projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
perf trace: Call bpf__apply_obj_config in 'perf trace'
[deliverable/linux.git] / tools / perf / builtin-trace.c
CommitLineData
a598bb5e
ACM		 1/*
2 * builtin-trace.c
3 *
4 * Builtin 'trace' command:
5 *
6 * Display a continuously updated trace of any workload, CPU, specific PID,
7 * system wide, etc. Default format is loosely strace like, but any other
8 * event may be specified using --event.
9 *
10 * Copyright (C) 2012, 2013, 2014, 2015 Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
11 *
12 * Initially based on the 'trace' prototype by Thomas Gleixner:
13 *
14 * http://lwn.net/Articles/415728/ ("Announcing a new utility: 'trace'")
15 *
16 * Released under the GPL v2. (and only v2, not any later version)
17 */
18
4e319027 19#include <traceevent/event-parse.h>
988bdb31 20#include <api/fs/tracing_path.h>
514f1c67 21#include "builtin.h"
752fde44 22#include "util/color.h"
7c304ee0 23#include "util/debug.h"
514f1c67 24#include "util/evlist.h"
4b6ab94e 25#include <subcmd/exec-cmd.h>
752fde44 26#include "util/machine.h"
6810fc91 27#include "util/session.h"
752fde44 28#include "util/thread.h"
4b6ab94e 29#include <subcmd/parse-options.h>
2ae3a312 30#include "util/strlist.h"
bdc89661 31#include "util/intlist.h"
514f1c67 32#include "util/thread_map.h"
bf2575c1 33#include "util/stat.h"
97978b3e 34#include "trace-event.h"
9aca7f17 35#include "util/parse-events.h"
ba504235 36#include "util/bpf-loader.h"
514f1c67
ACM		 37
38#include <libaudit.h>
39#include <stdlib.h>
ae685380 40#include <sys/mman.h>
f9da0b0c 41#include <linux/futex.h>
8dd2a131 42#include <linux/err.h>
514f1c67 43
456857bd
IM		 44/* For older distros: */
45#ifndef MAP_STACK
46# define MAP_STACK 0x20000
47#endif
48
49#ifndef MADV_HWPOISON
50# define MADV_HWPOISON 100
a598bb5e 51
456857bd
IM		 52#endif
53
54#ifndef MADV_MERGEABLE
55# define MADV_MERGEABLE 12
56#endif
57
58#ifndef MADV_UNMERGEABLE
59# define MADV_UNMERGEABLE 13
60#endif
61
79d26a6a
BH		 62#ifndef EFD_SEMAPHORE
63# define EFD_SEMAPHORE 1
64#endif
65
c188e7ac
ACM		 66#ifndef EFD_NONBLOCK
67# define EFD_NONBLOCK 00004000
68#endif
69
70#ifndef EFD_CLOEXEC
71# define EFD_CLOEXEC 02000000
72#endif
73
74#ifndef O_CLOEXEC
75# define O_CLOEXEC 02000000
76#endif
77
78#ifndef SOCK_DCCP
79# define SOCK_DCCP 6
80#endif
81
82#ifndef SOCK_CLOEXEC
83# define SOCK_CLOEXEC 02000000
84#endif
85
86#ifndef SOCK_NONBLOCK
87# define SOCK_NONBLOCK 00004000
88#endif
89
90#ifndef MSG_CMSG_CLOEXEC
91# define MSG_CMSG_CLOEXEC 0x40000000
92#endif
93
a1c2552d
ACM		 94#ifndef PERF_FLAG_FD_NO_GROUP
95# define PERF_FLAG_FD_NO_GROUP (1UL << 0)
96#endif
97
98#ifndef PERF_FLAG_FD_OUTPUT
99# define PERF_FLAG_FD_OUTPUT (1UL << 1)
100#endif
101
102#ifndef PERF_FLAG_PID_CGROUP
103# define PERF_FLAG_PID_CGROUP (1UL << 2) /* pid=cgroup id, per-cpu mode only */
104#endif
105
106#ifndef PERF_FLAG_FD_CLOEXEC
107# define PERF_FLAG_FD_CLOEXEC (1UL << 3) /* O_CLOEXEC */
108#endif
109
110
77170988
ACM		 111struct tp_field {
112 int offset;
113 union {
114 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
115 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
116 };
117};
118
119#define TP_UINT_FIELD(bits) \
120static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
121{ \
55d43bca
DA		 122 u##bits value; \
123 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
124 return value; \
77170988
ACM		 125}
126
127TP_UINT_FIELD(8);
128TP_UINT_FIELD(16);
129TP_UINT_FIELD(32);
130TP_UINT_FIELD(64);
131
132#define TP_UINT_FIELD__SWAPPED(bits) \
133static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
134{ \
55d43bca
DA		 135 u##bits value; \
136 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
77170988
ACM		 137 return bswap_##bits(value);\
138}
139
140TP_UINT_FIELD__SWAPPED(16);
141TP_UINT_FIELD__SWAPPED(32);
142TP_UINT_FIELD__SWAPPED(64);
143
144static int tp_field__init_uint(struct tp_field *field,
145 struct format_field *format_field,
146 bool needs_swap)
147{
148 field->offset = format_field->offset;
149
150 switch (format_field->size) {
151 case 1:
152 field->integer = tp_field__u8;
153 break;
154 case 2:
155 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
156 break;
157 case 4:
158 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
159 break;
160 case 8:
161 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
162 break;
163 default:
164 return -1;
165 }
166
167 return 0;
168}
169
170static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
171{
172 return sample->raw_data + field->offset;
173}
174
175static int tp_field__init_ptr(struct tp_field *field, struct format_field *format_field)
176{
177 field->offset = format_field->offset;
178 field->pointer = tp_field__ptr;
179 return 0;
180}
181
182struct syscall_tp {
183 struct tp_field id;
184 union {
185 struct tp_field args, ret;
186 };
187};
188
189static int perf_evsel__init_tp_uint_field(struct perf_evsel *evsel,
190 struct tp_field *field,
191 const char *name)
192{
193 struct format_field *format_field = perf_evsel__field(evsel, name);
194
195 if (format_field == NULL)
196 return -1;
197
198 return tp_field__init_uint(field, format_field, evsel->needs_swap);
199}
200
201#define perf_evsel__init_sc_tp_uint_field(evsel, name) \
202 ({ struct syscall_tp *sc = evsel->priv;\
203 perf_evsel__init_tp_uint_field(evsel, &sc->name, #name); })
204
205static int perf_evsel__init_tp_ptr_field(struct perf_evsel *evsel,
206 struct tp_field *field,
207 const char *name)
208{
209 struct format_field *format_field = perf_evsel__field(evsel, name);
210
211 if (format_field == NULL)
212 return -1;
213
214 return tp_field__init_ptr(field, format_field);
215}
216
217#define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
218 ({ struct syscall_tp *sc = evsel->priv;\
219 perf_evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
220
221static void perf_evsel__delete_priv(struct perf_evsel *evsel)
222{
04662523 223 zfree(&evsel->priv);
77170988
ACM		 224 perf_evsel__delete(evsel);
225}
226
96695d44
NK		 227static int perf_evsel__init_syscall_tp(struct perf_evsel *evsel, void *handler)
228{
229 evsel->priv = malloc(sizeof(struct syscall_tp));
230 if (evsel->priv != NULL) {
231 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
232 goto out_delete;
233
234 evsel->handler = handler;
235 return 0;
236 }
237
238 return -ENOMEM;
239
240out_delete:
04662523 241 zfree(&evsel->priv);
96695d44
NK		 242 return -ENOENT;
243}
244
ef503831 245static struct perf_evsel *perf_evsel__syscall_newtp(const char *direction, void *handler)
77170988 246{
ef503831 247 struct perf_evsel *evsel = perf_evsel__newtp("raw_syscalls", direction);
77170988 248
9aca7f17 249 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
8dd2a131 250 if (IS_ERR(evsel))
9aca7f17
DA		 251 evsel = perf_evsel__newtp("syscalls", direction);
252
8dd2a131
JO		 253 if (IS_ERR(evsel))
254 return NULL;
255
256 if (perf_evsel__init_syscall_tp(evsel, handler))
257 goto out_delete;
77170988
ACM		 258
259 return evsel;
260
261out_delete:
262 perf_evsel__delete_priv(evsel);
263 return NULL;
264}
265
266#define perf_evsel__sc_tp_uint(evsel, name, sample) \
267 ({ struct syscall_tp *fields = evsel->priv; \
268 fields->name.integer(&fields->name, sample); })
269
270#define perf_evsel__sc_tp_ptr(evsel, name, sample) \
271 ({ struct syscall_tp *fields = evsel->priv; \
272 fields->name.pointer(&fields->name, sample); })
273
01533e97
ACM		 274struct syscall_arg {
275 unsigned long val;
75b757ca
ACM		 276 struct thread *thread;
277 struct trace *trace;
1f115cb7 278 void *parm;
01533e97
ACM		 279 u8 idx;
280 u8 mask;
281};
282
1f115cb7 283struct strarray {
03e3adc9 284 int offset;
1f115cb7
ACM		 285 int nr_entries;
286 const char **entries;
287};
288
289#define DEFINE_STRARRAY(array) struct strarray strarray__##array = { \
290 .nr_entries = ARRAY_SIZE(array), \
291 .entries = array, \
292}
293
03e3adc9
ACM		 294#define DEFINE_STRARRAY_OFFSET(array, off) struct strarray strarray__##array = { \
295 .offset = off, \
296 .nr_entries = ARRAY_SIZE(array), \
297 .entries = array, \
298}
299
975b7c2f
ACM		 300static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
301 const char *intfmt,
302 struct syscall_arg *arg)
1f115cb7 303{
1f115cb7 304 struct strarray *sa = arg->parm;
03e3adc9 305 int idx = arg->val - sa->offset;
1f115cb7
ACM		 306
307 if (idx < 0 || idx >= sa->nr_entries)
975b7c2f 308 return scnprintf(bf, size, intfmt, arg->val);
1f115cb7
ACM		 309
310 return scnprintf(bf, size, "%s", sa->entries[idx]);
311}
312
975b7c2f
ACM		 313static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
314 struct syscall_arg *arg)
315{
316 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
317}
318
1f115cb7
ACM		 319#define SCA_STRARRAY syscall_arg__scnprintf_strarray
320
844ae5b4
ACM		 321#if defined(__i386__) || defined(__x86_64__)
322/*
323 * FIXME: Make this available to all arches as soon as the ioctl beautifier
324 * gets rewritten to support all arches.
325 */
78645cf3
ACM		 326static size_t syscall_arg__scnprintf_strhexarray(char *bf, size_t size,
327 struct syscall_arg *arg)
328{
329 return __syscall_arg__scnprintf_strarray(bf, size, "%#x", arg);
330}
331
332#define SCA_STRHEXARRAY syscall_arg__scnprintf_strhexarray
844ae5b4 333#endif /* defined(__i386__) || defined(__x86_64__) */
78645cf3 334
75b757ca
ACM		 335static size_t syscall_arg__scnprintf_fd(char *bf, size_t size,
336 struct syscall_arg *arg);
337
338#define SCA_FD syscall_arg__scnprintf_fd
339
340static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
341 struct syscall_arg *arg)
342{
343 int fd = arg->val;
344
345 if (fd == AT_FDCWD)
346 return scnprintf(bf, size, "CWD");
347
348 return syscall_arg__scnprintf_fd(bf, size, arg);
349}
350
351#define SCA_FDAT syscall_arg__scnprintf_fd_at
352
353static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
354 struct syscall_arg *arg);
355
356#define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
357
6e7eeb51 358static size_t syscall_arg__scnprintf_hex(char *bf, size_t size,
01533e97 359 struct syscall_arg *arg)
13d4ff3e 360{
01533e97 361 return scnprintf(bf, size, "%#lx", arg->val);
13d4ff3e
ACM		 362}
363
beccb2b5
ACM		 364#define SCA_HEX syscall_arg__scnprintf_hex
365
a1c2552d
ACM		 366static size_t syscall_arg__scnprintf_int(char *bf, size_t size,
367 struct syscall_arg *arg)
368{
369 return scnprintf(bf, size, "%d", arg->val);
370}
371
372#define SCA_INT syscall_arg__scnprintf_int
373
6e7eeb51 374static size_t syscall_arg__scnprintf_mmap_prot(char *bf, size_t size,
01533e97 375 struct syscall_arg *arg)
ae685380 376{
01533e97 377 int printed = 0, prot = arg->val;
ae685380
ACM		 378
379 if (prot == PROT_NONE)
380 return scnprintf(bf, size, "NONE");
381#define P_MMAP_PROT(n) \
382 if (prot & PROT_##n) { \
383 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
384 prot &= ~PROT_##n; \
385 }
386
387 P_MMAP_PROT(EXEC);
388 P_MMAP_PROT(READ);
389 P_MMAP_PROT(WRITE);
390#ifdef PROT_SEM
391 P_MMAP_PROT(SEM);
392#endif
393 P_MMAP_PROT(GROWSDOWN);
394 P_MMAP_PROT(GROWSUP);
395#undef P_MMAP_PROT
396
397 if (prot)
398 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", prot);
399
400 return printed;
401}
402
403#define SCA_MMAP_PROT syscall_arg__scnprintf_mmap_prot
404
6e7eeb51 405static size_t syscall_arg__scnprintf_mmap_flags(char *bf, size_t size,
01533e97 406 struct syscall_arg *arg)
941557e0 407{
01533e97 408 int printed = 0, flags = arg->val;
941557e0
ACM		 409
410#define P_MMAP_FLAG(n) \
411 if (flags & MAP_##n) { \
412 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
413 flags &= ~MAP_##n; \
414 }
415
416 P_MMAP_FLAG(SHARED);
417 P_MMAP_FLAG(PRIVATE);
41817815 418#ifdef MAP_32BIT
941557e0 419 P_MMAP_FLAG(32BIT);
41817815 420#endif
941557e0
ACM		 421 P_MMAP_FLAG(ANONYMOUS);
422 P_MMAP_FLAG(DENYWRITE);
423 P_MMAP_FLAG(EXECUTABLE);
424 P_MMAP_FLAG(FILE);
425 P_MMAP_FLAG(FIXED);
426 P_MMAP_FLAG(GROWSDOWN);
f2935f3e 427#ifdef MAP_HUGETLB
941557e0 428 P_MMAP_FLAG(HUGETLB);
f2935f3e 429#endif
941557e0
ACM		 430 P_MMAP_FLAG(LOCKED);
431 P_MMAP_FLAG(NONBLOCK);
432 P_MMAP_FLAG(NORESERVE);
433 P_MMAP_FLAG(POPULATE);
434 P_MMAP_FLAG(STACK);
435#ifdef MAP_UNINITIALIZED
436 P_MMAP_FLAG(UNINITIALIZED);
437#endif
438#undef P_MMAP_FLAG
439
440 if (flags)
441 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
442
443 return printed;
444}
445
446#define SCA_MMAP_FLAGS syscall_arg__scnprintf_mmap_flags
447
86998dda
AS		 448static size_t syscall_arg__scnprintf_mremap_flags(char *bf, size_t size,
449 struct syscall_arg *arg)
450{
451 int printed = 0, flags = arg->val;
452
453#define P_MREMAP_FLAG(n) \
454 if (flags & MREMAP_##n) { \
455 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
456 flags &= ~MREMAP_##n; \
457 }
458
459 P_MREMAP_FLAG(MAYMOVE);
460#ifdef MREMAP_FIXED
461 P_MREMAP_FLAG(FIXED);
462#endif
463#undef P_MREMAP_FLAG
464
465 if (flags)
466 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
467
468 return printed;
469}
470
471#define SCA_MREMAP_FLAGS syscall_arg__scnprintf_mremap_flags
472
6e7eeb51 473static size_t syscall_arg__scnprintf_madvise_behavior(char *bf, size_t size,
01533e97 474 struct syscall_arg *arg)
9e9716d1 475{
01533e97 476 int behavior = arg->val;
9e9716d1
ACM		 477
478 switch (behavior) {
479#define P_MADV_BHV(n) case MADV_##n: return scnprintf(bf, size, #n)
480 P_MADV_BHV(NORMAL);
481 P_MADV_BHV(RANDOM);
482 P_MADV_BHV(SEQUENTIAL);
483 P_MADV_BHV(WILLNEED);
484 P_MADV_BHV(DONTNEED);
485 P_MADV_BHV(REMOVE);
