xref: /linux/kernel/trace/ftrace.c (revision 5c44ddaf7df3a06391684dde65083a092e06052b)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Infrastructure for profiling code inserted by 'gcc -pg'.
4  *
5  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
6  * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
7  *
8  * Originally ported from the -rt patch by:
9  *   Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
10  *
11  * Based on code in the latency_tracer, that is:
12  *
13  *  Copyright (C) 2004-2006 Ingo Molnar
14  *  Copyright (C) 2004 Nadia Yvette Chambers
15  */
16 
17 #include <linux/stop_machine.h>
18 #include <linux/clocksource.h>
19 #include <linux/sched/task.h>
20 #include <linux/kallsyms.h>
21 #include <linux/security.h>
22 #include <linux/seq_file.h>
23 #include <linux/tracefs.h>
24 #include <linux/hardirq.h>
25 #include <linux/kthread.h>
26 #include <linux/uaccess.h>
27 #include <linux/bsearch.h>
28 #include <linux/module.h>
29 #include <linux/ftrace.h>
30 #include <linux/sysctl.h>
31 #include <linux/slab.h>
32 #include <linux/ctype.h>
33 #include <linux/sort.h>
34 #include <linux/list.h>
35 #include <linux/hash.h>
36 #include <linux/rcupdate.h>
37 #include <linux/kprobes.h>
38 
39 #include <trace/events/sched.h>
40 
41 #include <asm/sections.h>
42 #include <asm/setup.h>
43 
44 #include "ftrace_internal.h"
45 #include "trace_output.h"
46 #include "trace_stat.h"
47 
48 /* Flags that do not get reset */
49 #define FTRACE_NOCLEAR_FLAGS	(FTRACE_FL_DISABLED | FTRACE_FL_TOUCHED | \
50 				 FTRACE_FL_MODIFIED)
51 
52 #define FTRACE_INVALID_FUNCTION		"__ftrace_invalid_address__"
53 
54 #define FTRACE_WARN_ON(cond)			\
55 	({					\
56 		int ___r = cond;		\
57 		if (WARN_ON(___r))		\
58 			ftrace_kill();		\
59 		___r;				\
60 	})
61 
62 #define FTRACE_WARN_ON_ONCE(cond)		\
63 	({					\
64 		int ___r = cond;		\
65 		if (WARN_ON_ONCE(___r))		\
66 			ftrace_kill();		\
67 		___r;				\
68 	})
69 
70 /* hash bits for specific function selection */
71 #define FTRACE_HASH_DEFAULT_BITS 10
72 #define FTRACE_HASH_MAX_BITS 12
73 
74 #ifdef CONFIG_DYNAMIC_FTRACE
75 #define INIT_OPS_HASH(opsname)	\
76 	.func_hash		= &opsname.local_hash,			\
77 	.local_hash.regex_lock	= __MUTEX_INITIALIZER(opsname.local_hash.regex_lock), \
78 	.subop_list		= LIST_HEAD_INIT(opsname.subop_list),
79 #else
80 #define INIT_OPS_HASH(opsname)
81 #endif
82 
83 enum {
84 	FTRACE_MODIFY_ENABLE_FL		= (1 << 0),
85 	FTRACE_MODIFY_MAY_SLEEP_FL	= (1 << 1),
86 };
87 
88 struct ftrace_ops ftrace_list_end __read_mostly = {
89 	.func		= ftrace_stub,
90 	.flags		= FTRACE_OPS_FL_STUB,
91 	INIT_OPS_HASH(ftrace_list_end)
92 };
93 
94 /* ftrace_enabled is a method to turn ftrace on or off */
95 int ftrace_enabled __read_mostly;
96 static int __maybe_unused last_ftrace_enabled;
97 
98 /* Current function tracing op */
99 struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
100 /* What to set function_trace_op to */
101 static struct ftrace_ops *set_function_trace_op;
102 
ftrace_pids_enabled(struct ftrace_ops * ops)103 bool ftrace_pids_enabled(struct ftrace_ops *ops)
104 {
105 	struct trace_array *tr;
106 
107 	if (!(ops->flags & FTRACE_OPS_FL_PID) || !ops->private)
108 		return false;
109 
110 	tr = ops->private;
111 
112 	return tr->function_pids != NULL || tr->function_no_pids != NULL;
113 }
114 
115 static void ftrace_update_trampoline(struct ftrace_ops *ops);
116 
117 /*
118  * ftrace_disabled is set when an anomaly is discovered.
119  * ftrace_disabled is much stronger than ftrace_enabled.
120  */
121 static int ftrace_disabled __read_mostly;
122 
123 DEFINE_MUTEX(ftrace_lock);
124 
125 struct ftrace_ops __rcu *ftrace_ops_list __read_mostly = (struct ftrace_ops __rcu *)&ftrace_list_end;
126 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
127 struct ftrace_ops global_ops;
128 
129 /* Defined by vmlinux.lds.h see the comment above arch_ftrace_ops_list_func for details */
130 void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
131 			  struct ftrace_ops *op, struct ftrace_regs *fregs);
132 
133 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_CALL_OPS
134 /*
135  * Stub used to invoke the list ops without requiring a separate trampoline.
136  */
137 const struct ftrace_ops ftrace_list_ops = {
138 	.func	= ftrace_ops_list_func,
139 	.flags	= FTRACE_OPS_FL_STUB,
140 };
141 
ftrace_ops_nop_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)142 static void ftrace_ops_nop_func(unsigned long ip, unsigned long parent_ip,
143 				struct ftrace_ops *op,
144 				struct ftrace_regs *fregs)
145 {
146 	/* do nothing */
147 }
148 
149 /*
150  * Stub used when a call site is disabled. May be called transiently by threads
151  * which have made it into ftrace_caller but haven't yet recovered the ops at
152  * the point the call site is disabled.
153  */
154 const struct ftrace_ops ftrace_nop_ops = {
155 	.func	= ftrace_ops_nop_func,
156 	.flags  = FTRACE_OPS_FL_STUB,
157 };
158 #endif
159 
ftrace_ops_init(struct ftrace_ops * ops)160 static inline void ftrace_ops_init(struct ftrace_ops *ops)
161 {
162 #ifdef CONFIG_DYNAMIC_FTRACE
163 	if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
164 		mutex_init(&ops->local_hash.regex_lock);
165 		INIT_LIST_HEAD(&ops->subop_list);
166 		ops->func_hash = &ops->local_hash;
167 		ops->flags |= FTRACE_OPS_FL_INITIALIZED;
168 	}
169 #endif
170 }
171 
172 /* Call this function for when a callback filters on set_ftrace_pid */
ftrace_pid_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)173 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
174 			    struct ftrace_ops *op, struct ftrace_regs *fregs)
175 {
176 	struct trace_array *tr = op->private;
177 	int pid;
178 
179 	if (tr) {
180 		pid = this_cpu_read(tr->array_buffer.data->ftrace_ignore_pid);
181 		if (pid == FTRACE_PID_IGNORE)
182 			return;
183 		if (pid != FTRACE_PID_TRACE &&
184 		    pid != current->pid)
185 			return;
186 	}
187 
188 	op->saved_func(ip, parent_ip, op, fregs);
189 }
190 
ftrace_sync_ipi(void * data)191 static void ftrace_sync_ipi(void *data)
192 {
193 	/* Probably not needed, but do it anyway */
194 	smp_rmb();
195 }
196 
ftrace_ops_get_list_func(struct ftrace_ops * ops)197 static ftrace_func_t ftrace_ops_get_list_func(struct ftrace_ops *ops)
198 {
199 	/*
200 	 * If this is a dynamic or RCU ops, or we force list func,
201 	 * then it needs to call the list anyway.
202 	 */
203 	if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_RCU) ||
204 	    FTRACE_FORCE_LIST_FUNC)
205 		return ftrace_ops_list_func;
206 
207 	return ftrace_ops_get_func(ops);
208 }
209 
update_ftrace_function(void)210 static void update_ftrace_function(void)
211 {
212 	ftrace_func_t func;
213 
214 	/*
215 	 * Prepare the ftrace_ops that the arch callback will use.
216 	 * If there's only one ftrace_ops registered, the ftrace_ops_list
217 	 * will point to the ops we want.
218 	 */
219 	set_function_trace_op = rcu_dereference_protected(ftrace_ops_list,
220 						lockdep_is_held(&ftrace_lock));
221 
222 	/* If there's no ftrace_ops registered, just call the stub function */
223 	if (set_function_trace_op == &ftrace_list_end) {
224 		func = ftrace_stub;
225 
226 	/*
227 	 * If we are at the end of the list and this ops is
228 	 * recursion safe and not dynamic and the arch supports passing ops,
229 	 * then have the mcount trampoline call the function directly.
230 	 */
231 	} else if (rcu_dereference_protected(ftrace_ops_list->next,
232 			lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
233 		func = ftrace_ops_get_list_func(ftrace_ops_list);
234 
235 	} else {
236 		/* Just use the default ftrace_ops */
237 		set_function_trace_op = &ftrace_list_end;
238 		func = ftrace_ops_list_func;
239 	}
240 
241 	/* If there's no change, then do nothing more here */
242 	if (ftrace_trace_function == func)
243 		return;
244 
245 	/*
246 	 * If we are using the list function, it doesn't care
247 	 * about the function_trace_ops.
248 	 */
249 	if (func == ftrace_ops_list_func) {
250 		ftrace_trace_function = func;
251 		/*
252 		 * Don't even bother setting function_trace_ops,
253 		 * it would be racy to do so anyway.
254 		 */
255 		return;
256 	}
257 
258 #ifndef CONFIG_DYNAMIC_FTRACE
259 	/*
260 	 * For static tracing, we need to be a bit more careful.
261 	 * The function change takes affect immediately. Thus,
262 	 * we need to coordinate the setting of the function_trace_ops
263 	 * with the setting of the ftrace_trace_function.
264 	 *
265 	 * Set the function to the list ops, which will call the
266 	 * function we want, albeit indirectly, but it handles the
267 	 * ftrace_ops and doesn't depend on function_trace_op.
268 	 */
269 	ftrace_trace_function = ftrace_ops_list_func;
270 	/*
271 	 * Make sure all CPUs see this. Yes this is slow, but static
272 	 * tracing is slow and nasty to have enabled.
273 	 */
274 	synchronize_rcu_tasks_rude();
275 	/* Now all cpus are using the list ops. */
276 	function_trace_op = set_function_trace_op;
277 	/* Make sure the function_trace_op is visible on all CPUs */
278 	smp_wmb();
279 	/* Nasty way to force a rmb on all cpus */
280 	smp_call_function(ftrace_sync_ipi, NULL, 1);
281 	/* OK, we are all set to update the ftrace_trace_function now! */
282 #endif /* !CONFIG_DYNAMIC_FTRACE */
283 
284 	ftrace_trace_function = func;
285 }
286 
add_ftrace_ops(struct ftrace_ops __rcu ** list,struct ftrace_ops * ops)287 static void add_ftrace_ops(struct ftrace_ops __rcu **list,
288 			   struct ftrace_ops *ops)
289 {
290 	rcu_assign_pointer(ops->next, *list);
291 
292 	/*
293 	 * We are entering ops into the list but another
294 	 * CPU might be walking that list. We need to make sure
295 	 * the ops->next pointer is valid before another CPU sees
296 	 * the ops pointer included into the list.
297 	 */
298 	rcu_assign_pointer(*list, ops);
299 }
300 
remove_ftrace_ops(struct ftrace_ops __rcu ** list,struct ftrace_ops * ops)301 static int remove_ftrace_ops(struct ftrace_ops __rcu **list,
302 			     struct ftrace_ops *ops)
303 {
304 	struct ftrace_ops **p;
305 
306 	/*
307 	 * If we are removing the last function, then simply point
308 	 * to the ftrace_stub.
309 	 */
310 	if (rcu_dereference_protected(*list,
311 			lockdep_is_held(&ftrace_lock)) == ops &&
312 	    rcu_dereference_protected(ops->next,
313 			lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
314 		rcu_assign_pointer(*list, &ftrace_list_end);
315 		return 0;
316 	}
317 
318 	for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
319 		if (*p == ops)
320 			break;
321 
322 	if (*p != ops)
323 		return -1;
324 
325 	*p = (*p)->next;
326 	return 0;
327 }
328 
329 static void ftrace_update_trampoline(struct ftrace_ops *ops);
330 
__register_ftrace_function(struct ftrace_ops * ops)331 int __register_ftrace_function(struct ftrace_ops *ops)
332 {
333 	if (ops->flags & FTRACE_OPS_FL_DELETED)
334 		return -EINVAL;
335 
336 	if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
337 		return -EBUSY;
338 
339 #ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
340 	/*
341 	 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
342 	 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
343 	 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
344 	 */
345 	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
346 	    !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
347 		return -EINVAL;
348 
349 	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
350 		ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
351 #endif
352 	if (!ftrace_enabled && (ops->flags & FTRACE_OPS_FL_PERMANENT))
353 		return -EBUSY;
354 
355 	if (!is_kernel_core_data((unsigned long)ops))
356 		ops->flags |= FTRACE_OPS_FL_DYNAMIC;
357 
358 	add_ftrace_ops(&ftrace_ops_list, ops);
359 
360 	/* Always save the function, and reset at unregistering */
361 	ops->saved_func = ops->func;
362 
363 	if (ftrace_pids_enabled(ops))
364 		ops->func = ftrace_pid_func;
365 
366 	ftrace_update_trampoline(ops);
367 
368 	if (ftrace_enabled)
369 		update_ftrace_function();
370 
371 	return 0;
372 }
373 
__unregister_ftrace_function(struct ftrace_ops * ops)374 int __unregister_ftrace_function(struct ftrace_ops *ops)
375 {
376 	int ret;
377 
378 	if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
379 		return -EBUSY;
380 
381 	ret = remove_ftrace_ops(&ftrace_ops_list, ops);
382 
383 	if (ret < 0)
384 		return ret;
385 
386 	if (ftrace_enabled)
387 		update_ftrace_function();
388 
389 	ops->func = ops->saved_func;
390 
391 	return 0;
392 }
393 
ftrace_update_pid_func(void)394 static void ftrace_update_pid_func(void)
395 {
396 	struct ftrace_ops *op;
397 
398 	/* Only do something if we are tracing something */
399 	if (ftrace_trace_function == ftrace_stub)
400 		return;
401 
402 	do_for_each_ftrace_op(op, ftrace_ops_list) {
403 		if (op->flags & FTRACE_OPS_FL_PID) {
404 			op->func = ftrace_pids_enabled(op) ?
405 				ftrace_pid_func : op->saved_func;
406 			ftrace_update_trampoline(op);
407 		}
408 	} while_for_each_ftrace_op(op);
409 
410 	fgraph_update_pid_func();
411 
412 	update_ftrace_function();
413 }
414 
415 #ifdef CONFIG_FUNCTION_PROFILER
416 struct ftrace_profile {
417 	struct hlist_node		node;
418 	unsigned long			ip;
419 	unsigned long			counter;
420 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
421 	unsigned long long		time;
422 	unsigned long long		time_squared;
423 #endif
424 };
425 
426 struct ftrace_profile_page {
427 	struct ftrace_profile_page	*next;
428 	unsigned long			index;
429 	struct ftrace_profile		records[];
430 };
431 
432 struct ftrace_profile_stat {
433 	atomic_t			disabled;
434 	struct hlist_head		*hash;
435 	struct ftrace_profile_page	*pages;
436 	struct ftrace_profile_page	*start;
437 	struct tracer_stat		stat;
438 };
439 
440 #define PROFILE_RECORDS_SIZE						\
441 	(PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
442 
443 #define PROFILES_PER_PAGE					\
444 	(PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
445 
446 static int ftrace_profile_enabled __read_mostly;
447 
448 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
449 static DEFINE_MUTEX(ftrace_profile_lock);
450 
451 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
452 
453 #define FTRACE_PROFILE_HASH_BITS 10
454 #define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
455 
456 static void *
function_stat_next(void * v,int idx)457 function_stat_next(void *v, int idx)
458 {
459 	struct ftrace_profile *rec = v;
460 	struct ftrace_profile_page *pg;
461 
462 	pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
463 
464  again:
465 	if (idx != 0)
466 		rec++;
467 
468 	if ((void *)rec >= (void *)&pg->records[pg->index]) {
469 		pg = pg->next;
470 		if (!pg)
471 			return NULL;
472 		rec = &pg->records[0];
473 		if (!rec->counter)
474 			goto again;
475 	}
476 
477 	return rec;
478 }
479 
function_stat_start(struct tracer_stat * trace)480 static void *function_stat_start(struct tracer_stat *trace)
481 {
482 	struct ftrace_profile_stat *stat =
483 		container_of(trace, struct ftrace_profile_stat, stat);
484 
485 	if (!stat || !stat->start)
486 		return NULL;
487 
488 	return function_stat_next(&stat->start->records[0], 0);
489 }
490 
491 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
492 /* function graph compares on total time */
function_stat_cmp(const void * p1,const void * p2)493 static int function_stat_cmp(const void *p1, const void *p2)
494 {
495 	const struct ftrace_profile *a = p1;
496 	const struct ftrace_profile *b = p2;
497 
498 	if (a->time < b->time)
499 		return -1;
500 	if (a->time > b->time)
501 		return 1;
502 	else
503 		return 0;
504 }
505 #else
506 /* not function graph compares against hits */
function_stat_cmp(const void * p1,const void * p2)507 static int function_stat_cmp(const void *p1, const void *p2)
508 {
509 	const struct ftrace_profile *a = p1;
510 	const struct ftrace_profile *b = p2;
511 
512 	if (a->counter < b->counter)
513 		return -1;
514 	if (a->counter > b->counter)
515 		return 1;
516 	else
517 		return 0;
518 }
519 #endif
520 
function_stat_headers(struct seq_file * m)521 static int function_stat_headers(struct seq_file *m)
522 {
523 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
524 	seq_puts(m, "  Function                               "
525 		 "Hit    Time            Avg             s^2\n"
526 		    "  --------                               "
527 		 "---    ----            ---             ---\n");
528 #else
529 	seq_puts(m, "  Function                               Hit\n"
530 		    "  --------                               ---\n");
531 #endif
532 	return 0;
533 }
534 
function_stat_show(struct seq_file * m,void * v)535 static int function_stat_show(struct seq_file *m, void *v)
536 {
537 	struct ftrace_profile *rec = v;
538 	char str[KSYM_SYMBOL_LEN];
539 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
540 	static struct trace_seq s;
541 	unsigned long long avg;
542 	unsigned long long stddev;
543 	unsigned long long stddev_denom;
544 #endif
545 	guard(mutex)(&ftrace_profile_lock);
546 
547 	/* we raced with function_profile_reset() */
548 	if (unlikely(rec->counter == 0))
549 		return -EBUSY;
550 
551 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
552 	avg = div64_ul(rec->time, rec->counter);
553 	if (tracing_thresh && (avg < tracing_thresh))
554 		return 0;
555 #endif
556 
557 	kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
558 	seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
559 
560 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
561 	seq_puts(m, "    ");
562 
563 	/*
564 	 * Variance formula:
565 	 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
566 	 * Maybe Welford's method is better here?
567 	 * Divide only by 1000 for ns^2 -> us^2 conversion.
568 	 * trace_print_graph_duration will divide by 1000 again.
569 	 */
570 	stddev = 0;
571 	stddev_denom = rec->counter * (rec->counter - 1) * 1000;
572 	if (stddev_denom) {
573 		stddev = rec->counter * rec->time_squared -
574 			 rec->time * rec->time;
575 		stddev = div64_ul(stddev, stddev_denom);
576 	}
577 
578 	trace_seq_init(&s);
579 	trace_print_graph_duration(rec->time, &s);
580 	trace_seq_puts(&s, "    ");
581 	trace_print_graph_duration(avg, &s);
582 	trace_seq_puts(&s, "    ");
583 	trace_print_graph_duration(stddev, &s);
584 	trace_print_seq(m, &s);
585 #endif
586 	seq_putc(m, '\n');
587 
588 	return 0;
589 }
590 
ftrace_profile_reset(struct ftrace_profile_stat * stat)591 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
592 {
593 	struct ftrace_profile_page *pg;
594 
595 	pg = stat->pages = stat->start;
596 
597 	while (pg) {
598 		memset(pg->records, 0, PROFILE_RECORDS_SIZE);
599 		pg->index = 0;
600 		pg = pg->next;
601 	}
602 
603 	memset(stat->hash, 0,
604 	       FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
605 }
606 
ftrace_profile_pages_init(struct ftrace_profile_stat * stat)607 static int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
608 {
609 	struct ftrace_profile_page *pg;
610 	int functions;
611 	int pages;
612 	int i;
613 
614 	/* If we already allocated, do nothing */
615 	if (stat->pages)
616 		return 0;
617 
618 	stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
619 	if (!stat->pages)
620 		return -ENOMEM;
621 
622 #ifdef CONFIG_DYNAMIC_FTRACE
623 	functions = ftrace_update_tot_cnt;
624 #else
625 	/*
626 	 * We do not know the number of functions that exist because
627 	 * dynamic tracing is what counts them. With past experience
628 	 * we have around 20K functions. That should be more than enough.
629 	 * It is highly unlikely we will execute every function in
630 	 * the kernel.
631 	 */
632 	functions = 20000;
633 #endif
634 
635 	pg = stat->start = stat->pages;
636 
637 	pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
638 
639 	for (i = 1; i < pages; i++) {
640 		pg->next = (void *)get_zeroed_page(GFP_KERNEL);
641 		if (!pg->next)
642 			goto out_free;
643 		pg = pg->next;
644 	}
645 
646 	return 0;
647 
648  out_free:
649 	pg = stat->start;
650 	while (pg) {
651 		unsigned long tmp = (unsigned long)pg;
652 
653 		pg = pg->next;
654 		free_page(tmp);
655 	}
656 
657 	stat->pages = NULL;
658 	stat->start = NULL;
659 
660 	return -ENOMEM;
661 }
662 
ftrace_profile_init_cpu(int cpu)663 static int ftrace_profile_init_cpu(int cpu)
664 {
665 	struct ftrace_profile_stat *stat;
666 	int size;
667 
668 	stat = &per_cpu(ftrace_profile_stats, cpu);
669 
670 	if (stat->hash) {
671 		/* If the profile is already created, simply reset it */
672 		ftrace_profile_reset(stat);
673 		return 0;
674 	}
675 
676 	/*
677 	 * We are profiling all functions, but usually only a few thousand
678 	 * functions are hit. We'll make a hash of 1024 items.
679 	 */
680 	size = FTRACE_PROFILE_HASH_SIZE;
681 
682 	stat->hash = kcalloc(size, sizeof(struct hlist_head), GFP_KERNEL);
683 
684 	if (!stat->hash)
685 		return -ENOMEM;
686 
687 	/* Preallocate the function profiling pages */
688 	if (ftrace_profile_pages_init(stat) < 0) {
689 		kfree(stat->hash);
690 		stat->hash = NULL;
691 		return -ENOMEM;
692 	}
693 
694 	return 0;
695 }
696 
ftrace_profile_init(void)697 static int ftrace_profile_init(void)
698 {
699 	int cpu;
700 	int ret = 0;
701 
702 	for_each_possible_cpu(cpu) {
703 		ret = ftrace_profile_init_cpu(cpu);
704 		if (ret)
705 			break;
706 	}
707 
708 	return ret;
709 }
710 
711 /* interrupts must be disabled */
712 static struct ftrace_profile *
ftrace_find_profiled_func(struct ftrace_profile_stat * stat,unsigned long ip)713 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
714 {
715 	struct ftrace_profile *rec;
716 	struct hlist_head *hhd;
717 	unsigned long key;
718 
719 	key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
720 	hhd = &stat->hash[key];
721 
722 	if (hlist_empty(hhd))
723 		return NULL;
724 
725 	hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
726 		if (rec->ip == ip)
727 			return rec;
728 	}
729 
730 	return NULL;
731 }
732 
ftrace_add_profile(struct ftrace_profile_stat * stat,struct ftrace_profile * rec)733 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
734 			       struct ftrace_profile *rec)
735 {
736 	unsigned long key;
737 
738 	key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
739 	hlist_add_head_rcu(&rec->node, &stat->hash[key]);
740 }
741 
742 /*
743  * The memory is already allocated, this simply finds a new record to use.
744  */
745 static struct ftrace_profile *
ftrace_profile_alloc(struct ftrace_profile_stat * stat,unsigned long ip)746 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
747 {
748 	struct ftrace_profile *rec = NULL;
749 
750 	/* prevent recursion (from NMIs) */
751 	if (atomic_inc_return(&stat->disabled) != 1)
752 		goto out;
753 
754 	/*
755 	 * Try to find the function again since an NMI
756 	 * could have added it
757 	 */
758 	rec = ftrace_find_profiled_func(stat, ip);
759 	if (rec)
760 		goto out;
761 
762 	if (stat->pages->index == PROFILES_PER_PAGE) {
763 		if (!stat->pages->next)
764 			goto out;
765 		stat->pages = stat->pages->next;
766 	}
767 
768 	rec = &stat->pages->records[stat->pages->index++];
769 	rec->ip = ip;
770 	ftrace_add_profile(stat, rec);
771 
772  out:
773 	atomic_dec(&stat->disabled);
774 
775 	return rec;
776 }
777 
778 static void
function_profile_call(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * ops,struct ftrace_regs * fregs)779 function_profile_call(unsigned long ip, unsigned long parent_ip,
780 		      struct ftrace_ops *ops, struct ftrace_regs *fregs)
781 {
782 	struct ftrace_profile_stat *stat;
783 	struct ftrace_profile *rec;
784 
785 	if (!ftrace_profile_enabled)
786 		return;
787 
788 	guard(preempt_notrace)();
789 
790 	stat = this_cpu_ptr(&ftrace_profile_stats);
791 	if (!stat->hash || !ftrace_profile_enabled)
792 		return;
793 
794 	rec = ftrace_find_profiled_func(stat, ip);
795 	if (!rec) {
796 		rec = ftrace_profile_alloc(stat, ip);
797 		if (!rec)
798 			return;
799 	}
800 
801 	rec->counter++;
802 }
803 
804 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
805 static bool fgraph_graph_time = true;
806 
ftrace_graph_graph_time_control(bool enable)807 void ftrace_graph_graph_time_control(bool enable)
808 {
809 	fgraph_graph_time = enable;
810 }
811 
812 struct profile_fgraph_data {
813 	unsigned long long		calltime;
814 	unsigned long long		subtime;
815 	unsigned long long		sleeptime;
816 };
817 
profile_graph_entry(struct ftrace_graph_ent * trace,struct fgraph_ops * gops,struct ftrace_regs * fregs)818 static int profile_graph_entry(struct ftrace_graph_ent *trace,
819 			       struct fgraph_ops *gops,
820 			       struct ftrace_regs *fregs)
821 {
822 	struct profile_fgraph_data *profile_data;
823 
824 	function_profile_call(trace->func, 0, NULL, NULL);
825 
826 	/* If function graph is shutting down, ret_stack can be NULL */
827 	if (!current->ret_stack)
828 		return 0;
829 
830 	profile_data = fgraph_reserve_data(gops->idx, sizeof(*profile_data));
831 	if (!profile_data)
832 		return 0;
833 
834 	profile_data->subtime = 0;
835 	profile_data->sleeptime = current->ftrace_sleeptime;
836 	profile_data->calltime = trace_clock_local();
837 
838 	return 1;
839 }
840 
profile_graph_return(struct ftrace_graph_ret * trace,struct fgraph_ops * gops,struct ftrace_regs * fregs)841 static void profile_graph_return(struct ftrace_graph_ret *trace,
842 				 struct fgraph_ops *gops,
843 				 struct ftrace_regs *fregs)
844 {
845 	struct profile_fgraph_data *profile_data;
846 	struct ftrace_profile_stat *stat;
847 	unsigned long long calltime;
848 	unsigned long long rettime = trace_clock_local();
849 	struct ftrace_profile *rec;
850 	int size;
851 
852 	guard(preempt_notrace)();
853 
854 	stat = this_cpu_ptr(&ftrace_profile_stats);
855 	if (!stat->hash || !ftrace_profile_enabled)
856 		return;
857 
858 	profile_data = fgraph_retrieve_data(gops->idx, &size);
859 
860 	/* If the calltime was zero'd ignore it */
861 	if (!profile_data || !profile_data->calltime)
862 		return;
863 
864 	calltime = rettime - profile_data->calltime;
865 
866 	if (!fgraph_sleep_time) {
867 		if (current->ftrace_sleeptime)
868 			calltime -= current->ftrace_sleeptime - profile_data->sleeptime;
869 	}
870 
871 	if (!fgraph_graph_time) {
872 		struct profile_fgraph_data *parent_data;
873 
874 		/* Append this call time to the parent time to subtract */
875 		parent_data = fgraph_retrieve_parent_data(gops->idx, &size, 1);
876 		if (parent_data)
877 			parent_data->subtime += calltime;
878 
879 		if (profile_data->subtime && profile_data->subtime < calltime)
880 			calltime -= profile_data->subtime;
881 		else
882 			calltime = 0;
883 	}
884 
885 	rec = ftrace_find_profiled_func(stat, trace->func);
886 	if (rec) {
887 		rec->time += calltime;
888 		rec->time_squared += calltime * calltime;
889 	}
890 }
891 
892 static struct fgraph_ops fprofiler_ops = {
893 	.entryfunc = &profile_graph_entry,
894 	.retfunc = &profile_graph_return,
895 };
896 
register_ftrace_profiler(void)897 static int register_ftrace_profiler(void)
898 {
899 	ftrace_ops_set_global_filter(&fprofiler_ops.ops);
900 	return register_ftrace_graph(&fprofiler_ops);
901 }
902 
unregister_ftrace_profiler(void)903 static void unregister_ftrace_profiler(void)
904 {
905 	unregister_ftrace_graph(&fprofiler_ops);
906 }
907 #else
908 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
909 	.func		= function_profile_call,
910 };
911 
register_ftrace_profiler(void)912 static int register_ftrace_profiler(void)
913 {
914 	ftrace_ops_set_global_filter(&ftrace_profile_ops);
915 	return register_ftrace_function(&ftrace_profile_ops);
916 }
917 
unregister_ftrace_profiler(void)918 static void unregister_ftrace_profiler(void)
919 {
920 	unregister_ftrace_function(&ftrace_profile_ops);
921 }
922 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
923 
924 static ssize_t
ftrace_profile_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)925 ftrace_profile_write(struct file *filp, const char __user *ubuf,
926 		     size_t cnt, loff_t *ppos)
927 {
928 	unsigned long val;
929 	int ret;
930 
931 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
932 	if (ret)
933 		return ret;
934 
935 	val = !!val;
936 
937 	guard(mutex)(&ftrace_profile_lock);
938 	if (ftrace_profile_enabled ^ val) {
939 		if (val) {
940 			ret = ftrace_profile_init();
941 			if (ret < 0)
942 				return ret;
943 
944 			ret = register_ftrace_profiler();
945 			if (ret < 0)
946 				return ret;
947 			ftrace_profile_enabled = 1;
948 		} else {
949 			ftrace_profile_enabled = 0;
950 			/*
951 			 * unregister_ftrace_profiler calls stop_machine
952 			 * so this acts like an synchronize_rcu.
953 			 */
954 			unregister_ftrace_profiler();
955 		}
956 	}
957 
958 	*ppos += cnt;
959 
960 	return cnt;
961 }
962 
963 static ssize_t
ftrace_profile_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)964 ftrace_profile_read(struct file *filp, char __user *ubuf,
965 		     size_t cnt, loff_t *ppos)
966 {
967 	char buf[64];		/* big enough to hold a number */
968 	int r;
969 
970 	r = sprintf(buf, "%u\n", ftrace_profile_enabled);
971 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
972 }
973 
974 static const struct file_operations ftrace_profile_fops = {
975 	.open		= tracing_open_generic,
976 	.read		= ftrace_profile_read,
977 	.write		= ftrace_profile_write,
978 	.llseek		= default_llseek,
979 };
980 
981 /* used to initialize the real stat files */
982 static struct tracer_stat function_stats __initdata = {
983 	.name		= "functions",
984 	.stat_start	= function_stat_start,
985 	.stat_next	= function_stat_next,
986 	.stat_cmp	= function_stat_cmp,
987 	.stat_headers	= function_stat_headers,
988 	.stat_show	= function_stat_show
989 };
990 
ftrace_profile_tracefs(struct dentry * d_tracer)991 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
992 {
993 	struct ftrace_profile_stat *stat;
994 	char *name;
995 	int ret;
996 	int cpu;
997 
998 	for_each_possible_cpu(cpu) {
999 		stat = &per_cpu(ftrace_profile_stats, cpu);
1000 
1001 		name = kasprintf(GFP_KERNEL, "function%d", cpu);
1002 		if (!name) {
1003 			/*
1004 			 * The files created are permanent, if something happens
1005 			 * we still do not free memory.
1006 			 */
1007 			WARN(1,
1008 			     "Could not allocate stat file for cpu %d\n",
1009 			     cpu);
1010 			return;
1011 		}
1012 		stat->stat = function_stats;
1013 		stat->stat.name = name;
1014 		ret = register_stat_tracer(&stat->stat);
1015 		if (ret) {
1016 			WARN(1,
1017 			     "Could not register function stat for cpu %d\n",
1018 			     cpu);
1019 			kfree(name);
1020 			return;
1021 		}
1022 	}
1023 
1024 	trace_create_file("function_profile_enabled",
1025 			  TRACE_MODE_WRITE, d_tracer, NULL,
1026 			  &ftrace_profile_fops);
1027 }
1028 
1029 #else /* CONFIG_FUNCTION_PROFILER */
ftrace_profile_tracefs(struct dentry * d_tracer)1030 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1031 {
1032 }
1033 #endif /* CONFIG_FUNCTION_PROFILER */
1034 
1035 #ifdef CONFIG_DYNAMIC_FTRACE
1036 
1037 static struct ftrace_ops *removed_ops;
1038 
1039 /*
1040  * Set when doing a global update, like enabling all recs or disabling them.
1041  * It is not set when just updating a single ftrace_ops.
1042  */
1043 static bool update_all_ops;
1044 
1045 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1046 # error Dynamic ftrace depends on MCOUNT_RECORD
1047 #endif
1048 
1049 struct ftrace_func_probe {
1050 	struct ftrace_probe_ops	*probe_ops;
1051 	struct ftrace_ops	ops;
1052 	struct trace_array	*tr;
1053 	struct list_head	list;
1054 	void			*data;
1055 	int			ref;
1056 };
1057 
1058 /*
1059  * We make these constant because no one should touch them,
1060  * but they are used as the default "empty hash", to avoid allocating
1061  * it all the time. These are in a read only section such that if
1062  * anyone does try to modify it, it will cause an exception.
1063  */
1064 static const struct hlist_head empty_buckets[1];
1065 static const struct ftrace_hash empty_hash = {
1066 	.buckets = (struct hlist_head *)empty_buckets,
1067 };
1068 #define EMPTY_HASH	((struct ftrace_hash *)&empty_hash)
1069 
1070 struct ftrace_ops global_ops = {
1071 	.func				= ftrace_stub,
1072 	.local_hash.notrace_hash	= EMPTY_HASH,
1073 	.local_hash.filter_hash		= EMPTY_HASH,
1074 	INIT_OPS_HASH(global_ops)
1075 	.flags				= FTRACE_OPS_FL_INITIALIZED |
1076 					  FTRACE_OPS_FL_PID,
1077 };
1078 
1079 /*
1080  * Used by the stack unwinder to know about dynamic ftrace trampolines.
1081  */
ftrace_ops_trampoline(unsigned long addr)1082 struct ftrace_ops *ftrace_ops_trampoline(unsigned long addr)
1083 {
1084 	struct ftrace_ops *op = NULL;
1085 
1086 	/*
1087 	 * Some of the ops may be dynamically allocated,
1088 	 * they are freed after a synchronize_rcu().
1089 	 */
1090 	preempt_disable_notrace();
1091 
1092 	do_for_each_ftrace_op(op, ftrace_ops_list) {
1093 		/*
1094 		 * This is to check for dynamically allocated trampolines.
1095 		 * Trampolines that are in kernel text will have
1096 		 * core_kernel_text() return true.
1097 		 */
1098 		if (op->trampoline && op->trampoline_size)
1099 			if (addr >= op->trampoline &&
1100 			    addr < op->trampoline + op->trampoline_size) {
1101 				preempt_enable_notrace();
1102 				return op;
1103 			}
1104 	} while_for_each_ftrace_op(op);
1105 	preempt_enable_notrace();
1106 
1107 	return NULL;
1108 }
1109 
1110 /*
1111  * This is used by __kernel_text_address() to return true if the
1112  * address is on a dynamically allocated trampoline that would
1113  * not return true for either core_kernel_text() or
1114  * is_module_text_address().
1115  */
is_ftrace_trampoline(unsigned long addr)1116 bool is_ftrace_trampoline(unsigned long addr)
1117 {
1118 	return ftrace_ops_trampoline(addr) != NULL;
1119 }
1120 
1121 struct ftrace_page {
1122 	struct ftrace_page	*next;
1123 	struct dyn_ftrace	*records;
1124 	int			index;
1125 	int			order;
1126 };
1127 
1128 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1129 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1130 
1131 static struct ftrace_page	*ftrace_pages_start;
1132 static struct ftrace_page	*ftrace_pages;
1133 
1134 static __always_inline unsigned long
ftrace_hash_key(struct ftrace_hash * hash,unsigned long ip)1135 ftrace_hash_key(struct ftrace_hash *hash, unsigned long ip)
1136 {
1137 	if (hash->size_bits > 0)
1138 		return hash_long(ip, hash->size_bits);
1139 
1140 	return 0;
1141 }
1142 
1143 /* Only use this function if ftrace_hash_empty() has already been tested */
1144 static __always_inline struct ftrace_func_entry *
__ftrace_lookup_ip(struct ftrace_hash * hash,unsigned long ip)1145 __ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1146 {
1147 	unsigned long key;
1148 	struct ftrace_func_entry *entry;
1149 	struct hlist_head *hhd;
1150 
1151 	key = ftrace_hash_key(hash, ip);
1152 	hhd = &hash->buckets[key];
1153 
1154 	hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1155 		if (entry->ip == ip)
1156 			return entry;
1157 	}
1158 	return NULL;
1159 }
1160 
1161 /**
1162  * ftrace_lookup_ip - Test to see if an ip exists in an ftrace_hash
1163  * @hash: The hash to look at
1164  * @ip: The instruction pointer to test
1165  *
1166  * Search a given @hash to see if a given instruction pointer (@ip)
1167  * exists in it.
1168  *
1169  * Returns: the entry that holds the @ip if found. NULL otherwise.
