xref: /linux/kernel/trace/trace_fprobe.c (revision 6439079365dcb33c6701f5f0e480ac2c17312b6b)
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_obj(*tuser);
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 	if (!argc)
242 		return true;
243 
244 	if (strcmp(trace_fprobe_symbol(tf), argv[0]))
245 		return false;
246 	argc--; argv++;
247 
248 	return trace_probe_match_command_args(&tf->tp, argc, argv);
249 }
250 
trace_fprobe_match(const char * system,const char * event,int argc,const char ** argv,struct dyn_event * ev)251 static bool trace_fprobe_match(const char *system, const char *event,
252 			int argc, const char **argv, struct dyn_event *ev)
253 {
254 	struct trace_fprobe *tf = to_trace_fprobe(ev);
255 
256 	if (event[0] != '\0' && strcmp(trace_probe_name(&tf->tp), event))
257 		return false;
258 
259 	if (system && strcmp(trace_probe_group_name(&tf->tp), system))
260 		return false;
261 
262 	return trace_fprobe_match_command_head(tf, argc, argv);
263 }
264 
trace_fprobe_is_registered(struct trace_fprobe * tf)265 static bool trace_fprobe_is_registered(struct trace_fprobe *tf)
266 {
267 	return fprobe_is_registered(&tf->fp);
268 }
269 
270 /*
271  * Note that we don't verify the fetch_insn code, since it does not come
272  * from user space.
273  */
274 static int
process_fetch_insn(struct fetch_insn * code,void * rec,void * edata,void * dest,void * base)275 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
276 		   void *dest, void *base)
277 {
278 	struct ftrace_regs *fregs = rec;
279 	unsigned long val;
280 	int ret;
281 
282 retry:
283 	/* 1st stage: get value from context */
284 	switch (code->op) {
285 	case FETCH_OP_STACK:
286 		val = ftrace_regs_get_kernel_stack_nth(fregs, code->param);
287 		break;
288 	case FETCH_OP_STACKP:
289 		val = ftrace_regs_get_stack_pointer(fregs);
290 		break;
291 	case FETCH_OP_RETVAL:
292 		val = ftrace_regs_get_return_value(fregs);
293 		break;
294 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
295 	case FETCH_OP_ARG:
296 		val = ftrace_regs_get_argument(fregs, code->param);
297 		break;
298 	case FETCH_OP_EDATA:
299 		val = *(unsigned long *)((unsigned long)edata + code->offset);
300 		break;
301 #endif
302 	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
303 		code++;
304 		goto retry;
305 	default:
306 		ret = process_common_fetch_insn(code, &val);
307 		if (ret < 0)
308 			return ret;
309 	}
310 	code++;
311 
312 	return process_fetch_insn_bottom(code, val, dest, base);
313 }
NOKPROBE_SYMBOL(process_fetch_insn)314 NOKPROBE_SYMBOL(process_fetch_insn)
315 
316 /* function entry handler */
317 static nokprobe_inline void
318 __fentry_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
319 		    struct ftrace_regs *fregs,
320 		    struct trace_event_file *trace_file)
321 {
322 	struct fentry_trace_entry_head *entry;
323 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
324 	struct trace_event_buffer fbuffer;
325 	int dsize;
326 
327 	if (WARN_ON_ONCE(call != trace_file->event_call))
328 		return;
329 
330 	if (trace_trigger_soft_disabled(trace_file))
331 		return;
332 
333 	dsize = __get_data_size(&tf->tp, fregs, NULL);
334 
335 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
336 					   sizeof(*entry) + tf->tp.size + dsize);
337 	if (!entry)
338 		return;
339 
340 	fbuffer.regs = ftrace_get_regs(fregs);
341 	entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
342 	entry->ip = entry_ip;
343 	store_trace_args(&entry[1], &tf->tp, fregs, NULL, sizeof(*entry), dsize);
344 
345 	trace_event_buffer_commit(&fbuffer);
346 }
347 
348 static void
fentry_trace_func(struct trace_fprobe * tf,unsigned long entry_ip,struct ftrace_regs * fregs)349 fentry_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
350 		  struct ftrace_regs *fregs)
351 {
352 	struct event_file_link *link;
353 
354 	trace_probe_for_each_link_rcu(link, &tf->tp)
355 		__fentry_trace_func(tf, entry_ip, fregs, link->file);
356 }
357 NOKPROBE_SYMBOL(fentry_trace_func);
358 
359 static nokprobe_inline
store_fprobe_entry_data(void * edata,struct trace_probe * tp,struct ftrace_regs * fregs)360 void store_fprobe_entry_data(void *edata, struct trace_probe *tp, struct ftrace_regs *fregs)
361 {
362 	struct probe_entry_arg *earg = tp->entry_arg;
363 	unsigned long val = 0;
364 	int i;
365 
366 	if (!earg)
367 		return;
368 
369 	for (i = 0; i < earg->size; i++) {
370 		struct fetch_insn *code = &earg->code[i];
371 
372 		switch (code->op) {
373 		case FETCH_OP_ARG:
374 			val = ftrace_regs_get_argument(fregs, code->param);
375 			break;
376 		case FETCH_OP_ST_EDATA:
377 			*(unsigned long *)((unsigned long)edata + code->offset) = val;
378 			break;
379 		case FETCH_OP_END:
380 			goto end;
381 		default:
382 			break;
383 		}
384 	}
385 end:
386 	return;
387 }
388 
389 /* 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)390 static int trace_fprobe_entry_handler(struct fprobe *fp, unsigned long entry_ip,
391 				unsigned long ret_ip, struct ftrace_regs *fregs,
392 				void *entry_data)
393 {
394 	struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp);
395 
396 	if (tf->tp.entry_arg)
397 		store_fprobe_entry_data(entry_data, &tf->tp, fregs);
398 
399 	return 0;
400 }
NOKPROBE_SYMBOL(trace_fprobe_entry_handler)401 NOKPROBE_SYMBOL(trace_fprobe_entry_handler)
402 
403 static nokprobe_inline void
404 __fexit_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
405 		   unsigned long ret_ip, struct ftrace_regs *fregs,
406 		   void *entry_data, struct trace_event_file *trace_file)
407 {
408 	struct fexit_trace_entry_head *entry;
