Commit | Line | Data |
---|---|---|
47788c58 | 1 | #include <trace/syscall.h> |
1c569f02 | 2 | #include <trace/events/syscalls.h> |
5a0e3ad6 | 3 | #include <linux/slab.h> |
ee08c6ec | 4 | #include <linux/kernel.h> |
56d82e00 | 5 | #include <linux/module.h> /* for MODULE_NAME_LEN via KSYM_SYMBOL_LEN */ |
fb34a08c | 6 | #include <linux/ftrace.h> |
cdd6c482 | 7 | #include <linux/perf_event.h> |
ee08c6ec FW |
8 | #include <asm/syscall.h> |
9 | ||
10 | #include "trace_output.h" | |
11 | #include "trace.h" | |
12 | ||
5be71b61 | 13 | static DEFINE_MUTEX(syscall_trace_lock); |
fb34a08c JB |
14 | static int sys_refcount_enter; |
15 | static int sys_refcount_exit; | |
57421dbb JB |
16 | static DECLARE_BITMAP(enabled_enter_syscalls, NR_syscalls); |
17 | static DECLARE_BITMAP(enabled_exit_syscalls, NR_syscalls); | |
ee08c6ec | 18 | |
2239291a SR |
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 | ||
2e33af02 SR |
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 | ||
80decc70 | 35 | struct trace_event_functions enter_syscall_print_funcs = { |
7f85803a | 36 | .trace = print_syscall_enter, |
80decc70 SR |
37 | }; |
38 | ||
39 | struct trace_event_functions exit_syscall_print_funcs = { | |
7f85803a | 40 | .trace = print_syscall_exit, |
80decc70 SR |
41 | }; |
42 | ||
2239291a | 43 | struct ftrace_event_class event_class_syscall_enter = { |
7f85803a LJ |
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, | |
2239291a SR |
49 | }; |
50 | ||
51 | struct ftrace_event_class event_class_syscall_exit = { | |
7f85803a LJ |
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, | |
8f082018 SR |
57 | }; |
58 | ||
3d56e331 SR |
59 | extern struct syscall_metadata *__start_syscalls_metadata[]; |
60 | extern struct syscall_metadata *__stop_syscalls_metadata[]; | |
c44fc770 FW |
61 | |
62 | static struct syscall_metadata **syscalls_metadata; | |
63 | ||
b2d55496 IM |
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 | ||
3d56e331 SR |
77 | static __init struct syscall_metadata * |
78 | find_syscall_meta(unsigned long syscall) | |
c44fc770 | 79 | { |
3d56e331 SR |
80 | struct syscall_metadata **start; |
81 | struct syscall_metadata **stop; | |
c44fc770 FW |
82 | char str[KSYM_SYMBOL_LEN]; |
83 | ||
84 | ||
3d56e331 SR |
85 | start = __start_syscalls_metadata; |
86 | stop = __stop_syscalls_metadata; | |
c44fc770 FW |
87 | kallsyms_lookup(syscall, NULL, NULL, NULL, str); |
88 | ||
ae07f551 IM |
89 | if (arch_syscall_match_sym_name(str, "sys_ni_syscall")) |
90 | return NULL; | |
91 | ||
c44fc770 | 92 | for ( ; start < stop; start++) { |
b2d55496 | 93 | if ((*start)->name && arch_syscall_match_sym_name(str, (*start)->name)) |
3d56e331 | 94 | return *start; |
c44fc770 FW |
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 | ||
bed1ffca | 107 | enum print_line_t |
a9a57763 SR |
108 | print_syscall_enter(struct trace_iterator *iter, int flags, |
109 | struct trace_event *event) | |
bed1ffca FW |
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 | ||
64c12e04 | 117 | trace = (typeof(trace))ent; |
bed1ffca | 118 | syscall = trace->nr; |