1170  */
1171 struct ftrace_func_entry *
ftrace_lookup_ip(struct ftrace_hash * hash,unsigned long ip)1172 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1173 {
1174 	if (ftrace_hash_empty(hash))
1175 		return NULL;
1176 
1177 	return __ftrace_lookup_ip(hash, ip);
1178 }
1179 
__add_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1180 static void __add_hash_entry(struct ftrace_hash *hash,
1181 			     struct ftrace_func_entry *entry)
1182 {
1183 	struct hlist_head *hhd;
1184 	unsigned long key;
1185 
1186 	key = ftrace_hash_key(hash, entry->ip);
1187 	hhd = &hash->buckets[key];
1188 	hlist_add_head(&entry->hlist, hhd);
1189 	hash->count++;
1190 }
1191 
1192 static struct ftrace_func_entry *
add_hash_entry(struct ftrace_hash * hash,unsigned long ip)1193 add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1194 {
1195 	struct ftrace_func_entry *entry;
1196 
1197 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1198 	if (!entry)
1199 		return NULL;
1200 
1201 	entry->ip = ip;
1202 	__add_hash_entry(hash, entry);
1203 
1204 	return entry;
1205 }
1206 
1207 static void
free_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1208 free_hash_entry(struct ftrace_hash *hash,
1209 		  struct ftrace_func_entry *entry)
1210 {
1211 	hlist_del(&entry->hlist);
1212 	kfree(entry);
1213 	hash->count--;
1214 }
1215 
1216 static void
remove_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1217 remove_hash_entry(struct ftrace_hash *hash,
1218 		  struct ftrace_func_entry *entry)
1219 {
1220 	hlist_del_rcu(&entry->hlist);
1221 	hash->count--;
1222 }
1223 
ftrace_hash_clear(struct ftrace_hash * hash)1224 static void ftrace_hash_clear(struct ftrace_hash *hash)
1225 {
1226 	struct hlist_head *hhd;
1227 	struct hlist_node *tn;
1228 	struct ftrace_func_entry *entry;
1229 	int size = 1 << hash->size_bits;
1230 	int i;
1231 
1232 	if (!hash->count)
1233 		return;
1234 
1235 	for (i = 0; i < size; i++) {
1236 		hhd = &hash->buckets[i];
1237 		hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1238 			free_hash_entry(hash, entry);
1239 	}
1240 	FTRACE_WARN_ON(hash->count);
1241 }
1242 
free_ftrace_mod(struct ftrace_mod_load * ftrace_mod)1243 static void free_ftrace_mod(struct ftrace_mod_load *ftrace_mod)
1244 {
1245 	list_del(&ftrace_mod->list);
1246 	kfree(ftrace_mod->module);
1247 	kfree(ftrace_mod->func);
1248 	kfree(ftrace_mod);
1249 }
1250 
clear_ftrace_mod_list(struct list_head * head)1251 static void clear_ftrace_mod_list(struct list_head *head)
1252 {
1253 	struct ftrace_mod_load *p, *n;
1254 
1255 	/* stack tracer isn't supported yet */
1256 	if (!head)
1257 		return;
1258 
1259 	mutex_lock(&ftrace_lock);
1260 	list_for_each_entry_safe(p, n, head, list)
1261 		free_ftrace_mod(p);
1262 	mutex_unlock(&ftrace_lock);
1263 }
1264 
free_ftrace_hash(struct ftrace_hash * hash)1265 static void free_ftrace_hash(struct ftrace_hash *hash)
1266 {
1267 	if (!hash || hash == EMPTY_HASH)
1268 		return;
1269 	ftrace_hash_clear(hash);
1270 	kfree(hash->buckets);
1271 	kfree(hash);
1272 }
1273 
__free_ftrace_hash_rcu(struct rcu_head * rcu)1274 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1275 {
1276 	struct ftrace_hash *hash;
1277 
1278 	hash = container_of(rcu, struct ftrace_hash, rcu);
1279 	free_ftrace_hash(hash);
1280 }
1281 
free_ftrace_hash_rcu(struct ftrace_hash * hash)1282 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1283 {
1284 	if (!hash || hash == EMPTY_HASH)
1285 		return;
1286 	call_rcu(&hash->rcu, __free_ftrace_hash_rcu);
1287 }
1288 
1289 /**
1290  * ftrace_free_filter - remove all filters for an ftrace_ops
1291  * @ops: the ops to remove the filters from
1292  */
ftrace_free_filter(struct ftrace_ops * ops)1293 void ftrace_free_filter(struct ftrace_ops *ops)
1294 {
1295 	ftrace_ops_init(ops);
1296 	free_ftrace_hash(ops->func_hash->filter_hash);
1297 	free_ftrace_hash(ops->func_hash->notrace_hash);
1298 }
1299 EXPORT_SYMBOL_GPL(ftrace_free_filter);
1300 
alloc_ftrace_hash(int size_bits)1301 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1302 {
1303 	struct ftrace_hash *hash;
1304 	int size;
1305 
1306 	hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1307 	if (!hash)
1308 		return NULL;
1309 
1310 	size = 1 << size_bits;
1311 	hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1312 
1313 	if (!hash->buckets) {
1314 		kfree(hash);
1315 		return NULL;
1316 	}
1317 
1318 	hash->size_bits = size_bits;
1319 
1320 	return hash;
1321 }
1322 
1323 /* Used to save filters on functions for modules not loaded yet */
ftrace_add_mod(struct trace_array * tr,const char * func,const char * module,int enable)1324 static int ftrace_add_mod(struct trace_array *tr,
1325 			  const char *func, const char *module,
1326 			  int enable)
1327 {
1328 	struct ftrace_mod_load *ftrace_mod;
1329 	struct list_head *mod_head = enable ? &tr->mod_trace : &tr->mod_notrace;
1330 
1331 	ftrace_mod = kzalloc(sizeof(*ftrace_mod), GFP_KERNEL);
1332 	if (!ftrace_mod)
1333 		return -ENOMEM;
1334 
1335 	INIT_LIST_HEAD(&ftrace_mod->list);
1336 	ftrace_mod->func = kstrdup(func, GFP_KERNEL);
1337 	ftrace_mod->module = kstrdup(module, GFP_KERNEL);
1338 	ftrace_mod->enable = enable;
1339 
1340 	if (!ftrace_mod->func || !ftrace_mod->module)
1341 		goto out_free;
1342 
1343 	list_add(&ftrace_mod->list, mod_head);
1344 
1345 	return 0;
1346 
1347  out_free:
1348 	free_ftrace_mod(ftrace_mod);
1349 
1350 	return -ENOMEM;
1351 }
1352 
1353 static struct ftrace_hash *
alloc_and_copy_ftrace_hash(int size_bits,struct ftrace_hash * hash)1354 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1355 {
1356 	struct ftrace_func_entry *entry;
1357 	struct ftrace_hash *new_hash;
1358 	int size;
1359 	int i;
1360 
1361 	new_hash = alloc_ftrace_hash(size_bits);
1362 	if (!new_hash)
1363 		return NULL;
1364 
1365 	if (hash)
1366 		new_hash->flags = hash->flags;
1367 
1368 	/* Empty hash? */
1369 	if (ftrace_hash_empty(hash))
1370 		return new_hash;
1371 
1372 	size = 1 << hash->size_bits;
1373 	for (i = 0; i < size; i++) {
1374 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1375 			if (add_hash_entry(new_hash, entry->ip) == NULL)
1376 				goto free_hash;
1377 		}
1378 	}
1379 
1380 	FTRACE_WARN_ON(new_hash->count != hash->count);
1381 
1382 	return new_hash;
1383 
1384  free_hash:
1385 	free_ftrace_hash(new_hash);
1386 	return NULL;
1387 }
1388 
1389 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops);
1390 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops);
1391 
1392 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1393 				       struct ftrace_hash *new_hash);
1394 
1395 /*
1396  * Allocate a new hash and remove entries from @src and move them to the new hash.
1397  * On success, the @src hash will be empty and should be freed.
1398  */
__move_hash(struct ftrace_hash * src,int size)1399 static struct ftrace_hash *__move_hash(struct ftrace_hash *src, int size)
1400 {
1401 	struct ftrace_func_entry *entry;
1402 	struct ftrace_hash *new_hash;
1403 	struct hlist_head *hhd;
1404 	struct hlist_node *tn;
1405 	int bits = 0;
1406 	int i;
1407 
1408 	/*
1409 	 * Use around half the size (max bit of it), but
1410 	 * a minimum of 2 is fine (as size of 0 or 1 both give 1 for bits).
1411 	 */
1412 	bits = fls(size / 2);
1413 
1414 	/* Don't allocate too much */
1415 	if (bits > FTRACE_HASH_MAX_BITS)
1416 		bits = FTRACE_HASH_MAX_BITS;
1417 
1418 	new_hash = alloc_ftrace_hash(bits);
1419 	if (!new_hash)
1420 		return NULL;
1421 
1422 	new_hash->flags = src->flags;
1423 
1424 	size = 1 << src->size_bits;
1425 	for (i = 0; i < size; i++) {
1426 		hhd = &src->buckets[i];
1427 		hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1428 			remove_hash_entry(src, entry);
1429 			__add_hash_entry(new_hash, entry);
1430 		}
1431 	}
1432 	return new_hash;
1433 }
1434 
1435 /* Move the @src entries to a newly allocated hash */
1436 static struct ftrace_hash *
__ftrace_hash_move(struct ftrace_hash * src)1437 __ftrace_hash_move(struct ftrace_hash *src)
1438 {
1439 	int size = src->count;
1440 
1441 	/*
1442 	 * If the new source is empty, just return the empty_hash.
1443 	 */
1444 	if (ftrace_hash_empty(src))
1445 		return EMPTY_HASH;
1446 
1447 	return __move_hash(src, size);
1448 }
1449 
1450 /**
1451  * ftrace_hash_move - move a new hash to a filter and do updates
1452  * @ops: The ops with the hash that @dst points to
1453  * @enable: True if for the filter hash, false for the notrace hash
1454  * @dst: Points to the @ops hash that should be updated
1455  * @src: The hash to update @dst with
1456  *
1457  * This is called when an ftrace_ops hash is being updated and the
1458  * the kernel needs to reflect this. Note, this only updates the kernel
1459  * function callbacks if the @ops is enabled (not to be confused with
1460  * @enable above). If the @ops is enabled, its hash determines what
1461  * callbacks get called. This function gets called when the @ops hash
1462  * is updated and it requires new callbacks.
1463  *
1464  * On success the elements of @src is moved to @dst, and @dst is updated
1465  * properly, as well as the functions determined by the @ops hashes
1466  * are now calling the @ops callback function.
1467  *
1468  * Regardless of return type, @src should be freed with free_ftrace_hash().
1469  */
1470 static int
ftrace_hash_move(struct ftrace_ops * ops,int enable,struct ftrace_hash ** dst,struct ftrace_hash * src)1471 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1472 		 struct ftrace_hash **dst, struct ftrace_hash *src)
1473 {
1474 	struct ftrace_hash *new_hash;
1475 	int ret;
1476 
1477 	/* Reject setting notrace hash on IPMODIFY ftrace_ops */
1478 	if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1479 		return -EINVAL;
1480 
1481 	new_hash = __ftrace_hash_move(src);
1482 	if (!new_hash)
1483 		return -ENOMEM;
1484 
1485 	/* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1486 	if (enable) {
1487 		/* IPMODIFY should be updated only when filter_hash updating */
1488 		ret = ftrace_hash_ipmodify_update(ops, new_hash);
1489 		if (ret < 0) {
1490 			free_ftrace_hash(new_hash);
1491 			return ret;
1492 		}
1493 	}
1494 
1495 	/*
1496 	 * Remove the current set, update the hash and add
1497 	 * them back.
1498 	 */
1499 	ftrace_hash_rec_disable_modify(ops);
1500 
1501 	rcu_assign_pointer(*dst, new_hash);
1502 
1503 	ftrace_hash_rec_enable_modify(ops);
1504 
1505 	return 0;
1506 }
1507 
hash_contains_ip(unsigned long ip,struct ftrace_ops_hash * hash)1508 static bool hash_contains_ip(unsigned long ip,
1509 			     struct ftrace_ops_hash *hash)
1510 {
1511 	/*
1512 	 * The function record is a match if it exists in the filter
1513 	 * hash and not in the notrace hash. Note, an empty hash is
1514 	 * considered a match for the filter hash, but an empty
1515 	 * notrace hash is considered not in the notrace hash.
1516 	 */
1517 	return (ftrace_hash_empty(hash->filter_hash) ||
1518 		__ftrace_lookup_ip(hash->filter_hash, ip)) &&
1519 		(ftrace_hash_empty(hash->notrace_hash) ||
1520 		 !__ftrace_lookup_ip(hash->notrace_hash, ip));
1521 }
1522 
1523 /*
1524  * Test the hashes for this ops to see if we want to call
1525  * the ops->func or not.
1526  *
1527  * It's a match if the ip is in the ops->filter_hash or
1528  * the filter_hash does not exist or is empty,
1529  *  AND
1530  * the ip is not in the ops->notrace_hash.
1531  *
1532  * This needs to be called with preemption disabled as
1533  * the hashes are freed with call_rcu().
1534  */
1535 int
ftrace_ops_test(struct ftrace_ops * ops,unsigned long ip,void * regs)1536 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1537 {
1538 	struct ftrace_ops_hash hash;
1539 	int ret;
1540 
1541 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1542 	/*
1543 	 * There's a small race when adding ops that the ftrace handler
1544 	 * that wants regs, may be called without them. We can not
1545 	 * allow that handler to be called if regs is NULL.
1546 	 */
1547 	if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1548 		return 0;
1549 #endif
1550 
1551 	rcu_assign_pointer(hash.filter_hash, ops->func_hash->filter_hash);
1552 	rcu_assign_pointer(hash.notrace_hash, ops->func_hash->notrace_hash);
1553 
1554 	if (hash_contains_ip(ip, &hash))
1555 		ret = 1;
1556 	else
1557 		ret = 0;
1558 
1559 	return ret;
1560 }
1561 
1562 /*
1563  * This is a double for. Do not use 'break' to break out of the loop,
1564  * you must use a goto.
1565  */
1566 #define do_for_each_ftrace_rec(pg, rec)					\
1567 	for (pg = ftrace_pages_start; pg; pg = pg->next) {		\
1568 		int _____i;						\
1569 		for (_____i = 0; _____i < pg->index; _____i++) {	\
1570 			rec = &pg->records[_____i];
1571 
1572 #define while_for_each_ftrace_rec()		\
1573 		}				\
1574 	}
1575 
1576 
ftrace_cmp_recs(const void * a,const void * b)1577 static int ftrace_cmp_recs(const void *a, const void *b)
1578 {
1579 	const struct dyn_ftrace *key = a;
1580 	const struct dyn_ftrace *rec = b;
1581 
1582 	if (key->flags < rec->ip)
1583 		return -1;
1584 	if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1585 		return 1;
1586 	return 0;
1587 }
1588 
lookup_rec(unsigned long start,unsigned long end)1589 static struct dyn_ftrace *lookup_rec(unsigned long start, unsigned long end)
1590 {
1591 	struct ftrace_page *pg;
1592 	struct dyn_ftrace *rec = NULL;
1593 	struct dyn_ftrace key;
1594 
1595 	key.ip = start;
1596 	key.flags = end;	/* overload flags, as it is unsigned long */
1597 
1598 	for (pg = ftrace_pages_start; pg; pg = pg->next) {
1599 		if (pg->index == 0 ||
1600 		    end < pg->records[0].ip ||
1601 		    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1602 			continue;
1603 		rec = bsearch(&key, pg->records, pg->index,
1604 			      sizeof(struct dyn_ftrace),
1605 			      ftrace_cmp_recs);
1606 		if (rec)
1607 			break;
1608 	}
1609 	return rec;
1610 }
1611 
1612 /**
1613  * ftrace_location_range - return the first address of a traced location
1614  *	if it touches the given ip range
1615  * @start: start of range to search.
1616  * @end: end of range to search (inclusive). @end points to the last byte
1617  *	to check.
1618  *
1619  * Returns: rec->ip if the related ftrace location is a least partly within
1620  * the given address range. That is, the first address of the instruction
1621  * that is either a NOP or call to the function tracer. It checks the ftrace
1622  * internal tables to determine if the address belongs or not.
1623  */
ftrace_location_range(unsigned long start,unsigned long end)1624 unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1625 {
1626 	struct dyn_ftrace *rec;
1627 	unsigned long ip = 0;
1628 
1629 	rcu_read_lock();
1630 	rec = lookup_rec(start, end);
1631 	if (rec)
1632 		ip = rec->ip;
1633 	rcu_read_unlock();
1634 
1635 	return ip;
1636 }
1637 
1638 /**
1639  * ftrace_location - return the ftrace location
1640  * @ip: the instruction pointer to check
1641  *
1642  * Returns:
1643  * * If @ip matches the ftrace location, return @ip.
1644  * * If @ip matches sym+0, return sym's ftrace location.
1645  * * Otherwise, return 0.
1646  */
ftrace_location(unsigned long ip)1647 unsigned long ftrace_location(unsigned long ip)
1648 {
1649 	unsigned long loc;
1650 	unsigned long offset;
1651 	unsigned long size;
1652 
1653 	loc = ftrace_location_range(ip, ip);
1654 	if (!loc) {
1655 		if (!kallsyms_lookup_size_offset(ip, &size, &offset))
1656 			return 0;
1657 
1658 		/* map sym+0 to __fentry__ */
1659 		if (!offset)
1660 			loc = ftrace_location_range(ip, ip + size - 1);
1661 	}
1662 	return loc;
1663 }
1664 
1665 /**
1666  * ftrace_text_reserved - return true if range contains an ftrace location
1667  * @start: start of range to search
1668  * @end: end of range to search (inclusive). @end points to the last byte to check.
1669  *
1670  * Returns: 1 if @start and @end contains a ftrace location.
1671  * That is, the instruction that is either a NOP or call to
1672  * the function tracer. It checks the ftrace internal tables to
1673  * determine if the address belongs or not.
1674  */
ftrace_text_reserved(const void * start,const void * end)1675 int ftrace_text_reserved(const void *start, const void *end)
1676 {
1677 	unsigned long ret;
1678 
1679 	ret = ftrace_location_range((unsigned long)start,
1680 				    (unsigned long)end);
1681 
1682 	return (int)!!ret;
1683 }
1684 
1685 /* Test if ops registered to this rec needs regs */
test_rec_ops_needs_regs(struct dyn_ftrace * rec)1686 static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1687 {
1688 	struct ftrace_ops *ops;
1689 	bool keep_regs = false;
1690 
1691 	for (ops = ftrace_ops_list;
1692 	     ops != &ftrace_list_end; ops = ops->next) {
1693 		/* pass rec in as regs to have non-NULL val */
1694 		if (ftrace_ops_test(ops, rec->ip, rec)) {
1695 			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1696 				keep_regs = true;
1697 				break;
1698 			}
1699 		}
1700 	}
1701 
1702 	return  keep_regs;
1703 }
1704 
1705 static struct ftrace_ops *
1706 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
1707 static struct ftrace_ops *
1708 ftrace_find_tramp_ops_any_other(struct dyn_ftrace *rec, struct ftrace_ops *op_exclude);
1709 static struct ftrace_ops *
1710 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
1711 
skip_record(struct dyn_ftrace * rec)1712 static bool skip_record(struct dyn_ftrace *rec)
1713 {
1714 	/*
1715 	 * At boot up, weak functions are set to disable. Function tracing
1716 	 * can be enabled before they are, and they still need to be disabled now.
1717 	 * If the record is disabled, still continue if it is marked as already
1718 	 * enabled (this is needed to keep the accounting working).
1719 	 */
1720 	return rec->flags & FTRACE_FL_DISABLED &&
1721 		!(rec->flags & FTRACE_FL_ENABLED);
1722 }
1723 
1724 /*
1725  * This is the main engine to the ftrace updates to the dyn_ftrace records.
1726  *
1727  * It will iterate through all the available ftrace functions
1728  * (the ones that ftrace can have callbacks to) and set the flags
1729  * in the associated dyn_ftrace records.
1730  *
1731  * @inc: If true, the functions associated to @ops are added to
1732  *       the dyn_ftrace records, otherwise they are removed.
1733  */
__ftrace_hash_rec_update(struct ftrace_ops * ops,bool inc)1734 static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1735 				     bool inc)
1736 {
1737 	struct ftrace_hash *hash;
1738 	struct ftrace_hash *notrace_hash;
1739 	struct ftrace_page *pg;
1740 	struct dyn_ftrace *rec;
1741 	bool update = false;
1742 	int count = 0;
1743 	int all = false;
1744 
1745 	/* Only update if the ops has been registered */
1746 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1747 		return false;
1748 
1749 	/*
1750 	 *   If the count is zero, we update all records.
1751 	 *   Otherwise we just update the items in the hash.
1752 	 */
1753 	hash = ops->func_hash->filter_hash;
1754 	notrace_hash = ops->func_hash->notrace_hash;
1755 	if (ftrace_hash_empty(hash))
1756 		all = true;
1757 
1758 	do_for_each_ftrace_rec(pg, rec) {
1759 		int in_notrace_hash = 0;
1760 		int in_hash = 0;
1761 		int match = 0;
1762 
1763 		if (skip_record(rec))
1764 			continue;
1765 
1766 		if (all) {
1767 			/*
1768 			 * Only the filter_hash affects all records.
1769 			 * Update if the record is not in the notrace hash.
1770 			 */
1771 			if (!notrace_hash || !ftrace_lookup_ip(notrace_hash, rec->ip))
1772 				match = 1;
1773 		} else {
1774 			in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1775 			in_notrace_hash = !!ftrace_lookup_ip(notrace_hash, rec->ip);
1776 
1777 			/*
1778 			 * We want to match all functions that are in the hash but
1779 			 * not in the other hash.
1780 			 */
1781 			if (in_hash && !in_notrace_hash)
1782 				match = 1;
1783 		}
1784 		if (!match)
1785 			continue;
1786 
1787 		if (inc) {
1788 			rec->flags++;
1789 			if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1790 				return false;
1791 
1792 			if (ops->flags & FTRACE_OPS_FL_DIRECT)
1793 				rec->flags |= FTRACE_FL_DIRECT;
1794 
1795 			/*
1796 			 * If there's only a single callback registered to a
1797 			 * function, and the ops has a trampoline registered
1798 			 * for it, then we can call it directly.
1799 			 */
1800 			if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1801 				rec->flags |= FTRACE_FL_TRAMP;
1802 			else
1803 				/*
1804 				 * If we are adding another function callback
1805 				 * to this function, and the previous had a
1806 				 * custom trampoline in use, then we need to go
1807 				 * back to the default trampoline.
1808 				 */
1809 				rec->flags &= ~FTRACE_FL_TRAMP;
1810 
1811 			/*
1812 			 * If any ops wants regs saved for this function
1813 			 * then all ops will get saved regs.
1814 			 */
1815 			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1816 				rec->flags |= FTRACE_FL_REGS;
1817 		} else {
1818 			if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1819 				return false;
1820 			rec->flags--;
1821 
1822 			/*
1823 			 * Only the internal direct_ops should have the
1824 			 * DIRECT flag set. Thus, if it is removing a
1825 			 * function, then that function should no longer
1826 			 * be direct.
1827 			 */
1828 			if (ops->flags & FTRACE_OPS_FL_DIRECT)
1829 				rec->flags &= ~FTRACE_FL_DIRECT;
1830 
1831 			/*
1832 			 * If the rec had REGS enabled and the ops that is
1833 			 * being removed had REGS set, then see if there is
1834 			 * still any ops for this record that wants regs.
1835 			 * If not, we can stop recording them.
1836 			 */
1837 			if (ftrace_rec_count(rec) > 0 &&
1838 			    rec->flags & FTRACE_FL_REGS &&
1839 			    ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1840 				if (!test_rec_ops_needs_regs(rec))
1841 					rec->flags &= ~FTRACE_FL_REGS;
1842 			}
1843 
1844 			/*
1845 			 * The TRAMP needs to be set only if rec count
1846 			 * is decremented to one, and the ops that is
1847 			 * left has a trampoline. As TRAMP can only be
1848 			 * enabled if there is only a single ops attached
1849 			 * to it.
1850 			 */
1851 			if (ftrace_rec_count(rec) == 1 &&
1852 			    ftrace_find_tramp_ops_any_other(rec, ops))
1853 				rec->flags |= FTRACE_FL_TRAMP;
1854 			else
1855 				rec->flags &= ~FTRACE_FL_TRAMP;
1856 
1857 			/*
1858 			 * flags will be cleared in ftrace_check_record()
1859 			 * if rec count is zero.
1860 			 */
1861 		}
1862 
1863 		/*
1864 		 * If the rec has a single associated ops, and ops->func can be
1865 		 * called directly, allow the call site to call via the ops.
1866 		 */
1867 		if (IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_CALL_OPS) &&
1868 		    ftrace_rec_count(rec) == 1 &&
1869 		    ftrace_ops_get_func(ops) == ops->func)
1870 			rec->flags |= FTRACE_FL_CALL_OPS;
1871 		else
1872 			rec->flags &= ~FTRACE_FL_CALL_OPS;
1873 
1874 		count++;
1875 
1876 		/* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1877 		update |= ftrace_test_record(rec, true) != FTRACE_UPDATE_IGNORE;
1878 
1879 		/* Shortcut, if we handled all records, we are done. */
1880 		if (!all && count == hash->count)
1881 			return update;
1882 	} while_for_each_ftrace_rec();
1883 
1884 	return update;
1885 }
1886 
1887 /*
1888  * This is called when an ops is removed from tracing. It will decrement
1889  * the counters of the dyn_ftrace records for all the functions that
1890  * the @ops attached to.
1891  */
ftrace_hash_rec_disable(struct ftrace_ops * ops)1892 static bool ftrace_hash_rec_disable(struct ftrace_ops *ops)
1893 {
1894 	return __ftrace_hash_rec_update(ops, false);
1895 }
1896 
1897 /*
1898  * This is called when an ops is added to tracing. It will increment
1899  * the counters of the dyn_ftrace records for all the functions that
1900  * the @ops attached to.
1901  */
ftrace_hash_rec_enable(struct ftrace_ops * ops)1902 static bool ftrace_hash_rec_enable(struct ftrace_ops *ops)
1903 {
1904 	return __ftrace_hash_rec_update(ops, true);
1905 }
1906 
1907 /*
1908  * This function will update what functions @ops traces when its filter
1909  * changes.
1910  *
1911  * The @inc states if the @ops callbacks are going to be added or removed.
1912  * When one of the @ops hashes are updated to a "new_hash" the dyn_ftrace
1913  * records are update via:
1914  *
1915  * ftrace_hash_rec_disable_modify(ops);
1916  * ops->hash = new_hash
1917  * ftrace_hash_rec_enable_modify(ops);
1918  *
1919  * Where the @ops is removed from all the records it is tracing using
1920  * its old hash. The @ops hash is updated to the new hash, and then
1921  * the @ops is added back to the records so that it is tracing all
1922  * the new functions.
1923  */
ftrace_hash_rec_update_modify(struct ftrace_ops * ops,bool inc)1924 static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops, bool inc)
1925 {
1926 	struct ftrace_ops *op;
1927 
1928 	__ftrace_hash_rec_update(ops, inc);
1929 
1930 	if (ops->func_hash != &global_ops.local_hash)
1931 		return;
1932 
1933 	/*
1934 	 * If the ops shares the global_ops hash, then we need to update
1935 	 * all ops that are enabled and use this hash.
1936 	 */
1937 	do_for_each_ftrace_op(op, ftrace_ops_list) {
1938 		/* Already done */
1939 		if (op == ops)
1940 			continue;
1941 		if (op->func_hash == &global_ops.local_hash)
1942 			__ftrace_hash_rec_update(op, inc);
1943 	} while_for_each_ftrace_op(op);
1944 }
1945 
ftrace_hash_rec_disable_modify(struct ftrace_ops * ops)1946 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops)
1947 {
1948 	ftrace_hash_rec_update_modify(ops, false);
1949 }
1950 
ftrace_hash_rec_enable_modify(struct ftrace_ops * ops)1951 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops)
1952 {
1953 	ftrace_hash_rec_update_modify(ops, true);
1954 }
1955 
1956 /*
1957  * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1958  * or no-needed to update, -EBUSY if it detects a conflict of the flag
1959  * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1960  * Note that old_hash and new_hash has below meanings
1961  *  - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1962  *  - If the hash is EMPTY_HASH, it hits nothing
1963  *  - Anything else hits the recs which match the hash entries.
1964  *
1965  * DIRECT ops does not have IPMODIFY flag, but we still need to check it
1966  * against functions with FTRACE_FL_IPMODIFY. If there is any overlap, call
1967  * ops_func(SHARE_IPMODIFY_SELF) to make sure current ops can share with
1968  * IPMODIFY. If ops_func(SHARE_IPMODIFY_SELF) returns non-zero, propagate
1969  * the return value to the caller and eventually to the owner of the DIRECT
1970  * ops.
1971  */
__ftrace_hash_update_ipmodify(struct ftrace_ops * ops,struct ftrace_hash * old_hash,struct ftrace_hash * new_hash)1972 static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1973 					 struct ftrace_hash *old_hash,
1974 					 struct ftrace_hash *new_hash)
1975 {
1976 	struct ftrace_page *pg;
1977 	struct dyn_ftrace *rec, *end = NULL;
1978 	int in_old, in_new;
1979 	bool is_ipmodify, is_direct;
1980 
1981 	/* Only update if the ops has been registered */
1982 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1983 		return 0;
1984 
1985 	is_ipmodify = ops->flags & FTRACE_OPS_FL_IPMODIFY;
1986 	is_direct = ops->flags & FTRACE_OPS_FL_DIRECT;
1987 
1988 	/* neither IPMODIFY nor DIRECT, skip */
1989 	if (!is_ipmodify && !is_direct)
1990 		return 0;
1991 
1992 	if (WARN_ON_ONCE(is_ipmodify && is_direct))
1993 		return 0;
1994 
1995 	/*
1996 	 * Since the IPMODIFY and DIRECT are very address sensitive
1997 	 * actions, we do not allow ftrace_ops to set all functions to new
1998 	 * hash.
1999 	 */
2000 	if (!new_hash || !old_hash)
2001 		return -EINVAL;
2002 
2003 	/* Update rec->flags */
2004 	do_for_each_ftrace_rec(pg, rec) {
2005 
2006 		if (rec->flags & FTRACE_FL_DISABLED)
2007 			continue;
2008 
2009 		/* We need to update only differences of filter_hash */
2010 		in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
2011 		in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
2012 		if (in_old == in_new)
2013 			continue;
2014 
2015 		if (in_new) {
2016 			if (rec->flags & FTRACE_FL_IPMODIFY) {
2017 				int ret;
2018 
2019 				/* Cannot have two ipmodify on same rec */
2020 				if (is_ipmodify)
2021 					goto rollback;
2022 
2023 				FTRACE_WARN_ON(rec->flags & FTRACE_FL_DIRECT);
2024 
2025 				/*
2026 				 * Another ops with IPMODIFY is already
2027 				 * attached. We are now attaching a direct
2028 				 * ops. Run SHARE_IPMODIFY_SELF, to check
2029 				 * whether sharing is supported.
2030 				 */
2031 				if (!ops->ops_func)
2032 					return -EBUSY;
2033 				ret = ops->ops_func(ops, FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_SELF);
2034 				if (ret)
2035 					return ret;
2036 			} else if (is_ipmodify) {
2037 				rec->flags |= FTRACE_FL_IPMODIFY;
2038 			}
2039 		} else if (is_ipmodify) {
2040 			rec->flags &= ~FTRACE_FL_IPMODIFY;
2041 		}
2042 	} while_for_each_ftrace_rec();
2043 
2044 	return 0;
2045 
2046 rollback:
2047 	end = rec;
2048 
2049 	/* Roll back what we did above */
2050 	do_for_each_ftrace_rec(pg, rec) {
2051 
2052 		if (rec->flags & FTRACE_FL_DISABLED)
2053 			continue;
2054 
2055 		if (rec == end)
2056 			return -EBUSY;
2057 
2058 		in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
2059 		in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
2060 		if (in_old == in_new)
2061 			continue;
2062 
2063 		if (in_new)
2064 			rec->flags &= ~FTRACE_FL_IPMODIFY;
2065 		else
2066 			rec->flags |= FTRACE_FL_IPMODIFY;
2067 	} while_for_each_ftrace_rec();
2068 
2069 	return -EBUSY;
2070 }
2071 
ftrace_hash_ipmodify_enable(struct ftrace_ops * ops)2072 static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
2073 {
2074 	struct ftrace_hash *hash = ops->func_hash->filter_hash;
2075 
2076 	if (ftrace_hash_empty(hash))
2077 		hash = NULL;
2078 
2079 	return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
2080 }
2081 
2082 /* Disabling always succeeds */
ftrace_hash_ipmodify_disable(struct ftrace_ops * ops)2083 static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
2084 {
2085 	struct ftrace_hash *hash = ops->func_hash->filter_hash;
2086 
2087 	if (ftrace_hash_empty(hash))
2088 		hash = NULL;
2089 
2090 	__ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
2091 }
2092 
ftrace_hash_ipmodify_update(struct ftrace_ops * ops,struct ftrace_hash * new_hash)2093 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
2094 				       struct ftrace_hash *new_hash)
2095 {
2096 	struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
2097 
2098 	if (ftrace_hash_empty(old_hash))
2099 		old_hash = NULL;
2100 
2101 	if (ftrace_hash_empty(new_hash))
2102 		new_hash = NULL;
2103 
2104 	return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
2105 }
2106 
print_ip_ins(const char * fmt,const unsigned char * p)2107 static void print_ip_ins(const char *fmt, const unsigned char *p)
2108 {
2109 	char ins[MCOUNT_INSN_SIZE];
2110 
2111 	if (copy_from_kernel_nofault(ins, p, MCOUNT_INSN_SIZE)) {
2112 		printk(KERN_CONT "%s[FAULT] %px\n", fmt, p);
2113 		return;
2114 	}
2115 
2116 	printk(KERN_CONT "%s", fmt);
2117 	pr_cont("%*phC", MCOUNT_INSN_SIZE, ins);
2118 }
2119 
2120 enum ftrace_bug_type ftrace_bug_type;
2121 const void *ftrace_expected;
2122 
print_bug_type(void)2123 static void print_bug_type(void)
2124 {
2125 	switch (ftrace_bug_type) {
2126 	case FTRACE_BUG_UNKNOWN:
2127 		break;
2128 	case FTRACE_BUG_INIT:
2129 		pr_info("Initializing ftrace call sites\n");
2130 		break;
2131 	case FTRACE_BUG_NOP:
2132 		pr_info("Setting ftrace call site to NOP\n");
2133 		break;
2134 	case FTRACE_BUG_CALL:
2135 		pr_info("Setting ftrace call site to call ftrace function\n");
2136 		break;
2137 	case FTRACE_BUG_UPDATE:
2138 		pr_info("Updating ftrace call site to call a different ftrace function\n");
2139 		break;
2140 	}
2141 }
2142 
2143 /**
2144  * ftrace_bug - report and shutdown function tracer
2145  * @failed: The failed type (EFAULT, EINVAL, EPERM)
2146  * @rec: The record that failed
2147  *
2148  * The arch code that enables or disables the function tracing
2149  * can call ftrace_bug() when it has detected a problem in
2150  * modifying the code. @failed should be one of either:
2151  * EFAULT - if the problem happens on reading the @ip address
2152  * EINVAL - if what is read at @ip is not what was expected
2153  * EPERM - if the problem happens on writing to the @ip address
2154  */
ftrace_bug(int failed,struct dyn_ftrace * rec)2155 void ftrace_bug(int failed, struct dyn_ftrace *rec)
2156 {
2157 	unsigned long ip = rec ? rec->ip : 0;
2158 
2159 	pr_info("------------[ ftrace bug ]------------\n");
2160 
2161 	switch (failed) {
2162 	case -EFAULT:
2163 		pr_info("ftrace faulted on modifying ");
2164 		print_ip_sym(KERN_INFO, ip);
2165 		break;
2166 	case -EINVAL:
2167 		pr_info("ftrace failed to modify ");
2168 		print_ip_sym(KERN_INFO, ip);
2169 		print_ip_ins(" actual:   ", (unsigned char *)ip);
2170 		pr_cont("\n");
2171 		if (ftrace_expected) {
2172 			print_ip_ins(" expected: ", ftrace_expected);
2173 			pr_cont("\n");
2174 		}
2175 		break;
2176 	case -EPERM:
2177 		pr_info("ftrace faulted on writing ");
2178 		print_ip_sym(KERN_INFO, ip);
2179 		break;
2180 	default:
2181 		pr_info("ftrace faulted on unknown error ");
2182 		print_ip_sym(KERN_INFO, ip);
2183 	}
2184 	print_bug_type();
2185 	if (rec) {
2186 		struct ftrace_ops *ops = NULL;
2187 
2188 		pr_info("ftrace record flags: %lx\n", rec->flags);
2189 		pr_cont(" (%ld)%s%s", ftrace_rec_count(rec),
2190 			rec->flags & FTRACE_FL_REGS ? " R" : "  ",
2191 			rec->flags & FTRACE_FL_CALL_OPS ? " O" : "  ");
2192 		if (rec->flags & FTRACE_FL_TRAMP_EN) {
2193 			ops = ftrace_find_tramp_ops_any(rec);
2194 			if (ops) {
2195 				do {
2196 					pr_cont("\ttramp: %pS (%pS)",
2197 						(void *)ops->trampoline,
2198 						(void *)ops->func);
2199 					ops = ftrace_find_tramp_ops_next(rec, ops);
2200 				} while (ops);
2201 			} else
2202 				pr_cont("\ttramp: ERROR!");
2203 
2204 		}
2205 		ip = ftrace_get_addr_curr(rec);
2206 		pr_cont("\n expected tramp: %lx\n", ip);
2207 	}
2208 
2209 	FTRACE_WARN_ON_ONCE(1);
2210 }
2211 
ftrace_check_record(struct dyn_ftrace * rec,bool enable,bool update)2212 static int ftrace_check_record(struct dyn_ftrace *rec, bool enable, bool update)
2213 {
2214 	unsigned long flag = 0UL;
2215 
2216 	ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2217 
2218 	if (skip_record(rec))
2219 		return FTRACE_UPDATE_IGNORE;
2220 
2221 	/*
2222 	 * If we are updating calls:
2223 	 *
2224 	 *   If the record has a ref count, then we need to enable it
2225 	 *   because someone is using it.
2226 	 *
2227 	 *   Otherwise we make sure its disabled.
2228 	 *
2229 	 * If we are disabling calls, then disable all records that
2230 	 * are enabled.
2231 	 */
2232 	if (enable && ftrace_rec_count(rec))
2233 		flag = FTRACE_FL_ENABLED;
2234 
2235 	/*
2236 	 * If enabling and the REGS flag does not match the REGS_EN, or
2237 	 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2238 	 * this record. Set flags to fail the compare against ENABLED.
2239 	 * Same for direct calls.
2240 	 */
2241 	if (flag) {
2242 		if (!(rec->flags & FTRACE_FL_REGS) !=
2243 		    !(rec->flags & FTRACE_FL_REGS_EN))
2244 			flag |= FTRACE_FL_REGS;
2245 
2246 		if (!(rec->flags & FTRACE_FL_TRAMP) !=
2247 		    !(rec->flags & FTRACE_FL_TRAMP_EN))
2248 			flag |= FTRACE_FL_TRAMP;
2249 
2250 		/*
2251 		 * Direct calls are special, as count matters.
2252 		 * We must test the record for direct, if the
2253 		 * DIRECT and DIRECT_EN do not match, but only
2254 		 * if the count is 1. That's because, if the
2255 		 * count is something other than one, we do not
2256 		 * want the direct enabled (it will be done via the
2257 		 * direct helper). But if DIRECT_EN is set, and
2258 		 * the count is not one, we need to clear it.
2259 		 *
2260 		 */
2261 		if (ftrace_rec_count(rec) == 1) {
2262 			if (!(rec->flags & FTRACE_FL_DIRECT) !=
2263 			    !(rec->flags & FTRACE_FL_DIRECT_EN))
2264 				flag |= FTRACE_FL_DIRECT;
2265 		} else if (rec->flags & FTRACE_FL_DIRECT_EN) {
2266 			flag |= FTRACE_FL_DIRECT;
2267 		}
2268 
2269 		/*
2270 		 * Ops calls are special, as count matters.
2271 		 * As with direct calls, they must only be enabled when count
2272 		 * is one, otherwise they'll be handled via the list ops.