486 P_MADV_BHV(DONTFORK);
487 P_MADV_BHV(DOFORK);
488 P_MADV_BHV(HWPOISON);
489#ifdef MADV_SOFT_OFFLINE
490 P_MADV_BHV(SOFT_OFFLINE);
491#endif
492 P_MADV_BHV(MERGEABLE);
493 P_MADV_BHV(UNMERGEABLE);
f2935f3e 494#ifdef MADV_HUGEPAGE
9e9716d1 495 P_MADV_BHV(HUGEPAGE);
f2935f3e
DA		 496#endif
497#ifdef MADV_NOHUGEPAGE
9e9716d1 498 P_MADV_BHV(NOHUGEPAGE);
f2935f3e 499#endif
9e9716d1
ACM		 500#ifdef MADV_DONTDUMP
501 P_MADV_BHV(DONTDUMP);
502#endif
503#ifdef MADV_DODUMP
504 P_MADV_BHV(DODUMP);
505#endif
506#undef P_MADV_PHV
507 default: break;
508 }
509
510 return scnprintf(bf, size, "%#x", behavior);
511}
512
513#define SCA_MADV_BHV syscall_arg__scnprintf_madvise_behavior
514
5cea6ff2
ACM		 515static size_t syscall_arg__scnprintf_flock(char *bf, size_t size,
516 struct syscall_arg *arg)
517{
518 int printed = 0, op = arg->val;
519
520 if (op == 0)
521 return scnprintf(bf, size, "NONE");
522#define P_CMD(cmd) \
523 if ((op & LOCK_##cmd) == LOCK_##cmd) { \
524 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #cmd); \
525 op &= ~LOCK_##cmd; \
526 }
527
528 P_CMD(SH);
529 P_CMD(EX);
530 P_CMD(NB);
531 P_CMD(UN);
532 P_CMD(MAND);
533 P_CMD(RW);
534 P_CMD(READ);
535 P_CMD(WRITE);
536#undef P_OP
537
538 if (op)
539 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", op);
540
541 return printed;
542}
543
544#define SCA_FLOCK syscall_arg__scnprintf_flock
545
01533e97 546static size_t syscall_arg__scnprintf_futex_op(char *bf, size_t size, struct syscall_arg *arg)
f9da0b0c
ACM		 547{
548 enum syscall_futex_args {
549 SCF_UADDR = (1 << 0),
550 SCF_OP = (1 << 1),
551 SCF_VAL = (1 << 2),
552 SCF_TIMEOUT = (1 << 3),
553 SCF_UADDR2 = (1 << 4),
554 SCF_VAL3 = (1 << 5),
555 };
01533e97 556 int op = arg->val;
f9da0b0c
ACM		 557 int cmd = op & FUTEX_CMD_MASK;
558 size_t printed = 0;
559
560 switch (cmd) {
561#define P_FUTEX_OP(n) case FUTEX_##n: printed = scnprintf(bf, size, #n);
01533e97
ACM		 562 P_FUTEX_OP(WAIT); arg->mask |= SCF_VAL3|SCF_UADDR2; break;
563 P_FUTEX_OP(WAKE); arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
564 P_FUTEX_OP(FD); arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
565 P_FUTEX_OP(REQUEUE); arg->mask |= SCF_VAL3|SCF_TIMEOUT; break;
566 P_FUTEX_OP(CMP_REQUEUE); arg->mask |= SCF_TIMEOUT; break;
567 P_FUTEX_OP(CMP_REQUEUE_PI); arg->mask |= SCF_TIMEOUT; break;
f9da0b0c 568 P_FUTEX_OP(WAKE_OP); break;
01533e97
ACM		 569 P_FUTEX_OP(LOCK_PI); arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
570 P_FUTEX_OP(UNLOCK_PI); arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
571 P_FUTEX_OP(TRYLOCK_PI); arg->mask |= SCF_VAL3|SCF_UADDR2; break;
572 P_FUTEX_OP(WAIT_BITSET); arg->mask |= SCF_UADDR2; break;
573 P_FUTEX_OP(WAKE_BITSET); arg->mask |= SCF_UADDR2; break;
f9da0b0c
ACM		 574 P_FUTEX_OP(WAIT_REQUEUE_PI); break;
575 default: printed = scnprintf(bf, size, "%#x", cmd); break;
576 }
577
578 if (op & FUTEX_PRIVATE_FLAG)
579 printed += scnprintf(bf + printed, size - printed, "|PRIV");
580
581 if (op & FUTEX_CLOCK_REALTIME)
582 printed += scnprintf(bf + printed, size - printed, "|CLKRT");
583
584 return printed;
585}
586
efe6b882
ACM		 587#define SCA_FUTEX_OP syscall_arg__scnprintf_futex_op
588
729a7841
ACM		 589static const char *bpf_cmd[] = {
590 "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
591 "MAP_GET_NEXT_KEY", "PROG_LOAD",
592};
593static DEFINE_STRARRAY(bpf_cmd);
594
03e3adc9
ACM		 595static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
596static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, 1);
eac032c5 597
1f115cb7
ACM		 598static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
599static DEFINE_STRARRAY(itimers);
600
b62bee1b
ACM		 601static const char *keyctl_options[] = {
602 "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
603 "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
604 "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
605 "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
606 "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
607};
608static DEFINE_STRARRAY(keyctl_options);
609
efe6b882
ACM		 610static const char *whences[] = { "SET", "CUR", "END",
611#ifdef SEEK_DATA
612"DATA",
613#endif
614#ifdef SEEK_HOLE
615"HOLE",
616#endif
617};
618static DEFINE_STRARRAY(whences);
f9da0b0c 619
80f587d5
ACM		 620static const char *fcntl_cmds[] = {
621 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
622 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "F_GETLK64",
623 "F_SETLK64", "F_SETLKW64", "F_SETOWN_EX", "F_GETOWN_EX",
624 "F_GETOWNER_UIDS",
625};
626static DEFINE_STRARRAY(fcntl_cmds);
627
c045bf02
ACM		 628static const char *rlimit_resources[] = {
629 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
630 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
631 "RTTIME",
632};
633static DEFINE_STRARRAY(rlimit_resources);
634
eb5b1b14
ACM		 635static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
636static DEFINE_STRARRAY(sighow);
637
4f8c1b74
DA		 638static const char *clockid[] = {
639 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
28ebb87c
ACM		 640 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
641 "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
4f8c1b74
DA		 642};
643static DEFINE_STRARRAY(clockid);
644
e10bce81
ACM		 645static const char *socket_families[] = {
646 "UNSPEC", "LOCAL", "INET", "AX25", "IPX", "APPLETALK", "NETROM",
647 "BRIDGE", "ATMPVC", "X25", "INET6", "ROSE", "DECnet", "NETBEUI",
648 "SECURITY", "KEY", "NETLINK", "PACKET", "ASH", "ECONET", "ATMSVC",
649 "RDS", "SNA", "IRDA", "PPPOX", "WANPIPE", "LLC", "IB", "CAN", "TIPC",
650 "BLUETOOTH", "IUCV", "RXRPC", "ISDN", "PHONET", "IEEE802154", "CAIF",
651 "ALG", "NFC", "VSOCK",
652};
653static DEFINE_STRARRAY(socket_families);
654
a28b24b2
ACM		 655#ifndef SOCK_TYPE_MASK
656#define SOCK_TYPE_MASK 0xf
657#endif
658
659static size_t syscall_arg__scnprintf_socket_type(char *bf, size_t size,
660 struct syscall_arg *arg)
661{
662 size_t printed;
663 int type = arg->val,
664 flags = type & ~SOCK_TYPE_MASK;
665
666 type &= SOCK_TYPE_MASK;
667 /*
668 * Can't use a strarray, MIPS may override for ABI reasons.
669 */
670 switch (type) {
671#define P_SK_TYPE(n) case SOCK_##n: printed = scnprintf(bf, size, #n); break;
672 P_SK_TYPE(STREAM);
673 P_SK_TYPE(DGRAM);
674 P_SK_TYPE(RAW);
675 P_SK_TYPE(RDM);
676 P_SK_TYPE(SEQPACKET);
677 P_SK_TYPE(DCCP);
678 P_SK_TYPE(PACKET);
679#undef P_SK_TYPE
680 default:
681 printed = scnprintf(bf, size, "%#x", type);
682 }
683
684#define P_SK_FLAG(n) \
685 if (flags & SOCK_##n) { \
686 printed += scnprintf(bf + printed, size - printed, "|%s", #n); \
687 flags &= ~SOCK_##n; \
688 }
689
690 P_SK_FLAG(CLOEXEC);
691 P_SK_FLAG(NONBLOCK);
692#undef P_SK_FLAG
693
694 if (flags)
695 printed += scnprintf(bf + printed, size - printed, "|%#x", flags);
696
697 return printed;
698}
699
700#define SCA_SK_TYPE syscall_arg__scnprintf_socket_type
701
b2cc99fd
ACM		 702#ifndef MSG_PROBE
703#define MSG_PROBE 0x10
704#endif
b6e8f8f4
DA		 705#ifndef MSG_WAITFORONE
706#define MSG_WAITFORONE 0x10000
707#endif
b2cc99fd
ACM		 708#ifndef MSG_SENDPAGE_NOTLAST
709#define MSG_SENDPAGE_NOTLAST 0x20000
710#endif
711#ifndef MSG_FASTOPEN
712#define MSG_FASTOPEN 0x20000000
713#endif
714
715static size_t syscall_arg__scnprintf_msg_flags(char *bf, size_t size,
716 struct syscall_arg *arg)
717{
718 int printed = 0, flags = arg->val;
719
720 if (flags == 0)
721 return scnprintf(bf, size, "NONE");
722#define P_MSG_FLAG(n) \
723 if (flags & MSG_##n) { \
724 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
725 flags &= ~MSG_##n; \
726 }
727
728 P_MSG_FLAG(OOB);
729 P_MSG_FLAG(PEEK);
730 P_MSG_FLAG(DONTROUTE);
731 P_MSG_FLAG(TRYHARD);
732 P_MSG_FLAG(CTRUNC);
733 P_MSG_FLAG(PROBE);
734 P_MSG_FLAG(TRUNC);
735 P_MSG_FLAG(DONTWAIT);
736 P_MSG_FLAG(EOR);
737 P_MSG_FLAG(WAITALL);
738 P_MSG_FLAG(FIN);
739 P_MSG_FLAG(SYN);
740 P_MSG_FLAG(CONFIRM);
741 P_MSG_FLAG(RST);
742 P_MSG_FLAG(ERRQUEUE);
743 P_MSG_FLAG(NOSIGNAL);
744 P_MSG_FLAG(MORE);
745 P_MSG_FLAG(WAITFORONE);
746 P_MSG_FLAG(SENDPAGE_NOTLAST);
747 P_MSG_FLAG(FASTOPEN);
748 P_MSG_FLAG(CMSG_CLOEXEC);
749#undef P_MSG_FLAG
750
751 if (flags)
752 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
753
754 return printed;
755}
756
757#define SCA_MSG_FLAGS syscall_arg__scnprintf_msg_flags
758
51108999
ACM		 759static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
760 struct syscall_arg *arg)
761{
762 size_t printed = 0;
763 int mode = arg->val;
764
765 if (mode == F_OK) /* 0 */
766 return scnprintf(bf, size, "F");
767#define P_MODE(n) \
768 if (mode & n##_OK) { \
769 printed += scnprintf(bf + printed, size - printed, "%s", #n); \
770 mode &= ~n##_OK; \
771 }
772
773 P_MODE(R);
774 P_MODE(W);
775 P_MODE(X);
776#undef P_MODE
777
778 if (mode)
779 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
780
781 return printed;
782}
783
784#define SCA_ACCMODE syscall_arg__scnprintf_access_mode
785
f994592d
ACM		 786static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
787 struct syscall_arg *arg);
788
789#define SCA_FILENAME syscall_arg__scnprintf_filename
790
be65a89a 791static size_t syscall_arg__scnprintf_open_flags(char *bf, size_t size,
01533e97 792 struct syscall_arg *arg)
be65a89a 793{
01533e97 794 int printed = 0, flags = arg->val;
be65a89a
ACM		 795
796 if (!(flags & O_CREAT))
01533e97 797 arg->mask |= 1 << (arg->idx + 1); /* Mask the mode parm */
be65a89a
ACM		 798
799 if (flags == 0)
800 return scnprintf(bf, size, "RDONLY");
801#define P_FLAG(n) \
802 if (flags & O_##n) { \
803 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
804 flags &= ~O_##n; \
805 }
806
807 P_FLAG(APPEND);
808 P_FLAG(ASYNC);
809 P_FLAG(CLOEXEC);
810 P_FLAG(CREAT);
811 P_FLAG(DIRECT);
812 P_FLAG(DIRECTORY);
813 P_FLAG(EXCL);
814 P_FLAG(LARGEFILE);
815 P_FLAG(NOATIME);
816 P_FLAG(NOCTTY);
817#ifdef O_NONBLOCK
818 P_FLAG(NONBLOCK);
819#elif O_NDELAY
820 P_FLAG(NDELAY);
821#endif
822#ifdef O_PATH
823 P_FLAG(PATH);
824#endif
825 P_FLAG(RDWR);
826#ifdef O_DSYNC
827 if ((flags & O_SYNC) == O_SYNC)
828 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", "SYNC");
829 else {
830 P_FLAG(DSYNC);
831 }
832#else
833 P_FLAG(SYNC);
834#endif
835 P_FLAG(TRUNC);
836 P_FLAG(WRONLY);
837#undef P_FLAG
838
839 if (flags)
840 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
841
842 return printed;
843}
844
845#define SCA_OPEN_FLAGS syscall_arg__scnprintf_open_flags
846
a1c2552d
ACM		 847static size_t syscall_arg__scnprintf_perf_flags(char *bf, size_t size,
848 struct syscall_arg *arg)
849{
850 int printed = 0, flags = arg->val;
851
852 if (flags == 0)
853 return 0;
854
855#define P_FLAG(n) \
856 if (flags & PERF_FLAG_##n) { \
857 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
858 flags &= ~PERF_FLAG_##n; \
859 }
860
861 P_FLAG(FD_NO_GROUP);
862 P_FLAG(FD_OUTPUT);
863 P_FLAG(PID_CGROUP);
864 P_FLAG(FD_CLOEXEC);
865#undef P_FLAG
866
867 if (flags)
868 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
869
870 return printed;
871}
872
873#define SCA_PERF_FLAGS syscall_arg__scnprintf_perf_flags
874
49af9e93
ACM		 875static size_t syscall_arg__scnprintf_eventfd_flags(char *bf, size_t size,
876 struct syscall_arg *arg)
877{
878 int printed = 0, flags = arg->val;
879
880 if (flags == 0)
881 return scnprintf(bf, size, "NONE");
882#define P_FLAG(n) \
883 if (flags & EFD_##n) { \
884 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
885 flags &= ~EFD_##n; \
886 }
887
888 P_FLAG(SEMAPHORE);
889 P_FLAG(CLOEXEC);
890 P_FLAG(NONBLOCK);
891#undef P_FLAG
892
893 if (flags)
894 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
895
896 return printed;
897}
898
899#define SCA_EFD_FLAGS syscall_arg__scnprintf_eventfd_flags
900
46cce19b
ACM		 901static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
902 struct syscall_arg *arg)
903{
904 int printed = 0, flags = arg->val;
905
906#define P_FLAG(n) \
907 if (flags & O_##n) { \
908 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
909 flags &= ~O_##n; \
910 }
911
912 P_FLAG(CLOEXEC);
913 P_FLAG(NONBLOCK);
914#undef P_FLAG
915
916 if (flags)
917 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
918
919 return printed;
920}
921
922#define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
923
8bad5b0a
ACM		 924static size_t syscall_arg__scnprintf_signum(char *bf, size_t size, struct syscall_arg *arg)
925{
926 int sig = arg->val;
927
928 switch (sig) {
929#define P_SIGNUM(n) case SIG##n: return scnprintf(bf, size, #n)
930 P_SIGNUM(HUP);
931 P_SIGNUM(INT);
932 P_SIGNUM(QUIT);
933 P_SIGNUM(ILL);
934 P_SIGNUM(TRAP);
935 P_SIGNUM(ABRT);
936 P_SIGNUM(BUS);
937 P_SIGNUM(FPE);
938 P_SIGNUM(KILL);
939 P_SIGNUM(USR1);
940 P_SIGNUM(SEGV);
941 P_SIGNUM(USR2);
942 P_SIGNUM(PIPE);
943 P_SIGNUM(ALRM);
944 P_SIGNUM(TERM);
8bad5b0a
ACM		 945 P_SIGNUM(CHLD);
946 P_SIGNUM(CONT);
947 P_SIGNUM(STOP);
948 P_SIGNUM(TSTP);
949 P_SIGNUM(TTIN);
950 P_SIGNUM(TTOU);
951 P_SIGNUM(URG);
952 P_SIGNUM(XCPU);
953 P_SIGNUM(XFSZ);
954 P_SIGNUM(VTALRM);
955 P_SIGNUM(PROF);
956 P_SIGNUM(WINCH);
957 P_SIGNUM(IO);
958 P_SIGNUM(PWR);
959 P_SIGNUM(SYS);
02c5bb4a
BH		 960#ifdef SIGEMT
961 P_SIGNUM(EMT);
962#endif
963#ifdef SIGSTKFLT
964 P_SIGNUM(STKFLT);
965#endif
966#ifdef SIGSWI
967 P_SIGNUM(SWI);
968#endif
8bad5b0a
ACM		 969 default: break;
970 }
971
972 return scnprintf(bf, size, "%#x", sig);
973}
974
975#define SCA_SIGNUM syscall_arg__scnprintf_signum
976
844ae5b4
ACM		 977#if defined(__i386__) || defined(__x86_64__)
978/*
979 * FIXME: Make this available to all arches.