bed1ffca | 119 | entry = syscall_nr_to_meta(syscall); |
64c12e04 | 120 | |
bed1ffca FW |
121 | if (!entry) |
122 | goto end; | |
123 | ||
32c0edae | 124 | if (entry->enter_event->event.type != ent->type) { |
64c12e04 JB |
125 | WARN_ON_ONCE(1); |
126 | goto end; | |
127 | } | |
128 | ||
bed1ffca FW |
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 */ | |
ba8b3a40 | 135 | if (trace_flags & TRACE_ITER_VERBOSE) { |
bed1ffca FW |
136 | ret = trace_seq_printf(s, "%s ", entry->types[i]); |
137 | if (!ret) | |
138 | return TRACE_TYPE_PARTIAL_LINE; | |
139 | } | |
140 | /* parameter values */ | |
4539f077 | 141 | ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i], |
bed1ffca | 142 | trace->args[i], |
4539f077 | 143 | i == entry->nb_args - 1 ? "" : ", "); |
bed1ffca FW |
144 | if (!ret) |
145 | return TRACE_TYPE_PARTIAL_LINE; | |
146 | } | |
147 | ||
4539f077 LZ |
148 | ret = trace_seq_putc(s, ')'); |
149 | if (!ret) | |
150 | return TRACE_TYPE_PARTIAL_LINE; | |
151 | ||
bed1ffca | 152 | end: |
4539f077 LZ |
153 | ret = trace_seq_putc(s, '\n'); |
154 | if (!ret) | |
155 | return TRACE_TYPE_PARTIAL_LINE; | |
156 | ||
bed1ffca FW |
157 | return TRACE_TYPE_HANDLED; |
158 | } | |
159 | ||
160 | enum print_line_t | |
a9a57763 SR |
161 | print_syscall_exit(struct trace_iterator *iter, int flags, |
162 | struct trace_event *event) | |
bed1ffca FW |
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 | ||
64c12e04 | 171 | trace = (typeof(trace))ent; |
bed1ffca | 172 | syscall = trace->nr; |
bed1ffca | 173 | entry = syscall_nr_to_meta(syscall); |
64c12e04 | 174 | |
bed1ffca FW |
175 | if (!entry) { |
176 | trace_seq_printf(s, "\n"); | |
177 | return TRACE_TYPE_HANDLED; | |
178 | } | |
179 | ||
32c0edae | 180 | if (entry->exit_event->event.type != ent->type) { |
64c12e04 JB |
181 | WARN_ON_ONCE(1); |
182 | return TRACE_TYPE_UNHANDLED; | |
183 | } | |
184 | ||
bed1ffca FW |
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 | ||
e6971969 LZ |
193 | extern char *__bad_type_size(void); |
194 | ||
195 | #define SYSCALL_FIELD(type, name) \ | |
196 | sizeof(type) != sizeof(trace.name) ? \ | |
197 | __bad_type_size() : \ | |
26a50744 TZ |
198 | #type, #name, offsetof(typeof(trace), name), \ |
199 | sizeof(trace.name), is_signed_type(type) | |
e6971969 | 200 | |
50307a45 LJ |
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 | ||
2e33af02 | 262 | static int syscall_enter_define_fields(struct ftrace_event_call *call) |
540b7b8d LZ |
263 | { |
264 | struct syscall_trace_enter trace; | |
31c16b13 | 265 | struct syscall_metadata *meta = call->data; |
540b7b8d | 266 | int ret; |
540b7b8d LZ |
267 | int i; |
268 | int offset = offsetof(typeof(trace), args); | |
269 | ||
0f1ef51d LJ |
270 | ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER); |
271 | if (ret) | |
272 | return ret; | |
273 | ||
540b7b8d | 274 | for (i = 0; i < meta->nb_args; i++) { |
aeaeae11 FW |
275 | ret = trace_define_field(call, meta->types[i], |
276 | meta->args[i], offset, | |
43b51ead LZ |
277 | sizeof(unsigned long), 0, |