2273 		 */
2274 		if (ftrace_rec_count(rec) == 1) {
2275 			if (!(rec->flags & FTRACE_FL_CALL_OPS) !=
2276 			    !(rec->flags & FTRACE_FL_CALL_OPS_EN))
2277 				flag |= FTRACE_FL_CALL_OPS;
2278 		} else if (rec->flags & FTRACE_FL_CALL_OPS_EN) {
2279 			flag |= FTRACE_FL_CALL_OPS;
2280 		}
2281 	}
2282 
2283 	/* If the state of this record hasn't changed, then do nothing */
2284 	if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2285 		return FTRACE_UPDATE_IGNORE;
2286 
2287 	if (flag) {
2288 		/* Save off if rec is being enabled (for return value) */
2289 		flag ^= rec->flags & FTRACE_FL_ENABLED;
2290 
2291 		if (update) {
2292 			rec->flags |= FTRACE_FL_ENABLED | FTRACE_FL_TOUCHED;
2293 			if (flag & FTRACE_FL_REGS) {
2294 				if (rec->flags & FTRACE_FL_REGS)
2295 					rec->flags |= FTRACE_FL_REGS_EN;
2296 				else
2297 					rec->flags &= ~FTRACE_FL_REGS_EN;
2298 			}
2299 			if (flag & FTRACE_FL_TRAMP) {
2300 				if (rec->flags & FTRACE_FL_TRAMP)
2301 					rec->flags |= FTRACE_FL_TRAMP_EN;
2302 				else
2303 					rec->flags &= ~FTRACE_FL_TRAMP_EN;
2304 			}
2305 
2306 			/* Keep track of anything that modifies the function */
2307 			if (rec->flags & (FTRACE_FL_DIRECT | FTRACE_FL_IPMODIFY))
2308 				rec->flags |= FTRACE_FL_MODIFIED;
2309 
2310 			if (flag & FTRACE_FL_DIRECT) {
2311 				/*
2312 				 * If there's only one user (direct_ops helper)
2313 				 * then we can call the direct function
2314 				 * directly (no ftrace trampoline).
2315 				 */
2316 				if (ftrace_rec_count(rec) == 1) {
2317 					if (rec->flags & FTRACE_FL_DIRECT)
2318 						rec->flags |= FTRACE_FL_DIRECT_EN;
2319 					else
2320 						rec->flags &= ~FTRACE_FL_DIRECT_EN;
2321 				} else {
2322 					/*
2323 					 * Can only call directly if there's
2324 					 * only one callback to the function.
2325 					 */
2326 					rec->flags &= ~FTRACE_FL_DIRECT_EN;
2327 				}
2328 			}
2329 
2330 			if (flag & FTRACE_FL_CALL_OPS) {
2331 				if (ftrace_rec_count(rec) == 1) {
2332 					if (rec->flags & FTRACE_FL_CALL_OPS)
2333 						rec->flags |= FTRACE_FL_CALL_OPS_EN;
2334 					else
2335 						rec->flags &= ~FTRACE_FL_CALL_OPS_EN;
2336 				} else {
2337 					/*
2338 					 * Can only call directly if there's
2339 					 * only one set of associated ops.
2340 					 */
2341 					rec->flags &= ~FTRACE_FL_CALL_OPS_EN;
2342 				}
2343 			}
2344 		}
2345 
2346 		/*
2347 		 * If this record is being updated from a nop, then
2348 		 *   return UPDATE_MAKE_CALL.
2349 		 * Otherwise,
2350 		 *   return UPDATE_MODIFY_CALL to tell the caller to convert
2351 		 *   from the save regs, to a non-save regs function or
2352 		 *   vice versa, or from a trampoline call.
2353 		 */
2354 		if (flag & FTRACE_FL_ENABLED) {
2355 			ftrace_bug_type = FTRACE_BUG_CALL;
2356 			return FTRACE_UPDATE_MAKE_CALL;
2357 		}
2358 
2359 		ftrace_bug_type = FTRACE_BUG_UPDATE;
2360 		return FTRACE_UPDATE_MODIFY_CALL;
2361 	}
2362 
2363 	if (update) {
2364 		/* If there's no more users, clear all flags */
2365 		if (!ftrace_rec_count(rec))
2366 			rec->flags &= FTRACE_NOCLEAR_FLAGS;
2367 		else
2368 			/*
2369 			 * Just disable the record, but keep the ops TRAMP
2370 			 * and REGS states. The _EN flags must be disabled though.
2371 			 */
2372 			rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2373 					FTRACE_FL_REGS_EN | FTRACE_FL_DIRECT_EN |
2374 					FTRACE_FL_CALL_OPS_EN);
2375 	}
2376 
2377 	ftrace_bug_type = FTRACE_BUG_NOP;
2378 	return FTRACE_UPDATE_MAKE_NOP;
2379 }
2380 
2381 /**
2382  * ftrace_update_record - set a record that now is tracing or not
2383  * @rec: the record to update
2384  * @enable: set to true if the record is tracing, false to force disable
2385  *
2386  * The records that represent all functions that can be traced need
2387  * to be updated when tracing has been enabled.
2388  */
ftrace_update_record(struct dyn_ftrace * rec,bool enable)2389 int ftrace_update_record(struct dyn_ftrace *rec, bool enable)
2390 {
2391 	return ftrace_check_record(rec, enable, true);
2392 }
2393 
2394 /**
2395  * ftrace_test_record - check if the record has been enabled or not
2396  * @rec: the record to test
2397  * @enable: set to true to check if enabled, false if it is disabled
2398  *
2399  * The arch code may need to test if a record is already set to
2400  * tracing to determine how to modify the function code that it
2401  * represents.
2402  */
ftrace_test_record(struct dyn_ftrace * rec,bool enable)2403 int ftrace_test_record(struct dyn_ftrace *rec, bool enable)
2404 {
2405 	return ftrace_check_record(rec, enable, false);
2406 }
2407 
2408 static struct ftrace_ops *
ftrace_find_tramp_ops_any(struct dyn_ftrace * rec)2409 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2410 {
2411 	struct ftrace_ops *op;
2412 	unsigned long ip = rec->ip;
2413 
2414 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2415 
2416 		if (!op->trampoline)
2417 			continue;
2418 
2419 		if (hash_contains_ip(ip, op->func_hash))
2420 			return op;
2421 	} while_for_each_ftrace_op(op);
2422 
2423 	return NULL;
2424 }
2425 
2426 static struct ftrace_ops *
ftrace_find_tramp_ops_any_other(struct dyn_ftrace * rec,struct ftrace_ops * op_exclude)2427 ftrace_find_tramp_ops_any_other(struct dyn_ftrace *rec, struct ftrace_ops *op_exclude)
2428 {
2429 	struct ftrace_ops *op;
2430 	unsigned long ip = rec->ip;
2431 
2432 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2433 
2434 		if (op == op_exclude || !op->trampoline)
2435 			continue;
2436 
2437 		if (hash_contains_ip(ip, op->func_hash))
2438 			return op;
2439 	} while_for_each_ftrace_op(op);
2440 
2441 	return NULL;
2442 }
2443 
2444 static struct ftrace_ops *
ftrace_find_tramp_ops_next(struct dyn_ftrace * rec,struct ftrace_ops * op)2445 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2446 			   struct ftrace_ops *op)
2447 {
2448 	unsigned long ip = rec->ip;
2449 
2450 	while_for_each_ftrace_op(op) {
2451 
2452 		if (!op->trampoline)
2453 			continue;
2454 
2455 		if (hash_contains_ip(ip, op->func_hash))
2456 			return op;
2457 	}
2458 
2459 	return NULL;
2460 }
2461 
2462 static struct ftrace_ops *
ftrace_find_tramp_ops_curr(struct dyn_ftrace * rec)2463 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2464 {
2465 	struct ftrace_ops *op;
2466 	unsigned long ip = rec->ip;
2467 
2468 	/*
2469 	 * Need to check removed ops first.
2470 	 * If they are being removed, and this rec has a tramp,
2471 	 * and this rec is in the ops list, then it would be the
2472 	 * one with the tramp.
2473 	 */
2474 	if (removed_ops) {
2475 		if (hash_contains_ip(ip, &removed_ops->old_hash))
2476 			return removed_ops;
2477 	}
2478 
2479 	/*
2480 	 * Need to find the current trampoline for a rec.
2481 	 * Now, a trampoline is only attached to a rec if there
2482 	 * was a single 'ops' attached to it. But this can be called
2483 	 * when we are adding another op to the rec or removing the
2484 	 * current one. Thus, if the op is being added, we can
2485 	 * ignore it because it hasn't attached itself to the rec
2486 	 * yet.
2487 	 *
2488 	 * If an ops is being modified (hooking to different functions)
2489 	 * then we don't care about the new functions that are being
2490 	 * added, just the old ones (that are probably being removed).
2491 	 *
2492 	 * If we are adding an ops to a function that already is using
2493 	 * a trampoline, it needs to be removed (trampolines are only
2494 	 * for single ops connected), then an ops that is not being
2495 	 * modified also needs to be checked.
2496 	 */
2497 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2498 
2499 		if (!op->trampoline)
2500 			continue;
2501 
2502 		/*
2503 		 * If the ops is being added, it hasn't gotten to
2504 		 * the point to be removed from this tree yet.
2505 		 */
2506 		if (op->flags & FTRACE_OPS_FL_ADDING)
2507 			continue;
2508 
2509 
2510 		/*
2511 		 * If the ops is being modified and is in the old
2512 		 * hash, then it is probably being removed from this
2513 		 * function.
2514 		 */
2515 		if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2516 		    hash_contains_ip(ip, &op->old_hash))
2517 			return op;
2518 		/*
2519 		 * If the ops is not being added or modified, and it's
2520 		 * in its normal filter hash, then this must be the one
2521 		 * we want!
2522 		 */
2523 		if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2524 		    hash_contains_ip(ip, op->func_hash))
2525 			return op;
2526 
2527 	} while_for_each_ftrace_op(op);
2528 
2529 	return NULL;
2530 }
2531 
2532 static struct ftrace_ops *
ftrace_find_tramp_ops_new(struct dyn_ftrace * rec)2533 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2534 {
2535 	struct ftrace_ops *op;
2536 	unsigned long ip = rec->ip;
2537 
2538 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2539 		/* pass rec in as regs to have non-NULL val */
2540 		if (hash_contains_ip(ip, op->func_hash))
2541 			return op;
2542 	} while_for_each_ftrace_op(op);
2543 
2544 	return NULL;
2545 }
2546 
2547 struct ftrace_ops *
ftrace_find_unique_ops(struct dyn_ftrace * rec)2548 ftrace_find_unique_ops(struct dyn_ftrace *rec)
2549 {
2550 	struct ftrace_ops *op, *found = NULL;
2551 	unsigned long ip = rec->ip;
2552 
2553 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2554 
2555 		if (hash_contains_ip(ip, op->func_hash)) {
2556 			if (found)
2557 				return NULL;
2558 			found = op;
2559 		}
2560 
2561 	} while_for_each_ftrace_op(op);
2562 
2563 	return found;
2564 }
2565 
2566 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
2567 /* Protected by rcu_tasks for reading, and direct_mutex for writing */
2568 static struct ftrace_hash __rcu *direct_functions = EMPTY_HASH;
2569 static DEFINE_MUTEX(direct_mutex);
2570 
2571 /*
2572  * Search the direct_functions hash to see if the given instruction pointer
2573  * has a direct caller attached to it.
2574  */
ftrace_find_rec_direct(unsigned long ip)2575 unsigned long ftrace_find_rec_direct(unsigned long ip)
2576 {
2577 	struct ftrace_func_entry *entry;
2578 
2579 	entry = __ftrace_lookup_ip(direct_functions, ip);
2580 	if (!entry)
2581 		return 0;
2582 
2583 	return entry->direct;
2584 }
2585 
call_direct_funcs(unsigned long ip,unsigned long pip,struct ftrace_ops * ops,struct ftrace_regs * fregs)2586 static void call_direct_funcs(unsigned long ip, unsigned long pip,
2587 			      struct ftrace_ops *ops, struct ftrace_regs *fregs)
2588 {
2589 	unsigned long addr = READ_ONCE(ops->direct_call);
2590 
2591 	if (!addr)
2592 		return;
2593 
2594 	arch_ftrace_set_direct_caller(fregs, addr);
2595 }
2596 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
2597 
2598 /**
2599  * ftrace_get_addr_new - Get the call address to set to
2600  * @rec:  The ftrace record descriptor
2601  *
2602  * If the record has the FTRACE_FL_REGS set, that means that it
2603  * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2604  * is not set, then it wants to convert to the normal callback.
2605  *
2606  * Returns: the address of the trampoline to set to
2607  */
ftrace_get_addr_new(struct dyn_ftrace * rec)2608 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2609 {
2610 	struct ftrace_ops *ops;
2611 	unsigned long addr;
2612 
2613 	if ((rec->flags & FTRACE_FL_DIRECT) &&
2614 	    (ftrace_rec_count(rec) == 1)) {
2615 		addr = ftrace_find_rec_direct(rec->ip);
2616 		if (addr)
2617 			return addr;
2618 		WARN_ON_ONCE(1);
2619 	}
2620 
2621 	/* Trampolines take precedence over regs */
2622 	if (rec->flags & FTRACE_FL_TRAMP) {
2623 		ops = ftrace_find_tramp_ops_new(rec);
2624 		if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2625 			pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2626 				(void *)rec->ip, (void *)rec->ip, rec->flags);
2627 			/* Ftrace is shutting down, return anything */
2628 			return (unsigned long)FTRACE_ADDR;
2629 		}
2630 		return ops->trampoline;
2631 	}
2632 
2633 	if (rec->flags & FTRACE_FL_REGS)
2634 		return (unsigned long)FTRACE_REGS_ADDR;
2635 	else
2636 		return (unsigned long)FTRACE_ADDR;
2637 }
2638 
2639 /**
2640  * ftrace_get_addr_curr - Get the call address that is already there
2641  * @rec:  The ftrace record descriptor
2642  *
2643  * The FTRACE_FL_REGS_EN is set when the record already points to
2644  * a function that saves all the regs. Basically the '_EN' version
2645  * represents the current state of the function.
2646  *
2647  * Returns: the address of the trampoline that is currently being called
2648  */
ftrace_get_addr_curr(struct dyn_ftrace * rec)2649 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2650 {
2651 	struct ftrace_ops *ops;
2652 	unsigned long addr;
2653 
2654 	/* Direct calls take precedence over trampolines */
2655 	if (rec->flags & FTRACE_FL_DIRECT_EN) {
2656 		addr = ftrace_find_rec_direct(rec->ip);
2657 		if (addr)
2658 			return addr;
2659 		WARN_ON_ONCE(1);
2660 	}
2661 
2662 	/* Trampolines take precedence over regs */
2663 	if (rec->flags & FTRACE_FL_TRAMP_EN) {
2664 		ops = ftrace_find_tramp_ops_curr(rec);
2665 		if (FTRACE_WARN_ON(!ops)) {
2666 			pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2667 				(void *)rec->ip, (void *)rec->ip);
2668 			/* Ftrace is shutting down, return anything */
2669 			return (unsigned long)FTRACE_ADDR;
2670 		}
2671 		return ops->trampoline;
2672 	}
2673 
2674 	if (rec->flags & FTRACE_FL_REGS_EN)
2675 		return (unsigned long)FTRACE_REGS_ADDR;
2676 	else
2677 		return (unsigned long)FTRACE_ADDR;
2678 }
2679 
2680 static int
__ftrace_replace_code(struct dyn_ftrace * rec,bool enable)2681 __ftrace_replace_code(struct dyn_ftrace *rec, bool enable)
2682 {
2683 	unsigned long ftrace_old_addr;
2684 	unsigned long ftrace_addr;
2685 	int ret;
2686 
2687 	ftrace_addr = ftrace_get_addr_new(rec);
2688 
2689 	/* This needs to be done before we call ftrace_update_record */
2690 	ftrace_old_addr = ftrace_get_addr_curr(rec);
2691 
2692 	ret = ftrace_update_record(rec, enable);
2693 
2694 	ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2695 
2696 	switch (ret) {
2697 	case FTRACE_UPDATE_IGNORE:
2698 		return 0;
2699 
2700 	case FTRACE_UPDATE_MAKE_CALL:
2701 		ftrace_bug_type = FTRACE_BUG_CALL;
2702 		return ftrace_make_call(rec, ftrace_addr);
2703 
2704 	case FTRACE_UPDATE_MAKE_NOP:
2705 		ftrace_bug_type = FTRACE_BUG_NOP;
2706 		return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2707 
2708 	case FTRACE_UPDATE_MODIFY_CALL:
2709 		ftrace_bug_type = FTRACE_BUG_UPDATE;
2710 		return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2711 	}
2712 
2713 	return -1; /* unknown ftrace bug */
2714 }
2715 
ftrace_replace_code(int mod_flags)2716 void __weak ftrace_replace_code(int mod_flags)
2717 {
2718 	struct dyn_ftrace *rec;
2719 	struct ftrace_page *pg;
2720 	bool enable = mod_flags & FTRACE_MODIFY_ENABLE_FL;
2721 	int schedulable = mod_flags & FTRACE_MODIFY_MAY_SLEEP_FL;
2722 	int failed;
2723 
2724 	if (unlikely(ftrace_disabled))
2725 		return;
2726 
2727 	do_for_each_ftrace_rec(pg, rec) {
2728 
2729 		if (skip_record(rec))
2730 			continue;
2731 
2732 		failed = __ftrace_replace_code(rec, enable);
2733 		if (failed) {
2734 			ftrace_bug(failed, rec);
2735 			/* Stop processing */
2736 			return;
2737 		}
2738 		if (schedulable)
2739 			cond_resched();
2740 	} while_for_each_ftrace_rec();
2741 }
2742 
2743 struct ftrace_rec_iter {
2744 	struct ftrace_page	*pg;
2745 	int			index;
2746 };
2747 
2748 /**
2749  * ftrace_rec_iter_start - start up iterating over traced functions
2750  *
2751  * Returns: an iterator handle that is used to iterate over all
2752  * the records that represent address locations where functions
2753  * are traced.
2754  *
2755  * May return NULL if no records are available.
2756  */
ftrace_rec_iter_start(void)2757 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2758 {
2759 	/*
2760 	 * We only use a single iterator.
2761 	 * Protected by the ftrace_lock mutex.
2762 	 */
2763 	static struct ftrace_rec_iter ftrace_rec_iter;
2764 	struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2765 
2766 	iter->pg = ftrace_pages_start;
2767 	iter->index = 0;
2768 
2769 	/* Could have empty pages */
2770 	while (iter->pg && !iter->pg->index)
2771 		iter->pg = iter->pg->next;
2772 
2773 	if (!iter->pg)
2774 		return NULL;
2775 
2776 	return iter;
2777 }
2778 
2779 /**
2780  * ftrace_rec_iter_next - get the next record to process.
2781  * @iter: The handle to the iterator.
2782  *
2783  * Returns: the next iterator after the given iterator @iter.
2784  */
ftrace_rec_iter_next(struct ftrace_rec_iter * iter)2785 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2786 {
2787 	iter->index++;
2788 
2789 	if (iter->index >= iter->pg->index) {
2790 		iter->pg = iter->pg->next;
2791 		iter->index = 0;
2792 
2793 		/* Could have empty pages */
2794 		while (iter->pg && !iter->pg->index)
2795 			iter->pg = iter->pg->next;
2796 	}
2797 
2798 	if (!iter->pg)
2799 		return NULL;
2800 
2801 	return iter;
2802 }
2803 
2804 /**
2805  * ftrace_rec_iter_record - get the record at the iterator location
2806  * @iter: The current iterator location
2807  *
2808  * Returns: the record that the current @iter is at.
2809  */
ftrace_rec_iter_record(struct ftrace_rec_iter * iter)2810 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2811 {
2812 	return &iter->pg->records[iter->index];
2813 }
2814 
2815 static int
ftrace_nop_initialize(struct module * mod,struct dyn_ftrace * rec)2816 ftrace_nop_initialize(struct module *mod, struct dyn_ftrace *rec)
2817 {
2818 	int ret;
2819 
2820 	if (unlikely(ftrace_disabled))
2821 		return 0;
2822 
2823 	ret = ftrace_init_nop(mod, rec);
2824 	if (ret) {
2825 		ftrace_bug_type = FTRACE_BUG_INIT;
2826 		ftrace_bug(ret, rec);
2827 		return 0;
2828 	}
2829 	return 1;
2830 }
2831 
2832 /*
2833  * archs can override this function if they must do something
2834  * before the modifying code is performed.
2835  */
ftrace_arch_code_modify_prepare(void)2836 void __weak ftrace_arch_code_modify_prepare(void)
2837 {
2838 }
2839 
2840 /*
2841  * archs can override this function if they must do something
2842  * after the modifying code is performed.
2843  */
ftrace_arch_code_modify_post_process(void)2844 void __weak ftrace_arch_code_modify_post_process(void)
2845 {
2846 }
2847 
update_ftrace_func(ftrace_func_t func)2848 static int update_ftrace_func(ftrace_func_t func)
2849 {
2850 	static ftrace_func_t save_func;
2851 
2852 	/* Avoid updating if it hasn't changed */
2853 	if (func == save_func)
2854 		return 0;
2855 
2856 	save_func = func;
2857 
2858 	return ftrace_update_ftrace_func(func);
2859 }
2860 
ftrace_modify_all_code(int command)2861 void ftrace_modify_all_code(int command)
2862 {
2863 	int update = command & FTRACE_UPDATE_TRACE_FUNC;
2864 	int mod_flags = 0;
2865 	int err = 0;
2866 
2867 	if (command & FTRACE_MAY_SLEEP)
2868 		mod_flags = FTRACE_MODIFY_MAY_SLEEP_FL;
2869 
2870 	/*
2871 	 * If the ftrace_caller calls a ftrace_ops func directly,
2872 	 * we need to make sure that it only traces functions it
2873 	 * expects to trace. When doing the switch of functions,
2874 	 * we need to update to the ftrace_ops_list_func first
2875 	 * before the transition between old and new calls are set,
2876 	 * as the ftrace_ops_list_func will check the ops hashes
2877 	 * to make sure the ops are having the right functions
2878 	 * traced.
2879 	 */
2880 	if (update) {
2881 		err = update_ftrace_func(ftrace_ops_list_func);
2882 		if (FTRACE_WARN_ON(err))
2883 			return;
2884 	}
2885 
2886 	if (command & FTRACE_UPDATE_CALLS)
2887 		ftrace_replace_code(mod_flags | FTRACE_MODIFY_ENABLE_FL);
2888 	else if (command & FTRACE_DISABLE_CALLS)
2889 		ftrace_replace_code(mod_flags);
2890 
2891 	if (update && ftrace_trace_function != ftrace_ops_list_func) {
2892 		function_trace_op = set_function_trace_op;
2893 		smp_wmb();
2894 		/* If irqs are disabled, we are in stop machine */
2895 		if (!irqs_disabled())
2896 			smp_call_function(ftrace_sync_ipi, NULL, 1);
2897 		err = update_ftrace_func(ftrace_trace_function);
2898 		if (FTRACE_WARN_ON(err))
2899 			return;
2900 	}
2901 
2902 	if (command & FTRACE_START_FUNC_RET)
2903 		err = ftrace_enable_ftrace_graph_caller();
2904 	else if (command & FTRACE_STOP_FUNC_RET)
2905 		err = ftrace_disable_ftrace_graph_caller();
2906 	FTRACE_WARN_ON(err);
2907 }
2908 
__ftrace_modify_code(void * data)2909 static int __ftrace_modify_code(void *data)
2910 {
2911 	int *command = data;
2912 
2913 	ftrace_modify_all_code(*command);
2914 
2915 	return 0;
2916 }
2917 
2918 /**
2919  * ftrace_run_stop_machine - go back to the stop machine method
2920  * @command: The command to tell ftrace what to do
2921  *
2922  * If an arch needs to fall back to the stop machine method, the
2923  * it can call this function.
2924  */
ftrace_run_stop_machine(int command)2925 void ftrace_run_stop_machine(int command)
2926 {
2927 	stop_machine(__ftrace_modify_code, &command, NULL);
2928 }
2929 
2930 /**
2931  * arch_ftrace_update_code - modify the code to trace or not trace
2932  * @command: The command that needs to be done
2933  *
2934  * Archs can override this function if it does not need to
2935  * run stop_machine() to modify code.
2936  */
arch_ftrace_update_code(int command)2937 void __weak arch_ftrace_update_code(int command)
2938 {
2939 	ftrace_run_stop_machine(command);
2940 }
2941 
ftrace_run_update_code(int command)2942 static void ftrace_run_update_code(int command)
2943 {
2944 	ftrace_arch_code_modify_prepare();
2945 
2946 	/*
2947 	 * By default we use stop_machine() to modify the code.
2948 	 * But archs can do what ever they want as long as it
2949 	 * is safe. The stop_machine() is the safest, but also
2950 	 * produces the most overhead.
2951 	 */
2952 	arch_ftrace_update_code(command);
2953 
2954 	ftrace_arch_code_modify_post_process();
2955 }
2956 
ftrace_run_modify_code(struct ftrace_ops * ops,int command,struct ftrace_ops_hash * old_hash)2957 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2958 				   struct ftrace_ops_hash *old_hash)
2959 {
2960 	ops->flags |= FTRACE_OPS_FL_MODIFYING;
2961 	ops->old_hash.filter_hash = old_hash->filter_hash;
2962 	ops->old_hash.notrace_hash = old_hash->notrace_hash;
2963 	ftrace_run_update_code(command);
2964 	ops->old_hash.filter_hash = NULL;
2965 	ops->old_hash.notrace_hash = NULL;
2966 	ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2967 }
2968 
2969 static ftrace_func_t saved_ftrace_func;
2970 static int ftrace_start_up;
2971 
arch_ftrace_trampoline_free(struct ftrace_ops * ops)2972 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2973 {
2974 }
2975 
2976 /* List of trace_ops that have allocated trampolines */
2977 static LIST_HEAD(ftrace_ops_trampoline_list);
2978 
ftrace_add_trampoline_to_kallsyms(struct ftrace_ops * ops)2979 static void ftrace_add_trampoline_to_kallsyms(struct ftrace_ops *ops)
2980 {
2981 	lockdep_assert_held(&ftrace_lock);
2982 	list_add_rcu(&ops->list, &ftrace_ops_trampoline_list);
2983 }
2984 
ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops * ops)2985 static void ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops *ops)
2986 {
2987 	lockdep_assert_held(&ftrace_lock);
2988 	list_del_rcu(&ops->list);
2989 	synchronize_rcu();
2990 }
2991 
2992 /*
2993  * "__builtin__ftrace" is used as a module name in /proc/kallsyms for symbols
2994  * for pages allocated for ftrace purposes, even though "__builtin__ftrace" is
2995  * not a module.
2996  */
2997 #define FTRACE_TRAMPOLINE_MOD "__builtin__ftrace"
2998 #define FTRACE_TRAMPOLINE_SYM "ftrace_trampoline"
2999 
ftrace_trampoline_free(struct ftrace_ops * ops)3000 static void ftrace_trampoline_free(struct ftrace_ops *ops)
3001 {
3002 	if (ops && (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP) &&
3003 	    ops->trampoline) {
3004 		/*
3005 		 * Record the text poke event before the ksymbol unregister
3006 		 * event.
3007 		 */
3008 		perf_event_text_poke((void *)ops->trampoline,
3009 				     (void *)ops->trampoline,
3010 				     ops->trampoline_size, NULL, 0);
3011 		perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
3012 				   ops->trampoline, ops->trampoline_size,
3013 				   true, FTRACE_TRAMPOLINE_SYM);
3014 		/* Remove from kallsyms after the perf events */
3015 		ftrace_remove_trampoline_from_kallsyms(ops);
3016 	}
3017 
3018 	arch_ftrace_trampoline_free(ops);
3019 }
3020 
ftrace_startup_enable(int command)3021 static void ftrace_startup_enable(int command)
3022 {
3023 	if (saved_ftrace_func != ftrace_trace_function) {
3024 		saved_ftrace_func = ftrace_trace_function;
3025 		command |= FTRACE_UPDATE_TRACE_FUNC;
3026 	}
3027 
3028 	if (!command || !ftrace_enabled)
3029 		return;
3030 
3031 	ftrace_run_update_code(command);
3032 }
3033 
ftrace_startup_all(int command)3034 static void ftrace_startup_all(int command)
3035 {
3036 	update_all_ops = true;
3037 	ftrace_startup_enable(command);
3038 	update_all_ops = false;
3039 }
3040 
ftrace_startup(struct ftrace_ops * ops,int command)3041 int ftrace_startup(struct ftrace_ops *ops, int command)
3042 {
3043 	int ret;
3044 
3045 	if (unlikely(ftrace_disabled))
3046 		return -ENODEV;
3047 
3048 	ret = __register_ftrace_function(ops);
3049 	if (ret)
3050 		return ret;
3051 
3052 	ftrace_start_up++;
3053 
3054 	/*
3055 	 * Note that ftrace probes uses this to start up
3056 	 * and modify functions it will probe. But we still
3057 	 * set the ADDING flag for modification, as probes
3058 	 * do not have trampolines. If they add them in the
3059 	 * future, then the probes will need to distinguish
3060 	 * between adding and updating probes.
3061 	 */
3062 	ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
3063 
3064 	ret = ftrace_hash_ipmodify_enable(ops);
3065 	if (ret < 0) {
3066 		/* Rollback registration process */
3067 		__unregister_ftrace_function(ops);
3068 		ftrace_start_up--;
3069 		ops->flags &= ~FTRACE_OPS_FL_ENABLED;
3070 		if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
3071 			ftrace_trampoline_free(ops);
3072 		return ret;
3073 	}
3074 
3075 	if (ftrace_hash_rec_enable(ops))
3076 		command |= FTRACE_UPDATE_CALLS;
3077 
3078 	ftrace_startup_enable(command);
3079 
3080 	/*
3081 	 * If ftrace is in an undefined state, we just remove ops from list
3082 	 * to prevent the NULL pointer, instead of totally rolling it back and
3083 	 * free trampoline, because those actions could cause further damage.
3084 	 */
3085 	if (unlikely(ftrace_disabled)) {
3086 		__unregister_ftrace_function(ops);
3087 		return -ENODEV;
3088 	}
3089 
3090 	ops->flags &= ~FTRACE_OPS_FL_ADDING;
3091 
3092 	return 0;
3093 }
3094 
ftrace_shutdown(struct ftrace_ops * ops,int command)3095 int ftrace_shutdown(struct ftrace_ops *ops, int command)
3096 {
3097 	int ret;
3098 
3099 	if (unlikely(ftrace_disabled))
3100 		return -ENODEV;
3101 
3102 	ret = __unregister_ftrace_function(ops);
3103 	if (ret)
3104 		return ret;
3105 
3106 	ftrace_start_up--;
3107 	/*
3108 	 * Just warn in case of unbalance, no need to kill ftrace, it's not
3109 	 * critical but the ftrace_call callers may be never nopped again after
3110 	 * further ftrace uses.
3111 	 */
3112 	WARN_ON_ONCE(ftrace_start_up < 0);
3113 
3114 	/* Disabling ipmodify never fails */
3115 	ftrace_hash_ipmodify_disable(ops);
3116 
3117 	if (ftrace_hash_rec_disable(ops))
3118 		command |= FTRACE_UPDATE_CALLS;
3119 
3120 	ops->flags &= ~FTRACE_OPS_FL_ENABLED;
3121 
3122 	if (saved_ftrace_func != ftrace_trace_function) {
3123 		saved_ftrace_func = ftrace_trace_function;
3124 		command |= FTRACE_UPDATE_TRACE_FUNC;
3125 	}
3126 
3127 	if (!command || !ftrace_enabled)
3128 		goto out;
3129 
3130 	/*
3131 	 * If the ops uses a trampoline, then it needs to be
3132 	 * tested first on update.
3133 	 */
3134 	ops->flags |= FTRACE_OPS_FL_REMOVING;
3135 	removed_ops = ops;
3136 
3137 	/* The trampoline logic checks the old hashes */
3138 	ops->old_hash.filter_hash = ops->func_hash->filter_hash;
3139 	ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
3140 
3141 	ftrace_run_update_code(command);
3142 
3143 	/*
3144 	 * If there's no more ops registered with ftrace, run a
3145 	 * sanity check to make sure all rec flags are cleared.
3146 	 */
3147 	if (rcu_dereference_protected(ftrace_ops_list,
3148 			lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
3149 		struct ftrace_page *pg;
3150 		struct dyn_ftrace *rec;
3151 
3152 		do_for_each_ftrace_rec(pg, rec) {
3153 			if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_NOCLEAR_FLAGS))
3154 				pr_warn("  %pS flags:%lx\n",
3155 					(void *)rec->ip, rec->flags);
3156 		} while_for_each_ftrace_rec();
3157 	}
3158 
3159 	ops->old_hash.filter_hash = NULL;
3160 	ops->old_hash.notrace_hash = NULL;
3161 
3162 	removed_ops = NULL;
3163 	ops->flags &= ~FTRACE_OPS_FL_REMOVING;
3164 
3165 out:
3166 	/*
3167 	 * Dynamic ops may be freed, we must make sure that all
3168 	 * callers are done before leaving this function.
3169 	 */
3170 	if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
3171 		/*
3172 		 * We need to do a hard force of sched synchronization.
3173 		 * This is because we use preempt_disable() to do RCU, but
3174 		 * the function tracers can be called where RCU is not watching
3175 		 * (like before user_exit()). We can not rely on the RCU
3176 		 * infrastructure to do the synchronization, thus we must do it
3177 		 * ourselves.
3178 		 */
3179 		synchronize_rcu_tasks_rude();
3180 
3181 		/*
3182 		 * When the kernel is preemptive, tasks can be preempted
3183 		 * while on a ftrace trampoline. Just scheduling a task on
3184 		 * a CPU is not good enough to flush them. Calling
3185 		 * synchronize_rcu_tasks() will wait for those tasks to
3186 		 * execute and either schedule voluntarily or enter user space.
3187 		 */
3188 		synchronize_rcu_tasks();
3189 
3190 		ftrace_trampoline_free(ops);
3191 	}
3192 
3193 	return 0;
3194 }
3195 
3196 /* Simply make a copy of @src and return it */
copy_hash(struct ftrace_hash * src)3197 static struct ftrace_hash *copy_hash(struct ftrace_hash *src)
3198 {
3199 	if (ftrace_hash_empty(src))
3200 		return EMPTY_HASH;
3201 
3202 	return alloc_and_copy_ftrace_hash(src->size_bits, src);
3203 }
3204 
3205 /*
3206  * Append @new_hash entries to @hash:
3207  *
3208  *  If @hash is the EMPTY_HASH then it traces all functions and nothing
3209  *  needs to be done.
3210  *
3211  *  If @new_hash is the EMPTY_HASH, then make *hash the EMPTY_HASH so
3212  *  that it traces everything.
3213  *
3214  *  Otherwise, go through all of @new_hash and add anything that @hash
3215  *  doesn't already have, to @hash.
3216  *
3217  *  The filter_hash updates uses just the append_hash() function
3218  *  and the notrace_hash does not.
3219  */
append_hash(struct ftrace_hash ** hash,struct ftrace_hash * new_hash,int size_bits)3220 static int append_hash(struct ftrace_hash **hash, struct ftrace_hash *new_hash,
3221 		       int size_bits)
3222 {
3223 	struct ftrace_func_entry *entry;
3224 	int size;
3225 	int i;
3226 
3227 	if (*hash) {
3228 		/* An empty hash does everything */
3229 		if (ftrace_hash_empty(*hash))
3230 			return 0;
3231 	} else {
3232 		*hash = alloc_ftrace_hash(size_bits);
3233 		if (!*hash)
3234 			return -ENOMEM;
3235 	}
3236 
3237 	/* If new_hash has everything make hash have everything */
3238 	if (ftrace_hash_empty(new_hash)) {
3239 		free_ftrace_hash(*hash);
3240 		*hash = EMPTY_HASH;
3241 		return 0;
3242 	}
3243 
3244 	size = 1 << new_hash->size_bits;
3245 	for (i = 0; i < size; i++) {
3246 		hlist_for_each_entry(entry, &new_hash->buckets[i], hlist) {
3247 			/* Only add if not already in hash */
3248 			if (!__ftrace_lookup_ip(*hash, entry->ip) &&
3249 			    add_hash_entry(*hash, entry->ip) == NULL)
3250 				return -ENOMEM;
3251 		}
3252 	}
3253 	return 0;
3254 }
3255 
3256 /*
3257  * Add to @hash only those that are in both @new_hash1 and @new_hash2
3258  *
3259  * The notrace_hash updates uses just the intersect_hash() function
3260  * and the filter_hash does not.
3261  */
intersect_hash(struct ftrace_hash ** hash,struct ftrace_hash * new_hash1,struct ftrace_hash * new_hash2)3262 static int intersect_hash(struct ftrace_hash **hash, struct ftrace_hash *new_hash1,
3263 			  struct ftrace_hash *new_hash2)
3264 {
3265 	struct ftrace_func_entry *entry;
3266 	int size;
3267 	int i;
3268 
3269 	/*
3270 	 * If new_hash1 or new_hash2 is the EMPTY_HASH then make the hash
3271 	 * empty as well as empty for notrace means none are notraced.
