xref: /linux/kernel/trace/trace_fprobe.c (revision 4a0c9b3391999818e2c5b93719699b255be1f682)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Fprobe-based tracing events
4  * Copyright (C) 2022 Google LLC.
5  */
6 #define pr_fmt(fmt)	"trace_fprobe: " fmt
7 
8 #include <linux/fprobe.h>
9 #include <linux/list.h>
10 #include <linux/module.h>
11 #include <linux/mutex.h>
12 #include <linux/rculist.h>
13 #include <linux/security.h>
14 #include <linux/tracepoint.h>
15 #include <linux/uaccess.h>
16 
17 #include <asm/ptrace.h>
18 
19 #include "trace_dynevent.h"
20 #include "trace_probe.h"
21 #include "trace_probe_kernel.h"
22 #include "trace_probe_tmpl.h"
23 
24 #define FPROBE_EVENT_SYSTEM "fprobes"
25 #define TRACEPOINT_EVENT_SYSTEM "tracepoints"
26 #define RETHOOK_MAXACTIVE_MAX 4096
27 
28 static int trace_fprobe_create(const char *raw_command);
29 static int trace_fprobe_show(struct seq_file *m, struct dyn_event *ev);
30 static int trace_fprobe_release(struct dyn_event *ev);
31 static bool trace_fprobe_is_busy(struct dyn_event *ev);
32 static bool trace_fprobe_match(const char *system, const char *event,
33 			int argc, const char **argv, struct dyn_event *ev);
34 
35 static struct dyn_event_operations trace_fprobe_ops = {
36 	.create = trace_fprobe_create,
37 	.show = trace_fprobe_show,
38 	.is_busy = trace_fprobe_is_busy,
39 	.free = trace_fprobe_release,
40 	.match = trace_fprobe_match,
41 };
42 
43 /* List of tracepoint_user */
44 static LIST_HEAD(tracepoint_user_list);
45 static DEFINE_MUTEX(tracepoint_user_mutex);
46 
47 /* While living tracepoint_user, @tpoint can be NULL and @refcount != 0. */
48 struct tracepoint_user {
49 	struct list_head	list;
50 	const char		*name;
51 	struct tracepoint	*tpoint;
52 	unsigned int		refcount;
53 };
54 
55 /* NOTE: you must lock tracepoint_user_mutex. */
56 #define for_each_tracepoint_user(tuser)		\
57 	list_for_each_entry(tuser, &tracepoint_user_list, list)
58 
tracepoint_user_register(struct tracepoint_user * tuser)59 static int tracepoint_user_register(struct tracepoint_user *tuser)
60 {
61 	struct tracepoint *tpoint = tuser->tpoint;
62 
63 	if (!tpoint)
64 		return 0;
65 
66 	return tracepoint_probe_register_prio_may_exist(tpoint,
67 					tpoint->probestub, NULL, 0);
68 }
69 
tracepoint_user_unregister(struct tracepoint_user * tuser)70 static void tracepoint_user_unregister(struct tracepoint_user *tuser)
71 {
72 	if (!tuser->tpoint)
73 		return;
74 
75 	WARN_ON_ONCE(tracepoint_probe_unregister(tuser->tpoint, tuser->tpoint->probestub, NULL));
76 	tuser->tpoint = NULL;
77 }
78 
tracepoint_user_ip(struct tracepoint_user * tuser)79 static unsigned long tracepoint_user_ip(struct tracepoint_user *tuser)
80 {
81 	if (!tuser->tpoint)
82 		return 0UL;
83 
84 	return (unsigned long)tuser->tpoint->probestub;
85 }
86 
__tracepoint_user_free(struct tracepoint_user * tuser)87 static void __tracepoint_user_free(struct tracepoint_user *tuser)
88 {
89 	if (!tuser)
90 		return;
91 	kfree(tuser->name);
92 	kfree(tuser);
93 }
94 
DEFINE_FREE(tuser_free,struct tracepoint_user *,__tracepoint_user_free (_T))95 DEFINE_FREE(tuser_free, struct tracepoint_user *, __tracepoint_user_free(_T))
96 
97 static struct tracepoint_user *__tracepoint_user_init(const char *name, struct tracepoint *tpoint)
98 {
99 	struct tracepoint_user *tuser __free(tuser_free) = NULL;
100 	int ret;
101 
102 	tuser = kzalloc(sizeof(*tuser), GFP_KERNEL);
103 	if (!tuser)
104 		return NULL;
105 	tuser->name = kstrdup(name, GFP_KERNEL);
106 	if (!tuser->name)
107 		return NULL;
108 
109 	/* Register tracepoint if it is loaded. */
110 	if (tpoint) {
111 		tuser->tpoint = tpoint;
112 		ret = tracepoint_user_register(tuser);
113 		if (ret)
114 			return ERR_PTR(ret);
115 	}
116 
117 	tuser->refcount = 1;
118 	INIT_LIST_HEAD(&tuser->list);
119 	list_add(&tuser->list, &tracepoint_user_list);
120 
121 	return_ptr(tuser);
122 }
123 
124 static struct tracepoint *find_tracepoint(const char *tp_name,
125 	struct module **tp_mod);
126 
127 /*
128  * Get tracepoint_user if exist, or allocate new one and register it.
129  * If tracepoint is on a module, get its refcounter too.
130  * This returns errno or NULL (not loaded yet) or tracepoint_user.
131  */
tracepoint_user_find_get(const char * name,struct module ** pmod)132 static struct tracepoint_user *tracepoint_user_find_get(const char *name, struct module **pmod)
133 {
134 	struct module *mod __free(module_put) = NULL;
135 	struct tracepoint_user *tuser;
136 	struct tracepoint *tpoint;
137 
138 	if (!name || !pmod)
139 		return ERR_PTR(-EINVAL);
140 
141 	/* Get and lock the module which has tracepoint. */
142 	tpoint = find_tracepoint(name, &mod);
143 
144 	guard(mutex)(&tracepoint_user_mutex);
145 	/* Search existing tracepoint_user */
146 	for_each_tracepoint_user(tuser) {
147 		if (!strcmp(tuser->name, name)) {
148 			tuser->refcount++;
149 			*pmod = no_free_ptr(mod);
150 			return tuser;
151 		}
152 	}
153 
154 	/* The corresponding tracepoint_user is not found. */
155 	tuser = __tracepoint_user_init(name, tpoint);
156 	if (!IS_ERR_OR_NULL(tuser))
157 		*pmod = no_free_ptr(mod);
158 
159 	return tuser;
160 }
161 
tracepoint_user_put(struct tracepoint_user * tuser)162 static void tracepoint_user_put(struct tracepoint_user *tuser)
163 {
164 	scoped_guard(mutex, &tracepoint_user_mutex) {
165 		if (--tuser->refcount > 0)
166 			return;
167 
168 		list_del(&tuser->list);
169 		tracepoint_user_unregister(tuser);
170 	}
171 
172 	__tracepoint_user_free(tuser);
173 }
174 
175 DEFINE_FREE(tuser_put, struct tracepoint_user *,
176 	if (!IS_ERR_OR_NULL(_T))
177 		tracepoint_user_put(_T))
178 
179 /*
180  * Fprobe event core functions
181  */
182 
183 /*
184  * @tprobe is true for tracepoint probe.
185  * @tuser can be NULL if the trace_fprobe is disabled or the tracepoint is not
186  * loaded with a module. If @tuser != NULL, this trace_fprobe is enabled.
