• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #define FTRACE_WARN_ON(cond)			\
49 	({					\
50 		int ___r = cond;		\
51 		if (WARN_ON(___r))		\
52 			ftrace_kill();		\
53 		___r;				\
54 	})
55 
56 #define FTRACE_WARN_ON_ONCE(cond)		\
57 	({					\
58 		int ___r = cond;		\
59 		if (WARN_ON_ONCE(___r))		\
60 			ftrace_kill();		\
61 		___r;				\
62 	})
63 
64 /* hash bits for specific function selection */
65 #define FTRACE_HASH_BITS 7
66 #define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
67 #define FTRACE_HASH_DEFAULT_BITS 10
68 #define FTRACE_HASH_MAX_BITS 12
69 
70 #ifdef CONFIG_DYNAMIC_FTRACE
71 #define INIT_OPS_HASH(opsname)	\
72 	.func_hash		= &opsname.local_hash,			\
73 	.local_hash.regex_lock	= __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
74 #else
75 #define INIT_OPS_HASH(opsname)
76 #endif
77 
78 enum {
79 	FTRACE_MODIFY_ENABLE_FL		= (1 << 0),
80 	FTRACE_MODIFY_MAY_SLEEP_FL	= (1 << 1),
81 };
82 
83 struct ftrace_ops ftrace_list_end __read_mostly = {
84 	.func		= ftrace_stub,
85 	.flags		= FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_STUB,
86 	INIT_OPS_HASH(ftrace_list_end)
87 };
88 
89 /* ftrace_enabled is a method to turn ftrace on or off */
90 int ftrace_enabled __read_mostly;
91 static int last_ftrace_enabled;
92 
93 /* Current function tracing op */
94 struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
95 /* What to set function_trace_op to */
96 static struct ftrace_ops *set_function_trace_op;
97 
ftrace_pids_enabled(struct ftrace_ops * ops)98 static bool ftrace_pids_enabled(struct ftrace_ops *ops)
99 {
100 	struct trace_array *tr;
101 
102 	if (!(ops->flags & FTRACE_OPS_FL_PID) || !ops->private)
103 		return false;
104 
105 	tr = ops->private;
106 
107 	return tr->function_pids != NULL;
108 }
109 
110 static void ftrace_update_trampoline(struct ftrace_ops *ops);
111 
112 /*
113  * ftrace_disabled is set when an anomaly is discovered.
114  * ftrace_disabled is much stronger than ftrace_enabled.
115  */
116 static int ftrace_disabled __read_mostly;
117 
118 DEFINE_MUTEX(ftrace_lock);
119 
120 struct ftrace_ops __rcu *ftrace_ops_list __read_mostly = &ftrace_list_end;
121 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
122 struct ftrace_ops global_ops;
123 
124 #if ARCH_SUPPORTS_FTRACE_OPS
125 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
126 				 struct ftrace_ops *op, struct pt_regs *regs);
127 #else
128 /* See comment below, where ftrace_ops_list_func is defined */
129 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
130 #define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
131 #endif
132 
ftrace_ops_init(struct ftrace_ops * ops)133 static inline void ftrace_ops_init(struct ftrace_ops *ops)
134 {
135 #ifdef CONFIG_DYNAMIC_FTRACE
136 	if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
137 		mutex_init(&ops->local_hash.regex_lock);
138 		ops->func_hash = &ops->local_hash;
139 		ops->flags |= FTRACE_OPS_FL_INITIALIZED;
140 	}
141 #endif
142 }
143 
ftrace_pid_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct pt_regs * regs)144 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
145 			    struct ftrace_ops *op, struct pt_regs *regs)
146 {
147 	struct trace_array *tr = op->private;
148 
149 	if (tr && this_cpu_read(tr->trace_buffer.data->ftrace_ignore_pid))
150 		return;
151 
152 	op->saved_func(ip, parent_ip, op, regs);
153 }
154 
ftrace_sync(struct work_struct * work)155 static void ftrace_sync(struct work_struct *work)
156 {
157 	/*
158 	 * This function is just a stub to implement a hard force
159 	 * of synchronize_rcu(). This requires synchronizing
160 	 * tasks even in userspace and idle.
161 	 *
162 	 * Yes, function tracing is rude.
163 	 */
164 }
165 
ftrace_sync_ipi(void * data)166 static void ftrace_sync_ipi(void *data)
167 {
168 	/* Probably not needed, but do it anyway */
169 	smp_rmb();
170 }
171 
ftrace_ops_get_list_func(struct ftrace_ops * ops)172 static ftrace_func_t ftrace_ops_get_list_func(struct ftrace_ops *ops)
173 {
174 	/*
175 	 * If this is a dynamic, RCU, or per CPU ops, or we force list func,
176 	 * then it needs to call the list anyway.
177 	 */
178 	if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_RCU) ||
179 	    FTRACE_FORCE_LIST_FUNC)
180 		return ftrace_ops_list_func;
181 
182 	return ftrace_ops_get_func(ops);
183 }
184 
update_ftrace_function(void)185 static void update_ftrace_function(void)
186 {
187 	ftrace_func_t func;
188 
189 	/*
190 	 * Prepare the ftrace_ops that the arch callback will use.
191 	 * If there's only one ftrace_ops registered, the ftrace_ops_list
192 	 * will point to the ops we want.
193 	 */
194 	set_function_trace_op = rcu_dereference_protected(ftrace_ops_list,
195 						lockdep_is_held(&ftrace_lock));
196 
197 	/* If there's no ftrace_ops registered, just call the stub function */
198 	if (set_function_trace_op == &ftrace_list_end) {
199 		func = ftrace_stub;
200 
201 	/*
202 	 * If we are at the end of the list and this ops is
203 	 * recursion safe and not dynamic and the arch supports passing ops,
204 	 * then have the mcount trampoline call the function directly.
205 	 */
206 	} else if (rcu_dereference_protected(ftrace_ops_list->next,
207 			lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
208 		func = ftrace_ops_get_list_func(ftrace_ops_list);
209 
210 	} else {
211 		/* Just use the default ftrace_ops */
212 		set_function_trace_op = &ftrace_list_end;
213 		func = ftrace_ops_list_func;
214 	}
215 
216 	update_function_graph_func();
217 
218 	/* If there's no change, then do nothing more here */
219 	if (ftrace_trace_function == func)
220 		return;
221 
222 	/*
223 	 * If we are using the list function, it doesn't care
224 	 * about the function_trace_ops.
225 	 */
226 	if (func == ftrace_ops_list_func) {
227 		ftrace_trace_function = func;
228 		/*
229 		 * Don't even bother setting function_trace_ops,
230 		 * it would be racy to do so anyway.
231 		 */
232 		return;
233 	}
234 
235 #ifndef CONFIG_DYNAMIC_FTRACE
236 	/*
237 	 * For static tracing, we need to be a bit more careful.
238 	 * The function change takes affect immediately. Thus,
239 	 * we need to coorditate the setting of the function_trace_ops
240 	 * with the setting of the ftrace_trace_function.
241 	 *
242 	 * Set the function to the list ops, which will call the
243 	 * function we want, albeit indirectly, but it handles the
244 	 * ftrace_ops and doesn't depend on function_trace_op.
245 	 */
246 	ftrace_trace_function = ftrace_ops_list_func;
247 	/*
248 	 * Make sure all CPUs see this. Yes this is slow, but static
249 	 * tracing is slow and nasty to have enabled.
250 	 */
251 	schedule_on_each_cpu(ftrace_sync);
252 	/* Now all cpus are using the list ops. */
253 	function_trace_op = set_function_trace_op;
254 	/* Make sure the function_trace_op is visible on all CPUs */
255 	smp_wmb();
256 	/* Nasty way to force a rmb on all cpus */
257 	smp_call_function(ftrace_sync_ipi, NULL, 1);
258 	/* OK, we are all set to update the ftrace_trace_function now! */
259 #endif /* !CONFIG_DYNAMIC_FTRACE */
260 
261 	ftrace_trace_function = func;
262 }
263 
add_ftrace_ops(struct ftrace_ops __rcu ** list,struct ftrace_ops * ops)264 static void add_ftrace_ops(struct ftrace_ops __rcu **list,
265 			   struct ftrace_ops *ops)
266 {
267 	rcu_assign_pointer(ops->next, *list);
268 
269 	/*
270 	 * We are entering ops into the list but another
271 	 * CPU might be walking that list. We need to make sure
272 	 * the ops->next pointer is valid before another CPU sees
273 	 * the ops pointer included into the list.
274 	 */
275 	rcu_assign_pointer(*list, ops);
276 }
277 
remove_ftrace_ops(struct ftrace_ops __rcu ** list,struct ftrace_ops * ops)278 static int remove_ftrace_ops(struct ftrace_ops __rcu **list,
279 			     struct ftrace_ops *ops)
280 {
281 	struct ftrace_ops **p;
282 
283 	/*
284 	 * If we are removing the last function, then simply point
285 	 * to the ftrace_stub.
286 	 */
287 	if (rcu_dereference_protected(*list,
288 			lockdep_is_held(&ftrace_lock)) == ops &&
289 	    rcu_dereference_protected(ops->next,
290 			lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
291 		*list = &ftrace_list_end;
292 		return 0;
293 	}
294 
295 	for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
296 		if (*p == ops)
297 			break;
298 
299 	if (*p != ops)
300 		return -1;
301 
302 	*p = (*p)->next;
303 	return 0;
304 }
305 
306 static void ftrace_update_trampoline(struct ftrace_ops *ops);
307 
__register_ftrace_function(struct ftrace_ops * ops)308 int __register_ftrace_function(struct ftrace_ops *ops)
309 {
310 	if (ops->flags & FTRACE_OPS_FL_DELETED)
311 		return -EINVAL;
312 
313 	if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
314 		return -EBUSY;
315 
316 #ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
317 	/*
318 	 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
319 	 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
320 	 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
321 	 */
322 	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
323 	    !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
324 		return -EINVAL;
325 
326 	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
327 		ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
328 #endif
329 
330 	if (!core_kernel_data((unsigned long)ops))
331 		ops->flags |= FTRACE_OPS_FL_DYNAMIC;
332 
333 	add_ftrace_ops(&ftrace_ops_list, ops);
334 
335 	/* Always save the function, and reset at unregistering */
336 	ops->saved_func = ops->func;
337 
338 	if (ftrace_pids_enabled(ops))
339 		ops->func = ftrace_pid_func;
340 
341 	ftrace_update_trampoline(ops);
342 
343 	if (ftrace_enabled)
344 		update_ftrace_function();
345 
346 	return 0;
347 }
348 
__unregister_ftrace_function(struct ftrace_ops * ops)349 int __unregister_ftrace_function(struct ftrace_ops *ops)
350 {
351 	int ret;
352 
353 	if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
354 		return -EBUSY;
355 
356 	ret = remove_ftrace_ops(&ftrace_ops_list, ops);
357 
358 	if (ret < 0)
359 		return ret;
360 
361 	if (ftrace_enabled)
362 		update_ftrace_function();
363 
364 	ops->func = ops->saved_func;
365 
366 	return 0;
367 }
368 
ftrace_update_pid_func(void)369 static void ftrace_update_pid_func(void)
370 {
371 	struct ftrace_ops *op;
372 
373 	/* Only do something if we are tracing something */
374 	if (ftrace_trace_function == ftrace_stub)
375 		return;
376 
377 	do_for_each_ftrace_op(op, ftrace_ops_list) {
378 		if (op->flags & FTRACE_OPS_FL_PID) {
379 			op->func = ftrace_pids_enabled(op) ?