3272 	 */
3273 	if (ftrace_hash_empty(new_hash1) || ftrace_hash_empty(new_hash2)) {
3274 		free_ftrace_hash(*hash);
3275 		*hash = EMPTY_HASH;
3276 		return 0;
3277 	}
3278 
3279 	size = 1 << new_hash1->size_bits;
3280 	for (i = 0; i < size; i++) {
3281 		hlist_for_each_entry(entry, &new_hash1->buckets[i], hlist) {
3282 			/* Only add if in both @new_hash1 and @new_hash2 */
3283 			if (__ftrace_lookup_ip(new_hash2, entry->ip) &&
3284 			    add_hash_entry(*hash, entry->ip) == NULL)
3285 				return -ENOMEM;
3286 		}
3287 	}
3288 	/* If nothing intersects, make it the empty set */
3289 	if (ftrace_hash_empty(*hash)) {
3290 		free_ftrace_hash(*hash);
3291 		*hash = EMPTY_HASH;
3292 	}
3293 	return 0;
3294 }
3295 
3296 /* Return a new hash that has a union of all @ops->filter_hash entries */
append_hashes(struct ftrace_ops * ops)3297 static struct ftrace_hash *append_hashes(struct ftrace_ops *ops)
3298 {
3299 	struct ftrace_hash *new_hash = NULL;
3300 	struct ftrace_ops *subops;
3301 	int size_bits;
3302 	int ret;
3303 
3304 	if (ops->func_hash->filter_hash)
3305 		size_bits = ops->func_hash->filter_hash->size_bits;
3306 	else
3307 		size_bits = FTRACE_HASH_DEFAULT_BITS;
3308 
3309 	list_for_each_entry(subops, &ops->subop_list, list) {
3310 		ret = append_hash(&new_hash, subops->func_hash->filter_hash, size_bits);
3311 		if (ret < 0) {
3312 			free_ftrace_hash(new_hash);
3313 			return NULL;
3314 		}
3315 		/* Nothing more to do if new_hash is empty */
3316 		if (ftrace_hash_empty(new_hash))
3317 			break;
3318 	}
3319 	/* Can't return NULL as that means this failed */
3320 	return new_hash ? : EMPTY_HASH;
3321 }
3322 
3323 /* Make @ops trace evenything except what all its subops do not trace */
intersect_hashes(struct ftrace_ops * ops)3324 static struct ftrace_hash *intersect_hashes(struct ftrace_ops *ops)
3325 {
3326 	struct ftrace_hash *new_hash = NULL;
3327 	struct ftrace_ops *subops;
3328 	int size_bits;
3329 	int ret;
3330 
3331 	list_for_each_entry(subops, &ops->subop_list, list) {
3332 		struct ftrace_hash *next_hash;
3333 
3334 		if (!new_hash) {
3335 			size_bits = subops->func_hash->notrace_hash->size_bits;
3336 			new_hash = alloc_and_copy_ftrace_hash(size_bits, ops->func_hash->notrace_hash);
3337 			if (!new_hash)
3338 				return NULL;
3339 			continue;
3340 		}
3341 		size_bits = new_hash->size_bits;
3342 		next_hash = new_hash;
3343 		new_hash = alloc_ftrace_hash(size_bits);
3344 		ret = intersect_hash(&new_hash, next_hash, subops->func_hash->notrace_hash);
3345 		free_ftrace_hash(next_hash);
3346 		if (ret < 0) {
3347 			free_ftrace_hash(new_hash);
3348 			return NULL;
3349 		}
3350 		/* Nothing more to do if new_hash is empty */
3351 		if (ftrace_hash_empty(new_hash))
3352 			break;
3353 	}
3354 	return new_hash;
3355 }
3356 
ops_equal(struct ftrace_hash * A,struct ftrace_hash * B)3357 static bool ops_equal(struct ftrace_hash *A, struct ftrace_hash *B)
3358 {
3359 	struct ftrace_func_entry *entry;
3360 	int size;
3361 	int i;
3362 
3363 	if (ftrace_hash_empty(A))
3364 		return ftrace_hash_empty(B);
3365 
3366 	if (ftrace_hash_empty(B))
3367 		return ftrace_hash_empty(A);
3368 
3369 	if (A->count != B->count)
3370 		return false;
3371 
3372 	size = 1 << A->size_bits;
3373 	for (i = 0; i < size; i++) {
3374 		hlist_for_each_entry(entry, &A->buckets[i], hlist) {
3375 			if (!__ftrace_lookup_ip(B, entry->ip))
3376 				return false;
3377 		}
3378 	}
3379 
3380 	return true;
3381 }
3382 
3383 static void ftrace_ops_update_code(struct ftrace_ops *ops,
3384 				   struct ftrace_ops_hash *old_hash);
3385 
__ftrace_hash_move_and_update_ops(struct ftrace_ops * ops,struct ftrace_hash ** orig_hash,struct ftrace_hash * hash,int enable)3386 static int __ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
3387 					     struct ftrace_hash **orig_hash,
3388 					     struct ftrace_hash *hash,
3389 					     int enable)
3390 {
3391 	struct ftrace_ops_hash old_hash_ops;
3392 	struct ftrace_hash *old_hash;
3393 	int ret;
3394 
3395 	old_hash = *orig_hash;
3396 	old_hash_ops.filter_hash = ops->func_hash->filter_hash;
3397 	old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
3398 	ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3399 	if (!ret) {
3400 		ftrace_ops_update_code(ops, &old_hash_ops);
3401 		free_ftrace_hash_rcu(old_hash);
3402 	}
3403 	return ret;
3404 }
3405 
ftrace_update_ops(struct ftrace_ops * ops,struct ftrace_hash * filter_hash,struct ftrace_hash * notrace_hash)3406 static int ftrace_update_ops(struct ftrace_ops *ops, struct ftrace_hash *filter_hash,
3407 			     struct ftrace_hash *notrace_hash)
3408 {
3409 	int ret;
3410 
3411 	if (!ops_equal(filter_hash, ops->func_hash->filter_hash)) {
3412 		ret = __ftrace_hash_move_and_update_ops(ops, &ops->func_hash->filter_hash,
3413 							filter_hash, 1);
3414 		if (ret < 0)
3415 			return ret;
3416 	}
3417 
3418 	if (!ops_equal(notrace_hash, ops->func_hash->notrace_hash)) {
3419 		ret = __ftrace_hash_move_and_update_ops(ops, &ops->func_hash->notrace_hash,
3420 							notrace_hash, 0);
3421 		if (ret < 0)
3422 			return ret;
3423 	}
3424 
3425 	return 0;
3426 }
3427 
3428 /**
3429  * ftrace_startup_subops - enable tracing for subops of an ops
3430  * @ops: Manager ops (used to pick all the functions of its subops)
3431  * @subops: A new ops to add to @ops
3432  * @command: Extra commands to use to enable tracing
3433  *
3434  * The @ops is a manager @ops that has the filter that includes all the functions
3435  * that its list of subops are tracing. Adding a new @subops will add the
3436  * functions of @subops to @ops.
3437  */
ftrace_startup_subops(struct ftrace_ops * ops,struct ftrace_ops * subops,int command)3438 int ftrace_startup_subops(struct ftrace_ops *ops, struct ftrace_ops *subops, int command)
3439 {
3440 	struct ftrace_hash *filter_hash;
3441 	struct ftrace_hash *notrace_hash;
3442 	struct ftrace_hash *save_filter_hash;
3443 	struct ftrace_hash *save_notrace_hash;
3444 	int size_bits;
3445 	int ret;
3446 
3447 	if (unlikely(ftrace_disabled))
3448 		return -ENODEV;
3449 
3450 	ftrace_ops_init(ops);
3451 	ftrace_ops_init(subops);
3452 
3453 	if (WARN_ON_ONCE(subops->flags & FTRACE_OPS_FL_ENABLED))
3454 		return -EBUSY;
3455 
3456 	/* Make everything canonical (Just in case!) */
3457 	if (!ops->func_hash->filter_hash)
3458 		ops->func_hash->filter_hash = EMPTY_HASH;
3459 	if (!ops->func_hash->notrace_hash)
3460 		ops->func_hash->notrace_hash = EMPTY_HASH;
3461 	if (!subops->func_hash->filter_hash)
3462 		subops->func_hash->filter_hash = EMPTY_HASH;
3463 	if (!subops->func_hash->notrace_hash)
3464 		subops->func_hash->notrace_hash = EMPTY_HASH;
3465 
3466 	/* For the first subops to ops just enable it normally */
3467 	if (list_empty(&ops->subop_list)) {
3468 		/* Just use the subops hashes */
3469 		filter_hash = copy_hash(subops->func_hash->filter_hash);
3470 		notrace_hash = copy_hash(subops->func_hash->notrace_hash);
3471 		if (!filter_hash || !notrace_hash) {
3472 			free_ftrace_hash(filter_hash);
3473 			free_ftrace_hash(notrace_hash);
3474 			return -ENOMEM;
3475 		}
3476 
3477 		save_filter_hash = ops->func_hash->filter_hash;
3478 		save_notrace_hash = ops->func_hash->notrace_hash;
3479 
3480 		ops->func_hash->filter_hash = filter_hash;
3481 		ops->func_hash->notrace_hash = notrace_hash;
3482 		list_add(&subops->list, &ops->subop_list);
3483 		ret = ftrace_startup(ops, command);
3484 		if (ret < 0) {
3485 			list_del(&subops->list);
3486 			ops->func_hash->filter_hash = save_filter_hash;
3487 			ops->func_hash->notrace_hash = save_notrace_hash;
3488 			free_ftrace_hash(filter_hash);
3489 			free_ftrace_hash(notrace_hash);
3490 		} else {
3491 			free_ftrace_hash(save_filter_hash);
3492 			free_ftrace_hash(save_notrace_hash);
3493 			subops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP;
3494 			subops->managed = ops;
3495 		}
3496 		return ret;
3497 	}
3498 
3499 	/*
3500 	 * Here there's already something attached. Here are the rules:
3501 	 *   o If either filter_hash is empty then the final stays empty
3502 	 *      o Otherwise, the final is a superset of both hashes
3503 	 *   o If either notrace_hash is empty then the final stays empty
3504 	 *      o Otherwise, the final is an intersection between the hashes
3505 	 */
3506 	if (ftrace_hash_empty(ops->func_hash->filter_hash) ||
3507 	    ftrace_hash_empty(subops->func_hash->filter_hash)) {
3508 		filter_hash = EMPTY_HASH;
3509 	} else {
3510 		size_bits = max(ops->func_hash->filter_hash->size_bits,
3511 				subops->func_hash->filter_hash->size_bits);
3512 		filter_hash = alloc_and_copy_ftrace_hash(size_bits, ops->func_hash->filter_hash);
3513 		if (!filter_hash)
3514 			return -ENOMEM;
3515 		ret = append_hash(&filter_hash, subops->func_hash->filter_hash,
3516 				  size_bits);
3517 		if (ret < 0) {
3518 			free_ftrace_hash(filter_hash);
3519 			return ret;
3520 		}
3521 	}
3522 
3523 	if (ftrace_hash_empty(ops->func_hash->notrace_hash) ||
3524 	    ftrace_hash_empty(subops->func_hash->notrace_hash)) {
3525 		notrace_hash = EMPTY_HASH;
3526 	} else {
3527 		size_bits = max(ops->func_hash->filter_hash->size_bits,
3528 				subops->func_hash->filter_hash->size_bits);
3529 		notrace_hash = alloc_ftrace_hash(size_bits);
3530 		if (!notrace_hash) {
3531 			free_ftrace_hash(filter_hash);
3532 			return -ENOMEM;
3533 		}
3534 
3535 		ret = intersect_hash(&notrace_hash, ops->func_hash->filter_hash,
3536 				     subops->func_hash->filter_hash);
3537 		if (ret < 0) {
3538 			free_ftrace_hash(filter_hash);
3539 			free_ftrace_hash(notrace_hash);
3540 			return ret;
3541 		}
3542 	}
3543 
3544 	list_add(&subops->list, &ops->subop_list);
3545 
3546 	ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3547 	free_ftrace_hash(filter_hash);
3548 	free_ftrace_hash(notrace_hash);
3549 	if (ret < 0) {
3550 		list_del(&subops->list);
3551 	} else {
3552 		subops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP;
3553 		subops->managed = ops;
3554 	}
3555 	return ret;
3556 }
3557 
3558 /**
3559  * ftrace_shutdown_subops - Remove a subops from a manager ops
3560  * @ops: A manager ops to remove @subops from
3561  * @subops: The subops to remove from @ops
3562  * @command: Any extra command flags to add to modifying the text
3563  *
3564  * Removes the functions being traced by the @subops from @ops. Note, it
3565  * will not affect functions that are being traced by other subops that
3566  * still exist in @ops.
3567  *
3568  * If the last subops is removed from @ops, then @ops is shutdown normally.
3569  */
ftrace_shutdown_subops(struct ftrace_ops * ops,struct ftrace_ops * subops,int command)3570 int ftrace_shutdown_subops(struct ftrace_ops *ops, struct ftrace_ops *subops, int command)
3571 {
3572 	struct ftrace_hash *filter_hash;
3573 	struct ftrace_hash *notrace_hash;
3574 	int ret;
3575 
3576 	if (unlikely(ftrace_disabled))
3577 		return -ENODEV;
3578 
3579 	if (WARN_ON_ONCE(!(subops->flags & FTRACE_OPS_FL_ENABLED)))
3580 		return -EINVAL;
3581 
3582 	list_del(&subops->list);
3583 
3584 	if (list_empty(&ops->subop_list)) {
3585 		/* Last one, just disable the current ops */
3586 
3587 		ret = ftrace_shutdown(ops, command);
3588 		if (ret < 0) {
3589 			list_add(&subops->list, &ops->subop_list);
3590 			return ret;
3591 		}
3592 
3593 		subops->flags &= ~FTRACE_OPS_FL_ENABLED;
3594 
3595 		free_ftrace_hash(ops->func_hash->filter_hash);
3596 		free_ftrace_hash(ops->func_hash->notrace_hash);
3597 		ops->func_hash->filter_hash = EMPTY_HASH;
3598 		ops->func_hash->notrace_hash = EMPTY_HASH;
3599 		subops->flags &= ~(FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP);
3600 		subops->managed = NULL;
3601 
3602 		return 0;
3603 	}
3604 
3605 	/* Rebuild the hashes without subops */
3606 	filter_hash = append_hashes(ops);
3607 	notrace_hash = intersect_hashes(ops);
3608 	if (!filter_hash || !notrace_hash) {
3609 		free_ftrace_hash(filter_hash);
3610 		free_ftrace_hash(notrace_hash);
3611 		list_add(&subops->list, &ops->subop_list);
3612 		return -ENOMEM;
3613 	}
3614 
3615 	ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3616 	if (ret < 0) {
3617 		list_add(&subops->list, &ops->subop_list);
3618 	} else {
3619 		subops->flags &= ~(FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP);
3620 		subops->managed = NULL;
3621 	}
3622 	free_ftrace_hash(filter_hash);
3623 	free_ftrace_hash(notrace_hash);
3624 	return ret;
3625 }
3626 
ftrace_hash_move_and_update_subops(struct ftrace_ops * subops,struct ftrace_hash ** orig_subhash,struct ftrace_hash * hash,int enable)3627 static int ftrace_hash_move_and_update_subops(struct ftrace_ops *subops,
3628 					      struct ftrace_hash **orig_subhash,
3629 					      struct ftrace_hash *hash,
3630 					      int enable)
3631 {
3632 	struct ftrace_ops *ops = subops->managed;
3633 	struct ftrace_hash **orig_hash;
3634 	struct ftrace_hash *save_hash;
3635 	struct ftrace_hash *new_hash;
3636 	int ret;
3637 
3638 	/* Manager ops can not be subops (yet) */
3639 	if (WARN_ON_ONCE(!ops || ops->flags & FTRACE_OPS_FL_SUBOP))
3640 		return -EINVAL;
3641 
3642 	/* Move the new hash over to the subops hash */
3643 	save_hash = *orig_subhash;
3644 	*orig_subhash = __ftrace_hash_move(hash);
3645 	if (!*orig_subhash) {
3646 		*orig_subhash = save_hash;
3647 		return -ENOMEM;
3648 	}
3649 
3650 	/* Create a new_hash to hold the ops new functions */
3651 	if (enable) {
3652 		orig_hash = &ops->func_hash->filter_hash;
3653 		new_hash = append_hashes(ops);
3654 	} else {
3655 		orig_hash = &ops->func_hash->notrace_hash;
3656 		new_hash = intersect_hashes(ops);
3657 	}
3658 
3659 	/* Move the hash over to the new hash */
3660 	ret = __ftrace_hash_move_and_update_ops(ops, orig_hash, new_hash, enable);
3661 
3662 	free_ftrace_hash(new_hash);
3663 
3664 	if (ret) {
3665 		/* Put back the original hash */
3666 		free_ftrace_hash_rcu(*orig_subhash);
3667 		*orig_subhash = save_hash;
3668 	} else {
3669 		free_ftrace_hash_rcu(save_hash);
3670 	}
3671 	return ret;
3672 }
3673 
3674 
3675 u64			ftrace_update_time;
3676 u64			ftrace_total_mod_time;
3677 unsigned long		ftrace_update_tot_cnt;
3678 unsigned long		ftrace_number_of_pages;
3679 unsigned long		ftrace_number_of_groups;
3680 
ops_traces_mod(struct ftrace_ops * ops)3681 static inline int ops_traces_mod(struct ftrace_ops *ops)
3682 {
3683 	/*
3684 	 * Filter_hash being empty will default to trace module.
3685 	 * But notrace hash requires a test of individual module functions.
3686 	 */
3687 	return ftrace_hash_empty(ops->func_hash->filter_hash) &&
3688 		ftrace_hash_empty(ops->func_hash->notrace_hash);
3689 }
3690 
ftrace_update_code(struct module * mod,struct ftrace_page * new_pgs)3691 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
3692 {
3693 	bool init_nop = ftrace_need_init_nop();
3694 	struct ftrace_page *pg;
3695 	struct dyn_ftrace *p;
3696 	u64 start, stop, update_time;
3697 	unsigned long update_cnt = 0;
3698 	unsigned long rec_flags = 0;
3699 	int i;
3700 
3701 	start = ftrace_now(raw_smp_processor_id());
3702 
3703 	/*
3704 	 * When a module is loaded, this function is called to convert
3705 	 * the calls to mcount in its text to nops, and also to create
3706 	 * an entry in the ftrace data. Now, if ftrace is activated
3707 	 * after this call, but before the module sets its text to
3708 	 * read-only, the modification of enabling ftrace can fail if
3709 	 * the read-only is done while ftrace is converting the calls.
3710 	 * To prevent this, the module's records are set as disabled
3711 	 * and will be enabled after the call to set the module's text
3712 	 * to read-only.
3713 	 */
3714 	if (mod)
3715 		rec_flags |= FTRACE_FL_DISABLED;
3716 
3717 	for (pg = new_pgs; pg; pg = pg->next) {
3718 
3719 		for (i = 0; i < pg->index; i++) {
3720 
3721 			/* If something went wrong, bail without enabling anything */
3722 			if (unlikely(ftrace_disabled))
3723 				return -1;
3724 
3725 			p = &pg->records[i];
3726 			p->flags = rec_flags;
3727 
3728 			/*
3729 			 * Do the initial record conversion from mcount jump
3730 			 * to the NOP instructions.
3731 			 */
3732 			if (init_nop && !ftrace_nop_initialize(mod, p))
3733 				break;
3734 
3735 			update_cnt++;
3736 		}
3737 	}
3738 
3739 	stop = ftrace_now(raw_smp_processor_id());
3740 	update_time = stop - start;
3741 	if (mod)
3742 		ftrace_total_mod_time += update_time;
3743 	else
3744 		ftrace_update_time = update_time;
3745 	ftrace_update_tot_cnt += update_cnt;
3746 
3747 	return 0;
3748 }
3749 
ftrace_allocate_records(struct ftrace_page * pg,int count)3750 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
3751 {
3752 	int order;
3753 	int pages;
3754 	int cnt;
3755 
3756 	if (WARN_ON(!count))
3757 		return -EINVAL;
3758 
3759 	/* We want to fill as much as possible, with no empty pages */
3760 	pages = DIV_ROUND_UP(count, ENTRIES_PER_PAGE);
3761 	order = fls(pages) - 1;
3762 
3763  again:
3764 	pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
3765 
3766 	if (!pg->records) {
3767 		/* if we can't allocate this size, try something smaller */
3768 		if (!order)
3769 			return -ENOMEM;
3770 		order--;
3771 		goto again;
3772 	}
3773 
3774 	ftrace_number_of_pages += 1 << order;
3775 	ftrace_number_of_groups++;
3776 
3777 	cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
3778 	pg->order = order;
3779 
3780 	if (cnt > count)
3781 		cnt = count;
3782 
3783 	return cnt;
3784 }
3785 
ftrace_free_pages(struct ftrace_page * pages)3786 static void ftrace_free_pages(struct ftrace_page *pages)
3787 {
3788 	struct ftrace_page *pg = pages;
3789 
3790 	while (pg) {
3791 		if (pg->records) {
3792 			free_pages((unsigned long)pg->records, pg->order);
3793 			ftrace_number_of_pages -= 1 << pg->order;
3794 		}
3795 		pages = pg->next;
3796 		kfree(pg);
3797 		pg = pages;
3798 		ftrace_number_of_groups--;
3799 	}
3800 }
3801 
3802 static struct ftrace_page *
ftrace_allocate_pages(unsigned long num_to_init)3803 ftrace_allocate_pages(unsigned long num_to_init)
3804 {
3805 	struct ftrace_page *start_pg;
3806 	struct ftrace_page *pg;
3807 	int cnt;
3808 
3809 	if (!num_to_init)
3810 		return NULL;
3811 
3812 	start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3813 	if (!pg)
3814 		return NULL;
3815 
3816 	/*
3817 	 * Try to allocate as much as possible in one continues
3818 	 * location that fills in all of the space. We want to
3819 	 * waste as little space as possible.
3820 	 */
3821 	for (;;) {
3822 		cnt = ftrace_allocate_records(pg, num_to_init);
3823 		if (cnt < 0)
3824 			goto free_pages;
3825 
3826 		num_to_init -= cnt;
3827 		if (!num_to_init)
3828 			break;
3829 
3830 		pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3831 		if (!pg->next)
3832 			goto free_pages;
3833 
3834 		pg = pg->next;
3835 	}
3836 
3837 	return start_pg;
3838 
3839  free_pages:
3840 	ftrace_free_pages(start_pg);
3841 	pr_info("ftrace: FAILED to allocate memory for functions\n");
3842 	return NULL;
3843 }
3844 
3845 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3846 
3847 struct ftrace_iterator {
3848 	loff_t				pos;
3849 	loff_t				func_pos;
3850 	loff_t				mod_pos;
3851 	struct ftrace_page		*pg;
3852 	struct dyn_ftrace		*func;
3853 	struct ftrace_func_probe	*probe;
3854 	struct ftrace_func_entry	*probe_entry;
3855 	struct trace_parser		parser;
3856 	struct ftrace_hash		*hash;
3857 	struct ftrace_ops		*ops;
3858 	struct trace_array		*tr;
3859 	struct list_head		*mod_list;
3860 	int				pidx;
3861 	int				idx;
3862 	unsigned			flags;
3863 };
3864 
3865 static void *
t_probe_next(struct seq_file * m,loff_t * pos)3866 t_probe_next(struct seq_file *m, loff_t *pos)
3867 {
3868 	struct ftrace_iterator *iter = m->private;
3869 	struct trace_array *tr = iter->ops->private;
3870 	struct list_head *func_probes;
3871 	struct ftrace_hash *hash;
3872 	struct list_head *next;
3873 	struct hlist_node *hnd = NULL;
3874 	struct hlist_head *hhd;
3875 	int size;
3876 
3877 	(*pos)++;
3878 	iter->pos = *pos;
3879 
3880 	if (!tr)
3881 		return NULL;
3882 
3883 	func_probes = &tr->func_probes;
3884 	if (list_empty(func_probes))
3885 		return NULL;
3886 
3887 	if (!iter->probe) {
3888 		next = func_probes->next;
3889 		iter->probe = list_entry(next, struct ftrace_func_probe, list);
3890 	}
3891 
3892 	if (iter->probe_entry)
3893 		hnd = &iter->probe_entry->hlist;
3894 
3895 	hash = iter->probe->ops.func_hash->filter_hash;
3896 
3897 	/*
3898 	 * A probe being registered may temporarily have an empty hash
3899 	 * and it's at the end of the func_probes list.
3900 	 */
3901 	if (!hash || hash == EMPTY_HASH)
3902 		return NULL;
3903 
3904 	size = 1 << hash->size_bits;
3905 
3906  retry:
3907 	if (iter->pidx >= size) {
3908 		if (iter->probe->list.next == func_probes)
3909 			return NULL;
3910 		next = iter->probe->list.next;
3911 		iter->probe = list_entry(next, struct ftrace_func_probe, list);
3912 		hash = iter->probe->ops.func_hash->filter_hash;
3913 		size = 1 << hash->size_bits;
3914 		iter->pidx = 0;
3915 	}
3916 
3917 	hhd = &hash->buckets[iter->pidx];
3918 
3919 	if (hlist_empty(hhd)) {
3920 		iter->pidx++;
3921 		hnd = NULL;
3922 		goto retry;
3923 	}
3924 
3925 	if (!hnd)
3926 		hnd = hhd->first;
3927 	else {
3928 		hnd = hnd->next;
3929 		if (!hnd) {
3930 			iter->pidx++;
3931 			goto retry;
3932 		}
3933 	}
3934 
3935 	if (WARN_ON_ONCE(!hnd))
3936 		return NULL;
3937 
3938 	iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
3939 
3940 	return iter;
3941 }
3942 
t_probe_start(struct seq_file * m,loff_t * pos)3943 static void *t_probe_start(struct seq_file *m, loff_t *pos)
3944 {
3945 	struct ftrace_iterator *iter = m->private;
3946 	void *p = NULL;
3947 	loff_t l;
3948 
3949 	if (!(iter->flags & FTRACE_ITER_DO_PROBES))
3950 		return NULL;
3951 
3952 	if (iter->mod_pos > *pos)
3953 		return NULL;
3954 
3955 	iter->probe = NULL;
3956 	iter->probe_entry = NULL;
3957 	iter->pidx = 0;
3958 	for (l = 0; l <= (*pos - iter->mod_pos); ) {
3959 		p = t_probe_next(m, &l);
3960 		if (!p)
3961 			break;
3962 	}
3963 	if (!p)
3964 		return NULL;
3965 
3966 	/* Only set this if we have an item */
3967 	iter->flags |= FTRACE_ITER_PROBE;
3968 
3969 	return iter;
3970 }
3971 
3972 static int
t_probe_show(struct seq_file * m,struct ftrace_iterator * iter)3973 t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
3974 {
3975 	struct ftrace_func_entry *probe_entry;
3976 	struct ftrace_probe_ops *probe_ops;
3977 	struct ftrace_func_probe *probe;
3978 
3979 	probe = iter->probe;
3980 	probe_entry = iter->probe_entry;
3981 
3982 	if (WARN_ON_ONCE(!probe || !probe_entry))
3983 		return -EIO;
3984 
3985 	probe_ops = probe->probe_ops;
3986 
3987 	if (probe_ops->print)
3988 		return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
3989 
3990 	seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
3991 		   (void *)probe_ops->func);
3992 
3993 	return 0;
3994 }
3995 
3996 static void *
t_mod_next(struct seq_file * m,loff_t * pos)3997 t_mod_next(struct seq_file *m, loff_t *pos)
3998 {
3999 	struct ftrace_iterator *iter = m->private;
4000 	struct trace_array *tr = iter->tr;
4001 
4002 	(*pos)++;
4003 	iter->pos = *pos;
4004 
4005 	iter->mod_list = iter->mod_list->next;
4006 
4007 	if (iter->mod_list == &tr->mod_trace ||
4008 	    iter->mod_list == &tr->mod_notrace) {
4009 		iter->flags &= ~FTRACE_ITER_MOD;
4010 		return NULL;
4011 	}
4012 
4013 	iter->mod_pos = *pos;
4014 
4015 	return iter;
4016 }
4017 
t_mod_start(struct seq_file * m,loff_t * pos)4018 static void *t_mod_start(struct seq_file *m, loff_t *pos)
4019 {
4020 	struct ftrace_iterator *iter = m->private;
4021 	void *p = NULL;
4022 	loff_t l;
4023 
4024 	if (iter->func_pos > *pos)
4025 		return NULL;
4026 
4027 	iter->mod_pos = iter->func_pos;
4028 
4029 	/* probes are only available if tr is set */
4030 	if (!iter->tr)
4031 		return NULL;
4032 
4033 	for (l = 0; l <= (*pos - iter->func_pos); ) {
4034 		p = t_mod_next(m, &l);
4035 		if (!p)
4036 			break;
4037 	}
4038 	if (!p) {
4039 		iter->flags &= ~FTRACE_ITER_MOD;
4040 		return t_probe_start(m, pos);
4041 	}
4042 
4043 	/* Only set this if we have an item */
4044 	iter->flags |= FTRACE_ITER_MOD;
4045 
4046 	return iter;
4047 }
4048 
4049 static int
t_mod_show(struct seq_file * m,struct ftrace_iterator * iter)4050 t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
4051 {
4052 	struct ftrace_mod_load *ftrace_mod;
4053 	struct trace_array *tr = iter->tr;
4054 
4055 	if (WARN_ON_ONCE(!iter->mod_list) ||
4056 			 iter->mod_list == &tr->mod_trace ||
4057 			 iter->mod_list == &tr->mod_notrace)
4058 		return -EIO;
4059 
4060 	ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
4061 
4062 	if (ftrace_mod->func)
4063 		seq_printf(m, "%s", ftrace_mod->func);
4064 	else
4065 		seq_putc(m, '*');
4066 
4067 	seq_printf(m, ":mod:%s\n", ftrace_mod->module);
4068 
4069 	return 0;
4070 }
4071 
4072 static void *
t_func_next(struct seq_file * m,loff_t * pos)4073 t_func_next(struct seq_file *m, loff_t *pos)
4074 {
4075 	struct ftrace_iterator *iter = m->private;
4076 	struct dyn_ftrace *rec = NULL;
4077 
4078 	(*pos)++;
4079 
4080  retry:
4081 	if (iter->idx >= iter->pg->index) {
4082 		if (iter->pg->next) {
4083 			iter->pg = iter->pg->next;
4084 			iter->idx = 0;
4085 			goto retry;
4086 		}
4087 	} else {
4088 		rec = &iter->pg->records[iter->idx++];
4089 		if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
4090 		     !ftrace_lookup_ip(iter->hash, rec->ip)) ||
4091 
4092 		    ((iter->flags & FTRACE_ITER_ENABLED) &&
4093 		     !(rec->flags & FTRACE_FL_ENABLED)) ||
4094 
4095 		    ((iter->flags & FTRACE_ITER_TOUCHED) &&
4096 		     !(rec->flags & FTRACE_FL_TOUCHED))) {
4097 
4098 			rec = NULL;
4099 			goto retry;
4100 		}
4101 	}
4102 
4103 	if (!rec)
4104 		return NULL;
4105 
4106 	iter->pos = iter->func_pos = *pos;
4107 	iter->func = rec;
4108 
4109 	return iter;
4110 }
4111 
4112 static void *
t_next(struct seq_file * m,void * v,loff_t * pos)4113 t_next(struct seq_file *m, void *v, loff_t *pos)
4114 {
4115 	struct ftrace_iterator *iter = m->private;
4116 	loff_t l = *pos; /* t_probe_start() must use original pos */
4117 	void *ret;
4118 
4119 	if (unlikely(ftrace_disabled))
4120 		return NULL;
4121 
4122 	if (iter->flags & FTRACE_ITER_PROBE)
4123 		return t_probe_next(m, pos);
4124 
4125 	if (iter->flags & FTRACE_ITER_MOD)
4126 		return t_mod_next(m, pos);
4127 
4128 	if (iter->flags & FTRACE_ITER_PRINTALL) {
4129 		/* next must increment pos, and t_probe_start does not */
4130 		(*pos)++;
4131 		return t_mod_start(m, &l);
4132 	}
4133 
4134 	ret = t_func_next(m, pos);
4135 
4136 	if (!ret)
4137 		return t_mod_start(m, &l);
4138 
4139 	return ret;
4140 }
4141 
reset_iter_read(struct ftrace_iterator * iter)4142 static void reset_iter_read(struct ftrace_iterator *iter)
4143 {
4144 	iter->pos = 0;
4145 	iter->func_pos = 0;
4146 	iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
4147 }
4148 
t_start(struct seq_file * m,loff_t * pos)4149 static void *t_start(struct seq_file *m, loff_t *pos)
4150 {
4151 	struct ftrace_iterator *iter = m->private;
4152 	void *p = NULL;
4153 	loff_t l;
4154 
4155 	mutex_lock(&ftrace_lock);
4156 
4157 	if (unlikely(ftrace_disabled))
4158 		return NULL;
4159 
4160 	/*
4161 	 * If an lseek was done, then reset and start from beginning.
4162 	 */
4163 	if (*pos < iter->pos)
4164 		reset_iter_read(iter);
4165 
4166 	/*
4167 	 * For set_ftrace_filter reading, if we have the filter
4168 	 * off, we can short cut and just print out that all
4169 	 * functions are enabled.
4170 	 */
4171 	if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
4172 	    ftrace_hash_empty(iter->hash)) {
4173 		iter->func_pos = 1; /* Account for the message */
4174 		if (*pos > 0)
4175 			return t_mod_start(m, pos);
4176 		iter->flags |= FTRACE_ITER_PRINTALL;
4177 		/* reset in case of seek/pread */
4178 		iter->flags &= ~FTRACE_ITER_PROBE;
4179 		return iter;
4180 	}
4181 
4182 	if (iter->flags & FTRACE_ITER_MOD)
4183 		return t_mod_start(m, pos);
4184 
4185 	/*
4186 	 * Unfortunately, we need to restart at ftrace_pages_start
4187 	 * every time we let go of the ftrace_mutex. This is because
4188 	 * those pointers can change without the lock.
4189 	 */
4190 	iter->pg = ftrace_pages_start;
4191 	iter->idx = 0;
4192 	for (l = 0; l <= *pos; ) {
4193 		p = t_func_next(m, &l);
4194 		if (!p)
4195 			break;
4196 	}
4197 
4198 	if (!p)
4199 		return t_mod_start(m, pos);
4200 
4201 	return iter;
4202 }
4203 
t_stop(struct seq_file * m,void * p)4204 static void t_stop(struct seq_file *m, void *p)
4205 {
4206 	mutex_unlock(&ftrace_lock);
4207 }
4208 
4209 void * __weak
arch_ftrace_trampoline_func(struct ftrace_ops * ops,struct dyn_ftrace * rec)4210 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
4211 {
4212 	return NULL;
4213 }
4214 
add_trampoline_func(struct seq_file * m,struct ftrace_ops * ops,struct dyn_ftrace * rec)4215 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
4216 				struct dyn_ftrace *rec)
4217 {
4218 	void *ptr;
4219 
4220 	ptr = arch_ftrace_trampoline_func(ops, rec);
4221 	if (ptr)
4222 		seq_printf(m, " ->%pS", ptr);
4223 }
4224 
4225 #ifdef FTRACE_MCOUNT_MAX_OFFSET
4226 /*
4227  * Weak functions can still have an mcount/fentry that is saved in
4228  * the __mcount_loc section. These can be detected by having a
4229  * symbol offset of greater than FTRACE_MCOUNT_MAX_OFFSET, as the
4230  * symbol found by kallsyms is not the function that the mcount/fentry
4231  * is part of. The offset is much greater in these cases.
4232  *
4233  * Test the record to make sure that the ip points to a valid kallsyms
4234  * and if not, mark it disabled.
4235  */
test_for_valid_rec(struct dyn_ftrace * rec)4236 static int test_for_valid_rec(struct dyn_ftrace *rec)
4237 {
4238 	char str[KSYM_SYMBOL_LEN];
4239 	unsigned long offset;
4240 	const char *ret;
4241 
4242 	ret = kallsyms_lookup(rec->ip, NULL, &offset, NULL, str);
4243 
4244 	/* Weak functions can cause invalid addresses */
4245 	if (!ret || offset > FTRACE_MCOUNT_MAX_OFFSET) {
4246 		rec->flags |= FTRACE_FL_DISABLED;
4247 		return 0;
4248 	}
4249 	return 1;
4250 }
4251 
4252 static struct workqueue_struct *ftrace_check_wq __initdata;
4253 static struct work_struct ftrace_check_work __initdata;
4254 
4255 /*
4256  * Scan all the mcount/fentry entries to make sure they are valid.
4257  */
ftrace_check_work_func(struct work_struct * work)4258 static __init void ftrace_check_work_func(struct work_struct *work)
4259 {
4260 	struct ftrace_page *pg;
4261 	struct dyn_ftrace *rec;
4262 
4263 	mutex_lock(&ftrace_lock);
4264 	do_for_each_ftrace_rec(pg, rec) {
4265 		test_for_valid_rec(rec);
4266 	} while_for_each_ftrace_rec();
4267 	mutex_unlock(&ftrace_lock);
4268 }
4269 
ftrace_check_for_weak_functions(void)4270 static int __init ftrace_check_for_weak_functions(void)
4271 {
4272 	INIT_WORK(&ftrace_check_work, ftrace_check_work_func);
4273 
4274 	ftrace_check_wq = alloc_workqueue("ftrace_check_wq", WQ_UNBOUND, 0);
4275 
4276 	queue_work(ftrace_check_wq, &ftrace_check_work);
4277 	return 0;
4278 }
4279 
ftrace_check_sync(void)4280 static int __init ftrace_check_sync(void)
4281 {
4282 	/* Make sure the ftrace_check updates are finished */
4283 	if (ftrace_check_wq)
4284 		destroy_workqueue(ftrace_check_wq);
4285 	return 0;
4286 }
4287 
4288 late_initcall_sync(ftrace_check_sync);
4289 subsys_initcall(ftrace_check_for_weak_functions);
4290 
print_rec(struct seq_file * m,unsigned long ip)4291 static int print_rec(struct seq_file *m, unsigned long ip)
4292 {
4293 	unsigned long offset;
4294 	char str[KSYM_SYMBOL_LEN];
4295 	char *modname;
4296 	const char *ret;
4297 
4298 	ret = kallsyms_lookup(ip, NULL, &offset, &modname, str);
4299 	/* Weak functions can cause invalid addresses */
4300 	if (!ret || offset > FTRACE_MCOUNT_MAX_OFFSET) {
4301 		snprintf(str, KSYM_SYMBOL_LEN, "%s_%ld",
4302 			 FTRACE_INVALID_FUNCTION, offset);
4303 		ret = NULL;
4304 	}
4305 
4306 	seq_puts(m, str);
4307 	if (modname)
4308 		seq_printf(m, " [%s]", modname);
4309 	return ret == NULL ? -1 : 0;
4310 }
4311 #else
test_for_valid_rec(struct dyn_ftrace * rec)4312 static inline int test_for_valid_rec(struct dyn_ftrace *rec)
4313 {
4314 	return 1;
4315 }
4316 
print_rec(struct seq_file * m,unsigned long ip)4317 static inline int print_rec(struct seq_file *m, unsigned long ip)
4318 {
4319 	seq_printf(m, "%ps", (void *)ip);
4320 	return 0;
4321 }
4322 #endif
4323 
t_show(struct seq_file * m,void * v)4324 static int t_show(struct seq_file *m, void *v)
4325 {
4326 	struct ftrace_iterator *iter = m->private;
4327 	struct dyn_ftrace *rec;
4328 
4329 	if (iter->flags & FTRACE_ITER_PROBE)
4330 		return t_probe_show(m, iter);
4331 
4332 	if (iter->flags & FTRACE_ITER_MOD)
4333 		return t_mod_show(m, iter);
4334 
4335 	if (iter->flags & FTRACE_ITER_PRINTALL) {
4336 		if (iter->flags & FTRACE_ITER_NOTRACE)
4337 			seq_puts(m, "#### no functions disabled ####\n");
4338 		else
4339 			seq_puts(m, "#### all functions enabled ####\n");
4340 		return 0;
4341 	}
4342 
4343 	rec = iter->func;
4344 
4345 	if (!rec)
4346 		return 0;
4347 
4348 	if (iter->flags & FTRACE_ITER_ADDRS)
4349 		seq_printf(m, "%lx ", rec->ip);
4350 
4351 	if (print_rec(m, rec->ip)) {
4352 		/* This should only happen when a rec is disabled */
4353 		WARN_ON_ONCE(!(rec->flags & FTRACE_FL_DISABLED));
4354 		seq_putc(m, '\n');
4355 		return 0;
4356 	}
4357 
4358 	if (iter->flags & (FTRACE_ITER_ENABLED | FTRACE_ITER_TOUCHED)) {
4359 		struct ftrace_ops *ops;
4360 
4361 		seq_printf(m, " (%ld)%s%s%s%s%s",
4362 			   ftrace_rec_count(rec),
4363 			   rec->flags & FTRACE_FL_REGS ? " R" : "  ",
4364 			   rec->flags & FTRACE_FL_IPMODIFY ? " I" : "  ",
4365 			   rec->flags & FTRACE_FL_DIRECT ? " D" : "  ",
4366 			   rec->flags & FTRACE_FL_CALL_OPS ? " O" : "  ",
4367 			   rec->flags & FTRACE_FL_MODIFIED ? " M " : "   ");
4368 		if (rec->flags & FTRACE_FL_TRAMP_EN) {
4369 			ops = ftrace_find_tramp_ops_any(rec);
4370 			if (ops) {
4371 				do {
4372 					seq_printf(m, "\ttramp: %pS (%pS)",
4373 						   (void *)ops->trampoline,
4374 						   (void *)ops->func);
4375 					add_trampoline_func(m, ops, rec);
4376 					ops = ftrace_find_tramp_ops_next(rec, ops);
4377 				} while (ops);
4378 			} else
4379 				seq_puts(m, "\ttramp: ERROR!");
4380 		} else {
4381 			add_trampoline_func(m, NULL, rec);
4382 		}
4383 		if (rec->flags & FTRACE_FL_CALL_OPS_EN) {
4384 			ops = ftrace_find_unique_ops(rec);
4385 			if (ops) {
4386 				seq_printf(m, "\tops: %pS (%pS)",
4387 					   ops, ops->func);
4388 			} else {
4389 				seq_puts(m, "\tops: ERROR!");
4390 			}
4391 		}
4392 		if (rec->flags & FTRACE_FL_DIRECT) {
4393 			unsigned long direct;
4394 
4395 			direct = ftrace_find_rec_direct(rec->ip);
4396 			if (direct)
4397 				seq_printf(m, "\n\tdirect-->%pS", (void *)direct);
4398 		}
4399 	}
4400 
4401 	seq_putc(m, '\n');
4402 
4403 	return 0;
4404 }
4405 
4406 static const struct seq_operations show_ftrace_seq_ops = {
4407 	.start = t_start,
4408 	.next = t_next,
4409 	.stop = t_stop,
4410 	.show = t_show,
4411 };
4412 
4413 static int
ftrace_avail_open(struct inode * inode,struct file * file)4414 ftrace_avail_open(struct inode *inode, struct file *file)
4415 {
4416 	struct ftrace_iterator *iter;
4417 	int ret;
4418 
4419 	ret = security_locked_down(LOCKDOWN_TRACEFS);
4420 	if (ret)
4421 		return ret;
4422 
4423 	if (unlikely(ftrace_disabled))
4424 		return -ENODEV;
4425 
4426 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4427 	if (!iter)
4428 		return -ENOMEM;
4429 
4430 	iter->pg = ftrace_pages_start;
4431 	iter->ops = &global_ops;
4432 
4433 	return 0;
4434 }
4435 
4436 static int
ftrace_enabled_open(struct inode * inode,struct file * file)4437 ftrace_enabled_open(struct inode *inode, struct file *file)
4438 {
4439 	struct ftrace_iterator *iter;
4440 
4441 	/*
4442 	 * This shows us what functions are currently being
4443 	 * traced and by what. Not sure if we want lockdown
4444 	 * to hide such critical information for an admin.
