Commit | Line | Data |
---|---|---|
47788c58 | 1 | #include <trace/syscall.h> |
1c569f02 | 2 | #include <trace/events/syscalls.h> |
f431b634 | 3 | #include <linux/syscalls.h> |
5a0e3ad6 | 4 | #include <linux/slab.h> |
ee08c6ec | 5 | #include <linux/kernel.h> |
56d82e00 | 6 | #include <linux/module.h> /* for MODULE_NAME_LEN via KSYM_SYMBOL_LEN */ |
fb34a08c | 7 | #include <linux/ftrace.h> |
cdd6c482 | 8 | #include <linux/perf_event.h> |
ee08c6ec FW |
9 | #include <asm/syscall.h> |
10 | ||
11 | #include "trace_output.h" | |
12 | #include "trace.h" | |
13 | ||
5be71b61 | 14 | static DEFINE_MUTEX(syscall_trace_lock); |
ee08c6ec | 15 | |
2239291a | 16 | static int syscall_enter_register(struct ftrace_event_call *event, |
ceec0b6f | 17 | enum trace_reg type, void *data); |
2239291a | 18 | static int syscall_exit_register(struct ftrace_event_call *event, |
ceec0b6f | 19 | enum trace_reg type, void *data); |
2239291a | 20 | |
2e33af02 SR |
21 | static struct list_head * |
22 | syscall_get_enter_fields(struct ftrace_event_call *call) | |
23 | { | |
24 | struct syscall_metadata *entry = call->data; | |
25 | ||
26 | return &entry->enter_fields; | |
27 | } | |
28 | ||
3d56e331 SR |
29 | extern struct syscall_metadata *__start_syscalls_metadata[]; |
30 | extern struct syscall_metadata *__stop_syscalls_metadata[]; | |
c44fc770 FW |
31 | |
32 | static struct syscall_metadata **syscalls_metadata; | |
33 | ||
b2d55496 IM |
34 | #ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME |
35 | static inline bool arch_syscall_match_sym_name(const char *sym, const char *name) | |
36 | { | |
37 | /* | |
38 | * Only compare after the "sys" prefix. Archs that use | |
39 | * syscall wrappers may have syscalls symbols aliases prefixed | |
36a78e9e | 40 | * with ".SyS" or ".sys" instead of "sys", leading to an unwanted |
b2d55496 IM |
41 | * mismatch. |
42 | */ | |
43 | return !strcmp(sym + 3, name + 3); | |
44 | } | |
45 | #endif | |
46 | ||
f431b634 SR |
47 | #ifdef ARCH_TRACE_IGNORE_COMPAT_SYSCALLS |
48 | /* | |
49 | * Some architectures that allow for 32bit applications | |
50 | * to run on a 64bit kernel, do not map the syscalls for | |
51 | * the 32bit tasks the same as they do for 64bit tasks. | |
52 | * | |
53 | * *cough*x86*cough* | |
54 | * | |
55 | * In such a case, instead of reporting the wrong syscalls, | |
56 | * simply ignore them. | |
57 | * | |
58 | * For an arch to ignore the compat syscalls it needs to | |
59 | * define ARCH_TRACE_IGNORE_COMPAT_SYSCALLS as well as | |
60 | * define the function arch_trace_is_compat_syscall() to let | |
61 | * the tracing system know that it should ignore it. | |
62 | */ | |
63 | static int | |
64 | trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs) | |
65 | { | |
66 | if (unlikely(arch_trace_is_compat_syscall(regs))) | |
67 | return -1; | |
68 | ||
69 | return syscall_get_nr(task, regs); | |
70 | } | |
71 | #else | |
72 | static inline int | |
73 | trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs) | |
74 | { | |
75 | return syscall_get_nr(task, regs); | |
76 | } | |
77 | #endif /* ARCH_TRACE_IGNORE_COMPAT_SYSCALLS */ | |
78 | ||
3d56e331 SR |
79 | static __init struct syscall_metadata * |
80 | find_syscall_meta(unsigned long syscall) | |
c44fc770 | 81 | { |
3d56e331 SR |
82 | struct syscall_metadata **start; |
83 | struct syscall_metadata **stop; | |
c44fc770 FW |
84 | char str[KSYM_SYMBOL_LEN]; |
85 | ||
86 | ||
3d56e331 SR |
87 | start = __start_syscalls_metadata; |
88 | stop = __stop_syscalls_metadata; | |
c44fc770 FW |
