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); |
fb34a08c JB |
15 | static int sys_refcount_enter; |
16 | static int sys_refcount_exit; | |
57421dbb JB |
17 | static DECLARE_BITMAP(enabled_enter_syscalls, NR_syscalls); |
18 | static DECLARE_BITMAP(enabled_exit_syscalls, NR_syscalls); | |
ee08c6ec | 19 | |
2239291a | 20 | static int syscall_enter_register(struct ftrace_event_call *event, |
ceec0b6f | 21 | enum trace_reg type, void *data); |
2239291a | 22 | static int syscall_exit_register(struct ftrace_event_call *event, |
ceec0b6f | 23 | enum trace_reg type, void *data); |
2239291a | 24 | |
2e33af02 SR |
25 | static struct list_head * |
26 | syscall_get_enter_fields(struct ftrace_event_call *call) | |
27 | { | |
28 | struct syscall_metadata *entry = call->data; | |
29 | ||
30 | return &entry->enter_fields; | |
31 | } | |
32 | ||
3d56e331 SR |
33 | extern struct syscall_metadata *__start_syscalls_metadata[]; |
34 | extern struct syscall_metadata *__stop_syscalls_metadata[]; | |
c44fc770 FW |
35 | |
36 | static struct syscall_metadata **syscalls_metadata; | |
37 | ||
b2d55496 IM |
38 | #ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME |
39 | static inline bool arch_syscall_match_sym_name(const char *sym, const char *name) | |
40 | { | |
41 | /* | |
42 | * Only compare after the "sys" prefix. Archs that use | |
43 | * syscall wrappers may have syscalls symbols aliases prefixed | |
44 | * with "SyS" instead of "sys", leading to an unwanted | |
45 | * mismatch. | |
46 | */ | |
47 | return !strcmp(sym + 3, name + 3); | |
48 | } | |
49 | #endif | |
50 | ||
f431b634 SR |
51 | #ifdef ARCH_TRACE_IGNORE_COMPAT_SYSCALLS |
52 | /* | |
53 | * Some architectures that allow for 32bit applications | |
54 | * to run on a 64bit kernel, do not map the syscalls for | |
55 | * the 32bit tasks the same as they do for 64bit tasks. | |
56 | * | |
57 | * *cough*x86*cough* | |
58 | * | |
59 | * In such a case, instead of reporting the wrong syscalls, | |
60 | * simply ignore them. | |
61 | * | |
62 | * For an arch to ignore the compat syscalls it needs to | |
63 | * define ARCH_TRACE_IGNORE_COMPAT_SYSCALLS as well as | |
64 | * define the function arch_trace_is_compat_syscall() to let | |
65 | * the tracing system know that it should ignore it. | |
66 | */ | |
67 | static int | |
68 | trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs) | |
69 | { | |
70 | if (unlikely(arch_trace_is_compat_syscall(regs))) | |
71 | return -1; | |
72 | ||
73 | return syscall_get_nr(task, regs); | |
74 | } | |
75 | #else | |
76 | static inline int | |
77 | trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs) | |
78 | { | |
79 | return syscall_get_nr(task, regs); | |
80 | } | |
81 | #endif /* ARCH_TRACE_IGNORE_COMPAT_SYSCALLS */ | |
82 | ||
3d56e331 SR |
83 | static __init struct syscall_metadata * |
84 | find_syscall_meta(unsigned long syscall) | |
c44fc770 | 85 | { |
3d56e331 SR |
86 | struct syscall_metadata **start; |
87 | struct syscall_metadata **stop; | |
c44fc770 FW |
88 | char str[KSYM_SYMBOL_LEN]; |
89 | ||
90 | ||
3d56e331 SR |
91 | start = __start_syscalls_metadata; |
92 | stop = __stop_syscalls_metadata; | |
c44fc770 FW |
93 | kallsyms_lookup(syscall, NULL, NULL, NULL, str); |
94 | ||
ae07f551 IM |
95 | if (arch_syscall_match_sym_name(str, "sys_ni_syscall")) |
96 | return NULL; | |
97 | ||
