0;
4082
4083 if (!strcmp(var, "trace.add_events")) {
4084 struct option o = OPT_CALLBACK('e', "event", &trace->evlist, "event",
4085 "event selector. use 'perf list' to list available events",
4086 parse_events_option);
4087 /*
4088 * We can't propagate parse_event_option() return, as it is 1
4089 * for failure while perf_config() expects -1.
4090 */
4091 if (parse_events_option(&o, value, 0))
4092 err = -1;
4093 } else if (!strcmp(var, "trace.show_timestamp")) {
4094 trace->show_tstamp = perf_config_bool(var, value);
4095 } else if (!strcmp(var, "trace.show_duration")) {
4096 trace->show_duration = perf_config_bool(var, value);
4097 } else if (!strcmp(var, "trace.show_arg_names")) {
4098 trace->show_arg_names = perf_config_bool(var, value);
4099 if (!trace->show_arg_names)
4100 trace->show_zeros = true;
4101 } else if (!strcmp(var, "trace.show_zeros")) {
4102 bool new_show_zeros = perf_config_bool(var, value);
4103 if (!trace->show_arg_names && !new_show_zeros) {
4104 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4105 goto out;
4106 }
4107 trace->show_zeros = new_show_zeros;
4108 } else if (!strcmp(var, "trace.show_prefix")) {
4109 trace->show_string_prefix = perf_config_bool(var, value);
4110 } else if (!strcmp(var, "trace.no_inherit")) {
4111 trace->opts.no_inherit = perf_config_bool(var, value);
4112 } else if (!strcmp(var, "trace.args_alignment")) {
4113 int args_alignment = 0;
4114 if (perf_config_int(&args_alignment, var, value) == 0)
4115 trace->args_alignment = args_alignment;
4116 }
4117 out:
4118 return err;
4119 }
4120
cmd_trace(int argc,const char ** argv)4121 int cmd_trace(int argc, const char **argv)
4122 {
4123 const char *trace_usage[] = {
4124 "perf trace [<options>] [<command>]",
4125 "perf trace [<options>] -- <command> [<options>]",
4126 "perf trace record [<options>] [<command>]",
4127 "perf trace record [<options>] -- <command> [<options>]",
4128 NULL
4129 };
4130 struct trace trace = {
4131 .opts = {
4132 .target = {
4133 .uid = UINT_MAX,
4134 .uses_mmap = true,
4135 },
4136 .user_freq = UINT_MAX,
4137 .user_interval = ULLONG_MAX,
4138 .no_buffering = true,
4139 .mmap_pages = UINT_MAX,
4140 },
4141 .output = stderr,
4142 .show_comm = true,
4143 .show_tstamp = true,
4144 .show_duration = true,
4145 .show_arg_names = true,
4146 .args_alignment = 70,
4147 .trace_syscalls = false,
4148 .kernel_syscallchains = false,
4149 .max_stack = UINT_MAX,
4150 .max_events = ULONG_MAX,
4151 };
4152 const char *map_dump_str = NULL;
4153 const char *output_name = NULL;
4154 const struct option trace_options[] = {
4155 OPT_CALLBACK('e', "event", &trace, "event",
4156 "event/syscall selector. use 'perf list' to list available events",
4157 trace__parse_events_option),
4158 OPT_BOOLEAN(0, "comm", &trace.show_comm,
4159 "show the thread COMM next to its id"),
4160 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
4161 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4162 trace__parse_events_option),
4163 OPT_STRING('o', "output", &output_name, "file", "output file name"),
4164 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
4165 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4166 "trace events on existing process id"),
4167 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
4168 "trace events on existing thread id"),
4169 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
4170 "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
4171 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
4172 "system-wide collection from all CPUs"),
4173 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
4174 "list of cpus to monitor"),
4175 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
4176 "child tasks do not inherit counters"),
4177 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
4178 "number of mmap data pages",
4179 perf_evlist__parse_mmap_pages),
4180 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
4181 "user to profile"),
4182 OPT_CALLBACK(0, "duration", &trace, "float",
4183 "show only events with duration > N.M ms",
4184 trace__set_duration),
4185 #ifdef HAVE_LIBBPF_SUPPORT
4186 OPT_STRING(0, "map-dump", &map_dump_str, "BPF map", "BPF map to periodically dump"),
4187 #endif
4188 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
4189 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4190 OPT_BOOLEAN('T', "time", &trace.full_time,
4191 "Show full timestamp, not time relative to first start"),
4192 OPT_BOOLEAN(0, "failure", &trace.failure_only,
4193 "Show only syscalls that failed"),
4194 OPT_BOOLEAN('s', "summary", &trace.summary_only,
4195 "Show only syscall summary with statistics"),
4196 OPT_BOOLEAN('S', "with-summary", &trace.summary,
4197 "Show all syscalls and summary with statistics"),
4198 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4199 "Trace pagefaults", parse_pagefaults, "maj"),
4200 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
4201 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
4202 OPT_CALLBACK(0, "call-graph", &trace.opts,
4203 "record_mode[,record_size]", record_callchain_help,
4204 &record_parse_callchain_opt),
4205 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4206 "Show the kernel callchains on the syscall exit path"),
4207 OPT_ULONG(0, "max-events", &trace.max_events,
4208 "Set the maximum number of events to print, exit after that is reached. "),