380 				ftrace_pid_func : op->saved_func;
381 			ftrace_update_trampoline(op);
382 		}
383 	} while_for_each_ftrace_op(op);
384 
385 	update_ftrace_function();
386 }
387 
388 #ifdef CONFIG_FUNCTION_PROFILER
389 struct ftrace_profile {
390 	struct hlist_node		node;
391 	unsigned long			ip;
392 	unsigned long			counter;
393 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
394 	unsigned long long		time;
395 	unsigned long long		time_squared;
396 #endif
397 };
398 
399 struct ftrace_profile_page {
400 	struct ftrace_profile_page	*next;
401 	unsigned long			index;
402 	struct ftrace_profile		records[];
403 };
404 
405 struct ftrace_profile_stat {
406 	atomic_t			disabled;
407 	struct hlist_head		*hash;
408 	struct ftrace_profile_page	*pages;
409 	struct ftrace_profile_page	*start;
410 	struct tracer_stat		stat;
411 };
412 
413 #define PROFILE_RECORDS_SIZE						\
414 	(PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
415 
416 #define PROFILES_PER_PAGE					\
417 	(PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
418 
419 static int ftrace_profile_enabled __read_mostly;
420 
421 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
422 static DEFINE_MUTEX(ftrace_profile_lock);
423 
424 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
425 
426 #define FTRACE_PROFILE_HASH_BITS 10
427 #define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
428 
429 static void *
function_stat_next(void * v,int idx)430 function_stat_next(void *v, int idx)
431 {
432 	struct ftrace_profile *rec = v;
433 	struct ftrace_profile_page *pg;
434 
435 	pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
436 
437  again:
438 	if (idx != 0)
439 		rec++;
440 
441 	if ((void *)rec >= (void *)&pg->records[pg->index]) {
442 		pg = pg->next;
443 		if (!pg)
444 			return NULL;
445 		rec = &pg->records[0];
446 		if (!rec->counter)
447 			goto again;
448 	}
449 
450 	return rec;
451 }
452 
function_stat_start(struct tracer_stat * trace)453 static void *function_stat_start(struct tracer_stat *trace)
454 {
455 	struct ftrace_profile_stat *stat =
456 		container_of(trace, struct ftrace_profile_stat, stat);
457 
458 	if (!stat || !stat->start)
459 		return NULL;
460 
461 	return function_stat_next(&stat->start->records[0], 0);
462 }
463 
464 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
465 /* function graph compares on total time */
function_stat_cmp(void * p1,void * p2)466 static int function_stat_cmp(void *p1, void *p2)
467 {
468 	struct ftrace_profile *a = p1;
469 	struct ftrace_profile *b = p2;
470 
471 	if (a->time < b->time)
472 		return -1;
473 	if (a->time > b->time)
474 		return 1;
475 	else
476 		return 0;
477 }
478 #else
479 /* not function graph compares against hits */
function_stat_cmp(void * p1,void * p2)480 static int function_stat_cmp(void *p1, void *p2)
481 {
482 	struct ftrace_profile *a = p1;
483 	struct ftrace_profile *b = p2;
484 
485 	if (a->counter < b->counter)
486 		return -1;
487 	if (a->counter > b->counter)
488 		return 1;
489 	else
490 		return 0;
491 }
492 #endif
493 
function_stat_headers(struct seq_file * m)494 static int function_stat_headers(struct seq_file *m)
495 {
496 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
497 	seq_puts(m, "  Function                               "
498 		 "Hit    Time            Avg             s^2\n"
499 		    "  --------                               "
500 		 "---    ----            ---             ---\n");
501 #else
502 	seq_puts(m, "  Function                               Hit\n"
503 		    "  --------                               ---\n");
504 #endif
505 	return 0;
506 }
507 
function_stat_show(struct seq_file * m,void * v)508 static int function_stat_show(struct seq_file *m, void *v)
509 {
510 	struct ftrace_profile *rec = v;
511 	char str[KSYM_SYMBOL_LEN];
512 	int ret = 0;
513 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
514 	static struct trace_seq s;
515 	unsigned long long avg;
516 	unsigned long long stddev;
517 #endif
518 	mutex_lock(&ftrace_profile_lock);
519 
520 	/* we raced with function_profile_reset() */
521 	if (unlikely(rec->counter == 0)) {
522 		ret = -EBUSY;
523 		goto out;
524 	}
525 
526 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
527 	avg = div64_ul(rec->time, rec->counter);
528 	if (tracing_thresh && (avg < tracing_thresh))
529 		goto out;
530 #endif
531 
532 	kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
533 	seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
534 
535 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
536 	seq_puts(m, "    ");
537 
538 	/* Sample standard deviation (s^2) */
539 	if (rec->counter <= 1)
540 		stddev = 0;
541 	else {
542 		/*
543 		 * Apply Welford's method:
544 		 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
545 		 */
546 		stddev = rec->counter * rec->time_squared -
547 			 rec->time * rec->time;
548 
549 		/*
550 		 * Divide only 1000 for ns^2 -> us^2 conversion.
551 		 * trace_print_graph_duration will divide 1000 again.
552 		 */
553 		stddev = div64_ul(stddev,
554 				  rec->counter * (rec->counter - 1) * 1000);
555 	}
556 
557 	trace_seq_init(&s);
558 	trace_print_graph_duration(rec->time, &s);
559 	trace_seq_puts(&s, "    ");
560 	trace_print_graph_duration(avg, &s);
561 	trace_seq_puts(&s, "    ");
562 	trace_print_graph_duration(stddev, &s);
563 	trace_print_seq(m, &s);
564 #endif
565 	seq_putc(m, '\n');
566 out:
567 	mutex_unlock(&ftrace_profile_lock);
568 
569 	return ret;
570 }
571 
ftrace_profile_reset(struct ftrace_profile_stat * stat)572 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
573 {
574 	struct ftrace_profile_page *pg;
575 
576 	pg = stat->pages = stat->start;
577 
578 	while (pg) {
579 		memset(pg->records, 0, PROFILE_RECORDS_SIZE);
580 		pg->index = 0;
581 		pg = pg->next;
582 	}
583 
584 	memset(stat->hash, 0,
585 	       FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
586 }
587 
ftrace_profile_pages_init(struct ftrace_profile_stat * stat)588 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
589 {
590 	struct ftrace_profile_page *pg;
591 	int functions;
592 	int pages;
593 	int i;
594 
595 	/* If we already allocated, do nothing */
596 	if (stat->pages)
597 		return 0;
598 
599 	stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
600 	if (!stat->pages)
601 		return -ENOMEM;
602 
603 #ifdef CONFIG_DYNAMIC_FTRACE
604 	functions = ftrace_update_tot_cnt;
605 #else
606 	/*
607 	 * We do not know the number of functions that exist because
608 	 * dynamic tracing is what counts them. With past experience
609 	 * we have around 20K functions. That should be more than enough.
610 	 * It is highly unlikely we will execute every function in
611 	 * the kernel.
612 	 */
613 	functions = 20000;
614 #endif
615 
616 	pg = stat->start = stat->pages;
617 
618 	pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
619 
620 	for (i = 1; i < pages; i++) {
621 		pg->next = (void *)get_zeroed_page(GFP_KERNEL);
622 		if (!pg->next)
623 			goto out_free;
624 		pg = pg->next;
625 	}
626 
627 	return 0;
628 
629  out_free:
630 	pg = stat->start;
631 	while (pg) {
632 		unsigned long tmp = (unsigned long)pg;
633 
634 		pg = pg->next;
635 		free_page(tmp);
636 	}
637 
638 	stat->pages = NULL;
639 	stat->start = NULL;
640 
641 	return -ENOMEM;
642 }
643 
ftrace_profile_init_cpu(int cpu)644 static int ftrace_profile_init_cpu(int cpu)
645 {
646 	struct ftrace_profile_stat *stat;
647 	int size;
648 
649 	stat = &per_cpu(ftrace_profile_stats, cpu);
650 
651 	if (stat->hash) {
652 		/* If the profile is already created, simply reset it */
653 		ftrace_profile_reset(stat);
654 		return 0;
655 	}
656 
657 	/*
658 	 * We are profiling all functions, but usually only a few thousand
659 	 * functions are hit. We'll make a hash of 1024 items.
660 	 */
661 	size = FTRACE_PROFILE_HASH_SIZE;
662 
663 	stat->hash = kcalloc(size, sizeof(struct hlist_head), GFP_KERNEL);
664 
665 	if (!stat->hash)
666 		return -ENOMEM;
667 
668 	/* Preallocate the function profiling pages */
669 	if (ftrace_profile_pages_init(stat) < 0) {
670 		kfree(stat->hash);
671 		stat->hash = NULL;
672 		return -ENOMEM;
673 	}
674 
675 	return 0;
676 }
677 
ftrace_profile_init(void)678 static int ftrace_profile_init(void)
679 {
680 	int cpu;
681 	int ret = 0;
682 
683 	for_each_possible_cpu(cpu) {
684 		ret = ftrace_profile_init_cpu(cpu);
685 		if (ret)
686 			break;
687 	}
688 
689 	return ret;
690 }
691 
692 /* interrupts must be disabled */
693 static struct ftrace_profile *
ftrace_find_profiled_func(struct ftrace_profile_stat * stat,unsigned long ip)694 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
695 {
696 	struct ftrace_profile *rec;
697 	struct hlist_head *hhd;
698 	unsigned long key;
699 
700 	key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
701 	hhd = &stat->hash[key];
702 
703 	if (hlist_empty(hhd))
704 		return NULL;
705 
706 	hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
707 		if (rec->ip == ip)
708 			return rec;
709 	}
710 
711 	return NULL;
712 }
713 
ftrace_add_profile(struct ftrace_profile_stat * stat,struct ftrace_profile * rec)714 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
715 			       struct ftrace_profile *rec)
716 {
717 	unsigned long key;
718 
719 	key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
720 	hlist_add_head_rcu(&rec->node, &stat->hash[key]);
721 }
722 
723 /*
724  * The memory is already allocated, this simply finds a new record to use.