4445 	 * Although, perhaps it can show information we don't
4446 	 * want people to see, but if something is tracing
4447 	 * something, we probably want to know about it.
4448 	 */
4449 
4450 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4451 	if (!iter)
4452 		return -ENOMEM;
4453 
4454 	iter->pg = ftrace_pages_start;
4455 	iter->flags = FTRACE_ITER_ENABLED;
4456 	iter->ops = &global_ops;
4457 
4458 	return 0;
4459 }
4460 
4461 static int
ftrace_touched_open(struct inode * inode,struct file * file)4462 ftrace_touched_open(struct inode *inode, struct file *file)
4463 {
4464 	struct ftrace_iterator *iter;
4465 
4466 	/*
4467 	 * This shows us what functions have ever been enabled
4468 	 * (traced, direct, patched, etc). Not sure if we want lockdown
4469 	 * to hide such critical information for an admin.
4470 	 * Although, perhaps it can show information we don't
4471 	 * want people to see, but if something had traced
4472 	 * something, we probably want to know about it.
4473 	 */
4474 
4475 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4476 	if (!iter)
4477 		return -ENOMEM;
4478 
4479 	iter->pg = ftrace_pages_start;
4480 	iter->flags = FTRACE_ITER_TOUCHED;
4481 	iter->ops = &global_ops;
4482 
4483 	return 0;
4484 }
4485 
4486 static int
ftrace_avail_addrs_open(struct inode * inode,struct file * file)4487 ftrace_avail_addrs_open(struct inode *inode, struct file *file)
4488 {
4489 	struct ftrace_iterator *iter;
4490 	int ret;
4491 
4492 	ret = security_locked_down(LOCKDOWN_TRACEFS);
4493 	if (ret)
4494 		return ret;
4495 
4496 	if (unlikely(ftrace_disabled))
4497 		return -ENODEV;
4498 
4499 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4500 	if (!iter)
4501 		return -ENOMEM;
4502 
4503 	iter->pg = ftrace_pages_start;
4504 	iter->flags = FTRACE_ITER_ADDRS;
4505 	iter->ops = &global_ops;
4506 
4507 	return 0;
4508 }
4509 
4510 /**
4511  * ftrace_regex_open - initialize function tracer filter files
4512  * @ops: The ftrace_ops that hold the hash filters
4513  * @flag: The type of filter to process
4514  * @inode: The inode, usually passed in to your open routine
4515  * @file: The file, usually passed in to your open routine
4516  *
4517  * ftrace_regex_open() initializes the filter files for the
4518  * @ops. Depending on @flag it may process the filter hash or
4519  * the notrace hash of @ops. With this called from the open
4520  * routine, you can use ftrace_filter_write() for the write
4521  * routine if @flag has FTRACE_ITER_FILTER set, or
4522  * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
4523  * tracing_lseek() should be used as the lseek routine, and
4524  * release must call ftrace_regex_release().
4525  *
4526  * Returns: 0 on success or a negative errno value on failure
4527  */
4528 int
ftrace_regex_open(struct ftrace_ops * ops,int flag,struct inode * inode,struct file * file)4529 ftrace_regex_open(struct ftrace_ops *ops, int flag,
4530 		  struct inode *inode, struct file *file)
4531 {
4532 	struct ftrace_iterator *iter;
4533 	struct ftrace_hash *hash;
4534 	struct list_head *mod_head;
4535 	struct trace_array *tr = ops->private;
4536 	int ret = -ENOMEM;
4537 
4538 	ftrace_ops_init(ops);
4539 
4540 	if (unlikely(ftrace_disabled))
4541 		return -ENODEV;
4542 
4543 	if (tracing_check_open_get_tr(tr))
4544 		return -ENODEV;
4545 
4546 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4547 	if (!iter)
4548 		goto out;
4549 
4550 	if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX))
4551 		goto out;
4552 
4553 	iter->ops = ops;
4554 	iter->flags = flag;
4555 	iter->tr = tr;
4556 
4557 	mutex_lock(&ops->func_hash->regex_lock);
4558 
4559 	if (flag & FTRACE_ITER_NOTRACE) {
4560 		hash = ops->func_hash->notrace_hash;
4561 		mod_head = tr ? &tr->mod_notrace : NULL;
4562 	} else {
4563 		hash = ops->func_hash->filter_hash;
4564 		mod_head = tr ? &tr->mod_trace : NULL;
4565 	}
4566 
4567 	iter->mod_list = mod_head;
4568 
4569 	if (file->f_mode & FMODE_WRITE) {
4570 		const int size_bits = FTRACE_HASH_DEFAULT_BITS;
4571 
4572 		if (file->f_flags & O_TRUNC) {
4573 			iter->hash = alloc_ftrace_hash(size_bits);
4574 			clear_ftrace_mod_list(mod_head);
4575 	        } else {
4576 			iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
4577 		}
4578 
4579 		if (!iter->hash) {
4580 			trace_parser_put(&iter->parser);
4581 			goto out_unlock;
4582 		}
4583 	} else
4584 		iter->hash = hash;
4585 
4586 	ret = 0;
4587 
4588 	if (file->f_mode & FMODE_READ) {
4589 		iter->pg = ftrace_pages_start;
4590 
4591 		ret = seq_open(file, &show_ftrace_seq_ops);
4592 		if (!ret) {
4593 			struct seq_file *m = file->private_data;
4594 			m->private = iter;
4595 		} else {
4596 			/* Failed */
4597 			free_ftrace_hash(iter->hash);
4598 			trace_parser_put(&iter->parser);
4599 		}
4600 	} else
4601 		file->private_data = iter;
4602 
4603  out_unlock:
4604 	mutex_unlock(&ops->func_hash->regex_lock);
4605 
4606  out:
4607 	if (ret) {
4608 		kfree(iter);
4609 		if (tr)
4610 			trace_array_put(tr);
4611 	}
4612 
4613 	return ret;
4614 }
4615 
4616 static int
ftrace_filter_open(struct inode * inode,struct file * file)4617 ftrace_filter_open(struct inode *inode, struct file *file)
4618 {
4619 	struct ftrace_ops *ops = inode->i_private;
4620 
4621 	/* Checks for tracefs lockdown */
4622 	return ftrace_regex_open(ops,
4623 			FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
4624 			inode, file);
4625 }
4626 
4627 static int
ftrace_notrace_open(struct inode * inode,struct file * file)4628 ftrace_notrace_open(struct inode *inode, struct file *file)
4629 {
4630 	struct ftrace_ops *ops = inode->i_private;
4631 
4632 	/* Checks for tracefs lockdown */
4633 	return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
4634 				 inode, file);
4635 }
4636 
4637 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
4638 struct ftrace_glob {
4639 	char *search;
4640 	unsigned len;
4641 	int type;
4642 };
4643 
4644 /*
4645  * If symbols in an architecture don't correspond exactly to the user-visible
4646  * name of what they represent, it is possible to define this function to
4647  * perform the necessary adjustments.
4648 */
arch_ftrace_match_adjust(char * str,const char * search)4649 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
4650 {
4651 	return str;
4652 }
4653 
ftrace_match(char * str,struct ftrace_glob * g)4654 static int ftrace_match(char *str, struct ftrace_glob *g)
4655 {
4656 	int matched = 0;
4657 	int slen;
4658 
4659 	str = arch_ftrace_match_adjust(str, g->search);
4660 
4661 	switch (g->type) {
4662 	case MATCH_FULL:
4663 		if (strcmp(str, g->search) == 0)
4664 			matched = 1;
4665 		break;
4666 	case MATCH_FRONT_ONLY:
4667 		if (strncmp(str, g->search, g->len) == 0)
4668 			matched = 1;
4669 		break;
4670 	case MATCH_MIDDLE_ONLY:
4671 		if (strstr(str, g->search))
4672 			matched = 1;
4673 		break;
4674 	case MATCH_END_ONLY:
4675 		slen = strlen(str);
4676 		if (slen >= g->len &&
4677 		    memcmp(str + slen - g->len, g->search, g->len) == 0)
4678 			matched = 1;
4679 		break;
4680 	case MATCH_GLOB:
4681 		if (glob_match(g->search, str))
4682 			matched = 1;
4683 		break;
4684 	}
4685 
4686 	return matched;
4687 }
4688 
4689 static int
enter_record(struct ftrace_hash * hash,struct dyn_ftrace * rec,int clear_filter)4690 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
4691 {
4692 	struct ftrace_func_entry *entry;
4693 	int ret = 0;
4694 
4695 	entry = ftrace_lookup_ip(hash, rec->ip);
4696 	if (clear_filter) {
4697 		/* Do nothing if it doesn't exist */
4698 		if (!entry)
4699 			return 0;
4700 
4701 		free_hash_entry(hash, entry);
4702 	} else {
4703 		/* Do nothing if it exists */
4704 		if (entry)
4705 			return 0;
4706 		if (add_hash_entry(hash, rec->ip) == NULL)
4707 			ret = -ENOMEM;
4708 	}
4709 	return ret;
4710 }
4711 
4712 static int
add_rec_by_index(struct ftrace_hash * hash,struct ftrace_glob * func_g,int clear_filter)4713 add_rec_by_index(struct ftrace_hash *hash, struct ftrace_glob *func_g,
4714 		 int clear_filter)
4715 {
4716 	long index;
4717 	struct ftrace_page *pg;
4718 	struct dyn_ftrace *rec;
4719 
4720 	/* The index starts at 1 */
4721 	if (kstrtoul(func_g->search, 0, &index) || --index < 0)
4722 		return 0;
4723 
4724 	do_for_each_ftrace_rec(pg, rec) {
4725 		if (pg->index <= index) {
4726 			index -= pg->index;
4727 			/* this is a double loop, break goes to the next page */
4728 			break;
4729 		}
4730 		rec = &pg->records[index];
4731 		enter_record(hash, rec, clear_filter);
4732 		return 1;
4733 	} while_for_each_ftrace_rec();
4734 	return 0;
4735 }
4736 
4737 #ifdef FTRACE_MCOUNT_MAX_OFFSET
lookup_ip(unsigned long ip,char ** modname,char * str)4738 static int lookup_ip(unsigned long ip, char **modname, char *str)
4739 {
4740 	unsigned long offset;
4741 
4742 	kallsyms_lookup(ip, NULL, &offset, modname, str);
4743 	if (offset > FTRACE_MCOUNT_MAX_OFFSET)
4744 		return -1;
4745 	return 0;
4746 }
4747 #else
lookup_ip(unsigned long ip,char ** modname,char * str)4748 static int lookup_ip(unsigned long ip, char **modname, char *str)
4749 {
4750 	kallsyms_lookup(ip, NULL, NULL, modname, str);
4751 	return 0;
4752 }
4753 #endif
4754 
4755 static int
ftrace_match_record(struct dyn_ftrace * rec,struct ftrace_glob * func_g,struct ftrace_glob * mod_g,int exclude_mod)4756 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
4757 		struct ftrace_glob *mod_g, int exclude_mod)
4758 {
4759 	char str[KSYM_SYMBOL_LEN];
4760 	char *modname;
4761 
4762 	if (lookup_ip(rec->ip, &modname, str)) {
4763 		/* This should only happen when a rec is disabled */
4764 		WARN_ON_ONCE(system_state == SYSTEM_RUNNING &&
4765 			     !(rec->flags & FTRACE_FL_DISABLED));
4766 		return 0;
4767 	}
4768 
4769 	if (mod_g) {
4770 		int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
4771 
4772 		/* blank module name to match all modules */
4773 		if (!mod_g->len) {
4774 			/* blank module globbing: modname xor exclude_mod */
4775 			if (!exclude_mod != !modname)
4776 				goto func_match;
4777 			return 0;
4778 		}
4779 
4780 		/*
4781 		 * exclude_mod is set to trace everything but the given
4782 		 * module. If it is set and the module matches, then
4783 		 * return 0. If it is not set, and the module doesn't match
4784 		 * also return 0. Otherwise, check the function to see if
4785 		 * that matches.
4786 		 */
4787 		if (!mod_matches == !exclude_mod)
4788 			return 0;
4789 func_match:
4790 		/* blank search means to match all funcs in the mod */
4791 		if (!func_g->len)
4792 			return 1;
4793 	}
4794 
4795 	return ftrace_match(str, func_g);
4796 }
4797 
4798 static int
match_records(struct ftrace_hash * hash,char * func,int len,char * mod)4799 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
4800 {
4801 	struct ftrace_page *pg;
4802 	struct dyn_ftrace *rec;
4803 	struct ftrace_glob func_g = { .type = MATCH_FULL };
4804 	struct ftrace_glob mod_g = { .type = MATCH_FULL };
4805 	struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
4806 	int exclude_mod = 0;
4807 	int found = 0;
4808 	int ret;
4809 	int clear_filter = 0;
4810 
4811 	if (func) {
4812 		func_g.type = filter_parse_regex(func, len, &func_g.search,
4813 						 &clear_filter);
4814 		func_g.len = strlen(func_g.search);
4815 	}
4816 
4817 	if (mod) {
4818 		mod_g.type = filter_parse_regex(mod, strlen(mod),
4819 				&mod_g.search, &exclude_mod);
4820 		mod_g.len = strlen(mod_g.search);
4821 	}
4822 
4823 	guard(mutex)(&ftrace_lock);
4824 
4825 	if (unlikely(ftrace_disabled))
4826 		return 0;
4827 
4828 	if (func_g.type == MATCH_INDEX)
4829 		return add_rec_by_index(hash, &func_g, clear_filter);
4830 
4831 	do_for_each_ftrace_rec(pg, rec) {
4832 
4833 		if (rec->flags & FTRACE_FL_DISABLED)
4834 			continue;
4835 
4836 		if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
4837 			ret = enter_record(hash, rec, clear_filter);
4838 			if (ret < 0)
4839 				return ret;
4840 			found = 1;
4841 		}
4842 		cond_resched();
4843 	} while_for_each_ftrace_rec();
4844 
4845 	return found;
4846 }
4847 
4848 static int
ftrace_match_records(struct ftrace_hash * hash,char * buff,int len)4849 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
4850 {
4851 	return match_records(hash, buff, len, NULL);
4852 }
4853 
ftrace_ops_update_code(struct ftrace_ops * ops,struct ftrace_ops_hash * old_hash)4854 static void ftrace_ops_update_code(struct ftrace_ops *ops,
4855 				   struct ftrace_ops_hash *old_hash)
4856 {
4857 	struct ftrace_ops *op;
4858 
4859 	if (!ftrace_enabled)
4860 		return;
4861 
4862 	if (ops->flags & FTRACE_OPS_FL_ENABLED) {
4863 		ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
4864 		return;
4865 	}
4866 
4867 	/*
4868 	 * If this is the shared global_ops filter, then we need to
4869 	 * check if there is another ops that shares it, is enabled.
4870 	 * If so, we still need to run the modify code.
4871 	 */
4872 	if (ops->func_hash != &global_ops.local_hash)
4873 		return;
4874 
4875 	do_for_each_ftrace_op(op, ftrace_ops_list) {
4876 		if (op->func_hash == &global_ops.local_hash &&
4877 		    op->flags & FTRACE_OPS_FL_ENABLED) {
4878 			ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
4879 			/* Only need to do this once */
4880 			return;
4881 		}
4882 	} while_for_each_ftrace_op(op);
4883 }
4884 
ftrace_hash_move_and_update_ops(struct ftrace_ops * ops,struct ftrace_hash ** orig_hash,struct ftrace_hash * hash,int enable)4885 static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
4886 					   struct ftrace_hash **orig_hash,
4887 					   struct ftrace_hash *hash,
4888 					   int enable)
4889 {
4890 	if (ops->flags & FTRACE_OPS_FL_SUBOP)
4891 		return ftrace_hash_move_and_update_subops(ops, orig_hash, hash, enable);
4892 
4893 	/*
4894 	 * If this ops is not enabled, it could be sharing its filters
4895 	 * with a subop. If that's the case, update the subop instead of
4896 	 * this ops. Shared filters are only allowed to have one ops set
4897 	 * at a time, and if we update the ops that is not enabled,
4898 	 * it will not affect subops that share it.
4899 	 */
4900 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED)) {
4901 		struct ftrace_ops *op;
4902 
4903 		/* Check if any other manager subops maps to this hash */
4904 		do_for_each_ftrace_op(op, ftrace_ops_list) {
4905 			struct ftrace_ops *subops;
4906 
4907 			list_for_each_entry(subops, &op->subop_list, list) {
4908 				if ((subops->flags & FTRACE_OPS_FL_ENABLED) &&
4909 				     subops->func_hash == ops->func_hash) {
4910 					return ftrace_hash_move_and_update_subops(subops, orig_hash, hash, enable);
4911 				}
4912 			}
4913 		} while_for_each_ftrace_op(op);
4914 	}
4915 
4916 	return __ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
4917 }
4918 
cache_mod(struct trace_array * tr,const char * func,char * module,int enable)4919 static int cache_mod(struct trace_array *tr,
4920 		     const char *func, char *module, int enable)
4921 {
4922 	struct ftrace_mod_load *ftrace_mod, *n;
4923 	struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
4924 
4925 	guard(mutex)(&ftrace_lock);
4926 
4927 	/* We do not cache inverse filters */
4928 	if (func[0] == '!') {
4929 		int ret = -EINVAL;
4930 
4931 		func++;
4932 
4933 		/* Look to remove this hash */
4934 		list_for_each_entry_safe(ftrace_mod, n, head, list) {
4935 			if (strcmp(ftrace_mod->module, module) != 0)
4936 				continue;
4937 
4938 			/* no func matches all */
4939 			if (strcmp(func, "*") == 0 ||
4940 			    (ftrace_mod->func &&
4941 			     strcmp(ftrace_mod->func, func) == 0)) {
4942 				ret = 0;
4943 				free_ftrace_mod(ftrace_mod);
4944 				continue;
4945 			}
4946 		}
4947 		return ret;
4948 	}
4949 
4950 	/* We only care about modules that have not been loaded yet */
4951 	if (module_exists(module))
4952 		return -EINVAL;
4953 
4954 	/* Save this string off, and execute it when the module is loaded */
4955 	return ftrace_add_mod(tr, func, module, enable);
4956 }
4957 
4958 #ifdef CONFIG_MODULES
process_mod_list(struct list_head * head,struct ftrace_ops * ops,char * mod,bool enable)4959 static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
4960 			     char *mod, bool enable)
4961 {
4962 	struct ftrace_mod_load *ftrace_mod, *n;
4963 	struct ftrace_hash **orig_hash, *new_hash;
4964 	LIST_HEAD(process_mods);
4965 	char *func;
4966 
4967 	mutex_lock(&ops->func_hash->regex_lock);
4968 
4969 	if (enable)
4970 		orig_hash = &ops->func_hash->filter_hash;
4971 	else
4972 		orig_hash = &ops->func_hash->notrace_hash;
4973 
4974 	new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
4975 					      *orig_hash);
4976 	if (!new_hash)
4977 		goto out; /* warn? */
4978 
4979 	mutex_lock(&ftrace_lock);
4980 
4981 	list_for_each_entry_safe(ftrace_mod, n, head, list) {
4982 
4983 		if (strcmp(ftrace_mod->module, mod) != 0)
4984 			continue;
4985 
4986 		if (ftrace_mod->func)
4987 			func = kstrdup(ftrace_mod->func, GFP_KERNEL);
4988 		else
4989 			func = kstrdup("*", GFP_KERNEL);
4990 
4991 		if (!func) /* warn? */
4992 			continue;
4993 
4994 		list_move(&ftrace_mod->list, &process_mods);
4995 
4996 		/* Use the newly allocated func, as it may be "*" */
4997 		kfree(ftrace_mod->func);
4998 		ftrace_mod->func = func;
4999 	}
5000 
5001 	mutex_unlock(&ftrace_lock);
5002 
5003 	list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
5004 
5005 		func = ftrace_mod->func;
5006 
5007 		/* Grabs ftrace_lock, which is why we have this extra step */
5008 		match_records(new_hash, func, strlen(func), mod);
5009 		free_ftrace_mod(ftrace_mod);
5010 	}
5011 
5012 	if (enable && list_empty(head))
5013 		new_hash->flags &= ~FTRACE_HASH_FL_MOD;
5014 
5015 	mutex_lock(&ftrace_lock);
5016 
5017 	ftrace_hash_move_and_update_ops(ops, orig_hash,
5018 					      new_hash, enable);
5019 	mutex_unlock(&ftrace_lock);
5020 
5021  out:
5022 	mutex_unlock(&ops->func_hash->regex_lock);
5023 
5024 	free_ftrace_hash(new_hash);
5025 }
5026 
process_cached_mods(const char * mod_name)5027 static void process_cached_mods(const char *mod_name)
5028 {
5029 	struct trace_array *tr;
5030 	char *mod;
5031 
5032 	mod = kstrdup(mod_name, GFP_KERNEL);
5033 	if (!mod)
5034 		return;
5035 
5036 	mutex_lock(&trace_types_lock);
5037 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5038 		if (!list_empty(&tr->mod_trace))
5039 			process_mod_list(&tr->mod_trace, tr->ops, mod, true);
5040 		if (!list_empty(&tr->mod_notrace))
5041 			process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
5042 	}
5043 	mutex_unlock(&trace_types_lock);
5044 
5045 	kfree(mod);
5046 }
5047 #endif
5048 
5049 /*
5050  * We register the module command as a template to show others how
5051  * to register the a command as well.
5052  */
5053 
5054 static int
ftrace_mod_callback(struct trace_array * tr,struct ftrace_hash * hash,char * func_orig,char * cmd,char * module,int enable)5055 ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
5056 		    char *func_orig, char *cmd, char *module, int enable)
5057 {
5058 	char *func;
5059 	int ret;
5060 
5061 	if (!tr)
5062 		return -ENODEV;
5063 
5064 	/* match_records() modifies func, and we need the original */
5065 	func = kstrdup(func_orig, GFP_KERNEL);
5066 	if (!func)
5067 		return -ENOMEM;
5068 
5069 	/*
5070 	 * cmd == 'mod' because we only registered this func
5071 	 * for the 'mod' ftrace_func_command.
5072 	 * But if you register one func with multiple commands,
5073 	 * you can tell which command was used by the cmd
5074 	 * parameter.
5075 	 */
5076 	ret = match_records(hash, func, strlen(func), module);
5077 	kfree(func);
5078 
5079 	if (!ret)
5080 		return cache_mod(tr, func_orig, module, enable);
5081 	if (ret < 0)
5082 		return ret;
5083 	return 0;
5084 }
5085 
5086 static struct ftrace_func_command ftrace_mod_cmd = {
5087 	.name			= "mod",
5088 	.func			= ftrace_mod_callback,
5089 };
5090 
ftrace_mod_cmd_init(void)5091 static int __init ftrace_mod_cmd_init(void)
5092 {
5093 	return register_ftrace_command(&ftrace_mod_cmd);
5094 }
5095 core_initcall(ftrace_mod_cmd_init);
5096 
function_trace_probe_call(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)5097 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
5098 				      struct ftrace_ops *op, struct ftrace_regs *fregs)
5099 {
5100 	struct ftrace_probe_ops *probe_ops;
5101 	struct ftrace_func_probe *probe;
5102 
5103 	probe = container_of(op, struct ftrace_func_probe, ops);
5104 	probe_ops = probe->probe_ops;
5105 
5106 	/*
5107 	 * Disable preemption for these calls to prevent a RCU grace
5108 	 * period. This syncs the hash iteration and freeing of items
5109 	 * on the hash. rcu_read_lock is too dangerous here.
5110 	 */
5111 	preempt_disable_notrace();
5112 	probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
5113 	preempt_enable_notrace();
5114 }
5115 
5116 struct ftrace_func_map {
5117 	struct ftrace_func_entry	entry;
5118 	void				*data;
5119 };
5120 
5121 struct ftrace_func_mapper {
5122 	struct ftrace_hash		hash;
5123 };
5124 
5125 /**
5126  * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
5127  *
5128  * Returns: a ftrace_func_mapper descriptor that can be used to map ips to data.
5129  */
allocate_ftrace_func_mapper(void)5130 struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
5131 {
5132 	struct ftrace_hash *hash;
5133 
5134 	/*
5135 	 * The mapper is simply a ftrace_hash, but since the entries
5136 	 * in the hash are not ftrace_func_entry type, we define it
5137 	 * as a separate structure.
5138 	 */
5139 	hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5140 	return (struct ftrace_func_mapper *)hash;
5141 }
5142 
5143 /**
5144  * ftrace_func_mapper_find_ip - Find some data mapped to an ip
5145  * @mapper: The mapper that has the ip maps
5146  * @ip: the instruction pointer to find the data for
5147  *
5148  * Returns: the data mapped to @ip if found otherwise NULL. The return
5149  * is actually the address of the mapper data pointer. The address is
5150  * returned for use cases where the data is no bigger than a long, and
5151  * the user can use the data pointer as its data instead of having to
5152  * allocate more memory for the reference.
5153  */
ftrace_func_mapper_find_ip(struct ftrace_func_mapper * mapper,unsigned long ip)5154 void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
5155 				  unsigned long ip)
5156 {
5157 	struct ftrace_func_entry *entry;
5158 	struct ftrace_func_map *map;
5159 
5160 	entry = ftrace_lookup_ip(&mapper->hash, ip);
5161 	if (!entry)
5162 		return NULL;
5163 
5164 	map = (struct ftrace_func_map *)entry;
5165 	return &map->data;
5166 }
5167 
5168 /**
5169  * ftrace_func_mapper_add_ip - Map some data to an ip
5170  * @mapper: The mapper that has the ip maps
5171  * @ip: The instruction pointer address to map @data to
5172  * @data: The data to map to @ip
5173  *
5174  * Returns: 0 on success otherwise an error.
5175  */
ftrace_func_mapper_add_ip(struct ftrace_func_mapper * mapper,unsigned long ip,void * data)5176 int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
5177 			      unsigned long ip, void *data)
5178 {
5179 	struct ftrace_func_entry *entry;
5180 	struct ftrace_func_map *map;
5181 
5182 	entry = ftrace_lookup_ip(&mapper->hash, ip);
5183 	if (entry)
5184 		return -EBUSY;
5185 
5186 	map = kmalloc(sizeof(*map), GFP_KERNEL);
5187 	if (!map)
5188 		return -ENOMEM;
5189 
5190 	map->entry.ip = ip;
5191 	map->data = data;
5192 
5193 	__add_hash_entry(&mapper->hash, &map->entry);
5194 
5195 	return 0;
5196 }
5197 
5198 /**
5199  * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
5200  * @mapper: The mapper that has the ip maps
5201  * @ip: The instruction pointer address to remove the data from
5202  *
5203  * Returns: the data if it is found, otherwise NULL.
5204  * Note, if the data pointer is used as the data itself, (see
5205  * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
5206  * if the data pointer was set to zero.
5207  */
ftrace_func_mapper_remove_ip(struct ftrace_func_mapper * mapper,unsigned long ip)5208 void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
5209 				   unsigned long ip)
5210 {
5211 	struct ftrace_func_entry *entry;
5212 	struct ftrace_func_map *map;
5213 	void *data;
5214 
5215 	entry = ftrace_lookup_ip(&mapper->hash, ip);
5216 	if (!entry)
5217 		return NULL;
5218 
5219 	map = (struct ftrace_func_map *)entry;
5220 	data = map->data;
5221 
5222 	remove_hash_entry(&mapper->hash, entry);
5223 	kfree(entry);
5224 
5225 	return data;
5226 }
5227 
5228 /**
5229  * free_ftrace_func_mapper - free a mapping of ips and data
5230  * @mapper: The mapper that has the ip maps
5231  * @free_func: A function to be called on each data item.
5232  *
5233  * This is used to free the function mapper. The @free_func is optional
5234  * and can be used if the data needs to be freed as well.
5235  */
free_ftrace_func_mapper(struct ftrace_func_mapper * mapper,ftrace_mapper_func free_func)5236 void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
5237 			     ftrace_mapper_func free_func)
5238 {
5239 	struct ftrace_func_entry *entry;
5240 	struct ftrace_func_map *map;
5241 	struct hlist_head *hhd;
5242 	int size, i;
5243 
5244 	if (!mapper)
5245 		return;
5246 
5247 	if (free_func && mapper->hash.count) {
5248 		size = 1 << mapper->hash.size_bits;
5249 		for (i = 0; i < size; i++) {
5250 			hhd = &mapper->hash.buckets[i];
5251 			hlist_for_each_entry(entry, hhd, hlist) {
5252 				map = (struct ftrace_func_map *)entry;
5253 				free_func(map);
5254 			}
5255 		}
5256 	}
5257 	free_ftrace_hash(&mapper->hash);
5258 }
5259 
release_probe(struct ftrace_func_probe * probe)5260 static void release_probe(struct ftrace_func_probe *probe)
5261 {
5262 	struct ftrace_probe_ops *probe_ops;
5263 
5264 	guard(mutex)(&ftrace_lock);
5265 
5266 	WARN_ON(probe->ref <= 0);
5267 
5268 	/* Subtract the ref that was used to protect this instance */
5269 	probe->ref--;
5270 
5271 	if (!probe->ref) {
5272 		probe_ops = probe->probe_ops;
5273 		/*
5274 		 * Sending zero as ip tells probe_ops to free
5275 		 * the probe->data itself
5276 		 */
5277 		if (probe_ops->free)
5278 			probe_ops->free(probe_ops, probe->tr, 0, probe->data);
5279 		list_del(&probe->list);
5280 		kfree(probe);
5281 	}
5282 }
5283 
acquire_probe_locked(struct ftrace_func_probe * probe)5284 static void acquire_probe_locked(struct ftrace_func_probe *probe)
5285 {
5286 	/*
5287 	 * Add one ref to keep it from being freed when releasing the
5288 	 * ftrace_lock mutex.
5289 	 */
5290 	probe->ref++;
5291 }
5292 
5293 int
register_ftrace_function_probe(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops,void * data)5294 register_ftrace_function_probe(char *glob, struct trace_array *tr,
5295 			       struct ftrace_probe_ops *probe_ops,
5296 			       void *data)
5297 {
5298 	struct ftrace_func_probe *probe = NULL, *iter;
5299 	struct ftrace_func_entry *entry;
5300 	struct ftrace_hash **orig_hash;
5301 	struct ftrace_hash *old_hash;
5302 	struct ftrace_hash *hash;
5303 	int count = 0;
5304 	int size;
5305 	int ret;
5306 	int i;
5307 
5308 	if (WARN_ON(!tr))
5309 		return -EINVAL;
5310 
5311 	/* We do not support '!' for function probes */
5312 	if (WARN_ON(glob[0] == '!'))
5313 		return -EINVAL;
5314 
5315 
5316 	mutex_lock(&ftrace_lock);
5317 	/* Check if the probe_ops is already registered */
5318 	list_for_each_entry(iter, &tr->func_probes, list) {
5319 		if (iter->probe_ops == probe_ops) {
5320 			probe = iter;
5321 			break;
5322 		}
5323 	}
5324 	if (!probe) {
5325 		probe = kzalloc(sizeof(*probe), GFP_KERNEL);
5326 		if (!probe) {
5327 			mutex_unlock(&ftrace_lock);
5328 			return -ENOMEM;
5329 		}
5330 		probe->probe_ops = probe_ops;
5331 		probe->ops.func = function_trace_probe_call;
5332 		probe->tr = tr;
5333 		ftrace_ops_init(&probe->ops);
5334 		list_add(&probe->list, &tr->func_probes);
5335 	}
5336 
5337 	acquire_probe_locked(probe);
5338 
5339 	mutex_unlock(&ftrace_lock);
5340 
5341 	/*
5342 	 * Note, there's a small window here that the func_hash->filter_hash
5343 	 * may be NULL or empty. Need to be careful when reading the loop.
5344 	 */
5345 	mutex_lock(&probe->ops.func_hash->regex_lock);
5346 
5347 	orig_hash = &probe->ops.func_hash->filter_hash;
5348 	old_hash = *orig_hash;
5349 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
5350 
5351 	if (!hash) {
5352 		ret = -ENOMEM;
5353 		goto out;
5354 	}
5355 
5356 	ret = ftrace_match_records(hash, glob, strlen(glob));
5357 
5358 	/* Nothing found? */
5359 	if (!ret)
5360 		ret = -EINVAL;
5361 
5362 	if (ret < 0)
5363 		goto out;
5364 
5365 	size = 1 << hash->size_bits;
5366 	for (i = 0; i < size; i++) {
5367 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5368 			if (ftrace_lookup_ip(old_hash, entry->ip))
5369 				continue;
5370 			/*
5371 			 * The caller might want to do something special
5372 			 * for each function we find. We call the callback
5373 			 * to give the caller an opportunity to do so.
5374 			 */
5375 			if (probe_ops->init) {
5376 				ret = probe_ops->init(probe_ops, tr,
5377 						      entry->ip, data,
5378 						      &probe->data);
5379 				if (ret < 0) {
5380 					if (probe_ops->free && count)
5381 						probe_ops->free(probe_ops, tr,
5382 								0, probe->data);
5383 					probe->data = NULL;
5384 					goto out;
5385 				}
5386 			}
5387 			count++;
5388 		}
5389 	}
5390 
5391 	mutex_lock(&ftrace_lock);
5392 
5393 	if (!count) {
5394 		/* Nothing was added? */
5395 		ret = -EINVAL;
5396 		goto out_unlock;
5397 	}
5398 
5399 	ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
5400 					      hash, 1);
5401 	if (ret < 0)
5402 		goto err_unlock;
5403 
5404 	/* One ref for each new function traced */
5405 	probe->ref += count;
5406 
5407 	if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
5408 		ret = ftrace_startup(&probe->ops, 0);
5409 
5410  out_unlock:
5411 	mutex_unlock(&ftrace_lock);
5412 
5413 	if (!ret)
5414 		ret = count;
5415  out:
5416 	mutex_unlock(&probe->ops.func_hash->regex_lock);
5417 	free_ftrace_hash(hash);
5418 
5419 	release_probe(probe);
5420 
5421 	return ret;
5422 
5423  err_unlock:
5424 	if (!probe_ops->free || !count)
5425 		goto out_unlock;
5426 
5427 	/* Failed to do the move, need to call the free functions */
5428 	for (i = 0; i < size; i++) {
5429 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5430 			if (ftrace_lookup_ip(old_hash, entry->ip))
5431 				continue;
5432 			probe_ops->free(probe_ops, tr, entry->ip, probe->data);
5433 		}
5434 	}
5435 	goto out_unlock;
5436 }
5437 
5438 int
unregister_ftrace_function_probe_func(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops)5439 unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
5440 				      struct ftrace_probe_ops *probe_ops)
5441 {
5442 	struct ftrace_func_probe *probe = NULL, *iter;
5443 	struct ftrace_ops_hash old_hash_ops;
5444 	struct ftrace_func_entry *entry;
5445 	struct ftrace_glob func_g;
5446 	struct ftrace_hash **orig_hash;
5447 	struct ftrace_hash *old_hash;
5448 	struct ftrace_hash *hash = NULL;
5449 	struct hlist_node *tmp;
5450 	struct hlist_head hhd;
5451 	char str[KSYM_SYMBOL_LEN];
5452 	int count = 0;
5453 	int i, ret = -ENODEV;
5454 	int size;
5455 
5456 	if (!glob || !strlen(glob) || !strcmp(glob, "*"))
5457 		func_g.search = NULL;
5458 	else {
5459 		int not;
5460 
5461 		func_g.type = filter_parse_regex(glob, strlen(glob),
5462 						 &func_g.search, &not);
5463 		func_g.len = strlen(func_g.search);
5464 
5465 		/* we do not support '!' for function probes */
5466 		if (WARN_ON(not))
5467 			return -EINVAL;
5468 	}
5469 
5470 	mutex_lock(&ftrace_lock);
5471 	/* Check if the probe_ops is already registered */
5472 	list_for_each_entry(iter, &tr->func_probes, list) {
5473 		if (iter->probe_ops == probe_ops) {
5474 			probe = iter;
5475 			break;
5476 		}
5477 	}
5478 	if (!probe)
5479 		goto err_unlock_ftrace;
5480 
5481 	ret = -EINVAL;
5482 	if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
5483 		goto err_unlock_ftrace;
5484 
5485 	acquire_probe_locked(probe);
5486 
5487 	mutex_unlock(&ftrace_lock);
5488 
5489 	mutex_lock(&probe->ops.func_hash->regex_lock);
5490 
5491 	orig_hash = &probe->ops.func_hash->filter_hash;
5492 	old_hash = *orig_hash;
5493 
5494 	if (ftrace_hash_empty(old_hash))
5495 		goto out_unlock;
5496 
5497 	old_hash_ops.filter_hash = old_hash;
5498 	/* Probes only have filters */
5499 	old_hash_ops.notrace_hash = NULL;
5500 
5501 	ret = -ENOMEM;
5502 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
5503 	if (!hash)
5504 		goto out_unlock;
5505 
5506 	INIT_HLIST_HEAD(&hhd);
5507 
5508 	size = 1 << hash->size_bits;
5509 	for (i = 0; i < size; i++) {
5510 		hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
5511 
5512 			if (func_g.search) {
5513 				kallsyms_lookup(entry->ip, NULL, NULL,
5514 						NULL, str);
5515 				if (!ftrace_match(str, &func_g))
5516 					continue;
5517 			}
5518 			count++;
5519 			remove_hash_entry(hash, entry);
5520 			hlist_add_head(&entry->hlist, &hhd);
5521 		}
5522 	}
5523 
5524 	/* Nothing found? */
5525 	if (!count) {
5526 		ret = -EINVAL;
5527 		goto out_unlock;
5528 	}
5529 
5530 	mutex_lock(&ftrace_lock);
5531 
5532 	WARN_ON(probe->ref < count);
5533 
5534 	probe->ref -= count;
5535 
5536 	if (ftrace_hash_empty(hash))
5537 		ftrace_shutdown(&probe->ops, 0);
5538 
5539 	ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
5540 					      hash, 1);
5541 
5542 	/* still need to update the function call sites */
5543 	if (ftrace_enabled && !ftrace_hash_empty(hash))
5544 		ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
5545 				       &old_hash_ops);
5546 	synchronize_rcu();
5547 
5548 	hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
5549 		hlist_del(&entry->hlist);
5550 		if (probe_ops->free)
5551 			probe_ops->free(probe_ops, tr, entry->ip, probe->data);
5552 		kfree(entry);
5553 	}
5554 	mutex_unlock(&ftrace_lock);
5555 
5556  out_unlock:
5557 	mutex_unlock(&probe->ops.func_hash->regex_lock);
5558 	free_ftrace_hash(hash);
5559 
5560 	release_probe(probe);
5561 
5562 	return ret;
5563 
5564  err_unlock_ftrace:
5565 	mutex_unlock(&ftrace_lock);
5566 	return ret;
5567 }
5568 
clear_ftrace_function_probes(struct trace_array * tr)5569 void clear_ftrace_function_probes(struct trace_array *tr)
5570 {
5571 	struct ftrace_func_probe *probe, *n;
5572 
5573 	list_for_each_entry_safe(probe, n, &tr->func_probes, list)
5574 		unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
5575 }
5576 
5577 static LIST_HEAD(ftrace_commands);
5578 static DEFINE_MUTEX(ftrace_cmd_mutex);
5579 
5580 /*
5581  * Currently we only register ftrace commands from __init, so mark this
5582  * __init too.