187  */
188 struct trace_fprobe {
189 	struct dyn_event	devent;
190 	struct fprobe		fp;
191 	const char		*symbol;
192 	bool			tprobe;
193 	struct tracepoint_user	*tuser;
194 	struct trace_probe	tp;
195 };
196 
is_trace_fprobe(struct dyn_event * ev)197 static bool is_trace_fprobe(struct dyn_event *ev)
198 {
199 	return ev->ops == &trace_fprobe_ops;
200 }
201 
to_trace_fprobe(struct dyn_event * ev)202 static struct trace_fprobe *to_trace_fprobe(struct dyn_event *ev)
203 {
204 	return container_of(ev, struct trace_fprobe, devent);
205 }
206 
207 /**
208  * for_each_trace_fprobe - iterate over the trace_fprobe list
209  * @pos:	the struct trace_fprobe * for each entry
210  * @dpos:	the struct dyn_event * to use as a loop cursor
211  */
212 #define for_each_trace_fprobe(pos, dpos)	\
213 	for_each_dyn_event(dpos)		\
214 		if (is_trace_fprobe(dpos) && (pos = to_trace_fprobe(dpos)))
215 
trace_fprobe_is_return(struct trace_fprobe * tf)216 static bool trace_fprobe_is_return(struct trace_fprobe *tf)
217 {
218 	return tf->fp.exit_handler != NULL;
219 }
220 
trace_fprobe_is_tracepoint(struct trace_fprobe * tf)221 static bool trace_fprobe_is_tracepoint(struct trace_fprobe *tf)
222 {
223 	return tf->tprobe;
224 }
225 
trace_fprobe_symbol(struct trace_fprobe * tf)226 static const char *trace_fprobe_symbol(struct trace_fprobe *tf)
227 {
228 	return tf->symbol ? tf->symbol : "unknown";
229 }
230 
trace_fprobe_is_busy(struct dyn_event * ev)231 static bool trace_fprobe_is_busy(struct dyn_event *ev)
232 {
233 	struct trace_fprobe *tf = to_trace_fprobe(ev);
234 
235 	return trace_probe_is_enabled(&tf->tp);
236 }
237 
trace_fprobe_match_command_head(struct trace_fprobe * tf,int argc,const char ** argv)238 static bool trace_fprobe_match_command_head(struct trace_fprobe *tf,
239 					    int argc, const char **argv)
240 {
241 	char buf[MAX_ARGSTR_LEN + 1];
242 
243 	if (!argc)
244 		return true;
245 
246 	snprintf(buf, sizeof(buf), "%s", trace_fprobe_symbol(tf));
247 	if (strcmp(buf, argv[0]))
248 		return false;
249 	argc--; argv++;
250 
251 	return trace_probe_match_command_args(&tf->tp, argc, argv);
252 }
253 
trace_fprobe_match(const char * system,const char * event,int argc,const char ** argv,struct dyn_event * ev)254 static bool trace_fprobe_match(const char *system, const char *event,
255 			int argc, const char **argv, struct dyn_event *ev)
256 {
257 	struct trace_fprobe *tf = to_trace_fprobe(ev);
258 
259 	if (event[0] != '\0' && strcmp(trace_probe_name(&tf->tp), event))
260 		return false;
261 
262 	if (system && strcmp(trace_probe_group_name(&tf->tp), system))
263 		return false;
264 
265 	return trace_fprobe_match_command_head(tf, argc, argv);
266 }
267 
trace_fprobe_is_registered(struct trace_fprobe * tf)268 static bool trace_fprobe_is_registered(struct trace_fprobe *tf)
269 {
270 	return fprobe_is_registered(&tf->fp);
271 }
272 
273 /*
274  * Note that we don't verify the fetch_insn code, since it does not come
275  * from user space.
276  */
277 static int
process_fetch_insn(struct fetch_insn * code,void * rec,void * edata,void * dest,void * base)278 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
279 		   void *dest, void *base)
280 {
281 	struct ftrace_regs *fregs = rec;
282 	unsigned long val;
283 	int ret;
284 
285 retry:
286 	/* 1st stage: get value from context */
287 	switch (code->op) {
288 	case FETCH_OP_STACK:
289 		val = ftrace_regs_get_kernel_stack_nth(fregs, code->param);
290 		break;
291 	case FETCH_OP_STACKP:
292 		val = ftrace_regs_get_stack_pointer(fregs);
293 		break;
294 	case FETCH_OP_RETVAL:
295 		val = ftrace_regs_get_return_value(fregs);
296 		break;
297 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
298 	case FETCH_OP_ARG:
299 		val = ftrace_regs_get_argument(fregs, code->param);
300 		break;
301 	case FETCH_OP_EDATA:
302 		val = *(unsigned long *)((unsigned long)edata + code->offset);
303 		break;
304 #endif
305 	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
306 		code++;
307 		goto retry;
308 	default:
309 		ret = process_common_fetch_insn(code, &val);
310 		if (ret < 0)
311 			return ret;
312 	}
313 	code++;
314 
315 	return process_fetch_insn_bottom(code, val, dest, base);
316 }
NOKPROBE_SYMBOL(process_fetch_insn)317 NOKPROBE_SYMBOL(process_fetch_insn)
318 
319 /* function entry handler */
320 static nokprobe_inline void
321 __fentry_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
322 		    struct ftrace_regs *fregs,
323 		    struct trace_event_file *trace_file)
324 {
325 	struct fentry_trace_entry_head *entry;
326 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
327 	struct trace_event_buffer fbuffer;
328 	int dsize;
329 
330 	if (WARN_ON_ONCE(call != trace_file->event_call))
331 		return;
332 
333 	if (trace_trigger_soft_disabled(trace_file))
334 		return;
335 
336 	dsize = __get_data_size(&tf->tp, fregs, NULL);
337 
338 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
339 					   sizeof(*entry) + tf->tp.size + dsize);
340 	if (!entry)
341 		return;
342 
343 	fbuffer.regs = ftrace_get_regs(fregs);
344 	entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
345 	entry->ip = entry_ip;
346 	store_trace_args(&entry[1], &tf->tp, fregs, NULL, sizeof(*entry), dsize);
347 
348 	trace_event_buffer_commit(&fbuffer);
349 }
350 
351 static void
fentry_trace_func(struct trace_fprobe * tf,unsigned long entry_ip,struct ftrace_regs * fregs)352 fentry_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
353 		  struct ftrace_regs *fregs)
354 {
355 	struct event_file_link *link;
356 
357 	trace_probe_for_each_link_rcu(link, &tf->tp)
358 		__fentry_trace_func(tf, entry_ip, fregs, link->file);
359 }
360 NOKPROBE_SYMBOL(fentry_trace_func);
361 
362 static nokprobe_inline
store_fprobe_entry_data(void * edata,struct trace_probe * tp,struct ftrace_regs * fregs)363 void store_fprobe_entry_data(void *edata, struct trace_probe *tp, struct ftrace_regs *fregs)
364 {
365 	struct probe_entry_arg *earg = tp->entry_arg;
366 	unsigned long val = 0;
367 	int i;
368 
369 	if (!earg)
370 		return;
371 
372 	for (i = 0; i < earg->size; i++) {
373 		struct fetch_insn *code = &earg->code[i];
374 
375 		switch (code->op) {
376 		case FETCH_OP_ARG:
377 			val = ftrace_regs_get_argument(fregs, code->param);
378 			break;
379 		case FETCH_OP_ST_EDATA:
380 			*(unsigned long *)((unsigned long)edata + code->offset) = val;
381 			break;
382 		case FETCH_OP_END:
383 			goto end;
384 		default:
385 			break;
386 		}
387 	}