278 | FILTER_OTHER); | |
540b7b8d LZ |
279 | offset += sizeof(unsigned long); |
280 | } | |
281 | ||
282 | return ret; | |
283 | } | |
284 | ||
2e33af02 | 285 | static int syscall_exit_define_fields(struct ftrace_event_call *call) |
540b7b8d LZ |
286 | { |
287 | struct syscall_trace_exit trace; | |
288 | int ret; | |
289 | ||
0f1ef51d LJ |
290 | ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER); |
291 | if (ret) | |
292 | return ret; | |
293 | ||
26a50744 | 294 | ret = trace_define_field(call, SYSCALL_FIELD(long, ret), |
43b51ead | 295 | FILTER_OTHER); |
540b7b8d LZ |
296 | |
297 | return ret; | |
298 | } | |
299 | ||
38516ab5 | 300 | void ftrace_syscall_enter(void *ignore, struct pt_regs *regs, long id) |
ee08c6ec | 301 | { |
bed1ffca FW |
302 | struct syscall_trace_enter *entry; |
303 | struct syscall_metadata *sys_data; | |
304 | struct ring_buffer_event *event; | |
e77405ad | 305 | struct ring_buffer *buffer; |
bed1ffca | 306 | int size; |
ee08c6ec FW |
307 | int syscall_nr; |
308 | ||
309 | syscall_nr = syscall_get_nr(current, regs); | |
cd0980fc HB |
310 | if (syscall_nr < 0) |
311 | return; | |
fb34a08c JB |
312 | if (!test_bit(syscall_nr, enabled_enter_syscalls)) |
313 | return; | |
ee08c6ec | 314 | |
bed1ffca FW |
315 | sys_data = syscall_nr_to_meta(syscall_nr); |
316 | if (!sys_data) | |
317 | return; | |
318 | ||
319 | size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args; | |
320 | ||
fcc19438 | 321 | event = trace_current_buffer_lock_reserve(&buffer, |
32c0edae | 322 | sys_data->enter_event->event.type, size, 0, 0); |
bed1ffca FW |
323 | if (!event) |
324 | return; | |
325 | ||
326 | entry = ring_buffer_event_data(event); | |
327 | entry->nr = syscall_nr; | |
328 | syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args); | |
329 | ||
e77405ad SR |
330 | if (!filter_current_check_discard(buffer, sys_data->enter_event, |
331 | entry, event)) | |
332 | trace_current_buffer_unlock_commit(buffer, event, 0, 0); | |
ee08c6ec FW |
333 | } |
334 | ||
38516ab5 | 335 | void ftrace_syscall_exit(void *ignore, struct pt_regs *regs, long ret) |
ee08c6ec | 336 | { |
bed1ffca FW |
337 | struct syscall_trace_exit *entry; |
338 | struct syscall_metadata *sys_data; | |
339 | struct ring_buffer_event *event; | |
e77405ad | 340 | struct ring_buffer *buffer; |
ee08c6ec FW |
341 | int syscall_nr; |
342 | ||
343 | syscall_nr = syscall_get_nr(current, regs); | |
cd0980fc HB |
344 | if (syscall_nr < 0) |
345 | return; | |
fb34a08c JB |
346 | if (!test_bit(syscall_nr, enabled_exit_syscalls)) |
347 | return; | |
ee08c6ec | 348 | |
bed1ffca FW |
349 | sys_data = syscall_nr_to_meta(syscall_nr); |
350 | if (!sys_data) | |
351 | return; | |
352 | ||
fcc19438 | 353 | event = trace_current_buffer_lock_reserve(&buffer, |
32c0edae | 354 | sys_data->exit_event->event.type, sizeof(*entry), 0, 0); |
bed1ffca FW |
355 | if (!event) |
356 | return; | |
357 | ||
358 | entry = ring_buffer_event_data(event); | |
359 | entry->nr = syscall_nr; | |
360 | entry->ret = syscall_get_return_value(current, regs); | |
361 | ||