980 */
78645cf3
ACM		 981#define TCGETS 0x5401
982
983static const char *tioctls[] = {
984 "TCGETS", "TCSETS", "TCSETSW", "TCSETSF", "TCGETA", "TCSETA", "TCSETAW",
985 "TCSETAF", "TCSBRK", "TCXONC", "TCFLSH", "TIOCEXCL", "TIOCNXCL",
986 "TIOCSCTTY", "TIOCGPGRP", "TIOCSPGRP", "TIOCOUTQ", "TIOCSTI",
987 "TIOCGWINSZ", "TIOCSWINSZ", "TIOCMGET", "TIOCMBIS", "TIOCMBIC",
988 "TIOCMSET", "TIOCGSOFTCAR", "TIOCSSOFTCAR", "FIONREAD", "TIOCLINUX",
989 "TIOCCONS", "TIOCGSERIAL", "TIOCSSERIAL", "TIOCPKT", "FIONBIO",
990 "TIOCNOTTY", "TIOCSETD", "TIOCGETD", "TCSBRKP", [0x27] = "TIOCSBRK",
991 "TIOCCBRK", "TIOCGSID", "TCGETS2", "TCSETS2", "TCSETSW2", "TCSETSF2",
992 "TIOCGRS485", "TIOCSRS485", "TIOCGPTN", "TIOCSPTLCK",
993 "TIOCGDEV||TCGETX", "TCSETX", "TCSETXF", "TCSETXW", "TIOCSIG",
994 "TIOCVHANGUP", "TIOCGPKT", "TIOCGPTLCK", "TIOCGEXCL",
995 [0x50] = "FIONCLEX", "FIOCLEX", "FIOASYNC", "TIOCSERCONFIG",
996 "TIOCSERGWILD", "TIOCSERSWILD", "TIOCGLCKTRMIOS", "TIOCSLCKTRMIOS",
997 "TIOCSERGSTRUCT", "TIOCSERGETLSR", "TIOCSERGETMULTI", "TIOCSERSETMULTI",
998 "TIOCMIWAIT", "TIOCGICOUNT", [0x60] = "FIOQSIZE",
999};
1000
1001static DEFINE_STRARRAY_OFFSET(tioctls, 0x5401);
844ae5b4 1002#endif /* defined(__i386__) || defined(__x86_64__) */
78645cf3 1003
453350dd
ACM		 1004#define STRARRAY(arg, name, array) \
1005 .arg_scnprintf = { [arg] = SCA_STRARRAY, }, \
1006 .arg_parm = { [arg] = &strarray__##array, }
1007
514f1c67
ACM		 1008static struct syscall_fmt {
1009 const char *name;
aec1930b 1010 const char *alias;
01533e97 1011 size_t (*arg_scnprintf[6])(char *bf, size_t size, struct syscall_arg *arg);
1f115cb7 1012 void *arg_parm[6];
514f1c67
ACM		 1013 bool errmsg;
1014 bool timeout;
04b34729 1015 bool hexret;
514f1c67 1016} syscall_fmts[] = {
51108999 1017 { .name = "access", .errmsg = true,
34221118
ACM		 1018 .arg_scnprintf = { [0] = SCA_FILENAME, /* filename */
1019 [1] = SCA_ACCMODE, /* mode */ }, },
aec1930b 1020 { .name = "arch_prctl", .errmsg = true, .alias = "prctl", },
729a7841 1021 { .name = "bpf", .errmsg = true, STRARRAY(0, cmd, bpf_cmd), },
beccb2b5
ACM		 1022 { .name = "brk", .hexret = true,
1023 .arg_scnprintf = { [0] = SCA_HEX, /* brk */ }, },
34221118
ACM		 1024 { .name = "chdir", .errmsg = true,
1025 .arg_scnprintf = { [0] = SCA_FILENAME, /* filename */ }, },
1026 { .name = "chmod", .errmsg = true,
1027 .arg_scnprintf = { [0] = SCA_FILENAME, /* filename */ }, },
1028 { .name = "chroot", .errmsg = true,
1029 .arg_scnprintf = { [0] = SCA_FILENAME, /* filename */ }, },
4f8c1b74 1030 { .name = "clock_gettime", .errmsg = true, STRARRAY(0, clk_id, clockid), },
75b757ca 1031 { .name = "close", .errmsg = true,
48000a1a 1032 .arg_scnprintf = { [0] = SCA_CLOSE_FD, /* fd */ }, },
a14bb860 1033 { .name = "connect", .errmsg = true, },
34221118
ACM		 1034 { .name = "creat", .errmsg = true,
1035 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
75b757ca 1036 { .name = "dup", .errmsg = true,
48000a1a 1037 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1038 { .name = "dup2", .errmsg = true,
48000a1a 1039 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1040 { .name = "dup3", .errmsg = true,
48000a1a 1041 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
453350dd 1042 { .name = "epoll_ctl", .errmsg = true, STRARRAY(1, op, epoll_ctl_ops), },
49af9e93
ACM		 1043 { .name = "eventfd2", .errmsg = true,
1044 .arg_scnprintf = { [1] = SCA_EFD_FLAGS, /* flags */ }, },
75b757ca 1045 { .name = "faccessat", .errmsg = true,
34221118
ACM		 1046 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
1047 [1] = SCA_FILENAME, /* filename */ }, },
75b757ca 1048 { .name = "fadvise64", .errmsg = true,
48000a1a 1049 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1050 { .name = "fallocate", .errmsg = true,
48000a1a 1051 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1052 { .name = "fchdir", .errmsg = true,
48000a1a 1053 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1054 { .name = "fchmod", .errmsg = true,
48000a1a 1055 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1056 { .name = "fchmodat", .errmsg = true,
090389b6
ACM		 1057 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */
1058 [1] = SCA_FILENAME, /* filename */ }, },
75b757ca 1059 { .name = "fchown", .errmsg = true,
48000a1a 1060 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1061 { .name = "fchownat", .errmsg = true,
34221118
ACM		 1062 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */
1063 [1] = SCA_FILENAME, /* filename */ }, },
75b757ca
ACM		 1064 { .name = "fcntl", .errmsg = true,
1065 .arg_scnprintf = { [0] = SCA_FD, /* fd */
1066 [1] = SCA_STRARRAY, /* cmd */ },
1067 .arg_parm = { [1] = &strarray__fcntl_cmds, /* cmd */ }, },
1068 { .name = "fdatasync", .errmsg = true,
48000a1a 1069 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
5cea6ff2 1070 { .name = "flock", .errmsg = true,
75b757ca
ACM		 1071 .arg_scnprintf = { [0] = SCA_FD, /* fd */
1072 [1] = SCA_FLOCK, /* cmd */ }, },
1073 { .name = "fsetxattr", .errmsg = true,
48000a1a 1074 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1075 { .name = "fstat", .errmsg = true, .alias = "newfstat",
48000a1a 1076 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1077 { .name = "fstatat", .errmsg = true, .alias = "newfstatat",
34221118
ACM		 1078 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
1079 [1] = SCA_FILENAME, /* filename */ }, },
75b757ca 1080 { .name = "fstatfs", .errmsg = true,
48000a1a 1081 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1082 { .name = "fsync", .errmsg = true,
48000a1a 1083 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1084 { .name = "ftruncate", .errmsg = true,
48000a1a 1085 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
f9da0b0c
ACM		 1086 { .name = "futex", .errmsg = true,
1087 .arg_scnprintf = { [1] = SCA_FUTEX_OP, /* op */ }, },
75b757ca 1088 { .name = "futimesat", .errmsg = true,
090389b6
ACM		 1089 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */
1090 [1] = SCA_FILENAME, /* filename */ }, },
75b757ca 1091 { .name = "getdents", .errmsg = true,
48000a1a 1092 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1093 { .name = "getdents64", .errmsg = true,
48000a1a 1094 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
453350dd
ACM		 1095 { .name = "getitimer", .errmsg = true, STRARRAY(0, which, itimers), },
1096 { .name = "getrlimit", .errmsg = true, STRARRAY(0, resource, rlimit_resources), },
34221118
ACM		 1097 { .name = "getxattr", .errmsg = true,
1098 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
1099 { .name = "inotify_add_watch", .errmsg = true,
1100 .arg_scnprintf = { [1] = SCA_FILENAME, /* pathname */ }, },
beccb2b5 1101 { .name = "ioctl", .errmsg = true,
48000a1a 1102 .arg_scnprintf = { [0] = SCA_FD, /* fd */
844ae5b4
ACM		 1103#if defined(__i386__) || defined(__x86_64__)
1104/*
1105 * FIXME: Make this available to all arches.
1106 */
78645cf3
ACM		 1107 [1] = SCA_STRHEXARRAY, /* cmd */
1108 [2] = SCA_HEX, /* arg */ },
1109 .arg_parm = { [1] = &strarray__tioctls, /* cmd */ }, },
844ae5b4
ACM		 1110#else
1111 [2] = SCA_HEX, /* arg */ }, },
1112#endif
b62bee1b 1113 { .name = "keyctl", .errmsg = true, STRARRAY(0, option, keyctl_options), },
8bad5b0a
ACM		 1114 { .name = "kill", .errmsg = true,
1115 .arg_scnprintf = { [1] = SCA_SIGNUM, /* sig */ }, },
34221118
ACM		 1116 { .name = "lchown", .errmsg = true,
1117 .arg_scnprintf = { [0] = SCA_FILENAME, /* filename */ }, },
1118 { .name = "lgetxattr", .errmsg = true,
1119 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
75b757ca 1120 { .name = "linkat", .errmsg = true,
48000a1a 1121 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
34221118
ACM		 1122 { .name = "listxattr", .errmsg = true,
1123 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
090389b6
ACM		 1124 { .name = "llistxattr", .errmsg = true,
1125 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
1126 { .name = "lremovexattr", .errmsg = true,
1127 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
75b757ca
ACM		 1128 { .name = "lseek", .errmsg = true,
1129 .arg_scnprintf = { [0] = SCA_FD, /* fd */
1130 [2] = SCA_STRARRAY, /* whence */ },
1131 .arg_parm = { [2] = &strarray__whences, /* whence */ }, },
34221118
ACM		 1132 { .name = "lsetxattr", .errmsg = true,
1133 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
090389b6
ACM		 1134 { .name = "lstat", .errmsg = true, .alias = "newlstat",
1135 .arg_scnprintf = { [0] = SCA_FILENAME, /* filename */ }, },
34221118
ACM		 1136 { .name = "lsxattr", .errmsg = true,
1137 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
9e9716d1
ACM		 1138 { .name = "madvise", .errmsg = true,
1139 .arg_scnprintf = { [0] = SCA_HEX, /* start */
1140 [2] = SCA_MADV_BHV, /* behavior */ }, },
34221118
ACM		 1141 { .name = "mkdir", .errmsg = true,
1142 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
75b757ca 1143 { .name = "mkdirat", .errmsg = true,
34221118
ACM		 1144 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */
1145 [1] = SCA_FILENAME, /* pathname */ }, },
1146 { .name = "mknod", .errmsg = true,
1147 .arg_scnprintf = { [0] = SCA_FILENAME, /* filename */ }, },
75b757ca 1148 { .name = "mknodat", .errmsg = true,
090389b6
ACM		 1149 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */
1150 [1] = SCA_FILENAME, /* filename */ }, },
3d903aa7
ACM		 1151 { .name = "mlock", .errmsg = true,
1152 .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
1153 { .name = "mlockall", .errmsg = true,
1154 .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
beccb2b5 1155 { .name = "mmap", .hexret = true,
ae685380 1156 .arg_scnprintf = { [0] = SCA_HEX, /* addr */
941557e0 1157 [2] = SCA_MMAP_PROT, /* prot */
73faab3a
NK		 1158 [3] = SCA_MMAP_FLAGS, /* flags */
1159 [4] = SCA_FD, /* fd */ }, },
beccb2b5 1160 { .name = "mprotect", .errmsg = true,
ae685380
ACM		 1161 .arg_scnprintf = { [0] = SCA_HEX, /* start */
1162 [2] = SCA_MMAP_PROT, /* prot */ }, },
090389b6
ACM		 1163 { .name = "mq_unlink", .errmsg = true,
1164 .arg_scnprintf = { [0] = SCA_FILENAME, /* u_name */ }, },
ae685380
ACM		 1165 { .name = "mremap", .hexret = true,
1166 .arg_scnprintf = { [0] = SCA_HEX, /* addr */
86998dda 1167 [3] = SCA_MREMAP_FLAGS, /* flags */
ae685380 1168 [4] = SCA_HEX, /* new_addr */ }, },
3d903aa7
ACM		 1169 { .name = "munlock", .errmsg = true,
1170 .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
beccb2b5
ACM		 1171 { .name = "munmap", .errmsg = true,
1172 .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
75b757ca 1173 { .name = "name_to_handle_at", .errmsg = true,
48000a1a 1174 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
75b757ca 1175 { .name = "newfstatat", .errmsg = true,
34221118
ACM		 1176 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
1177 [1] = SCA_FILENAME, /* filename */ }, },
be65a89a 1178 { .name = "open", .errmsg = true,
f994592d
ACM		 1179 .arg_scnprintf = { [0] = SCA_FILENAME, /* filename */
1180 [1] = SCA_OPEN_FLAGS, /* flags */ }, },
31cd3855 1181 { .name = "open_by_handle_at", .errmsg = true,
75b757ca
ACM		 1182 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
1183 [2] = SCA_OPEN_FLAGS, /* flags */ }, },
31cd3855 1184 { .name = "openat", .errmsg = true,
75b757ca 1185 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
34221118 1186 [1] = SCA_FILENAME, /* filename */
75b757ca 1187 [2] = SCA_OPEN_FLAGS, /* flags */ }, },
a1c2552d
ACM		 1188 { .name = "perf_event_open", .errmsg = true,
1189 .arg_scnprintf = { [1] = SCA_INT, /* pid */
1190 [2] = SCA_INT, /* cpu */
1191 [3] = SCA_FD, /* group_fd */
1192 [4] = SCA_PERF_FLAGS, /* flags */ }, },
46cce19b
ACM		 1193 { .name = "pipe2", .errmsg = true,
1194 .arg_scnprintf = { [1] = SCA_PIPE_FLAGS, /* flags */ }, },
aec1930b
ACM		 1195 { .name = "poll", .errmsg = true, .timeout = true, },
1196 { .name = "ppoll", .errmsg = true, .timeout = true, },
75b757ca 1197 { .name = "pread", .errmsg = true, .alias = "pread64",
48000a1a 1198 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1199 { .name = "preadv", .errmsg = true, .alias = "pread",
