tracing: Fix irqs-off tag display in syscall tracing
[pandora-kernel.git] / kernel / trace / trace_syscalls.c
1 #include <trace/syscall.h>
2 #include <trace/events/syscalls.h>
3 #include <linux/slab.h>
4 #include <linux/kernel.h>
5 #include <linux/module.h>       /* for MODULE_NAME_LEN via KSYM_SYMBOL_LEN */
6 #include <linux/ftrace.h>
7 #include <linux/perf_event.h>
8 #include <asm/syscall.h>
9
10 #include "trace_output.h"
11 #include "trace.h"
12
13 static DEFINE_MUTEX(syscall_trace_lock);
14 static int sys_refcount_enter;
15 static int sys_refcount_exit;
16 static DECLARE_BITMAP(enabled_enter_syscalls, NR_syscalls);
17 static DECLARE_BITMAP(enabled_exit_syscalls, NR_syscalls);
18
19 static int syscall_enter_register(struct ftrace_event_call *event,
20                                  enum trace_reg type);
21 static int syscall_exit_register(struct ftrace_event_call *event,
22                                  enum trace_reg type);
23
24 static int syscall_enter_define_fields(struct ftrace_event_call *call);
25 static int syscall_exit_define_fields(struct ftrace_event_call *call);
26
27 static struct list_head *
28 syscall_get_enter_fields(struct ftrace_event_call *call)
29 {
30         struct syscall_metadata *entry = call->data;
31
32         return &entry->enter_fields;
33 }
34
35 struct trace_event_functions enter_syscall_print_funcs = {
36         .trace          = print_syscall_enter,
37 };
38
39 struct trace_event_functions exit_syscall_print_funcs = {
40         .trace          = print_syscall_exit,
41 };
42
43 struct ftrace_event_class event_class_syscall_enter = {
44         .system         = "syscalls",
45         .reg            = syscall_enter_register,
46         .define_fields  = syscall_enter_define_fields,
47         .get_fields     = syscall_get_enter_fields,
48         .raw_init       = init_syscall_trace,
49 };
50
51 struct ftrace_event_class event_class_syscall_exit = {
52         .system         = "syscalls",
53         .reg            = syscall_exit_register,
54         .define_fields  = syscall_exit_define_fields,
55         .fields         = LIST_HEAD_INIT(event_class_syscall_exit.fields),
56         .raw_init       = init_syscall_trace,
57 };
58
59 extern struct syscall_metadata *__start_syscalls_metadata[];
60 extern struct syscall_metadata *__stop_syscalls_metadata[];
61
62 static struct syscall_metadata **syscalls_metadata;
63
64 #ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME
65 static inline bool arch_syscall_match_sym_name(const char *sym, const char *name)
66 {
67         /*
68          * Only compare after the "sys" prefix. Archs that use
69          * syscall wrappers may have syscalls symbols aliases prefixed
70          * with "SyS" instead of "sys", leading to an unwanted
71          * mismatch.
72          */
73         return !strcmp(sym + 3, name + 3);
74 }
75 #endif
76
77 static __init struct syscall_metadata *
78 find_syscall_meta(unsigned long syscall)
79 {
80         struct syscall_metadata **start;
81         struct syscall_metadata **stop;
82         char str[KSYM_SYMBOL_LEN];
83
84
85         start = __start_syscalls_metadata;
86         stop = __stop_syscalls_metadata;
87         kallsyms_lookup(syscall, NULL, NULL, NULL, str);
88
89         if (arch_syscall_match_sym_name(str, "sys_ni_syscall"))
90                 return NULL;
91
92         for ( ; start < stop; start++) {
93                 if ((*start)->name && arch_syscall_match_sym_name(str, (*start)->name))
94                         return *start;
95         }
96         return NULL;
97 }
98
99 static struct syscall_metadata *syscall_nr_to_meta(int nr)
100 {
101         if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
102                 return NULL;
103
104         return syscalls_metadata[nr];
105 }
106
107 enum print_line_t
108 print_syscall_enter(struct trace_iterator *iter, int flags,
109                     struct trace_event *event)
110 {
111         struct trace_seq *s = &iter->seq;
112         struct trace_entry *ent = iter->ent;
113         struct syscall_trace_enter *trace;
114         struct syscall_metadata *entry;
115         int i, ret, syscall;
116
117         trace = (typeof(trace))ent;
118         syscall = trace->nr;
119         entry = syscall_nr_to_meta(syscall);
120
121         if (!entry)
122                 goto end;
123
124         if (entry->enter_event->event.type != ent->type) {
125                 WARN_ON_ONCE(1);
126                 goto end;
127         }
128
129         ret = trace_seq_printf(s, "%s(", entry->name);
130         if (!ret)
131                 return TRACE_TYPE_PARTIAL_LINE;
132
133         for (i = 0; i < entry->nb_args; i++) {
134                 /* parameter types */
135                 if (trace_flags & TRACE_ITER_VERBOSE) {
136                         ret = trace_seq_printf(s, "%s ", entry->types[i]);
137                         if (!ret)