c44fc770 | 98 | for ( ; start < stop; start++) { |
b2d55496 | 99 | if ((*start)->name && arch_syscall_match_sym_name(str, (*start)->name)) |
3d56e331 | 100 | return *start; |
c44fc770 FW |
101 | } |
102 | return NULL; | |
103 | } | |
104 | ||
105 | static struct syscall_metadata *syscall_nr_to_meta(int nr) | |
106 | { | |
107 | if (!syscalls_metadata || nr >= NR_syscalls || nr < 0) | |
108 | return NULL; | |
109 | ||
110 | return syscalls_metadata[nr]; | |
111 | } | |
112 | ||
6aea49cb | 113 | static enum print_line_t |
a9a57763 SR |
114 | print_syscall_enter(struct trace_iterator *iter, int flags, |
115 | struct trace_event *event) | |
bed1ffca FW |
116 | { |
117 | struct trace_seq *s = &iter->seq; | |
118 | struct trace_entry *ent = iter->ent; | |
119 | struct syscall_trace_enter *trace; | |
120 | struct syscall_metadata *entry; | |
121 | int i, ret, syscall; | |
122 | ||
64c12e04 | 123 | trace = (typeof(trace))ent; |
bed1ffca | 124 | syscall = trace->nr; |
bed1ffca | 125 | entry = syscall_nr_to_meta(syscall); |
64c12e04 | 126 | |
bed1ffca FW |
127 | if (!entry) |
128 | goto end; | |
129 | ||
32c0edae | 130 | if (entry->enter_event->event.type != ent->type) { |
64c12e04 JB |
131 | WARN_ON_ONCE(1); |
132 | goto end; | |
133 | } | |
134 | ||
bed1ffca FW |
135 | ret = trace_seq_printf(s, "%s(", entry->name); |
136 | if (!ret) | |
137 | return TRACE_TYPE_PARTIAL_LINE; | |
138 | ||
139 | for (i = 0; i < entry->nb_args; i++) { | |
140 | /* parameter types */ | |
ba8b3a40 | 141 | if (trace_flags & TRACE_ITER_VERBOSE) { |
bed1ffca FW |
142 | ret = trace_seq_printf(s, "%s ", entry->types[i]); |
143 | if (!ret) | |
144 | return TRACE_TYPE_PARTIAL_LINE; | |
145 | } | |
146 | /* parameter values */ | |
4539f077 | 147 | ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i], |
bed1ffca | 148 | trace->args[i], |
4539f077 | 149 | i == entry->nb_args - 1 ? "" : ", "); |
bed1ffca FW |
150 | if (!ret) |
151 | return TRACE_TYPE_PARTIAL_LINE; | |
152 | } | |
153 | ||
4539f077 LZ |
154 | ret = trace_seq_putc(s, ')'); |
155 | if (!ret) | |
156 | return TRACE_TYPE_PARTIAL_LINE; | |
157 | ||
bed1ffca | 158 | end: |
4539f077 LZ |
159 | ret = trace_seq_putc(s, '\n'); |
160 | if (!ret) | |
161 | return TRACE_TYPE_PARTIAL_LINE; | |
162 | ||
bed1ffca FW |
163 | return TRACE_TYPE_HANDLED; |
164 | } | |
165 | ||
6aea49cb | 166 | static enum print_line_t |
a9a57763 SR |
167 | print_syscall_exit(struct trace_iterator *iter, int flags, |
168 | struct trace_event *event) | |
bed1ffca FW |
169 | { |
170 | struct trace_seq *s = &iter->seq; | |
171 | struct trace_entry *ent = iter->ent; | |
172 | struct syscall_trace_exit *trace; | |
173 | int syscall; | |
174 | struct syscall_metadata *entry; | |
175 | int ret; | |
176 | ||
64c12e04 | 177 | trace = (typeof(trace))ent; |
bed1ffca | 178 | syscall = trace->nr; |
bed1ffca | 179 | entry = syscall_nr_to_meta(syscall); |
64c12e04 | 180 | |
bed1ffca FW |
181 | if (!entry) { |
182 | trace_seq_printf(s, "\n"); | |
183 | return TRACE_TYPE_HANDLED; | |
184 | } | |
185 | ||
32c0edae | 186 | if (entry->exit_event->event.type != ent->type) { |
64c12e04 JB |
187 | WARN_ON_ONCE(1); |
188 | return TRACE_TYPE_UNHANDLED; | |
189 | } | |
190 | ||
bed1ffca FW |
191 | ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name, |
192 | trace->ret); | |
193 | if (!ret) | |