725  */
726 static struct ftrace_profile *
ftrace_profile_alloc(struct ftrace_profile_stat * stat,unsigned long ip)727 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
728 {
729 	struct ftrace_profile *rec = NULL;
730 
731 	/* prevent recursion (from NMIs) */
732 	if (atomic_inc_return(&stat->disabled) != 1)
733 		goto out;
734 
735 	/*
736 	 * Try to find the function again since an NMI
737 	 * could have added it
738 	 */
739 	rec = ftrace_find_profiled_func(stat, ip);
740 	if (rec)
741 		goto out;
742 
743 	if (stat->pages->index == PROFILES_PER_PAGE) {
744 		if (!stat->pages->next)
745 			goto out;
746 		stat->pages = stat->pages->next;
747 	}
748 
749 	rec = &stat->pages->records[stat->pages->index++];
750 	rec->ip = ip;
751 	ftrace_add_profile(stat, rec);
752 
753  out:
754 	atomic_dec(&stat->disabled);
755 
756 	return rec;
757 }
758 
759 static void
function_profile_call(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * ops,struct pt_regs * regs)760 function_profile_call(unsigned long ip, unsigned long parent_ip,
761 		      struct ftrace_ops *ops, struct pt_regs *regs)
762 {
763 	struct ftrace_profile_stat *stat;
764 	struct ftrace_profile *rec;
765 	unsigned long flags;
766 
767 	if (!ftrace_profile_enabled)
768 		return;
769 
770 	local_irq_save(flags);
771 
772 	stat = this_cpu_ptr(&ftrace_profile_stats);
773 	if (!stat->hash || !ftrace_profile_enabled)
774 		goto out;
775 
776 	rec = ftrace_find_profiled_func(stat, ip);
777 	if (!rec) {
778 		rec = ftrace_profile_alloc(stat, ip);
779 		if (!rec)
780 			goto out;
781 	}
782 
783 	rec->counter++;
784  out:
785 	local_irq_restore(flags);
786 }
787 
788 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
789 static bool fgraph_graph_time = true;
790 
ftrace_graph_graph_time_control(bool enable)791 void ftrace_graph_graph_time_control(bool enable)
792 {
793 	fgraph_graph_time = enable;
794 }
795 
profile_graph_entry(struct ftrace_graph_ent * trace)796 static int profile_graph_entry(struct ftrace_graph_ent *trace)
797 {
798 	struct ftrace_ret_stack *ret_stack;
799 
800 	function_profile_call(trace->func, 0, NULL, NULL);
801 
802 	/* If function graph is shutting down, ret_stack can be NULL */
803 	if (!current->ret_stack)
804 		return 0;
805 
806 	ret_stack = ftrace_graph_get_ret_stack(current, 0);
807 	if (ret_stack)
808 		ret_stack->subtime = 0;
809 
810 	return 1;
811 }
812 
profile_graph_return(struct ftrace_graph_ret * trace)813 static void profile_graph_return(struct ftrace_graph_ret *trace)
814 {
815 	struct ftrace_ret_stack *ret_stack;
816 	struct ftrace_profile_stat *stat;
817 	unsigned long long calltime;
818 	struct ftrace_profile *rec;
819 	unsigned long flags;
820 
821 	local_irq_save(flags);
822 	stat = this_cpu_ptr(&ftrace_profile_stats);
823 	if (!stat->hash || !ftrace_profile_enabled)
824 		goto out;
825 
826 	/* If the calltime was zero'd ignore it */
827 	if (!trace->calltime)
828 		goto out;
829 
830 	calltime = trace->rettime - trace->calltime;
831 
832 	if (!fgraph_graph_time) {
833 
834 		/* Append this call time to the parent time to subtract */
835 		ret_stack = ftrace_graph_get_ret_stack(current, 1);
836 		if (ret_stack)
837 			ret_stack->subtime += calltime;
838 
839 		ret_stack = ftrace_graph_get_ret_stack(current, 0);
840 		if (ret_stack && ret_stack->subtime < calltime)
841 			calltime -= ret_stack->subtime;
842 		else
843 			calltime = 0;
844 	}
845 
846 	rec = ftrace_find_profiled_func(stat, trace->func);
847 	if (rec) {
848 		rec->time += calltime;
849 		rec->time_squared += calltime * calltime;
850 	}
851 
852  out:
853 	local_irq_restore(flags);
854 }
855 
856 static struct fgraph_ops fprofiler_ops = {
857 	.entryfunc = &profile_graph_entry,
858 	.retfunc = &profile_graph_return,
859 };
860 
register_ftrace_profiler(void)861 static int register_ftrace_profiler(void)
862 {
863 	return register_ftrace_graph(&fprofiler_ops);
864 }
865 
unregister_ftrace_profiler(void)866 static void unregister_ftrace_profiler(void)
867 {
868 	unregister_ftrace_graph(&fprofiler_ops);
869 }
870 #else
871 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
872 	.func		= function_profile_call,
873 	.flags		= FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
874 	INIT_OPS_HASH(ftrace_profile_ops)
875 };
876 
register_ftrace_profiler(void)877 static int register_ftrace_profiler(void)
878 {
879 	return register_ftrace_function(&ftrace_profile_ops);
880 }
881 
unregister_ftrace_profiler(void)882 static void unregister_ftrace_profiler(void)
883 {
884 	unregister_ftrace_function(&ftrace_profile_ops);
885 }
886 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
887 
888 static ssize_t
ftrace_profile_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)889 ftrace_profile_write(struct file *filp, const char __user *ubuf,
890 		     size_t cnt, loff_t *ppos)
891 {
892 	unsigned long val;
893 	int ret;
894 
895 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
896 	if (ret)
897 		return ret;
898 
899 	val = !!val;
900 
901 	mutex_lock(&ftrace_profile_lock);
902 	if (ftrace_profile_enabled ^ val) {
903 		if (val) {
904 			ret = ftrace_profile_init();
905 			if (ret < 0) {
906 				cnt = ret;
907 				goto out;
908 			}
909 
910 			ret = register_ftrace_profiler();
911 			if (ret < 0) {
912 				cnt = ret;
913 				goto out;
914 			}
915 			ftrace_profile_enabled = 1;
916 		} else {
917 			ftrace_profile_enabled = 0;
918 			/*
919 			 * unregister_ftrace_profiler calls stop_machine
920 			 * so this acts like an synchronize_rcu.
921 			 */
922 			unregister_ftrace_profiler();
923 		}
924 	}
925  out:
926 	mutex_unlock(&ftrace_profile_lock);
927 
928 	*ppos += cnt;
929 
930 	return cnt;
931 }
932 
933 static ssize_t
ftrace_profile_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)934 ftrace_profile_read(struct file *filp, char __user *ubuf,
935 		     size_t cnt, loff_t *ppos)
936 {
937 	char buf[64];		/* big enough to hold a number */
938 	int r;
939 
940 	r = sprintf(buf, "%u\n", ftrace_profile_enabled);
941 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
942 }
943 
944 static const struct file_operations ftrace_profile_fops = {
945 	.open		= tracing_open_generic,
946 	.read		= ftrace_profile_read,
947 	.write		= ftrace_profile_write,
948 	.llseek		= default_llseek,
949 };
950 
951 /* used to initialize the real stat files */
952 static struct tracer_stat function_stats __initdata = {
953 	.name		= "functions",
954 	.stat_start	= function_stat_start,
955 	.stat_next	= function_stat_next,
956 	.stat_cmp	= function_stat_cmp,
957 	.stat_headers	= function_stat_headers,
958 	.stat_show	= function_stat_show
959 };
960 
ftrace_profile_tracefs(struct dentry * d_tracer)961 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
962 {
963 	struct ftrace_profile_stat *stat;
964 	struct dentry *entry;
965 	char *name;
966 	int ret;
967 	int cpu;
968 
969 	for_each_possible_cpu(cpu) {
970 		stat = &per_cpu(ftrace_profile_stats, cpu);
971 
972 		name = kasprintf(GFP_KERNEL, "function%d", cpu);
973 		if (!name) {
974 			/*
975 			 * The files created are permanent, if something happens
976 			 * we still do not free memory.
977 			 */
978 			WARN(1,
979 			     "Could not allocate stat file for cpu %d\n",
980 			     cpu);
981 			return;
982 		}
983 		stat->stat = function_stats;
984 		stat->stat.name = name;
985 		ret = register_stat_tracer(&stat->stat);
986 		if (ret) {
987 			WARN(1,
988 			     "Could not register function stat for cpu %d\n",
989 			     cpu);
990 			kfree(name);
991 			return;
992 		}
993 	}
994 
995 	entry = tracefs_create_file("function_profile_enabled", 0644,
996 				    d_tracer, NULL, &ftrace_profile_fops);
997 	if (!entry)
998 		pr_warn("Could not create tracefs 'function_profile_enabled' entry\n");
999 }
1000 
1001 #else /* CONFIG_FUNCTION_PROFILER */
ftrace_profile_tracefs(struct dentry * d_tracer)1002 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1003 {
1004 }
1005 #endif /* CONFIG_FUNCTION_PROFILER */
1006 
1007 #ifdef CONFIG_DYNAMIC_FTRACE
1008 
1009 static struct ftrace_ops *removed_ops;
1010 
1011 /*
1012  * Set when doing a global update, like enabling all recs or disabling them.
1013  * It is not set when just updating a single ftrace_ops.
1014  */
1015 static bool update_all_ops;
1016 
1017 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1018 # error Dynamic ftrace depends on MCOUNT_RECORD
1019 #endif
1020 
1021 struct ftrace_func_entry {
1022 	struct hlist_node hlist;
1023 	unsigned long ip;
1024 };
1025 
1026 struct ftrace_func_probe {
1027 	struct ftrace_probe_ops	*probe_ops;
1028 	struct ftrace_ops	ops;
1029 	struct trace_array	*tr;
1030 	struct list_head	list;
1031 	void			*data;
1032 	int			ref;
1033 };
1034 
1035 /*
1036  * We make these constant because no one should touch them,
1037  * but they are used as the default "empty hash", to avoid allocating
1038  * it all the time. These are in a read only section such that if
1039  * anyone does try to modify it, it will cause an exception.
1040  */
1041 static const struct hlist_head empty_buckets[1];
1042 static const struct ftrace_hash empty_hash = {
1043 	.buckets = (struct hlist_head *)empty_buckets,
1044 };
1045 #define EMPTY_HASH	((struct ftrace_hash *)&empty_hash)
1046 
1047 struct ftrace_ops global_ops = {
1048 	.func				= ftrace_stub,
1049 	.local_hash.notrace_hash	= EMPTY_HASH,
1050 	.local_hash.filter_hash		= EMPTY_HASH,
1051 	INIT_OPS_HASH(global_ops)
1052 	.flags				= FTRACE_OPS_FL_RECURSION_SAFE |
1053 					  FTRACE_OPS_FL_INITIALIZED |
1054 					  FTRACE_OPS_FL_PID,
1055 };
1056 
1057 /*
1058  * Used by the stack undwinder to know about dynamic ftrace trampolines.
1059  */
ftrace_ops_trampoline(unsigned long addr)1060 struct ftrace_ops *ftrace_ops_trampoline(unsigned long addr)
1061 {
1062 	struct ftrace_ops *op = NULL;
1063 
1064 	/*
1065 	 * Some of the ops may be dynamically allocated,
1066 	 * they are freed after a synchronize_rcu().
1067 	 */
1068 	preempt_disable_notrace();
1069 
1070 	do_for_each_ftrace_op(op, ftrace_ops_list) {
1071 		/*
1072 		 * This is to check for dynamically allocated trampolines.
1073 		 * Trampolines that are in kernel text will have
1074 		 * core_kernel_text() return true.
1075 		 */
1076 		if (op->trampoline && op->trampoline_size)
1077 			if (addr >= op->trampoline &&
1078 			    addr < op->trampoline + op->trampoline_size) {
1079 				preempt_enable_notrace();
1080 				return op;
1081 			}
1082 	} while_for_each_ftrace_op(op);
1083 	preempt_enable_notrace();
1084 
1085 	return NULL;
1086 }
1087 
1088 /*
1089  * This is used by __kernel_text_address() to return true if the
1090  * address is on a dynamically allocated trampoline that would
1091  * not return true for either core_kernel_text() or
1092  * is_module_text_address().
1093  */
is_ftrace_trampoline(unsigned long addr)1094 bool is_ftrace_trampoline(unsigned long addr)
1095 {
1096 	return ftrace_ops_trampoline(addr) != NULL;
1097 }
1098 
1099 struct ftrace_page {
1100 	struct ftrace_page	*next;
1101 	struct dyn_ftrace	*records;
1102 	int			index;
1103 	int			size;
1104 };
1105 
1106 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1107 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1108 
1109 /* estimate from running different kernels */
1110 #define NR_TO_INIT		10000
1111 
1112 static struct ftrace_page	*ftrace_pages_start;
1113 static struct ftrace_page	*ftrace_pages;
1114 
1115 static __always_inline unsigned long
ftrace_hash_key(struct ftrace_hash * hash,unsigned long ip)1116 ftrace_hash_key(struct ftrace_hash *hash, unsigned long ip)
1117 {
1118 	if (hash->size_bits > 0)
1119 		return hash_long(ip, hash->size_bits);
1120 
1121 	return 0;
1122 }
1123 
1124 /* Only use this function if ftrace_hash_empty() has already been tested */
1125 static __always_inline struct ftrace_func_entry *
__ftrace_lookup_ip(struct ftrace_hash * hash,unsigned long ip)1126 __ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1127 {
1128 	unsigned long key;
1129 	struct ftrace_func_entry *entry;
1130 	struct hlist_head *hhd;
1131 
1132 	key = ftrace_hash_key(hash, ip);
1133 	hhd = &hash->buckets[key];
1134 
1135 	hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1136 		if (entry->ip == ip)
1137 			return entry;
1138 	}
1139 	return NULL;
1140 }
1141 
1142 /**
1143  * ftrace_lookup_ip - Test to see if an ip exists in an ftrace_hash
1144  * @hash: The hash to look at
1145  * @ip: The instruction pointer to test
1146  *
1147  * Search a given @hash to see if a given instruction pointer (@ip)
1148  * exists in it.
1149  *
1150  * Returns the entry that holds the @ip if found. NULL otherwise.