388 end:
389 	return;
390 }
391 
392 /* function exit handler */
trace_fprobe_entry_handler(struct fprobe * fp,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)393 static int trace_fprobe_entry_handler(struct fprobe *fp, unsigned long entry_ip,
394 				unsigned long ret_ip, struct ftrace_regs *fregs,
395 				void *entry_data)
396 {
397 	struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp);
398 
399 	if (tf->tp.entry_arg)
400 		store_fprobe_entry_data(entry_data, &tf->tp, fregs);
401 
402 	return 0;
403 }
NOKPROBE_SYMBOL(trace_fprobe_entry_handler)404 NOKPROBE_SYMBOL(trace_fprobe_entry_handler)
405 
406 static nokprobe_inline void
407 __fexit_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
408 		   unsigned long ret_ip, struct ftrace_regs *fregs,
409 		   void *entry_data, struct trace_event_file *trace_file)
410 {
411 	struct fexit_trace_entry_head *entry;
412 	struct trace_event_buffer fbuffer;
413 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
414 	int dsize;
415 
416 	if (WARN_ON_ONCE(call != trace_file->event_call))
417 		return;
418 
419 	if (trace_trigger_soft_disabled(trace_file))
420 		return;
421 
422 	dsize = __get_data_size(&tf->tp, fregs, entry_data);
423 
424 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
425 					   sizeof(*entry) + tf->tp.size + dsize);
426 	if (!entry)
427 		return;
428 
429 	fbuffer.regs = ftrace_get_regs(fregs);
430 	entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
431 	entry->func = entry_ip;
432 	entry->ret_ip = ret_ip;
433 	store_trace_args(&entry[1], &tf->tp, fregs, entry_data, sizeof(*entry), dsize);
434 
435 	trace_event_buffer_commit(&fbuffer);
436 }
437 
438 static void
fexit_trace_func(struct trace_fprobe * tf,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)439 fexit_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
440 		 unsigned long ret_ip, struct ftrace_regs *fregs, void *entry_data)
441 {
442 	struct event_file_link *link;
443 
444 	trace_probe_for_each_link_rcu(link, &tf->tp)
445 		__fexit_trace_func(tf, entry_ip, ret_ip, fregs, entry_data, link->file);
446 }
447 NOKPROBE_SYMBOL(fexit_trace_func);
448 
449 #ifdef CONFIG_PERF_EVENTS
450 
fentry_perf_func(struct trace_fprobe * tf,unsigned long entry_ip,struct ftrace_regs * fregs)451 static int fentry_perf_func(struct trace_fprobe *tf, unsigned long entry_ip,
452 			    struct ftrace_regs *fregs)
453 {
454 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
455 	struct fentry_trace_entry_head *entry;
456 	struct hlist_head *head;
457 	int size, __size, dsize;
458 	struct pt_regs *regs;
459 	int rctx;
460 
461 	head = this_cpu_ptr(call->perf_events);
462 	if (hlist_empty(head))
463 		return 0;
464 
465 	dsize = __get_data_size(&tf->tp, fregs, NULL);
466 	__size = sizeof(*entry) + tf->tp.size + dsize;
467 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
468 	size -= sizeof(u32);
469 
470 	entry = perf_trace_buf_alloc(size, &regs, &rctx);
471 	if (!entry)
472 		return 0;
473 
474 	regs = ftrace_fill_perf_regs(fregs, regs);
475 
476 	entry->ip = entry_ip;
477 	memset(&entry[1], 0, dsize);
478 	store_trace_args(&entry[1], &tf->tp, fregs, NULL, sizeof(*entry), dsize);
479 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
480 			      head, NULL);
481 	return 0;
482 }
483 NOKPROBE_SYMBOL(fentry_perf_func);
484 
485 static void
fexit_perf_func(struct trace_fprobe * tf,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)486 fexit_perf_func(struct trace_fprobe *tf, unsigned long entry_ip,
487 		unsigned long ret_ip, struct ftrace_regs *fregs,
488 		void *entry_data)
489 {
490 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
491 	struct fexit_trace_entry_head *entry;
492 	struct hlist_head *head;
493 	int size, __size, dsize;
494 	struct pt_regs *regs;
495 	int rctx;
496 
497 	head = this_cpu_ptr(call->perf_events);
498 	if (hlist_empty(head))
499 		return;
500 
501 	dsize = __get_data_size(&tf->tp, fregs, entry_data);
502 	__size = sizeof(*entry) + tf->tp.size + dsize;
503 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
504 	size -= sizeof(u32);
505 
506 	entry = perf_trace_buf_alloc(size, &regs, &rctx);
507 	if (!entry)
508 		return;
509 
510 	regs = ftrace_fill_perf_regs(fregs, regs);
511 
512 	entry->func = entry_ip;
513 	entry->ret_ip = ret_ip;
514 	store_trace_args(&entry[1], &tf->tp, fregs, entry_data, sizeof(*entry), dsize);
515 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
516 			      head, NULL);
517 }
518 NOKPROBE_SYMBOL(fexit_perf_func);
519 #endif	/* CONFIG_PERF_EVENTS */
520 
fentry_dispatcher(struct fprobe * fp,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)521 static int fentry_dispatcher(struct fprobe *fp, unsigned long entry_ip,
522 			     unsigned long ret_ip, struct ftrace_regs *fregs,
523 			     void *entry_data)
524 {
525 	struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp);
526 	unsigned int flags = trace_probe_load_flag(&tf->tp);
527 	int ret = 0;
528 
529 	if (flags & TP_FLAG_TRACE)
530 		fentry_trace_func(tf, entry_ip, fregs);
531 
532 #ifdef CONFIG_PERF_EVENTS
533 	if (flags & TP_FLAG_PROFILE)
534 		ret = fentry_perf_func(tf, entry_ip, fregs);
535 #endif
536 	return ret;
537 }
538 NOKPROBE_SYMBOL(fentry_dispatcher);
539 
fexit_dispatcher(struct fprobe * fp,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)540 static void fexit_dispatcher(struct fprobe *fp, unsigned long entry_ip,
541 			     unsigned long ret_ip, struct ftrace_regs *fregs,
542 			     void *entry_data)
543 {
544 	struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp);
545 	unsigned int flags = trace_probe_load_flag(&tf->tp);
546 
547 	if (flags & TP_FLAG_TRACE)
548 		fexit_trace_func(tf, entry_ip, ret_ip, fregs, entry_data);
549 #ifdef CONFIG_PERF_EVENTS
550 	if (flags & TP_FLAG_PROFILE)
551 		fexit_perf_func(tf, entry_ip, ret_ip, fregs, entry_data);
552 #endif
553 }
554 NOKPROBE_SYMBOL(fexit_dispatcher);
555 
free_trace_fprobe(struct trace_fprobe * tf)556 static void free_trace_fprobe(struct trace_fprobe *tf)
557 {
558 	if (tf) {
559 		trace_probe_cleanup(&tf->tp);
560 		if (tf->tuser)
561 			tracepoint_user_put(tf->tuser);
562 		kfree(tf->symbol);
563 		kfree(tf);
564 	}
565 }
566 
567 /* Since alloc_trace_fprobe() can return error, check the pointer is ERR too. */
568 DEFINE_FREE(free_trace_fprobe, struct trace_fprobe *, if (!IS_ERR_OR_NULL(_T)) free_trace_fprobe(_T))
569 
570 /*
571  * Allocate new trace_probe and initialize it (including fprobe).