194 | return TRACE_TYPE_PARTIAL_LINE; | |
195 | ||
196 | return TRACE_TYPE_HANDLED; | |
197 | } | |
198 | ||
e6971969 LZ |
199 | extern char *__bad_type_size(void); |
200 | ||
201 | #define SYSCALL_FIELD(type, name) \ | |
202 | sizeof(type) != sizeof(trace.name) ? \ | |
203 | __bad_type_size() : \ | |
26a50744 TZ |
204 | #type, #name, offsetof(typeof(trace), name), \ |
205 | sizeof(trace.name), is_signed_type(type) | |
e6971969 | 206 | |
50307a45 LJ |
207 | static |
208 | int __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len) | |
209 | { | |
210 | int i; | |
211 | int pos = 0; | |
212 | ||
213 | /* When len=0, we just calculate the needed length */ | |
214 | #define LEN_OR_ZERO (len ? len - pos : 0) | |
215 | ||
216 | pos += snprintf(buf + pos, LEN_OR_ZERO, "\""); | |
217 | for (i = 0; i < entry->nb_args; i++) { | |
218 | pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s", | |
219 | entry->args[i], sizeof(unsigned long), | |
220 | i == entry->nb_args - 1 ? "" : ", "); | |
221 | } | |
222 | pos += snprintf(buf + pos, LEN_OR_ZERO, "\""); | |
223 | ||
224 | for (i = 0; i < entry->nb_args; i++) { | |
225 | pos += snprintf(buf + pos, LEN_OR_ZERO, | |
226 | ", ((unsigned long)(REC->%s))", entry->args[i]); | |
227 | } | |
228 | ||
229 | #undef LEN_OR_ZERO | |
230 | ||
231 | /* return the length of print_fmt */ | |
232 | return pos; | |
233 | } | |
234 | ||
235 | static int set_syscall_print_fmt(struct ftrace_event_call *call) | |
236 | { | |
237 | char *print_fmt; | |
238 | int len; | |
239 | struct syscall_metadata *entry = call->data; | |
240 | ||
241 | if (entry->enter_event != call) { | |
242 | call->print_fmt = "\"0x%lx\", REC->ret"; | |
243 | return 0; | |
244 | } | |
245 | ||
246 | /* First: called with 0 length to calculate the needed length */ | |
247 | len = __set_enter_print_fmt(entry, NULL, 0); | |
248 | ||
249 | print_fmt = kmalloc(len + 1, GFP_KERNEL); | |
250 | if (!print_fmt) | |
251 | return -ENOMEM; | |
252 | ||
253 | /* Second: actually write the @print_fmt */ | |
254 | __set_enter_print_fmt(entry, print_fmt, len + 1); | |
255 | call->print_fmt = print_fmt; | |
256 | ||
257 | return 0; | |
258 | } | |
259 | ||
260 | static void free_syscall_print_fmt(struct ftrace_event_call *call) | |
261 | { | |
262 | struct syscall_metadata *entry = call->data; | |
263 | ||
264 | if (entry->enter_event == call) | |
265 | kfree(call->print_fmt); | |
266 | } | |
267 | ||
2e33af02 | 268 | static int syscall_enter_define_fields(struct ftrace_event_call *call) |
540b7b8d LZ |
269 | { |
270 | struct syscall_trace_enter trace; | |
31c16b13 | 271 | struct syscall_metadata *meta = call->data; |
540b7b8d | 272 | int ret; |
540b7b8d LZ |
273 | int i; |
274 | int offset = offsetof(typeof(trace), args); | |
275 | ||
0f1ef51d LJ |
276 | ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER); |
277 | if (ret) | |
278 | return ret; | |
279 | ||
540b7b8d | 280 | for (i = 0; i < meta->nb_args; i++) { |
aeaeae11 FW |
281 | ret = trace_define_field(call, meta->types[i], |
282 | meta->args[i], offset, | |
43b51ead LZ |
283 | sizeof(unsigned long), 0, |
284 | FILTER_OTHER); | |
540b7b8d LZ |
285 | offset += sizeof(unsigned long); |
286 | } | |
287 | ||
288 | return ret; | |
289 | } | |
290 | ||
2e33af02 | 291 | static int syscall_exit_define_fields(struct ftrace_event_call *call) |
540b7b8d LZ |