138                                 return TRACE_TYPE_PARTIAL_LINE;
139                 }
140                 /* parameter values */
141                 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
142                                        trace->args[i],
143                                        i == entry->nb_args - 1 ? "" : ", ");
144                 if (!ret)
145                         return TRACE_TYPE_PARTIAL_LINE;
146         }
147
148         ret = trace_seq_putc(s, ')');
149         if (!ret)
150                 return TRACE_TYPE_PARTIAL_LINE;
151
152 end:
153         ret =  trace_seq_putc(s, '\n');
154         if (!ret)
155                 return TRACE_TYPE_PARTIAL_LINE;
156
157         return TRACE_TYPE_HANDLED;
158 }
159
160 enum print_line_t
161 print_syscall_exit(struct trace_iterator *iter, int flags,
162                    struct trace_event *event)
163 {
164         struct trace_seq *s = &iter->seq;
165         struct trace_entry *ent = iter->ent;
166         struct syscall_trace_exit *trace;
167         int syscall;
168         struct syscall_metadata *entry;
169         int ret;
170
171         trace = (typeof(trace))ent;
172         syscall = trace->nr;
173         entry = syscall_nr_to_meta(syscall);
174
175         if (!entry) {
176                 trace_seq_printf(s, "\n");
177                 return TRACE_TYPE_HANDLED;
178         }
179
180         if (entry->exit_event->event.type != ent->type) {
181                 WARN_ON_ONCE(1);
182                 return TRACE_TYPE_UNHANDLED;
183         }
184
185         ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
186                                 trace->ret);
187         if (!ret)
188                 return TRACE_TYPE_PARTIAL_LINE;
189
190         return TRACE_TYPE_HANDLED;
191 }
192
193 extern char *__bad_type_size(void);
194
195 #define SYSCALL_FIELD(type, name)                                       \
196         sizeof(type) != sizeof(trace.name) ?                            \
197                 __bad_type_size() :                                     \
198                 #type, #name, offsetof(typeof(trace), name),            \
199                 sizeof(trace.name), is_signed_type(type)
200
201 static
202 int  __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
203 {
204         int i;
205         int pos = 0;
206
207         /* When len=0, we just calculate the needed length */
208 #define LEN_OR_ZERO (len ? len - pos : 0)
209
210         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
211         for (i = 0; i < entry->nb_args; i++) {
212                 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
213                                 entry->args[i], sizeof(unsigned long),
214                                 i == entry->nb_args - 1 ? "" : ", ");
215         }
216         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
217
218         for (i = 0; i < entry->nb_args; i++) {
219                 pos += snprintf(buf + pos, LEN_OR_ZERO,
220                                 ", ((unsigned long)(REC->%s))", entry->args[i]);
221         }
222
223 #undef LEN_OR_ZERO
224
225         /* return the length of print_fmt */
226         return pos;
227 }
228
229 static int set_syscall_print_fmt(struct ftrace_event_call *call)
230 {
231         char *print_fmt;
232         int len;
233         struct syscall_metadata *entry = call->data;
234
235         if (entry->enter_event != call) {
236                 call->print_fmt = "\"0x%lx\", REC->ret";
237                 return 0;
238         }
239
240         /* First: called with 0 length to calculate the needed length */
241         len = __set_enter_print_fmt(entry, NULL, 0);
242
243         print_fmt = kmalloc(len + 1, GFP_KERNEL);
244         if (!print_fmt)
245                 return -ENOMEM;
246
247         /* Second: actually write the @print_fmt */
248         __set_enter_print_fmt(entry, print_fmt, len + 1);
249         call->print_fmt = print_fmt;
250
251         return 0;
252 }
253
254 static void free_syscall_print_fmt(struct ftrace_event_call *call)
255 {
256         struct syscall_metadata *entry = call->data;
257
258         if (entry->enter_event == call)
259                 kfree(call->print_fmt);
260 }
261
262 static int syscall_enter_define_fields(struct ftrace_event_call *call)
263 {
264         struct syscall_trace_enter trace;
265         struct syscall_metadata *meta = call->data;
266         int ret;
267         int i;
268         int offset = offsetof(typeof(trace), args);
269
270         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
271         if (ret)
272                 return ret;
273
274         for (i = 0; i < meta->nb_args; i++) {
275                 ret = trace_define_field(call, meta->types[i],
276                                          meta->args[i], offset,
277                                          sizeof(unsigned long), 0,