572  */
alloc_trace_fprobe(const char * group,const char * event,const char * symbol,int nargs,bool is_return,bool is_tracepoint)573 static struct trace_fprobe *alloc_trace_fprobe(const char *group,
574 					       const char *event,
575 					       const char *symbol,
576 					       int nargs, bool is_return,
577 					       bool is_tracepoint)
578 {
579 	struct trace_fprobe *tf __free(free_trace_fprobe) = NULL;
580 	int ret = -ENOMEM;
581 
582 	tf = kzalloc(struct_size(tf, tp.args, nargs), GFP_KERNEL);
583 	if (!tf)
584 		return ERR_PTR(ret);
585 
586 	tf->symbol = kstrdup(symbol, GFP_KERNEL);
587 	if (!tf->symbol)
588 		return ERR_PTR(-ENOMEM);
589 
590 	if (is_return)
591 		tf->fp.exit_handler = fexit_dispatcher;
592 	else
593 		tf->fp.entry_handler = fentry_dispatcher;
594 
595 	tf->tprobe = is_tracepoint;
596 
597 	ret = trace_probe_init(&tf->tp, event, group, false, nargs);
598 	if (ret < 0)
599 		return ERR_PTR(ret);
600 
601 	dyn_event_init(&tf->devent, &trace_fprobe_ops);
602 	return_ptr(tf);
603 }
604 
find_trace_fprobe(const char * event,const char * group)605 static struct trace_fprobe *find_trace_fprobe(const char *event,
606 					      const char *group)
607 {
608 	struct dyn_event *pos;
609 	struct trace_fprobe *tf;
610 
611 	for_each_trace_fprobe(tf, pos)
612 		if (strcmp(trace_probe_name(&tf->tp), event) == 0 &&
613 		    strcmp(trace_probe_group_name(&tf->tp), group) == 0)
614 			return tf;
615 	return NULL;
616 }
617 
618 /* Event entry printers */
619 static enum print_line_t
print_fentry_event(struct trace_iterator * iter,int flags,struct trace_event * event)620 print_fentry_event(struct trace_iterator *iter, int flags,
621 		   struct trace_event *event)
622 {
623 	struct fentry_trace_entry_head *field;
624 	struct trace_seq *s = &iter->seq;
625 	struct trace_probe *tp;
626 
627 	field = (struct fentry_trace_entry_head *)iter->ent;
628 	tp = trace_probe_primary_from_call(
629 		container_of(event, struct trace_event_call, event));
630 	if (WARN_ON_ONCE(!tp))
631 		goto out;
632 
633 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
634 
635 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
636 		goto out;
637 
638 	trace_seq_putc(s, ')');
639 
640 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
641 			     (u8 *)&field[1], field) < 0)
642 		goto out;
643 
644 	trace_seq_putc(s, '\n');
645  out:
646 	return trace_handle_return(s);
647 }
648 
649 static enum print_line_t
print_fexit_event(struct trace_iterator * iter,int flags,struct trace_event * event)650 print_fexit_event(struct trace_iterator *iter, int flags,
651 		  struct trace_event *event)
652 {
653 	struct fexit_trace_entry_head *field;
654 	struct trace_seq *s = &iter->seq;
655 	struct trace_probe *tp;
656 
657 	field = (struct fexit_trace_entry_head *)iter->ent;
658 	tp = trace_probe_primary_from_call(
659 		container_of(event, struct trace_event_call, event));
660 	if (WARN_ON_ONCE(!tp))
661 		goto out;
662 
663 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
664 
665 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
666 		goto out;
667 
668 	trace_seq_puts(s, " <- ");
669 
670 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
671 		goto out;
672 
673 	trace_seq_putc(s, ')');
674 
675 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
676 			     (u8 *)&field[1], field) < 0)
677 		goto out;
678 
679 	trace_seq_putc(s, '\n');
680 
681  out:
682 	return trace_handle_return(s);
683 }
684 
fentry_event_define_fields(struct trace_event_call * event_call)685 static int fentry_event_define_fields(struct trace_event_call *event_call)
686 {
687 	int ret;
688 	struct fentry_trace_entry_head field;
689 	struct trace_probe *tp;
690 
691 	tp = trace_probe_primary_from_call(event_call);
692 	if (WARN_ON_ONCE(!tp))
693 		return -ENOENT;
694 
695 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
696 
697 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
698 }
699 
fexit_event_define_fields(struct trace_event_call * event_call)700 static int fexit_event_define_fields(struct trace_event_call *event_call)
701 {
702 	int ret;
703 	struct fexit_trace_entry_head field;
704 	struct trace_probe *tp;
705 
706 	tp = trace_probe_primary_from_call(event_call);
707 	if (WARN_ON_ONCE(!tp))
708 		return -ENOENT;
709 
710 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
711 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
712 
713 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
714 }
715 
716 static struct trace_event_functions fentry_funcs = {
717 	.trace		= print_fentry_event
718 };
719 
720 static struct trace_event_functions fexit_funcs = {
721 	.trace		= print_fexit_event
722 };
723 
724 static struct trace_event_fields fentry_fields_array[] = {
725 	{ .type = TRACE_FUNCTION_TYPE,
726 	  .define_fields = fentry_event_define_fields },
727 	{}
728 };
729 
730 static struct trace_event_fields fexit_fields_array[] = {
731 	{ .type = TRACE_FUNCTION_TYPE,
732 	  .define_fields = fexit_event_define_fields },
733 	{}
734 };
735 
736 static int fprobe_register(struct trace_event_call *event,
737 			   enum trace_reg type, void *data);
738 
init_trace_event_call(struct trace_fprobe * tf)739 static inline void init_trace_event_call(struct trace_fprobe *tf)
740 {
741 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
742 
743 	if (trace_fprobe_is_return(tf)) {
744 		call->event.funcs = &fexit_funcs;
745 		call->class->fields_array = fexit_fields_array;
746 	} else {
747 		call->event.funcs = &fentry_funcs;
748 		call->class->fields_array = fentry_fields_array;
749 	}
750 
751 	call->flags = TRACE_EVENT_FL_FPROBE;
752 	call->class->reg = fprobe_register;
753 }
754 
register_fprobe_event(struct trace_fprobe * tf)755 static int register_fprobe_event(struct trace_fprobe *tf)
756 {
757 	init_trace_event_call(tf);
758 
759 	return trace_probe_register_event_call(&tf->tp);
760 }
761 
unregister_fprobe_event(struct trace_fprobe * tf)762 static int unregister_fprobe_event(struct trace_fprobe *tf)
763 {
764 	return trace_probe_unregister_event_call(&tf->tp);
765 }
766 
__regsiter_tracepoint_fprobe(struct trace_fprobe * tf)767 static int __regsiter_tracepoint_fprobe(struct trace_fprobe *tf)
768 {
769 	struct tracepoint_user *tuser __free(tuser_put) = NULL;
770 	struct module *mod __free(module_put) = NULL;
771 	unsigned long ip;
772 	int ret;
773 
774 	if (WARN_ON_ONCE(tf->tuser))
775 		return -EINVAL;
776 
777 	/* If the tracepoint is in a module, it must be locked in this function. */
778 	tuser = tracepoint_user_find_get(tf->symbol, &mod);
779 	/* This tracepoint is not loaded yet */
780 	if (IS_ERR(tuser))
781 		return PTR_ERR(tuser);
782 	if (!tuser)
783 		return -ENOMEM;
784 
785 	/* Register fprobe only if the tracepoint is loaded. */
786 	if (tuser->tpoint) {
787 		ip = tracepoint_user_ip(tuser);
788 		if (WARN_ON_ONCE(!ip))
789 			return -ENOENT;
790 
791 		ret = register_fprobe_ips(&tf->fp, &ip, 1);
792 		if (ret < 0)
793 			return ret;
794 	}
795 
796 	tf->tuser = no_free_ptr(tuser);
797 	return 0;
798 }
799 
800 /* Returns an error if the target function is not available, or 0 */
trace_fprobe_verify_target(struct trace_fprobe * tf)801 static int trace_fprobe_verify_target(struct trace_fprobe *tf)
802 {
803 	int ret;
804 
805 	/* Tracepoint should have a stub function. */
806 	if (trace_fprobe_is_tracepoint(tf))
807 		return 0;
808 
809 	/*
810 	 * Note: since we don't lock the module, even if this succeeded,
811 	 * register_fprobe() later can fail.