409 	struct trace_event_buffer fbuffer;
410 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
411 	int dsize;
412 
413 	if (WARN_ON_ONCE(call != trace_file->event_call))
414 		return;
415 
416 	if (trace_trigger_soft_disabled(trace_file))
417 		return;
418 
419 	dsize = __get_data_size(&tf->tp, fregs, entry_data);
420 
421 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
422 					   sizeof(*entry) + tf->tp.size + dsize);
423 	if (!entry)
424 		return;
425 
426 	fbuffer.regs = ftrace_get_regs(fregs);
427 	entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
428 	entry->func = entry_ip;
429 	entry->ret_ip = ret_ip;
430 	store_trace_args(&entry[1], &tf->tp, fregs, entry_data, sizeof(*entry), dsize);
431 
432 	trace_event_buffer_commit(&fbuffer);
433 }
434 
435 static void
fexit_trace_func(struct trace_fprobe * tf,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)436 fexit_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
437 		 unsigned long ret_ip, struct ftrace_regs *fregs, void *entry_data)
438 {
439 	struct event_file_link *link;
440 
441 	trace_probe_for_each_link_rcu(link, &tf->tp)
442 		__fexit_trace_func(tf, entry_ip, ret_ip, fregs, entry_data, link->file);
443 }
444 NOKPROBE_SYMBOL(fexit_trace_func);
445 
446 #ifdef CONFIG_PERF_EVENTS
447 
fentry_perf_func(struct trace_fprobe * tf,unsigned long entry_ip,struct ftrace_regs * fregs)448 static int fentry_perf_func(struct trace_fprobe *tf, unsigned long entry_ip,
449 			    struct ftrace_regs *fregs)
450 {
451 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
452 	struct fentry_trace_entry_head *entry;
453 	struct hlist_head *head;
454 	int size, __size, dsize;
455 	struct pt_regs *regs;
456 	int rctx;
457 
458 	head = this_cpu_ptr(call->perf_events);
459 	if (hlist_empty(head))
460 		return 0;
461 
462 	dsize = __get_data_size(&tf->tp, fregs, NULL);
463 	__size = sizeof(*entry) + tf->tp.size + dsize;
464 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
465 	size -= sizeof(u32);
466 
467 	entry = perf_trace_buf_alloc(size, &regs, &rctx);
468 	if (!entry)
469 		return 0;
470 
471 	regs = ftrace_fill_perf_regs(fregs, regs);
472 
473 	entry->ip = entry_ip;
474 	store_trace_args(&entry[1], &tf->tp, fregs, NULL, sizeof(*entry), dsize);
475 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
476 			      head, NULL);
477 	return 0;
478 }
479 NOKPROBE_SYMBOL(fentry_perf_func);
480 
481 static void
fexit_perf_func(struct trace_fprobe * tf,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)482 fexit_perf_func(struct trace_fprobe *tf, unsigned long entry_ip,
483 		unsigned long ret_ip, struct ftrace_regs *fregs,
484 		void *entry_data)
485 {
486 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
487 	struct fexit_trace_entry_head *entry;
488 	struct hlist_head *head;
489 	int size, __size, dsize;
490 	struct pt_regs *regs;
491 	int rctx;
492 
493 	head = this_cpu_ptr(call->perf_events);
494 	if (hlist_empty(head))
495 		return;
496 
497 	dsize = __get_data_size(&tf->tp, fregs, entry_data);
498 	__size = sizeof(*entry) + tf->tp.size + dsize;
499 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
500 	size -= sizeof(u32);
501 
502 	entry = perf_trace_buf_alloc(size, &regs, &rctx);
503 	if (!entry)
504 		return;
505 
506 	regs = ftrace_fill_perf_regs(fregs, regs);
507 
508 	entry->func = entry_ip;
509 	entry->ret_ip = ret_ip;
510 	store_trace_args(&entry[1], &tf->tp, fregs, entry_data, sizeof(*entry), dsize);
511 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
512 			      head, NULL);
513 }
514 NOKPROBE_SYMBOL(fexit_perf_func);
515 #endif	/* CONFIG_PERF_EVENTS */
516 
fentry_dispatcher(struct fprobe * fp,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)517 static int fentry_dispatcher(struct fprobe *fp, unsigned long entry_ip,
518 			     unsigned long ret_ip, struct ftrace_regs *fregs,
519 			     void *entry_data)
520 {
521 	struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp);
522 	unsigned int flags = trace_probe_load_flag(&tf->tp);
523 	int ret = 0;
524 
525 	if (flags & TP_FLAG_TRACE)
526 		fentry_trace_func(tf, entry_ip, fregs);
527 
528 #ifdef CONFIG_PERF_EVENTS
529 	if (flags & TP_FLAG_PROFILE)
530 		ret = fentry_perf_func(tf, entry_ip, fregs);
531 #endif
532 	return ret;
533 }
534 NOKPROBE_SYMBOL(fentry_dispatcher);
535 
fexit_dispatcher(struct fprobe * fp,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)536 static void fexit_dispatcher(struct fprobe *fp, unsigned long entry_ip,
537 			     unsigned long ret_ip, struct ftrace_regs *fregs,
538 			     void *entry_data)
539 {
540 	struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp);
541 	unsigned int flags = trace_probe_load_flag(&tf->tp);
542 
543 	if (flags & TP_FLAG_TRACE)
544 		fexit_trace_func(tf, entry_ip, ret_ip, fregs, entry_data);
545 #ifdef CONFIG_PERF_EVENTS
546 	if (flags & TP_FLAG_PROFILE)
547 		fexit_perf_func(tf, entry_ip, ret_ip, fregs, entry_data);
548 #endif
549 }
550 NOKPROBE_SYMBOL(fexit_dispatcher);
551 
free_trace_fprobe(struct trace_fprobe * tf)552 static void free_trace_fprobe(struct trace_fprobe *tf)
553 {
554 	if (tf) {
555 		trace_probe_cleanup(&tf->tp);
556 		if (tf->tuser)
557 			tracepoint_user_put(tf->tuser);
558 		kfree(tf->symbol);
559 		kfree(tf);
560 	}
561 }
562 
563 /* Since alloc_trace_fprobe() can return error, check the pointer is ERR too. */
564 DEFINE_FREE(free_trace_fprobe, struct trace_fprobe *, if (!IS_ERR_OR_NULL(_T)) free_trace_fprobe(_T))
565 
566 /*
567  * Allocate new trace_probe and initialize it (including fprobe).