5583  */
register_ftrace_command(struct ftrace_func_command * cmd)5584 __init int register_ftrace_command(struct ftrace_func_command *cmd)
5585 {
5586 	struct ftrace_func_command *p;
5587 
5588 	guard(mutex)(&ftrace_cmd_mutex);
5589 	list_for_each_entry(p, &ftrace_commands, list) {
5590 		if (strcmp(cmd->name, p->name) == 0)
5591 			return -EBUSY;
5592 	}
5593 	list_add(&cmd->list, &ftrace_commands);
5594 
5595 	return 0;
5596 }
5597 
5598 /*
5599  * Currently we only unregister ftrace commands from __init, so mark
5600  * this __init too.
5601  */
unregister_ftrace_command(struct ftrace_func_command * cmd)5602 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
5603 {
5604 	struct ftrace_func_command *p, *n;
5605 
5606 	guard(mutex)(&ftrace_cmd_mutex);
5607 
5608 	list_for_each_entry_safe(p, n, &ftrace_commands, list) {
5609 		if (strcmp(cmd->name, p->name) == 0) {
5610 			list_del_init(&p->list);
5611 			return 0;
5612 		}
5613 	}
5614 
5615 	return -ENODEV;
5616 }
5617 
ftrace_process_regex(struct ftrace_iterator * iter,char * buff,int len,int enable)5618 static int ftrace_process_regex(struct ftrace_iterator *iter,
5619 				char *buff, int len, int enable)
5620 {
5621 	struct ftrace_hash *hash = iter->hash;
5622 	struct trace_array *tr = iter->ops->private;
5623 	char *func, *command, *next = buff;
5624 	struct ftrace_func_command *p;
5625 	int ret;
5626 
5627 	func = strsep(&next, ":");
5628 
5629 	if (!next) {
5630 		ret = ftrace_match_records(hash, func, len);
5631 		if (!ret)
5632 			ret = -EINVAL;
5633 		if (ret < 0)
5634 			return ret;
5635 		return 0;
5636 	}
5637 
5638 	/* command found */
5639 
5640 	command = strsep(&next, ":");
5641 
5642 	guard(mutex)(&ftrace_cmd_mutex);
5643 
5644 	list_for_each_entry(p, &ftrace_commands, list) {
5645 		if (strcmp(p->name, command) == 0)
5646 			return p->func(tr, hash, func, command, next, enable);
5647 	}
5648 
5649 	return -EINVAL;
5650 }
5651 
5652 static ssize_t
ftrace_regex_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos,int enable)5653 ftrace_regex_write(struct file *file, const char __user *ubuf,
5654 		   size_t cnt, loff_t *ppos, int enable)
5655 {
5656 	struct ftrace_iterator *iter;
5657 	struct trace_parser *parser;
5658 	ssize_t ret, read;
5659 
5660 	if (!cnt)
5661 		return 0;
5662 
5663 	if (file->f_mode & FMODE_READ) {
5664 		struct seq_file *m = file->private_data;
5665 		iter = m->private;
5666 	} else
5667 		iter = file->private_data;
5668 
5669 	if (unlikely(ftrace_disabled))
5670 		return -ENODEV;
5671 
5672 	/* iter->hash is a local copy, so we don't need regex_lock */
5673 
5674 	parser = &iter->parser;
5675 	read = trace_get_user(parser, ubuf, cnt, ppos);
5676 
5677 	if (read >= 0 && trace_parser_loaded(parser) &&
5678 	    !trace_parser_cont(parser)) {
5679 		ret = ftrace_process_regex(iter, parser->buffer,
5680 					   parser->idx, enable);
5681 		trace_parser_clear(parser);
5682 		if (ret < 0)
5683 			return ret;
5684 	}
5685 
5686 	return read;
5687 }
5688 
5689 ssize_t
ftrace_filter_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5690 ftrace_filter_write(struct file *file, const char __user *ubuf,
5691 		    size_t cnt, loff_t *ppos)
5692 {
5693 	return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
5694 }
5695 
5696 ssize_t
ftrace_notrace_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5697 ftrace_notrace_write(struct file *file, const char __user *ubuf,
5698 		     size_t cnt, loff_t *ppos)
5699 {
5700 	return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
5701 }
5702 
5703 static int
__ftrace_match_addr(struct ftrace_hash * hash,unsigned long ip,int remove)5704 __ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
5705 {
5706 	struct ftrace_func_entry *entry;
5707 
5708 	ip = ftrace_location(ip);
5709 	if (!ip)
5710 		return -EINVAL;
5711 
5712 	if (remove) {
5713 		entry = ftrace_lookup_ip(hash, ip);
5714 		if (!entry)
5715 			return -ENOENT;
5716 		free_hash_entry(hash, entry);
5717 		return 0;
5718 	} else if (__ftrace_lookup_ip(hash, ip) != NULL) {
5719 		/* Already exists */
5720 		return 0;
5721 	}
5722 
5723 	entry = add_hash_entry(hash, ip);
5724 	return entry ? 0 :  -ENOMEM;
5725 }
5726 
5727 static int
ftrace_match_addr(struct ftrace_hash * hash,unsigned long * ips,unsigned int cnt,int remove)5728 ftrace_match_addr(struct ftrace_hash *hash, unsigned long *ips,
5729 		  unsigned int cnt, int remove)
5730 {
5731 	unsigned int i;
5732 	int err;
5733 
5734 	for (i = 0; i < cnt; i++) {
5735 		err = __ftrace_match_addr(hash, ips[i], remove);
5736 		if (err) {
5737 			/*
5738 			 * This expects the @hash is a temporary hash and if this
5739 			 * fails the caller must free the @hash.
5740 			 */
5741 			return err;
5742 		}
5743 	}
5744 	return 0;
5745 }
5746 
5747 static int
ftrace_set_hash(struct ftrace_ops * ops,unsigned char * buf,int len,unsigned long * ips,unsigned int cnt,int remove,int reset,int enable,char * mod)5748 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
5749 		unsigned long *ips, unsigned int cnt,
5750 		int remove, int reset, int enable, char *mod)
5751 {
5752 	struct ftrace_hash **orig_hash;
5753 	struct ftrace_hash *hash;
5754 	int ret;
5755 
5756 	if (unlikely(ftrace_disabled))
5757 		return -ENODEV;
5758 
5759 	mutex_lock(&ops->func_hash->regex_lock);
5760 
5761 	if (enable)
5762 		orig_hash = &ops->func_hash->filter_hash;
5763 	else
5764 		orig_hash = &ops->func_hash->notrace_hash;
5765 
5766 	if (reset)
5767 		hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5768 	else
5769 		hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
5770 
5771 	if (!hash) {
5772 		ret = -ENOMEM;
5773 		goto out_regex_unlock;
5774 	}
5775 
5776 	if (buf && !match_records(hash, buf, len, mod)) {
5777 		/* If this was for a module and nothing was enabled, flag it */
5778 		if (mod)
5779 			(*orig_hash)->flags |= FTRACE_HASH_FL_MOD;
5780 
5781 		/*
5782 		 * Even if it is a mod, return error to let caller know
5783 		 * nothing was added
5784 		 */
5785 		ret = -EINVAL;
5786 		goto out_regex_unlock;
5787 	}
5788 	if (ips) {
5789 		ret = ftrace_match_addr(hash, ips, cnt, remove);
5790 		if (ret < 0)
5791 			goto out_regex_unlock;
5792 	}
5793 
5794 	mutex_lock(&ftrace_lock);
5795 	ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
5796 	mutex_unlock(&ftrace_lock);
5797 
5798  out_regex_unlock:
5799 	mutex_unlock(&ops->func_hash->regex_lock);
5800 
5801 	free_ftrace_hash(hash);
5802 	return ret;
5803 }
5804 
5805 static int
ftrace_set_addr(struct ftrace_ops * ops,unsigned long * ips,unsigned int cnt,int remove,int reset,int enable)5806 ftrace_set_addr(struct ftrace_ops *ops, unsigned long *ips, unsigned int cnt,
5807 		int remove, int reset, int enable)
5808 {
5809 	return ftrace_set_hash(ops, NULL, 0, ips, cnt, remove, reset, enable, NULL);
5810 }
5811 
5812 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
5813 
5814 static int register_ftrace_function_nolock(struct ftrace_ops *ops);
5815 
5816 /*
5817  * If there are multiple ftrace_ops, use SAVE_REGS by default, so that direct
5818  * call will be jumped from ftrace_regs_caller. Only if the architecture does
5819  * not support ftrace_regs_caller but direct_call, use SAVE_ARGS so that it
5820  * jumps from ftrace_caller for multiple ftrace_ops.
5821  */
5822 #ifndef CONFIG_HAVE_DYNAMIC_FTRACE_WITH_REGS
5823 #define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_ARGS)
5824 #else
5825 #define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_REGS)
5826 #endif
5827 
check_direct_multi(struct ftrace_ops * ops)5828 static int check_direct_multi(struct ftrace_ops *ops)
5829 {
5830 	if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED))
5831 		return -EINVAL;
5832 	if ((ops->flags & MULTI_FLAGS) != MULTI_FLAGS)
5833 		return -EINVAL;
5834 	return 0;
5835 }
5836 
remove_direct_functions_hash(struct ftrace_hash * hash,unsigned long addr)5837 static void remove_direct_functions_hash(struct ftrace_hash *hash, unsigned long addr)
5838 {
5839 	struct ftrace_func_entry *entry, *del;
5840 	int size, i;
5841 
5842 	size = 1 << hash->size_bits;
5843 	for (i = 0; i < size; i++) {
5844 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5845 			del = __ftrace_lookup_ip(direct_functions, entry->ip);
5846 			if (del && del->direct == addr) {
5847 				remove_hash_entry(direct_functions, del);
5848 				kfree(del);
5849 			}
5850 		}
5851 	}
5852 }
5853 
register_ftrace_direct_cb(struct rcu_head * rhp)5854 static void register_ftrace_direct_cb(struct rcu_head *rhp)
5855 {
5856 	struct ftrace_hash *fhp = container_of(rhp, struct ftrace_hash, rcu);
5857 
5858 	free_ftrace_hash(fhp);
5859 }
5860 
5861 /**
5862  * register_ftrace_direct - Call a custom trampoline directly
5863  * for multiple functions registered in @ops
5864  * @ops: The address of the struct ftrace_ops object
5865  * @addr: The address of the trampoline to call at @ops functions
5866  *
5867  * This is used to connect a direct calls to @addr from the nop locations
5868  * of the functions registered in @ops (with by ftrace_set_filter_ip
5869  * function).
5870  *
5871  * The location that it calls (@addr) must be able to handle a direct call,
5872  * and save the parameters of the function being traced, and restore them
5873  * (or inject new ones if needed), before returning.
5874  *
5875  * Returns:
5876  *  0 on success
5877  *  -EINVAL  - The @ops object was already registered with this call or
5878  *             when there are no functions in @ops object.
5879  *  -EBUSY   - Another direct function is already attached (there can be only one)
5880  *  -ENODEV  - @ip does not point to a ftrace nop location (or not supported)
5881  *  -ENOMEM  - There was an allocation failure.
5882  */
register_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)5883 int register_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
5884 {
5885 	struct ftrace_hash *hash, *new_hash = NULL, *free_hash = NULL;
5886 	struct ftrace_func_entry *entry, *new;
5887 	int err = -EBUSY, size, i;
5888 
5889 	if (ops->func || ops->trampoline)
5890 		return -EINVAL;
5891 	if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED))
5892 		return -EINVAL;
5893 	if (ops->flags & FTRACE_OPS_FL_ENABLED)
5894 		return -EINVAL;
5895 
5896 	hash = ops->func_hash->filter_hash;
5897 	if (ftrace_hash_empty(hash))
5898 		return -EINVAL;
5899 
5900 	mutex_lock(&direct_mutex);
5901 
5902 	/* Make sure requested entries are not already registered.. */
5903 	size = 1 << hash->size_bits;
5904 	for (i = 0; i < size; i++) {
5905 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5906 			if (ftrace_find_rec_direct(entry->ip))
5907 				goto out_unlock;
5908 		}
5909 	}
5910 
5911 	err = -ENOMEM;
5912 
5913 	/* Make a copy hash to place the new and the old entries in */
5914 	size = hash->count + direct_functions->count;
5915 	if (size > 32)
5916 		size = 32;
5917 	new_hash = alloc_ftrace_hash(fls(size));
5918 	if (!new_hash)
5919 		goto out_unlock;
5920 
5921 	/* Now copy over the existing direct entries */
5922 	size = 1 << direct_functions->size_bits;
5923 	for (i = 0; i < size; i++) {
5924 		hlist_for_each_entry(entry, &direct_functions->buckets[i], hlist) {
5925 			new = add_hash_entry(new_hash, entry->ip);
5926 			if (!new)
5927 				goto out_unlock;
5928 			new->direct = entry->direct;
5929 		}
5930 	}
5931 
5932 	/* ... and add the new entries */
5933 	size = 1 << hash->size_bits;
5934 	for (i = 0; i < size; i++) {
5935 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5936 			new = add_hash_entry(new_hash, entry->ip);
5937 			if (!new)
5938 				goto out_unlock;
5939 			/* Update both the copy and the hash entry */
5940 			new->direct = addr;
5941 			entry->direct = addr;
5942 		}
5943 	}
5944 
5945 	free_hash = direct_functions;
5946 	rcu_assign_pointer(direct_functions, new_hash);
5947 	new_hash = NULL;
5948 
5949 	ops->func = call_direct_funcs;
5950 	ops->flags = MULTI_FLAGS;
5951 	ops->trampoline = FTRACE_REGS_ADDR;
5952 	ops->direct_call = addr;
5953 
5954 	err = register_ftrace_function_nolock(ops);
5955 
5956  out_unlock:
5957 	mutex_unlock(&direct_mutex);
5958 
5959 	if (free_hash && free_hash != EMPTY_HASH)
5960 		call_rcu_tasks(&free_hash->rcu, register_ftrace_direct_cb);
5961 
5962 	if (new_hash)
5963 		free_ftrace_hash(new_hash);
5964 
5965 	return err;
5966 }
5967 EXPORT_SYMBOL_GPL(register_ftrace_direct);
5968 
5969 /**
5970  * unregister_ftrace_direct - Remove calls to custom trampoline
5971  * previously registered by register_ftrace_direct for @ops object.
5972  * @ops: The address of the struct ftrace_ops object
5973  * @addr: The address of the direct function that is called by the @ops functions
5974  * @free_filters: Set to true to remove all filters for the ftrace_ops, false otherwise
5975  *
5976  * This is used to remove a direct calls to @addr from the nop locations
5977  * of the functions registered in @ops (with by ftrace_set_filter_ip
5978  * function).
5979  *
5980  * Returns:
5981  *  0 on success
5982  *  -EINVAL - The @ops object was not properly registered.
5983  */
unregister_ftrace_direct(struct ftrace_ops * ops,unsigned long addr,bool free_filters)5984 int unregister_ftrace_direct(struct ftrace_ops *ops, unsigned long addr,
5985 			     bool free_filters)
5986 {
5987 	struct ftrace_hash *hash = ops->func_hash->filter_hash;
5988 	int err;
5989 
5990 	if (check_direct_multi(ops))
5991 		return -EINVAL;
5992 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
5993 		return -EINVAL;
5994 
5995 	mutex_lock(&direct_mutex);
5996 	err = unregister_ftrace_function(ops);
5997 	remove_direct_functions_hash(hash, addr);
5998 	mutex_unlock(&direct_mutex);
5999 
6000 	/* cleanup for possible another register call */
6001 	ops->func = NULL;
6002 	ops->trampoline = 0;
6003 
6004 	if (free_filters)
6005 		ftrace_free_filter(ops);
6006 	return err;
6007 }
6008 EXPORT_SYMBOL_GPL(unregister_ftrace_direct);
6009 
6010 static int
__modify_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)6011 __modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
6012 {
6013 	struct ftrace_hash *hash;
6014 	struct ftrace_func_entry *entry, *iter;
6015 	static struct ftrace_ops tmp_ops = {
6016 		.func		= ftrace_stub,
6017 		.flags		= FTRACE_OPS_FL_STUB,
6018 	};
6019 	int i, size;
6020 	int err;
6021 
6022 	lockdep_assert_held_once(&direct_mutex);
6023 
6024 	/* Enable the tmp_ops to have the same functions as the direct ops */
6025 	ftrace_ops_init(&tmp_ops);
6026 	tmp_ops.func_hash = ops->func_hash;
6027 	tmp_ops.direct_call = addr;
6028 
6029 	err = register_ftrace_function_nolock(&tmp_ops);
6030 	if (err)
6031 		return err;
6032 
6033 	/*
6034 	 * Now the ftrace_ops_list_func() is called to do the direct callers.
6035 	 * We can safely change the direct functions attached to each entry.
6036 	 */
6037 	mutex_lock(&ftrace_lock);
6038 
6039 	hash = ops->func_hash->filter_hash;
6040 	size = 1 << hash->size_bits;
6041 	for (i = 0; i < size; i++) {
6042 		hlist_for_each_entry(iter, &hash->buckets[i], hlist) {
6043 			entry = __ftrace_lookup_ip(direct_functions, iter->ip);
6044 			if (!entry)
6045 				continue;
6046 			entry->direct = addr;
6047 		}
6048 	}
6049 	/* Prevent store tearing if a trampoline concurrently accesses the value */
6050 	WRITE_ONCE(ops->direct_call, addr);
6051 
6052 	mutex_unlock(&ftrace_lock);
6053 
6054 	/* Removing the tmp_ops will add the updated direct callers to the functions */
6055 	unregister_ftrace_function(&tmp_ops);
6056 
6057 	return err;
6058 }
6059 
6060 /**
6061  * modify_ftrace_direct_nolock - Modify an existing direct 'multi' call
6062  * to call something else
6063  * @ops: The address of the struct ftrace_ops object
6064  * @addr: The address of the new trampoline to call at @ops functions
6065  *
6066  * This is used to unregister currently registered direct caller and
6067  * register new one @addr on functions registered in @ops object.
6068  *
6069  * Note there's window between ftrace_shutdown and ftrace_startup calls
6070  * where there will be no callbacks called.
6071  *
6072  * Caller should already have direct_mutex locked, so we don't lock
6073  * direct_mutex here.
6074  *
6075  * Returns: zero on success. Non zero on error, which includes:
6076  *  -EINVAL - The @ops object was not properly registered.
6077  */
modify_ftrace_direct_nolock(struct ftrace_ops * ops,unsigned long addr)6078 int modify_ftrace_direct_nolock(struct ftrace_ops *ops, unsigned long addr)
6079 {
6080 	if (check_direct_multi(ops))
6081 		return -EINVAL;
6082 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6083 		return -EINVAL;
6084 
6085 	return __modify_ftrace_direct(ops, addr);
6086 }
6087 EXPORT_SYMBOL_GPL(modify_ftrace_direct_nolock);
6088 
6089 /**
6090  * modify_ftrace_direct - Modify an existing direct 'multi' call
6091  * to call something else
6092  * @ops: The address of the struct ftrace_ops object
6093  * @addr: The address of the new trampoline to call at @ops functions
6094  *
6095  * This is used to unregister currently registered direct caller and
6096  * register new one @addr on functions registered in @ops object.
6097  *
6098  * Note there's window between ftrace_shutdown and ftrace_startup calls
6099  * where there will be no callbacks called.
6100  *
6101  * Returns: zero on success. Non zero on error, which includes:
6102  *  -EINVAL - The @ops object was not properly registered.
6103  */
modify_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)6104 int modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
6105 {
6106 	int err;
6107 
6108 	if (check_direct_multi(ops))
6109 		return -EINVAL;
6110 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6111 		return -EINVAL;
6112 
6113 	mutex_lock(&direct_mutex);
6114 	err = __modify_ftrace_direct(ops, addr);
6115 	mutex_unlock(&direct_mutex);
6116 	return err;
6117 }
6118 EXPORT_SYMBOL_GPL(modify_ftrace_direct);
6119 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
6120 
6121 /**
6122  * ftrace_set_filter_ip - set a function to filter on in ftrace by address
6123  * @ops: the ops to set the filter with
6124  * @ip: the address to add to or remove from the filter.
6125  * @remove: non zero to remove the ip from the filter
6126  * @reset: non zero to reset all filters before applying this filter.
6127  *
6128  * Filters denote which functions should be enabled when tracing is enabled
6129  * If @ip is NULL, it fails to update filter.
6130  *
6131  * This can allocate memory which must be freed before @ops can be freed,
6132  * either by removing each filtered addr or by using
6133  * ftrace_free_filter(@ops).
6134  */
ftrace_set_filter_ip(struct ftrace_ops * ops,unsigned long ip,int remove,int reset)6135 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
6136 			 int remove, int reset)
6137 {
6138 	ftrace_ops_init(ops);
6139 	return ftrace_set_addr(ops, &ip, 1, remove, reset, 1);
6140 }
6141 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
6142 
6143 /**
6144  * ftrace_set_filter_ips - set functions to filter on in ftrace by addresses
6145  * @ops: the ops to set the filter with
6146  * @ips: the array of addresses to add to or remove from the filter.
6147  * @cnt: the number of addresses in @ips
6148  * @remove: non zero to remove ips from the filter
6149  * @reset: non zero to reset all filters before applying this filter.
6150  *
6151  * Filters denote which functions should be enabled when tracing is enabled
6152  * If @ips array or any ip specified within is NULL , it fails to update filter.
6153  *
6154  * This can allocate memory which must be freed before @ops can be freed,
6155  * either by removing each filtered addr or by using
6156  * ftrace_free_filter(@ops).
6157 */
ftrace_set_filter_ips(struct ftrace_ops * ops,unsigned long * ips,unsigned int cnt,int remove,int reset)6158 int ftrace_set_filter_ips(struct ftrace_ops *ops, unsigned long *ips,
6159 			  unsigned int cnt, int remove, int reset)
6160 {
6161 	ftrace_ops_init(ops);
6162 	return ftrace_set_addr(ops, ips, cnt, remove, reset, 1);
6163 }
6164 EXPORT_SYMBOL_GPL(ftrace_set_filter_ips);
6165 
6166 /**
6167  * ftrace_ops_set_global_filter - setup ops to use global filters
6168  * @ops: the ops which will use the global filters
6169  *
6170  * ftrace users who need global function trace filtering should call this.
6171  * It can set the global filter only if ops were not initialized before.
6172  */
ftrace_ops_set_global_filter(struct ftrace_ops * ops)6173 void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
6174 {
6175 	if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
6176 		return;
6177 
6178 	ftrace_ops_init(ops);
6179 	ops->func_hash = &global_ops.local_hash;
6180 }
6181 EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
6182 
6183 static int
ftrace_set_regex(struct ftrace_ops * ops,unsigned char * buf,int len,int reset,int enable)6184 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
6185 		 int reset, int enable)
6186 {
6187 	char *mod = NULL, *func, *command, *next = buf;
6188 	char *tmp __free(kfree) = NULL;
6189 	struct trace_array *tr = ops->private;
6190 	int ret;
6191 
6192 	func = strsep(&next, ":");
6193 
6194 	/* This can also handle :mod: parsing */
6195 	if (next) {
6196 		if (!tr)
6197 			return -EINVAL;
6198 
6199 		command = strsep(&next, ":");
6200 		if (strcmp(command, "mod") != 0)
6201 			return -EINVAL;
6202 
6203 		mod = next;
6204 		len = command - func;
6205 		/* Save the original func as ftrace_set_hash() can modify it */
6206 		tmp = kstrdup(func, GFP_KERNEL);
6207 	}
6208 
6209 	ret = ftrace_set_hash(ops, func, len, NULL, 0, 0, reset, enable, mod);
6210 
6211 	if (tr && mod && ret < 0) {
6212 		/* Did tmp fail to allocate? */
6213 		if (!tmp)
6214 			return -ENOMEM;
6215 		ret = cache_mod(tr, tmp, mod, enable);
6216 	}
6217 
6218 	return ret;
6219 }
6220 
6221 /**
6222  * ftrace_set_filter - set a function to filter on in ftrace
6223  * @ops: the ops to set the filter with
6224  * @buf: the string that holds the function filter text.
6225  * @len: the length of the string.
6226  * @reset: non-zero to reset all filters before applying this filter.
6227  *
6228  * Filters denote which functions should be enabled when tracing is enabled.
6229  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
6230  *
6231  * This can allocate memory which must be freed before @ops can be freed,
6232  * either by removing each filtered addr or by using
6233  * ftrace_free_filter(@ops).
6234  */
ftrace_set_filter(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)6235 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
6236 		       int len, int reset)
6237 {
6238 	ftrace_ops_init(ops);
6239 	return ftrace_set_regex(ops, buf, len, reset, 1);
6240 }
6241 EXPORT_SYMBOL_GPL(ftrace_set_filter);
6242 
6243 /**
6244  * ftrace_set_notrace - set a function to not trace in ftrace
6245  * @ops: the ops to set the notrace filter with
6246  * @buf: the string that holds the function notrace text.
6247  * @len: the length of the string.
6248  * @reset: non-zero to reset all filters before applying this filter.
6249  *
6250  * Notrace Filters denote which functions should not be enabled when tracing
6251  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
6252  * for tracing.
6253  *
6254  * This can allocate memory which must be freed before @ops can be freed,
6255  * either by removing each filtered addr or by using
6256  * ftrace_free_filter(@ops).
6257  */
ftrace_set_notrace(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)6258 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
6259 			int len, int reset)
6260 {
6261 	ftrace_ops_init(ops);
6262 	return ftrace_set_regex(ops, buf, len, reset, 0);
6263 }
6264 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
6265 /**
6266  * ftrace_set_global_filter - set a function to filter on with global tracers
6267  * @buf: the string that holds the function filter text.
6268  * @len: the length of the string.
6269  * @reset: non-zero to reset all filters before applying this filter.
6270  *
6271  * Filters denote which functions should be enabled when tracing is enabled.
6272  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
6273  */
ftrace_set_global_filter(unsigned char * buf,int len,int reset)6274 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
6275 {
6276 	ftrace_set_regex(&global_ops, buf, len, reset, 1);
6277 }
6278 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
6279 
6280 /**
6281  * ftrace_set_global_notrace - set a function to not trace with global tracers
6282  * @buf: the string that holds the function notrace text.
6283  * @len: the length of the string.
6284  * @reset: non-zero to reset all filters before applying this filter.
6285  *
6286  * Notrace Filters denote which functions should not be enabled when tracing
6287  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
6288  * for tracing.
6289  */
ftrace_set_global_notrace(unsigned char * buf,int len,int reset)6290 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
6291 {
6292 	ftrace_set_regex(&global_ops, buf, len, reset, 0);
6293 }
6294 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
6295 
6296 /*
6297  * command line interface to allow users to set filters on boot up.
6298  */
6299 #define FTRACE_FILTER_SIZE		COMMAND_LINE_SIZE
6300 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
6301 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
6302 
6303 /* Used by function selftest to not test if filter is set */
6304 bool ftrace_filter_param __initdata;
6305 
set_ftrace_notrace(char * str)6306 static int __init set_ftrace_notrace(char *str)
6307 {
6308 	ftrace_filter_param = true;
6309 	strscpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
6310 	return 1;
6311 }
6312 __setup("ftrace_notrace=", set_ftrace_notrace);
6313 
set_ftrace_filter(char * str)6314 static int __init set_ftrace_filter(char *str)
6315 {
6316 	ftrace_filter_param = true;
6317 	strscpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
6318 	return 1;
6319 }
6320 __setup("ftrace_filter=", set_ftrace_filter);
6321 
6322 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6323 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
6324 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
6325 static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
6326 
set_graph_function(char * str)6327 static int __init set_graph_function(char *str)
6328 {
6329 	strscpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
6330 	return 1;
6331 }
6332 __setup("ftrace_graph_filter=", set_graph_function);
6333 
set_graph_notrace_function(char * str)6334 static int __init set_graph_notrace_function(char *str)
6335 {
6336 	strscpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
6337 	return 1;
6338 }
6339 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
6340 
set_graph_max_depth_function(char * str)6341 static int __init set_graph_max_depth_function(char *str)
6342 {
6343 	if (!str || kstrtouint(str, 0, &fgraph_max_depth))
6344 		return 0;
6345 	return 1;
6346 }
6347 __setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
6348 
set_ftrace_early_graph(char * buf,int enable)6349 static void __init set_ftrace_early_graph(char *buf, int enable)
6350 {
6351 	int ret;
6352 	char *func;
6353 	struct ftrace_hash *hash;
6354 
6355 	hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
6356 	if (MEM_FAIL(!hash, "Failed to allocate hash\n"))
6357 		return;
6358 
6359 	while (buf) {
6360 		func = strsep(&buf, ",");
6361 		/* we allow only one expression at a time */
6362 		ret = ftrace_graph_set_hash(hash, func);
6363 		if (ret)
6364 			printk(KERN_DEBUG "ftrace: function %s not "
6365 					  "traceable\n", func);
6366 	}
6367 
6368 	if (enable)
6369 		ftrace_graph_hash = hash;
6370 	else
6371 		ftrace_graph_notrace_hash = hash;
6372 }
6373 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6374 
6375 void __init
ftrace_set_early_filter(struct ftrace_ops * ops,char * buf,int enable)6376 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
6377 {
6378 	char *func;
6379 
6380 	ftrace_ops_init(ops);
6381 
6382 	/* The trace_array is needed for caching module function filters */
6383 	if (!ops->private) {
6384 		struct trace_array *tr = trace_get_global_array();
6385 
6386 		ops->private = tr;
6387 		ftrace_init_trace_array(tr);
6388 	}
6389 
6390 	while (buf) {
6391 		func = strsep(&buf, ",");
6392 		ftrace_set_regex(ops, func, strlen(func), 0, enable);
6393 	}
6394 }
6395 
set_ftrace_early_filters(void)6396 static void __init set_ftrace_early_filters(void)
6397 {
6398 	if (ftrace_filter_buf[0])
6399 		ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
6400 	if (ftrace_notrace_buf[0])
6401 		ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
6402 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6403 	if (ftrace_graph_buf[0])
6404 		set_ftrace_early_graph(ftrace_graph_buf, 1);
6405 	if (ftrace_graph_notrace_buf[0])
6406 		set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
6407 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6408 }
6409 
ftrace_regex_release(struct inode * inode,struct file * file)6410 int ftrace_regex_release(struct inode *inode, struct file *file)
6411 {
6412 	struct seq_file *m = (struct seq_file *)file->private_data;
6413 	struct ftrace_iterator *iter;
6414 	struct ftrace_hash **orig_hash;
6415 	struct trace_parser *parser;
6416 	int filter_hash;
6417 
6418 	if (file->f_mode & FMODE_READ) {
6419 		iter = m->private;
6420 		seq_release(inode, file);
6421 	} else
6422 		iter = file->private_data;
6423 
6424 	parser = &iter->parser;
6425 	if (trace_parser_loaded(parser)) {
6426 		int enable = !(iter->flags & FTRACE_ITER_NOTRACE);
6427 
6428 		ftrace_process_regex(iter, parser->buffer,
6429 				     parser->idx, enable);
6430 	}
6431 
6432 	trace_parser_put(parser);
6433 
6434 	mutex_lock(&iter->ops->func_hash->regex_lock);
6435 
6436 	if (file->f_mode & FMODE_WRITE) {
6437 		filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
6438 
6439 		if (filter_hash) {
6440 			orig_hash = &iter->ops->func_hash->filter_hash;
6441 			if (iter->tr) {
6442 				if (list_empty(&iter->tr->mod_trace))
6443 					iter->hash->flags &= ~FTRACE_HASH_FL_MOD;
6444 				else
6445 					iter->hash->flags |= FTRACE_HASH_FL_MOD;
6446 			}
6447 		} else
6448 			orig_hash = &iter->ops->func_hash->notrace_hash;
6449 
6450 		mutex_lock(&ftrace_lock);
6451 		ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
6452 						      iter->hash, filter_hash);
6453 		mutex_unlock(&ftrace_lock);
6454 	} else {
6455 		/* For read only, the hash is the ops hash */
6456 		iter->hash = NULL;
6457 	}
6458 
6459 	mutex_unlock(&iter->ops->func_hash->regex_lock);
6460 	free_ftrace_hash(iter->hash);
6461 	if (iter->tr)
6462 		trace_array_put(iter->tr);
6463 	kfree(iter);
6464 
6465 	return 0;
6466 }
6467 
6468 static const struct file_operations ftrace_avail_fops = {
6469 	.open = ftrace_avail_open,
6470 	.read = seq_read,
6471 	.llseek = seq_lseek,
6472 	.release = seq_release_private,
6473 };
6474 
6475 static const struct file_operations ftrace_enabled_fops = {
6476 	.open = ftrace_enabled_open,
6477 	.read = seq_read,
6478 	.llseek = seq_lseek,
6479 	.release = seq_release_private,
6480 };
6481 
6482 static const struct file_operations ftrace_touched_fops = {
6483 	.open = ftrace_touched_open,
6484 	.read = seq_read,
6485 	.llseek = seq_lseek,
6486 	.release = seq_release_private,
6487 };
6488 
6489 static const struct file_operations ftrace_avail_addrs_fops = {
6490 	.open = ftrace_avail_addrs_open,
6491 	.read = seq_read,
6492 	.llseek = seq_lseek,
6493 	.release = seq_release_private,
6494 };
6495 
6496 static const struct file_operations ftrace_filter_fops = {
6497 	.open = ftrace_filter_open,
6498 	.read = seq_read,
6499 	.write = ftrace_filter_write,
6500 	.llseek = tracing_lseek,
6501 	.release = ftrace_regex_release,
6502 };
6503 
6504 static const struct file_operations ftrace_notrace_fops = {
6505 	.open = ftrace_notrace_open,
6506 	.read = seq_read,
6507 	.write = ftrace_notrace_write,
6508 	.llseek = tracing_lseek,
6509 	.release = ftrace_regex_release,
6510 };
6511 
6512 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6513 
6514 static DEFINE_MUTEX(graph_lock);
6515 
6516 struct ftrace_hash __rcu *ftrace_graph_hash = EMPTY_HASH;
6517 struct ftrace_hash __rcu *ftrace_graph_notrace_hash = EMPTY_HASH;
6518 
6519 enum graph_filter_type {
6520 	GRAPH_FILTER_NOTRACE	= 0,
6521 	GRAPH_FILTER_FUNCTION,
6522 };
6523 
6524 #define FTRACE_GRAPH_EMPTY	((void *)1)
6525 
6526 struct ftrace_graph_data {
6527 	struct ftrace_hash		*hash;
6528 	struct ftrace_func_entry	*entry;
6529 	int				idx;   /* for hash table iteration */
6530 	enum graph_filter_type		type;
6531 	struct ftrace_hash		*new_hash;
6532 	const struct seq_operations	*seq_ops;
6533 	struct trace_parser		parser;
6534 };
6535 
6536 static void *
__g_next(struct seq_file * m,loff_t * pos)6537 __g_next(struct seq_file *m, loff_t *pos)
6538 {
6539 	struct ftrace_graph_data *fgd = m->private;
6540 	struct ftrace_func_entry *entry = fgd->entry;
6541 	struct hlist_head *head;
6542 	int i, idx = fgd->idx;
6543 
6544 	if (*pos >= fgd->hash->count)
6545 		return NULL;
6546 
6547 	if (entry) {
6548 		hlist_for_each_entry_continue(entry, hlist) {
6549 			fgd->entry = entry;
6550 			return entry;
6551 		}
6552 
6553 		idx++;
6554 	}
6555 
6556 	for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
6557 		head = &fgd->hash->buckets[i];
6558 		hlist_for_each_entry(entry, head, hlist) {
6559 			fgd->entry = entry;
6560 			fgd->idx = i;
6561 			return entry;
6562 		}
6563 	}
6564 	return NULL;
6565 }
6566 
6567 static void *
g_next(struct seq_file * m,void * v,loff_t * pos)6568 g_next(struct seq_file *m, void *v, loff_t *pos)
6569 {
6570 	(*pos)++;
6571 	return __g_next(m, pos);
6572 }
6573 
g_start(struct seq_file * m,loff_t * pos)6574 static void *g_start(struct seq_file *m, loff_t *pos)
6575 {
6576 	struct ftrace_graph_data *fgd = m->private;
6577 
6578 	mutex_lock(&graph_lock);
6579 
6580 	if (fgd->type == GRAPH_FILTER_FUNCTION)
6581 		fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
6582 					lockdep_is_held(&graph_lock));
6583 	else
6584 		fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6585 					lockdep_is_held(&graph_lock));
6586 
6587 	/* Nothing, tell g_show to print all functions are enabled */
6588 	if (ftrace_hash_empty(fgd->hash) && !*pos)
6589 		return FTRACE_GRAPH_EMPTY;
6590 
6591 	fgd->idx = 0;
6592 	fgd->entry = NULL;
6593 	return __g_next(m, pos);
6594 }
6595 
g_stop(struct seq_file * m,void * p)6596 static void g_stop(struct seq_file *m, void *p)
6597 {
6598 	mutex_unlock(&graph_lock);
6599 }
6600 
g_show(struct seq_file * m,void * v)6601 static int g_show(struct seq_file *m, void *v)
6602 {
6603 	struct ftrace_func_entry *entry = v;
6604 
6605 	if (!entry)
6606 		return 0;
6607 
6608 	if (entry == FTRACE_GRAPH_EMPTY) {
6609 		struct ftrace_graph_data *fgd = m->private;
6610 
6611 		if (fgd->type == GRAPH_FILTER_FUNCTION)
6612 			seq_puts(m, "#### all functions enabled ####\n");
6613 		else
6614 			seq_puts(m, "#### no functions disabled ####\n");
6615 		return 0;
6616 	}
6617 
6618 	seq_printf(m, "%ps\n", (void *)entry->ip);
6619 
6620 	return 0;
6621 }
6622 
6623 static const struct seq_operations ftrace_graph_seq_ops = {
6624 	.start = g_start,
6625 	.next = g_next,
6626 	.stop = g_stop,
6627 	.show = g_show,
6628 };
6629 
6630 static int
__ftrace_graph_open(struct inode * inode,struct file * file,struct ftrace_graph_data * fgd)6631 __ftrace_graph_open(struct inode *inode, struct file *file,
6632 		    struct ftrace_graph_data *fgd)
6633 {
6634 	int ret;
6635 	struct ftrace_hash *new_hash = NULL;
6636 
6637 	ret = security_locked_down(LOCKDOWN_TRACEFS);
6638 	if (ret)
6639 		return ret;
6640 
6641 	if (file->f_mode & FMODE_WRITE) {
6642 		const int size_bits = FTRACE_HASH_DEFAULT_BITS;
6643 
6644 		if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
6645 			return -ENOMEM;
6646 
6647 		if (file->f_flags & O_TRUNC)
6648 			new_hash = alloc_ftrace_hash(size_bits);
6649 		else
6650 			new_hash = alloc_and_copy_ftrace_hash(size_bits,
6651 							      fgd->hash);
6652 		if (!new_hash) {
6653 			ret = -ENOMEM;
6654 			goto out;
6655 		}
6656 	}
6657 
6658 	if (file->f_mode & FMODE_READ) {
6659 		ret = seq_open(file, &ftrace_graph_seq_ops);
6660 		if (!ret) {
6661 			struct seq_file *m = file->private_data;
6662 			m->private = fgd;
6663 		} else {
6664 			/* Failed */
6665 			free_ftrace_hash(new_hash);
6666 			new_hash = NULL;
6667 		}
6668 	} else
6669 		file->private_data = fgd;
6670 
6671 out:
6672 	if (ret < 0 && file->f_mode & FMODE_WRITE)
6673 		trace_parser_put(&fgd->parser);
6674 
6675 	fgd->new_hash = new_hash;
6676 
6677 	/*
6678 	 * All uses of fgd->hash must be taken with the graph_lock
6679 	 * held. The graph_lock is going to be released, so force
6680 	 * fgd->hash to be reinitialized when it is taken again.