48000a1a 1200 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
453350dd 1201 { .name = "prlimit64", .errmsg = true, STRARRAY(1, resource, rlimit_resources), },
75b757ca 1202 { .name = "pwrite", .errmsg = true, .alias = "pwrite64",
48000a1a 1203 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1204 { .name = "pwritev", .errmsg = true,
48000a1a 1205 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1206 { .name = "read", .errmsg = true,
48000a1a 1207 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
34221118
ACM		 1208 { .name = "readlink", .errmsg = true,
1209 .arg_scnprintf = { [0] = SCA_FILENAME, /* path */ }, },
75b757ca 1210 { .name = "readlinkat", .errmsg = true,
34221118
ACM		 1211 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
1212 [1] = SCA_FILENAME, /* pathname */ }, },
75b757ca 1213 { .name = "readv", .errmsg = true,
48000a1a 1214 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
b2cc99fd 1215 { .name = "recvfrom", .errmsg = true,
8d8c66a2
ACM		 1216 .arg_scnprintf = { [0] = SCA_FD, /* fd */
1217 [3] = SCA_MSG_FLAGS, /* flags */ }, },
b2cc99fd 1218 { .name = "recvmmsg", .errmsg = true,
8d8c66a2
ACM		 1219 .arg_scnprintf = { [0] = SCA_FD, /* fd */
1220 [3] = SCA_MSG_FLAGS, /* flags */ }, },
b2cc99fd 1221 { .name = "recvmsg", .errmsg = true,
8d8c66a2
ACM		 1222 .arg_scnprintf = { [0] = SCA_FD, /* fd */
1223 [2] = SCA_MSG_FLAGS, /* flags */ }, },
34221118
ACM		 1224 { .name = "removexattr", .errmsg = true,
1225 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
75b757ca 1226 { .name = "renameat", .errmsg = true,
48000a1a 1227 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
34221118
ACM		 1228 { .name = "rmdir", .errmsg = true,
1229 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
8bad5b0a
ACM		 1230 { .name = "rt_sigaction", .errmsg = true,
1231 .arg_scnprintf = { [0] = SCA_SIGNUM, /* sig */ }, },
453350dd 1232 { .name = "rt_sigprocmask", .errmsg = true, STRARRAY(0, how, sighow), },
8bad5b0a
ACM		 1233 { .name = "rt_sigqueueinfo", .errmsg = true,
1234 .arg_scnprintf = { [1] = SCA_SIGNUM, /* sig */ }, },
1235 { .name = "rt_tgsigqueueinfo", .errmsg = true,
1236 .arg_scnprintf = { [2] = SCA_SIGNUM, /* sig */ }, },
aec1930b 1237 { .name = "select", .errmsg = true, .timeout = true, },
b2cc99fd 1238 { .name = "sendmmsg", .errmsg = true,
8d8c66a2
ACM		 1239 .arg_scnprintf = { [0] = SCA_FD, /* fd */
1240 [3] = SCA_MSG_FLAGS, /* flags */ }, },
b2cc99fd 1241 { .name = "sendmsg", .errmsg = true,
8d8c66a2
ACM		 1242 .arg_scnprintf = { [0] = SCA_FD, /* fd */
1243 [2] = SCA_MSG_FLAGS, /* flags */ }, },
b2cc99fd 1244 { .name = "sendto", .errmsg = true,
8d8c66a2
ACM		 1245 .arg_scnprintf = { [0] = SCA_FD, /* fd */
1246 [3] = SCA_MSG_FLAGS, /* flags */ }, },
453350dd
ACM		 1247 { .name = "setitimer", .errmsg = true, STRARRAY(0, which, itimers), },
1248 { .name = "setrlimit", .errmsg = true, STRARRAY(0, resource, rlimit_resources), },
34221118
ACM		 1249 { .name = "setxattr", .errmsg = true,
1250 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
75b757ca 1251 { .name = "shutdown", .errmsg = true,
48000a1a 1252 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
e10bce81 1253 { .name = "socket", .errmsg = true,
a28b24b2
ACM		 1254 .arg_scnprintf = { [0] = SCA_STRARRAY, /* family */
1255 [1] = SCA_SK_TYPE, /* type */ },
07120aa5
ACM		 1256 .arg_parm = { [0] = &strarray__socket_families, /* family */ }, },
1257 { .name = "socketpair", .errmsg = true,
1258 .arg_scnprintf = { [0] = SCA_STRARRAY, /* family */
1259 [1] = SCA_SK_TYPE, /* type */ },
e10bce81 1260 .arg_parm = { [0] = &strarray__socket_families, /* family */ }, },
090389b6
ACM		 1261 { .name = "stat", .errmsg = true, .alias = "newstat",
1262 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
34221118
ACM		 1263 { .name = "statfs", .errmsg = true,
1264 .arg_scnprintf = { [0] = SCA_FILENAME, /* pathname */ }, },
1265 { .name = "swapoff", .errmsg = true,
1266 .arg_scnprintf = { [0] = SCA_FILENAME, /* specialfile */ }, },
1267 { .name = "swapon", .errmsg = true,
1268 .arg_scnprintf = { [0] = SCA_FILENAME, /* specialfile */ }, },
75b757ca 1269 { .name = "symlinkat", .errmsg = true,
48000a1a 1270 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
8bad5b0a
ACM		 1271 { .name = "tgkill", .errmsg = true,
1272 .arg_scnprintf = { [2] = SCA_SIGNUM, /* sig */ }, },
1273 { .name = "tkill", .errmsg = true,
1274 .arg_scnprintf = { [1] = SCA_SIGNUM, /* sig */ }, },
34221118
ACM		 1275 { .name = "truncate", .errmsg = true,
1276 .arg_scnprintf = { [0] = SCA_FILENAME, /* path */ }, },
e5959683 1277 { .name = "uname", .errmsg = true, .alias = "newuname", },
75b757ca 1278 { .name = "unlinkat", .errmsg = true,
34221118
ACM		 1279 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
1280 [1] = SCA_FILENAME, /* pathname */ }, },
1281 { .name = "utime", .errmsg = true,
1282 .arg_scnprintf = { [0] = SCA_FILENAME, /* filename */ }, },
75b757ca 1283 { .name = "utimensat", .errmsg = true,
34221118
ACM		 1284 .arg_scnprintf = { [0] = SCA_FDAT, /* dirfd */
1285 [1] = SCA_FILENAME, /* filename */ }, },
1286 { .name = "utimes", .errmsg = true,
1287 .arg_scnprintf = { [0] = SCA_FILENAME, /* filename */ }, },
090389b6
ACM		 1288 { .name = "vmsplice", .errmsg = true,
1289 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1290 { .name = "write", .errmsg = true,
48000a1a 1291 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1292 { .name = "writev", .errmsg = true,
48000a1a 1293 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
514f1c67
ACM		 1294};
1295
1296static int syscall_fmt__cmp(const void *name, const void *fmtp)
1297{
1298 const struct syscall_fmt *fmt = fmtp;
1299 return strcmp(name, fmt->name);
1300}
1301
1302static struct syscall_fmt *syscall_fmt__find(const char *name)
1303{
1304 const int nmemb = ARRAY_SIZE(syscall_fmts);
1305 return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1306}
1307
1308struct syscall {
1309 struct event_format *tp_format;
f208bd8d
ACM		 1310 int nr_args;
1311 struct format_field *args;
514f1c67 1312 const char *name;
5089f20e 1313 bool is_exit;
514f1c67 1314 struct syscall_fmt *fmt;
01533e97 1315 size_t (**arg_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1f115cb7 1316 void **arg_parm;
514f1c67
ACM		 1317};
1318
60c907ab
ACM		 1319static size_t fprintf_duration(unsigned long t, FILE *fp)
1320{
1321 double duration = (double)t / NSEC_PER_MSEC;
1322 size_t printed = fprintf(fp, "(");
1323
1324 if (duration >= 1.0)
1325 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1326 else if (duration >= 0.01)
1327 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1328 else
1329 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
c24ff998 1330 return printed + fprintf(fp, "): ");
60c907ab
ACM		 1331}
1332
f994592d
ACM		 1333/**
1334 * filename.ptr: The filename char pointer that will be vfs_getname'd
1335 * filename.entry_str_pos: Where to insert the string translated from
1336 * filename.ptr by the vfs_getname tracepoint/kprobe.
1337 */
752fde44
ACM		 1338struct thread_trace {
1339 u64 entry_time;
1340 u64 exit_time;
1341 bool entry_pending;
efd5745e 1342 unsigned long nr_events;
a2ea67d7 1343 unsigned long pfmaj, pfmin;
752fde44 1344 char *entry_str;
1302d88e 1345 double runtime_ms;
f994592d
ACM		 1346 struct {
1347 unsigned long ptr;
7f4f8001
ACM		 1348 short int entry_str_pos;
1349 bool pending_open;
1350 unsigned int namelen;
1351 char *name;
f994592d 1352 } filename;
75b757ca
ACM		 1353 struct {
1354 int max;
1355 char **table;
1356 } paths;
bf2575c1
DA		 1357
1358 struct intlist *syscall_stats;
752fde44
ACM		 1359};
1360
1361static struct thread_trace *thread_trace__new(void)
1362{
75b757ca
ACM		 1363 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
1364
1365 if (ttrace)
1366 ttrace->paths.max = -1;
1367
bf2575c1
DA		 1368 ttrace->syscall_stats = intlist__new(NULL);
1369
75b757ca 1370 return ttrace;
752fde44
ACM		 1371}
1372
c24ff998 1373static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
752fde44 1374{
efd5745e
ACM		 1375 struct thread_trace *ttrace;
1376
752fde44
ACM		 1377 if (thread == NULL)
1378 goto fail;
1379
89dceb22
NK		 1380 if (thread__priv(thread) == NULL)
1381 thread__set_priv(thread, thread_trace__new());
48000a1a 1382
89dceb22 1383 if (thread__priv(thread) == NULL)
752fde44
ACM		 1384 goto fail;
1385
89dceb22 1386 ttrace = thread__priv(thread);
efd5745e
ACM		 1387 ++ttrace->nr_events;
1388
1389 return ttrace;
752fde44 1390fail:
c24ff998 1391 color_fprintf(fp, PERF_COLOR_RED,
752fde44
ACM		 1392 "WARNING: not enough memory, dropping samples!\n");
1393 return NULL;
1394}
1395
598d02c5
SF		 1396#define TRACE_PFMAJ (1 << 0)
1397#define TRACE_PFMIN (1 << 1)
1398
e4d44e83
ACM		 1399static const size_t trace__entry_str_size = 2048;
1400
514f1c67 1401struct trace {
c24ff998 1402 struct perf_tool tool;
c522739d
ACM		 1403 struct {
1404 int machine;
1405 int open_id;
1406 } audit;
514f1c67
ACM		 1407 struct {
1408 int max;
1409 struct syscall *table;
c27366f0 1410 struct {
8b3ce757
ACM		 1411 struct perf_evsel *sys_enter,
1412 *sys_exit;
c27366f0 1413 } events;
514f1c67 1414 } syscalls;
b4006796 1415 struct record_opts opts;
14a052df 1416 struct perf_evlist *evlist;
8fb598e5 1417 struct machine *host;
e596663e 1418 struct thread *current;
752fde44 1419 u64 base_time;
c24ff998 1420 FILE *output;
efd5745e 1421 unsigned long nr_events;
b059efdf 1422 struct strlist *ev_qualifier;
8b3ce757
ACM		 1423 struct {
1424 size_t nr;
1425 int *entries;
1426 } ev_qualifier_ids;
bdc89661
DA		 1427 struct intlist *tid_list;
1428 struct intlist *pid_list;
f078c385
ACM		 1429 struct {
1430 size_t nr;
1431 pid_t *entries;
1432 } filter_pids;
98eafce6
ACM		 1433 double duration_filter;
1434 double runtime_ms;
1435 struct {
1436 u64 vfs_getname,
1437 proc_getname;
1438 } stats;
1439 bool not_ev_qualifier;
1440 bool live;
1441 bool full_time;
1302d88e 1442 bool sched;
752fde44 1443 bool multiple_threads;
bf2575c1 1444 bool summary;
fd2eabaf 1445 bool summary_only;
50c95cbd 1446 bool show_comm;
c522739d 1447 bool show_tool_stats;
e281a960 1448 bool trace_syscalls;
e366a6d8 1449 bool force;
08c98776 1450 bool vfs_getname;
598d02c5 1451 int trace_pgfaults;
514f1c67
ACM		 1452};
1453
97119f37 1454static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
75b757ca 1455{
89dceb22 1456 struct thread_trace *ttrace = thread__priv(thread);
75b757ca
ACM		 1457
1458 if (fd > ttrace->paths.max) {
1459 char **npath = realloc(ttrace->paths.table, (fd + 1) * sizeof(char *));
1460
1461 if (npath == NULL)
1462 return -1;
1463
1464 if (ttrace->paths.max != -1) {
1465 memset(npath + ttrace->paths.max + 1, 0,
1466 (fd - ttrace->paths.max) * sizeof(char *));
1467 } else {
1468 memset(npath, 0, (fd + 1) * sizeof(char *));
1469 }
1470
1471 ttrace->paths.table = npath;
1472 ttrace->paths.max = fd;
1473 }
1474
1475 ttrace->paths.table[fd] = strdup(pathname);
1476
1477 return ttrace->paths.table[fd] != NULL ? 0 : -1;
1478}
1479
97119f37
ACM		 1480static int thread__read_fd_path(struct thread *thread, int fd)
1481{
1482 char linkname[PATH_MAX], pathname[PATH_MAX];
1483 struct stat st;
1484 int ret;
1485
1486 if (thread->pid_ == thread->tid) {
1487 scnprintf(linkname, sizeof(linkname),
1488 "/proc/%d/fd/%d", thread->pid_, fd);
1489 } else {
1490 scnprintf(linkname, sizeof(linkname),
1491 "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1492 }
1493
1494 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1495 return -1;
1496
1497 ret = readlink(linkname, pathname, sizeof(pathname));
1498
1499 if (ret < 0 || ret > st.st_size)
1500 return -1;
1501
1502 pathname[ret] = '\0';
1503 return trace__set_fd_pathname(thread, fd, pathname);
1504}
1505
c522739d
ACM		 1506static const char *thread__fd_path(struct thread *thread, int fd,
1507 struct trace *trace)
75b757ca 1508{
89dceb22 1509 struct thread_trace *ttrace = thread__priv(thread);
75b757ca
ACM		 1510
1511 if (ttrace == NULL)
1512 return NULL;
1513
1514 if (fd < 0)