568  */
alloc_trace_fprobe(const char * group,const char * event,const char * symbol,int nargs,bool is_return,bool is_tracepoint)569 static struct trace_fprobe *alloc_trace_fprobe(const char *group,
570 					       const char *event,
571 					       const char *symbol,
572 					       int nargs, bool is_return,
573 					       bool is_tracepoint)
574 {
575 	struct trace_fprobe *tf __free(free_trace_fprobe) = NULL;
576 	int ret = -ENOMEM;
577 
578 	tf = kzalloc_flex(*tf, tp.args, nargs);
579 	if (!tf)
580 		return ERR_PTR(ret);
581 
582 	tf->symbol = kstrdup(symbol, GFP_KERNEL);
583 	if (!tf->symbol)
584 		return ERR_PTR(-ENOMEM);
585 
586 	if (is_return)
587 		tf->fp.exit_handler = fexit_dispatcher;
588 	else
589 		tf->fp.entry_handler = fentry_dispatcher;
590 
591 	tf->tprobe = is_tracepoint;
592 
593 	ret = trace_probe_init(&tf->tp, event, group, false, nargs);
594 	if (ret < 0)
595 		return ERR_PTR(ret);
596 
597 	dyn_event_init(&tf->devent, &trace_fprobe_ops);
598 	return_ptr(tf);
599 }
600 
find_trace_fprobe(const char * event,const char * group)601 static struct trace_fprobe *find_trace_fprobe(const char *event,
602 					      const char *group)
603 {
604 	struct dyn_event *pos;
605 	struct trace_fprobe *tf;
606 
607 	for_each_trace_fprobe(tf, pos)
608 		if (strcmp(trace_probe_name(&tf->tp), event) == 0 &&
609 		    strcmp(trace_probe_group_name(&tf->tp), group) == 0)
610 			return tf;
611 	return NULL;
612 }
613 
614 /* Event entry printers */
615 static enum print_line_t
print_fentry_event(struct trace_iterator * iter,int flags,struct trace_event * event)616 print_fentry_event(struct trace_iterator *iter, int flags,
617 		   struct trace_event *event)
618 {
619 	struct fentry_trace_entry_head *field;
620 	struct trace_seq *s = &iter->seq;
621 	struct trace_probe *tp;
622 
623 	field = (struct fentry_trace_entry_head *)iter->ent;
624 	tp = trace_probe_primary_from_call(
625 		container_of(event, struct trace_event_call, event));
626 	if (WARN_ON_ONCE(!tp))
627 		goto out;
628 
629 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
630 
631 	if (!seq_print_ip_sym_offset(s, field->ip, flags))
632 		goto out;
633 
634 	trace_seq_putc(s, ')');
635 
636 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
637 			     (u8 *)&field[1], field) < 0)
638 		goto out;
639 
640 	trace_seq_putc(s, '\n');
641  out:
642 	return trace_handle_return(s);
643 }
644 
645 static enum print_line_t
print_fexit_event(struct trace_iterator * iter,int flags,struct trace_event * event)646 print_fexit_event(struct trace_iterator *iter, int flags,
647 		  struct trace_event *event)
648 {
649 	struct fexit_trace_entry_head *field;
650 	struct trace_seq *s = &iter->seq;
651 	struct trace_probe *tp;
652 
653 	field = (struct fexit_trace_entry_head *)iter->ent;
654 	tp = trace_probe_primary_from_call(
655 		container_of(event, struct trace_event_call, event));
656 	if (WARN_ON_ONCE(!tp))
657 		goto out;
658 
659 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
660 
661 	if (!seq_print_ip_sym_offset(s, field->ret_ip, flags))
662 		goto out;
663 
664 	trace_seq_puts(s, " <- ");
665 
666 	if (!seq_print_ip_sym_no_offset(s, field->func, flags))
667 		goto out;
668 
669 	trace_seq_putc(s, ')');
670 
671 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
672 			     (u8 *)&field[1], field) < 0)
673 		goto out;
674 
675 	trace_seq_putc(s, '\n');
676 
677  out:
678 	return trace_handle_return(s);
679 }
680 
fentry_event_define_fields(struct trace_event_call * event_call)681 static int fentry_event_define_fields(struct trace_event_call *event_call)
682 {
683 	int ret;
684 	struct fentry_trace_entry_head field;
685 	struct trace_probe *tp;
686 
687 	tp = trace_probe_primary_from_call(event_call);
688 	if (WARN_ON_ONCE(!tp))
689 		return -ENOENT;
690 
691 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
692 
693 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
694 }
695 
fexit_event_define_fields(struct trace_event_call * event_call)696 static int fexit_event_define_fields(struct trace_event_call *event_call)
697 {
698 	int ret;
699 	struct fexit_trace_entry_head field;
700 	struct trace_probe *tp;
701 
702 	tp = trace_probe_primary_from_call(event_call);
703 	if (WARN_ON_ONCE(!tp))
704 		return -ENOENT;
705 
706 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
707 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
708 
709 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
710 }
711 
712 static struct trace_event_functions fentry_funcs = {
713 	.trace		= print_fentry_event
714 };
715 
716 static struct trace_event_functions fexit_funcs = {
717 	.trace		= print_fexit_event
718 };
719 
720 static struct trace_event_fields fentry_fields_array[] = {
721 	{ .type = TRACE_FUNCTION_TYPE,
722 	  .define_fields = fentry_event_define_fields },
723 	{}
724 };
725 
726 static struct trace_event_fields fexit_fields_array[] = {
727 	{ .type = TRACE_FUNCTION_TYPE,
728 	  .define_fields = fexit_event_define_fields },
729 	{}
730 };
731 
732 static int fprobe_register(struct trace_event_call *event,
733 			   enum trace_reg type, void *data);
734 
init_trace_event_call(struct trace_fprobe * tf)735 static inline void init_trace_event_call(struct trace_fprobe *tf)
736 {
737 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
738 
739 	if (trace_fprobe_is_return(tf)) {
740 		call->event.funcs = &fexit_funcs;
741 		call->class->fields_array = fexit_fields_array;
742 	} else {
743 		call->event.funcs = &fentry_funcs;
744 		call->class->fields_array = fentry_fields_array;
745 	}
746 
747 	call->flags = TRACE_EVENT_FL_FPROBE;
748 	call->class->reg = fprobe_register;
749 }
750 
register_fprobe_event(struct trace_fprobe * tf)751 static int register_fprobe_event(struct trace_fprobe *tf)
752 {
753 	init_trace_event_call(tf);
754 
755 	return trace_probe_register_event_call(&tf->tp);
756 }
757 
unregister_fprobe_event(struct trace_fprobe * tf)758 static int unregister_fprobe_event(struct trace_fprobe *tf)
759 {
760 	return trace_probe_unregister_event_call(&tf->tp);
761 }
762 
__register_tracepoint_fprobe(struct trace_fprobe * tf)763 static int __register_tracepoint_fprobe(struct trace_fprobe *tf)
764 {
765 	struct tracepoint_user *tuser __free(tuser_put) = NULL;
766 	struct module *mod __free(module_put) = NULL;
767 	unsigned long ip;
768 	int ret;
769 
770 	if (WARN_ON_ONCE(tf->tuser))
771 		return -EINVAL;
772 
773 	/* If the tracepoint is in a module, it must be locked in this function. */
774 	tuser = tracepoint_user_find_get(tf->symbol, &mod);
775 	/* This tracepoint is not loaded yet */
776 	if (IS_ERR(tuser))
777 		return PTR_ERR(tuser);
778 	if (!tuser)
779 		return -ENOMEM;
780 
781 	/* Register fprobe only if the tracepoint is loaded. */
782 	if (tuser->tpoint) {
783 		ip = tracepoint_user_ip(tuser);
784 		if (WARN_ON_ONCE(!ip))
785 			return -ENOENT;
786 
787 		ret = register_fprobe_ips(&tf->fp, &ip, 1);
788 		if (ret < 0)
789 			return ret;
790 	}
791 
792 	tf->tuser = no_free_ptr(tuser);
793 	return 0;
794 }
795 
796 /* Returns an error if the target function is not available, or 0 */
trace_fprobe_verify_target(struct trace_fprobe * tf)797 static int trace_fprobe_verify_target(struct trace_fprobe *tf)
798 {
799 	int ret;
800 
801 	/* Tracepoint should have a stub function. */
802 	if (trace_fprobe_is_tracepoint(tf))
803 		return 0;
804 
805 	/*
806 	 * Note: since we don't lock the module, even if this succeeded,
807 	 * register_fprobe() later can fail.