89 | kallsyms_lookup(syscall, NULL, NULL, NULL, str); |
90 | ||
ae07f551 IM |
91 | if (arch_syscall_match_sym_name(str, "sys_ni_syscall")) |
92 | return NULL; | |
93 | ||
c44fc770 | 94 | for ( ; start < stop; start++) { |
b2d55496 | 95 | if ((*start)->name && arch_syscall_match_sym_name(str, (*start)->name)) |
3d56e331 | 96 | return *start; |
c44fc770 FW |
97 | } |
98 | return NULL; | |
99 | } | |
100 | ||
101 | static struct syscall_metadata *syscall_nr_to_meta(int nr) | |
102 | { | |
103 | if (!syscalls_metadata || nr >= NR_syscalls || nr < 0) | |
104 | return NULL; | |
105 | ||
106 | return syscalls_metadata[nr]; | |
107 | } | |
108 | ||
6aea49cb | 109 | static enum print_line_t |
a9a57763 SR |
110 | print_syscall_enter(struct trace_iterator *iter, int flags, |
111 | struct trace_event *event) | |
bed1ffca FW |
112 | { |
113 | struct trace_seq *s = &iter->seq; | |
114 | struct trace_entry *ent = iter->ent; | |
115 | struct syscall_trace_enter *trace; | |
116 | struct syscall_metadata *entry; | |
117 | int i, ret, syscall; | |
118 | ||
64c12e04 | 119 | trace = (typeof(trace))ent; |
bed1ffca | 120 | syscall = trace->nr; |
bed1ffca | 121 | entry = syscall_nr_to_meta(syscall); |
64c12e04 | 122 | |
bed1ffca FW |
123 | if (!entry) |
124 | goto end; | |
125 | ||
32c0edae | 126 | if (entry->enter_event->event.type != ent->type) { |
64c12e04 JB |
127 | WARN_ON_ONCE(1); |
128 | goto end; | |
129 | } | |
130 | ||
bed1ffca FW |
131 | ret = trace_seq_printf(s, "%s(", entry->name); |
132 | if (!ret) | |
133 | return TRACE_TYPE_PARTIAL_LINE; | |
134 | ||
135 | for (i = 0; i < entry->nb_args; i++) { | |
136 | /* parameter types */ | |
ba8b3a40 | 137 | if (trace_flags & TRACE_ITER_VERBOSE) { |
bed1ffca FW |
138 | ret = trace_seq_printf(s, "%s ", entry->types[i]); |
139 | if (!ret) | |
140 | return TRACE_TYPE_PARTIAL_LINE; | |
141 | } | |
142 | /* parameter values */ | |
4539f077 | 143 | ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i], |
bed1ffca | 144 | trace->args[i], |
4539f077 | 145 | i == entry->nb_args - 1 ? "" : ", "); |
bed1ffca FW |
146 | if (!ret) |
147 | return TRACE_TYPE_PARTIAL_LINE; | |
148 | } | |
149 | ||
4539f077 LZ |
150 | ret = trace_seq_putc(s, ')'); |
151 | if (!ret) | |
152 | return TRACE_TYPE_PARTIAL_LINE; | |
153 | ||
bed1ffca | 154 | end: |
4539f077 LZ |
155 | ret = trace_seq_putc(s, '\n'); |
156 | if (!ret) | |
157 | return TRACE_TYPE_PARTIAL_LINE; | |
158 | ||
bed1ffca FW |
159 | return TRACE_TYPE_HANDLED; |
160 | } | |
161 | ||
6aea49cb | 162 | static enum print_line_t |
a9a57763 SR |
163 | print_syscall_exit(struct trace_iterator *iter, int flags, |
164 | struct trace_event *event) | |
bed1ffca FW |
165 | { |
166 | struct trace_seq *s = &iter->seq; | |
167 | struct trace_entry *ent = iter->ent; | |
168 | struct syscall_trace_exit *trace; | |
169 | int syscall; | |
170 | struct syscall_metadata *entry; | |
171 | int ret; | |
172 | ||
64c12e04 | 173 | trace = (typeof(trace))ent; |
bed1ffca | 174 | syscall = trace->nr; |
bed1ffca | 175 | entry = syscall_nr_to_meta(syscall); |
64c12e04 | 176 | |
bed1ffca | 177 | if (!entry) { |
146c3442 | 178 | trace_seq_putc(s, '\n'); |
bed1ffca FW |
179 | return TRACE_TYPE_HANDLED; |
180 | } | |
181 | ||
32c0edae | 182 | if (entry->exit_event->event.type != ent->type) { |
64c12e04 JB |
183 | WARN_ON_ONCE(1); |
184 | return TRACE_TYPE_UNHANDLED; | |
185 | } | |
186 | ||
bed1ffca FW |
187 | ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name, |
188 | trace->ret); | |
189 | if (!ret) | |