1515 return NULL;
1516
cdcd1e6b 1517 if ((fd > ttrace->paths.max || ttrace->paths.table[fd] == NULL)) {
c522739d
ACM		 1518 if (!trace->live)
1519 return NULL;
1520 ++trace->stats.proc_getname;
cdcd1e6b 1521 if (thread__read_fd_path(thread, fd))
c522739d
ACM		 1522 return NULL;
1523 }
75b757ca
ACM		 1524
1525 return ttrace->paths.table[fd];
1526}
1527
1528static size_t syscall_arg__scnprintf_fd(char *bf, size_t size,
1529 struct syscall_arg *arg)
1530{
1531 int fd = arg->val;
1532 size_t printed = scnprintf(bf, size, "%d", fd);
c522739d 1533 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
75b757ca
ACM		 1534
1535 if (path)
1536 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1537
1538 return printed;
1539}
1540
1541static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1542 struct syscall_arg *arg)
1543{
1544 int fd = arg->val;
1545 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
89dceb22 1546 struct thread_trace *ttrace = thread__priv(arg->thread);
75b757ca 1547
04662523
ACM		 1548 if (ttrace && fd >= 0 && fd <= ttrace->paths.max)
1549 zfree(&ttrace->paths.table[fd]);
75b757ca
ACM		 1550
1551 return printed;
1552}
1553
f994592d
ACM		 1554static void thread__set_filename_pos(struct thread *thread, const char *bf,
1555 unsigned long ptr)
1556{
1557 struct thread_trace *ttrace = thread__priv(thread);
1558
1559 ttrace->filename.ptr = ptr;
1560 ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1561}
1562
1563static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1564 struct syscall_arg *arg)
1565{
1566 unsigned long ptr = arg->val;
1567
1568 if (!arg->trace->vfs_getname)
1569 return scnprintf(bf, size, "%#x", ptr);
1570
1571 thread__set_filename_pos(arg->thread, bf, ptr);
1572 return 0;
1573}
1574
ae9ed035
ACM		 1575static bool trace__filter_duration(struct trace *trace, double t)
1576{
1577 return t < (trace->duration_filter * NSEC_PER_MSEC);
1578}
1579
752fde44
ACM		 1580static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1581{
1582 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1583
60c907ab 1584 return fprintf(fp, "%10.3f ", ts);
752fde44
ACM		 1585}
1586
f15eb531 1587static bool done = false;
ba209f85 1588static bool interrupted = false;
f15eb531 1589
ba209f85 1590static void sig_handler(int sig)
f15eb531
NK		 1591{
1592 done = true;
ba209f85 1593 interrupted = sig == SIGINT;
f15eb531
NK		 1594}
1595
752fde44 1596static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
60c907ab 1597 u64 duration, u64 tstamp, FILE *fp)
752fde44
ACM		 1598{
1599 size_t printed = trace__fprintf_tstamp(trace, tstamp, fp);
60c907ab 1600 printed += fprintf_duration(duration, fp);
752fde44 1601
50c95cbd
ACM		 1602 if (trace->multiple_threads) {
1603 if (trace->show_comm)
1902efe7 1604 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
38051234 1605 printed += fprintf(fp, "%d ", thread->tid);
50c95cbd 1606 }
752fde44
ACM		 1607
1608 return printed;
1609}
1610
c24ff998 1611static int trace__process_event(struct trace *trace, struct machine *machine,
162f0bef 1612 union perf_event *event, struct perf_sample *sample)
752fde44
ACM		 1613{
1614 int ret = 0;
1615
1616 switch (event->header.type) {
1617 case PERF_RECORD_LOST:
c24ff998 1618 color_fprintf(trace->output, PERF_COLOR_RED,
752fde44 1619 "LOST %" PRIu64 " events!\n", event->lost.lost);
162f0bef 1620 ret = machine__process_lost_event(machine, event, sample);
752fde44 1621 default:
162f0bef 1622 ret = machine__process_event(machine, event, sample);
752fde44
ACM		 1623 break;
1624 }
1625
1626 return ret;
1627}
1628
c24ff998 1629static int trace__tool_process(struct perf_tool *tool,
752fde44 1630 union perf_event *event,
162f0bef 1631 struct perf_sample *sample,
752fde44
ACM		 1632 struct machine *machine)
1633{
c24ff998 1634 struct trace *trace = container_of(tool, struct trace, tool);
162f0bef 1635 return trace__process_event(trace, machine, event, sample);
752fde44
ACM		 1636}
1637
1638static int trace__symbols_init(struct trace *trace, struct perf_evlist *evlist)
1639{
0a7e6d1b 1640 int err = symbol__init(NULL);
752fde44
ACM		 1641
1642 if (err)
1643 return err;
1644
8fb598e5
DA		 1645 trace->host = machine__new_host();
1646 if (trace->host == NULL)
1647 return -ENOMEM;
752fde44 1648
959c2199 1649 if (trace_event__register_resolver(trace->host, machine__resolve_kernel_addr) < 0)
706c3da4
ACM		 1650 return -errno;
1651
a33fbd56 1652 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
9d9cad76
KL		 1653 evlist->threads, trace__tool_process, false,
1654 trace->opts.proc_map_timeout);
752fde44
ACM		 1655 if (err)
1656 symbol__exit();
1657
1658 return err;
1659}
1660
13d4ff3e
ACM		 1661static int syscall__set_arg_fmts(struct syscall *sc)
1662{
1663 struct format_field *field;
1664 int idx = 0;
1665
f208bd8d 1666 sc->arg_scnprintf = calloc(sc->nr_args, sizeof(void *));
13d4ff3e
ACM		 1667 if (sc->arg_scnprintf == NULL)
1668 return -1;
1669
1f115cb7
ACM		 1670 if (sc->fmt)
1671 sc->arg_parm = sc->fmt->arg_parm;
1672
f208bd8d 1673 for (field = sc->args; field; field = field->next) {
beccb2b5
ACM		 1674 if (sc->fmt && sc->fmt->arg_scnprintf[idx])
1675 sc->arg_scnprintf[idx] = sc->fmt->arg_scnprintf[idx];
1676 else if (field->flags & FIELD_IS_POINTER)
13d4ff3e
ACM		 1677 sc->arg_scnprintf[idx] = syscall_arg__scnprintf_hex;
1678 ++idx;
1679 }
1680
1681 return 0;
1682}
1683
514f1c67
ACM		 1684static int trace__read_syscall_info(struct trace *trace, int id)
1685{
1686 char tp_name[128];
1687 struct syscall *sc;
c522739d 1688 const char *name = audit_syscall_to_name(id, trace->audit.machine);
3a531260
ACM		 1689
1690 if (name == NULL)
1691 return -1;
514f1c67
ACM		 1692
1693 if (id > trace->syscalls.max) {
1694 struct syscall *nsyscalls = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1695
1696 if (nsyscalls == NULL)
1697 return -1;
1698
1699 if (trace->syscalls.max != -1) {
1700 memset(nsyscalls + trace->syscalls.max + 1, 0,
1701 (id - trace->syscalls.max) * sizeof(*sc));
1702 } else {
1703 memset(nsyscalls, 0, (id + 1) * sizeof(*sc));
1704 }
1705
1706 trace->syscalls.table = nsyscalls;
1707 trace->syscalls.max = id;
1708 }
1709
1710 sc = trace->syscalls.table + id;
3a531260 1711 sc->name = name;
2ae3a312 1712
3a531260 1713 sc->fmt = syscall_fmt__find(sc->name);
514f1c67 1714
aec1930b 1715 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
97978b3e 1716 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
aec1930b 1717
8dd2a131 1718 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
aec1930b 1719 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
97978b3e 1720 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
aec1930b 1721 }
514f1c67 1722
8dd2a131 1723 if (IS_ERR(sc->tp_format))
13d4ff3e
ACM		 1724 return -1;
1725
f208bd8d
ACM		 1726 sc->args = sc->tp_format->format.fields;
1727 sc->nr_args = sc->tp_format->format.nr_fields;
1728 /* drop nr field - not relevant here; does not exist on older kernels */
1729 if (sc->args && strcmp(sc->args->name, "nr") == 0) {
1730 sc->args = sc->args->next;
1731 --sc->nr_args;
1732 }
1733
5089f20e
ACM		 1734 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1735
13d4ff3e 1736 return syscall__set_arg_fmts(sc);
514f1c67
ACM		 1737}
1738
d0cc439b
ACM		 1739static int trace__validate_ev_qualifier(struct trace *trace)
1740{
8b3ce757 1741 int err = 0, i;
d0cc439b
ACM		 1742 struct str_node *pos;
1743
8b3ce757
ACM		 1744 trace->ev_qualifier_ids.nr = strlist__nr_entries(trace->ev_qualifier);
1745 trace->ev_qualifier_ids.entries = malloc(trace->ev_qualifier_ids.nr *
1746 sizeof(trace->ev_qualifier_ids.entries[0]));
1747
1748 if (trace->ev_qualifier_ids.entries == NULL) {
1749 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1750 trace->output);
1751 err = -EINVAL;
1752 goto out;
1753 }
1754
1755 i = 0;
1756
d0cc439b
ACM		 1757 strlist__for_each(pos, trace->ev_qualifier) {
1758 const char *sc = pos->s;
8b3ce757 1759 int id = audit_name_to_syscall(sc, trace->audit.machine);
d0cc439b 1760
8b3ce757 1761 if (id < 0) {
d0cc439b
ACM		 1762 if (err == 0) {
1763 fputs("Error:\tInvalid syscall ", trace->output);
1764 err = -EINVAL;
1765 } else {
1766 fputs(", ", trace->output);
1767 }
1768
1769 fputs(sc, trace->output);
1770 }
8b3ce757
ACM		 1771
1772 trace->ev_qualifier_ids.entries[i++] = id;
d0cc439b
ACM		 1773 }
1774
1775 if (err < 0) {
1776 fputs("\nHint:\ttry 'perf list syscalls:sys_enter_*'"
1777 "\nHint:\tand: 'man syscalls'\n", trace->output);
8b3ce757
ACM		 1778 zfree(&trace->ev_qualifier_ids.entries);
1779 trace->ev_qualifier_ids.nr = 0;
d0cc439b 1780 }
8b3ce757 1781out:
d0cc439b
ACM		 1782 return err;
1783}
1784
55d43bca
DA		 1785/*
1786 * args is to be interpreted as a series of longs but we need to handle
1787 * 8-byte unaligned accesses. args points to raw_data within the event
1788 * and raw_data is guaranteed to be 8-byte unaligned because it is
1789 * preceded by raw_size which is a u32. So we need to copy args to a temp
1790 * variable to read it. Most notably this avoids extended load instructions
1791 * on unaligned addresses
1792 */
1793
752fde44 1794static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
55d43bca 1795 unsigned char *args, struct trace *trace,
75b757ca 1796 struct thread *thread)
514f1c67 1797{
514f1c67 1798 size_t printed = 0;
55d43bca
DA		 1799 unsigned char *p;
1800 unsigned long val;
514f1c67 1801
f208bd8d 1802 if (sc->args != NULL) {
514f1c67 1803 struct format_field *field;
01533e97
ACM		 1804 u8 bit = 1;
1805 struct syscall_arg arg = {
75b757ca
ACM		 1806 .idx = 0,
1807 .mask = 0,
1808 .trace = trace,
1809 .thread = thread,
01533e97 1810 };
6e7eeb51 1811
f208bd8d 1812 for (field = sc->args; field;
01533e97
ACM		 1813 field = field->next, ++arg.idx, bit <<= 1) {
1814 if (arg.mask & bit)
6e7eeb51 1815 continue;
55d43bca
DA		 1816
1817 /* special care for unaligned accesses */
1818 p = args + sizeof(unsigned long) * arg.idx;
1819 memcpy(&val, p, sizeof(val));
1820
4aa58232
ACM		 1821 /*
1822 * Suppress this argument if its value is zero and
1823 * and we don't have a string associated in an
1824 * strarray for it.
1825 */
55d43bca 1826 if (val == 0 &&
4aa58232
ACM		 1827 !(sc->arg_scnprintf &&
1828 sc->arg_scnprintf[arg.idx] == SCA_STRARRAY &&
1829 sc->arg_parm[arg.idx]))
22ae5cf1
ACM		 1830 continue;
1831
752fde44 1832 printed += scnprintf(bf + printed, size - printed,
13d4ff3e 1833 "%s%s: ", printed ? ", " : "", field->name);
01533e97 1834 if (sc->arg_scnprintf && sc->arg_scnprintf[arg.idx]) {
55d43bca 1835 arg.val = val;
1f115cb7
ACM		 1836 if (sc->arg_parm)
1837 arg.parm = sc->arg_parm[arg.idx];
01533e97
ACM		 1838 printed += sc->arg_scnprintf[arg.idx](bf + printed,
1839 size - printed, &arg);
6e7eeb51 1840 } else {
13d4ff3e 1841 printed += scnprintf(bf + printed, size - printed,
55d43bca 1842 "%ld", val);
6e7eeb51 1843 }
514f1c67
ACM		 1844 }
1845 } else {
01533e97
ACM		 1846 int i = 0;
1847
514f1c67 1848 while (i < 6) {
55d43bca
DA		 1849 /* special care for unaligned accesses */
1850 p = args + sizeof(unsigned long) * i;
1851 memcpy(&val, p, sizeof(val));
752fde44
ACM		 1852 printed += scnprintf(bf + printed, size - printed,
1853 "%sarg%d: %ld",
55d43bca 1854 printed ? ", " : "", i, val);
514f1c67
ACM		 1855 ++i;
1856 }
1857 }
1858
1859 return printed;
1860}
1861
ba3d7dee 1862typedef int (*tracepoint_handler)(struct trace *trace, struct perf_evsel *evsel,
0c82adcf 1863 union perf_event *event,
ba3d7dee
ACM		 1864 struct perf_sample *sample);
1865
1866static struct syscall *trace__syscall_info(struct trace *trace,
bf2575c1 1867 struct perf_evsel *evsel, int id)
ba3d7dee 1868{
ba3d7dee
ACM		 1869
1870 if (id < 0) {
adaa18bf
ACM		 1871
1872 /*
1873 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
1874 * before that, leaving at a higher verbosity level till that is
1875 * explained. Reproduced with plain ftrace with:
1876 *
1877 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
1878 * grep "NR -1 " /t/trace_pipe
1879 *
1880 * After generating some load on the machine.