812 	 */
813 	ret = fprobe_count_ips_from_filter(tf->symbol, NULL);
814 	return (ret < 0) ? ret : 0;
815 }
816 
817 /* Internal register function - just handle fprobe and flags */
__register_trace_fprobe(struct trace_fprobe * tf)818 static int __register_trace_fprobe(struct trace_fprobe *tf)
819 {
820 	int i, ret;
821 
822 	/* Should we need new LOCKDOWN flag for fprobe? */
823 	ret = security_locked_down(LOCKDOWN_KPROBES);
824 	if (ret)
825 		return ret;
826 
827 	if (trace_fprobe_is_registered(tf))
828 		return -EINVAL;
829 
830 	for (i = 0; i < tf->tp.nr_args; i++) {
831 		ret = traceprobe_update_arg(&tf->tp.args[i]);
832 		if (ret)
833 			return ret;
834 	}
835 
836 	tf->fp.flags &= ~FPROBE_FL_DISABLED;
837 
838 	if (trace_fprobe_is_tracepoint(tf))
839 		return __regsiter_tracepoint_fprobe(tf);
840 
841 	/* TODO: handle filter, nofilter or symbol list */
842 	return register_fprobe(&tf->fp, tf->symbol, NULL);
843 }
844 
845 /* Internal unregister function - just handle fprobe and flags */
__unregister_trace_fprobe(struct trace_fprobe * tf)846 static void __unregister_trace_fprobe(struct trace_fprobe *tf)
847 {
848 	if (trace_fprobe_is_registered(tf))
849 		unregister_fprobe(&tf->fp);
850 	if (tf->tuser) {
851 		tracepoint_user_put(tf->tuser);
852 		tf->tuser = NULL;
853 	}
854 }
855 
856 /* TODO: make this trace_*probe common function */
857 /* Unregister a trace_probe and probe_event */
unregister_trace_fprobe(struct trace_fprobe * tf)858 static int unregister_trace_fprobe(struct trace_fprobe *tf)
859 {
860 	/* If other probes are on the event, just unregister fprobe */
861 	if (trace_probe_has_sibling(&tf->tp))
862 		goto unreg;
863 
864 	/* Enabled event can not be unregistered */
865 	if (trace_probe_is_enabled(&tf->tp))
866 		return -EBUSY;
867 
868 	/* If there's a reference to the dynamic event */
869 	if (trace_event_dyn_busy(trace_probe_event_call(&tf->tp)))
870 		return -EBUSY;
871 
872 	/* Will fail if probe is being used by ftrace or perf */
873 	if (unregister_fprobe_event(tf))
874 		return -EBUSY;
875 
876 unreg:
877 	__unregister_trace_fprobe(tf);
878 	dyn_event_remove(&tf->devent);
879 	trace_probe_unlink(&tf->tp);
880 
881 	return 0;
882 }
883 
trace_fprobe_has_same_fprobe(struct trace_fprobe * orig,struct trace_fprobe * comp)884 static bool trace_fprobe_has_same_fprobe(struct trace_fprobe *orig,
885 					 struct trace_fprobe *comp)
886 {
887 	struct trace_probe_event *tpe = orig->tp.event;
888 	int i;
889 
890 	list_for_each_entry(orig, &tpe->probes, tp.list) {
891 		if (strcmp(trace_fprobe_symbol(orig),
892 			   trace_fprobe_symbol(comp)))
893 			continue;
894 
895 		/*
896 		 * trace_probe_compare_arg_type() ensured that nr_args and
897 		 * each argument name and type are same. Let's compare comm.
898 		 */
899 		for (i = 0; i < orig->tp.nr_args; i++) {
900 			if (strcmp(orig->tp.args[i].comm,
901 				   comp->tp.args[i].comm))
902 				break;
903 		}
904 
905 		if (i == orig->tp.nr_args)
906 			return true;
907 	}
908 
909 	return false;
910 }
911 
append_trace_fprobe_event(struct trace_fprobe * tf,struct trace_fprobe * to)912 static int append_trace_fprobe_event(struct trace_fprobe *tf, struct trace_fprobe *to)
913 {
914 	int ret;
915 
916 	if (trace_fprobe_is_return(tf) != trace_fprobe_is_return(to) ||
917 	    trace_fprobe_is_tracepoint(tf) != trace_fprobe_is_tracepoint(to)) {
918 		trace_probe_log_set_index(0);
919 		trace_probe_log_err(0, DIFF_PROBE_TYPE);
920 		return -EEXIST;
921 	}
922 	ret = trace_probe_compare_arg_type(&tf->tp, &to->tp);
923 	if (ret) {
924 		/* Note that argument starts index = 2 */
925 		trace_probe_log_set_index(ret + 1);
926 		trace_probe_log_err(0, DIFF_ARG_TYPE);
927 		return -EEXIST;
928 	}
929 	if (trace_fprobe_has_same_fprobe(to, tf)) {
930 		trace_probe_log_set_index(0);
931 		trace_probe_log_err(0, SAME_PROBE);
932 		return -EEXIST;
933 	}
934 
935 	/* Append to existing event */
936 	ret = trace_probe_append(&tf->tp, &to->tp);
937 	if (ret)
938 		return ret;
939 
940 	ret = trace_fprobe_verify_target(tf);
941 	if (ret)
942 		trace_probe_unlink(&tf->tp);
943 	else
944 		dyn_event_add(&tf->devent, trace_probe_event_call(&tf->tp));
945 
946 	return ret;
947 }
948 
949 /* Register a trace_probe and probe_event, and check the fprobe is available. */
register_trace_fprobe_event(struct trace_fprobe * tf)950 static int register_trace_fprobe_event(struct trace_fprobe *tf)
951 {
952 	struct trace_fprobe *old_tf;
953 	int ret;
954 
955 	guard(mutex)(&event_mutex);
956 
957 	old_tf = find_trace_fprobe(trace_probe_name(&tf->tp),
958 				   trace_probe_group_name(&tf->tp));
959 	if (old_tf)
960 		return append_trace_fprobe_event(tf, old_tf);
961 
962 	/* Register new event */
963 	ret = register_fprobe_event(tf);
964 	if (ret) {
965 		if (ret == -EEXIST) {
966 			trace_probe_log_set_index(0);
967 			trace_probe_log_err(0, EVENT_EXIST);
968 		} else
969 			pr_warn("Failed to register probe event(%d)\n", ret);
970 		return ret;
971 	}
972 
973 	/* Verify fprobe is sane. */
974 	ret = trace_fprobe_verify_target(tf);
975 	if (ret < 0)
976 		unregister_fprobe_event(tf);
977 	else
978 		dyn_event_add(&tf->devent, trace_probe_event_call(&tf->tp));
979 
980 	return ret;
981 }
982 
983 struct __find_tracepoint_cb_data {
984 	const char *tp_name;
985 	struct tracepoint *tpoint;
986 	struct module *mod;
987 };
988 
__find_tracepoint_module_cb(struct tracepoint * tp,struct module * mod,void * priv)989 static void __find_tracepoint_module_cb(struct tracepoint *tp, struct module *mod, void *priv)
990 {
991 	struct __find_tracepoint_cb_data *data = priv;
992 
993 	if (!data->tpoint && !strcmp(data->tp_name, tp->name)) {
994 		/* If module is not specified, try getting module refcount. */
995 		if (!data->mod && mod) {
996 			/* If failed to get refcount, ignore this tracepoint. */
997 			if (!try_module_get(mod))
998 				return;
999 
1000 			data->mod = mod;
1001 		}
1002 		data->tpoint = tp;
1003 	}
1004 }
1005 
__find_tracepoint_cb(struct tracepoint * tp,void * priv)1006 static void __find_tracepoint_cb(struct tracepoint *tp, void *priv)
1007 {
1008 	struct __find_tracepoint_cb_data *data = priv;
1009 
1010 	if (!data->tpoint && !strcmp(data->tp_name, tp->name))
1011 		data->tpoint = tp;
1012 }
1013 
1014 /*
1015  * Find a tracepoint from kernel and module. If the tracepoint is on the module,
1016  * the module's refcount is incremented and returned as *@tp_mod. Thus, if it is
1017  * not NULL, caller must call module_put(*tp_mod) after used the tracepoint.