190 | return TRACE_TYPE_PARTIAL_LINE; | |
191 | ||
192 | return TRACE_TYPE_HANDLED; | |
193 | } | |
194 | ||
e6971969 LZ |
195 | extern char *__bad_type_size(void); |
196 | ||
197 | #define SYSCALL_FIELD(type, name) \ | |
198 | sizeof(type) != sizeof(trace.name) ? \ | |
199 | __bad_type_size() : \ | |
26a50744 TZ |
200 | #type, #name, offsetof(typeof(trace), name), \ |
201 | sizeof(trace.name), is_signed_type(type) | |
e6971969 | 202 | |
3ddc77f6 LZ |
203 | static int __init |
204 | __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len) | |
50307a45 LJ |
205 | { |
206 | int i; | |
207 | int pos = 0; | |
208 | ||
209 | /* When len=0, we just calculate the needed length */ | |
210 | #define LEN_OR_ZERO (len ? len - pos : 0) | |
211 | ||
212 | pos += snprintf(buf + pos, LEN_OR_ZERO, "\""); | |
213 | for (i = 0; i < entry->nb_args; i++) { | |
214 | pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s", | |
215 | entry->args[i], sizeof(unsigned long), | |
216 | i == entry->nb_args - 1 ? "" : ", "); | |
217 | } | |
218 | pos += snprintf(buf + pos, LEN_OR_ZERO, "\""); | |
219 | ||
220 | for (i = 0; i < entry->nb_args; i++) { | |
221 | pos += snprintf(buf + pos, LEN_OR_ZERO, | |
222 | ", ((unsigned long)(REC->%s))", entry->args[i]); | |
223 | } | |
224 | ||
225 | #undef LEN_OR_ZERO | |
226 | ||
227 | /* return the length of print_fmt */ | |
228 | return pos; | |
229 | } | |
230 | ||
3ddc77f6 | 231 | static int __init set_syscall_print_fmt(struct ftrace_event_call *call) |
50307a45 LJ |
232 | { |
233 | char *print_fmt; | |
234 | int len; | |
235 | struct syscall_metadata *entry = call->data; | |
236 | ||
237 | if (entry->enter_event != call) { | |
238 | call->print_fmt = "\"0x%lx\", REC->ret"; | |
239 | return 0; | |
240 | } | |
241 | ||
242 | /* First: called with 0 length to calculate the needed length */ | |
243 | len = __set_enter_print_fmt(entry, NULL, 0); | |
244 | ||
245 | print_fmt = kmalloc(len + 1, GFP_KERNEL); | |
246 | if (!print_fmt) | |
247 | return -ENOMEM; | |
248 | ||
249 | /* Second: actually write the @print_fmt */ | |
250 | __set_enter_print_fmt(entry, print_fmt, len + 1); | |
251 | call->print_fmt = print_fmt; | |
252 | ||
253 | return 0; | |
254 | } | |
255 | ||
3ddc77f6 | 256 | static void __init free_syscall_print_fmt(struct ftrace_event_call *call) |
50307a45 LJ |
257 | { |
258 | struct syscall_metadata *entry = call->data; | |
259 | ||
260 | if (entry->enter_event == call) | |
261 | kfree(call->print_fmt); | |
262 | } | |
263 | ||
b8aae39f | 264 | static int __init syscall_enter_define_fields(struct ftrace_event_call *call) |
540b7b8d LZ |
265 | { |
266 | struct syscall_trace_enter trace; | |
31c16b13 | 267 | struct syscall_metadata *meta = call->data; |
540b7b8d | 268 | int ret; |
540b7b8d LZ |
269 | int i; |
270 | int offset = offsetof(typeof(trace), args); | |
271 | ||
0f1ef51d LJ |
272 | ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER); |
273 | if (ret) | |
274 | return ret; | |
275 | ||
540b7b8d | 276 | for (i = 0; i < meta->nb_args; i++) { |
aeaeae11 FW |
277 | ret = trace_define_field(call, meta->types[i], |
278 | meta->args[i], offset, | |
43b51ead LZ |
279 | sizeof(unsigned long), 0, |
280 | FILTER_OTHER); | |
540b7b8d LZ |
281 | offset += sizeof(unsigned long); |
282 | } | |
283 | ||
284 | return ret; | |
285 | } | |
286 | ||
b8aae39f | 287 | static int __init syscall_exit_define_fields(struct ftrace_event_call *call) |
540b7b8d LZ |
288 | { |
289 | struct syscall_trace_exit trace; | |
290 | int ret; | |
291 | ||
0f1ef51d LJ |