1881 */
1882 if (verbose > 1) {
1883 static u64 n;
1884 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
1885 id, perf_evsel__name(evsel), ++n);
1886 }
ba3d7dee
ACM		 1887 return NULL;
1888 }
1889
1890 if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL) &&
1891 trace__read_syscall_info(trace, id))
1892 goto out_cant_read;
1893
1894 if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL))
1895 goto out_cant_read;
1896
1897 return &trace->syscalls.table[id];
1898
1899out_cant_read:
7c304ee0
ACM		 1900 if (verbose) {
1901 fprintf(trace->output, "Problems reading syscall %d", id);
1902 if (id <= trace->syscalls.max && trace->syscalls.table[id].name != NULL)
1903 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
1904 fputs(" information\n", trace->output);
1905 }
ba3d7dee
ACM		 1906 return NULL;
1907}
1908
bf2575c1
DA		 1909static void thread__update_stats(struct thread_trace *ttrace,
1910 int id, struct perf_sample *sample)
1911{
1912 struct int_node *inode;
1913 struct stats *stats;
1914 u64 duration = 0;
1915
1916 inode = intlist__findnew(ttrace->syscall_stats, id);
1917 if (inode == NULL)
1918 return;
1919
1920 stats = inode->priv;
1921 if (stats == NULL) {
1922 stats = malloc(sizeof(struct stats));
1923 if (stats == NULL)
1924 return;
1925 init_stats(stats);
1926 inode->priv = stats;
1927 }
1928
1929 if (ttrace->entry_time && sample->time > ttrace->entry_time)
1930 duration = sample->time - ttrace->entry_time;
1931
1932 update_stats(stats, duration);
1933}
1934
e596663e
ACM		 1935static int trace__printf_interrupted_entry(struct trace *trace, struct perf_sample *sample)
1936{
1937 struct thread_trace *ttrace;
1938 u64 duration;
1939 size_t printed;
1940
1941 if (trace->current == NULL)
1942 return 0;
1943
1944 ttrace = thread__priv(trace->current);
1945
1946 if (!ttrace->entry_pending)
1947 return 0;
1948
1949 duration = sample->time - ttrace->entry_time;
1950
1951 printed = trace__fprintf_entry_head(trace, trace->current, duration, sample->time, trace->output);
1952 printed += fprintf(trace->output, "%-70s) ...\n", ttrace->entry_str);
1953 ttrace->entry_pending = false;
1954
1955 return printed;
1956}
1957
ba3d7dee 1958static int trace__sys_enter(struct trace *trace, struct perf_evsel *evsel,
0c82adcf 1959 union perf_event *event __maybe_unused,
ba3d7dee
ACM		 1960 struct perf_sample *sample)
1961{
752fde44 1962 char *msg;
ba3d7dee 1963 void *args;
752fde44 1964 size_t printed = 0;
2ae3a312 1965 struct thread *thread;
b91fc39f 1966 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
bf2575c1 1967 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2ae3a312
ACM		 1968 struct thread_trace *ttrace;
1969
1970 if (sc == NULL)
1971 return -1;
ba3d7dee 1972
8fb598e5 1973 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
c24ff998 1974 ttrace = thread__trace(thread, trace->output);
2ae3a312 1975 if (ttrace == NULL)
b91fc39f 1976 goto out_put;
ba3d7dee 1977
77170988 1978 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
752fde44
ACM		 1979
1980 if (ttrace->entry_str == NULL) {
e4d44e83 1981 ttrace->entry_str = malloc(trace__entry_str_size);
752fde44 1982 if (!ttrace->entry_str)
b91fc39f 1983 goto out_put;
752fde44
ACM		 1984 }
1985
13f22a2d 1986 if (!trace->summary_only)
6ebad5c1 1987 trace__printf_interrupted_entry(trace, sample);
e596663e 1988
752fde44
ACM		 1989 ttrace->entry_time = sample->time;
1990 msg = ttrace->entry_str;
e4d44e83 1991 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
752fde44 1992
e4d44e83 1993 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
75b757ca 1994 args, trace, thread);
752fde44 1995
5089f20e 1996 if (sc->is_exit) {
fd2eabaf 1997 if (!trace->duration_filter && !trace->summary_only) {
c24ff998
ACM		 1998 trace__fprintf_entry_head(trace, thread, 1, sample->time, trace->output);
1999 fprintf(trace->output, "%-70s\n", ttrace->entry_str);
ae9ed035 2000 }
7f4f8001 2001 } else {
752fde44 2002 ttrace->entry_pending = true;
7f4f8001
ACM		 2003 /* See trace__vfs_getname & trace__sys_exit */
2004 ttrace->filename.pending_open = false;
2005 }
ba3d7dee 2006
f3b623b8
ACM		 2007 if (trace->current != thread) {
2008 thread__put(trace->current);
2009 trace->current = thread__get(thread);
2010 }
b91fc39f
ACM		 2011 err = 0;
2012out_put:
2013 thread__put(thread);
2014 return err;
ba3d7dee
ACM		 2015}
2016
2017static int trace__sys_exit(struct trace *trace, struct perf_evsel *evsel,
0c82adcf 2018 union perf_event *event __maybe_unused,
ba3d7dee
ACM		 2019 struct perf_sample *sample)
2020{
2c82c3ad 2021 long ret;
60c907ab 2022 u64 duration = 0;
2ae3a312 2023 struct thread *thread;
b91fc39f 2024 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
bf2575c1 2025 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2ae3a312
ACM		 2026 struct thread_trace *ttrace;
2027
2028 if (sc == NULL)
2029 return -1;
ba3d7dee 2030
8fb598e5 2031 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
c24ff998 2032 ttrace = thread__trace(thread, trace->output);
2ae3a312 2033 if (ttrace == NULL)
b91fc39f 2034 goto out_put;
ba3d7dee 2035
bf2575c1
DA		 2036 if (trace->summary)
2037 thread__update_stats(ttrace, id, sample);
2038
77170988 2039 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
ba3d7dee 2040
7f4f8001
ACM		 2041 if (id == trace->audit.open_id && ret >= 0 && ttrace->filename.pending_open) {
2042 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2043 ttrace->filename.pending_open = false;
c522739d
ACM		 2044 ++trace->stats.vfs_getname;
2045 }
2046
752fde44
ACM		 2047 ttrace->exit_time = sample->time;
2048
ae9ed035 2049 if (ttrace->entry_time) {
60c907ab 2050 duration = sample->time - ttrace->entry_time;
ae9ed035
ACM		 2051 if (trace__filter_duration(trace, duration))
2052 goto out;
2053 } else if (trace->duration_filter)
2054 goto out;
60c907ab 2055
fd2eabaf
DA		 2056 if (trace->summary_only)
2057 goto out;
2058
c24ff998 2059 trace__fprintf_entry_head(trace, thread, duration, sample->time, trace->output);
752fde44
ACM		 2060
2061 if (ttrace->entry_pending) {
c24ff998 2062 fprintf(trace->output, "%-70s", ttrace->entry_str);
752fde44 2063 } else {
c24ff998
ACM		 2064 fprintf(trace->output, " ... [");
2065 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2066 fprintf(trace->output, "]: %s()", sc->name);
752fde44
ACM		 2067 }
2068
da3c9a44
ACM		 2069 if (sc->fmt == NULL) {
2070signed_print:
2c82c3ad 2071 fprintf(trace->output, ") = %ld", ret);
da3c9a44 2072 } else if (ret < 0 && sc->fmt->errmsg) {
942a91ed 2073 char bf[STRERR_BUFSIZE];
ba3d7dee
ACM		 2074 const char *emsg = strerror_r(-ret, bf, sizeof(bf)),
2075 *e = audit_errno_to_name(-ret);
2076
c24ff998 2077 fprintf(trace->output, ") = -1 %s %s", e, emsg);
da3c9a44 2078 } else if (ret == 0 && sc->fmt->timeout)
c24ff998 2079 fprintf(trace->output, ") = 0 Timeout");
04b34729 2080 else if (sc->fmt->hexret)
2c82c3ad 2081 fprintf(trace->output, ") = %#lx", ret);
ba3d7dee 2082 else
da3c9a44 2083 goto signed_print;
ba3d7dee 2084
c24ff998 2085 fputc('\n', trace->output);
ae9ed035 2086out:
752fde44 2087 ttrace->entry_pending = false;
b91fc39f
ACM		 2088 err = 0;
2089out_put:
2090 thread__put(thread);
2091 return err;
ba3d7dee
ACM		 2092}
2093
c522739d 2094static int trace__vfs_getname(struct trace *trace, struct perf_evsel *evsel,
0c82adcf 2095 union perf_event *event __maybe_unused,
c522739d
ACM		 2096 struct perf_sample *sample)
2097{
f994592d
ACM		 2098 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2099 struct thread_trace *ttrace;
2100 size_t filename_len, entry_str_len, to_move;
2101 ssize_t remaining_space;
2102 char *pos;
7f4f8001 2103 const char *filename = perf_evsel__rawptr(evsel, sample, "pathname");
f994592d
ACM		 2104
2105 if (!thread)
2106 goto out;
2107
2108 ttrace = thread__priv(thread);
2109 if (!ttrace)
2110 goto out;
2111
7f4f8001
ACM		 2112 filename_len = strlen(filename);
2113
2114 if (ttrace->filename.namelen < filename_len) {
2115 char *f = realloc(ttrace->filename.name, filename_len + 1);
2116
2117 if (f == NULL)
2118 goto out;
2119
2120 ttrace->filename.namelen = filename_len;
2121 ttrace->filename.name = f;
2122 }
2123
2124 strcpy(ttrace->filename.name, filename);
2125 ttrace->filename.pending_open = true;
2126
f994592d
ACM		 2127 if (!ttrace->filename.ptr)
2128 goto out;
2129
2130 entry_str_len = strlen(ttrace->entry_str);
2131 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2132 if (remaining_space <= 0)
2133 goto out;
2134
f994592d
ACM		 2135 if (filename_len > (size_t)remaining_space) {
2136 filename += filename_len - remaining_space;
2137 filename_len = remaining_space;
2138 }
2139
2140 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2141 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2142 memmove(pos + filename_len, pos, to_move);
2143 memcpy(pos, filename, filename_len);
2144
2145 ttrace->filename.ptr = 0;
2146 ttrace->filename.entry_str_pos = 0;
2147out:
c522739d
ACM		 2148 return 0;
2149}
2150
1302d88e 2151static int trace__sched_stat_runtime(struct trace *trace, struct perf_evsel *evsel,
0c82adcf 2152 union perf_event *event __maybe_unused,
1302d88e
ACM		 2153 struct perf_sample *sample)
2154{
2155 u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
2156 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
8fb598e5 2157 struct thread *thread = machine__findnew_thread(trace->host,
314add6b
AH		 2158 sample->pid,
2159 sample->tid);
c24ff998 2160 struct thread_trace *ttrace = thread__trace(thread, trace->output);
1302d88e
ACM		 2161
2162 if (ttrace == NULL)
2163 goto out_dump;
2164
2165 ttrace->runtime_ms += runtime_ms;
2166 trace->runtime_ms += runtime_ms;
b91fc39f 2167 thread__put(thread);
1302d88e
ACM		 2168 return 0;
2169
2170out_dump:
c24ff998 2171 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
1302d88e
ACM		 2172 evsel->name,
2173 perf_evsel__strval(evsel, sample, "comm"),
2174 (pid_t)perf_evsel__intval(evsel, sample, "pid"),
2175 runtime,
2176 perf_evsel__intval(evsel, sample, "vruntime"));
b91fc39f 2177 thread__put(thread);
1302d88e
ACM		 2178 return 0;
2179}
2180
14a052df
ACM		 2181static int trace__event_handler(struct trace *trace, struct perf_evsel *evsel,
2182 union perf_event *event __maybe_unused,
2183 struct perf_sample *sample)
2184{
2185 trace__printf_interrupted_entry(trace, sample);
2186 trace__fprintf_tstamp(trace, sample->time, trace->output);
0808921a
ACM		 2187
2188 if (trace->trace_syscalls)
2189 fprintf(trace->output, "( ): ");
2190
2191 fprintf(trace->output, "%s:", evsel->name);
14a052df
ACM		 2192
2193 if (evsel->tp_format) {
2194 event_format__fprintf(evsel->tp_format, sample->cpu,
2195 sample->raw_data, sample->raw_size,
2196 trace->output);
2197 }
2198
2199 fprintf(trace->output, ")\n");
2200 return 0;
2201}
2202
598d02c5
SF		 2203static void print_location(FILE *f, struct perf_sample *sample,
2204 struct addr_location *al,
2205 bool print_dso, bool print_sym)
2206{
2207
2208 if ((verbose || print_dso) && al->map)
2209 fprintf(f, "%s@", al->map->dso->long_name);
2210
2211 if ((verbose || print_sym) && al->sym)
4414a3c5 2212 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
598d02c5
SF		 2213 al->addr - al->sym->start);
2214 else if (al->map)
4414a3c5 2215 fprintf(f, "0x%" PRIx64, al->addr);
598d02c5 2216 else
4414a3c5 2217 fprintf(f, "0x%" PRIx64, sample->addr);
598d02c5
SF		 2218}
2219
2220static int trace__pgfault(struct trace *trace,
2221 struct perf_evsel *evsel,
2222 union perf_event *event,
2223 struct perf_sample *sample)
2224{
2225 struct thread *thread;
2226 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
2227 struct addr_location al;
2228 char map_type = 'd';
a2ea67d7 2229 struct thread_trace *ttrace;
b91fc39f 2230 int err = -1;
598d02c5
SF		 2231
2232 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
a2ea67d7
SF		 2233 ttrace = thread__trace(thread, trace->output);
2234 if (ttrace == NULL)
b91fc39f 2235 goto out_put;
a2ea67d7
SF		 2236
2237 if (evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2238 ttrace->pfmaj++;
2239 else
2240 ttrace->pfmin++;
2241
2242 if (trace->summary_only)
b91fc39f 2243 goto out;
598d02c5 2244
bb871a9c 2245 thread__find_addr_location(thread, cpumode, MAP__FUNCTION,
598d02c5
SF		 2246 sample->ip, &al);
2247
2248 trace__fprintf_entry_head(trace, thread, 0, sample->time, trace->output);
2249
2250 fprintf(trace->output, "%sfault [",
2251 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2252 "maj" : "min");
2253
2254 print_location(trace->output, sample, &al, false, true);
2255
2256 fprintf(trace->output, "] => ");
2257
bb871a9c 2258 thread__find_addr_location(thread, cpumode, MAP__VARIABLE,
598d02c5
SF		 2259 sample->addr, &al);
2260
2261 if (!al.map) {
bb871a9c 2262 thread__find_addr_location(thread, cpumode,
598d02c5
SF		 2263 MAP__FUNCTION, sample->addr, &al);
2264
2265 if (al.map)
2266 map_type = 'x';
2267 else
2268 map_type = '?';
2269 }
2270
2271 print_location(trace->output, sample, &al, true, false);
2272
2273 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
b91fc39f
ACM		 2274out:
2275 err = 0;
2276out_put:
2277 thread__put(thread);
2278 return err;
598d02c5
SF		 2279}
2280
bdc89661
DA		 2281static bool skip_sample(struct trace *trace, struct perf_sample *sample)
2282{
2283 if ((trace->pid_list && intlist__find(trace->pid_list, sample->pid)) ||
2284 (trace->tid_list && intlist__find(trace->tid_list, sample->tid)))
2285 return false;
2286
2287 if (trace->pid_list || trace->tid_list)
2288 return true;
2289
2290 return false;
2291}
2292
6810fc91 2293static int trace__process_sample(struct perf_tool *tool,
0c82adcf 2294 union perf_event *event,
6810fc91
DA		 2295 struct perf_sample *sample,
2296 struct perf_evsel *evsel,
2297 struct machine *machine __maybe_unused)
2298{
2299 struct trace *trace = container_of(tool, struct trace, tool);
2300 int err = 0;
2301
744a9719 2302 tracepoint_handler handler = evsel->handler;
6810fc91 2303
bdc89661
DA		 2304 if (skip_sample(trace, sample))
2305 return 0;
2306
4bb09192 2307 if (!trace->full_time && trace->base_time == 0)
6810fc91
DA		 2308 trace->base_time = sample->time;
2309
3160565f
DA		 2310 if (handler) {
2311 ++trace->nr_events;
0c82adcf 2312 handler(trace, evsel, event, sample);
3160565f 2313 }
6810fc91
DA		 2314
2315 return err;
2316}
2317
bdc89661
DA		 2318static int parse_target_str(struct trace *trace)
2319{
2320 if (trace->opts.target.pid) {
2321 trace->pid_list = intlist__new(trace->opts.target.pid);
2322 if (trace->pid_list == NULL) {
2323 pr_err("Error parsing process id string\n");
2324 return -EINVAL;
2325 }
2326 }
2327
2328 if (trace->opts.target.tid) {
2329 trace->tid_list = intlist__new(trace->opts.target.tid);
2330 if (trace->tid_list == NULL) {
2331 pr_err("Error parsing thread id string\n");