292 | { |
293 | struct syscall_trace_exit trace; | |
294 | int ret; | |
295 | ||
0f1ef51d LJ |
296 | ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER); |
297 | if (ret) | |
298 | return ret; | |
299 | ||
26a50744 | 300 | ret = trace_define_field(call, SYSCALL_FIELD(long, ret), |
43b51ead | 301 | FILTER_OTHER); |
540b7b8d LZ |
302 | |
303 | return ret; | |
304 | } | |
305 | ||
6aea49cb | 306 | static void ftrace_syscall_enter(void *ignore, struct pt_regs *regs, long id) |
ee08c6ec | 307 | { |
bed1ffca FW |
308 | struct syscall_trace_enter *entry; |
309 | struct syscall_metadata *sys_data; | |
310 | struct ring_buffer_event *event; | |
e77405ad | 311 | struct ring_buffer *buffer; |
ee08c6ec | 312 | int syscall_nr; |
f431b634 | 313 | int size; |
ee08c6ec | 314 | |
f431b634 | 315 | syscall_nr = trace_get_syscall_nr(current, regs); |
cd0980fc HB |
316 | if (syscall_nr < 0) |
317 | return; | |
fb34a08c JB |
318 | if (!test_bit(syscall_nr, enabled_enter_syscalls)) |
319 | return; | |
ee08c6ec | 320 | |
bed1ffca FW |
321 | sys_data = syscall_nr_to_meta(syscall_nr); |
322 | if (!sys_data) | |
323 | return; | |
324 | ||
325 | size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args; | |
326 | ||
fcc19438 | 327 | event = trace_current_buffer_lock_reserve(&buffer, |
32c0edae | 328 | sys_data->enter_event->event.type, size, 0, 0); |
bed1ffca FW |
329 | if (!event) |
330 | return; | |
331 | ||
332 | entry = ring_buffer_event_data(event); | |
333 | entry->nr = syscall_nr; | |
334 | syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args); | |
335 | ||
e77405ad SR |
336 | if (!filter_current_check_discard(buffer, sys_data->enter_event, |
337 | entry, event)) | |
338 | trace_current_buffer_unlock_commit(buffer, event, 0, 0); | |
ee08c6ec FW |
339 | } |
340 | ||
6aea49cb | 341 | static void ftrace_syscall_exit(void *ignore, struct pt_regs *regs, long ret) |
ee08c6ec | 342 | { |
bed1ffca FW |
343 | struct syscall_trace_exit *entry; |
344 | struct syscall_metadata *sys_data; | |
345 | struct ring_buffer_event *event; | |
e77405ad | 346 | struct ring_buffer *buffer; |
ee08c6ec FW |
347 | int syscall_nr; |
348 | ||
f431b634 | 349 | syscall_nr = trace_get_syscall_nr(current, regs); |
cd0980fc HB |
350 | if (syscall_nr < 0) |
351 | return; | |
fb34a08c JB |
352 | if (!test_bit(syscall_nr, enabled_exit_syscalls)) |
353 | return; | |
ee08c6ec | 354 | |
bed1ffca FW |
355 | sys_data = syscall_nr_to_meta(syscall_nr); |
356 | if (!sys_data) | |
357 | return; | |
358 | ||
fcc19438 | 359 | event = trace_current_buffer_lock_reserve(&buffer, |
32c0edae | 360 | sys_data->exit_event->event.type, sizeof(*entry), 0, 0); |
bed1ffca FW |
361 | if (!event) |
362 | return; | |
363 | ||
364 | entry = ring_buffer_event_data(event); | |
365 | entry->nr = syscall_nr; | |
366 | entry->ret = syscall_get_return_value(current, regs); | |
367 | ||
e77405ad SR |
368 | if (!filter_current_check_discard(buffer, sys_data->exit_event, |
369 | entry, event)) | |
370 | trace_current_buffer_unlock_commit(buffer, event, 0, 0); | |
ee08c6ec FW |
371 | } |
372 | ||
6aea49cb | 373 | static int reg_event_syscall_enter(struct ftrace_event_call *call) |
ee08c6ec | 374 | { |
fb34a08c JB |
375 | int ret = 0; |
376 | int num; | |
fb34a08c | 377 | |
c252f657 | 378 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 379 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