1151  */
1152 struct ftrace_func_entry *
ftrace_lookup_ip(struct ftrace_hash * hash,unsigned long ip)1153 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1154 {
1155 	if (ftrace_hash_empty(hash))
1156 		return NULL;
1157 
1158 	return __ftrace_lookup_ip(hash, ip);
1159 }
1160 
__add_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1161 static void __add_hash_entry(struct ftrace_hash *hash,
1162 			     struct ftrace_func_entry *entry)
1163 {
1164 	struct hlist_head *hhd;
1165 	unsigned long key;
1166 
1167 	key = ftrace_hash_key(hash, entry->ip);
1168 	hhd = &hash->buckets[key];
1169 	hlist_add_head(&entry->hlist, hhd);
1170 	hash->count++;
1171 }
1172 
add_hash_entry(struct ftrace_hash * hash,unsigned long ip)1173 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1174 {
1175 	struct ftrace_func_entry *entry;
1176 
1177 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1178 	if (!entry)
1179 		return -ENOMEM;
1180 
1181 	entry->ip = ip;
1182 	__add_hash_entry(hash, entry);
1183 
1184 	return 0;
1185 }
1186 
1187 static void
free_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1188 free_hash_entry(struct ftrace_hash *hash,
1189 		  struct ftrace_func_entry *entry)
1190 {
1191 	hlist_del(&entry->hlist);
1192 	kfree(entry);
1193 	hash->count--;
1194 }
1195 
1196 static void
remove_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1197 remove_hash_entry(struct ftrace_hash *hash,
1198 		  struct ftrace_func_entry *entry)
1199 {
1200 	hlist_del_rcu(&entry->hlist);
1201 	hash->count--;
1202 }
1203 
ftrace_hash_clear(struct ftrace_hash * hash)1204 static void ftrace_hash_clear(struct ftrace_hash *hash)
1205 {
1206 	struct hlist_head *hhd;
1207 	struct hlist_node *tn;
1208 	struct ftrace_func_entry *entry;
1209 	int size = 1 << hash->size_bits;
1210 	int i;
1211 
1212 	if (!hash->count)
1213 		return;
1214 
1215 	for (i = 0; i < size; i++) {
1216 		hhd = &hash->buckets[i];
1217 		hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1218 			free_hash_entry(hash, entry);
1219 	}
1220 	FTRACE_WARN_ON(hash->count);
1221 }
1222 
free_ftrace_mod(struct ftrace_mod_load * ftrace_mod)1223 static void free_ftrace_mod(struct ftrace_mod_load *ftrace_mod)
1224 {
1225 	list_del(&ftrace_mod->list);
1226 	kfree(ftrace_mod->module);
1227 	kfree(ftrace_mod->func);
1228 	kfree(ftrace_mod);
1229 }
1230 
clear_ftrace_mod_list(struct list_head * head)1231 static void clear_ftrace_mod_list(struct list_head *head)
1232 {
1233 	struct ftrace_mod_load *p, *n;
1234 
1235 	/* stack tracer isn't supported yet */
1236 	if (!head)
1237 		return;
1238 
1239 	mutex_lock(&ftrace_lock);
1240 	list_for_each_entry_safe(p, n, head, list)
1241 		free_ftrace_mod(p);
1242 	mutex_unlock(&ftrace_lock);
1243 }
1244 
free_ftrace_hash(struct ftrace_hash * hash)1245 static void free_ftrace_hash(struct ftrace_hash *hash)
1246 {
1247 	if (!hash || hash == EMPTY_HASH)
1248 		return;
1249 	ftrace_hash_clear(hash);
1250 	kfree(hash->buckets);
1251 	kfree(hash);
1252 }
1253 
__free_ftrace_hash_rcu(struct rcu_head * rcu)1254 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1255 {
1256 	struct ftrace_hash *hash;
1257 
1258 	hash = container_of(rcu, struct ftrace_hash, rcu);
1259 	free_ftrace_hash(hash);
1260 }
1261 
free_ftrace_hash_rcu(struct ftrace_hash * hash)1262 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1263 {
1264 	if (!hash || hash == EMPTY_HASH)
1265 		return;
1266 	call_rcu(&hash->rcu, __free_ftrace_hash_rcu);
1267 }
1268 
ftrace_free_filter(struct ftrace_ops * ops)1269 void ftrace_free_filter(struct ftrace_ops *ops)
1270 {
1271 	ftrace_ops_init(ops);
1272 	free_ftrace_hash(ops->func_hash->filter_hash);
1273 	free_ftrace_hash(ops->func_hash->notrace_hash);
1274 }
1275 
alloc_ftrace_hash(int size_bits)1276 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1277 {
1278 	struct ftrace_hash *hash;
1279 	int size;
1280 
1281 	hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1282 	if (!hash)
1283 		return NULL;
1284 
1285 	size = 1 << size_bits;
1286 	hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1287 
1288 	if (!hash->buckets) {
1289 		kfree(hash);
1290 		return NULL;
1291 	}
1292 
1293 	hash->size_bits = size_bits;
1294 
1295 	return hash;
1296 }
1297 
1298 
ftrace_add_mod(struct trace_array * tr,const char * func,const char * module,int enable)1299 static int ftrace_add_mod(struct trace_array *tr,
1300 			  const char *func, const char *module,
1301 			  int enable)
1302 {
1303 	struct ftrace_mod_load *ftrace_mod;
1304 	struct list_head *mod_head = enable ? &tr->mod_trace : &tr->mod_notrace;
1305 
1306 	ftrace_mod = kzalloc(sizeof(*ftrace_mod), GFP_KERNEL);
1307 	if (!ftrace_mod)
1308 		return -ENOMEM;
1309 
1310 	ftrace_mod->func = kstrdup(func, GFP_KERNEL);
1311 	ftrace_mod->module = kstrdup(module, GFP_KERNEL);
1312 	ftrace_mod->enable = enable;
1313 
1314 	if (!ftrace_mod->func || !ftrace_mod->module)
1315 		goto out_free;
1316 
1317 	list_add(&ftrace_mod->list, mod_head);
1318 
1319 	return 0;
1320 
1321  out_free:
1322 	free_ftrace_mod(ftrace_mod);
1323 
1324 	return -ENOMEM;
1325 }
1326 
1327 static struct ftrace_hash *
alloc_and_copy_ftrace_hash(int size_bits,struct ftrace_hash * hash)1328 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1329 {
1330 	struct ftrace_func_entry *entry;
1331 	struct ftrace_hash *new_hash;
1332 	int size;
1333 	int ret;
1334 	int i;
1335 
1336 	new_hash = alloc_ftrace_hash(size_bits);
1337 	if (!new_hash)
1338 		return NULL;
1339 
1340 	if (hash)
1341 		new_hash->flags = hash->flags;
1342 
1343 	/* Empty hash? */
1344 	if (ftrace_hash_empty(hash))
1345 		return new_hash;
1346 
1347 	size = 1 << hash->size_bits;
1348 	for (i = 0; i < size; i++) {
1349 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1350 			ret = add_hash_entry(new_hash, entry->ip);
1351 			if (ret < 0)
1352 				goto free_hash;
1353 		}
1354 	}
1355 
1356 	FTRACE_WARN_ON(new_hash->count != hash->count);
1357 
1358 	return new_hash;
1359 
1360  free_hash:
1361 	free_ftrace_hash(new_hash);
1362 	return NULL;
1363 }
1364 
1365 static void
1366 ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1367 static void
1368 ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1369 
1370 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1371 				       struct ftrace_hash *new_hash);
1372 
1373 static struct ftrace_hash *
__ftrace_hash_move(struct ftrace_hash * src)1374 __ftrace_hash_move(struct ftrace_hash *src)
1375 {
1376 	struct ftrace_func_entry *entry;
1377 	struct hlist_node *tn;
1378 	struct hlist_head *hhd;
1379 	struct ftrace_hash *new_hash;
1380 	int size = src->count;
1381 	int bits = 0;
1382 	int i;
1383 
1384 	/*
1385 	 * If the new source is empty, just return the empty_hash.
1386 	 */
1387 	if (ftrace_hash_empty(src))
1388 		return EMPTY_HASH;
1389 
1390 	/*
1391 	 * Make the hash size about 1/2 the # found
1392 	 */
1393 	for (size /= 2; size; size >>= 1)
1394 		bits++;
1395 
1396 	/* Don't allocate too much */
1397 	if (bits > FTRACE_HASH_MAX_BITS)
1398 		bits = FTRACE_HASH_MAX_BITS;
1399 
1400 	new_hash = alloc_ftrace_hash(bits);
1401 	if (!new_hash)
1402 		return NULL;
1403 
1404 	new_hash->flags = src->flags;
1405 
1406 	size = 1 << src->size_bits;
1407 	for (i = 0; i < size; i++) {
1408 		hhd = &src->buckets[i];
1409 		hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1410 			remove_hash_entry(src, entry);
1411 			__add_hash_entry(new_hash, entry);
1412 		}
1413 	}
1414 
1415 	return new_hash;
1416 }
1417 
1418 static int
ftrace_hash_move(struct ftrace_ops * ops,int enable,struct ftrace_hash ** dst,struct ftrace_hash * src)1419 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1420 		 struct ftrace_hash **dst, struct ftrace_hash *src)
1421 {
1422 	struct ftrace_hash *new_hash;
1423 	int ret;
1424 
1425 	/* Reject setting notrace hash on IPMODIFY ftrace_ops */
1426 	if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1427 		return -EINVAL;
1428 
1429 	new_hash = __ftrace_hash_move(src);
1430 	if (!new_hash)
1431 		return -ENOMEM;
1432 
1433 	/* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1434 	if (enable) {
1435 		/* IPMODIFY should be updated only when filter_hash updating */
1436 		ret = ftrace_hash_ipmodify_update(ops, new_hash);
1437 		if (ret < 0) {
1438 			free_ftrace_hash(new_hash);
1439 			return ret;
1440 		}
1441 	}
1442 
1443 	/*
1444 	 * Remove the current set, update the hash and add
1445 	 * them back.
1446 	 */
1447 	ftrace_hash_rec_disable_modify(ops, enable);
1448 
1449 	rcu_assign_pointer(*dst, new_hash);
1450 
1451 	ftrace_hash_rec_enable_modify(ops, enable);
1452 
1453 	return 0;
1454 }
1455 
hash_contains_ip(unsigned long ip,struct ftrace_ops_hash * hash)1456 static bool hash_contains_ip(unsigned long ip,
1457 			     struct ftrace_ops_hash *hash)
1458 {
1459 	/*
1460 	 * The function record is a match if it exists in the filter
1461 	 * hash and not in the notrace hash. Note, an emty hash is
1462 	 * considered a match for the filter hash, but an empty
1463 	 * notrace hash is considered not in the notrace hash.
1464 	 */
1465 	return (ftrace_hash_empty(hash->filter_hash) ||
1466 		__ftrace_lookup_ip(hash->filter_hash, ip)) &&
1467 		(ftrace_hash_empty(hash->notrace_hash) ||
1468 		 !__ftrace_lookup_ip(hash->notrace_hash, ip));
1469 }
1470 
1471 /*
1472  * Test the hashes for this ops to see if we want to call
1473  * the ops->func or not.
1474  *
1475  * It's a match if the ip is in the ops->filter_hash or
1476  * the filter_hash does not exist or is empty,
1477  *  AND
1478  * the ip is not in the ops->notrace_hash.
1479  *
1480  * This needs to be called with preemption disabled as
1481  * the hashes are freed with call_rcu().
1482  */
1483 int
ftrace_ops_test(struct ftrace_ops * ops,unsigned long ip,void * regs)1484 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1485 {
1486 	struct ftrace_ops_hash hash;
1487 	int ret;
1488 
1489 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1490 	/*
1491 	 * There's a small race when adding ops that the ftrace handler
1492 	 * that wants regs, may be called without them. We can not
1493 	 * allow that handler to be called if regs is NULL.
1494 	 */
1495 	if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1496 		return 0;
1497 #endif
1498 
1499 	rcu_assign_pointer(hash.filter_hash, ops->func_hash->filter_hash);
1500 	rcu_assign_pointer(hash.notrace_hash, ops->func_hash->notrace_hash);
1501 
1502 	if (hash_contains_ip(ip, &hash))
1503 		ret = 1;
1504 	else
1505 		ret = 0;
1506 
1507 	return ret;
1508 }
1509 
1510 /*
1511  * This is a double for. Do not use 'break' to break out of the loop,
1512  * you must use a goto.