292 | ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER); |
293 | if (ret) | |
294 | return ret; | |
295 | ||
26a50744 | 296 | ret = trace_define_field(call, SYSCALL_FIELD(long, ret), |
43b51ead | 297 | FILTER_OTHER); |
540b7b8d LZ |
298 | |
299 | return ret; | |
300 | } | |
301 | ||
12ab74ee | 302 | static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id) |
ee08c6ec | 303 | { |
12ab74ee | 304 | struct trace_array *tr = data; |
bed1ffca FW |
305 | struct syscall_trace_enter *entry; |
306 | struct syscall_metadata *sys_data; | |
307 | struct ring_buffer_event *event; | |
e77405ad | 308 | struct ring_buffer *buffer; |
11034ae9 J |
309 | unsigned long irq_flags; |
310 | int pc; | |
ee08c6ec | 311 | int syscall_nr; |
f431b634 | 312 | int size; |
ee08c6ec | 313 | |
f431b634 | 314 | syscall_nr = trace_get_syscall_nr(current, regs); |
cd0980fc HB |
315 | if (syscall_nr < 0) |
316 | return; | |
12ab74ee | 317 | if (!test_bit(syscall_nr, tr->enabled_enter_syscalls)) |
fb34a08c | 318 | return; |
ee08c6ec | 319 | |
bed1ffca FW |
320 | sys_data = syscall_nr_to_meta(syscall_nr); |
321 | if (!sys_data) | |
322 | return; | |
323 | ||
324 | size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args; | |
325 | ||
11034ae9 J |
326 | local_save_flags(irq_flags); |
327 | pc = preempt_count(); | |
328 | ||
12883efb | 329 | buffer = tr->trace_buffer.buffer; |
12ab74ee | 330 | event = trace_buffer_lock_reserve(buffer, |
11034ae9 | 331 | sys_data->enter_event->event.type, size, irq_flags, pc); |
bed1ffca FW |
332 | if (!event) |
333 | return; | |
334 | ||
335 | entry = ring_buffer_event_data(event); | |
336 | entry->nr = syscall_nr; | |
337 | syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args); | |
338 | ||
e77405ad SR |
339 | if (!filter_current_check_discard(buffer, sys_data->enter_event, |
340 | entry, event)) | |
11034ae9 J |
341 | trace_current_buffer_unlock_commit(buffer, event, |
342 | irq_flags, pc); | |
ee08c6ec FW |
343 | } |
344 | ||
12ab74ee | 345 | static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret) |
ee08c6ec | 346 | { |
12ab74ee | 347 | struct trace_array *tr = data; |
bed1ffca FW |
348 | struct syscall_trace_exit *entry; |
349 | struct syscall_metadata *sys_data; | |
350 | struct ring_buffer_event *event; | |
e77405ad | 351 | struct ring_buffer *buffer; |
11034ae9 J |
352 | unsigned long irq_flags; |
353 | int pc; | |
ee08c6ec FW |
354 | int syscall_nr; |
355 | ||
f431b634 | 356 | syscall_nr = trace_get_syscall_nr(current, regs); |
cd0980fc HB |
357 | if (syscall_nr < 0) |
358 | return; | |
12ab74ee | 359 | if (!test_bit(syscall_nr, tr->enabled_exit_syscalls)) |
fb34a08c | 360 | return; |
ee08c6ec | 361 | |
bed1ffca FW |
362 | sys_data = syscall_nr_to_meta(syscall_nr); |
363 | if (!sys_data) | |
364 | return; | |
365 | ||
11034ae9 J |
366 | local_save_flags(irq_flags); |
367 | pc = preempt_count(); | |
368 | ||
12883efb | 369 | buffer = tr->trace_buffer.buffer; |
12ab74ee | 370 | event = trace_buffer_lock_reserve(buffer, |
11034ae9 J |
371 | sys_data->exit_event->event.type, sizeof(*entry), |
372 | irq_flags, pc); | |
bed1ffca FW |
373 | if (!event) |
374 | return; | |
375 | ||
376 | entry = ring_buffer_event_data(event); | |
377 | entry->nr = syscall_nr; | |
378 | entry->ret = syscall_get_return_value(current, regs); | |
379 | ||
e77405ad SR |
380 | if (!filter_current_check_discard(buffer, sys_data->exit_event, |
381 | entry, event)) | |
11034ae9 J |
382 | trace_current_buffer_unlock_commit(buffer, event, |
383 | irq_flags, pc); | |