380 | return -ENOSYS; |
381 | mutex_lock(&syscall_trace_lock); | |
382 | if (!sys_refcount_enter) | |
38516ab5 | 383 | ret = register_trace_sys_enter(ftrace_syscall_enter, NULL); |
3b8e4273 | 384 | if (!ret) { |
fb34a08c JB |
385 | set_bit(num, enabled_enter_syscalls); |
386 | sys_refcount_enter++; | |
387 | } | |
388 | mutex_unlock(&syscall_trace_lock); | |
389 | return ret; | |
ee08c6ec FW |
390 | } |
391 | ||
6aea49cb | 392 | static void unreg_event_syscall_enter(struct ftrace_event_call *call) |
ee08c6ec | 393 | { |
fb34a08c | 394 | int num; |
ee08c6ec | 395 | |
c252f657 | 396 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 397 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
398 | return; |
399 | mutex_lock(&syscall_trace_lock); | |
400 | sys_refcount_enter--; | |
401 | clear_bit(num, enabled_enter_syscalls); | |
402 | if (!sys_refcount_enter) | |
38516ab5 | 403 | unregister_trace_sys_enter(ftrace_syscall_enter, NULL); |
fb34a08c JB |
404 | mutex_unlock(&syscall_trace_lock); |
405 | } | |
ee08c6ec | 406 | |
6aea49cb | 407 | static int reg_event_syscall_exit(struct ftrace_event_call *call) |
ee08c6ec | 408 | { |
fb34a08c JB |
409 | int ret = 0; |
410 | int num; | |
fb34a08c | 411 | |
c252f657 | 412 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 413 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
414 | return -ENOSYS; |
415 | mutex_lock(&syscall_trace_lock); | |
416 | if (!sys_refcount_exit) | |
38516ab5 | 417 | ret = register_trace_sys_exit(ftrace_syscall_exit, NULL); |
3b8e4273 | 418 | if (!ret) { |
fb34a08c JB |
419 | set_bit(num, enabled_exit_syscalls); |
420 | sys_refcount_exit++; | |
ee08c6ec | 421 | } |
fb34a08c JB |
422 | mutex_unlock(&syscall_trace_lock); |
423 | return ret; | |
424 | } | |
ee08c6ec | 425 | |
6aea49cb | 426 | static void unreg_event_syscall_exit(struct ftrace_event_call *call) |
fb34a08c JB |
427 | { |
428 | int num; | |
ee08c6ec | 429 | |
c252f657 | 430 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 431 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
432 | return; |
433 | mutex_lock(&syscall_trace_lock); | |
434 | sys_refcount_exit--; | |
435 | clear_bit(num, enabled_exit_syscalls); | |
436 | if (!sys_refcount_exit) | |
38516ab5 | 437 | unregister_trace_sys_exit(ftrace_syscall_exit, NULL); |
fb34a08c | 438 | mutex_unlock(&syscall_trace_lock); |
ee08c6ec | 439 | } |
fb34a08c | 440 | |
6f86ab9f | 441 | static int init_syscall_trace(struct ftrace_event_call *call) |
a1301da0 LJ |
442 | { |
443 | int id; | |
ba976970 IM |
444 | int num; |
445 | ||
446 | num = ((struct syscall_metadata *)call->data)->syscall_nr; | |
447 | if (num < 0 || num >= NR_syscalls) { | |
448 | pr_debug("syscall %s metadata not mapped, disabling ftrace event\n", | |
449 | ((struct syscall_metadata *)call->data)->name); | |
450 | return -ENOSYS; | |
451 | } | |
a1301da0 | 452 | |
50307a45 LJ |
453 | if (set_syscall_print_fmt(call) < 0) |
454 | return -ENOMEM; | |
455 | ||
c7ef3a90 SR |
456 | id = trace_event_raw_init(call); |
457 | ||
458 | if (id < 0) { | |
50307a45 | 459 | free_syscall_print_fmt(call); |
c7ef3a90 | 460 | return id; |
50307a45 | 461 | } |
c7ef3a90 SR |
462 | |
463 | return id; | |
a1301da0 LJ |
464 | } |
465 | ||
6f86ab9f VN |
466 | struct trace_event_functions enter_syscall_print_funcs = { |