278                                          FILTER_OTHER);
279                 offset += sizeof(unsigned long);
280         }
281
282         return ret;
283 }
284
285 static int syscall_exit_define_fields(struct ftrace_event_call *call)
286 {
287         struct syscall_trace_exit trace;
288         int ret;
289
290         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
291         if (ret)
292                 return ret;
293
294         ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
295                                  FILTER_OTHER);
296
297         return ret;
298 }
299
300 void ftrace_syscall_enter(void *ignore, struct pt_regs *regs, long id)
301 {
302         struct syscall_trace_enter *entry;
303         struct syscall_metadata *sys_data;
304         struct ring_buffer_event *event;
305         struct ring_buffer *buffer;
306         unsigned long irq_flags;
307         int pc;
308         int size;
309         int syscall_nr;
310
311         syscall_nr = syscall_get_nr(current, regs);
312         if (syscall_nr < 0)
313                 return;
314         if (!test_bit(syscall_nr, enabled_enter_syscalls))
315                 return;
316
317         sys_data = syscall_nr_to_meta(syscall_nr);
318         if (!sys_data)
319                 return;
320
321         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
322
323         local_save_flags(irq_flags);
324         pc = preempt_count();
325
326         event = trace_current_buffer_lock_reserve(&buffer,
327                         sys_data->enter_event->event.type, size, irq_flags, pc);
328         if (!event)
329                 return;
330
331         entry = ring_buffer_event_data(event);
332         entry->nr = syscall_nr;
333         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
334
335         if (!filter_current_check_discard(buffer, sys_data->enter_event,
336                                           entry, event))
337                 trace_current_buffer_unlock_commit(buffer, event,
338                                                    irq_flags, pc);
339 }
340
341 void ftrace_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
342 {
343         struct syscall_trace_exit *entry;
344         struct syscall_metadata *sys_data;
345         struct ring_buffer_event *event;
346         struct ring_buffer *buffer;
347         unsigned long irq_flags;
348         int pc;
349         int syscall_nr;
350
351         syscall_nr = syscall_get_nr(current, regs);
352         if (syscall_nr < 0)
353                 return;
354         if (!test_bit(syscall_nr, enabled_exit_syscalls))
355                 return;
356
357         sys_data = syscall_nr_to_meta(syscall_nr);
358         if (!sys_data)
359                 return;
360
361         local_save_flags(irq_flags);
362         pc = preempt_count();
363
364         event = trace_current_buffer_lock_reserve(&buffer,
365                         sys_data->exit_event->event.type, sizeof(*entry),
366                         irq_flags, pc);
367         if (!event)
368                 return;
369
370         entry = ring_buffer_event_data(event);
371         entry->nr = syscall_nr;
372         entry->ret = syscall_get_return_value(current, regs);
373
374         if (!filter_current_check_discard(buffer, sys_data->exit_event,
375                                           entry, event))
376                 trace_current_buffer_unlock_commit(buffer, event,
377                                                    irq_flags, pc);
378 }
379
380 int reg_event_syscall_enter(struct ftrace_event_call *call)
381 {
382         int ret = 0;
383         int num;
384
385         num = ((struct syscall_metadata *)call->data)->syscall_nr;
386         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
387                 return -ENOSYS;
388         mutex_lock(&syscall_trace_lock);
389         if (!sys_refcount_enter)
390                 ret = register_trace_sys_enter(ftrace_syscall_enter, NULL);
391         if (!ret) {
392                 set_bit(num, enabled_enter_syscalls);
393                 sys_refcount_enter++;
394         }
395         mutex_unlock(&syscall_trace_lock);
396         return ret;
397 }
398
399 void unreg_event_syscall_enter(struct ftrace_event_call *call)
400 {
401         int num;
402
403         num = ((struct syscall_metadata *)call->data)->syscall_nr;
404         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
405                 return;
406         mutex_lock(&syscall_trace_lock);
407         sys_refcount_enter--;
408         clear_bit(num, enabled_enter_syscalls);
409         if (!sys_refcount_enter)
410                 unregister_trace_sys_enter(ftrace_syscall_enter, NULL);
411         mutex_unlock(&syscall_trace_lock);
412 }
413
414 int reg_event_syscall_exit(struct ftrace_event_call *call)
415 {
416         int ret = 0;
417         int num;