808 	 */
809 	ret = fprobe_count_ips_from_filter(tf->symbol, NULL);
810 	return (ret < 0) ? ret : 0;
811 }
812 
813 /* Internal register function - just handle fprobe and flags */
__register_trace_fprobe(struct trace_fprobe * tf)814 static int __register_trace_fprobe(struct trace_fprobe *tf)
815 {
816 	int i, ret;
817 
818 	/* Should we need new LOCKDOWN flag for fprobe? */
819 	ret = security_locked_down(LOCKDOWN_KPROBES);
820 	if (ret)
821 		return ret;
822 
823 	if (trace_fprobe_is_registered(tf))
824 		return -EINVAL;
825 
826 	for (i = 0; i < tf->tp.nr_args; i++) {
827 		ret = traceprobe_update_arg(&tf->tp.args[i]);
828 		if (ret)
829 			return ret;
830 	}
831 
832 	tf->fp.flags &= ~FPROBE_FL_DISABLED;
833 
834 	if (trace_fprobe_is_tracepoint(tf))
835 		return __register_tracepoint_fprobe(tf);
836 
837 	/* TODO: handle filter, nofilter or symbol list */
838 	return register_fprobe(&tf->fp, tf->symbol, NULL);
839 }
840 
841 /* Internal unregister function - just handle fprobe and flags */
__unregister_trace_fprobe(struct trace_fprobe * tf)842 static void __unregister_trace_fprobe(struct trace_fprobe *tf)
843 {
844 	if (trace_fprobe_is_registered(tf))
845 		unregister_fprobe(&tf->fp);
846 	if (tf->tuser) {
847 		tracepoint_user_put(tf->tuser);
848 		tf->tuser = NULL;
849 	}
850 }
851 
852 /* TODO: make this trace_*probe common function */
853 /* Unregister a trace_probe and probe_event */
unregister_trace_fprobe(struct trace_fprobe * tf)854 static int unregister_trace_fprobe(struct trace_fprobe *tf)
855 {
856 	/* If other probes are on the event, just unregister fprobe */
857 	if (trace_probe_has_sibling(&tf->tp))
858 		goto unreg;
859 
860 	/* Enabled event can not be unregistered */
861 	if (trace_probe_is_enabled(&tf->tp))
862 		return -EBUSY;
863 
864 	/* If there's a reference to the dynamic event */
865 	if (trace_event_dyn_busy(trace_probe_event_call(&tf->tp)))
866 		return -EBUSY;
867 
868 	/* Will fail if probe is being used by ftrace or perf */
869 	if (unregister_fprobe_event(tf))
870 		return -EBUSY;
871 
872 unreg:
873 	__unregister_trace_fprobe(tf);
874 	dyn_event_remove(&tf->devent);
875 	trace_probe_unlink(&tf->tp);
876 
877 	return 0;
878 }
879 
trace_fprobe_has_same_fprobe(struct trace_fprobe * orig,struct trace_fprobe * comp)880 static bool trace_fprobe_has_same_fprobe(struct trace_fprobe *orig,
881 					 struct trace_fprobe *comp)
882 {
883 	struct trace_probe_event *tpe = orig->tp.event;
884 	int i;
885 
886 	list_for_each_entry(orig, &tpe->probes, tp.list) {
887 		if (strcmp(trace_fprobe_symbol(orig),
888 			   trace_fprobe_symbol(comp)))
889 			continue;
890 
891 		/*
892 		 * trace_probe_compare_arg_type() ensured that nr_args and
893 		 * each argument name and type are same. Let's compare comm.
894 		 */
895 		for (i = 0; i < orig->tp.nr_args; i++) {
896 			if (strcmp(orig->tp.args[i].comm,
897 				   comp->tp.args[i].comm))
898 				break;
899 		}
900 
901 		if (i == orig->tp.nr_args)
902 			return true;
903 	}
904 
905 	return false;
906 }
907 
append_trace_fprobe_event(struct trace_fprobe * tf,struct trace_fprobe * to)908 static int append_trace_fprobe_event(struct trace_fprobe *tf, struct trace_fprobe *to)
909 {
910 	int ret;
911 
912 	if (trace_fprobe_is_return(tf) != trace_fprobe_is_return(to) ||
913 	    trace_fprobe_is_tracepoint(tf) != trace_fprobe_is_tracepoint(to)) {
914 		trace_probe_log_set_index(0);
915 		trace_probe_log_err(0, DIFF_PROBE_TYPE);
916 		return -EEXIST;
917 	}
918 	ret = trace_probe_compare_arg_type(&tf->tp, &to->tp);
919 	if (ret) {
920 		/* Note that argument starts index = 2 */
921 		trace_probe_log_set_index(ret + 1);
922 		trace_probe_log_err(0, DIFF_ARG_TYPE);
923 		return -EEXIST;
924 	}
925 	if (trace_fprobe_has_same_fprobe(to, tf)) {
926 		trace_probe_log_set_index(0);
927 		trace_probe_log_err(0, SAME_PROBE);
928 		return -EEXIST;
929 	}
930 
931 	/* Append to existing event */
932 	ret = trace_probe_append(&tf->tp, &to->tp);
933 	if (ret)
934 		return ret;
935 
936 	ret = trace_fprobe_verify_target(tf);
937 	if (ret)
938 		trace_probe_unlink(&tf->tp);
939 	else
940 		dyn_event_add(&tf->devent, trace_probe_event_call(&tf->tp));
941 
942 	return ret;
943 }
944 
945 /* Register a trace_probe and probe_event, and check the fprobe is available. */
register_trace_fprobe_event(struct trace_fprobe * tf)946 static int register_trace_fprobe_event(struct trace_fprobe *tf)
947 {
948 	struct trace_fprobe *old_tf;
949 	int ret;
950 
951 	guard(mutex)(&event_mutex);
952 
953 	old_tf = find_trace_fprobe(trace_probe_name(&tf->tp),
954 				   trace_probe_group_name(&tf->tp));
955 	if (old_tf)
956 		return append_trace_fprobe_event(tf, old_tf);
957 
958 	/* Register new event */
959 	ret = register_fprobe_event(tf);
960 	if (ret) {
961 		if (ret == -EEXIST) {
962 			trace_probe_log_set_index(0);
963 			trace_probe_log_err(0, EVENT_EXIST);
964 		} else
965 			pr_warn("Failed to register probe event(%d)\n", ret);
966 		return ret;
967 	}
968 
969 	/* Verify fprobe is sane. */
970 	ret = trace_fprobe_verify_target(tf);
971 	if (ret < 0)
972 		unregister_fprobe_event(tf);
973 	else
974 		dyn_event_add(&tf->devent, trace_probe_event_call(&tf->tp));
975 
976 	return ret;
977 }
978 
979 struct __find_tracepoint_cb_data {
980 	const char *tp_name;
981 	struct tracepoint *tpoint;
982 	struct module *mod;
983 };
984 
__find_tracepoint_module_cb(struct tracepoint * tp,struct module * mod,void * priv)985 static void __find_tracepoint_module_cb(struct tracepoint *tp, struct module *mod, void *priv)
986 {