e77405ad SR |
362 | if (!filter_current_check_discard(buffer, sys_data->exit_event, |
363 | entry, event)) | |
364 | trace_current_buffer_unlock_commit(buffer, event, 0, 0); | |
ee08c6ec FW |
365 | } |
366 | ||
bd1a5c84 | 367 | int reg_event_syscall_enter(struct ftrace_event_call *call) |
ee08c6ec | 368 | { |
fb34a08c JB |
369 | int ret = 0; |
370 | int num; | |
fb34a08c | 371 | |
c252f657 | 372 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 373 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
374 | return -ENOSYS; |
375 | mutex_lock(&syscall_trace_lock); | |
376 | if (!sys_refcount_enter) | |
38516ab5 | 377 | ret = register_trace_sys_enter(ftrace_syscall_enter, NULL); |
3b8e4273 | 378 | if (!ret) { |
fb34a08c JB |
379 | set_bit(num, enabled_enter_syscalls); |
380 | sys_refcount_enter++; | |
381 | } | |
382 | mutex_unlock(&syscall_trace_lock); | |
383 | return ret; | |
ee08c6ec FW |
384 | } |
385 | ||
bd1a5c84 | 386 | void unreg_event_syscall_enter(struct ftrace_event_call *call) |
ee08c6ec | 387 | { |
fb34a08c | 388 | int num; |
ee08c6ec | 389 | |
c252f657 | 390 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 391 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
392 | return; |
393 | mutex_lock(&syscall_trace_lock); | |
394 | sys_refcount_enter--; | |
395 | clear_bit(num, enabled_enter_syscalls); | |
396 | if (!sys_refcount_enter) | |
38516ab5 | 397 | unregister_trace_sys_enter(ftrace_syscall_enter, NULL); |
fb34a08c JB |
398 | mutex_unlock(&syscall_trace_lock); |
399 | } | |
ee08c6ec | 400 | |
bd1a5c84 | 401 | int reg_event_syscall_exit(struct ftrace_event_call *call) |
ee08c6ec | 402 | { |
fb34a08c JB |
403 | int ret = 0; |
404 | int num; | |
fb34a08c | 405 | |
c252f657 | 406 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 407 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
408 | return -ENOSYS; |
409 | mutex_lock(&syscall_trace_lock); | |
410 | if (!sys_refcount_exit) | |
38516ab5 | 411 | ret = register_trace_sys_exit(ftrace_syscall_exit, NULL); |
3b8e4273 | 412 | if (!ret) { |
fb34a08c JB |
413 | set_bit(num, enabled_exit_syscalls); |
414 | sys_refcount_exit++; | |
ee08c6ec | 415 | } |
fb34a08c JB |
416 | mutex_unlock(&syscall_trace_lock); |
417 | return ret; | |
418 | } | |
ee08c6ec | 419 | |
bd1a5c84 | 420 | void unreg_event_syscall_exit(struct ftrace_event_call *call) |
fb34a08c JB |
421 | { |
422 | int num; | |
ee08c6ec | 423 | |
c252f657 | 424 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
3773b389 | 425 | if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls)) |
fb34a08c JB |
426 | return; |
427 | mutex_lock(&syscall_trace_lock); | |
428 | sys_refcount_exit--; | |
429 | clear_bit(num, enabled_exit_syscalls); | |
430 | if (!sys_refcount_exit) | |
38516ab5 | 431 | unregister_trace_sys_exit(ftrace_syscall_exit, NULL); |
fb34a08c | 432 | mutex_unlock(&syscall_trace_lock); |
ee08c6ec | 433 | } |
fb34a08c | 434 | |
a1301da0 LJ |
435 | int init_syscall_trace(struct ftrace_event_call *call) |
436 | { | |
437 | int id; | |
ba976970 IM |
438 | int num; |
439 | ||