418
419         num = ((struct syscall_metadata *)call->data)->syscall_nr;
420         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
421                 return -ENOSYS;
422         mutex_lock(&syscall_trace_lock);
423         if (!sys_refcount_exit)
424                 ret = register_trace_sys_exit(ftrace_syscall_exit, NULL);
425         if (!ret) {
426                 set_bit(num, enabled_exit_syscalls);
427                 sys_refcount_exit++;
428         }
429         mutex_unlock(&syscall_trace_lock);
430         return ret;
431 }
432
433 void unreg_event_syscall_exit(struct ftrace_event_call *call)
434 {
435         int num;
436
437         num = ((struct syscall_metadata *)call->data)->syscall_nr;
438         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
439                 return;
440         mutex_lock(&syscall_trace_lock);
441         sys_refcount_exit--;
442         clear_bit(num, enabled_exit_syscalls);
443         if (!sys_refcount_exit)
444                 unregister_trace_sys_exit(ftrace_syscall_exit, NULL);
445         mutex_unlock(&syscall_trace_lock);
446 }
447
448 int init_syscall_trace(struct ftrace_event_call *call)
449 {
450         int id;
451         int num;
452
453         num = ((struct syscall_metadata *)call->data)->syscall_nr;
454         if (num < 0 || num >= NR_syscalls) {
455                 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
456                                 ((struct syscall_metadata *)call->data)->name);
457                 return -ENOSYS;
458         }
459
460         if (set_syscall_print_fmt(call) < 0)
461                 return -ENOMEM;
462
463         id = trace_event_raw_init(call);
464
465         if (id < 0) {
466                 free_syscall_print_fmt(call);
467                 return id;
468         }
469
470         return id;
471 }
472
473 unsigned long __init __weak arch_syscall_addr(int nr)
474 {
475         return (unsigned long)sys_call_table[nr];
476 }
477
478 int __init init_ftrace_syscalls(void)
479 {
480         struct syscall_metadata *meta;
481         unsigned long addr;
482         int i;
483
484         syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
485                                         NR_syscalls, GFP_KERNEL);
486         if (!syscalls_metadata) {
487                 WARN_ON(1);
488                 return -ENOMEM;
489         }
490
491         for (i = 0; i < NR_syscalls; i++) {
492                 addr = arch_syscall_addr(i);
493                 meta = find_syscall_meta(addr);
494                 if (!meta)
495                         continue;
496
497                 meta->syscall_nr = i;
498                 syscalls_metadata[i] = meta;
499         }
500
501         return 0;
502 }
503 core_initcall(init_ftrace_syscalls);
504
505 #ifdef CONFIG_PERF_EVENTS
506
507 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
508 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
509 static int sys_perf_refcount_enter;
510 static int sys_perf_refcount_exit;
511
512 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
513 {
514         struct syscall_metadata *sys_data;
515         struct syscall_trace_enter *rec;
516         struct hlist_head *head;
517         int syscall_nr;
518         int rctx;
519         int size;
520
521         syscall_nr = syscall_get_nr(current, regs);
522         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
523                 return;
524
525         sys_data = syscall_nr_to_meta(syscall_nr);
526         if (!sys_data)
527                 return;
528
529         /* get the size after alignment with the u32 buffer size field */
530         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
531         size = ALIGN(size + sizeof(u32), sizeof(u64));
532         size -= sizeof(u32);
533
534         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
535                       "perf buffer not large enough"))
536                 return;
537
538         rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size,
539                                 sys_data->enter_event->event.type, regs, &rctx);
540         if (!rec)
541                 return;
542
543         rec->nr = syscall_nr;
544         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
545                                (unsigned long *)&rec->args);
546
547         head = this_cpu_ptr(sys_data->enter_event->perf_events);
548         perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head);
549 }
550
551 int perf_sysenter_enable(struct ftrace_event_call *call)
552 {
553         int ret = 0;
554         int num;
555
556         num = ((struct syscall_metadata *)call->data)->syscall_nr;
557
558         mutex_lock(&syscall_trace_lock);
559         if (!sys_perf_refcount_enter)
560                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
561         if (ret) {
562                 pr_info("event trace: Could not activate"