987 	struct __find_tracepoint_cb_data *data = priv;
988 
989 	if (!data->tpoint && !strcmp(data->tp_name, tp->name)) {
990 		/* If module is not specified, try getting module refcount. */
991 		if (!data->mod && mod) {
992 			/* If failed to get refcount, ignore this tracepoint. */
993 			if (!try_module_get(mod))
994 				return;
995 
996 			data->mod = mod;
997 		}
998 		data->tpoint = tp;
999 	}
1000 }
1001 
__find_tracepoint_cb(struct tracepoint * tp,void * priv)1002 static void __find_tracepoint_cb(struct tracepoint *tp, void *priv)
1003 {
1004 	struct __find_tracepoint_cb_data *data = priv;
1005 
1006 	if (!data->tpoint && !strcmp(data->tp_name, tp->name))
1007 		data->tpoint = tp;
1008 }
1009 
1010 /*
1011  * Find a tracepoint from kernel and module. If the tracepoint is on the module,
1012  * the module's refcount is incremented and returned as *@tp_mod. Thus, if it is
1013  * not NULL, caller must call module_put(*tp_mod) after used the tracepoint.
1014  */
find_tracepoint(const char * tp_name,struct module ** tp_mod)1015 static struct tracepoint *find_tracepoint(const char *tp_name,
1016 					  struct module **tp_mod)
1017 {
1018 	struct __find_tracepoint_cb_data data = {
1019 		.tp_name = tp_name,
1020 		.mod = NULL,
1021 	};
1022 
1023 	for_each_kernel_tracepoint(__find_tracepoint_cb, &data);
1024 
1025 	if (!data.tpoint && IS_ENABLED(CONFIG_MODULES)) {
1026 		for_each_module_tracepoint(__find_tracepoint_module_cb, &data);
1027 		*tp_mod = data.mod;
1028 	}
1029 
1030 	return data.tpoint;
1031 }
1032 
1033 #ifdef CONFIG_MODULES
1034 /*
1035  * Find a tracepoint from specified module. In this case, this does not get the
1036  * module's refcount. The caller must ensure the module is not freed.
1037  */
find_tracepoint_in_module(struct module * mod,const char * tp_name)1038 static struct tracepoint *find_tracepoint_in_module(struct module *mod,
1039 						    const char *tp_name)
1040 {
1041 	struct __find_tracepoint_cb_data data = {
1042 		.tp_name = tp_name,
1043 		.mod = mod,
1044 	};
1045 
1046 	for_each_tracepoint_in_module(mod, __find_tracepoint_module_cb, &data);
1047 	return data.tpoint;
1048 }
1049 
1050 /* These are CONFIG_MODULES=y specific functions. */
tracepoint_user_within_module(struct tracepoint_user * tuser,struct module * mod)1051 static bool tracepoint_user_within_module(struct tracepoint_user *tuser,
1052 					  struct module *mod)
1053 {
1054 	return within_module(tracepoint_user_ip(tuser), mod);
1055 }
1056 
tracepoint_user_register_again(struct tracepoint_user * tuser,struct tracepoint * tpoint)1057 static int tracepoint_user_register_again(struct tracepoint_user *tuser,
1058 					  struct tracepoint *tpoint)
1059 {
1060 	tuser->tpoint = tpoint;
1061 	return tracepoint_user_register(tuser);
1062 }
1063 
tracepoint_user_unregister_clear(struct tracepoint_user * tuser)1064 static void tracepoint_user_unregister_clear(struct tracepoint_user *tuser)
1065 {
1066 	tracepoint_user_unregister(tuser);
1067 	tuser->tpoint = NULL;
1068 }
1069 
1070 /* module callback for tracepoint_user */
__tracepoint_probe_module_cb(struct notifier_block * self,unsigned long val,void * data)1071 static int __tracepoint_probe_module_cb(struct notifier_block *self,
1072 					unsigned long val, void *data)
1073 {
1074 	struct tp_module *tp_mod = data;
1075 	struct tracepoint_user *tuser;
1076 	struct tracepoint *tpoint;
1077 
1078 	if (val != MODULE_STATE_GOING && val != MODULE_STATE_COMING)
1079 		return NOTIFY_DONE;
1080 
1081 	mutex_lock(&tracepoint_user_mutex);
1082 	for_each_tracepoint_user(tuser) {
1083 		if (val == MODULE_STATE_COMING) {
1084 			/* This is not a tracepoint in this module. Skip it. */
1085 			tpoint = find_tracepoint_in_module(tp_mod->mod, tuser->name);
1086 			if (!tpoint)
1087 				continue;
1088 			WARN_ON_ONCE(tracepoint_user_register_again(tuser, tpoint));
1089 		} else if (val == MODULE_STATE_GOING &&
1090 			  tracepoint_user_within_module(tuser, tp_mod->mod)) {
1091 			/* Unregister all tracepoint_user in this module. */
1092 			tracepoint_user_unregister_clear(tuser);
1093 		}
1094 	}
1095 	mutex_unlock(&tracepoint_user_mutex);
1096 
1097 	return NOTIFY_DONE;
1098 }
1099 
1100 static struct notifier_block tracepoint_module_nb = {
1101 	.notifier_call = __tracepoint_probe_module_cb,
1102 };
1103 
1104 /* module callback for tprobe events */
__tprobe_event_module_cb(struct notifier_block * self,unsigned long val,void * data)1105 static int __tprobe_event_module_cb(struct notifier_block *self,
1106 				     unsigned long val, void *data)
1107 {
1108 	struct trace_fprobe *tf;
1109 	struct dyn_event *pos;
1110 	struct module *mod = data;
1111 
1112 	if (val != MODULE_STATE_GOING && val != MODULE_STATE_COMING)
1113 		return NOTIFY_DONE;
1114 
1115 	mutex_lock(&event_mutex);
1116 	for_each_trace_fprobe(tf, pos) {
1117 		/* Skip fprobe and disabled tprobe events. */
1118 		if (!trace_fprobe_is_tracepoint(tf) || !tf->tuser)
1119 			continue;
1120 
1121 		/* Before this notification, tracepoint notifier has already done. */
1122 		if (val == MODULE_STATE_COMING &&
1123 		    tracepoint_user_within_module(tf->tuser, mod)) {
1124 			unsigned long ip = tracepoint_user_ip(tf->tuser);
1125 
1126 			WARN_ON_ONCE(register_fprobe_ips(&tf->fp, &ip, 1));
1127 		} else if (val == MODULE_STATE_GOING &&
1128 			   /*
1129 			    * tracepoint_user_within_module() does not work here because
1130 			    * tracepoint_user is already unregistered and cleared tpoint.