1513  */
1514 #define do_for_each_ftrace_rec(pg, rec)					\
1515 	for (pg = ftrace_pages_start; pg; pg = pg->next) {		\
1516 		int _____i;						\
1517 		for (_____i = 0; _____i < pg->index; _____i++) {	\
1518 			rec = &pg->records[_____i];
1519 
1520 #define while_for_each_ftrace_rec()		\
1521 		}				\
1522 	}
1523 
1524 
ftrace_cmp_recs(const void * a,const void * b)1525 static int ftrace_cmp_recs(const void *a, const void *b)
1526 {
1527 	const struct dyn_ftrace *key = a;
1528 	const struct dyn_ftrace *rec = b;
1529 
1530 	if (key->flags < rec->ip)
1531 		return -1;
1532 	if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1533 		return 1;
1534 	return 0;
1535 }
1536 
1537 /**
1538  * ftrace_location_range - return the first address of a traced location
1539  *	if it touches the given ip range
1540  * @start: start of range to search.
1541  * @end: end of range to search (inclusive). @end points to the last byte
1542  *	to check.
1543  *
1544  * Returns rec->ip if the related ftrace location is a least partly within
1545  * the given address range. That is, the first address of the instruction
1546  * that is either a NOP or call to the function tracer. It checks the ftrace
1547  * internal tables to determine if the address belongs or not.
1548  */
ftrace_location_range(unsigned long start,unsigned long end)1549 unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1550 {
1551 	struct ftrace_page *pg;
1552 	struct dyn_ftrace *rec;
1553 	struct dyn_ftrace key;
1554 
1555 	key.ip = start;
1556 	key.flags = end;	/* overload flags, as it is unsigned long */
1557 
1558 	for (pg = ftrace_pages_start; pg; pg = pg->next) {
1559 		if (end < pg->records[0].ip ||
1560 		    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1561 			continue;
1562 		rec = bsearch(&key, pg->records, pg->index,
1563 			      sizeof(struct dyn_ftrace),
1564 			      ftrace_cmp_recs);
1565 		if (rec)
1566 			return rec->ip;
1567 	}
1568 
1569 	return 0;
1570 }
1571 
1572 /**
1573  * ftrace_location - return true if the ip giving is a traced location
1574  * @ip: the instruction pointer to check
1575  *
1576  * Returns rec->ip if @ip given is a pointer to a ftrace location.
1577  * That is, the instruction that is either a NOP or call to
1578  * the function tracer. It checks the ftrace internal tables to
1579  * determine if the address belongs or not.
1580  */
ftrace_location(unsigned long ip)1581 unsigned long ftrace_location(unsigned long ip)
1582 {
1583 	return ftrace_location_range(ip, ip);
1584 }
1585 
1586 /**
1587  * ftrace_text_reserved - return true if range contains an ftrace location
1588  * @start: start of range to search
1589  * @end: end of range to search (inclusive). @end points to the last byte to check.
1590  *
1591  * Returns 1 if @start and @end contains a ftrace location.
1592  * That is, the instruction that is either a NOP or call to
1593  * the function tracer. It checks the ftrace internal tables to
1594  * determine if the address belongs or not.
1595  */
ftrace_text_reserved(const void * start,const void * end)1596 int ftrace_text_reserved(const void *start, const void *end)
1597 {
1598 	unsigned long ret;
1599 
1600 	ret = ftrace_location_range((unsigned long)start,
1601 				    (unsigned long)end);
1602 
1603 	return (int)!!ret;
1604 }
1605 
1606 /* Test if ops registered to this rec needs regs */
test_rec_ops_needs_regs(struct dyn_ftrace * rec)1607 static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1608 {
1609 	struct ftrace_ops *ops;
1610 	bool keep_regs = false;
1611 
1612 	for (ops = ftrace_ops_list;
1613 	     ops != &ftrace_list_end; ops = ops->next) {
1614 		/* pass rec in as regs to have non-NULL val */
1615 		if (ftrace_ops_test(ops, rec->ip, rec)) {
1616 			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1617 				keep_regs = true;
1618 				break;
1619 			}
1620 		}
1621 	}
1622 
1623 	return  keep_regs;
1624 }
1625 
1626 static struct ftrace_ops *
1627 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
1628 static struct ftrace_ops *
1629 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
1630 
__ftrace_hash_rec_update(struct ftrace_ops * ops,int filter_hash,bool inc)1631 static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1632 				     int filter_hash,
1633 				     bool inc)
1634 {
1635 	struct ftrace_hash *hash;
1636 	struct ftrace_hash *other_hash;
1637 	struct ftrace_page *pg;
1638 	struct dyn_ftrace *rec;
1639 	bool update = false;
1640 	int count = 0;
1641 	int all = false;
1642 
1643 	/* Only update if the ops has been registered */
1644 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1645 		return false;
1646 
1647 	/*
1648 	 * In the filter_hash case:
1649 	 *   If the count is zero, we update all records.
1650 	 *   Otherwise we just update the items in the hash.
1651 	 *
1652 	 * In the notrace_hash case:
1653 	 *   We enable the update in the hash.
1654 	 *   As disabling notrace means enabling the tracing,
1655 	 *   and enabling notrace means disabling, the inc variable
1656 	 *   gets inversed.
1657 	 */
1658 	if (filter_hash) {
1659 		hash = ops->func_hash->filter_hash;
1660 		other_hash = ops->func_hash->notrace_hash;
1661 		if (ftrace_hash_empty(hash))
1662 			all = true;
1663 	} else {
1664 		inc = !inc;
1665 		hash = ops->func_hash->notrace_hash;
1666 		other_hash = ops->func_hash->filter_hash;
1667 		/*
1668 		 * If the notrace hash has no items,
1669 		 * then there's nothing to do.
1670 		 */
1671 		if (ftrace_hash_empty(hash))
1672 			return false;
1673 	}
1674 
1675 	do_for_each_ftrace_rec(pg, rec) {
1676 		int in_other_hash = 0;
1677 		int in_hash = 0;
1678 		int match = 0;
1679 
1680 		if (rec->flags & FTRACE_FL_DISABLED)
1681 			continue;
1682 
1683 		if (all) {
1684 			/*
1685 			 * Only the filter_hash affects all records.
1686 			 * Update if the record is not in the notrace hash.
1687 			 */
1688 			if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1689 				match = 1;
1690 		} else {
1691 			in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1692 			in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1693 
1694 			/*
1695 			 * If filter_hash is set, we want to match all functions
1696 			 * that are in the hash but not in the other hash.
1697 			 *
1698 			 * If filter_hash is not set, then we are decrementing.
1699 			 * That means we match anything that is in the hash
1700 			 * and also in the other_hash. That is, we need to turn
1701 			 * off functions in the other hash because they are disabled
1702 			 * by this hash.
1703 			 */
1704 			if (filter_hash && in_hash && !in_other_hash)
1705 				match = 1;
1706 			else if (!filter_hash && in_hash &&
1707 				 (in_other_hash || ftrace_hash_empty(other_hash)))
1708 				match = 1;
1709 		}
1710 		if (!match)
1711 			continue;
1712 
1713 		if (inc) {
1714 			rec->flags++;
1715 			if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1716 				return false;
1717 
1718 			/*
1719 			 * If there's only a single callback registered to a
1720 			 * function, and the ops has a trampoline registered
1721 			 * for it, then we can call it directly.
1722 			 */
1723 			if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1724 				rec->flags |= FTRACE_FL_TRAMP;
1725 			else
1726 				/*
1727 				 * If we are adding another function callback
1728 				 * to this function, and the previous had a
1729 				 * custom trampoline in use, then we need to go
1730 				 * back to the default trampoline.
1731 				 */
1732 				rec->flags &= ~FTRACE_FL_TRAMP;
1733 
1734 			/*
1735 			 * If any ops wants regs saved for this function
1736 			 * then all ops will get saved regs.
1737 			 */
1738 			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1739 				rec->flags |= FTRACE_FL_REGS;
1740 		} else {
1741 			if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1742 				return false;
1743 			rec->flags--;
1744 
1745 			/*
1746 			 * If the rec had REGS enabled and the ops that is
1747 			 * being removed had REGS set, then see if there is
1748 			 * still any ops for this record that wants regs.
1749 			 * If not, we can stop recording them.
1750 			 */
1751 			if (ftrace_rec_count(rec) > 0 &&
1752 			    rec->flags & FTRACE_FL_REGS &&
1753 			    ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1754 				if (!test_rec_ops_needs_regs(rec))
1755 					rec->flags &= ~FTRACE_FL_REGS;
1756 			}
1757 
1758 			/*
1759 			 * The TRAMP needs to be set only if rec count
1760 			 * is decremented to one, and the ops that is
1761 			 * left has a trampoline. As TRAMP can only be
1762 			 * enabled if there is only a single ops attached
1763 			 * to it.
1764 			 */
1765 			if (ftrace_rec_count(rec) == 1 &&
1766 			    ftrace_find_tramp_ops_any(rec))
1767 				rec->flags |= FTRACE_FL_TRAMP;
1768 			else
1769 				rec->flags &= ~FTRACE_FL_TRAMP;
1770 
1771 			/*
1772 			 * flags will be cleared in ftrace_check_record()
1773 			 * if rec count is zero.
1774 			 */
1775 		}
1776 		count++;
1777 
1778 		/* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1779 		update |= ftrace_test_record(rec, true) != FTRACE_UPDATE_IGNORE;
1780 
1781 		/* Shortcut, if we handled all records, we are done. */
1782 		if (!all && count == hash->count)
1783 			return update;
1784 	} while_for_each_ftrace_rec();
1785 
1786 	return update;
1787 }
1788 
ftrace_hash_rec_disable(struct ftrace_ops * ops,int filter_hash)1789 static bool ftrace_hash_rec_disable(struct ftrace_ops *ops,
1790 				    int filter_hash)
1791 {
1792 	return __ftrace_hash_rec_update(ops, filter_hash, 0);
1793 }
1794 
ftrace_hash_rec_enable(struct ftrace_ops * ops,int filter_hash)1795 static bool ftrace_hash_rec_enable(struct ftrace_ops *ops,
1796 				   int filter_hash)
1797 {
1798 	return __ftrace_hash_rec_update(ops, filter_hash, 1);
1799 }
1800 
ftrace_hash_rec_update_modify(struct ftrace_ops * ops,int filter_hash,int inc)1801 static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1802 					  int filter_hash, int inc)
1803 {
1804 	struct ftrace_ops *op;
1805 
1806 	__ftrace_hash_rec_update(ops, filter_hash, inc);
1807 
1808 	if (ops->func_hash != &global_ops.local_hash)
1809 		return;
1810 
1811 	/*
1812 	 * If the ops shares the global_ops hash, then we need to update
1813 	 * all ops that are enabled and use this hash.
1814 	 */
1815 	do_for_each_ftrace_op(op, ftrace_ops_list) {
1816 		/* Already done */
1817 		if (op == ops)
1818 			continue;
1819 		if (op->func_hash == &global_ops.local_hash)
1820 			__ftrace_hash_rec_update(op, filter_hash, inc);
1821 	} while_for_each_ftrace_op(op);
1822 }
1823 
ftrace_hash_rec_disable_modify(struct ftrace_ops * ops,int filter_hash)1824 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1825 					   int filter_hash)
1826 {
1827 	ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1828 }
1829 
ftrace_hash_rec_enable_modify(struct ftrace_ops * ops,int filter_hash)1830 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1831 					  int filter_hash)
1832 {
1833 	ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1834 }
1835 
1836 /*
1837  * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1838  * or no-needed to update, -EBUSY if it detects a conflict of the flag
1839  * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1840  * Note that old_hash and new_hash has below meanings
1841  *  - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1842  *  - If the hash is EMPTY_HASH, it hits nothing
1843  *  - Anything else hits the recs which match the hash entries.
1844  */
__ftrace_hash_update_ipmodify(struct ftrace_ops * ops,struct ftrace_hash * old_hash,struct ftrace_hash * new_hash)1845 static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1846 					 struct ftrace_hash *old_hash,
1847 					 struct ftrace_hash *new_hash)
1848 {
1849 	struct ftrace_page *pg;
1850 	struct dyn_ftrace *rec, *end = NULL;
1851 	int in_old, in_new;
1852 
1853 	/* Only update if the ops has been registered */
1854 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1855 		return 0;
1856 
1857 	if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
1858 		return 0;
1859 
1860 	/*
1861 	 * Since the IPMODIFY is a very address sensitive action, we do not
1862 	 * allow ftrace_ops to set all functions to new hash.