563                                 "syscall entry trace point");
564         } else {
565                 set_bit(num, enabled_perf_enter_syscalls);
566                 sys_perf_refcount_enter++;
567         }
568         mutex_unlock(&syscall_trace_lock);
569         return ret;
570 }
571
572 void perf_sysenter_disable(struct ftrace_event_call *call)
573 {
574         int num;
575
576         num = ((struct syscall_metadata *)call->data)->syscall_nr;
577
578         mutex_lock(&syscall_trace_lock);
579         sys_perf_refcount_enter--;
580         clear_bit(num, enabled_perf_enter_syscalls);
581         if (!sys_perf_refcount_enter)
582                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
583         mutex_unlock(&syscall_trace_lock);
584 }
585
586 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
587 {
588         struct syscall_metadata *sys_data;
589         struct syscall_trace_exit *rec;
590         struct hlist_head *head;
591         int syscall_nr;
592         int rctx;
593         int size;
594
595         syscall_nr = syscall_get_nr(current, regs);
596         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
597                 return;
598
599         sys_data = syscall_nr_to_meta(syscall_nr);
600         if (!sys_data)
601                 return;
602
603         /* We can probably do that at build time */
604         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
605         size -= sizeof(u32);
606
607         /*
608          * Impossible, but be paranoid with the future
609          * How to put this check outside runtime?
610          */
611         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
612                 "exit event has grown above perf buffer size"))
613                 return;
614
615         rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size,
616                                 sys_data->exit_event->event.type, regs, &rctx);
617         if (!rec)
618                 return;
619
620         rec->nr = syscall_nr;
621         rec->ret = syscall_get_return_value(current, regs);
622
623         head = this_cpu_ptr(sys_data->exit_event->perf_events);
624         perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head);
625 }
626
627 int perf_sysexit_enable(struct ftrace_event_call *call)
628 {
629         int ret = 0;
630         int num;
631
632         num = ((struct syscall_metadata *)call->data)->syscall_nr;
633
634         mutex_lock(&syscall_trace_lock);
635         if (!sys_perf_refcount_exit)
636                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
637         if (ret) {
638                 pr_info("event trace: Could not activate"
639                                 "syscall exit trace point");
640         } else {
641                 set_bit(num, enabled_perf_exit_syscalls);
642                 sys_perf_refcount_exit++;
643         }
644         mutex_unlock(&syscall_trace_lock);
645         return ret;
646 }
647
648 void perf_sysexit_disable(struct ftrace_event_call *call)
649 {
650         int num;
651
652         num = ((struct syscall_metadata *)call->data)->syscall_nr;
653
654         mutex_lock(&syscall_trace_lock);
655         sys_perf_refcount_exit--;
656         clear_bit(num, enabled_perf_exit_syscalls);
657         if (!sys_perf_refcount_exit)
658                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
659         mutex_unlock(&syscall_trace_lock);
660 }
661
662 #endif /* CONFIG_PERF_EVENTS */
663
664 static int syscall_enter_register(struct ftrace_event_call *event,
665                                  enum trace_reg type)
666 {
667         switch (type) {
668         case TRACE_REG_REGISTER:
669                 return reg_event_syscall_enter(event);
670         case TRACE_REG_UNREGISTER:
671                 unreg_event_syscall_enter(event);
672                 return 0;
673
674 #ifdef CONFIG_PERF_EVENTS
675         case TRACE_REG_PERF_REGISTER:
676                 return perf_sysenter_enable(event);
677         case TRACE_REG_PERF_UNREGISTER:
678                 perf_sysenter_disable(event);
679                 return 0;
680 #endif
681         }
682         return 0;
683 }
684
685 static int syscall_exit_register(struct ftrace_event_call *event,
686                                  enum trace_reg type)
687 {
688         switch (type) {
689         case TRACE_REG_REGISTER:
690                 return reg_event_syscall_exit(event);
691         case TRACE_REG_UNREGISTER:
692                 unreg_event_syscall_exit(event);
693                 return 0;
694
695 #ifdef CONFIG_PERF_EVENTS
696         case TRACE_REG_PERF_REGISTER:
697                 return perf_sysexit_enable(event);
698         case TRACE_REG_PERF_UNREGISTER:
699                 perf_sysexit_disable(event);
700                 return 0;
701 #endif
702         }
703         return 0;
704 }