440 | num = ((struct syscall_metadata *)call->data)->syscall_nr; | |
441 | if (num < 0 || num >= NR_syscalls) { | |
442 | pr_debug("syscall %s metadata not mapped, disabling ftrace event\n", | |
443 | ((struct syscall_metadata *)call->data)->name); | |
444 | return -ENOSYS; | |
445 | } | |
a1301da0 | 446 | |
50307a45 LJ |
447 | if (set_syscall_print_fmt(call) < 0) |
448 | return -ENOMEM; | |
449 | ||
c7ef3a90 SR |
450 | id = trace_event_raw_init(call); |
451 | ||
452 | if (id < 0) { | |
50307a45 | 453 | free_syscall_print_fmt(call); |
c7ef3a90 | 454 | return id; |
50307a45 | 455 | } |
c7ef3a90 SR |
456 | |
457 | return id; | |
a1301da0 LJ |
458 | } |
459 | ||
c763ba06 | 460 | unsigned long __init __weak arch_syscall_addr(int nr) |
e7b8e675 MF |
461 | { |
462 | return (unsigned long)sys_call_table[nr]; | |
463 | } | |
464 | ||
c44fc770 FW |
465 | int __init init_ftrace_syscalls(void) |
466 | { | |
467 | struct syscall_metadata *meta; | |
468 | unsigned long addr; | |
469 | int i; | |
470 | ||
471 | syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) * | |
472 | NR_syscalls, GFP_KERNEL); | |
473 | if (!syscalls_metadata) { | |
474 | WARN_ON(1); | |
475 | return -ENOMEM; | |
476 | } | |
477 | ||
478 | for (i = 0; i < NR_syscalls; i++) { | |
479 | addr = arch_syscall_addr(i); | |
480 | meta = find_syscall_meta(addr); | |
c252f657 LJ |
481 | if (!meta) |
482 | continue; | |
483 | ||
484 | meta->syscall_nr = i; | |
c44fc770 FW |
485 | syscalls_metadata[i] = meta; |
486 | } | |
487 | ||
488 | return 0; | |
489 | } | |
490 | core_initcall(init_ftrace_syscalls); | |
491 | ||
07b139c8 | 492 | #ifdef CONFIG_PERF_EVENTS |
19007a67 | 493 | |
97d5a220 FW |
494 | static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls); |
495 | static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls); | |
496 | static int sys_perf_refcount_enter; | |
497 | static int sys_perf_refcount_exit; | |
f4b5ffcc | 498 | |
38516ab5 | 499 | static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id) |
f4b5ffcc JB |
500 | { |
501 | struct syscall_metadata *sys_data; | |
20ab4425 | 502 | struct syscall_trace_enter *rec; |
1c024eca | 503 | struct hlist_head *head; |
f4b5ffcc | 504 | int syscall_nr; |
4ed7c92d | 505 | int rctx; |
19007a67 | 506 | int size; |
f4b5ffcc JB |
507 | |
508 | syscall_nr = syscall_get_nr(current, regs); | |
97d5a220 | 509 | if (!test_bit(syscall_nr, enabled_perf_enter_syscalls)) |
f4b5ffcc JB |
510 | return; |
511 | ||
512 | sys_data = syscall_nr_to_meta(syscall_nr); | |
513 | if (!sys_data) | |
514 | return; | |
515 | ||
19007a67 FW |
516 | /* get the size after alignment with the u32 buffer size field */ |
517 | size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec); | |
518 | size = ALIGN(size + sizeof(u32), sizeof(u64)); | |
519 | size -= sizeof(u32); | |
520 | ||
97d5a220 FW |
521 | if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, |
522 | "perf buffer not large enough")) | |
20ab4425 FW |
523 | return; |
524 | ||
97d5a220 | 525 | rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size, |