4209 OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4210 "Set the minimum stack depth when parsing the callchain, "
4211 "anything below the specified depth will be ignored."),
4212 OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4213 "Set the maximum stack depth when parsing the callchain, "
4214 "anything beyond the specified depth will be ignored. "
4215 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
4216 OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4217 "Sort batch of events before processing, use if getting out of order events"),
4218 OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4219 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
4220 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
4221 "per thread proc mmap processing timeout in ms"),
4222 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4223 trace__parse_cgroups),
4224 OPT_UINTEGER('D', "delay", &trace.opts.initial_delay,
4225 "ms to wait before starting measurement after program "
4226 "start"),
4227 OPTS_EVSWITCH(&trace.evswitch),
4228 OPT_END()
4229 };
4230 bool __maybe_unused max_stack_user_set = true;
4231 bool mmap_pages_user_set = true;
4232 struct evsel *evsel;
4233 const char * const trace_subcommands[] = { "record", NULL };
4234 int err = -1;
4235 char bf[BUFSIZ];
4236
4237 signal(SIGSEGV, sighandler_dump_stack);
4238 signal(SIGFPE, sighandler_dump_stack);
4239
4240 trace.evlist = evlist__new();
4241 trace.sctbl = syscalltbl__new();
4242
4243 if (trace.evlist == NULL || trace.sctbl == NULL) {
4244 pr_err("Not enough memory to run!\n");
4245 err = -ENOMEM;
4246 goto out;
4247 }
4248
4249 /*
4250 * Parsing .perfconfig may entail creating a BPF event, that may need
4251 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4252 * is too small. This affects just this process, not touching the
4253 * global setting. If it fails we'll get something in 'perf trace -v'
4254 * to help diagnose the problem.
4255 */
4256 rlimit__bump_memlock();
4257
4258 err = perf_config(trace__config, &trace);
4259 if (err)
4260 goto out;
4261
4262 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4263 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
4264
4265 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4266 usage_with_options_msg(trace_usage, trace_options,
4267 "cgroup monitoring only available in system-wide mode");
4268 }
4269
4270 evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
4271 if (IS_ERR(evsel)) {
4272 bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
4273 pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
4274 goto out;
4275 }
4276
4277 if (evsel) {
4278 trace.syscalls.events.augmented = evsel;
4279
4280 evsel = perf_evlist__find_tracepoint_by_name(trace.evlist, "raw_syscalls:sys_enter");
4281 if (evsel == NULL) {
4282 pr_err("ERROR: raw_syscalls:sys_enter not found in the augmented BPF object\n");
4283 goto out;
4284 }
4285
4286 if (evsel->bpf_obj == NULL) {
4287 pr_err("ERROR: raw_syscalls:sys_enter not associated to a BPF object\n");
4288 goto out;
4289 }
4290
4291 trace.bpf_obj = evsel->bpf_obj;
4292
4293 trace__set_bpf_map_filtered_pids(&trace);
4294 trace__set_bpf_map_syscalls(&trace);
4295 trace.syscalls.unaugmented_prog = trace__find_bpf_program_by_title(&trace, "!raw_syscalls:unaugmented");
4296 }
4297
4298 err = bpf__setup_stdout(trace.evlist);
4299 if (err) {
4300 bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
4301 pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
4302 goto out;
4303 }
4304
4305 err = -1;
4306
4307 if (map_dump_str) {
4308 trace.dump.map = trace__find_bpf_map_by_name(&trace, map_dump_str);
4309 if (trace.dump.map == NULL) {
4310 pr_err("ERROR: BPF map \"%s\" not found\n", map_dump_str);
4311 goto out;
4312 }
4313 }
4314
4315 if (trace.trace_pgfaults) {
4316 trace.opts.sample_address = true;
4317 trace.opts.sample_time = true;
4318 }
4319
4320 if (trace.opts.mmap_pages == UINT_MAX)
4321 mmap_pages_user_set = false;
4322
4323 if (trace.max_stack == UINT_MAX) {
4324 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
4325 max_stack_user_set = false;
4326 }
4327
4328 #ifdef HAVE_DWARF_UNWIND_SUPPORT
4329 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
4330 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
4331 }
4332 #endif
4333
4334 if (callchain_param.enabled) {
4335 if (!mmap_pages_user_set && geteuid() == 0)
4336 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
4337
4338 symbol_conf.use_callchain = true;
4339 }
4340
4341 if (trace.evlist->core.nr_entries > 0) {
4342 evlist__set_evsel_handler(trace.evlist, trace__event_handler);
4343 if (evlist__set_syscall_tp_fields(trace.evlist)) {
4344 perror("failed to set syscalls:* tracepoint fields");
4345 goto out;
4346 }
4347 }
4348
4349 if (trace.sort_events) {
4350 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
4351 ordered_events__set_copy_on_queue(&trace.oe.data, true);
4352 }
4353
4354 /*
4355 * If we are augmenting syscalls, then combine what we put in the
4356 * __augmented_syscalls__ BPF map with what is in the
4357 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
4358 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
4359 *
4360 * We'll switch to look at two BPF maps, one for sys_enter and the
4361 * other for sys_exit when we start augmenting the sys_exit paths with
4362 * buffers that are being copied from kernel to userspace, think 'read'
4363 * syscall.