1863 	 */
1864 	if (!new_hash || !old_hash)
1865 		return -EINVAL;
1866 
1867 	/* Update rec->flags */
1868 	do_for_each_ftrace_rec(pg, rec) {
1869 
1870 		if (rec->flags & FTRACE_FL_DISABLED)
1871 			continue;
1872 
1873 		/* We need to update only differences of filter_hash */
1874 		in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1875 		in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1876 		if (in_old == in_new)
1877 			continue;
1878 
1879 		if (in_new) {
1880 			/* New entries must ensure no others are using it */
1881 			if (rec->flags & FTRACE_FL_IPMODIFY)
1882 				goto rollback;
1883 			rec->flags |= FTRACE_FL_IPMODIFY;
1884 		} else /* Removed entry */
1885 			rec->flags &= ~FTRACE_FL_IPMODIFY;
1886 	} while_for_each_ftrace_rec();
1887 
1888 	return 0;
1889 
1890 rollback:
1891 	end = rec;
1892 
1893 	/* Roll back what we did above */
1894 	do_for_each_ftrace_rec(pg, rec) {
1895 
1896 		if (rec->flags & FTRACE_FL_DISABLED)
1897 			continue;
1898 
1899 		if (rec == end)
1900 			goto err_out;
1901 
1902 		in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1903 		in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1904 		if (in_old == in_new)
1905 			continue;
1906 
1907 		if (in_new)
1908 			rec->flags &= ~FTRACE_FL_IPMODIFY;
1909 		else
1910 			rec->flags |= FTRACE_FL_IPMODIFY;
1911 	} while_for_each_ftrace_rec();
1912 
1913 err_out:
1914 	return -EBUSY;
1915 }
1916 
ftrace_hash_ipmodify_enable(struct ftrace_ops * ops)1917 static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
1918 {
1919 	struct ftrace_hash *hash = ops->func_hash->filter_hash;
1920 
1921 	if (ftrace_hash_empty(hash))
1922 		hash = NULL;
1923 
1924 	return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
1925 }
1926 
1927 /* Disabling always succeeds */
ftrace_hash_ipmodify_disable(struct ftrace_ops * ops)1928 static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
1929 {
1930 	struct ftrace_hash *hash = ops->func_hash->filter_hash;
1931 
1932 	if (ftrace_hash_empty(hash))
1933 		hash = NULL;
1934 
1935 	__ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
1936 }
1937 
ftrace_hash_ipmodify_update(struct ftrace_ops * ops,struct ftrace_hash * new_hash)1938 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1939 				       struct ftrace_hash *new_hash)
1940 {
1941 	struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
1942 
1943 	if (ftrace_hash_empty(old_hash))
1944 		old_hash = NULL;
1945 
1946 	if (ftrace_hash_empty(new_hash))
1947 		new_hash = NULL;
1948 
1949 	return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
1950 }
1951 
print_ip_ins(const char * fmt,const unsigned char * p)1952 static void print_ip_ins(const char *fmt, const unsigned char *p)
1953 {
1954 	int i;
1955 
1956 	printk(KERN_CONT "%s", fmt);
1957 
1958 	for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1959 		printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1960 }
1961 
1962 enum ftrace_bug_type ftrace_bug_type;
1963 const void *ftrace_expected;
1964 
print_bug_type(void)1965 static void print_bug_type(void)
1966 {
1967 	switch (ftrace_bug_type) {
1968 	case FTRACE_BUG_UNKNOWN:
1969 		break;
1970 	case FTRACE_BUG_INIT:
1971 		pr_info("Initializing ftrace call sites\n");
1972 		break;
1973 	case FTRACE_BUG_NOP:
1974 		pr_info("Setting ftrace call site to NOP\n");
1975 		break;
1976 	case FTRACE_BUG_CALL:
1977 		pr_info("Setting ftrace call site to call ftrace function\n");
1978 		break;
1979 	case FTRACE_BUG_UPDATE:
1980 		pr_info("Updating ftrace call site to call a different ftrace function\n");
1981 		break;
1982 	}
1983 }
1984 
1985 /**
1986  * ftrace_bug - report and shutdown function tracer
1987  * @failed: The failed type (EFAULT, EINVAL, EPERM)
1988  * @rec: The record that failed
1989  *
1990  * The arch code that enables or disables the function tracing
1991  * can call ftrace_bug() when it has detected a problem in
1992  * modifying the code. @failed should be one of either:
1993  * EFAULT - if the problem happens on reading the @ip address
1994  * EINVAL - if what is read at @ip is not what was expected
1995  * EPERM - if the problem happens on writing to the @ip address
1996  */
ftrace_bug(int failed,struct dyn_ftrace * rec)1997 void ftrace_bug(int failed, struct dyn_ftrace *rec)
1998 {
1999 	unsigned long ip = rec ? rec->ip : 0;
2000 
2001 	switch (failed) {
2002 	case -EFAULT:
2003 		FTRACE_WARN_ON_ONCE(1);
2004 		pr_info("ftrace faulted on modifying ");
2005 		print_ip_sym(ip);
2006 		break;
2007 	case -EINVAL:
2008 		FTRACE_WARN_ON_ONCE(1);
2009 		pr_info("ftrace failed to modify ");
2010 		print_ip_sym(ip);
2011 		print_ip_ins(" actual:   ", (unsigned char *)ip);
2012 		pr_cont("\n");
2013 		if (ftrace_expected) {
2014 			print_ip_ins(" expected: ", ftrace_expected);
2015 			pr_cont("\n");
2016 		}
2017 		break;
2018 	case -EPERM:
2019 		FTRACE_WARN_ON_ONCE(1);
2020 		pr_info("ftrace faulted on writing ");
2021 		print_ip_sym(ip);
2022 		break;
2023 	default:
2024 		FTRACE_WARN_ON_ONCE(1);
2025 		pr_info("ftrace faulted on unknown error ");
2026 		print_ip_sym(ip);
2027 	}
2028 	print_bug_type();
2029 	if (rec) {
2030 		struct ftrace_ops *ops = NULL;
2031 
2032 		pr_info("ftrace record flags: %lx\n", rec->flags);
2033 		pr_cont(" (%ld)%s", ftrace_rec_count(rec),
2034 			rec->flags & FTRACE_FL_REGS ? " R" : "  ");
2035 		if (rec->flags & FTRACE_FL_TRAMP_EN) {
2036 			ops = ftrace_find_tramp_ops_any(rec);
2037 			if (ops) {
2038 				do {
2039 					pr_cont("\ttramp: %pS (%pS)",
2040 						(void *)ops->trampoline,
2041 						(void *)ops->func);
2042 					ops = ftrace_find_tramp_ops_next(rec, ops);
2043 				} while (ops);
2044 			} else
2045 				pr_cont("\ttramp: ERROR!");
2046 
2047 		}
2048 		ip = ftrace_get_addr_curr(rec);
2049 		pr_cont("\n expected tramp: %lx\n", ip);
2050 	}
2051 }
2052 
ftrace_check_record(struct dyn_ftrace * rec,bool enable,bool update)2053 static int ftrace_check_record(struct dyn_ftrace *rec, bool enable, bool update)
2054 {
2055 	unsigned long flag = 0UL;
2056 
2057 	ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2058 
2059 	if (rec->flags & FTRACE_FL_DISABLED)
2060 		return FTRACE_UPDATE_IGNORE;
2061 
2062 	/*
2063 	 * If we are updating calls:
2064 	 *
2065 	 *   If the record has a ref count, then we need to enable it
2066 	 *   because someone is using it.
2067 	 *
2068 	 *   Otherwise we make sure its disabled.
2069 	 *
2070 	 * If we are disabling calls, then disable all records that
2071 	 * are enabled.
2072 	 */
2073 	if (enable && ftrace_rec_count(rec))
2074 		flag = FTRACE_FL_ENABLED;
2075 
2076 	/*
2077 	 * If enabling and the REGS flag does not match the REGS_EN, or
2078 	 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2079 	 * this record. Set flags to fail the compare against ENABLED.
2080 	 */
2081 	if (flag) {
2082 		if (!(rec->flags & FTRACE_FL_REGS) !=
2083 		    !(rec->flags & FTRACE_FL_REGS_EN))
2084 			flag |= FTRACE_FL_REGS;
2085 
2086 		if (!(rec->flags & FTRACE_FL_TRAMP) !=
2087 		    !(rec->flags & FTRACE_FL_TRAMP_EN))
2088 			flag |= FTRACE_FL_TRAMP;
2089 	}
2090 
2091 	/* If the state of this record hasn't changed, then do nothing */
2092 	if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2093 		return FTRACE_UPDATE_IGNORE;
2094 
2095 	if (flag) {
2096 		/* Save off if rec is being enabled (for return value) */
2097 		flag ^= rec->flags & FTRACE_FL_ENABLED;
2098 
2099 		if (update) {
2100 			rec->flags |= FTRACE_FL_ENABLED;
2101 			if (flag & FTRACE_FL_REGS) {
2102 				if (rec->flags & FTRACE_FL_REGS)
2103 					rec->flags |= FTRACE_FL_REGS_EN;
2104 				else
2105 					rec->flags &= ~FTRACE_FL_REGS_EN;
2106 			}
2107 			if (flag & FTRACE_FL_TRAMP) {
2108 				if (rec->flags & FTRACE_FL_TRAMP)
2109 					rec->flags |= FTRACE_FL_TRAMP_EN;
2110 				else
2111 					rec->flags &= ~FTRACE_FL_TRAMP_EN;
2112 			}
2113 		}
2114 
2115 		/*
2116 		 * If this record is being updated from a nop, then
2117 		 *   return UPDATE_MAKE_CALL.
2118 		 * Otherwise,
2119 		 *   return UPDATE_MODIFY_CALL to tell the caller to convert
2120 		 *   from the save regs, to a non-save regs function or
2121 		 *   vice versa, or from a trampoline call.
2122 		 */
2123 		if (flag & FTRACE_FL_ENABLED) {
2124 			ftrace_bug_type = FTRACE_BUG_CALL;
2125 			return FTRACE_UPDATE_MAKE_CALL;
2126 		}
2127 
2128 		ftrace_bug_type = FTRACE_BUG_UPDATE;
2129 		return FTRACE_UPDATE_MODIFY_CALL;
2130 	}
2131 
2132 	if (update) {
2133 		/* If there's no more users, clear all flags */
2134 		if (!ftrace_rec_count(rec))
2135 			rec->flags = 0;
2136 		else
2137 			/*
2138 			 * Just disable the record, but keep the ops TRAMP
2139 			 * and REGS states. The _EN flags must be disabled though.
2140 			 */
2141 			rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2142 					FTRACE_FL_REGS_EN);
2143 	}
2144 
2145 	ftrace_bug_type = FTRACE_BUG_NOP;
2146 	return FTRACE_UPDATE_MAKE_NOP;
2147 }
2148 
2149 /**
2150  * ftrace_update_record, set a record that now is tracing or not
2151  * @rec: the record to update
2152  * @enable: set to true if the record is tracing, false to force disable
2153  *
2154  * The records that represent all functions that can be traced need
2155  * to be updated when tracing has been enabled.
2156  */
ftrace_update_record(struct dyn_ftrace * rec,bool enable)2157 int ftrace_update_record(struct dyn_ftrace *rec, bool enable)
2158 {
2159 	return ftrace_check_record(rec, enable, true);
2160 }
2161 
2162 /**
2163  * ftrace_test_record, check if the record has been enabled or not
2164  * @rec: the record to test
2165  * @enable: set to true to check if enabled, false if it is disabled
2166  *
2167  * The arch code may need to test if a record is already set to
2168  * tracing to determine how to modify the function code that it
2169  * represents.