ff5f149b | 526 | sys_data->enter_event->event.type, regs, &rctx); |
430ad5a6 XG |
527 | if (!rec) |
528 | return; | |
20ab4425 | 529 | |
20ab4425 FW |
530 | rec->nr = syscall_nr; |
531 | syscall_get_arguments(current, regs, 0, sys_data->nb_args, | |
532 | (unsigned long *)&rec->args); | |
1c024eca | 533 | |
3771f077 | 534 | head = this_cpu_ptr(sys_data->enter_event->perf_events); |
1c024eca | 535 | perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head); |
f4b5ffcc JB |
536 | } |
537 | ||
97d5a220 | 538 | int perf_sysenter_enable(struct ftrace_event_call *call) |
f4b5ffcc JB |
539 | { |
540 | int ret = 0; | |
541 | int num; | |
542 | ||
3bbe84e9 | 543 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
544 | |
545 | mutex_lock(&syscall_trace_lock); | |
97d5a220 | 546 | if (!sys_perf_refcount_enter) |
38516ab5 | 547 | ret = register_trace_sys_enter(perf_syscall_enter, NULL); |
f4b5ffcc JB |
548 | if (ret) { |
549 | pr_info("event trace: Could not activate" | |
550 | "syscall entry trace point"); | |
551 | } else { | |
97d5a220 FW |
552 | set_bit(num, enabled_perf_enter_syscalls); |
553 | sys_perf_refcount_enter++; | |
f4b5ffcc JB |
554 | } |
555 | mutex_unlock(&syscall_trace_lock); | |
556 | return ret; | |
557 | } | |
558 | ||
97d5a220 | 559 | void perf_sysenter_disable(struct ftrace_event_call *call) |
f4b5ffcc JB |
560 | { |
561 | int num; | |
562 | ||
3bbe84e9 | 563 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
564 | |
565 | mutex_lock(&syscall_trace_lock); | |
97d5a220 FW |
566 | sys_perf_refcount_enter--; |
567 | clear_bit(num, enabled_perf_enter_syscalls); | |
568 | if (!sys_perf_refcount_enter) | |
38516ab5 | 569 | unregister_trace_sys_enter(perf_syscall_enter, NULL); |
f4b5ffcc JB |
570 | mutex_unlock(&syscall_trace_lock); |
571 | } | |
572 | ||
38516ab5 | 573 | static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret) |
f4b5ffcc JB |
574 | { |
575 | struct syscall_metadata *sys_data; | |
20ab4425 | 576 | struct syscall_trace_exit *rec; |
1c024eca | 577 | struct hlist_head *head; |
f4b5ffcc | 578 | int syscall_nr; |
4ed7c92d | 579 | int rctx; |
20ab4425 | 580 | int size; |
f4b5ffcc JB |
581 | |
582 | syscall_nr = syscall_get_nr(current, regs); | |
97d5a220 | 583 | if (!test_bit(syscall_nr, enabled_perf_exit_syscalls)) |
f4b5ffcc JB |
584 | return; |
585 | ||
586 | sys_data = syscall_nr_to_meta(syscall_nr); | |
587 | if (!sys_data) | |
588 | return; | |
589 | ||
20ab4425 FW |
590 | /* We can probably do that at build time */ |
591 | size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64)); | |
592 | size -= sizeof(u32); | |
19007a67 | 593 | |
20ab4425 FW |
594 | /* |
595 | * Impossible, but be paranoid with the future | |
596 | * How to put this check outside runtime? | |
597 | */ | |
97d5a220 FW |
598 | if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, |
599 | "exit event has grown above perf buffer size")) | |
20ab4425 FW |
600 | return; |
601 | ||
97d5a220 | 602 | rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size, |
ff5f149b | 603 | sys_data->exit_event->event.type, regs, &rctx); |