1018  */
find_tracepoint(const char * tp_name,struct module ** tp_mod)1019 static struct tracepoint *find_tracepoint(const char *tp_name,
1020 					  struct module **tp_mod)
1021 {
1022 	struct __find_tracepoint_cb_data data = {
1023 		.tp_name = tp_name,
1024 		.mod = NULL,
1025 	};
1026 
1027 	for_each_kernel_tracepoint(__find_tracepoint_cb, &data);
1028 
1029 	if (!data.tpoint && IS_ENABLED(CONFIG_MODULES)) {
1030 		for_each_module_tracepoint(__find_tracepoint_module_cb, &data);
1031 		*tp_mod = data.mod;
1032 	}
1033 
1034 	return data.tpoint;
1035 }
1036 
1037 #ifdef CONFIG_MODULES
1038 /*
1039  * Find a tracepoint from specified module. In this case, this does not get the
1040  * module's refcount. The caller must ensure the module is not freed.
1041  */
find_tracepoint_in_module(struct module * mod,const char * tp_name)1042 static struct tracepoint *find_tracepoint_in_module(struct module *mod,
1043 						    const char *tp_name)
1044 {
1045 	struct __find_tracepoint_cb_data data = {
1046 		.tp_name = tp_name,
1047 		.mod = mod,
1048 	};
1049 
1050 	for_each_tracepoint_in_module(mod, __find_tracepoint_module_cb, &data);
1051 	return data.tpoint;
1052 }
1053 
1054 /* These are CONFIG_MODULES=y specific functions. */
tracepoint_user_within_module(struct tracepoint_user * tuser,struct module * mod)1055 static bool tracepoint_user_within_module(struct tracepoint_user *tuser,
1056 					  struct module *mod)
1057 {
1058 	return within_module(tracepoint_user_ip(tuser), mod);
1059 }
1060 
tracepoint_user_register_again(struct tracepoint_user * tuser,struct tracepoint * tpoint)1061 static int tracepoint_user_register_again(struct tracepoint_user *tuser,
1062 					  struct tracepoint *tpoint)
1063 {
1064 	tuser->tpoint = tpoint;
1065 	return tracepoint_user_register(tuser);
1066 }
1067 
tracepoint_user_unregister_clear(struct tracepoint_user * tuser)1068 static void tracepoint_user_unregister_clear(struct tracepoint_user *tuser)
1069 {
1070 	tracepoint_user_unregister(tuser);
1071 	tuser->tpoint = NULL;
1072 }
1073 
1074 /* module callback for tracepoint_user */
__tracepoint_probe_module_cb(struct notifier_block * self,unsigned long val,void * data)1075 static int __tracepoint_probe_module_cb(struct notifier_block *self,
1076 					unsigned long val, void *data)
1077 {
1078 	struct tp_module *tp_mod = data;
1079 	struct tracepoint_user *tuser;
1080 	struct tracepoint *tpoint;
1081 
1082 	if (val != MODULE_STATE_GOING && val != MODULE_STATE_COMING)
1083 		return NOTIFY_DONE;
1084 
1085 	mutex_lock(&tracepoint_user_mutex);
1086 	for_each_tracepoint_user(tuser) {
1087 		if (val == MODULE_STATE_COMING) {
1088 			/* This is not a tracepoint in this module. Skip it. */
1089 			tpoint = find_tracepoint_in_module(tp_mod->mod, tuser->name);
1090 			if (!tpoint)
1091 				continue;
1092 			WARN_ON_ONCE(tracepoint_user_register_again(tuser, tpoint));
1093 		} else if (val == MODULE_STATE_GOING &&
1094 			  tracepoint_user_within_module(tuser, tp_mod->mod)) {
1095 			/* Unregister all tracepoint_user in this module. */
1096 			tracepoint_user_unregister_clear(tuser);
1097 		}
1098 	}
1099 	mutex_unlock(&tracepoint_user_mutex);
1100 
1101 	return NOTIFY_DONE;
1102 }
1103 
1104 static struct notifier_block tracepoint_module_nb = {
1105 	.notifier_call = __tracepoint_probe_module_cb,
1106 };
1107 
1108 /* module callback for tprobe events */
__tprobe_event_module_cb(struct notifier_block * self,unsigned long val,void * data)1109 static int __tprobe_event_module_cb(struct notifier_block *self,
1110 				     unsigned long val, void *data)
1111 {
1112 	struct trace_fprobe *tf;
1113 	struct dyn_event *pos;
1114 	struct module *mod = data;
1115 
1116 	if (val != MODULE_STATE_GOING && val != MODULE_STATE_COMING)
1117 		return NOTIFY_DONE;
1118 
1119 	mutex_lock(&event_mutex);
1120 	for_each_trace_fprobe(tf, pos) {
1121 		/* Skip fprobe and disabled tprobe events. */
1122 		if (!trace_fprobe_is_tracepoint(tf) || !tf->tuser)
1123 			continue;
1124 
1125 		/* Before this notification, tracepoint notifier has already done. */
1126 		if (val == MODULE_STATE_COMING &&
1127 		    tracepoint_user_within_module(tf->tuser, mod)) {
1128 			unsigned long ip = tracepoint_user_ip(tf->tuser);
1129 
1130 			WARN_ON_ONCE(register_fprobe_ips(&tf->fp, &ip, 1));
1131 		} else if (val == MODULE_STATE_GOING &&
1132 			   /*
1133 			    * tracepoint_user_within_module() does not work here because
1134 			    * tracepoint_user is already unregistered and cleared tpoint.
1135 			    * Instead, checking whether the fprobe is registered but
1136 			    * tpoint is cleared(unregistered). Such unbalance probes
1137 			    * must be adjusted anyway.