1131 			    * Instead, checking whether the fprobe is registered but
1132 			    * tpoint is cleared(unregistered). Such unbalance probes
1133 			    * must be adjusted anyway.
1134 			    */
1135 			    trace_fprobe_is_registered(tf) &&
1136 			    !tf->tuser->tpoint) {
1137 			unregister_fprobe(&tf->fp);
1138 		}
1139 	}
1140 	mutex_unlock(&event_mutex);
1141 
1142 	return NOTIFY_DONE;
1143 }
1144 
1145 /* NOTE: this must be called after tracepoint callback */
1146 static struct notifier_block tprobe_event_module_nb = {
1147 	.notifier_call = __tprobe_event_module_cb,
1148 	/* Make sure this is later than tracepoint module notifier. */
1149 	.priority = -10,
1150 };
1151 #endif /* CONFIG_MODULES */
1152 
parse_symbol_and_return(int argc,const char * argv[],char ** symbol,bool * is_return,bool is_tracepoint)1153 static int parse_symbol_and_return(int argc, const char *argv[],
1154 				   char **symbol, bool *is_return,
1155 				   bool is_tracepoint)
1156 {
1157 	char *tmp = strchr(argv[1], '%');
1158 	int i;
1159 
1160 	if (tmp) {
1161 		int len = tmp - argv[1];
1162 
1163 		if (!is_tracepoint && !strcmp(tmp, "%return")) {
1164 			*is_return = true;
1165 		} else {
1166 			trace_probe_log_err(len, BAD_ADDR_SUFFIX);
1167 			return -EINVAL;
1168 		}
1169 		*symbol = kmemdup_nul(argv[1], len, GFP_KERNEL);
1170 	} else
1171 		*symbol = kstrdup(argv[1], GFP_KERNEL);
1172 	if (!*symbol)
1173 		return -ENOMEM;
1174 
1175 	if (*is_return)
1176 		return 0;
1177 
1178 	if (is_tracepoint) {
1179 		tmp = *symbol;
1180 		while (*tmp && (isalnum(*tmp) || *tmp == '_'))
1181 			tmp++;
1182 		if (*tmp) {
1183 			/* find a wrong character. */
1184 			trace_probe_log_err(tmp - *symbol, BAD_TP_NAME);
1185 			kfree(*symbol);
1186 			*symbol = NULL;
1187 			return -EINVAL;
1188 		}
1189 	}
1190 
1191 	/* If there is $retval, this should be a return fprobe. */
1192 	for (i = 2; i < argc; i++) {
1193 		tmp = strstr(argv[i], "$retval");
1194 		if (tmp && !isalnum(tmp[7]) && tmp[7] != '_') {
1195 			if (is_tracepoint) {
1196 				trace_probe_log_set_index(i);
1197 				trace_probe_log_err(tmp - argv[i], RETVAL_ON_PROBE);
1198 				kfree(*symbol);
1199 				*symbol = NULL;
1200 				return -EINVAL;
1201 			}
1202 			*is_return = true;
1203 			break;
1204 		}
1205 	}
1206 	return 0;
1207 }
1208 
trace_fprobe_create_internal(int argc,const char * argv[],struct traceprobe_parse_context * ctx)1209 static int trace_fprobe_create_internal(int argc, const char *argv[],
1210 					struct traceprobe_parse_context *ctx)
1211 {
1212 	/*
1213 	 * Argument syntax:
1214 	 *  - Add fentry probe:
1215 	 *      f[:[GRP/][EVENT]] [MOD:]KSYM [FETCHARGS]
1216 	 *  - Add fexit probe:
1217 	 *      f[N][:[GRP/][EVENT]] [MOD:]KSYM%return [FETCHARGS]
1218 	 *  - Add tracepoint probe:
1219 	 *      t[:[GRP/][EVENT]] TRACEPOINT [FETCHARGS]
1220 	 *
1221 	 * Fetch args:
1222 	 *  $retval	: fetch return value
1223 	 *  $stack	: fetch stack address
1224 	 *  $stackN	: fetch Nth entry of stack (N:0-)
1225 	 *  $argN	: fetch Nth argument (N:1-)
1226 	 *  $comm       : fetch current task comm
1227 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
1228 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
1229 	 * Dereferencing memory fetch:
1230 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
1231 	 * Alias name of args:
1232 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
1233 	 * Type of args:
1234 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
1235 	 */
1236 	struct trace_fprobe *tf __free(free_trace_fprobe) = NULL;
1237 	const char *event = NULL, *group = FPROBE_EVENT_SYSTEM;
1238 	struct module *mod __free(module_put) = NULL;
1239 	const char **new_argv __free(kfree) = NULL;
1240 	char *symbol __free(kfree) = NULL;
1241 	char *ebuf __free(kfree) = NULL;
1242 	char *gbuf __free(kfree) = NULL;
1243 	char *sbuf __free(kfree) = NULL;
1244 	char *abuf __free(kfree) = NULL;
1245 	char *dbuf __free(kfree) = NULL;
1246 	int i, new_argc = 0, ret = 0;
1247 	bool is_tracepoint = false;
1248 	bool is_return = false;
1249 
1250 	if ((argv[0][0] != 'f' && argv[0][0] != 't') || argc < 2)
1251 		return -ECANCELED;
1252 
1253 	if (argv[0][0] == 't') {
1254 		is_tracepoint = true;
1255 		group = TRACEPOINT_EVENT_SYSTEM;
1256 	}
1257 
1258 	if (argv[0][1] != '\0') {
1259 		if (argv[0][1] != ':') {
1260 			trace_probe_log_set_index(0);
1261 			trace_probe_log_err(1, BAD_MAXACT);
1262 			return -EINVAL;
1263 		}
1264 		event = &argv[0][2];
1265 	}
1266 
1267 	trace_probe_log_set_index(1);
1268 
1269 	/* a symbol(or tracepoint) must be specified */
1270 	ret = parse_symbol_and_return(argc, argv, &symbol, &is_return, is_tracepoint);
1271 	if (ret < 0)
1272 		return -EINVAL;
1273 
1274 	trace_probe_log_set_index(0);
1275 	if (event) {
1276 		gbuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
1277 		if (!gbuf)