2170  */
ftrace_test_record(struct dyn_ftrace * rec,bool enable)2171 int ftrace_test_record(struct dyn_ftrace *rec, bool enable)
2172 {
2173 	return ftrace_check_record(rec, enable, false);
2174 }
2175 
2176 static struct ftrace_ops *
ftrace_find_tramp_ops_any(struct dyn_ftrace * rec)2177 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2178 {
2179 	struct ftrace_ops *op;
2180 	unsigned long ip = rec->ip;
2181 
2182 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2183 
2184 		if (!op->trampoline)
2185 			continue;
2186 
2187 		if (hash_contains_ip(ip, op->func_hash))
2188 			return op;
2189 	} while_for_each_ftrace_op(op);
2190 
2191 	return NULL;
2192 }
2193 
2194 static struct ftrace_ops *
ftrace_find_tramp_ops_next(struct dyn_ftrace * rec,struct ftrace_ops * op)2195 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2196 			   struct ftrace_ops *op)
2197 {
2198 	unsigned long ip = rec->ip;
2199 
2200 	while_for_each_ftrace_op(op) {
2201 
2202 		if (!op->trampoline)
2203 			continue;
2204 
2205 		if (hash_contains_ip(ip, op->func_hash))
2206 			return op;
2207 	}
2208 
2209 	return NULL;
2210 }
2211 
2212 static struct ftrace_ops *
ftrace_find_tramp_ops_curr(struct dyn_ftrace * rec)2213 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2214 {
2215 	struct ftrace_ops *op;
2216 	unsigned long ip = rec->ip;
2217 
2218 	/*
2219 	 * Need to check removed ops first.
2220 	 * If they are being removed, and this rec has a tramp,
2221 	 * and this rec is in the ops list, then it would be the
2222 	 * one with the tramp.
2223 	 */
2224 	if (removed_ops) {
2225 		if (hash_contains_ip(ip, &removed_ops->old_hash))
2226 			return removed_ops;
2227 	}
2228 
2229 	/*
2230 	 * Need to find the current trampoline for a rec.
2231 	 * Now, a trampoline is only attached to a rec if there
2232 	 * was a single 'ops' attached to it. But this can be called
2233 	 * when we are adding another op to the rec or removing the
2234 	 * current one. Thus, if the op is being added, we can
2235 	 * ignore it because it hasn't attached itself to the rec
2236 	 * yet.
2237 	 *
2238 	 * If an ops is being modified (hooking to different functions)
2239 	 * then we don't care about the new functions that are being
2240 	 * added, just the old ones (that are probably being removed).
2241 	 *
2242 	 * If we are adding an ops to a function that already is using
2243 	 * a trampoline, it needs to be removed (trampolines are only
2244 	 * for single ops connected), then an ops that is not being
2245 	 * modified also needs to be checked.
2246 	 */
2247 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2248 
2249 		if (!op->trampoline)
2250 			continue;
2251 
2252 		/*
2253 		 * If the ops is being added, it hasn't gotten to
2254 		 * the point to be removed from this tree yet.
2255 		 */
2256 		if (op->flags & FTRACE_OPS_FL_ADDING)
2257 			continue;
2258 
2259 
2260 		/*
2261 		 * If the ops is being modified and is in the old
2262 		 * hash, then it is probably being removed from this
2263 		 * function.
2264 		 */
2265 		if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2266 		    hash_contains_ip(ip, &op->old_hash))
2267 			return op;
2268 		/*
2269 		 * If the ops is not being added or modified, and it's
2270 		 * in its normal filter hash, then this must be the one
2271 		 * we want!
2272 		 */
2273 		if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2274 		    hash_contains_ip(ip, op->func_hash))
2275 			return op;
2276 
2277 	} while_for_each_ftrace_op(op);
2278 
2279 	return NULL;
2280 }
2281 
2282 static struct ftrace_ops *
ftrace_find_tramp_ops_new(struct dyn_ftrace * rec)2283 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2284 {
2285 	struct ftrace_ops *op;
2286 	unsigned long ip = rec->ip;
2287 
2288 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2289 		/* pass rec in as regs to have non-NULL val */
2290 		if (hash_contains_ip(ip, op->func_hash))
2291 			return op;
2292 	} while_for_each_ftrace_op(op);
2293 
2294 	return NULL;
2295 }
2296 
2297 /**
2298  * ftrace_get_addr_new - Get the call address to set to
2299  * @rec:  The ftrace record descriptor
2300  *
2301  * If the record has the FTRACE_FL_REGS set, that means that it
2302  * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2303  * is not not set, then it wants to convert to the normal callback.
2304  *
2305  * Returns the address of the trampoline to set to
2306  */
ftrace_get_addr_new(struct dyn_ftrace * rec)2307 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2308 {
2309 	struct ftrace_ops *ops;
2310 
2311 	/* Trampolines take precedence over regs */
2312 	if (rec->flags & FTRACE_FL_TRAMP) {
2313 		ops = ftrace_find_tramp_ops_new(rec);
2314 		if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2315 			pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2316 				(void *)rec->ip, (void *)rec->ip, rec->flags);
2317 			/* Ftrace is shutting down, return anything */
2318 			return (unsigned long)FTRACE_ADDR;
2319 		}
2320 		return ops->trampoline;
2321 	}
2322 
2323 	if (rec->flags & FTRACE_FL_REGS)
2324 		return (unsigned long)FTRACE_REGS_ADDR;
2325 	else
2326 		return (unsigned long)FTRACE_ADDR;
2327 }
2328 
2329 /**
2330  * ftrace_get_addr_curr - Get the call address that is already there
2331  * @rec:  The ftrace record descriptor
2332  *
2333  * The FTRACE_FL_REGS_EN is set when the record already points to
2334  * a function that saves all the regs. Basically the '_EN' version
2335  * represents the current state of the function.
2336  *
2337  * Returns the address of the trampoline that is currently being called
2338  */
ftrace_get_addr_curr(struct dyn_ftrace * rec)2339 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2340 {
2341 	struct ftrace_ops *ops;
2342 
2343 	/* Trampolines take precedence over regs */
2344 	if (rec->flags & FTRACE_FL_TRAMP_EN) {
2345 		ops = ftrace_find_tramp_ops_curr(rec);
2346 		if (FTRACE_WARN_ON(!ops)) {
2347 			pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2348 				(void *)rec->ip, (void *)rec->ip);
2349 			/* Ftrace is shutting down, return anything */
2350 			return (unsigned long)FTRACE_ADDR;
2351 		}
2352 		return ops->trampoline;
2353 	}
2354 
2355 	if (rec->flags & FTRACE_FL_REGS_EN)
2356 		return (unsigned long)FTRACE_REGS_ADDR;
2357 	else
2358 		return (unsigned long)FTRACE_ADDR;
2359 }
2360 
2361 static int
__ftrace_replace_code(struct dyn_ftrace * rec,bool enable)2362 __ftrace_replace_code(struct dyn_ftrace *rec, bool enable)
2363 {
2364 	unsigned long ftrace_old_addr;
2365 	unsigned long ftrace_addr;
2366 	int ret;
2367 
2368 	ftrace_addr = ftrace_get_addr_new(rec);
2369 
2370 	/* This needs to be done before we call ftrace_update_record */
2371 	ftrace_old_addr = ftrace_get_addr_curr(rec);
2372 
2373 	ret = ftrace_update_record(rec, enable);
2374 
2375 	ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2376 
2377 	switch (ret) {
2378 	case FTRACE_UPDATE_IGNORE:
2379 		return 0;
2380 
2381 	case FTRACE_UPDATE_MAKE_CALL:
2382 		ftrace_bug_type = FTRACE_BUG_CALL;
2383 		return ftrace_make_call(rec, ftrace_addr);
2384 
2385 	case FTRACE_UPDATE_MAKE_NOP:
2386 		ftrace_bug_type = FTRACE_BUG_NOP;
2387 		return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2388 
2389 	case FTRACE_UPDATE_MODIFY_CALL:
2390 		ftrace_bug_type = FTRACE_BUG_UPDATE;
2391 		return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2392 	}
2393 
2394 	return -1; /* unknown ftrace bug */
2395 }
2396 
ftrace_replace_code(int mod_flags)2397 void __weak ftrace_replace_code(int mod_flags)
2398 {
2399 	struct dyn_ftrace *rec;
2400 	struct ftrace_page *pg;
2401 	bool enable = mod_flags & FTRACE_MODIFY_ENABLE_FL;
2402 	int schedulable = mod_flags & FTRACE_MODIFY_MAY_SLEEP_FL;
2403 	int failed;
2404 
2405 	if (unlikely(ftrace_disabled))
2406 		return;
2407 
2408 	do_for_each_ftrace_rec(pg, rec) {
2409 
2410 		if (rec->flags & FTRACE_FL_DISABLED)
2411 			continue;
2412 
2413 		failed = __ftrace_replace_code(rec, enable);
2414 		if (failed) {
2415 			ftrace_bug(failed, rec);
2416 			/* Stop processing */
2417 			return;
2418 		}
2419 		if (schedulable)
2420 			cond_resched();
2421 	} while_for_each_ftrace_rec();
2422 }
2423 
2424 struct ftrace_rec_iter {
2425 	struct ftrace_page	*pg;
2426 	int			index;
2427 };
2428 
2429 /**
2430  * ftrace_rec_iter_start, start up iterating over traced functions
2431  *
2432  * Returns an iterator handle that is used to iterate over all
2433  * the records that represent address locations where functions
2434  * are traced.
2435  *
2436  * May return NULL if no records are available.
2437  */
ftrace_rec_iter_start(void)2438 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2439 {
2440 	/*
2441 	 * We only use a single iterator.
2442 	 * Protected by the ftrace_lock mutex.
2443 	 */
2444 	static struct ftrace_rec_iter ftrace_rec_iter;
2445 	struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2446 
2447 	iter->pg = ftrace_pages_start;
2448 	iter->index = 0;
2449 
2450 	/* Could have empty pages */
2451 	while (iter->pg && !iter->pg->index)
2452 		iter->pg = iter->pg->next;
2453 
2454 	if (!iter->pg)
2455 		return NULL;
2456 
2457 	return iter;
2458 }
2459 
2460 /**
2461  * ftrace_rec_iter_next, get the next record to process.
2462  * @iter: The handle to the iterator.
2463  *
2464  * Returns the next iterator after the given iterator @iter.
2465  */
ftrace_rec_iter_next(struct ftrace_rec_iter * iter)2466 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2467 {
2468 	iter->index++;
2469 
2470 	if (iter->index >= iter->pg->index) {
2471 		iter->pg = iter->pg->next;
2472 		iter->index = 0;
2473 
2474 		/* Could have empty pages */
2475 		while (iter->pg && !iter->pg->index)
2476 			iter->pg = iter->pg->next;
2477 	}
2478 
2479 	if (!iter->pg)
2480 		return NULL;
2481 
2482 	return iter;
2483 }
2484 
2485 /**
2486  * ftrace_rec_iter_record, get the record at the iterator location
2487  * @iter: The current iterator location
2488  *
2489  * Returns the record that the current @iter is at.
2490  */
ftrace_rec_iter_record(struct ftrace_rec_iter * iter)2491 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2492 {
2493 	return &iter->pg->records[iter->index];
2494 }
2495 
2496 static int
ftrace_code_disable(struct module * mod,struct dyn_ftrace * rec)2497 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
2498 {
2499 	int ret;
2500 
2501 	if (unlikely(ftrace_disabled))
2502 		return 0;
2503 
2504 	ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
2505 	if (ret) {
2506 		ftrace_bug_type = FTRACE_BUG_INIT;
2507 		ftrace_bug(ret, rec);
2508 		return 0;
2509 	}
2510 	return 1;
2511 }
2512 
2513 /*
2514  * archs can override this function if they must do something
2515  * before the modifying code is performed.
2516  */
ftrace_arch_code_modify_prepare(void)2517 int __weak ftrace_arch_code_modify_prepare(void)
2518 {
2519 	return 0;
2520 }
2521 
2522 /*
2523  * archs can override this function if they must do something
2524  * after the modifying code is performed.