1138 			    */
1139 			    trace_fprobe_is_registered(tf) &&
1140 			    !tf->tuser->tpoint) {
1141 			unregister_fprobe(&tf->fp);
1142 		}
1143 	}
1144 	mutex_unlock(&event_mutex);
1145 
1146 	return NOTIFY_DONE;
1147 }
1148 
1149 /* NOTE: this must be called after tracepoint callback */
1150 static struct notifier_block tprobe_event_module_nb = {
1151 	.notifier_call = __tprobe_event_module_cb,
1152 	/* Make sure this is later than tracepoint module notifier. */
1153 	.priority = -10,
1154 };
1155 #endif /* CONFIG_MODULES */
1156 
parse_symbol_and_return(int argc,const char * argv[],char ** symbol,bool * is_return,bool is_tracepoint)1157 static int parse_symbol_and_return(int argc, const char *argv[],
1158 				   char **symbol, bool *is_return,
1159 				   bool is_tracepoint)
1160 {
1161 	char *tmp = strchr(argv[1], '%');
1162 	int i;
1163 
1164 	if (tmp) {
1165 		int len = tmp - argv[1];
1166 
1167 		if (!is_tracepoint && !strcmp(tmp, "%return")) {
1168 			*is_return = true;
1169 		} else {
1170 			trace_probe_log_err(len, BAD_ADDR_SUFFIX);
1171 			return -EINVAL;
1172 		}
1173 		*symbol = kmemdup_nul(argv[1], len, GFP_KERNEL);
1174 	} else
1175 		*symbol = kstrdup(argv[1], GFP_KERNEL);
1176 	if (!*symbol)
1177 		return -ENOMEM;
1178 
1179 	if (*is_return)
1180 		return 0;
1181 
1182 	if (is_tracepoint) {
1183 		tmp = *symbol;
1184 		while (*tmp && (isalnum(*tmp) || *tmp == '_'))
1185 			tmp++;
1186 		if (*tmp) {
1187 			/* find a wrong character. */
1188 			trace_probe_log_err(tmp - *symbol, BAD_TP_NAME);
1189 			kfree(*symbol);
1190 			*symbol = NULL;
1191 			return -EINVAL;
1192 		}
1193 	}
1194 
1195 	/* If there is $retval, this should be a return fprobe. */
1196 	for (i = 2; i < argc; i++) {
1197 		tmp = strstr(argv[i], "$retval");
1198 		if (tmp && !isalnum(tmp[7]) && tmp[7] != '_') {
1199 			if (is_tracepoint) {
1200 				trace_probe_log_set_index(i);
1201 				trace_probe_log_err(tmp - argv[i], RETVAL_ON_PROBE);
1202 				kfree(*symbol);
1203 				*symbol = NULL;
1204 				return -EINVAL;
1205 			}
1206 			*is_return = true;
1207 			break;
1208 		}
1209 	}
1210 	return 0;
1211 }
1212 
trace_fprobe_create_internal(int argc,const char * argv[],struct traceprobe_parse_context * ctx)1213 static int trace_fprobe_create_internal(int argc, const char *argv[],
1214 					struct traceprobe_parse_context *ctx)
1215 {
1216 	/*
1217 	 * Argument syntax:
1218 	 *  - Add fentry probe:
1219 	 *      f[:[GRP/][EVENT]] [MOD:]KSYM [FETCHARGS]
1220 	 *  - Add fexit probe:
1221 	 *      f[N][:[GRP/][EVENT]] [MOD:]KSYM%return [FETCHARGS]
1222 	 *  - Add tracepoint probe:
1223 	 *      t[:[GRP/][EVENT]] TRACEPOINT [FETCHARGS]
1224 	 *
1225 	 * Fetch args:
1226 	 *  $retval	: fetch return value
1227 	 *  $stack	: fetch stack address
1228 	 *  $stackN	: fetch Nth entry of stack (N:0-)
1229 	 *  $argN	: fetch Nth argument (N:1-)
1230 	 *  $comm       : fetch current task comm
1231 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
1232 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
1233 	 * Dereferencing memory fetch:
1234 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
1235 	 * Alias name of args:
1236 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
1237 	 * Type of args:
1238 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
1239 	 */
1240 	struct trace_fprobe *tf __free(free_trace_fprobe) = NULL;
1241 	const char *event = NULL, *group = FPROBE_EVENT_SYSTEM;
1242 	struct module *mod __free(module_put) = NULL;
1243 	const char **new_argv __free(kfree) = NULL;
1244 	char *symbol __free(kfree) = NULL;
1245 	char *ebuf __free(kfree) = NULL;
1246 	char *gbuf __free(kfree) = NULL;
1247 	char *sbuf __free(kfree) = NULL;
1248 	char *abuf __free(kfree) = NULL;
1249 	char *dbuf __free(kfree) = NULL;
1250 	int i, new_argc = 0, ret = 0;
1251 	bool is_tracepoint = false;
1252 	bool is_return = false;
1253 
1254 	if ((argv[0][0] != 'f' && argv[0][0] != 't') || argc < 2)
1255 		return -ECANCELED;
1256 
1257 	if (argv[0][0] == 't') {
1258 		is_tracepoint = true;
1259 		group = TRACEPOINT_EVENT_SYSTEM;
1260 	}
1261 
1262 	if (argv[0][1] != '\0') {
1263 		if (argv[0][1] != ':') {
1264 			trace_probe_log_set_index(0);
1265 			trace_probe_log_err(1, BAD_MAXACT);
1266 			return -EINVAL;
1267 		}
1268 		event = &argv[0][2];
1269 	}
1270 
1271 	trace_probe_log_set_index(1);
1272 
1273 	/* a symbol(or tracepoint) must be specified */
1274 	ret = parse_symbol_and_return(argc, argv, &symbol, &is_return, is_tracepoint);
1275 	if (ret < 0)
1276 		return -EINVAL;
1277 
1278 	trace_probe_log_set_index(0);
1279 	if (event) {
1280 		gbuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
1281 		if (!gbuf)
1282 			return -ENOMEM;
1283 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
1284 						  event - argv[0]);
1285 		if (ret)
1286 			return -EINVAL;
1287 	}
1288 
1289 	if (!event) {
1290 		ebuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
1291 		if (!ebuf)
1292 			return -ENOMEM;
1293 		/* Make a new event name */
1294 		if (is_tracepoint)
1295 			snprintf(ebuf, MAX_EVENT_NAME_LEN, "%s%s",
1296 				 isdigit(*symbol) ? "_" : "", symbol);
1297 		else
1298 			snprintf(ebuf, MAX_EVENT_NAME_LEN, "%s__%s", symbol,
1299 				 is_return ? "exit" : "entry");
1300 		sanitize_event_name(ebuf);
1301 		event = ebuf;
1302 	}
1303 
1304 	if (is_return)
1305 		ctx->flags |= TPARG_FL_RETURN;
1306 	else
1307 		ctx->flags |= TPARG_FL_FENTRY;
1308 
1309 	ctx->funcname = NULL;
1310 	if (is_tracepoint) {
1311 		/* Get tracepoint and lock its module until the end of the registration. */
1312 		struct tracepoint *tpoint;
1313 
1314 		ctx->flags |= TPARG_FL_TPOINT;
1315 		mod = NULL;
1316 		tpoint = find_tracepoint(symbol, &mod);
1317 		if (tpoint) {
1318 			sbuf = kmalloc(KSYM_NAME_LEN, GFP_KERNEL);
1319 			if (!sbuf)
1320 				return -ENOMEM;
1321 			ctx->funcname = kallsyms_lookup((unsigned long)tpoint->probestub,
1322 							NULL, NULL, NULL, sbuf);
1323 		}
1324 	}
1325 	if (!ctx->funcname)
1326 		ctx->funcname = symbol;
1327 
1328 	abuf = kmalloc(MAX_BTF_ARGS_LEN, GFP_KERNEL);
1329 	if (!abuf)
1330 		return -ENOMEM;
1331 	argc -= 2; argv += 2;
1332 	new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
1333 					       abuf, MAX_BTF_ARGS_LEN, ctx);
1334 	if (IS_ERR(new_argv))
1335 		return PTR_ERR(new_argv);
1336 	if (new_argv) {
1337 		argc = new_argc;
1338 		argv = new_argv;
1339 	}
1340 	if (argc > MAX_TRACE_ARGS) {
1341 		trace_probe_log_set_index(2);
1342 		trace_probe_log_err(0, TOO_MANY_ARGS);
1343 		return -E2BIG;
1344 	}
1345 
1346 	ret = traceprobe_expand_dentry_args(argc, argv, &dbuf);
1347 	if (ret)
1348 		return ret;
1349 
1350 	/* setup a probe */
1351 	tf = alloc_trace_fprobe(group, event, symbol, argc, is_return, is_tracepoint);
1352 	if (IS_ERR(tf)) {
1353 		ret = PTR_ERR(tf);
1354 		/* This must return -ENOMEM, else there is a bug */
1355 		WARN_ON_ONCE(ret != -ENOMEM);
1356 		return ret;
1357 	}
1358 
1359 	/* parse arguments */
1360 	for (i = 0; i < argc; i++) {
1361 		trace_probe_log_set_index(i + 2);