ee08c6ec FW |
384 | } |
385 | ||
12ab74ee SR |
386 | static int reg_event_syscall_enter(struct ftrace_event_file *file, |
387 | struct ftrace_event_call *call) | |
ee08c6ec | 388 | { |
12ab74ee | 389 | struct trace_array *tr = file->tr; |
fb34a08c JB |
390 | int ret = 0; |
391 | int num; | |
fb34a08c | 392 | |
c252f657 | 393 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 394 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
395 | return -ENOSYS; |
396 | mutex_lock(&syscall_trace_lock); | |
12ab74ee SR |
397 | if (!tr->sys_refcount_enter) |
398 | ret = register_trace_sys_enter(ftrace_syscall_enter, tr); | |
3b8e4273 | 399 | if (!ret) { |
12ab74ee SR |
400 | set_bit(num, tr->enabled_enter_syscalls); |
401 | tr->sys_refcount_enter++; | |
fb34a08c JB |
402 | } |
403 | mutex_unlock(&syscall_trace_lock); | |
404 | return ret; | |
ee08c6ec FW |
405 | } |
406 | ||
12ab74ee SR |
407 | static void unreg_event_syscall_enter(struct ftrace_event_file *file, |
408 | struct ftrace_event_call *call) | |
ee08c6ec | 409 | { |
12ab74ee | 410 | struct trace_array *tr = file->tr; |
fb34a08c | 411 | int num; |
ee08c6ec | 412 | |
c252f657 | 413 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 414 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
415 | return; |
416 | mutex_lock(&syscall_trace_lock); | |
12ab74ee SR |
417 | tr->sys_refcount_enter--; |
418 | clear_bit(num, tr->enabled_enter_syscalls); | |
419 | if (!tr->sys_refcount_enter) | |
420 | unregister_trace_sys_enter(ftrace_syscall_enter, tr); | |
fb34a08c JB |
421 | mutex_unlock(&syscall_trace_lock); |
422 | } | |
ee08c6ec | 423 | |
12ab74ee SR |
424 | static int reg_event_syscall_exit(struct ftrace_event_file *file, |
425 | struct ftrace_event_call *call) | |
ee08c6ec | 426 | { |
12ab74ee | 427 | struct trace_array *tr = file->tr; |
fb34a08c JB |
428 | int ret = 0; |
429 | int num; | |
fb34a08c | 430 | |
c252f657 | 431 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 432 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
433 | return -ENOSYS; |
434 | mutex_lock(&syscall_trace_lock); | |
12ab74ee SR |
435 | if (!tr->sys_refcount_exit) |
436 | ret = register_trace_sys_exit(ftrace_syscall_exit, tr); | |
3b8e4273 | 437 | if (!ret) { |
12ab74ee SR |
438 | set_bit(num, tr->enabled_exit_syscalls); |
439 | tr->sys_refcount_exit++; | |
ee08c6ec | 440 | } |
fb34a08c JB |
441 | mutex_unlock(&syscall_trace_lock); |
442 | return ret; | |
443 | } | |
ee08c6ec | 444 | |
12ab74ee SR |
445 | static void unreg_event_syscall_exit(struct ftrace_event_file *file, |
446 | struct ftrace_event_call *call) | |
fb34a08c | 447 | { |
12ab74ee | 448 | struct trace_array *tr = file->tr; |
fb34a08c | 449 | int num; |
ee08c6ec | 450 | |
c252f657 | 451 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 452 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
453 | return; |
454 | mutex_lock(&syscall_trace_lock); | |
12ab74ee SR |
455 | tr->sys_refcount_exit--; |
456 | clear_bit(num, tr->enabled_exit_syscalls); | |
457 | if (!tr->sys_refcount_exit) | |
458 | unregister_trace_sys_exit(ftrace_syscall_exit, tr); | |
fb34a08c | 459 | mutex_unlock(&syscall_trace_lock); |
ee08c6ec | 460 | } |
fb34a08c | 461 | |
3ddc77f6 | 462 | static int __init init_syscall_trace(struct ftrace_event_call *call) |
a1301da0 LJ |
463 | { |
464 | int id; | |
ba976970 IM |
465 | int num; |
466 | ||
467 | num = ((struct syscall_metadata *)call->data)->syscall_nr; | |
468 | if (num < 0 || num >= NR_syscalls) { | |