467 | .trace = print_syscall_enter, | |
468 | }; | |
469 | ||
470 | struct trace_event_functions exit_syscall_print_funcs = { | |
471 | .trace = print_syscall_exit, | |
472 | }; | |
473 | ||
474 | struct ftrace_event_class event_class_syscall_enter = { | |
475 | .system = "syscalls", | |
476 | .reg = syscall_enter_register, | |
477 | .define_fields = syscall_enter_define_fields, | |
478 | .get_fields = syscall_get_enter_fields, | |
479 | .raw_init = init_syscall_trace, | |
480 | }; | |
481 | ||
482 | struct ftrace_event_class event_class_syscall_exit = { | |
483 | .system = "syscalls", | |
484 | .reg = syscall_exit_register, | |
485 | .define_fields = syscall_exit_define_fields, | |
486 | .fields = LIST_HEAD_INIT(event_class_syscall_exit.fields), | |
487 | .raw_init = init_syscall_trace, | |
488 | }; | |
489 | ||
c763ba06 | 490 | unsigned long __init __weak arch_syscall_addr(int nr) |
e7b8e675 MF |
491 | { |
492 | return (unsigned long)sys_call_table[nr]; | |
493 | } | |
494 | ||
6aea49cb | 495 | static int __init init_ftrace_syscalls(void) |
c44fc770 FW |
496 | { |
497 | struct syscall_metadata *meta; | |
498 | unsigned long addr; | |
499 | int i; | |
500 | ||
47b0edcb TM |
501 | syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata), |
502 | GFP_KERNEL); | |
c44fc770 FW |
503 | if (!syscalls_metadata) { |
504 | WARN_ON(1); | |
505 | return -ENOMEM; | |
506 | } | |
507 | ||
508 | for (i = 0; i < NR_syscalls; i++) { | |
509 | addr = arch_syscall_addr(i); | |
510 | meta = find_syscall_meta(addr); | |
c252f657 LJ |
511 | if (!meta) |
512 | continue; | |
513 | ||
514 | meta->syscall_nr = i; | |
c44fc770 FW |
515 | syscalls_metadata[i] = meta; |
516 | } | |
517 | ||
518 | return 0; | |
519 | } | |
8781915a | 520 | early_initcall(init_ftrace_syscalls); |
c44fc770 | 521 | |
07b139c8 | 522 | #ifdef CONFIG_PERF_EVENTS |
19007a67 | 523 | |
97d5a220 FW |
524 | static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls); |
525 | static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls); | |
526 | static int sys_perf_refcount_enter; | |
527 | static int sys_perf_refcount_exit; | |
f4b5ffcc | 528 | |
38516ab5 | 529 | static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id) |
f4b5ffcc JB |
530 | { |
531 | struct syscall_metadata *sys_data; | |
20ab4425 | 532 | struct syscall_trace_enter *rec; |
1c024eca | 533 | struct hlist_head *head; |
f4b5ffcc | 534 | int syscall_nr; |
4ed7c92d | 535 | int rctx; |
19007a67 | 536 | int size; |
f4b5ffcc | 537 | |
f431b634 | 538 | syscall_nr = trace_get_syscall_nr(current, regs); |
60916a93 WD |
539 | if (syscall_nr < 0) |
540 | return; | |
97d5a220 | 541 | if (!test_bit(syscall_nr, enabled_perf_enter_syscalls)) |
f4b5ffcc JB |
542 | return; |
543 | ||
544 | sys_data = syscall_nr_to_meta(syscall_nr); | |
545 | if (!sys_data) | |
546 | return; | |
547 | ||
19007a67 FW |
548 | /* get the size after alignment with the u32 buffer size field */ |
549 | size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec); | |
550 | size = ALIGN(size + sizeof(u32), sizeof(u64)); | |
551 | size -= sizeof(u32); | |
552 | ||
97d5a220 FW |
553 | if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, |
554 | "perf buffer not large enough")) | |
20ab4425 FW |
555 | return; |
556 | ||
97d5a220 | 557 | rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size, |