2332 return -EINVAL;
2333 }
2334 }
2335
2336 return 0;
2337}
2338
1e28fe0a 2339static int trace__record(struct trace *trace, int argc, const char **argv)
5e2485b1
DA		 2340{
2341 unsigned int rec_argc, i, j;
2342 const char **rec_argv;
2343 const char * const record_args[] = {
2344 "record",
2345 "-R",
2346 "-m", "1024",
2347 "-c", "1",
5e2485b1
DA		 2348 };
2349
1e28fe0a
SF		 2350 const char * const sc_args[] = { "-e", };
2351 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
2352 const char * const majpf_args[] = { "-e", "major-faults" };
2353 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
2354 const char * const minpf_args[] = { "-e", "minor-faults" };
2355 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
2356
9aca7f17 2357 /* +1 is for the event string below */
1e28fe0a
SF		 2358 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 1 +
2359 majpf_args_nr + minpf_args_nr + argc;
5e2485b1
DA		 2360 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2361
2362 if (rec_argv == NULL)
2363 return -ENOMEM;
2364
1e28fe0a 2365 j = 0;
5e2485b1 2366 for (i = 0; i < ARRAY_SIZE(record_args); i++)
1e28fe0a
SF		 2367 rec_argv[j++] = record_args[i];
2368
e281a960
SF		 2369 if (trace->trace_syscalls) {
2370 for (i = 0; i < sc_args_nr; i++)
2371 rec_argv[j++] = sc_args[i];
2372
2373 /* event string may be different for older kernels - e.g., RHEL6 */
2374 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
2375 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
2376 else if (is_valid_tracepoint("syscalls:sys_enter"))
2377 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
2378 else {
2379 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
2380 return -1;
2381 }
9aca7f17 2382 }
9aca7f17 2383
1e28fe0a
SF		 2384 if (trace->trace_pgfaults & TRACE_PFMAJ)
2385 for (i = 0; i < majpf_args_nr; i++)
2386 rec_argv[j++] = majpf_args[i];
2387
2388 if (trace->trace_pgfaults & TRACE_PFMIN)
2389 for (i = 0; i < minpf_args_nr; i++)
2390 rec_argv[j++] = minpf_args[i];
2391
2392 for (i = 0; i < (unsigned int)argc; i++)
2393 rec_argv[j++] = argv[i];
5e2485b1 2394
1e28fe0a 2395 return cmd_record(j, rec_argv, NULL);
5e2485b1
DA		 2396}
2397
bf2575c1
DA		 2398static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
2399
08c98776 2400static bool perf_evlist__add_vfs_getname(struct perf_evlist *evlist)
c522739d 2401{
ef503831 2402 struct perf_evsel *evsel = perf_evsel__newtp("probe", "vfs_getname");
8dd2a131
JO		 2403
2404 if (IS_ERR(evsel))
08c98776 2405 return false;
c522739d
ACM		 2406
2407 if (perf_evsel__field(evsel, "pathname") == NULL) {
2408 perf_evsel__delete(evsel);
08c98776 2409 return false;
c522739d
ACM		 2410 }
2411
744a9719 2412 evsel->handler = trace__vfs_getname;
c522739d 2413 perf_evlist__add(evlist, evsel);
08c98776 2414 return true;
c522739d
ACM		 2415}
2416
598d02c5
SF		 2417static int perf_evlist__add_pgfault(struct perf_evlist *evlist,
2418 u64 config)
2419{
2420 struct perf_evsel *evsel;
2421 struct perf_event_attr attr = {
2422 .type = PERF_TYPE_SOFTWARE,
2423 .mmap_data = 1,
598d02c5
SF		 2424 };
2425
2426 attr.config = config;
0524798c 2427 attr.sample_period = 1;
598d02c5
SF		 2428
2429 event_attr_init(&attr);
2430
2431 evsel = perf_evsel__new(&attr);
2432 if (!evsel)
2433 return -ENOMEM;
2434
2435 evsel->handler = trace__pgfault;
2436 perf_evlist__add(evlist, evsel);
2437
2438 return 0;
2439}
2440
ddbb1b13
ACM		 2441static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
2442{
2443 const u32 type = event->header.type;
2444 struct perf_evsel *evsel;
2445
2446 if (!trace->full_time && trace->base_time == 0)
2447 trace->base_time = sample->time;
2448
2449 if (type != PERF_RECORD_SAMPLE) {
2450 trace__process_event(trace, trace->host, event, sample);
2451 return;
2452 }
2453
2454 evsel = perf_evlist__id2evsel(trace->evlist, sample->id);
2455 if (evsel == NULL) {
2456 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
2457 return;
2458 }
2459
2460 if (evsel->attr.type == PERF_TYPE_TRACEPOINT &&
2461 sample->raw_data == NULL) {
2462 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
2463 perf_evsel__name(evsel), sample->tid,
2464 sample->cpu, sample->raw_size);
2465 } else {
2466 tracepoint_handler handler = evsel->handler;
2467 handler(trace, evsel, event, sample);
2468 }
2469}
2470
c27366f0
ACM		 2471static int trace__add_syscall_newtp(struct trace *trace)
2472{
2473 int ret = -1;
2474 struct perf_evlist *evlist = trace->evlist;
2475 struct perf_evsel *sys_enter, *sys_exit;
2476
2477 sys_enter = perf_evsel__syscall_newtp("sys_enter", trace__sys_enter);
2478 if (sys_enter == NULL)
2479 goto out;
2480
2481 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
2482 goto out_delete_sys_enter;
2483
2484 sys_exit = perf_evsel__syscall_newtp("sys_exit", trace__sys_exit);
2485 if (sys_exit == NULL)
2486 goto out_delete_sys_enter;
2487
2488 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
2489 goto out_delete_sys_exit;
2490
2491 perf_evlist__add(evlist, sys_enter);
2492 perf_evlist__add(evlist, sys_exit);
2493
8b3ce757
ACM		 2494 trace->syscalls.events.sys_enter = sys_enter;
2495 trace->syscalls.events.sys_exit = sys_exit;
c27366f0
ACM		 2496
2497 ret = 0;
2498out:
2499 return ret;
2500
2501out_delete_sys_exit:
2502 perf_evsel__delete_priv(sys_exit);
2503out_delete_sys_enter:
2504 perf_evsel__delete_priv(sys_enter);
2505 goto out;
2506}
2507
19867b61
ACM		 2508static int trace__set_ev_qualifier_filter(struct trace *trace)
2509{
2510 int err = -1;
2511 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
2512 trace->ev_qualifier_ids.nr,
2513 trace->ev_qualifier_ids.entries);
2514
2515 if (filter == NULL)
2516 goto out_enomem;
2517
2518 if (!perf_evsel__append_filter(trace->syscalls.events.sys_enter, "&&", filter))
2519 err = perf_evsel__append_filter(trace->syscalls.events.sys_exit, "&&", filter);
2520
2521 free(filter);
2522out:
2523 return err;
2524out_enomem:
2525 errno = ENOMEM;
2526 goto out;
2527}
c27366f0 2528
f15eb531 2529static int trace__run(struct trace *trace, int argc, const char **argv)
514f1c67 2530{
14a052df 2531 struct perf_evlist *evlist = trace->evlist;
94ad89bc 2532 struct perf_evsel *evsel;
efd5745e
ACM		 2533 int err = -1, i;
2534 unsigned long before;
f15eb531 2535 const bool forks = argc > 0;
46fb3c21 2536 bool draining = false;
514f1c67 2537
75b757ca
ACM		 2538 trace->live = true;
2539
c27366f0 2540 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
801c67b0 2541 goto out_error_raw_syscalls;
514f1c67 2542
e281a960 2543 if (trace->trace_syscalls)
08c98776 2544 trace->vfs_getname = perf_evlist__add_vfs_getname(evlist);
c522739d 2545
598d02c5 2546 if ((trace->trace_pgfaults & TRACE_PFMAJ) &&
e2726d99 2547 perf_evlist__add_pgfault(evlist, PERF_COUNT_SW_PAGE_FAULTS_MAJ)) {
5ed08dae 2548 goto out_error_mem;
e2726d99 2549 }
598d02c5
SF		 2550
2551 if ((trace->trace_pgfaults & TRACE_PFMIN) &&
2552 perf_evlist__add_pgfault(evlist, PERF_COUNT_SW_PAGE_FAULTS_MIN))
5ed08dae 2553 goto out_error_mem;
598d02c5 2554
1302d88e 2555 if (trace->sched &&
2cc990ba
ACM		 2556 perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
2557 trace__sched_stat_runtime))
2558 goto out_error_sched_stat_runtime;
1302d88e 2559
514f1c67
ACM		 2560 err = perf_evlist__create_maps(evlist, &trace->opts.target);
2561 if (err < 0) {
c24ff998 2562 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
514f1c67
ACM		 2563 goto out_delete_evlist;
2564 }
2565
752fde44
ACM		 2566 err = trace__symbols_init(trace, evlist);
2567 if (err < 0) {
c24ff998 2568 fprintf(trace->output, "Problems initializing symbol libraries!\n");
03ad9747 2569 goto out_delete_evlist;
752fde44
ACM		 2570 }
2571
f77a9518 2572 perf_evlist__config(evlist, &trace->opts);
514f1c67 2573
f15eb531
NK		 2574 signal(SIGCHLD, sig_handler);
2575 signal(SIGINT, sig_handler);
2576
2577 if (forks) {
6ef73ec4 2578 err = perf_evlist__prepare_workload(evlist, &trace->opts.target,
735f7e0b 2579 argv, false, NULL);
f15eb531 2580 if (err < 0) {
c24ff998 2581 fprintf(trace->output, "Couldn't run the workload!\n");
03ad9747 2582 goto out_delete_evlist;
f15eb531
NK		 2583 }
2584 }
2585
514f1c67 2586 err = perf_evlist__open(evlist);
a8f23d8f
ACM		 2587 if (err < 0)
2588 goto out_error_open;
514f1c67 2589
ba504235
WN		 2590 err = bpf__apply_obj_config();
2591 if (err) {
2592 char errbuf[BUFSIZ];
2593
2594 bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
2595 pr_err("ERROR: Apply config to BPF failed: %s\n",
2596 errbuf);
2597 goto out_error_open;
2598 }
2599
241b057c
ACM		 2600 /*
2601 * Better not use !target__has_task() here because we need to cover the
2602 * case where no threads were specified in the command line, but a
2603 * workload was, and in that case we will fill in the thread_map when
2604 * we fork the workload in perf_evlist__prepare_workload.
2605 */
f078c385
ACM		 2606 if (trace->filter_pids.nr > 0)
2607 err = perf_evlist__set_filter_pids(evlist, trace->filter_pids.nr, trace->filter_pids.entries);
e13798c7 2608 else if (thread_map__pid(evlist->threads, 0) == -1)
f078c385
ACM		 2609 err = perf_evlist__set_filter_pid(evlist, getpid());
2610
94ad89bc
ACM		 2611 if (err < 0)
2612 goto out_error_mem;
2613
19867b61
ACM		 2614 if (trace->ev_qualifier_ids.nr > 0) {
2615 err = trace__set_ev_qualifier_filter(trace);
2616 if (err < 0)
2617 goto out_errno;
19867b61 2618
2e5e5f87
ACM		 2619 pr_debug("event qualifier tracepoint filter: %s\n",
2620 trace->syscalls.events.sys_exit->filter);
2621 }
19867b61 2622
94ad89bc
ACM		 2623 err = perf_evlist__apply_filters(evlist, &evsel);
2624 if (err < 0)
2625 goto out_error_apply_filters;
241b057c 2626
f885037e 2627 err = perf_evlist__mmap(evlist, trace->opts.mmap_pages, false);
e09b18d4
ACM		 2628 if (err < 0)
2629 goto out_error_mmap;
514f1c67 2630
cb24d01d
ACM		 2631 if (!target__none(&trace->opts.target))
2632 perf_evlist__enable(evlist);
2633
f15eb531
NK		 2634 if (forks)
2635 perf_evlist__start_workload(evlist);
2636
e13798c7 2637 trace->multiple_threads = thread_map__pid(evlist->threads, 0) == -1 ||
42052bea
ACM		 2638 evlist->threads->nr > 1 ||
2639 perf_evlist__first(evlist)->attr.inherit;
514f1c67 2640again:
efd5745e 2641 before = trace->nr_events;
514f1c67
ACM		 2642
2643 for (i = 0; i < evlist->nr_mmaps; i++) {
2644 union perf_event *event;
2645
2646 while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
514f1c67 2647 struct perf_sample sample;
514f1c67 2648
efd5745e 2649 ++trace->nr_events;
514f1c67 2650
514f1c67
ACM		 2651 err = perf_evlist__parse_sample(evlist, event, &sample);
2652 if (err) {
c24ff998 2653 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
8e50d384 2654 goto next_event;
514f1c67
ACM		 2655 }
2656
ddbb1b13 2657 trace__handle_event(trace, event, &sample);
8e50d384
ZZ		 2658next_event:
2659 perf_evlist__mmap_consume(evlist, i);
20c5f10e 2660
ba209f85
ACM		 2661 if (interrupted)
2662 goto out_disable;
02ac5421
ACM		 2663
2664 if (done && !draining) {
2665 perf_evlist__disable(evlist);
2666 draining = true;
2667 }
514f1c67
ACM		 2668 }
2669 }
2670
efd5745e 2671 if (trace->nr_events == before) {
ba209f85 2672 int timeout = done ? 100 : -1;
f15eb531 2673
46fb3c21
ACM		 2674 if (!draining && perf_evlist__poll(evlist, timeout) > 0) {
2675 if (perf_evlist__filter_pollfd(evlist, POLLERR | POLLHUP) == 0)
2676 draining = true;
2677
ba209f85 2678 goto again;
46fb3c21 2679 }
ba209f85
ACM		 2680 } else {
2681 goto again;
f15eb531
NK		 2682 }
2683
ba209f85 2684out_disable:
f3b623b8
ACM		 2685 thread__zput(trace->current);
2686
ba209f85 2687 perf_evlist__disable(evlist);
514f1c67 2688
c522739d
ACM		 2689 if (!err) {
2690 if (trace->summary)
2691 trace__fprintf_thread_summary(trace, trace->output);
2692
2693 if (trace->show_tool_stats) {
2694 fprintf(trace->output, "Stats:\n "
2695 " vfs_getname : %" PRIu64 "\n"
2696 " proc_getname: %" PRIu64 "\n",
2697 trace->stats.vfs_getname,
2698 trace->stats.proc_getname);
2699 }
2700 }
bf2575c1 2701
514f1c67
ACM		 2702out_delete_evlist:
2703 perf_evlist__delete(evlist);
14a052df 2704 trace->evlist = NULL;
75b757ca 2705 trace->live = false;
514f1c67 2706 return err;
6ef068cb
ACM		 2707{
2708 char errbuf[BUFSIZ];
a8f23d8f 2709
2cc990ba 2710out_error_sched_stat_runtime:
988bdb31 2711 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
2cc990ba
ACM		 2712 goto out_error;
2713
801c67b0 2714out_error_raw_syscalls:
988bdb31 2715 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
a8f23d8f
ACM		 2716 goto out_error;
2717
e09b18d4
ACM		 2718out_error_mmap:
2719 perf_evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
2720 goto out_error;
2721
a8f23d8f
ACM		 2722out_error_open:
2723 perf_evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
2724
2725out_error:
6ef068cb 2726 fprintf(trace->output, "%s\n", errbuf);
87f91868 2727 goto out_delete_evlist;
94ad89bc
ACM		 2728
2729out_error_apply_filters:
2730 fprintf(trace->output,
2731 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
2732 evsel->filter, perf_evsel__name(evsel), errno,
2733 strerror_r(errno, errbuf, sizeof(errbuf)));
2734 goto out_delete_evlist;
514f1c67 2735}
5ed08dae
ACM		 2736out_error_mem:
2737 fprintf(trace->output, "Not enough memory to run!\n");
2738 goto out_delete_evlist;
19867b61
ACM		 2739
2740out_errno:
2741 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
2742 goto out_delete_evlist;
a8f23d8f 2743}
514f1c67 2744
6810fc91
DA		 2745static int trace__replay(struct trace *trace)
2746{
2747 const struct perf_evsel_str_handler handlers[] = {
c522739d 2748 { "probe:vfs_getname", trace__vfs_getname, },
6810fc91 2749 };
f5fc1412
JO		 2750 struct perf_data_file file = {
2751 .path = input_name,
2752 .mode = PERF_DATA_MODE_READ,
e366a6d8 2753 .force = trace->force,
f5fc1412 2754 };
6810fc91 2755 struct perf_session *session;
003824e8 2756 struct perf_evsel *evsel;
6810fc91
DA		 2757 int err = -1;
2758
2759 trace->tool.sample = trace__process_sample;
2760 trace->tool.mmap = perf_event__process_mmap;
384c671e 2761 trace->tool.mmap2 = perf_event__process_mmap2;
6810fc91
DA		 2762 trace->tool.comm = perf_event__process_comm;
2763 trace->tool.exit = perf_event__process_exit;
2764 trace->tool.fork = perf_event__process_fork;
2765 trace->tool.attr = perf_event__process_attr;
2766 trace->tool.tracing_data = perf_event__process_tracing_data;
2767 trace->tool.build_id = perf_event__process_build_id;
2768
0a8cb85c 2769 trace->tool.ordered_events = true;
6810fc91
DA		 2770 trace->tool.ordering_requires_timestamps = true;
2771
2772 /* add tid to output */
2773 trace->multiple_threads = true;
2774
f5fc1412 2775 session = perf_session__new(&file, false, &trace->tool);
6810fc91 2776 if (session == NULL)
52e02834 2777 return -1;
6810fc91 2778
0a7e6d1b 2779 if (symbol__init(&session->header.env) < 0)