469 | pr_debug("syscall %s metadata not mapped, disabling ftrace event\n", | |
470 | ((struct syscall_metadata *)call->data)->name); | |
471 | return -ENOSYS; | |
472 | } | |
a1301da0 | 473 | |
50307a45 LJ |
474 | if (set_syscall_print_fmt(call) < 0) |
475 | return -ENOMEM; | |
476 | ||
c7ef3a90 SR |
477 | id = trace_event_raw_init(call); |
478 | ||
479 | if (id < 0) { | |
50307a45 | 480 | free_syscall_print_fmt(call); |
c7ef3a90 | 481 | return id; |
50307a45 | 482 | } |
c7ef3a90 SR |
483 | |
484 | return id; | |
a1301da0 LJ |
485 | } |
486 | ||
6f86ab9f VN |
487 | struct trace_event_functions enter_syscall_print_funcs = { |
488 | .trace = print_syscall_enter, | |
489 | }; | |
490 | ||
491 | struct trace_event_functions exit_syscall_print_funcs = { | |
492 | .trace = print_syscall_exit, | |
493 | }; | |
494 | ||
523c8113 | 495 | struct ftrace_event_class __refdata event_class_syscall_enter = { |
6f86ab9f VN |
496 | .system = "syscalls", |
497 | .reg = syscall_enter_register, | |
498 | .define_fields = syscall_enter_define_fields, | |
499 | .get_fields = syscall_get_enter_fields, | |
500 | .raw_init = init_syscall_trace, | |
501 | }; | |
502 | ||
523c8113 | 503 | struct ftrace_event_class __refdata event_class_syscall_exit = { |
6f86ab9f VN |
504 | .system = "syscalls", |
505 | .reg = syscall_exit_register, | |
506 | .define_fields = syscall_exit_define_fields, | |
507 | .fields = LIST_HEAD_INIT(event_class_syscall_exit.fields), | |
508 | .raw_init = init_syscall_trace, | |
509 | }; | |
510 | ||
c763ba06 | 511 | unsigned long __init __weak arch_syscall_addr(int nr) |
e7b8e675 MF |
512 | { |
513 | return (unsigned long)sys_call_table[nr]; | |
514 | } | |
515 | ||
6aea49cb | 516 | static int __init init_ftrace_syscalls(void) |
c44fc770 FW |
517 | { |
518 | struct syscall_metadata *meta; | |
519 | unsigned long addr; | |
520 | int i; | |
521 | ||
47b0edcb TM |
522 | syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata), |
523 | GFP_KERNEL); | |
c44fc770 FW |
524 | if (!syscalls_metadata) { |
525 | WARN_ON(1); | |
526 | return -ENOMEM; | |
527 | } | |
528 | ||
529 | for (i = 0; i < NR_syscalls; i++) { | |
530 | addr = arch_syscall_addr(i); | |
531 | meta = find_syscall_meta(addr); | |
c252f657 LJ |
532 | if (!meta) |
533 | continue; | |
534 | ||
535 | meta->syscall_nr = i; | |
c44fc770 FW |
536 | syscalls_metadata[i] = meta; |
537 | } | |
538 | ||
539 | return 0; | |
540 | } | |
8781915a | 541 | early_initcall(init_ftrace_syscalls); |
c44fc770 | 542 | |
07b139c8 | 543 | #ifdef CONFIG_PERF_EVENTS |
19007a67 | 544 | |
97d5a220 FW |
545 | static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls); |
546 | static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls); | |
547 | static int sys_perf_refcount_enter; | |
548 | static int sys_perf_refcount_exit; | |
f4b5ffcc | 549 | |
38516ab5 | 550 | static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id) |
f4b5ffcc JB |
551 | { |
552 | struct syscall_metadata *sys_data; | |
20ab4425 | 553 | struct syscall_trace_enter *rec; |
1c024eca | 554 | struct hlist_head *head; |
f4b5ffcc | 555 | int syscall_nr; |
4ed7c92d | 556 | int rctx; |
19007a67 | 557 | int size; |
f4b5ffcc | 558 | |
f431b634 | 559 | syscall_nr = trace_get_syscall_nr(current, regs); |
60916a93 WD |
560 | if (syscall_nr < 0) |
561 | return; | |
97d5a220 | 562 | if (!test_bit(syscall_nr, enabled_perf_enter_syscalls)) |
f4b5ffcc JB |
563 | return; |
564 | ||
565 | sys_data = syscall_nr_to_meta(syscall_nr); | |
566 | if (!sys_data) | |
567 | return; | |
568 | ||