ff5f149b | 558 | sys_data->enter_event->event.type, regs, &rctx); |
430ad5a6 XG |
559 | if (!rec) |
560 | return; | |
20ab4425 | 561 | |
20ab4425 FW |
562 | rec->nr = syscall_nr; |
563 | syscall_get_arguments(current, regs, 0, sys_data->nb_args, | |
564 | (unsigned long *)&rec->args); | |
1c024eca | 565 | |
3771f077 | 566 | head = this_cpu_ptr(sys_data->enter_event->perf_events); |
e6dab5ff | 567 | perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL); |
f4b5ffcc JB |
568 | } |
569 | ||
6f86ab9f | 570 | static int perf_sysenter_enable(struct ftrace_event_call *call) |
f4b5ffcc JB |
571 | { |
572 | int ret = 0; | |
573 | int num; | |
574 | ||
3bbe84e9 | 575 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
576 | |
577 | mutex_lock(&syscall_trace_lock); | |
97d5a220 | 578 | if (!sys_perf_refcount_enter) |
38516ab5 | 579 | ret = register_trace_sys_enter(perf_syscall_enter, NULL); |
f4b5ffcc JB |
580 | if (ret) { |
581 | pr_info("event trace: Could not activate" | |
582 | "syscall entry trace point"); | |
583 | } else { | |
97d5a220 FW |
584 | set_bit(num, enabled_perf_enter_syscalls); |
585 | sys_perf_refcount_enter++; | |
f4b5ffcc JB |
586 | } |
587 | mutex_unlock(&syscall_trace_lock); | |
588 | return ret; | |
589 | } | |
590 | ||
6f86ab9f | 591 | static void perf_sysenter_disable(struct ftrace_event_call *call) |
f4b5ffcc JB |
592 | { |
593 | int num; | |
594 | ||
3bbe84e9 | 595 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
596 | |
597 | mutex_lock(&syscall_trace_lock); | |
97d5a220 FW |
598 | sys_perf_refcount_enter--; |
599 | clear_bit(num, enabled_perf_enter_syscalls); | |
600 | if (!sys_perf_refcount_enter) | |
38516ab5 | 601 | unregister_trace_sys_enter(perf_syscall_enter, NULL); |
f4b5ffcc JB |
602 | mutex_unlock(&syscall_trace_lock); |
603 | } | |
604 | ||
38516ab5 | 605 | static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret) |
f4b5ffcc JB |
606 | { |
607 | struct syscall_metadata *sys_data; | |
20ab4425 | 608 | struct syscall_trace_exit *rec; |
1c024eca | 609 | struct hlist_head *head; |
f4b5ffcc | 610 | int syscall_nr; |
4ed7c92d | 611 | int rctx; |
20ab4425 | 612 | int size; |
f4b5ffcc | 613 | |
f431b634 | 614 | syscall_nr = trace_get_syscall_nr(current, regs); |
60916a93 WD |
615 | if (syscall_nr < 0) |
616 | return; | |
97d5a220 | 617 | if (!test_bit(syscall_nr, enabled_perf_exit_syscalls)) |
f4b5ffcc JB |
618 | return; |
619 | ||
620 | sys_data = syscall_nr_to_meta(syscall_nr); | |
621 | if (!sys_data) | |
622 | return; | |
623 | ||
20ab4425 FW |
624 | /* We can probably do that at build time */ |
625 | size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64)); | |
626 | size -= sizeof(u32); | |
19007a67 | 627 | |
20ab4425 FW |
628 | /* |
629 | * Impossible, but be paranoid with the future | |
630 | * How to put this check outside runtime? | |
631 | */ | |
97d5a220 FW |
632 | if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, |
633 | "exit event has grown above perf buffer size")) | |
20ab4425 FW |
634 | return; |
635 | ||
97d5a220 | 636 | rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size, |
ff5f149b | 637 | sys_data->exit_event->event.type, regs, &rctx); |
430ad5a6 XG |
638 | if (!rec) |
639 | return; | |
20ab4425 | 640 | |
20ab4425 FW |
641 | rec->nr = syscall_nr; |
642 | rec->ret = syscall_get_return_value(current, regs); | |