1362 		ctx->offset = 0;
1363 		ret = traceprobe_parse_probe_arg(&tf->tp, i, argv[i], ctx);
1364 		if (ret)
1365 			return ret;	/* This can be -ENOMEM */
1366 	}
1367 
1368 	if (is_return && tf->tp.entry_arg) {
1369 		tf->fp.entry_handler = trace_fprobe_entry_handler;
1370 		tf->fp.entry_data_size = traceprobe_get_entry_data_size(&tf->tp);
1371 		if (ALIGN(tf->fp.entry_data_size, sizeof(long)) > MAX_FPROBE_DATA_SIZE) {
1372 			trace_probe_log_set_index(2);
1373 			trace_probe_log_err(0, TOO_MANY_EARGS);
1374 			return -E2BIG;
1375 		}
1376 	}
1377 
1378 	ret = traceprobe_set_print_fmt(&tf->tp,
1379 			is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL);
1380 	if (ret < 0)
1381 		return ret;
1382 
1383 	ret = register_trace_fprobe_event(tf);
1384 	if (ret) {
1385 		trace_probe_log_set_index(1);
1386 		if (ret == -EILSEQ)
1387 			trace_probe_log_err(0, BAD_INSN_BNDRY);
1388 		else if (ret == -ENOENT)
1389 			trace_probe_log_err(0, BAD_PROBE_ADDR);
1390 		else if (ret != -ENOMEM && ret != -EEXIST)
1391 			trace_probe_log_err(0, FAIL_REG_PROBE);
1392 		return -EINVAL;
1393 	}
1394 
1395 	/* 'tf' is successfully registered. To avoid freeing, assign NULL. */
1396 	tf = NULL;
1397 
1398 	return 0;
1399 }
1400 
trace_fprobe_create_cb(int argc,const char * argv[])1401 static int trace_fprobe_create_cb(int argc, const char *argv[])
1402 {
1403 	struct traceprobe_parse_context *ctx __free(traceprobe_parse_context) = NULL;
1404 	int ret;
1405 
1406 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1407 	if (!ctx)
1408 		return -ENOMEM;
1409 
1410 	ctx->flags = TPARG_FL_KERNEL | TPARG_FL_FPROBE;
1411 
1412 	trace_probe_log_init("trace_fprobe", argc, argv);
1413 	ret = trace_fprobe_create_internal(argc, argv, ctx);
1414 	trace_probe_log_clear();
1415 	return ret;
1416 }
1417 
trace_fprobe_create(const char * raw_command)1418 static int trace_fprobe_create(const char *raw_command)
1419 {
1420 	return trace_probe_create(raw_command, trace_fprobe_create_cb);
1421 }
1422 
trace_fprobe_release(struct dyn_event * ev)1423 static int trace_fprobe_release(struct dyn_event *ev)
1424 {
1425 	struct trace_fprobe *tf = to_trace_fprobe(ev);
1426 	int ret = unregister_trace_fprobe(tf);
1427 
1428 	if (!ret)
1429 		free_trace_fprobe(tf);
1430 	return ret;
1431 }
1432 
trace_fprobe_show(struct seq_file * m,struct dyn_event * ev)1433 static int trace_fprobe_show(struct seq_file *m, struct dyn_event *ev)
1434 {
1435 	struct trace_fprobe *tf = to_trace_fprobe(ev);
1436 	int i;
1437 
1438 	if (trace_fprobe_is_tracepoint(tf))
1439 		seq_putc(m, 't');
1440 	else
1441 		seq_putc(m, 'f');
1442 	seq_printf(m, ":%s/%s", trace_probe_group_name(&tf->tp),
1443 				trace_probe_name(&tf->tp));
1444 
1445 	seq_printf(m, " %s%s", trace_fprobe_symbol(tf),
1446 			       trace_fprobe_is_return(tf) ? "%return" : "");
1447 
1448 	for (i = 0; i < tf->tp.nr_args; i++)
1449 		seq_printf(m, " %s=%s", tf->tp.args[i].name, tf->tp.args[i].comm);
1450 	seq_putc(m, '\n');
1451 
1452 	return 0;
1453 }
1454 
1455 /*
1456  * Enable trace_probe
1457  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
1458  */
enable_trace_fprobe(struct trace_event_call * call,struct trace_event_file * file)1459 static int enable_trace_fprobe(struct trace_event_call *call,
1460 			       struct trace_event_file *file)
1461 {
1462 	struct trace_probe *tp;
1463 	struct trace_fprobe *tf;
1464 	bool enabled;
1465 	int ret = 0;
1466 
1467 	tp = trace_probe_primary_from_call(call);
1468 	if (WARN_ON_ONCE(!tp))
1469 		return -ENODEV;
1470 	enabled = trace_probe_is_enabled(tp);
1471 
1472 	/* This also changes "enabled" state */
1473 	if (file) {
1474 		ret = trace_probe_add_file(tp, file);
1475 		if (ret)
1476 			return ret;
1477 	} else
1478 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1479 
1480 	if (!enabled) {
1481 		list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) {
1482 			ret = __register_trace_fprobe(tf);
1483 			if (ret < 0)
1484 				return ret;
1485 		}
1486 	}
1487 
1488 	return 0;
1489 }
1490 
1491 /*
1492  * Disable trace_probe
1493  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
1494  */
disable_trace_fprobe(struct trace_event_call * call,struct trace_event_file * file)1495 static int disable_trace_fprobe(struct trace_event_call *call,
1496 				struct trace_event_file *file)
1497 {
1498 	struct trace_fprobe *tf;
1499 	struct trace_probe *tp;
1500 
1501 	tp = trace_probe_primary_from_call(call);
1502 	if (WARN_ON_ONCE(!tp))
1503 		return -ENODEV;
1504 
1505 	if (file) {
1506 		if (!trace_probe_get_file_link(tp, file))
1507 			return -ENOENT;
1508 		if (!trace_probe_has_single_file(tp))
1509 			goto out;
1510 		trace_probe_clear_flag(tp, TP_FLAG_TRACE);
1511 	} else
1512 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1513 
1514 	if (!trace_probe_is_enabled(tp)) {
1515 		list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) {
1516 			unregister_fprobe(&tf->fp);
1517 			if (tf->tuser) {
1518 				tracepoint_user_put(tf->tuser);
1519 				tf->tuser = NULL;
1520 			}
1521 		}
1522 	}
1523 
1524  out:
1525 	if (file)
1526 		/*
1527 		 * Synchronization is done in below function. For perf event,
1528 		 * file == NULL and perf_trace_event_unreg() calls
1529 		 * tracepoint_synchronize_unregister() to ensure synchronize
1530 		 * event. We don't need to care about it.
1531 		 */
1532 		trace_probe_remove_file(tp, file);
1533 
1534 	return 0;
1535 }
1536 
1537 /*
1538  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1539  */
fprobe_register(struct trace_event_call * event,enum trace_reg type,void * data)1540 static int fprobe_register(struct trace_event_call *event,
1541 			   enum trace_reg type, void *data)
1542 {
1543 	struct trace_event_file *file = data;
1544 
1545 	switch (type) {
1546 	case TRACE_REG_REGISTER:
1547 		return enable_trace_fprobe(event, file);
1548 	case TRACE_REG_UNREGISTER:
1549 		return disable_trace_fprobe(event, file);
1550 
1551 #ifdef CONFIG_PERF_EVENTS
1552 	case TRACE_REG_PERF_REGISTER:
1553 		return enable_trace_fprobe(event, NULL);
1554 	case TRACE_REG_PERF_UNREGISTER:
1555 		return disable_trace_fprobe(event, NULL);
1556 	case TRACE_REG_PERF_OPEN:
1557 	case TRACE_REG_PERF_CLOSE:
1558 	case TRACE_REG_PERF_ADD:
1559 	case TRACE_REG_PERF_DEL:
1560 		return 0;
1561 #endif
1562 	}
1563 	return 0;
1564 }
1565 
1566 /*
1567  * Register dynevent at core_initcall. This allows kernel to setup fprobe
1568  * events in postcore_initcall without tracefs.
1569  */
init_fprobe_trace_early(void)1570 static __init int init_fprobe_trace_early(void)
1571 {
1572 	int ret;
1573 
1574 	ret = dyn_event_register(&trace_fprobe_ops);
1575 	if (ret)
1576 		return ret;
1577 
1578 #ifdef CONFIG_MODULES
1579 	ret = register_tracepoint_module_notifier(&tracepoint_module_nb);
1580 	if (ret)
1581 		return ret;
1582 	ret = register_module_notifier(&tprobe_event_module_nb);
1583 	if (ret)
1584 		return ret;
1585 #endif
1586 
1587 	return 0;
1588 }
1589 core_initcall(init_fprobe_trace_early);
1590