1278 			return -ENOMEM;
1279 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
1280 						  event - argv[0]);
1281 		if (ret)
1282 			return -EINVAL;
1283 	}
1284 
1285 	if (!event) {
1286 		ebuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
1287 		if (!ebuf)
1288 			return -ENOMEM;
1289 		/* Make a new event name */
1290 		if (is_tracepoint)
1291 			snprintf(ebuf, MAX_EVENT_NAME_LEN, "%s%s",
1292 				 isdigit(*symbol) ? "_" : "", symbol);
1293 		else
1294 			snprintf(ebuf, MAX_EVENT_NAME_LEN, "%s__%s", symbol,
1295 				 is_return ? "exit" : "entry");
1296 		sanitize_event_name(ebuf);
1297 		event = ebuf;
1298 	}
1299 
1300 	if (is_return)
1301 		ctx->flags |= TPARG_FL_RETURN;
1302 	else
1303 		ctx->flags |= TPARG_FL_FENTRY;
1304 
1305 	ctx->funcname = NULL;
1306 	if (is_tracepoint) {
1307 		/* Get tracepoint and lock its module until the end of the registration. */
1308 		struct tracepoint *tpoint;
1309 
1310 		ctx->flags |= TPARG_FL_TPOINT;
1311 		mod = NULL;
1312 		tpoint = find_tracepoint(symbol, &mod);
1313 		if (tpoint) {
1314 			sbuf = kmalloc(KSYM_NAME_LEN, GFP_KERNEL);
1315 			if (!sbuf)
1316 				return -ENOMEM;
1317 			ctx->funcname = kallsyms_lookup((unsigned long)tpoint->probestub,
1318 							NULL, NULL, NULL, sbuf);
1319 		}
1320 	}
1321 	if (!ctx->funcname)
1322 		ctx->funcname = symbol;
1323 
1324 	abuf = kmalloc(MAX_BTF_ARGS_LEN, GFP_KERNEL);
1325 	if (!abuf)
1326 		return -ENOMEM;
1327 	argc -= 2; argv += 2;
1328 	new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
1329 					       abuf, MAX_BTF_ARGS_LEN, ctx);
1330 	if (IS_ERR(new_argv))
1331 		return PTR_ERR(new_argv);
1332 	if (new_argv) {
1333 		argc = new_argc;
1334 		argv = new_argv;
1335 	}
1336 	if (argc > MAX_TRACE_ARGS) {
1337 		trace_probe_log_set_index(2);
1338 		trace_probe_log_err(0, TOO_MANY_ARGS);
1339 		return -E2BIG;
1340 	}
1341 
1342 	ret = traceprobe_expand_dentry_args(argc, argv, &dbuf);
1343 	if (ret)
1344 		return ret;
1345 
1346 	/* setup a probe */
1347 	tf = alloc_trace_fprobe(group, event, symbol, argc, is_return, is_tracepoint);
1348 	if (IS_ERR(tf)) {
1349 		ret = PTR_ERR(tf);
1350 		/* This must return -ENOMEM, else there is a bug */
1351 		WARN_ON_ONCE(ret != -ENOMEM);
1352 		return ret;
1353 	}
1354 
1355 	/* parse arguments */
1356 	for (i = 0; i < argc; i++) {
1357 		trace_probe_log_set_index(i + 2);
1358 		ctx->offset = 0;
1359 		ret = traceprobe_parse_probe_arg(&tf->tp, i, argv[i], ctx);
1360 		if (ret)
1361 			return ret;	/* This can be -ENOMEM */
1362 	}
1363 
1364 	if (is_return && tf->tp.entry_arg) {
1365 		tf->fp.entry_handler = trace_fprobe_entry_handler;
1366 		tf->fp.entry_data_size = traceprobe_get_entry_data_size(&tf->tp);
1367 		if (ALIGN(tf->fp.entry_data_size, sizeof(long)) > MAX_FPROBE_DATA_SIZE) {
1368 			trace_probe_log_set_index(2);
1369 			trace_probe_log_err(0, TOO_MANY_EARGS);
1370 			return -E2BIG;
1371 		}
1372 	}
1373 
1374 	ret = traceprobe_set_print_fmt(&tf->tp,
1375 			is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL);
1376 	if (ret < 0)
1377 		return ret;
1378 
1379 	ret = register_trace_fprobe_event(tf);
1380 	if (ret) {
1381 		trace_probe_log_set_index(1);
1382 		if (ret == -EILSEQ)
1383 			trace_probe_log_err(0, BAD_INSN_BNDRY);
1384 		else if (ret == -ENOENT)
1385 			trace_probe_log_err(0, BAD_PROBE_ADDR);
1386 		else if (ret != -ENOMEM && ret != -EEXIST)
1387 			trace_probe_log_err(0, FAIL_REG_PROBE);
1388 		return -EINVAL;
1389 	}
1390 
1391 	/* 'tf' is successfully registered. To avoid freeing, assign NULL. */
1392 	tf = NULL;
1393 
1394 	return 0;
1395 }
1396 
trace_fprobe_create_cb(int argc,const char * argv[])1397 static int trace_fprobe_create_cb(int argc, const char *argv[])
1398 {
1399 	struct traceprobe_parse_context *ctx __free(traceprobe_parse_context) = NULL;
1400 	int ret;
1401 
1402 	ctx = kzalloc_obj(*ctx);
1403 	if (!ctx)
1404 		return -ENOMEM;
1405 
1406 	ctx->flags = TPARG_FL_KERNEL | TPARG_FL_FPROBE;
1407 
1408 	trace_probe_log_init("trace_fprobe", argc, argv);
1409 	ret = trace_fprobe_create_internal(argc, argv, ctx);
1410 	trace_probe_log_clear();
1411 	return ret;
1412 }
1413 
trace_fprobe_create(const char * raw_command)1414 static int trace_fprobe_create(const char *raw_command)
1415 {
1416 	return trace_probe_create(raw_command, trace_fprobe_create_cb);
1417 }
1418 
trace_fprobe_release(struct dyn_event * ev)1419 static int trace_fprobe_release(struct dyn_event *ev)
1420 {
1421 	struct trace_fprobe *tf = to_trace_fprobe(ev);
1422 	int ret = unregister_trace_fprobe(tf);
1423 
1424 	if (!ret)
1425 		free_trace_fprobe(tf);
1426 	return ret;
1427 }
1428 
trace_fprobe_show(struct seq_file * m,struct dyn_event * ev)1429 static int trace_fprobe_show(struct seq_file *m, struct dyn_event *ev)
1430 {
1431 	struct trace_fprobe *tf = to_trace_fprobe(ev);
1432 	int i;
1433 