421c7860 ON |
569 | head = this_cpu_ptr(sys_data->enter_event->perf_events); |
570 | if (hlist_empty(head)) | |
571 | return; | |
572 | ||
19007a67 FW |
573 | /* get the size after alignment with the u32 buffer size field */ |
574 | size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec); | |
575 | size = ALIGN(size + sizeof(u32), sizeof(u64)); | |
576 | size -= sizeof(u32); | |
577 | ||
97d5a220 | 578 | rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size, |
ff5f149b | 579 | sys_data->enter_event->event.type, regs, &rctx); |
430ad5a6 XG |
580 | if (!rec) |
581 | return; | |
20ab4425 | 582 | |
20ab4425 FW |
583 | rec->nr = syscall_nr; |
584 | syscall_get_arguments(current, regs, 0, sys_data->nb_args, | |
585 | (unsigned long *)&rec->args); | |
e6dab5ff | 586 | perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL); |
f4b5ffcc JB |
587 | } |
588 | ||
6f86ab9f | 589 | static int perf_sysenter_enable(struct ftrace_event_call *call) |
f4b5ffcc JB |
590 | { |
591 | int ret = 0; | |
592 | int num; | |
593 | ||
3bbe84e9 | 594 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
595 | |
596 | mutex_lock(&syscall_trace_lock); | |
97d5a220 | 597 | if (!sys_perf_refcount_enter) |
38516ab5 | 598 | ret = register_trace_sys_enter(perf_syscall_enter, NULL); |
f4b5ffcc JB |
599 | if (ret) { |
600 | pr_info("event trace: Could not activate" | |
601 | "syscall entry trace point"); | |
602 | } else { | |
97d5a220 FW |
603 | set_bit(num, enabled_perf_enter_syscalls); |
604 | sys_perf_refcount_enter++; | |
f4b5ffcc JB |
605 | } |
606 | mutex_unlock(&syscall_trace_lock); | |
607 | return ret; | |
608 | } | |
609 | ||
6f86ab9f | 610 | static void perf_sysenter_disable(struct ftrace_event_call *call) |
f4b5ffcc JB |
611 | { |
612 | int num; | |
613 | ||
3bbe84e9 | 614 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
615 | |
616 | mutex_lock(&syscall_trace_lock); | |
97d5a220 FW |
617 | sys_perf_refcount_enter--; |
618 | clear_bit(num, enabled_perf_enter_syscalls); | |
619 | if (!sys_perf_refcount_enter) | |
38516ab5 | 620 | unregister_trace_sys_enter(perf_syscall_enter, NULL); |
f4b5ffcc JB |
621 | mutex_unlock(&syscall_trace_lock); |
622 | } | |
623 | ||
38516ab5 | 624 | static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret) |
f4b5ffcc JB |
625 | { |
626 | struct syscall_metadata *sys_data; | |
20ab4425 | 627 | struct syscall_trace_exit *rec; |
1c024eca | 628 | struct hlist_head *head; |
f4b5ffcc | 629 | int syscall_nr; |
4ed7c92d | 630 | int rctx; |
20ab4425 | 631 | int size; |
f4b5ffcc | 632 | |
f431b634 | 633 | syscall_nr = trace_get_syscall_nr(current, regs); |
60916a93 WD |
634 | if (syscall_nr < 0) |
635 | return; | |
97d5a220 | 636 | if (!test_bit(syscall_nr, enabled_perf_exit_syscalls)) |
f4b5ffcc JB |
637 | return; |
638 | ||
639 | sys_data = syscall_nr_to_meta(syscall_nr); | |
640 | if (!sys_data) | |
641 | return; | |
642 | ||
421c7860 ON |
643 | head = this_cpu_ptr(sys_data->exit_event->perf_events); |
644 | if (hlist_empty(head)) | |
645 | return; | |
646 | ||
20ab4425 FW |
647 | /* We can probably do that at build time */ |
648 | size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64)); | |
649 | size -= sizeof(u32); | |
19007a67 | 650 | |
97d5a220 | 651 | rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size, |
ff5f149b | 652 | sys_data->exit_event->event.type, regs, &rctx); |
430ad5a6 XG |
653 | if (!rec) |
654 | return; | |
20ab4425 | 655 | |
20ab4425 FW |
656 | rec->nr = syscall_nr; |
657 | rec->ret = syscall_get_return_value(current, regs); | |