6681 	 */
6682 	fgd->hash = NULL;
6683 
6684 	return ret;
6685 }
6686 
6687 static int
ftrace_graph_open(struct inode * inode,struct file * file)6688 ftrace_graph_open(struct inode *inode, struct file *file)
6689 {
6690 	struct ftrace_graph_data *fgd;
6691 	int ret;
6692 
6693 	if (unlikely(ftrace_disabled))
6694 		return -ENODEV;
6695 
6696 	fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
6697 	if (fgd == NULL)
6698 		return -ENOMEM;
6699 
6700 	mutex_lock(&graph_lock);
6701 
6702 	fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
6703 					lockdep_is_held(&graph_lock));
6704 	fgd->type = GRAPH_FILTER_FUNCTION;
6705 	fgd->seq_ops = &ftrace_graph_seq_ops;
6706 
6707 	ret = __ftrace_graph_open(inode, file, fgd);
6708 	if (ret < 0)
6709 		kfree(fgd);
6710 
6711 	mutex_unlock(&graph_lock);
6712 	return ret;
6713 }
6714 
6715 static int
ftrace_graph_notrace_open(struct inode * inode,struct file * file)6716 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
6717 {
6718 	struct ftrace_graph_data *fgd;
6719 	int ret;
6720 
6721 	if (unlikely(ftrace_disabled))
6722 		return -ENODEV;
6723 
6724 	fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
6725 	if (fgd == NULL)
6726 		return -ENOMEM;
6727 
6728 	mutex_lock(&graph_lock);
6729 
6730 	fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6731 					lockdep_is_held(&graph_lock));
6732 	fgd->type = GRAPH_FILTER_NOTRACE;
6733 	fgd->seq_ops = &ftrace_graph_seq_ops;
6734 
6735 	ret = __ftrace_graph_open(inode, file, fgd);
6736 	if (ret < 0)
6737 		kfree(fgd);
6738 
6739 	mutex_unlock(&graph_lock);
6740 	return ret;
6741 }
6742 
6743 static int
ftrace_graph_release(struct inode * inode,struct file * file)6744 ftrace_graph_release(struct inode *inode, struct file *file)
6745 {
6746 	struct ftrace_graph_data *fgd;
6747 	struct ftrace_hash *old_hash, *new_hash;
6748 	struct trace_parser *parser;
6749 	int ret = 0;
6750 
6751 	if (file->f_mode & FMODE_READ) {
6752 		struct seq_file *m = file->private_data;
6753 
6754 		fgd = m->private;
6755 		seq_release(inode, file);
6756 	} else {
6757 		fgd = file->private_data;
6758 	}
6759 
6760 
6761 	if (file->f_mode & FMODE_WRITE) {
6762 
6763 		parser = &fgd->parser;
6764 
6765 		if (trace_parser_loaded((parser))) {
6766 			ret = ftrace_graph_set_hash(fgd->new_hash,
6767 						    parser->buffer);
6768 		}
6769 
6770 		trace_parser_put(parser);
6771 
6772 		new_hash = __ftrace_hash_move(fgd->new_hash);
6773 		if (!new_hash) {
6774 			ret = -ENOMEM;
6775 			goto out;
6776 		}
6777 
6778 		mutex_lock(&graph_lock);
6779 
6780 		if (fgd->type == GRAPH_FILTER_FUNCTION) {
6781 			old_hash = rcu_dereference_protected(ftrace_graph_hash,
6782 					lockdep_is_held(&graph_lock));
6783 			rcu_assign_pointer(ftrace_graph_hash, new_hash);
6784 		} else {
6785 			old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6786 					lockdep_is_held(&graph_lock));
6787 			rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
6788 		}
6789 
6790 		mutex_unlock(&graph_lock);
6791 
6792 		/*
6793 		 * We need to do a hard force of sched synchronization.
6794 		 * This is because we use preempt_disable() to do RCU, but
6795 		 * the function tracers can be called where RCU is not watching
6796 		 * (like before user_exit()). We can not rely on the RCU
6797 		 * infrastructure to do the synchronization, thus we must do it
6798 		 * ourselves.
6799 		 */
6800 		if (old_hash != EMPTY_HASH)
6801 			synchronize_rcu_tasks_rude();
6802 
6803 		free_ftrace_hash(old_hash);
6804 	}
6805 
6806  out:
6807 	free_ftrace_hash(fgd->new_hash);
6808 	kfree(fgd);
6809 
6810 	return ret;
6811 }
6812 
6813 static int
ftrace_graph_set_hash(struct ftrace_hash * hash,char * buffer)6814 ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
6815 {
6816 	struct ftrace_glob func_g;
6817 	struct dyn_ftrace *rec;
6818 	struct ftrace_page *pg;
6819 	struct ftrace_func_entry *entry;
6820 	int fail = 1;
6821 	int not;
6822 
6823 	/* decode regex */
6824 	func_g.type = filter_parse_regex(buffer, strlen(buffer),
6825 					 &func_g.search, &not);
6826 
6827 	func_g.len = strlen(func_g.search);
6828 
6829 	guard(mutex)(&ftrace_lock);
6830 
6831 	if (unlikely(ftrace_disabled))
6832 		return -ENODEV;
6833 
6834 	do_for_each_ftrace_rec(pg, rec) {
6835 
6836 		if (rec->flags & FTRACE_FL_DISABLED)
6837 			continue;
6838 
6839 		if (ftrace_match_record(rec, &func_g, NULL, 0)) {
6840 			entry = ftrace_lookup_ip(hash, rec->ip);
6841 
6842 			if (!not) {
6843 				fail = 0;
6844 
6845 				if (entry)
6846 					continue;
6847 				if (add_hash_entry(hash, rec->ip) == NULL)
6848 					return 0;
6849 			} else {
6850 				if (entry) {
6851 					free_hash_entry(hash, entry);
6852 					fail = 0;
6853 				}
6854 			}
6855 		}
6856 	} while_for_each_ftrace_rec();
6857 
6858 	return fail ? -EINVAL : 0;
6859 }
6860 
6861 static ssize_t
ftrace_graph_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)6862 ftrace_graph_write(struct file *file, const char __user *ubuf,
6863 		   size_t cnt, loff_t *ppos)
6864 {
6865 	ssize_t read, ret = 0;
6866 	struct ftrace_graph_data *fgd = file->private_data;
6867 	struct trace_parser *parser;
6868 
6869 	if (!cnt)
6870 		return 0;
6871 
6872 	/* Read mode uses seq functions */
6873 	if (file->f_mode & FMODE_READ) {
6874 		struct seq_file *m = file->private_data;
6875 		fgd = m->private;
6876 	}
6877 
6878 	parser = &fgd->parser;
6879 
6880 	read = trace_get_user(parser, ubuf, cnt, ppos);
6881 
6882 	if (read >= 0 && trace_parser_loaded(parser) &&
6883 	    !trace_parser_cont(parser)) {
6884 
6885 		ret = ftrace_graph_set_hash(fgd->new_hash,
6886 					    parser->buffer);
6887 		trace_parser_clear(parser);
6888 	}
6889 
6890 	if (!ret)
6891 		ret = read;
6892 
6893 	return ret;
6894 }
6895 
6896 static const struct file_operations ftrace_graph_fops = {
6897 	.open		= ftrace_graph_open,
6898 	.read		= seq_read,
6899 	.write		= ftrace_graph_write,
6900 	.llseek		= tracing_lseek,
6901 	.release	= ftrace_graph_release,
6902 };
6903 
6904 static const struct file_operations ftrace_graph_notrace_fops = {
6905 	.open		= ftrace_graph_notrace_open,
6906 	.read		= seq_read,
6907 	.write		= ftrace_graph_write,
6908 	.llseek		= tracing_lseek,
6909 	.release	= ftrace_graph_release,
6910 };
6911 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6912 
ftrace_create_filter_files(struct ftrace_ops * ops,struct dentry * parent)6913 void ftrace_create_filter_files(struct ftrace_ops *ops,
6914 				struct dentry *parent)
6915 {
6916 
6917 	trace_create_file("set_ftrace_filter", TRACE_MODE_WRITE, parent,
6918 			  ops, &ftrace_filter_fops);
6919 
6920 	trace_create_file("set_ftrace_notrace", TRACE_MODE_WRITE, parent,
6921 			  ops, &ftrace_notrace_fops);
6922 }
6923 
6924 /*
6925  * The name "destroy_filter_files" is really a misnomer. Although
6926  * in the future, it may actually delete the files, but this is
6927  * really intended to make sure the ops passed in are disabled
6928  * and that when this function returns, the caller is free to
6929  * free the ops.
6930  *
6931  * The "destroy" name is only to match the "create" name that this
6932  * should be paired with.
6933  */
ftrace_destroy_filter_files(struct ftrace_ops * ops)6934 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
6935 {
6936 	mutex_lock(&ftrace_lock);
6937 	if (ops->flags & FTRACE_OPS_FL_ENABLED)
6938 		ftrace_shutdown(ops, 0);
6939 	ops->flags |= FTRACE_OPS_FL_DELETED;
6940 	ftrace_free_filter(ops);
6941 	mutex_unlock(&ftrace_lock);
6942 }
6943 
ftrace_init_dyn_tracefs(struct dentry * d_tracer)6944 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
6945 {
6946 
6947 	trace_create_file("available_filter_functions", TRACE_MODE_READ,
6948 			d_tracer, NULL, &ftrace_avail_fops);
6949 
6950 	trace_create_file("available_filter_functions_addrs", TRACE_MODE_READ,
6951 			d_tracer, NULL, &ftrace_avail_addrs_fops);
6952 
6953 	trace_create_file("enabled_functions", TRACE_MODE_READ,
6954 			d_tracer, NULL, &ftrace_enabled_fops);
6955 
6956 	trace_create_file("touched_functions", TRACE_MODE_READ,
6957 			d_tracer, NULL, &ftrace_touched_fops);
6958 
6959 	ftrace_create_filter_files(&global_ops, d_tracer);
6960 
6961 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6962 	trace_create_file("set_graph_function", TRACE_MODE_WRITE, d_tracer,
6963 				    NULL,
6964 				    &ftrace_graph_fops);
6965 	trace_create_file("set_graph_notrace", TRACE_MODE_WRITE, d_tracer,
6966 				    NULL,
6967 				    &ftrace_graph_notrace_fops);
6968 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6969 
6970 	return 0;
6971 }
6972 
ftrace_cmp_ips(const void * a,const void * b)6973 static int ftrace_cmp_ips(const void *a, const void *b)
6974 {
6975 	const unsigned long *ipa = a;
6976 	const unsigned long *ipb = b;
6977 
6978 	if (*ipa > *ipb)
6979 		return 1;
6980 	if (*ipa < *ipb)
6981 		return -1;
6982 	return 0;
6983 }
6984 
6985 #ifdef CONFIG_FTRACE_SORT_STARTUP_TEST
test_is_sorted(unsigned long * start,unsigned long count)6986 static void test_is_sorted(unsigned long *start, unsigned long count)
6987 {
6988 	int i;
6989 
6990 	for (i = 1; i < count; i++) {
6991 		if (WARN(start[i - 1] > start[i],
6992 			 "[%d] %pS at %lx is not sorted with %pS at %lx\n", i,
6993 			 (void *)start[i - 1], start[i - 1],
6994 			 (void *)start[i], start[i]))
6995 			break;
6996 	}
6997 	if (i == count)
6998 		pr_info("ftrace section at %px sorted properly\n", start);
6999 }
7000 #else
test_is_sorted(unsigned long * start,unsigned long count)7001 static void test_is_sorted(unsigned long *start, unsigned long count)
7002 {
7003 }
7004 #endif
7005 
ftrace_process_locs(struct module * mod,unsigned long * start,unsigned long * end)7006 static int ftrace_process_locs(struct module *mod,
7007 			       unsigned long *start,
7008 			       unsigned long *end)
7009 {
7010 	struct ftrace_page *pg_unuse = NULL;
7011 	struct ftrace_page *start_pg;
7012 	struct ftrace_page *pg;
7013 	struct dyn_ftrace *rec;
7014 	unsigned long skipped = 0;
7015 	unsigned long count;
7016 	unsigned long *p;
7017 	unsigned long addr;
7018 	unsigned long flags = 0; /* Shut up gcc */
7019 	int ret = -ENOMEM;
7020 
7021 	count = end - start;
7022 
7023 	if (!count)
7024 		return 0;
7025 
7026 	/*
7027 	 * Sorting mcount in vmlinux at build time depend on
7028 	 * CONFIG_BUILDTIME_MCOUNT_SORT, while mcount loc in
7029 	 * modules can not be sorted at build time.
7030 	 */
7031 	if (!IS_ENABLED(CONFIG_BUILDTIME_MCOUNT_SORT) || mod) {
7032 		sort(start, count, sizeof(*start),
7033 		     ftrace_cmp_ips, NULL);
7034 	} else {
7035 		test_is_sorted(start, count);
7036 	}
7037 
7038 	start_pg = ftrace_allocate_pages(count);
7039 	if (!start_pg)
7040 		return -ENOMEM;
7041 
7042 	mutex_lock(&ftrace_lock);
7043 
7044 	/*
7045 	 * Core and each module needs their own pages, as
7046 	 * modules will free them when they are removed.
7047 	 * Force a new page to be allocated for modules.
7048 	 */
7049 	if (!mod) {
7050 		WARN_ON(ftrace_pages || ftrace_pages_start);
7051 		/* First initialization */
7052 		ftrace_pages = ftrace_pages_start = start_pg;
7053 	} else {
7054 		if (!ftrace_pages)
7055 			goto out;
7056 
7057 		if (WARN_ON(ftrace_pages->next)) {
7058 			/* Hmm, we have free pages? */
7059 			while (ftrace_pages->next)
7060 				ftrace_pages = ftrace_pages->next;
7061 		}
7062 
7063 		ftrace_pages->next = start_pg;
7064 	}
7065 
7066 	p = start;
7067 	pg = start_pg;
7068 	while (p < end) {
7069 		unsigned long end_offset;
7070 		addr = ftrace_call_adjust(*p++);
7071 		/*
7072 		 * Some architecture linkers will pad between
7073 		 * the different mcount_loc sections of different
7074 		 * object files to satisfy alignments.
7075 		 * Skip any NULL pointers.
7076 		 */
7077 		if (!addr) {
7078 			skipped++;
7079 			continue;
7080 		}
7081 
7082 		end_offset = (pg->index+1) * sizeof(pg->records[0]);
7083 		if (end_offset > PAGE_SIZE << pg->order) {
7084 			/* We should have allocated enough */
7085 			if (WARN_ON(!pg->next))
7086 				break;
7087 			pg = pg->next;
7088 		}
7089 
7090 		rec = &pg->records[pg->index++];
7091 		rec->ip = addr;
7092 	}
7093 
7094 	if (pg->next) {
7095 		pg_unuse = pg->next;
7096 		pg->next = NULL;
7097 	}
7098 
7099 	/* Assign the last page to ftrace_pages */
7100 	ftrace_pages = pg;
7101 
7102 	/*
7103 	 * We only need to disable interrupts on start up
7104 	 * because we are modifying code that an interrupt
7105 	 * may execute, and the modification is not atomic.
7106 	 * But for modules, nothing runs the code we modify
7107 	 * until we are finished with it, and there's no
7108 	 * reason to cause large interrupt latencies while we do it.
7109 	 */
7110 	if (!mod)
7111 		local_irq_save(flags);
7112 	ftrace_update_code(mod, start_pg);
7113 	if (!mod)
7114 		local_irq_restore(flags);
7115 	ret = 0;
7116  out:
7117 	mutex_unlock(&ftrace_lock);
7118 
7119 	/* We should have used all pages unless we skipped some */
7120 	if (pg_unuse) {
7121 		WARN_ON(!skipped);
7122 		/* Need to synchronize with ftrace_location_range() */
7123 		synchronize_rcu();
7124 		ftrace_free_pages(pg_unuse);
7125 	}
7126 	return ret;
7127 }
7128 
7129 struct ftrace_mod_func {
7130 	struct list_head	list;
7131 	char			*name;
7132 	unsigned long		ip;
7133 	unsigned int		size;
7134 };
7135 
7136 struct ftrace_mod_map {
7137 	struct rcu_head		rcu;
7138 	struct list_head	list;
7139 	struct module		*mod;
7140 	unsigned long		start_addr;
7141 	unsigned long		end_addr;
7142 	struct list_head	funcs;
7143 	unsigned int		num_funcs;
7144 };
7145 
ftrace_get_trampoline_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7146 static int ftrace_get_trampoline_kallsym(unsigned int symnum,
7147 					 unsigned long *value, char *type,
7148 					 char *name, char *module_name,
7149 					 int *exported)
7150 {
7151 	struct ftrace_ops *op;
7152 
7153 	list_for_each_entry_rcu(op, &ftrace_ops_trampoline_list, list) {
7154 		if (!op->trampoline || symnum--)
7155 			continue;
7156 		*value = op->trampoline;
7157 		*type = 't';
7158 		strscpy(name, FTRACE_TRAMPOLINE_SYM, KSYM_NAME_LEN);
7159 		strscpy(module_name, FTRACE_TRAMPOLINE_MOD, MODULE_NAME_LEN);
7160 		*exported = 0;
7161 		return 0;
7162 	}
7163 
7164 	return -ERANGE;
7165 }
7166 
7167 #if defined(CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS) || defined(CONFIG_MODULES)
7168 /*
7169  * Check if the current ops references the given ip.
7170  *
7171  * If the ops traces all functions, then it was already accounted for.
7172  * If the ops does not trace the current record function, skip it.
7173  * If the ops ignores the function via notrace filter, skip it.
7174  */
7175 static bool
ops_references_ip(struct ftrace_ops * ops,unsigned long ip)7176 ops_references_ip(struct ftrace_ops *ops, unsigned long ip)
7177 {
7178 	/* If ops isn't enabled, ignore it */
7179 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
7180 		return false;
7181 
7182 	/* If ops traces all then it includes this function */
7183 	if (ops_traces_mod(ops))
7184 		return true;
7185 
7186 	/* The function must be in the filter */
7187 	if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
7188 	    !__ftrace_lookup_ip(ops->func_hash->filter_hash, ip))
7189 		return false;
7190 
7191 	/* If in notrace hash, we ignore it too */
7192 	if (ftrace_lookup_ip(ops->func_hash->notrace_hash, ip))
7193 		return false;
7194 
7195 	return true;
7196 }
7197 #endif
7198 
7199 #ifdef CONFIG_MODULES
7200 
7201 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
7202 
7203 static LIST_HEAD(ftrace_mod_maps);
7204 
referenced_filters(struct dyn_ftrace * rec)7205 static int referenced_filters(struct dyn_ftrace *rec)
7206 {
7207 	struct ftrace_ops *ops;
7208 	int cnt = 0;
7209 
7210 	for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
7211 		if (ops_references_ip(ops, rec->ip)) {
7212 			if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_DIRECT))
7213 				continue;
7214 			if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_IPMODIFY))
7215 				continue;
7216 			cnt++;
7217 			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
7218 				rec->flags |= FTRACE_FL_REGS;
7219 			if (cnt == 1 && ops->trampoline)
7220 				rec->flags |= FTRACE_FL_TRAMP;
7221 			else
7222 				rec->flags &= ~FTRACE_FL_TRAMP;
7223 		}
7224 	}
7225 
7226 	return cnt;
7227 }
7228 
7229 static void
clear_mod_from_hash(struct ftrace_page * pg,struct ftrace_hash * hash)7230 clear_mod_from_hash(struct ftrace_page *pg, struct ftrace_hash *hash)
7231 {
7232 	struct ftrace_func_entry *entry;
7233 	struct dyn_ftrace *rec;
7234 	int i;
7235 
7236 	if (ftrace_hash_empty(hash))
7237 		return;
7238 
7239 	for (i = 0; i < pg->index; i++) {
7240 		rec = &pg->records[i];
7241 		entry = __ftrace_lookup_ip(hash, rec->ip);
7242 		/*
7243 		 * Do not allow this rec to match again.
7244 		 * Yeah, it may waste some memory, but will be removed
7245 		 * if/when the hash is modified again.
7246 		 */
7247 		if (entry)
7248 			entry->ip = 0;
7249 	}
7250 }
7251 
7252 /* Clear any records from hashes */
clear_mod_from_hashes(struct ftrace_page * pg)7253 static void clear_mod_from_hashes(struct ftrace_page *pg)
7254 {
7255 	struct trace_array *tr;
7256 
7257 	mutex_lock(&trace_types_lock);
7258 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7259 		if (!tr->ops || !tr->ops->func_hash)
7260 			continue;
7261 		mutex_lock(&tr->ops->func_hash->regex_lock);
7262 		clear_mod_from_hash(pg, tr->ops->func_hash->filter_hash);
7263 		clear_mod_from_hash(pg, tr->ops->func_hash->notrace_hash);
7264 		mutex_unlock(&tr->ops->func_hash->regex_lock);
7265 	}
7266 	mutex_unlock(&trace_types_lock);
7267 }
7268 
ftrace_free_mod_map(struct rcu_head * rcu)7269 static void ftrace_free_mod_map(struct rcu_head *rcu)
7270 {
7271 	struct ftrace_mod_map *mod_map = container_of(rcu, struct ftrace_mod_map, rcu);
7272 	struct ftrace_mod_func *mod_func;
7273 	struct ftrace_mod_func *n;
7274 
7275 	/* All the contents of mod_map are now not visible to readers */
7276 	list_for_each_entry_safe(mod_func, n, &mod_map->funcs, list) {
7277 		kfree(mod_func->name);
7278 		list_del(&mod_func->list);
7279 		kfree(mod_func);
7280 	}
7281 
7282 	kfree(mod_map);
7283 }
7284 
ftrace_release_mod(struct module * mod)7285 void ftrace_release_mod(struct module *mod)
7286 {
7287 	struct ftrace_mod_map *mod_map;
7288 	struct ftrace_mod_map *n;
7289 	struct dyn_ftrace *rec;
7290 	struct ftrace_page **last_pg;
7291 	struct ftrace_page *tmp_page = NULL;
7292 	struct ftrace_page *pg;
7293 
7294 	mutex_lock(&ftrace_lock);
7295 
7296 	if (ftrace_disabled)
7297 		goto out_unlock;
7298 
7299 	list_for_each_entry_safe(mod_map, n, &ftrace_mod_maps, list) {
7300 		if (mod_map->mod == mod) {
7301 			list_del_rcu(&mod_map->list);
7302 			call_rcu(&mod_map->rcu, ftrace_free_mod_map);
7303 			break;
7304 		}
7305 	}
7306 
7307 	/*
7308 	 * Each module has its own ftrace_pages, remove
7309 	 * them from the list.
7310 	 */
7311 	last_pg = &ftrace_pages_start;
7312 	for (pg = ftrace_pages_start; pg; pg = *last_pg) {
7313 		rec = &pg->records[0];
7314 		if (within_module(rec->ip, mod)) {
7315 			/*
7316 			 * As core pages are first, the first
7317 			 * page should never be a module page.
7318 			 */
7319 			if (WARN_ON(pg == ftrace_pages_start))
7320 				goto out_unlock;
7321 
7322 			/* Check if we are deleting the last page */
7323 			if (pg == ftrace_pages)
7324 				ftrace_pages = next_to_ftrace_page(last_pg);
7325 
7326 			ftrace_update_tot_cnt -= pg->index;
7327 			*last_pg = pg->next;
7328 
7329 			pg->next = tmp_page;
7330 			tmp_page = pg;
7331 		} else
7332 			last_pg = &pg->next;
7333 	}
7334  out_unlock:
7335 	mutex_unlock(&ftrace_lock);
7336 
7337 	/* Need to synchronize with ftrace_location_range() */
7338 	if (tmp_page)
7339 		synchronize_rcu();
7340 	for (pg = tmp_page; pg; pg = tmp_page) {
7341 
7342 		/* Needs to be called outside of ftrace_lock */
7343 		clear_mod_from_hashes(pg);
7344 
7345 		if (pg->records) {
7346 			free_pages((unsigned long)pg->records, pg->order);
7347 			ftrace_number_of_pages -= 1 << pg->order;
7348 		}
7349 		tmp_page = pg->next;
7350 		kfree(pg);
7351 		ftrace_number_of_groups--;
7352 	}
7353 }
7354 
ftrace_module_enable(struct module * mod)7355 void ftrace_module_enable(struct module *mod)
7356 {
7357 	struct dyn_ftrace *rec;
7358 	struct ftrace_page *pg;
7359 
7360 	mutex_lock(&ftrace_lock);
7361 
7362 	if (ftrace_disabled)
7363 		goto out_unlock;
7364 
7365 	/*
7366 	 * If the tracing is enabled, go ahead and enable the record.
7367 	 *
7368 	 * The reason not to enable the record immediately is the
7369 	 * inherent check of ftrace_make_nop/ftrace_make_call for
7370 	 * correct previous instructions.  Making first the NOP
7371 	 * conversion puts the module to the correct state, thus
7372 	 * passing the ftrace_make_call check.
7373 	 *
7374 	 * We also delay this to after the module code already set the
7375 	 * text to read-only, as we now need to set it back to read-write
7376 	 * so that we can modify the text.
7377 	 */
7378 	if (ftrace_start_up)
7379 		ftrace_arch_code_modify_prepare();
7380 
7381 	do_for_each_ftrace_rec(pg, rec) {
7382 		int cnt;
7383 		/*
7384 		 * do_for_each_ftrace_rec() is a double loop.
7385 		 * module text shares the pg. If a record is
7386 		 * not part of this module, then skip this pg,
7387 		 * which the "break" will do.
7388 		 */
7389 		if (!within_module(rec->ip, mod))
7390 			break;
7391 
7392 		/* Weak functions should still be ignored */
7393 		if (!test_for_valid_rec(rec)) {
7394 			/* Clear all other flags. Should not be enabled anyway */
7395 			rec->flags = FTRACE_FL_DISABLED;
7396 			continue;
7397 		}
7398 
7399 		cnt = 0;
7400 
7401 		/*
7402 		 * When adding a module, we need to check if tracers are
7403 		 * currently enabled and if they are, and can trace this record,
7404 		 * we need to enable the module functions as well as update the
7405 		 * reference counts for those function records.
7406 		 */
7407 		if (ftrace_start_up)
7408 			cnt += referenced_filters(rec);
7409 
7410 		rec->flags &= ~FTRACE_FL_DISABLED;
7411 		rec->flags += cnt;
7412 
7413 		if (ftrace_start_up && cnt) {
7414 			int failed = __ftrace_replace_code(rec, 1);
7415 			if (failed) {
7416 				ftrace_bug(failed, rec);
7417 				goto out_loop;
7418 			}
7419 		}
7420 
7421 	} while_for_each_ftrace_rec();
7422 
7423  out_loop:
7424 	if (ftrace_start_up)
7425 		ftrace_arch_code_modify_post_process();
7426 
7427  out_unlock:
7428 	mutex_unlock(&ftrace_lock);
7429 
7430 	process_cached_mods(mod->name);
7431 }
7432 
ftrace_module_init(struct module * mod)7433 void ftrace_module_init(struct module *mod)
7434 {
7435 	int ret;
7436 
7437 	if (ftrace_disabled || !mod->num_ftrace_callsites)
7438 		return;
7439 
7440 	ret = ftrace_process_locs(mod, mod->ftrace_callsites,
7441 				  mod->ftrace_callsites + mod->num_ftrace_callsites);
7442 	if (ret)
7443 		pr_warn("ftrace: failed to allocate entries for module '%s' functions\n",
7444 			mod->name);
7445 }
7446 
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)7447 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
7448 				struct dyn_ftrace *rec)
7449 {
7450 	struct ftrace_mod_func *mod_func;
7451 	unsigned long symsize;
7452 	unsigned long offset;
7453 	char str[KSYM_SYMBOL_LEN];
7454 	char *modname;
7455 	const char *ret;
7456 
7457 	ret = kallsyms_lookup(rec->ip, &symsize, &offset, &modname, str);
7458 	if (!ret)
7459 		return;
7460 
7461 	mod_func = kmalloc(sizeof(*mod_func), GFP_KERNEL);
7462 	if (!mod_func)
7463 		return;
7464 
7465 	mod_func->name = kstrdup(str, GFP_KERNEL);
7466 	if (!mod_func->name) {
7467 		kfree(mod_func);
7468 		return;
7469 	}
7470 
7471 	mod_func->ip = rec->ip - offset;
7472 	mod_func->size = symsize;
7473 
7474 	mod_map->num_funcs++;
7475 
7476 	list_add_rcu(&mod_func->list, &mod_map->funcs);
7477 }
7478 
7479 static struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)7480 allocate_ftrace_mod_map(struct module *mod,
7481 			unsigned long start, unsigned long end)
7482 {
7483 	struct ftrace_mod_map *mod_map;
7484 
7485 	mod_map = kmalloc(sizeof(*mod_map), GFP_KERNEL);
7486 	if (!mod_map)
7487 		return NULL;
7488 
7489 	mod_map->mod = mod;
7490 	mod_map->start_addr = start;
7491 	mod_map->end_addr = end;
7492 	mod_map->num_funcs = 0;
7493 
7494 	INIT_LIST_HEAD_RCU(&mod_map->funcs);
7495 
7496 	list_add_rcu(&mod_map->list, &ftrace_mod_maps);
7497 
7498 	return mod_map;
7499 }
7500 
7501 static int
ftrace_func_address_lookup(struct ftrace_mod_map * mod_map,unsigned long addr,unsigned long * size,unsigned long * off,char * sym)7502 ftrace_func_address_lookup(struct ftrace_mod_map *mod_map,
7503 			   unsigned long addr, unsigned long *size,
7504 			   unsigned long *off, char *sym)
7505 {
7506 	struct ftrace_mod_func *found_func =  NULL;
7507 	struct ftrace_mod_func *mod_func;
7508 
7509 	list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
7510 		if (addr >= mod_func->ip &&
7511 		    addr < mod_func->ip + mod_func->size) {
7512 			found_func = mod_func;
7513 			break;
7514 		}
7515 	}
7516 
7517 	if (found_func) {
7518 		if (size)
7519 			*size = found_func->size;
7520 		if (off)
7521 			*off = addr - found_func->ip;
7522 		return strscpy(sym, found_func->name, KSYM_NAME_LEN);
7523 	}
7524 
7525 	return 0;
7526 }
7527 
7528 int
ftrace_mod_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)7529 ftrace_mod_address_lookup(unsigned long addr, unsigned long *size,
7530 		   unsigned long *off, char **modname, char *sym)
7531 {
7532 	struct ftrace_mod_map *mod_map;
7533 	int ret = 0;
7534 
7535 	/* mod_map is freed via call_rcu() */
7536 	preempt_disable();
7537 	list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
7538 		ret = ftrace_func_address_lookup(mod_map, addr, size, off, sym);
7539 		if (ret) {
7540 			if (modname)
7541 				*modname = mod_map->mod->name;
7542 			break;
7543 		}
7544 	}
7545 	preempt_enable();
7546 
7547 	return ret;
7548 }
7549 
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7550 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
7551 			   char *type, char *name,
7552 			   char *module_name, int *exported)
7553 {
7554 	struct ftrace_mod_map *mod_map;
7555 	struct ftrace_mod_func *mod_func;
7556 	int ret;
7557 
7558 	preempt_disable();
7559 	list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
7560 
7561 		if (symnum >= mod_map->num_funcs) {
7562 			symnum -= mod_map->num_funcs;
7563 			continue;
7564 		}
7565 
7566 		list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
7567 			if (symnum > 1) {
7568 				symnum--;
7569 				continue;
7570 			}
7571 
7572 			*value = mod_func->ip;
7573 			*type = 'T';
7574 			strscpy(name, mod_func->name, KSYM_NAME_LEN);
7575 			strscpy(module_name, mod_map->mod->name, MODULE_NAME_LEN);
7576 			*exported = 1;
7577 			preempt_enable();
7578 			return 0;
7579 		}
7580 		WARN_ON(1);
7581 		break;
7582 	}
7583 	ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
7584 					    module_name, exported);
7585 	preempt_enable();
7586 	return ret;
7587 }
7588 
7589 #else
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)7590 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
7591 				struct dyn_ftrace *rec) { }
7592 static inline struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)7593 allocate_ftrace_mod_map(struct module *mod,
7594 			unsigned long start, unsigned long end)
7595 {
7596 	return NULL;
7597 }
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7598 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
7599 			   char *type, char *name, char *module_name,
7600 			   int *exported)
7601 {
7602 	int ret;
7603 
7604 	preempt_disable();
7605 	ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
7606 					    module_name, exported);
7607 	preempt_enable();
7608 	return ret;
7609 }
7610 #endif /* CONFIG_MODULES */
7611 
7612 struct ftrace_init_func {
7613 	struct list_head list;
7614 	unsigned long ip;
7615 };
7616 
7617 /* Clear any init ips from hashes */
7618 static void
clear_func_from_hash(struct ftrace_init_func * func,struct ftrace_hash * hash)7619 clear_func_from_hash(struct ftrace_init_func *func, struct ftrace_hash *hash)
7620 {
7621 	struct ftrace_func_entry *entry;
7622 
7623 	entry = ftrace_lookup_ip(hash, func->ip);
7624 	/*
7625 	 * Do not allow this rec to match again.
7626 	 * Yeah, it may waste some memory, but will be removed
7627 	 * if/when the hash is modified again.
7628 	 */
7629 	if (entry)
7630 		entry->ip = 0;
7631 }
7632 
7633 static void
clear_func_from_hashes(struct ftrace_init_func * func)7634 clear_func_from_hashes(struct ftrace_init_func *func)
7635 {
7636 	struct trace_array *tr;
7637 
7638 	mutex_lock(&trace_types_lock);
7639 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7640 		if (!tr->ops || !tr->ops->func_hash)
7641 			continue;
7642 		mutex_lock(&tr->ops->func_hash->regex_lock);
7643 		clear_func_from_hash(func, tr->ops->func_hash->filter_hash);
7644 		clear_func_from_hash(func, tr->ops->func_hash->notrace_hash);
7645 		mutex_unlock(&tr->ops->func_hash->regex_lock);
7646 	}
7647 	mutex_unlock(&trace_types_lock);
7648 }
7649 
add_to_clear_hash_list(struct list_head * clear_list,struct dyn_ftrace * rec)7650 static void add_to_clear_hash_list(struct list_head *clear_list,
7651 				   struct dyn_ftrace *rec)
7652 {
7653 	struct ftrace_init_func *func;
7654 
7655 	func = kmalloc(sizeof(*func), GFP_KERNEL);
7656 	if (!func) {
7657 		MEM_FAIL(1, "alloc failure, ftrace filter could be stale\n");
7658 		return;
7659 	}
7660 
7661 	func->ip = rec->ip;
7662 	list_add(&func->list, clear_list);
7663 }
7664 
ftrace_free_mem(struct module * mod,void * start_ptr,void * end_ptr)7665 void ftrace_free_mem(struct module *mod, void *start_ptr, void *end_ptr)
7666 {
7667 	unsigned long start = (unsigned long)(start_ptr);
7668 	unsigned long end = (unsigned long)(end_ptr);
7669 	struct ftrace_page **last_pg = &ftrace_pages_start;
7670 	struct ftrace_page *tmp_page = NULL;
7671 	struct ftrace_page *pg;
7672 	struct dyn_ftrace *rec;
7673 	struct dyn_ftrace key;
7674 	struct ftrace_mod_map *mod_map = NULL;
7675 	struct ftrace_init_func *func, *func_next;
7676 	LIST_HEAD(clear_hash);
7677 
7678 	key.ip = start;
7679 	key.flags = end;	/* overload flags, as it is unsigned long */
7680 
7681 	mutex_lock(&ftrace_lock);
7682 
7683 	/*
7684 	 * If we are freeing module init memory, then check if
7685 	 * any tracer is active. If so, we need to save a mapping of
7686 	 * the module functions being freed with the address.