1434 	if (trace_fprobe_is_tracepoint(tf))
1435 		seq_putc(m, 't');
1436 	else
1437 		seq_putc(m, 'f');
1438 	seq_printf(m, ":%s/%s", trace_probe_group_name(&tf->tp),
1439 				trace_probe_name(&tf->tp));
1440 
1441 	seq_printf(m, " %s%s", trace_fprobe_symbol(tf),
1442 			       trace_fprobe_is_return(tf) ? "%return" : "");
1443 
1444 	for (i = 0; i < tf->tp.nr_args; i++)
1445 		seq_printf(m, " %s=%s", tf->tp.args[i].name, tf->tp.args[i].comm);
1446 	seq_putc(m, '\n');
1447 
1448 	trace_probe_dump_args(m, &tf->tp);
1449 
1450 	return 0;
1451 }
1452 
1453 /*
1454  * Enable trace_probe
1455  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
1456  */
enable_trace_fprobe(struct trace_event_call * call,struct trace_event_file * file)1457 static int enable_trace_fprobe(struct trace_event_call *call,
1458 			       struct trace_event_file *file)
1459 {
1460 	struct trace_probe *tp;
1461 	struct trace_fprobe *tf;
1462 	bool enabled;
1463 	int ret = 0;
1464 
1465 	tp = trace_probe_primary_from_call(call);
1466 	if (WARN_ON_ONCE(!tp))
1467 		return -ENODEV;
1468 	enabled = trace_probe_is_enabled(tp);
1469 
1470 	/* This also changes "enabled" state */
1471 	if (file) {
1472 		ret = trace_probe_add_file(tp, file);
1473 		if (ret)
1474 			return ret;
1475 	} else
1476 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1477 
1478 	if (!enabled) {
1479 		list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) {
1480 			ret = __register_trace_fprobe(tf);
1481 			if (ret < 0)
1482 				goto err;
1483 		}
1484 	}
1485 
1486 	return 0;
1487 
1488 err:
1489 	/* Failed to enable one of them. Roll back all */
1490 	list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list)
1491 		__unregister_trace_fprobe(tf);
1492 	if (file)
1493 		trace_probe_remove_file(tp, file);
1494 	else
1495 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1496 	return ret;
1497 }
1498 
1499 /*
1500  * Disable trace_probe
1501  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
1502  */
disable_trace_fprobe(struct trace_event_call * call,struct trace_event_file * file)1503 static int disable_trace_fprobe(struct trace_event_call *call,
1504 				struct trace_event_file *file)
1505 {
1506 	struct trace_fprobe *tf;
1507 	struct trace_probe *tp;
1508 
1509 	tp = trace_probe_primary_from_call(call);
1510 	if (WARN_ON_ONCE(!tp))
1511 		return -ENODEV;
1512 
1513 	if (file) {
1514 		if (!trace_probe_get_file_link(tp, file))
1515 			return -ENOENT;
1516 		if (!trace_probe_has_single_file(tp))
1517 			goto out;
1518 		trace_probe_clear_flag(tp, TP_FLAG_TRACE);
1519 	} else
1520 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1521 
1522 	if (!trace_probe_is_enabled(tp)) {
1523 		list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) {
1524 			unregister_fprobe(&tf->fp);
1525 			if (tf->tuser) {
1526 				tracepoint_user_put(tf->tuser);
1527 				tf->tuser = NULL;
1528 			}
1529 		}
1530 	}
1531 
1532  out:
1533 	if (file)
1534 		/*
1535 		 * Synchronization is done in below function. For perf event,
1536 		 * file == NULL and perf_trace_event_unreg() calls
1537 		 * tracepoint_synchronize_unregister() to ensure synchronize
1538 		 * event. We don't need to care about it.
1539 		 */
1540 		trace_probe_remove_file(tp, file);
1541 
1542 	return 0;
1543 }
1544 
1545 /*
1546  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1547  */
fprobe_register(struct trace_event_call * event,enum trace_reg type,void * data)1548 static int fprobe_register(struct trace_event_call *event,
1549 			   enum trace_reg type, void *data)
1550 {
1551 	struct trace_event_file *file = data;
1552 
1553 	switch (type) {
1554 	case TRACE_REG_REGISTER:
1555 		return enable_trace_fprobe(event, file);
1556 	case TRACE_REG_UNREGISTER:
1557 		return disable_trace_fprobe(event, file);
1558 
1559 #ifdef CONFIG_PERF_EVENTS
1560 	case TRACE_REG_PERF_REGISTER:
1561 		return enable_trace_fprobe(event, NULL);
1562 	case TRACE_REG_PERF_UNREGISTER:
1563 		return disable_trace_fprobe(event, NULL);
1564 	case TRACE_REG_PERF_OPEN:
1565 	case TRACE_REG_PERF_CLOSE:
1566 	case TRACE_REG_PERF_ADD:
1567 	case TRACE_REG_PERF_DEL:
1568 		return 0;
1569 #endif
1570 	}
1571 	return 0;
1572 }
1573 
1574 /*
1575  * Register dynevent at core_initcall. This allows kernel to setup fprobe
1576  * events in postcore_initcall without tracefs.
1577  */
init_fprobe_trace_early(void)1578 static __init int init_fprobe_trace_early(void)
1579 {
1580 	int ret;
1581 
1582 	ret = dyn_event_register(&trace_fprobe_ops);
1583 	if (ret)
1584 		return ret;
1585 
1586 #ifdef CONFIG_MODULES
1587 	ret = register_tracepoint_module_notifier(&tracepoint_module_nb);
1588 	if (ret)
1589 		return ret;
1590 	ret = register_module_notifier(&tprobe_event_module_nb);
1591 	if (ret)
1592 		return ret;
1593 #endif
1594 
1595 	return 0;
1596 }
1597 core_initcall(init_fprobe_trace_early);
1598