4364 */
4365 if (trace.syscalls.events.augmented) {
4366 evlist__for_each_entry(trace.evlist, evsel) {
4367 bool raw_syscalls_sys_exit = strcmp(perf_evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
4368
4369 if (raw_syscalls_sys_exit) {
4370 trace.raw_augmented_syscalls = true;
4371 goto init_augmented_syscall_tp;
4372 }
4373
4374 if (trace.syscalls.events.augmented->priv == NULL &&
4375 strstr(perf_evsel__name(evsel), "syscalls:sys_enter")) {
4376 struct evsel *augmented = trace.syscalls.events.augmented;
4377 if (perf_evsel__init_augmented_syscall_tp(augmented, evsel) ||
4378 perf_evsel__init_augmented_syscall_tp_args(augmented))
4379 goto out;
4380 /*
4381 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
4382 * Above we made sure we can get from the payload the tp fields
4383 * that we get from syscalls:sys_enter tracefs format file.
4384 */
4385 augmented->handler = trace__sys_enter;
4386 /*
4387 * Now we do the same for the *syscalls:sys_enter event so that
4388 * if we handle it directly, i.e. if the BPF prog returns 0 so
4389 * as not to filter it, then we'll handle it just like we would
4390 * for the BPF_OUTPUT one:
4391 */
4392 if (perf_evsel__init_augmented_syscall_tp(evsel, evsel) ||
4393 perf_evsel__init_augmented_syscall_tp_args(evsel))
4394 goto out;
4395 evsel->handler = trace__sys_enter;
4396 }
4397
4398 if (strstarts(perf_evsel__name(evsel), "syscalls:sys_exit_")) {
4399 struct syscall_tp *sc;
4400 init_augmented_syscall_tp:
4401 if (perf_evsel__init_augmented_syscall_tp(evsel, evsel))
4402 goto out;
4403 sc = evsel->priv;
4404 /*
4405 * For now with BPF raw_augmented we hook into
4406 * raw_syscalls:sys_enter and there we get all
4407 * 6 syscall args plus the tracepoint common
4408 * fields and the syscall_nr (another long).
4409 * So we check if that is the case and if so
4410 * don't look after the sc->args_size but
4411 * always after the full raw_syscalls:sys_enter
4412 * payload, which is fixed.
4413 *
4414 * We'll revisit this later to pass
4415 * s->args_size to the BPF augmenter (now
4416 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
4417 * so that it copies only what we need for each
4418 * syscall, like what happens when we use
4419 * syscalls:sys_enter_NAME, so that we reduce
4420 * the kernel/userspace traffic to just what is
4421 * needed for each syscall.
4422 */
4423 if (trace.raw_augmented_syscalls)
4424 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
4425 perf_evsel__init_augmented_syscall_tp_ret(evsel);
4426 evsel->handler = trace__sys_exit;
4427 }
4428 }
4429 }
4430
4431 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
4432 return trace__record(&trace, argc-1, &argv[1]);
4433
4434 /* summary_only implies summary option, but don't overwrite summary if set */
4435 if (trace.summary_only)
4436 trace.summary = trace.summary_only;
4437
4438 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4439 trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4440 trace.trace_syscalls = true;
4441 }
4442
4443 if (output_name != NULL) {
4444 err = trace__open_output(&trace, output_name);
4445 if (err < 0) {
4446 perror("failed to create output file");
4447 goto out;
4448 }
4449 }
4450
4451 err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
4452 if (err)
4453 goto out_close;
4454
4455 err = target__validate(&trace.opts.target);
4456 if (err) {
4457 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
4458 fprintf(trace.output, "%s", bf);
4459 goto out_close;
4460 }
4461
4462 err = target__parse_uid(&trace.opts.target);
4463 if (err) {
4464 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
4465 fprintf(trace.output, "%s", bf);
4466 goto out_close;
4467 }
4468
4469 if (!argc && target__none(&trace.opts.target))
4470 trace.opts.target.system_wide = true;
4471
4472 if (input_name)
4473 err = trace__replay(&trace);
4474 else
4475 err = trace__run(&trace, argc, argv);
4476
4477 out_close:
4478 if (output_name != NULL)
4479 fclose(trace.output);
4480 out:
4481 return err;
4482 }
4483