430ad5a6 XG |
604 | if (!rec) |
605 | return; | |
20ab4425 | 606 | |
20ab4425 FW |
607 | rec->nr = syscall_nr; |
608 | rec->ret = syscall_get_return_value(current, regs); | |
609 | ||
3771f077 | 610 | head = this_cpu_ptr(sys_data->exit_event->perf_events); |
1c024eca | 611 | perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head); |
f4b5ffcc JB |
612 | } |
613 | ||
97d5a220 | 614 | int perf_sysexit_enable(struct ftrace_event_call *call) |
f4b5ffcc JB |
615 | { |
616 | int ret = 0; | |
617 | int num; | |
618 | ||
3bbe84e9 | 619 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
620 | |
621 | mutex_lock(&syscall_trace_lock); | |
97d5a220 | 622 | if (!sys_perf_refcount_exit) |
38516ab5 | 623 | ret = register_trace_sys_exit(perf_syscall_exit, NULL); |
f4b5ffcc JB |
624 | if (ret) { |
625 | pr_info("event trace: Could not activate" | |
6574658b | 626 | "syscall exit trace point"); |
f4b5ffcc | 627 | } else { |
97d5a220 FW |
628 | set_bit(num, enabled_perf_exit_syscalls); |
629 | sys_perf_refcount_exit++; | |
f4b5ffcc JB |
630 | } |
631 | mutex_unlock(&syscall_trace_lock); | |
632 | return ret; | |
633 | } | |
634 | ||
97d5a220 | 635 | void perf_sysexit_disable(struct ftrace_event_call *call) |
f4b5ffcc JB |
636 | { |
637 | int num; | |
638 | ||
3bbe84e9 | 639 | num = ((struct syscall_metadata *)call->data)->syscall_nr; |
f4b5ffcc JB |
640 | |
641 | mutex_lock(&syscall_trace_lock); | |
97d5a220 FW |
642 | sys_perf_refcount_exit--; |
643 | clear_bit(num, enabled_perf_exit_syscalls); | |
644 | if (!sys_perf_refcount_exit) | |
38516ab5 | 645 | unregister_trace_sys_exit(perf_syscall_exit, NULL); |
f4b5ffcc JB |
646 | mutex_unlock(&syscall_trace_lock); |
647 | } | |
648 | ||
07b139c8 | 649 | #endif /* CONFIG_PERF_EVENTS */ |
f4b5ffcc | 650 | |
2239291a SR |
651 | static int syscall_enter_register(struct ftrace_event_call *event, |
652 | enum trace_reg type) | |
653 | { | |
654 | switch (type) { | |
655 | case TRACE_REG_REGISTER: | |
656 | return reg_event_syscall_enter(event); | |
657 | case TRACE_REG_UNREGISTER: | |
658 | unreg_event_syscall_enter(event); | |
659 | return 0; | |
660 | ||
661 | #ifdef CONFIG_PERF_EVENTS | |
662 | case TRACE_REG_PERF_REGISTER: | |
663 | return perf_sysenter_enable(event); | |
664 | case TRACE_REG_PERF_UNREGISTER: | |
665 | perf_sysenter_disable(event); | |
666 | return 0; | |
667 | #endif | |
668 | } | |
669 | return 0; | |
670 | } | |
671 | ||
672 | static int syscall_exit_register(struct ftrace_event_call *event, | |
673 | enum trace_reg type) | |
674 | { | |
675 | switch (type) { | |
676 | case TRACE_REG_REGISTER: | |
677 | return reg_event_syscall_exit(event); | |
678 | case TRACE_REG_UNREGISTER: | |
679 | unreg_event_syscall_exit(event); | |
680 | return 0; | |
681 | ||
682 | #ifdef CONFIG_PERF_EVENTS | |
683 | case TRACE_REG_PERF_REGISTER: | |
684 | return perf_sysexit_enable(event); | |
685 | case TRACE_REG_PERF_UNREGISTER: | |
686 | perf_sysexit_disable(event); | |
687 | return 0; | |
688 | #endif | |
689 | } | |
690 | return 0; | |
691 | } |