7687 	 */
7688 	if (mod && ftrace_ops_list != &ftrace_list_end)
7689 		mod_map = allocate_ftrace_mod_map(mod, start, end);
7690 
7691 	for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
7692 		if (end < pg->records[0].ip ||
7693 		    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
7694 			continue;
7695  again:
7696 		rec = bsearch(&key, pg->records, pg->index,
7697 			      sizeof(struct dyn_ftrace),
7698 			      ftrace_cmp_recs);
7699 		if (!rec)
7700 			continue;
7701 
7702 		/* rec will be cleared from hashes after ftrace_lock unlock */
7703 		add_to_clear_hash_list(&clear_hash, rec);
7704 
7705 		if (mod_map)
7706 			save_ftrace_mod_rec(mod_map, rec);
7707 
7708 		pg->index--;
7709 		ftrace_update_tot_cnt--;
7710 		if (!pg->index) {
7711 			*last_pg = pg->next;
7712 			pg->next = tmp_page;
7713 			tmp_page = pg;
7714 			pg = container_of(last_pg, struct ftrace_page, next);
7715 			if (!(*last_pg))
7716 				ftrace_pages = pg;
7717 			continue;
7718 		}
7719 		memmove(rec, rec + 1,
7720 			(pg->index - (rec - pg->records)) * sizeof(*rec));
7721 		/* More than one function may be in this block */
7722 		goto again;
7723 	}
7724 	mutex_unlock(&ftrace_lock);
7725 
7726 	list_for_each_entry_safe(func, func_next, &clear_hash, list) {
7727 		clear_func_from_hashes(func);
7728 		kfree(func);
7729 	}
7730 	/* Need to synchronize with ftrace_location_range() */
7731 	if (tmp_page) {
7732 		synchronize_rcu();
7733 		ftrace_free_pages(tmp_page);
7734 	}
7735 }
7736 
ftrace_free_init_mem(void)7737 void __init ftrace_free_init_mem(void)
7738 {
7739 	void *start = (void *)(&__init_begin);
7740 	void *end = (void *)(&__init_end);
7741 
7742 	ftrace_boot_snapshot();
7743 
7744 	ftrace_free_mem(NULL, start, end);
7745 }
7746 
ftrace_dyn_arch_init(void)7747 int __init __weak ftrace_dyn_arch_init(void)
7748 {
7749 	return 0;
7750 }
7751 
ftrace_init(void)7752 void __init ftrace_init(void)
7753 {
7754 	extern unsigned long __start_mcount_loc[];
7755 	extern unsigned long __stop_mcount_loc[];
7756 	unsigned long count, flags;
7757 	int ret;
7758 
7759 	local_irq_save(flags);
7760 	ret = ftrace_dyn_arch_init();
7761 	local_irq_restore(flags);
7762 	if (ret)
7763 		goto failed;
7764 
7765 	count = __stop_mcount_loc - __start_mcount_loc;
7766 	if (!count) {
7767 		pr_info("ftrace: No functions to be traced?\n");
7768 		goto failed;
7769 	}
7770 
7771 	pr_info("ftrace: allocating %ld entries in %ld pages\n",
7772 		count, DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
7773 
7774 	ret = ftrace_process_locs(NULL,
7775 				  __start_mcount_loc,
7776 				  __stop_mcount_loc);
7777 	if (ret) {
7778 		pr_warn("ftrace: failed to allocate entries for functions\n");
7779 		goto failed;
7780 	}
7781 
7782 	pr_info("ftrace: allocated %ld pages with %ld groups\n",
7783 		ftrace_number_of_pages, ftrace_number_of_groups);
7784 
7785 	last_ftrace_enabled = ftrace_enabled = 1;
7786 
7787 	set_ftrace_early_filters();
7788 
7789 	return;
7790  failed:
7791 	ftrace_disabled = 1;
7792 }
7793 
7794 /* Do nothing if arch does not support this */
arch_ftrace_update_trampoline(struct ftrace_ops * ops)7795 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
7796 {
7797 }
7798 
ftrace_update_trampoline(struct ftrace_ops * ops)7799 static void ftrace_update_trampoline(struct ftrace_ops *ops)
7800 {
7801 	unsigned long trampoline = ops->trampoline;
7802 
7803 	arch_ftrace_update_trampoline(ops);
7804 	if (ops->trampoline && ops->trampoline != trampoline &&
7805 	    (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP)) {
7806 		/* Add to kallsyms before the perf events */
7807 		ftrace_add_trampoline_to_kallsyms(ops);
7808 		perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
7809 				   ops->trampoline, ops->trampoline_size, false,
7810 				   FTRACE_TRAMPOLINE_SYM);
7811 		/*
7812 		 * Record the perf text poke event after the ksymbol register
7813 		 * event.
7814 		 */
7815 		perf_event_text_poke((void *)ops->trampoline, NULL, 0,
7816 				     (void *)ops->trampoline,
7817 				     ops->trampoline_size);
7818 	}
7819 }
7820 
ftrace_init_trace_array(struct trace_array * tr)7821 void ftrace_init_trace_array(struct trace_array *tr)
7822 {
7823 	if (tr->flags & TRACE_ARRAY_FL_MOD_INIT)
7824 		return;
7825 
7826 	INIT_LIST_HEAD(&tr->func_probes);
7827 	INIT_LIST_HEAD(&tr->mod_trace);
7828 	INIT_LIST_HEAD(&tr->mod_notrace);
7829 
7830 	tr->flags |= TRACE_ARRAY_FL_MOD_INIT;
7831 }
7832 #else
7833 
7834 struct ftrace_ops global_ops = {
7835 	.func			= ftrace_stub,
7836 	.flags			= FTRACE_OPS_FL_INITIALIZED |
7837 				  FTRACE_OPS_FL_PID,
7838 };
7839 
ftrace_nodyn_init(void)7840 static int __init ftrace_nodyn_init(void)
7841 {
7842 	ftrace_enabled = 1;
7843 	return 0;
7844 }
7845 core_initcall(ftrace_nodyn_init);
7846 
ftrace_init_dyn_tracefs(struct dentry * d_tracer)7847 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
ftrace_startup_all(int command)7848 static inline void ftrace_startup_all(int command) { }
7849 
ftrace_update_trampoline(struct ftrace_ops * ops)7850 static void ftrace_update_trampoline(struct ftrace_ops *ops)
7851 {
7852 }
7853 
7854 #endif /* CONFIG_DYNAMIC_FTRACE */
7855 
ftrace_init_global_array_ops(struct trace_array * tr)7856 __init void ftrace_init_global_array_ops(struct trace_array *tr)
7857 {
7858 	tr->ops = &global_ops;
7859 	if (!global_ops.private)
7860 		global_ops.private = tr;
7861 	ftrace_init_trace_array(tr);
7862 	init_array_fgraph_ops(tr, tr->ops);
7863 }
7864 
ftrace_init_array_ops(struct trace_array * tr,ftrace_func_t func)7865 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
7866 {
7867 	/* If we filter on pids, update to use the pid function */
7868 	if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
7869 		if (WARN_ON(tr->ops->func != ftrace_stub))
7870 			printk("ftrace ops had %pS for function\n",
7871 			       tr->ops->func);
7872 	}
7873 	tr->ops->func = func;
7874 	tr->ops->private = tr;
7875 }
7876 
ftrace_reset_array_ops(struct trace_array * tr)7877 void ftrace_reset_array_ops(struct trace_array *tr)
7878 {
7879 	tr->ops->func = ftrace_stub;
7880 }
7881 
7882 static nokprobe_inline void
__ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * ignored,struct ftrace_regs * fregs)7883 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
7884 		       struct ftrace_ops *ignored, struct ftrace_regs *fregs)
7885 {
7886 	struct pt_regs *regs = ftrace_get_regs(fregs);
7887 	struct ftrace_ops *op;
7888 	int bit;
7889 
7890 	/*
7891 	 * The ftrace_test_and_set_recursion() will disable preemption,
7892 	 * which is required since some of the ops may be dynamically
7893 	 * allocated, they must be freed after a synchronize_rcu().
7894 	 */
7895 	bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
7896 	if (bit < 0)
7897 		return;
7898 
7899 	do_for_each_ftrace_op(op, ftrace_ops_list) {
7900 		/* Stub functions don't need to be called nor tested */
7901 		if (op->flags & FTRACE_OPS_FL_STUB)
7902 			continue;
7903 		/*
7904 		 * Check the following for each ops before calling their func:
7905 		 *  if RCU flag is set, then rcu_is_watching() must be true
7906 		 *  Otherwise test if the ip matches the ops filter
7907 		 *
7908 		 * If any of the above fails then the op->func() is not executed.
7909 		 */
7910 		if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
7911 		    ftrace_ops_test(op, ip, regs)) {
7912 			if (FTRACE_WARN_ON(!op->func)) {
7913 				pr_warn("op=%p %pS\n", op, op);
7914 				goto out;
7915 			}
7916 			op->func(ip, parent_ip, op, fregs);
7917 		}
7918 	} while_for_each_ftrace_op(op);
7919 out:
7920 	trace_clear_recursion(bit);
7921 }
7922 
7923 /*
7924  * Some archs only support passing ip and parent_ip. Even though
7925  * the list function ignores the op parameter, we do not want any
7926  * C side effects, where a function is called without the caller
7927  * sending a third parameter.
7928  * Archs are to support both the regs and ftrace_ops at the same time.
7929  * If they support ftrace_ops, it is assumed they support regs.
7930  * If call backs want to use regs, they must either check for regs
7931  * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
7932  * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
7933  * An architecture can pass partial regs with ftrace_ops and still
7934  * set the ARCH_SUPPORTS_FTRACE_OPS.
7935  *
7936  * In vmlinux.lds.h, ftrace_ops_list_func() is defined to be
7937  * arch_ftrace_ops_list_func.
7938  */
7939 #if ARCH_SUPPORTS_FTRACE_OPS
arch_ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)7940 void arch_ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
7941 			       struct ftrace_ops *op, struct ftrace_regs *fregs)
7942 {
7943 	kmsan_unpoison_memory(fregs, ftrace_regs_size());
7944 	__ftrace_ops_list_func(ip, parent_ip, NULL, fregs);
7945 }
7946 #else
arch_ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip)7947 void arch_ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
7948 {
7949 	__ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
7950 }
7951 #endif
7952 NOKPROBE_SYMBOL(arch_ftrace_ops_list_func);
7953 
7954 /*
7955  * If there's only one function registered but it does not support
7956  * recursion, needs RCU protection, then this function will be called
7957  * by the mcount trampoline.
7958  */
ftrace_ops_assist_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)7959 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
7960 				   struct ftrace_ops *op, struct ftrace_regs *fregs)
7961 {
7962 	int bit;
7963 
7964 	bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
7965 	if (bit < 0)
7966 		return;
7967 
7968 	if (!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching())
7969 		op->func(ip, parent_ip, op, fregs);
7970 
7971 	trace_clear_recursion(bit);
7972 }
7973 NOKPROBE_SYMBOL(ftrace_ops_assist_func);
7974 
7975 /**
7976  * ftrace_ops_get_func - get the function a trampoline should call
7977  * @ops: the ops to get the function for
7978  *
7979  * Normally the mcount trampoline will call the ops->func, but there
7980  * are times that it should not. For example, if the ops does not
7981  * have its own recursion protection, then it should call the
7982  * ftrace_ops_assist_func() instead.
7983  *
7984  * Returns: the function that the trampoline should call for @ops.
7985  */
ftrace_ops_get_func(struct ftrace_ops * ops)7986 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
7987 {
7988 	/*
7989 	 * If the function does not handle recursion or needs to be RCU safe,
7990 	 * then we need to call the assist handler.
7991 	 */
7992 	if (ops->flags & (FTRACE_OPS_FL_RECURSION |
7993 			  FTRACE_OPS_FL_RCU))
7994 		return ftrace_ops_assist_func;
7995 
7996 	return ops->func;
7997 }
7998 
7999 static void
ftrace_filter_pid_sched_switch_probe(void * data,bool preempt,struct task_struct * prev,struct task_struct * next,unsigned int prev_state)8000 ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
8001 				     struct task_struct *prev,
8002 				     struct task_struct *next,
8003 				     unsigned int prev_state)
8004 {
8005 	struct trace_array *tr = data;
8006 	struct trace_pid_list *pid_list;
8007 	struct trace_pid_list *no_pid_list;
8008 
8009 	pid_list = rcu_dereference_sched(tr->function_pids);
8010 	no_pid_list = rcu_dereference_sched(tr->function_no_pids);
8011 
8012 	if (trace_ignore_this_task(pid_list, no_pid_list, next))
8013 		this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8014 			       FTRACE_PID_IGNORE);
8015 	else
8016 		this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8017 			       next->pid);
8018 }
8019 
8020 static void
ftrace_pid_follow_sched_process_fork(void * data,struct task_struct * self,struct task_struct * task)8021 ftrace_pid_follow_sched_process_fork(void *data,
8022 				     struct task_struct *self,
8023 				     struct task_struct *task)
8024 {
8025 	struct trace_pid_list *pid_list;
8026 	struct trace_array *tr = data;
8027 
8028 	pid_list = rcu_dereference_sched(tr->function_pids);
8029 	trace_filter_add_remove_task(pid_list, self, task);
8030 
8031 	pid_list = rcu_dereference_sched(tr->function_no_pids);
8032 	trace_filter_add_remove_task(pid_list, self, task);
8033 }
8034 
8035 static void
ftrace_pid_follow_sched_process_exit(void * data,struct task_struct * task)8036 ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
8037 {
8038 	struct trace_pid_list *pid_list;
8039 	struct trace_array *tr = data;
8040 
8041 	pid_list = rcu_dereference_sched(tr->function_pids);
8042 	trace_filter_add_remove_task(pid_list, NULL, task);
8043 
8044 	pid_list = rcu_dereference_sched(tr->function_no_pids);
8045 	trace_filter_add_remove_task(pid_list, NULL, task);
8046 }
8047 
ftrace_pid_follow_fork(struct trace_array * tr,bool enable)8048 void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
8049 {
8050 	if (enable) {
8051 		register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
8052 						  tr);
8053 		register_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
8054 						  tr);
8055 	} else {
8056 		unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
8057 						    tr);
8058 		unregister_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
8059 						    tr);
8060 	}
8061 }
8062 
clear_ftrace_pids(struct trace_array * tr,int type)8063 static void clear_ftrace_pids(struct trace_array *tr, int type)
8064 {
8065 	struct trace_pid_list *pid_list;
8066 	struct trace_pid_list *no_pid_list;
8067 	int cpu;
8068 
8069 	pid_list = rcu_dereference_protected(tr->function_pids,
8070 					     lockdep_is_held(&ftrace_lock));
8071 	no_pid_list = rcu_dereference_protected(tr->function_no_pids,
8072 						lockdep_is_held(&ftrace_lock));
8073 
8074 	/* Make sure there's something to do */
8075 	if (!pid_type_enabled(type, pid_list, no_pid_list))
8076 		return;
8077 
8078 	/* See if the pids still need to be checked after this */
8079 	if (!still_need_pid_events(type, pid_list, no_pid_list)) {
8080 		unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
8081 		for_each_possible_cpu(cpu)
8082 			per_cpu_ptr(tr->array_buffer.data, cpu)->ftrace_ignore_pid = FTRACE_PID_TRACE;
8083 	}
8084 
8085 	if (type & TRACE_PIDS)
8086 		rcu_assign_pointer(tr->function_pids, NULL);
8087 
8088 	if (type & TRACE_NO_PIDS)
8089 		rcu_assign_pointer(tr->function_no_pids, NULL);
8090 
8091 	/* Wait till all users are no longer using pid filtering */
8092 	synchronize_rcu();
8093 
8094 	if ((type & TRACE_PIDS) && pid_list)
8095 		trace_pid_list_free(pid_list);
8096 
8097 	if ((type & TRACE_NO_PIDS) && no_pid_list)
8098 		trace_pid_list_free(no_pid_list);
8099 }
8100 
ftrace_clear_pids(struct trace_array * tr)8101 void ftrace_clear_pids(struct trace_array *tr)
8102 {
8103 	mutex_lock(&ftrace_lock);
8104 
8105 	clear_ftrace_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
8106 
8107 	mutex_unlock(&ftrace_lock);
8108 }
8109 
ftrace_pid_reset(struct trace_array * tr,int type)8110 static void ftrace_pid_reset(struct trace_array *tr, int type)
8111 {
8112 	mutex_lock(&ftrace_lock);
8113 	clear_ftrace_pids(tr, type);
8114 
8115 	ftrace_update_pid_func();
8116 	ftrace_startup_all(0);
8117 
8118 	mutex_unlock(&ftrace_lock);
8119 }
8120 
8121 /* Greater than any max PID */
8122 #define FTRACE_NO_PIDS		(void *)(PID_MAX_LIMIT + 1)
8123 
fpid_start(struct seq_file * m,loff_t * pos)8124 static void *fpid_start(struct seq_file *m, loff_t *pos)
8125 	__acquires(RCU)
8126 {
8127 	struct trace_pid_list *pid_list;
8128 	struct trace_array *tr = m->private;
8129 
8130 	mutex_lock(&ftrace_lock);
8131 	rcu_read_lock_sched();
8132 
8133 	pid_list = rcu_dereference_sched(tr->function_pids);
8134 
8135 	if (!pid_list)
8136 		return !(*pos) ? FTRACE_NO_PIDS : NULL;
8137 
8138 	return trace_pid_start(pid_list, pos);
8139 }
8140 
fpid_next(struct seq_file * m,void * v,loff_t * pos)8141 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
8142 {
8143 	struct trace_array *tr = m->private;
8144 	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
8145 
8146 	if (v == FTRACE_NO_PIDS) {
8147 		(*pos)++;
8148 		return NULL;
8149 	}
8150 	return trace_pid_next(pid_list, v, pos);
8151 }
8152 
fpid_stop(struct seq_file * m,void * p)8153 static void fpid_stop(struct seq_file *m, void *p)
8154 	__releases(RCU)
8155 {
8156 	rcu_read_unlock_sched();
8157 	mutex_unlock(&ftrace_lock);
8158 }
8159 
fpid_show(struct seq_file * m,void * v)8160 static int fpid_show(struct seq_file *m, void *v)
8161 {
8162 	if (v == FTRACE_NO_PIDS) {
8163 		seq_puts(m, "no pid\n");
8164 		return 0;
8165 	}
8166 
8167 	return trace_pid_show(m, v);
8168 }
8169 
8170 static const struct seq_operations ftrace_pid_sops = {
8171 	.start = fpid_start,
8172 	.next = fpid_next,
8173 	.stop = fpid_stop,
8174 	.show = fpid_show,
8175 };
8176 
fnpid_start(struct seq_file * m,loff_t * pos)8177 static void *fnpid_start(struct seq_file *m, loff_t *pos)
8178 	__acquires(RCU)
8179 {
8180 	struct trace_pid_list *pid_list;
8181 	struct trace_array *tr = m->private;
8182 
8183 	mutex_lock(&ftrace_lock);
8184 	rcu_read_lock_sched();
8185 
8186 	pid_list = rcu_dereference_sched(tr->function_no_pids);
8187 
8188 	if (!pid_list)
8189 		return !(*pos) ? FTRACE_NO_PIDS : NULL;
8190 
8191 	return trace_pid_start(pid_list, pos);
8192 }
8193 
fnpid_next(struct seq_file * m,void * v,loff_t * pos)8194 static void *fnpid_next(struct seq_file *m, void *v, loff_t *pos)
8195 {
8196 	struct trace_array *tr = m->private;
8197 	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_no_pids);
8198 
8199 	if (v == FTRACE_NO_PIDS) {
8200 		(*pos)++;
8201 		return NULL;
8202 	}
8203 	return trace_pid_next(pid_list, v, pos);
8204 }
8205 
8206 static const struct seq_operations ftrace_no_pid_sops = {
8207 	.start = fnpid_start,
8208 	.next = fnpid_next,
8209 	.stop = fpid_stop,
8210 	.show = fpid_show,
8211 };
8212 
pid_open(struct inode * inode,struct file * file,int type)8213 static int pid_open(struct inode *inode, struct file *file, int type)
8214 {
8215 	const struct seq_operations *seq_ops;
8216 	struct trace_array *tr = inode->i_private;
8217 	struct seq_file *m;
8218 	int ret = 0;
8219 
8220 	ret = tracing_check_open_get_tr(tr);
8221 	if (ret)
8222 		return ret;
8223 
8224 	if ((file->f_mode & FMODE_WRITE) &&
8225 	    (file->f_flags & O_TRUNC))
8226 		ftrace_pid_reset(tr, type);
8227 
8228 	switch (type) {
8229 	case TRACE_PIDS:
8230 		seq_ops = &ftrace_pid_sops;
8231 		break;
8232 	case TRACE_NO_PIDS:
8233 		seq_ops = &ftrace_no_pid_sops;
8234 		break;
8235 	default:
8236 		trace_array_put(tr);
8237 		WARN_ON_ONCE(1);
8238 		return -EINVAL;
8239 	}
8240 
8241 	ret = seq_open(file, seq_ops);
8242 	if (ret < 0) {
8243 		trace_array_put(tr);
8244 	} else {
8245 		m = file->private_data;
8246 		/* copy tr over to seq ops */
8247 		m->private = tr;
8248 	}
8249 
8250 	return ret;
8251 }
8252 
8253 static int
ftrace_pid_open(struct inode * inode,struct file * file)8254 ftrace_pid_open(struct inode *inode, struct file *file)
8255 {
8256 	return pid_open(inode, file, TRACE_PIDS);
8257 }
8258 
8259 static int
ftrace_no_pid_open(struct inode * inode,struct file * file)8260 ftrace_no_pid_open(struct inode *inode, struct file *file)
8261 {
8262 	return pid_open(inode, file, TRACE_NO_PIDS);
8263 }
8264 
ignore_task_cpu(void * data)8265 static void ignore_task_cpu(void *data)
8266 {
8267 	struct trace_array *tr = data;
8268 	struct trace_pid_list *pid_list;
8269 	struct trace_pid_list *no_pid_list;
8270 
8271 	/*
8272 	 * This function is called by on_each_cpu() while the
8273 	 * event_mutex is held.
8274 	 */
8275 	pid_list = rcu_dereference_protected(tr->function_pids,
8276 					     mutex_is_locked(&ftrace_lock));
8277 	no_pid_list = rcu_dereference_protected(tr->function_no_pids,
8278 						mutex_is_locked(&ftrace_lock));
8279 
8280 	if (trace_ignore_this_task(pid_list, no_pid_list, current))
8281 		this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8282 			       FTRACE_PID_IGNORE);
8283 	else
8284 		this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8285 			       current->pid);
8286 }
8287 
8288 static ssize_t
pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos,int type)8289 pid_write(struct file *filp, const char __user *ubuf,
8290 	  size_t cnt, loff_t *ppos, int type)
8291 {
8292 	struct seq_file *m = filp->private_data;
8293 	struct trace_array *tr = m->private;
8294 	struct trace_pid_list *filtered_pids;
8295 	struct trace_pid_list *other_pids;
8296 	struct trace_pid_list *pid_list;
8297 	ssize_t ret;
8298 
8299 	if (!cnt)
8300 		return 0;
8301 
8302 	guard(mutex)(&ftrace_lock);
8303 
8304 	switch (type) {
8305 	case TRACE_PIDS:
8306 		filtered_pids = rcu_dereference_protected(tr->function_pids,
8307 					     lockdep_is_held(&ftrace_lock));
8308 		other_pids = rcu_dereference_protected(tr->function_no_pids,
8309 					     lockdep_is_held(&ftrace_lock));
8310 		break;
8311 	case TRACE_NO_PIDS:
8312 		filtered_pids = rcu_dereference_protected(tr->function_no_pids,
8313 					     lockdep_is_held(&ftrace_lock));
8314 		other_pids = rcu_dereference_protected(tr->function_pids,
8315 					     lockdep_is_held(&ftrace_lock));
8316 		break;
8317 	default:
8318 		WARN_ON_ONCE(1);
8319 		return -EINVAL;
8320 	}
8321 
8322 	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
8323 	if (ret < 0)
8324 		return ret;
8325 
8326 	switch (type) {
8327 	case TRACE_PIDS:
8328 		rcu_assign_pointer(tr->function_pids, pid_list);
8329 		break;
8330 	case TRACE_NO_PIDS:
8331 		rcu_assign_pointer(tr->function_no_pids, pid_list);
8332 		break;
8333 	}
8334 
8335 
8336 	if (filtered_pids) {
8337 		synchronize_rcu();
8338 		trace_pid_list_free(filtered_pids);
8339 	} else if (pid_list && !other_pids) {
8340 		/* Register a probe to set whether to ignore the tracing of a task */
8341 		register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
8342 	}
8343 
8344 	/*
8345 	 * Ignoring of pids is done at task switch. But we have to
8346 	 * check for those tasks that are currently running.
8347 	 * Always do this in case a pid was appended or removed.
8348 	 */
8349 	on_each_cpu(ignore_task_cpu, tr, 1);
8350 
8351 	ftrace_update_pid_func();
8352 	ftrace_startup_all(0);
8353 
8354 	*ppos += ret;
8355 
8356 	return ret;
8357 }
8358 
8359 static ssize_t
ftrace_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)8360 ftrace_pid_write(struct file *filp, const char __user *ubuf,
8361 		 size_t cnt, loff_t *ppos)
8362 {
8363 	return pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
8364 }
8365 
8366 static ssize_t
ftrace_no_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)8367 ftrace_no_pid_write(struct file *filp, const char __user *ubuf,
8368 		    size_t cnt, loff_t *ppos)
8369 {
8370 	return pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
8371 }
8372 
8373 static int
ftrace_pid_release(struct inode * inode,struct file * file)8374 ftrace_pid_release(struct inode *inode, struct file *file)
8375 {
8376 	struct trace_array *tr = inode->i_private;
8377 
8378 	trace_array_put(tr);
8379 
8380 	return seq_release(inode, file);
8381 }
8382 
8383 static const struct file_operations ftrace_pid_fops = {
8384 	.open		= ftrace_pid_open,
8385 	.write		= ftrace_pid_write,
8386 	.read		= seq_read,
8387 	.llseek		= tracing_lseek,
8388 	.release	= ftrace_pid_release,
8389 };
8390 
8391 static const struct file_operations ftrace_no_pid_fops = {
8392 	.open		= ftrace_no_pid_open,
8393 	.write		= ftrace_no_pid_write,
8394 	.read		= seq_read,
8395 	.llseek		= tracing_lseek,
8396 	.release	= ftrace_pid_release,
8397 };
8398 
ftrace_init_tracefs(struct trace_array * tr,struct dentry * d_tracer)8399 void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
8400 {
8401 	trace_create_file("set_ftrace_pid", TRACE_MODE_WRITE, d_tracer,
8402 			    tr, &ftrace_pid_fops);
8403 	trace_create_file("set_ftrace_notrace_pid", TRACE_MODE_WRITE,
8404 			  d_tracer, tr, &ftrace_no_pid_fops);
8405 }
8406 
ftrace_init_tracefs_toplevel(struct trace_array * tr,struct dentry * d_tracer)8407 void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
8408 					 struct dentry *d_tracer)
8409 {
8410 	/* Only the top level directory has the dyn_tracefs and profile */
8411 	WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
8412 
8413 	ftrace_init_dyn_tracefs(d_tracer);
8414 	ftrace_profile_tracefs(d_tracer);
8415 }
8416 
8417 /**
8418  * ftrace_kill - kill ftrace
8419  *
8420  * This function should be used by panic code. It stops ftrace
8421  * but in a not so nice way. If you need to simply kill ftrace
8422  * from a non-atomic section, use ftrace_kill.
8423  */
ftrace_kill(void)8424 void ftrace_kill(void)
8425 {
8426 	ftrace_disabled = 1;
8427 	ftrace_enabled = 0;
8428 	ftrace_trace_function = ftrace_stub;
8429 	kprobe_ftrace_kill();
8430 }
8431 
8432 /**
8433  * ftrace_is_dead - Test if ftrace is dead or not.
8434  *
8435  * Returns: 1 if ftrace is "dead", zero otherwise.
8436  */
ftrace_is_dead(void)8437 int ftrace_is_dead(void)
8438 {
8439 	return ftrace_disabled;
8440 }
8441 
8442 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
8443 /*
8444  * When registering ftrace_ops with IPMODIFY, it is necessary to make sure
8445  * it doesn't conflict with any direct ftrace_ops. If there is existing
8446  * direct ftrace_ops on a kernel function being patched, call
8447  * FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_PEER on it to enable sharing.
8448  *
8449  * @ops:     ftrace_ops being registered.
8450  *
8451  * Returns:
8452  *         0 on success;
8453  *         Negative on failure.
8454  */
prepare_direct_functions_for_ipmodify(struct ftrace_ops * ops)8455 static int prepare_direct_functions_for_ipmodify(struct ftrace_ops *ops)
8456 {
8457 	struct ftrace_func_entry *entry;
8458 	struct ftrace_hash *hash;
8459 	struct ftrace_ops *op;
8460 	int size, i, ret;
8461 
8462 	lockdep_assert_held_once(&direct_mutex);
8463 
8464 	if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
8465 		return 0;
8466 
8467 	hash = ops->func_hash->filter_hash;
8468 	size = 1 << hash->size_bits;
8469 	for (i = 0; i < size; i++) {
8470 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
8471 			unsigned long ip = entry->ip;
8472 			bool found_op = false;
8473 
8474 			mutex_lock(&ftrace_lock);
8475 			do_for_each_ftrace_op(op, ftrace_ops_list) {
8476 				if (!(op->flags & FTRACE_OPS_FL_DIRECT))
8477 					continue;
8478 				if (ops_references_ip(op, ip)) {
8479 					found_op = true;
8480 					break;
8481 				}
8482 			} while_for_each_ftrace_op(op);
8483 			mutex_unlock(&ftrace_lock);
8484 
8485 			if (found_op) {
8486 				if (!op->ops_func)
8487 					return -EBUSY;
8488 
8489 				ret = op->ops_func(op, FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_PEER);
8490 				if (ret)
8491 					return ret;
8492 			}
8493 		}
8494 	}
8495 
8496 	return 0;
8497 }
8498 
8499 /*
8500  * Similar to prepare_direct_functions_for_ipmodify, clean up after ops
8501  * with IPMODIFY is unregistered. The cleanup is optional for most DIRECT
8502  * ops.
8503  */
cleanup_direct_functions_after_ipmodify(struct ftrace_ops * ops)8504 static void cleanup_direct_functions_after_ipmodify(struct ftrace_ops *ops)
8505 {
8506 	struct ftrace_func_entry *entry;
8507 	struct ftrace_hash *hash;
8508 	struct ftrace_ops *op;
8509 	int size, i;
8510 
8511 	if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
8512 		return;
8513 
8514 	mutex_lock(&direct_mutex);
8515 
8516 	hash = ops->func_hash->filter_hash;
8517 	size = 1 << hash->size_bits;
8518 	for (i = 0; i < size; i++) {
8519 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
8520 			unsigned long ip = entry->ip;
8521 			bool found_op = false;
8522 
8523 			mutex_lock(&ftrace_lock);
8524 			do_for_each_ftrace_op(op, ftrace_ops_list) {
8525 				if (!(op->flags & FTRACE_OPS_FL_DIRECT))
8526 					continue;
8527 				if (ops_references_ip(op, ip)) {
8528 					found_op = true;
8529 					break;
8530 				}
8531 			} while_for_each_ftrace_op(op);
8532 			mutex_unlock(&ftrace_lock);
8533 
8534 			/* The cleanup is optional, ignore any errors */
8535 			if (found_op && op->ops_func)
8536 				op->ops_func(op, FTRACE_OPS_CMD_DISABLE_SHARE_IPMODIFY_PEER);
8537 		}
8538 	}
8539 	mutex_unlock(&direct_mutex);
8540 }
8541 
8542 #define lock_direct_mutex()	mutex_lock(&direct_mutex)
8543 #define unlock_direct_mutex()	mutex_unlock(&direct_mutex)
8544 
8545 #else  /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
8546 
prepare_direct_functions_for_ipmodify(struct ftrace_ops * ops)8547 static int prepare_direct_functions_for_ipmodify(struct ftrace_ops *ops)
8548 {
8549 	return 0;
8550 }
8551 
cleanup_direct_functions_after_ipmodify(struct ftrace_ops * ops)8552 static void cleanup_direct_functions_after_ipmodify(struct ftrace_ops *ops)
8553 {
8554 }
8555 
8556 #define lock_direct_mutex()	do { } while (0)
8557 #define unlock_direct_mutex()	do { } while (0)
8558 
8559 #endif  /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
8560 
8561 /*
8562  * Similar to register_ftrace_function, except we don't lock direct_mutex.
8563  */
register_ftrace_function_nolock(struct ftrace_ops * ops)8564 static int register_ftrace_function_nolock(struct ftrace_ops *ops)
8565 {
8566 	int ret;
8567 
8568 	ftrace_ops_init(ops);
8569 
8570 	mutex_lock(&ftrace_lock);
8571 
8572 	ret = ftrace_startup(ops, 0);
8573 
8574 	mutex_unlock(&ftrace_lock);
8575 
8576 	return ret;
8577 }
8578 
8579 /**
8580  * register_ftrace_function - register a function for profiling
8581  * @ops:	ops structure that holds the function for profiling.
8582  *
8583  * Register a function to be called by all functions in the
8584  * kernel.
8585  *
8586  * Note: @ops->func and all the functions it calls must be labeled
8587  *       with "notrace", otherwise it will go into a
8588  *       recursive loop.
8589  */
register_ftrace_function(struct ftrace_ops * ops)8590 int register_ftrace_function(struct ftrace_ops *ops)
8591 {
8592 	int ret;
8593 
8594 	lock_direct_mutex();
8595 	ret = prepare_direct_functions_for_ipmodify(ops);
8596 	if (ret < 0)
8597 		goto out_unlock;
8598 
8599 	ret = register_ftrace_function_nolock(ops);
8600 
8601 out_unlock:
8602 	unlock_direct_mutex();
8603 	return ret;
8604 }
8605 EXPORT_SYMBOL_GPL(register_ftrace_function);
8606 
8607 /**
8608  * unregister_ftrace_function - unregister a function for profiling.
8609  * @ops:	ops structure that holds the function to unregister
8610  *
8611  * Unregister a function that was added to be called by ftrace profiling.
8612  */
unregister_ftrace_function(struct ftrace_ops * ops)8613 int unregister_ftrace_function(struct ftrace_ops *ops)
8614 {
8615 	int ret;
8616 
8617 	mutex_lock(&ftrace_lock);
8618 	ret = ftrace_shutdown(ops, 0);
8619 	mutex_unlock(&ftrace_lock);
8620 
8621 	cleanup_direct_functions_after_ipmodify(ops);
8622 	return ret;
8623 }
8624 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
8625 
symbols_cmp(const void * a,const void * b)8626 static int symbols_cmp(const void *a, const void *b)
8627 {
8628 	const char **str_a = (const char **) a;
8629 	const char **str_b = (const char **) b;
8630 
8631 	return strcmp(*str_a, *str_b);
8632 }
8633 
8634 struct kallsyms_data {
8635 	unsigned long *addrs;
8636 	const char **syms;
8637 	size_t cnt;
8638 	size_t found;
8639 };
8640 
8641 /* This function gets called for all kernel and module symbols
8642  * and returns 1 in case we resolved all the requested symbols,
8643  * 0 otherwise.
8644  */
kallsyms_callback(void * data,const char * name,unsigned long addr)8645 static int kallsyms_callback(void *data, const char *name, unsigned long addr)
8646 {
8647 	struct kallsyms_data *args = data;
8648 	const char **sym;
8649 	int idx;
8650 
8651 	sym = bsearch(&name, args->syms, args->cnt, sizeof(*args->syms), symbols_cmp);
8652 	if (!sym)
8653 		return 0;
8654 
8655 	idx = sym - args->syms;
8656 	if (args->addrs[idx])
8657 		return 0;
8658 
8659 	if (!ftrace_location(addr))
8660 		return 0;
8661 
8662 	args->addrs[idx] = addr;
8663 	args->found++;
8664 	return args->found == args->cnt ? 1 : 0;
8665 }
8666 
8667 /**
8668  * ftrace_lookup_symbols - Lookup addresses for array of symbols
8669  *
8670  * @sorted_syms: array of symbols pointers symbols to resolve,
8671  * must be alphabetically sorted
8672  * @cnt: number of symbols/addresses in @syms/@addrs arrays
8673  * @addrs: array for storing resulting addresses
8674  *
8675  * This function looks up addresses for array of symbols provided in
8676  * @syms array (must be alphabetically sorted) and stores them in
8677  * @addrs array, which needs to be big enough to store at least @cnt
8678  * addresses.
8679  *
8680  * Returns: 0 if all provided symbols are found, -ESRCH otherwise.
8681  */
ftrace_lookup_symbols(const char ** sorted_syms,size_t cnt,unsigned long * addrs)8682 int ftrace_lookup_symbols(const char **sorted_syms, size_t cnt, unsigned long *addrs)
8683 {
8684 	struct kallsyms_data args;
8685 	int found_all;
8686 
8687 	memset(addrs, 0, sizeof(*addrs) * cnt);
8688 	args.addrs = addrs;
8689 	args.syms = sorted_syms;
8690 	args.cnt = cnt;
8691 	args.found = 0;
8692 
8693 	found_all = kallsyms_on_each_symbol(kallsyms_callback, &args);
8694 	if (found_all)
8695 		return 0;
8696 	found_all = module_kallsyms_on_each_symbol(NULL, kallsyms_callback, &args);
8697 	return found_all ? 0 : -ESRCH;
8698 }
8699 
8700 #ifdef CONFIG_SYSCTL
8701 
8702 #ifdef CONFIG_DYNAMIC_FTRACE
ftrace_startup_sysctl(void)8703 static void ftrace_startup_sysctl(void)
8704 {
8705 	int command;
8706 
8707 	if (unlikely(ftrace_disabled))
8708 		return;
8709 
8710 	/* Force update next time */
8711 	saved_ftrace_func = NULL;
8712 	/* ftrace_start_up is true if we want ftrace running */
8713 	if (ftrace_start_up) {
8714 		command = FTRACE_UPDATE_CALLS;
8715 		if (ftrace_graph_active)
8716 			command |= FTRACE_START_FUNC_RET;
8717 		ftrace_startup_enable(command);
8718 	}
8719 }
8720 
ftrace_shutdown_sysctl(void)8721 static void ftrace_shutdown_sysctl(void)
8722 {
8723 	int command;
8724 
8725 	if (unlikely(ftrace_disabled))
8726 		return;
8727 
8728 	/* ftrace_start_up is true if ftrace is running */
8729 	if (ftrace_start_up) {
8730 		command = FTRACE_DISABLE_CALLS;
8731 		if (ftrace_graph_active)
8732 			command |= FTRACE_STOP_FUNC_RET;
8733 		ftrace_run_update_code(command);
8734 	}
8735 }
8736 #else
8737 # define ftrace_startup_sysctl()       do { } while (0)
8738 # define ftrace_shutdown_sysctl()      do { } while (0)
8739 #endif /* CONFIG_DYNAMIC_FTRACE */
8740 
is_permanent_ops_registered(void)8741 static bool is_permanent_ops_registered(void)
8742 {
8743 	struct ftrace_ops *op;
8744 
8745 	do_for_each_ftrace_op(op, ftrace_ops_list) {
8746 		if (op->flags & FTRACE_OPS_FL_PERMANENT)
8747 			return true;
8748 	} while_for_each_ftrace_op(op);
8749 
8750 	return false;
8751 }
8752 
8753 static int
ftrace_enable_sysctl(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)8754 ftrace_enable_sysctl(const struct ctl_table *table, int write,
8755 		     void *buffer, size_t *lenp, loff_t *ppos)
8756 {
8757 	int ret;
8758 
8759 	guard(mutex)(&ftrace_lock);
8760 
8761 	if (unlikely(ftrace_disabled))
8762 		return -ENODEV;
8763 
8764 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
8765 
8766 	if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
8767 		return ret;
8768 
8769 	if (ftrace_enabled) {
8770 
8771 		/* we are starting ftrace again */
8772 		if (rcu_dereference_protected(ftrace_ops_list,
8773 			lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
8774 			update_ftrace_function();
8775 
8776 		ftrace_startup_sysctl();
8777 
8778 	} else {
8779 		if (is_permanent_ops_registered()) {
8780 			ftrace_enabled = true;
8781 			return -EBUSY;
8782 		}
8783 
8784 		/* stopping ftrace calls (just send to ftrace_stub) */
8785 		ftrace_trace_function = ftrace_stub;
8786 
8787 		ftrace_shutdown_sysctl();
8788 	}
8789 
8790 	last_ftrace_enabled = !!ftrace_enabled;
8791 	return 0;
8792 }
8793 
8794 static const struct ctl_table ftrace_sysctls[] = {
8795 	{
8796 		.procname       = "ftrace_enabled",
8797 		.data           = &ftrace_enabled,
8798 		.maxlen         = sizeof(int),
8799 		.mode           = 0644,
8800 		.proc_handler   = ftrace_enable_sysctl,
8801 	},
8802 };
8803 
ftrace_sysctl_init(void)8804 static int __init ftrace_sysctl_init(void)
8805 {
8806 	register_sysctl_init("kernel", ftrace_sysctls);
8807 	return 0;
8808 }
8809 late_initcall(ftrace_sysctl_init);
8810 #endif
8811