643 | ||
3771f077 | 644 | head = this_cpu_ptr(sys_data->exit_event->perf_events); |
e6dab5ff | 645 | perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL); |
f4b5ffcc JB |
646 | } |
647 | ||
6f86ab9f | 648 | static int perf_sysexit_enable(struct ftrace_event_call *call) |
f4b5ffcc JB |
649 | { |
650 | int ret = 0; | |
651 | int num; | |
652 | ||
3bbe84e9 | 653 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
654 | |
655 | mutex_lock(&syscall_trace_lock); | |
97d5a220 | 656 | if (!sys_perf_refcount_exit) |
38516ab5 | 657 | ret = register_trace_sys_exit(perf_syscall_exit, NULL); |
f4b5ffcc JB |
658 | if (ret) { |
659 | pr_info("event trace: Could not activate" | |
6574658b | 660 | "syscall exit trace point"); |
f4b5ffcc | 661 | } else { |
97d5a220 FW |
662 | set_bit(num, enabled_perf_exit_syscalls); |
663 | sys_perf_refcount_exit++; | |
f4b5ffcc JB |
664 | } |
665 | mutex_unlock(&syscall_trace_lock); | |
666 | return ret; | |
667 | } | |
668 | ||
6f86ab9f | 669 | static void perf_sysexit_disable(struct ftrace_event_call *call) |
f4b5ffcc JB |
670 | { |
671 | int num; | |
672 | ||
3bbe84e9 | 673 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
674 | |
675 | mutex_lock(&syscall_trace_lock); | |
97d5a220 FW |
676 | sys_perf_refcount_exit--; |
677 | clear_bit(num, enabled_perf_exit_syscalls); | |
678 | if (!sys_perf_refcount_exit) | |
38516ab5 | 679 | unregister_trace_sys_exit(perf_syscall_exit, NULL); |
f4b5ffcc JB |
680 | mutex_unlock(&syscall_trace_lock); |
681 | } | |
682 | ||
07b139c8 | 683 | #endif /* CONFIG_PERF_EVENTS */ |
f4b5ffcc | 684 | |
2239291a | 685 | static int syscall_enter_register(struct ftrace_event_call *event, |
ceec0b6f | 686 | enum trace_reg type, void *data) |
2239291a SR |
687 | { |
688 | switch (type) { | |
689 | case TRACE_REG_REGISTER: | |
690 | return reg_event_syscall_enter(event); | |
691 | case TRACE_REG_UNREGISTER: | |
692 | unreg_event_syscall_enter(event); | |
693 | return 0; | |
694 | ||
695 | #ifdef CONFIG_PERF_EVENTS | |
696 | case TRACE_REG_PERF_REGISTER: | |
697 | return perf_sysenter_enable(event); | |
698 | case TRACE_REG_PERF_UNREGISTER: | |
699 | perf_sysenter_disable(event); | |
700 | return 0; | |
ceec0b6f JO |
701 | case TRACE_REG_PERF_OPEN: |
702 | case TRACE_REG_PERF_CLOSE: | |
489c75c3 JO |
703 | case TRACE_REG_PERF_ADD: |
704 | case TRACE_REG_PERF_DEL: | |
ceec0b6f | 705 | return 0; |
2239291a SR |
706 | #endif |
707 | } | |
708 | return 0; | |
709 | } | |
710 | ||
711 | static int syscall_exit_register(struct ftrace_event_call *event, | |
ceec0b6f | 712 | enum trace_reg type, void *data) |
2239291a SR |
713 | { |
714 | switch (type) { | |
715 | case TRACE_REG_REGISTER: | |
716 | return reg_event_syscall_exit(event); | |
717 | case TRACE_REG_UNREGISTER: | |
718 | unreg_event_syscall_exit(event); | |
719 | return 0; | |
720 | ||
721 | #ifdef CONFIG_PERF_EVENTS | |
722 | case TRACE_REG_PERF_REGISTER: | |
723 | return perf_sysexit_enable(event); | |
724 | case TRACE_REG_PERF_UNREGISTER: | |
725 | perf_sysexit_disable(event); | |
726 | return 0; | |
ceec0b6f JO |
727 | case TRACE_REG_PERF_OPEN: |
728 | case TRACE_REG_PERF_CLOSE: | |
489c75c3 JO |
729 | case TRACE_REG_PERF_ADD: |
730 | case TRACE_REG_PERF_DEL: | |
ceec0b6f | 731 | return 0; |
2239291a SR |
732 | #endif |
733 | } | |
734 | return 0; | |
735 | } |