e6dab5ff | 658 | perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL); |
f4b5ffcc JB |
659 | } |
660 | ||
6f86ab9f | 661 | static int perf_sysexit_enable(struct ftrace_event_call *call) |
f4b5ffcc JB |
662 | { |
663 | int ret = 0; | |
664 | int num; | |
665 | ||
3bbe84e9 | 666 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
667 | |
668 | mutex_lock(&syscall_trace_lock); | |
97d5a220 | 669 | if (!sys_perf_refcount_exit) |
38516ab5 | 670 | ret = register_trace_sys_exit(perf_syscall_exit, NULL); |
f4b5ffcc JB |
671 | if (ret) { |
672 | pr_info("event trace: Could not activate" | |
6574658b | 673 | "syscall exit trace point"); |
f4b5ffcc | 674 | } else { |
97d5a220 FW |
675 | set_bit(num, enabled_perf_exit_syscalls); |
676 | sys_perf_refcount_exit++; | |
f4b5ffcc JB |
677 | } |
678 | mutex_unlock(&syscall_trace_lock); | |
679 | return ret; | |
680 | } | |
681 | ||
6f86ab9f | 682 | static void perf_sysexit_disable(struct ftrace_event_call *call) |
f4b5ffcc JB |
683 | { |
684 | int num; | |
685 | ||
3bbe84e9 | 686 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
687 | |
688 | mutex_lock(&syscall_trace_lock); | |
97d5a220 FW |
689 | sys_perf_refcount_exit--; |
690 | clear_bit(num, enabled_perf_exit_syscalls); | |
691 | if (!sys_perf_refcount_exit) | |
38516ab5 | 692 | unregister_trace_sys_exit(perf_syscall_exit, NULL); |
f4b5ffcc JB |
693 | mutex_unlock(&syscall_trace_lock); |
694 | } | |
695 | ||
07b139c8 | 696 | #endif /* CONFIG_PERF_EVENTS */ |
f4b5ffcc | 697 | |
2239291a | 698 | static int syscall_enter_register(struct ftrace_event_call *event, |
ceec0b6f | 699 | enum trace_reg type, void *data) |
2239291a | 700 | { |
12ab74ee SR |
701 | struct ftrace_event_file *file = data; |
702 | ||
2239291a SR |
703 | switch (type) { |
704 | case TRACE_REG_REGISTER: | |
12ab74ee | 705 | return reg_event_syscall_enter(file, event); |
2239291a | 706 | case TRACE_REG_UNREGISTER: |
12ab74ee | 707 | unreg_event_syscall_enter(file, event); |
2239291a SR |
708 | return 0; |
709 | ||
710 | #ifdef CONFIG_PERF_EVENTS | |
711 | case TRACE_REG_PERF_REGISTER: | |
712 | return perf_sysenter_enable(event); | |
713 | case TRACE_REG_PERF_UNREGISTER: | |
714 | perf_sysenter_disable(event); | |
715 | return 0; | |
ceec0b6f JO |
716 | case TRACE_REG_PERF_OPEN: |
717 | case TRACE_REG_PERF_CLOSE: | |
489c75c3 JO |
718 | case TRACE_REG_PERF_ADD: |
719 | case TRACE_REG_PERF_DEL: | |
ceec0b6f | 720 | return 0; |
2239291a SR |
721 | #endif |
722 | } | |
723 | return 0; | |
724 | } | |
725 | ||
726 | static int syscall_exit_register(struct ftrace_event_call *event, | |
ceec0b6f | 727 | enum trace_reg type, void *data) |
2239291a | 728 | { |
12ab74ee SR |
729 | struct ftrace_event_file *file = data; |
730 | ||
2239291a SR |
731 | switch (type) { |
732 | case TRACE_REG_REGISTER: | |
12ab74ee | 733 | return reg_event_syscall_exit(file, event); |
2239291a | 734 | case TRACE_REG_UNREGISTER: |
12ab74ee | 735 | unreg_event_syscall_exit(file, event); |
2239291a SR |
736 | return 0; |
737 | ||
738 | #ifdef CONFIG_PERF_EVENTS | |
739 | case TRACE_REG_PERF_REGISTER: | |
740 | return perf_sysexit_enable(event); | |
741 | case TRACE_REG_PERF_UNREGISTER: | |
742 | perf_sysexit_disable(event); | |
743 | return 0; | |
ceec0b6f JO |
744 | case TRACE_REG_PERF_OPEN: |
745 | case TRACE_REG_PERF_CLOSE: | |
489c75c3 JO |
746 | case TRACE_REG_PERF_ADD: |
747 | case TRACE_REG_PERF_DEL: | |
ceec0b6f | 748 | return 0; |
2239291a SR |
749 | #endif |
750 | } | |
751 | return 0; | |
752 | } |