cb2ffae2
NK		 2780 goto out;
2781
8fb598e5
DA		 2782 trace->host = &session->machines.host;
2783
6810fc91
DA		 2784 err = perf_session__set_tracepoints_handlers(session, handlers);
2785 if (err)
2786 goto out;
2787
003824e8
NK		 2788 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2789 "raw_syscalls:sys_enter");
9aca7f17
DA		 2790 /* older kernels have syscalls tp versus raw_syscalls */
2791 if (evsel == NULL)
2792 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2793 "syscalls:sys_enter");
003824e8 2794
e281a960
SF		 2795 if (evsel &&
2796 (perf_evsel__init_syscall_tp(evsel, trace__sys_enter) < 0 ||
2797 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
003824e8
NK		 2798 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
2799 goto out;
2800 }
2801
2802 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2803 "raw_syscalls:sys_exit");
9aca7f17
DA		 2804 if (evsel == NULL)
2805 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2806 "syscalls:sys_exit");
e281a960
SF		 2807 if (evsel &&
2808 (perf_evsel__init_syscall_tp(evsel, trace__sys_exit) < 0 ||
2809 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
003824e8 2810 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
6810fc91
DA		 2811 goto out;
2812 }
2813
1e28fe0a
SF		 2814 evlist__for_each(session->evlist, evsel) {
2815 if (evsel->attr.type == PERF_TYPE_SOFTWARE &&
2816 (evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
2817 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
2818 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS))
2819 evsel->handler = trace__pgfault;
2820 }
2821
bdc89661
DA		 2822 err = parse_target_str(trace);
2823 if (err != 0)
2824 goto out;
2825
6810fc91
DA		 2826 setup_pager();
2827
b7b61cbe 2828 err = perf_session__process_events(session);
6810fc91
DA		 2829 if (err)
2830 pr_err("Failed to process events, error %d", err);
2831
bf2575c1
DA		 2832 else if (trace->summary)
2833 trace__fprintf_thread_summary(trace, trace->output);
2834
6810fc91
DA		 2835out:
2836 perf_session__delete(session);
2837
2838 return err;
2839}
2840
1302d88e
ACM		 2841static size_t trace__fprintf_threads_header(FILE *fp)
2842{
2843 size_t printed;
2844
99ff7150 2845 printed = fprintf(fp, "\n Summary of events:\n\n");
bf2575c1
DA		 2846
2847 return printed;
2848}
2849
2850static size_t thread__dump_stats(struct thread_trace *ttrace,
2851 struct trace *trace, FILE *fp)
2852{
2853 struct stats *stats;
2854 size_t printed = 0;
2855 struct syscall *sc;
2856 struct int_node *inode = intlist__first(ttrace->syscall_stats);
2857
2858 if (inode == NULL)
2859 return 0;
2860
2861 printed += fprintf(fp, "\n");
2862
834fd46d
MW		 2863 printed += fprintf(fp, " syscall calls total min avg max stddev\n");
2864 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
2865 printed += fprintf(fp, " --------------- -------- --------- --------- --------- --------- ------\n");
99ff7150 2866
bf2575c1
DA		 2867 /* each int_node is a syscall */
2868 while (inode) {
2869 stats = inode->priv;
2870 if (stats) {
2871 double min = (double)(stats->min) / NSEC_PER_MSEC;
2872 double max = (double)(stats->max) / NSEC_PER_MSEC;
2873 double avg = avg_stats(stats);
2874 double pct;
2875 u64 n = (u64) stats->n;
2876
2877 pct = avg ? 100.0 * stddev_stats(stats)/avg : 0.0;
2878 avg /= NSEC_PER_MSEC;
2879
2880 sc = &trace->syscalls.table[inode->i];
99ff7150 2881 printed += fprintf(fp, " %-15s", sc->name);
834fd46d
MW		 2882 printed += fprintf(fp, " %8" PRIu64 " %9.3f %9.3f %9.3f",
2883 n, avg * n, min, avg);
27a778b5 2884 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
bf2575c1
DA		 2885 }
2886
2887 inode = intlist__next(inode);
2888 }
2889
2890 printed += fprintf(fp, "\n\n");
1302d88e
ACM		 2891
2892 return printed;
2893}
2894
896cbb56
DA		 2895/* struct used to pass data to per-thread function */
2896struct summary_data {
2897 FILE *fp;
2898 struct trace *trace;
2899 size_t printed;
2900};
2901
2902static int trace__fprintf_one_thread(struct thread *thread, void *priv)
2903{
2904 struct summary_data *data = priv;
2905 FILE *fp = data->fp;
2906 size_t printed = data->printed;
2907 struct trace *trace = data->trace;
89dceb22 2908 struct thread_trace *ttrace = thread__priv(thread);
896cbb56
DA		 2909 double ratio;
2910
2911 if (ttrace == NULL)
2912 return 0;
2913
2914 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
2915
15e65c69 2916 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
99ff7150 2917 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
15e65c69 2918 printed += fprintf(fp, "%.1f%%", ratio);
a2ea67d7
SF		 2919 if (ttrace->pfmaj)
2920 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
2921 if (ttrace->pfmin)
2922 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
99ff7150 2923 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
bf2575c1 2924 printed += thread__dump_stats(ttrace, trace, fp);
896cbb56
DA		 2925
2926 data->printed += printed;
2927
2928 return 0;
2929}
2930
1302d88e
ACM		 2931static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
2932{
896cbb56
DA		 2933 struct summary_data data = {
2934 .fp = fp,
2935 .trace = trace
2936 };
2937 data.printed = trace__fprintf_threads_header(fp);
1302d88e 2938
896cbb56
DA		 2939 machine__for_each_thread(trace->host, trace__fprintf_one_thread, &data);
2940
2941 return data.printed;
1302d88e
ACM		 2942}
2943
ae9ed035
ACM		 2944static int trace__set_duration(const struct option *opt, const char *str,
2945 int unset __maybe_unused)
2946{
2947 struct trace *trace = opt->value;
2948
2949 trace->duration_filter = atof(str);
2950 return 0;
2951}
2952
f078c385
ACM		 2953static int trace__set_filter_pids(const struct option *opt, const char *str,
2954 int unset __maybe_unused)
2955{
2956 int ret = -1;
2957 size_t i;
2958 struct trace *trace = opt->value;
2959 /*
2960 * FIXME: introduce a intarray class, plain parse csv and create a
2961 * { int nr, int entries[] } struct...
2962 */
2963 struct intlist *list = intlist__new(str);
2964
2965 if (list == NULL)
2966 return -1;
2967
2968 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
2969 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
2970
2971 if (trace->filter_pids.entries == NULL)
2972 goto out;
2973
2974 trace->filter_pids.entries[0] = getpid();
2975
2976 for (i = 1; i < trace->filter_pids.nr; ++i)
2977 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
2978
2979 intlist__delete(list);
2980 ret = 0;
2981out:
2982 return ret;
2983}
2984
c24ff998
ACM		 2985static int trace__open_output(struct trace *trace, const char *filename)
2986{
2987 struct stat st;
2988
2989 if (!stat(filename, &st) && st.st_size) {
2990 char oldname[PATH_MAX];
2991
2992 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
2993 unlink(oldname);
2994 rename(filename, oldname);
2995 }
2996
2997 trace->output = fopen(filename, "w");
2998
2999 return trace->output == NULL ? -errno : 0;
3000}
3001
598d02c5
SF		 3002static int parse_pagefaults(const struct option *opt, const char *str,
3003 int unset __maybe_unused)
3004{
3005 int *trace_pgfaults = opt->value;
3006
3007 if (strcmp(str, "all") == 0)
3008 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
3009 else if (strcmp(str, "maj") == 0)
3010 *trace_pgfaults |= TRACE_PFMAJ;
3011 else if (strcmp(str, "min") == 0)
3012 *trace_pgfaults |= TRACE_PFMIN;
3013 else
3014 return -1;
3015
3016 return 0;
3017}
3018
14a052df
ACM		 3019static void evlist__set_evsel_handler(struct perf_evlist *evlist, void *handler)
3020{
3021 struct perf_evsel *evsel;
3022
3023 evlist__for_each(evlist, evsel)
3024 evsel->handler = handler;
3025}
3026
514f1c67
ACM		 3027int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
3028{
6fdd9cb7 3029 const char *trace_usage[] = {
f15eb531
NK		 3030 "perf trace [<options>] [<command>]",
3031 "perf trace [<options>] -- <command> [<options>]",
5e2485b1
DA		 3032 "perf trace record [<options>] [<command>]",
3033 "perf trace record [<options>] -- <command> [<options>]",
514f1c67
ACM		 3034 NULL
3035 };
3036 struct trace trace = {
c522739d
ACM		 3037 .audit = {
3038 .machine = audit_detect_machine(),
3039 .open_id = audit_name_to_syscall("open", trace.audit.machine),
3040 },
514f1c67
ACM		 3041 .syscalls = {
3042 . max = -1,
3043 },
3044 .opts = {
3045 .target = {
3046 .uid = UINT_MAX,
3047 .uses_mmap = true,
3048 },
3049 .user_freq = UINT_MAX,
3050 .user_interval = ULLONG_MAX,
509051ea 3051 .no_buffering = true,
38d5447d 3052 .mmap_pages = UINT_MAX,
9d9cad76 3053 .proc_map_timeout = 500,
514f1c67 3054 },
007d66a0 3055 .output = stderr,
50c95cbd 3056 .show_comm = true,
e281a960 3057 .trace_syscalls = true,
514f1c67 3058 };
c24ff998 3059 const char *output_name = NULL;
2ae3a312 3060 const char *ev_qualifier_str = NULL;
514f1c67 3061 const struct option trace_options[] = {
14a052df
ACM		 3062 OPT_CALLBACK(0, "event", &trace.evlist, "event",
3063 "event selector. use 'perf list' to list available events",
3064 parse_events_option),
50c95cbd
ACM		 3065 OPT_BOOLEAN(0, "comm", &trace.show_comm,
3066 "show the thread COMM next to its id"),
c522739d 3067 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
d303e85a 3068 OPT_STRING('e', "expr", &ev_qualifier_str, "expr", "list of syscalls to trace"),
c24ff998 3069 OPT_STRING('o', "output", &output_name, "file", "output file name"),
6810fc91 3070 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
514f1c67
ACM		 3071 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
3072 "trace events on existing process id"),
ac9be8ee 3073 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
514f1c67 3074 "trace events on existing thread id"),
fa0e4ffe
ACM		 3075 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
3076 "pids to filter (by the kernel)", trace__set_filter_pids),
ac9be8ee 3077 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
514f1c67 3078 "system-wide collection from all CPUs"),
ac9be8ee 3079 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
514f1c67 3080 "list of cpus to monitor"),
6810fc91 3081 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
514f1c67 3082 "child tasks do not inherit counters"),
994a1f78
JO		 3083 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
3084 "number of mmap data pages",
3085 perf_evlist__parse_mmap_pages),
ac9be8ee 3086 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
514f1c67 3087 "user to profile"),
ae9ed035
ACM		 3088 OPT_CALLBACK(0, "duration", &trace, "float",
3089 "show only events with duration > N.M ms",
3090 trace__set_duration),
1302d88e 3091 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
7c304ee0 3092 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4bb09192
DA		 3093 OPT_BOOLEAN('T', "time", &trace.full_time,
3094 "Show full timestamp, not time relative to first start"),
fd2eabaf
DA		 3095 OPT_BOOLEAN('s', "summary", &trace.summary_only,
3096 "Show only syscall summary with statistics"),
3097 OPT_BOOLEAN('S', "with-summary", &trace.summary,
3098 "Show all syscalls and summary with statistics"),
598d02c5
SF		 3099 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
3100 "Trace pagefaults", parse_pagefaults, "maj"),
e281a960 3101 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
e366a6d8 3102 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
9d9cad76
KL		 3103 OPT_UINTEGER(0, "proc-map-timeout", &trace.opts.proc_map_timeout,
3104 "per thread proc mmap processing timeout in ms"),
514f1c67
ACM		 3105 OPT_END()
3106 };
6fdd9cb7 3107 const char * const trace_subcommands[] = { "record", NULL };
514f1c67 3108 int err;
32caf0d1 3109 char bf[BUFSIZ];
514f1c67 3110
4d08cb80
ACM		 3111 signal(SIGSEGV, sighandler_dump_stack);
3112 signal(SIGFPE, sighandler_dump_stack);
3113
14a052df 3114 trace.evlist = perf_evlist__new();
14a052df
ACM		 3115
3116 if (trace.evlist == NULL) {
3117 pr_err("Not enough memory to run!\n");
ff8f695c 3118 err = -ENOMEM;
14a052df
ACM		 3119 goto out;
3120 }
3121
6fdd9cb7
YS		 3122 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
3123 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
fd2eabaf 3124
598d02c5
SF		 3125 if (trace.trace_pgfaults) {
3126 trace.opts.sample_address = true;
3127 trace.opts.sample_time = true;
3128 }
3129
14a052df
ACM		 3130 if (trace.evlist->nr_entries > 0)
3131 evlist__set_evsel_handler(trace.evlist, trace__event_handler);
3132
1e28fe0a
SF		 3133 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
3134 return trace__record(&trace, argc-1, &argv[1]);
3135
3136 /* summary_only implies summary option, but don't overwrite summary if set */
3137 if (trace.summary_only)
3138 trace.summary = trace.summary_only;
3139
726f3234
ACM		 3140 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
3141 trace.evlist->nr_entries == 0 /* Was --events used? */) {
e281a960
SF		 3142 pr_err("Please specify something to trace.\n");
3143 return -1;
3144 }
3145
c24ff998
ACM		 3146 if (output_name != NULL) {
3147 err = trace__open_output(&trace, output_name);
3148 if (err < 0) {
3149 perror("failed to create output file");
3150 goto out;
3151 }
3152 }
3153
2ae3a312 3154 if (ev_qualifier_str != NULL) {
b059efdf 3155 const char *s = ev_qualifier_str;
005438a8
ACM		 3156 struct strlist_config slist_config = {
3157 .dirname = system_path(STRACE_GROUPS_DIR),
3158 };
b059efdf
ACM		 3159
3160 trace.not_ev_qualifier = *s == '!';
3161 if (trace.not_ev_qualifier)
3162 ++s;
005438a8 3163 trace.ev_qualifier = strlist__new(s, &slist_config);
2ae3a312 3164 if (trace.ev_qualifier == NULL) {
c24ff998
ACM		 3165 fputs("Not enough memory to parse event qualifier",
3166 trace.output);
3167 err = -ENOMEM;
3168 goto out_close;
2ae3a312 3169 }
d0cc439b
ACM		 3170
3171 err = trace__validate_ev_qualifier(&trace);
3172 if (err)
3173 goto out_close;
2ae3a312
ACM		 3174 }
3175
602ad878 3176 err = target__validate(&trace.opts.target);
32caf0d1 3177 if (err) {
602ad878 3178 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
c24ff998
ACM		 3179 fprintf(trace.output, "%s", bf);
3180 goto out_close;
32caf0d1
NK		 3181 }
3182
602ad878 3183 err = target__parse_uid(&trace.opts.target);
514f1c67 3184 if (err) {
602ad878 3185 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
c24ff998
ACM		 3186 fprintf(trace.output, "%s", bf);
3187 goto out_close;
514f1c67
ACM		 3188 }
3189
602ad878 3190 if (!argc && target__none(&trace.opts.target))
ee76120e
NK		 3191 trace.opts.target.system_wide = true;
3192
6810fc91
DA		 3193 if (input_name)
3194 err = trace__replay(&trace);
3195 else
3196 err = trace__run(&trace, argc, argv);
1302d88e 3197
c24ff998
ACM		 3198out_close:
3199 if (output_name != NULL)
3200 fclose(trace.output);
3201out:
1302d88e 3202 return err;
514f1c67 3203}
Linux kernel - Deliverables
This page took 0.354608 seconds and 5 git commands to generate.