2525  */
ftrace_arch_code_modify_post_process(void)2526 int __weak ftrace_arch_code_modify_post_process(void)
2527 {
2528 	return 0;
2529 }
2530 
ftrace_modify_all_code(int command)2531 void ftrace_modify_all_code(int command)
2532 {
2533 	int update = command & FTRACE_UPDATE_TRACE_FUNC;
2534 	int mod_flags = 0;
2535 	int err = 0;
2536 
2537 	if (command & FTRACE_MAY_SLEEP)
2538 		mod_flags = FTRACE_MODIFY_MAY_SLEEP_FL;
2539 
2540 	/*
2541 	 * If the ftrace_caller calls a ftrace_ops func directly,
2542 	 * we need to make sure that it only traces functions it
2543 	 * expects to trace. When doing the switch of functions,
2544 	 * we need to update to the ftrace_ops_list_func first
2545 	 * before the transition between old and new calls are set,
2546 	 * as the ftrace_ops_list_func will check the ops hashes
2547 	 * to make sure the ops are having the right functions
2548 	 * traced.
2549 	 */
2550 	if (update) {
2551 		err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2552 		if (FTRACE_WARN_ON(err))
2553 			return;
2554 	}
2555 
2556 	if (command & FTRACE_UPDATE_CALLS)
2557 		ftrace_replace_code(mod_flags | FTRACE_MODIFY_ENABLE_FL);
2558 	else if (command & FTRACE_DISABLE_CALLS)
2559 		ftrace_replace_code(mod_flags);
2560 
2561 	if (update && ftrace_trace_function != ftrace_ops_list_func) {
2562 		function_trace_op = set_function_trace_op;
2563 		smp_wmb();
2564 		/* If irqs are disabled, we are in stop machine */
2565 		if (!irqs_disabled())
2566 			smp_call_function(ftrace_sync_ipi, NULL, 1);
2567 		err = ftrace_update_ftrace_func(ftrace_trace_function);
2568 		if (FTRACE_WARN_ON(err))
2569 			return;
2570 	}
2571 
2572 	if (command & FTRACE_START_FUNC_RET)
2573 		err = ftrace_enable_ftrace_graph_caller();
2574 	else if (command & FTRACE_STOP_FUNC_RET)
2575 		err = ftrace_disable_ftrace_graph_caller();
2576 	FTRACE_WARN_ON(err);
2577 }
2578 
__ftrace_modify_code(void * data)2579 static int __ftrace_modify_code(void *data)
2580 {
2581 	int *command = data;
2582 
2583 	ftrace_modify_all_code(*command);
2584 
2585 	return 0;
2586 }
2587 
2588 /**
2589  * ftrace_run_stop_machine, go back to the stop machine method
2590  * @command: The command to tell ftrace what to do
2591  *
2592  * If an arch needs to fall back to the stop machine method, the
2593  * it can call this function.
2594  */
ftrace_run_stop_machine(int command)2595 void ftrace_run_stop_machine(int command)
2596 {
2597 	stop_machine(__ftrace_modify_code, &command, NULL);
2598 }
2599 
2600 /**
2601  * arch_ftrace_update_code, modify the code to trace or not trace
2602  * @command: The command that needs to be done
2603  *
2604  * Archs can override this function if it does not need to
2605  * run stop_machine() to modify code.
2606  */
arch_ftrace_update_code(int command)2607 void __weak arch_ftrace_update_code(int command)
2608 {
2609 	ftrace_run_stop_machine(command);
2610 }
2611 
ftrace_run_update_code(int command)2612 static void ftrace_run_update_code(int command)
2613 {
2614 	int ret;
2615 
2616 	ret = ftrace_arch_code_modify_prepare();
2617 	FTRACE_WARN_ON(ret);
2618 	if (ret)
2619 		return;
2620 
2621 	/*
2622 	 * By default we use stop_machine() to modify the code.
2623 	 * But archs can do what ever they want as long as it
2624 	 * is safe. The stop_machine() is the safest, but also
2625 	 * produces the most overhead.
2626 	 */
2627 	arch_ftrace_update_code(command);
2628 
2629 	ret = ftrace_arch_code_modify_post_process();
2630 	FTRACE_WARN_ON(ret);
2631 }
2632 
ftrace_run_modify_code(struct ftrace_ops * ops,int command,struct ftrace_ops_hash * old_hash)2633 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2634 				   struct ftrace_ops_hash *old_hash)
2635 {
2636 	ops->flags |= FTRACE_OPS_FL_MODIFYING;
2637 	ops->old_hash.filter_hash = old_hash->filter_hash;
2638 	ops->old_hash.notrace_hash = old_hash->notrace_hash;
2639 	ftrace_run_update_code(command);
2640 	ops->old_hash.filter_hash = NULL;
2641 	ops->old_hash.notrace_hash = NULL;
2642 	ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2643 }
2644 
2645 static ftrace_func_t saved_ftrace_func;
2646 static int ftrace_start_up;
2647 
arch_ftrace_trampoline_free(struct ftrace_ops * ops)2648 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2649 {
2650 }
2651 
ftrace_startup_enable(int command)2652 static void ftrace_startup_enable(int command)
2653 {
2654 	if (saved_ftrace_func != ftrace_trace_function) {
2655 		saved_ftrace_func = ftrace_trace_function;
2656 		command |= FTRACE_UPDATE_TRACE_FUNC;
2657 	}
2658 
2659 	if (!command || !ftrace_enabled)
2660 		return;
2661 
2662 	ftrace_run_update_code(command);
2663 }
2664 
ftrace_startup_all(int command)2665 static void ftrace_startup_all(int command)
2666 {
2667 	update_all_ops = true;
2668 	ftrace_startup_enable(command);
2669 	update_all_ops = false;
2670 }
2671 
ftrace_startup(struct ftrace_ops * ops,int command)2672 int ftrace_startup(struct ftrace_ops *ops, int command)
2673 {
2674 	int ret;
2675 
2676 	if (unlikely(ftrace_disabled))
2677 		return -ENODEV;
2678 
2679 	ret = __register_ftrace_function(ops);
2680 	if (ret)
2681 		return ret;
2682 
2683 	ftrace_start_up++;
2684 
2685 	/*
2686 	 * Note that ftrace probes uses this to start up
2687 	 * and modify functions it will probe. But we still
2688 	 * set the ADDING flag for modification, as probes
2689 	 * do not have trampolines. If they add them in the
2690 	 * future, then the probes will need to distinguish
2691 	 * between adding and updating probes.
2692 	 */
2693 	ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
2694 
2695 	ret = ftrace_hash_ipmodify_enable(ops);
2696 	if (ret < 0) {
2697 		/* Rollback registration process */
2698 		__unregister_ftrace_function(ops);
2699 		ftrace_start_up--;
2700 		ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2701 		return ret;
2702 	}
2703 
2704 	if (ftrace_hash_rec_enable(ops, 1))
2705 		command |= FTRACE_UPDATE_CALLS;
2706 
2707 	ftrace_startup_enable(command);
2708 
2709 	ops->flags &= ~FTRACE_OPS_FL_ADDING;
2710 
2711 	return 0;
2712 }
2713 
ftrace_shutdown(struct ftrace_ops * ops,int command)2714 int ftrace_shutdown(struct ftrace_ops *ops, int command)
2715 {
2716 	int ret;
2717 
2718 	if (unlikely(ftrace_disabled))
2719 		return -ENODEV;
2720 
2721 	ret = __unregister_ftrace_function(ops);
2722 	if (ret)
2723 		return ret;
2724 
2725 	ftrace_start_up--;
2726 	/*
2727 	 * Just warn in case of unbalance, no need to kill ftrace, it's not
2728 	 * critical but the ftrace_call callers may be never nopped again after
2729 	 * further ftrace uses.
2730 	 */
2731 	WARN_ON_ONCE(ftrace_start_up < 0);
2732 
2733 	/* Disabling ipmodify never fails */
2734 	ftrace_hash_ipmodify_disable(ops);
2735 
2736 	if (ftrace_hash_rec_disable(ops, 1))
2737 		command |= FTRACE_UPDATE_CALLS;
2738 
2739 	ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2740 
2741 	if (saved_ftrace_func != ftrace_trace_function) {
2742 		saved_ftrace_func = ftrace_trace_function;
2743 		command |= FTRACE_UPDATE_TRACE_FUNC;
2744 	}
2745 
2746 	if (!command || !ftrace_enabled) {
2747 		/*
2748 		 * If these are dynamic or per_cpu ops, they still
2749 		 * need their data freed. Since, function tracing is
2750 		 * not currently active, we can just free them
2751 		 * without synchronizing all CPUs.
2752 		 */
2753 		if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
2754 			goto free_ops;
2755 
2756 		return 0;
2757 	}
2758 
2759 	/*
2760 	 * If the ops uses a trampoline, then it needs to be
2761 	 * tested first on update.
2762 	 */
2763 	ops->flags |= FTRACE_OPS_FL_REMOVING;
2764 	removed_ops = ops;
2765 
2766 	/* The trampoline logic checks the old hashes */
2767 	ops->old_hash.filter_hash = ops->func_hash->filter_hash;
2768 	ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
2769 
2770 	ftrace_run_update_code(command);
2771 
2772 	/*
2773 	 * If there's no more ops registered with ftrace, run a
2774 	 * sanity check to make sure all rec flags are cleared.
2775 	 */
2776 	if (rcu_dereference_protected(ftrace_ops_list,
2777 			lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
2778 		struct ftrace_page *pg;
2779 		struct dyn_ftrace *rec;
2780 
2781 		do_for_each_ftrace_rec(pg, rec) {
2782 			if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_FL_DISABLED))
2783 				pr_warn("  %pS flags:%lx\n",
2784 					(void *)rec->ip, rec->flags);
2785 		} while_for_each_ftrace_rec();
2786 	}
2787 
2788 	ops->old_hash.filter_hash = NULL;
2789 	ops->old_hash.notrace_hash = NULL;
2790 
2791 	removed_ops = NULL;
2792 	ops->flags &= ~FTRACE_OPS_FL_REMOVING;
2793 
2794 	/*
2795 	 * Dynamic ops may be freed, we must make sure that all
2796 	 * callers are done before leaving this function.
2797 	 * The same goes for freeing the per_cpu data of the per_cpu
2798 	 * ops.
2799 	 */
2800 	if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
2801 		/*
2802 		 * We need to do a hard force of sched synchronization.
2803 		 * This is because we use preempt_disable() to do RCU, but
2804 		 * the function tracers can be called where RCU is not watching
2805 		 * (like before user_exit()). We can not rely on the RCU
2806 		 * infrastructure to do the synchronization, thus we must do it
2807 		 * ourselves.
2808 		 */
2809 		schedule_on_each_cpu(ftrace_sync);
2810 
2811 		/*
2812 		 * When the kernel is preeptive, tasks can be preempted
2813 		 * while on a ftrace trampoline. Just scheduling a task on
2814 		 * a CPU is not good enough to flush them. Calling
2815 		 * synchornize_rcu_tasks() will wait for those tasks to
2816 		 * execute and either schedule voluntarily or enter user space.
2817 		 */
2818 		if (IS_ENABLED(CONFIG_PREEMPTION))
2819 			synchronize_rcu_tasks();
2820 
2821  free_ops:
2822 		arch_ftrace_trampoline_free(ops);
2823 	}
2824 
2825 	return 0;
2826 }
2827 
ftrace_startup_sysctl(void)2828 static void ftrace_startup_sysctl(void)
2829 {
2830 	int command;
2831 
2832 	if (unlikely(ftrace_disabled))
2833 		return;
2834 
2835 	/* Force update next time */
2836 	saved_ftrace_func = NULL;
2837 	/* ftrace_start_up is true if we want ftrace running */
2838 	if (ftrace_start_up) {
2839 		command = FTRACE_UPDATE_CALLS;
2840 		if (ftrace_graph_active)
2841 			command |= FTRACE_START_FUNC_RET;
2842 		ftrace_startup_enable(command);
2843 	}
2844 }
2845 
ftrace_shutdown_sysctl(void)2846 static void ftrace_shutdown_sysctl(void)
2847 {
2848 	int command;
2849 
2850 	if (unlikely(ftrace_disabled))
2851 		return;
2852 
2853 	/* ftrace_start_up is true if ftrace is running */
2854 	if (ftrace_start_up) {
2855 		command = FTRACE_DISABLE_CALLS;
2856 		if (ftrace_graph_active)
2857 			command |= FTRACE_STOP_FUNC_RET;
2858 		ftrace_run_update_code(command);
2859 	}
2860 }
2861 
2862 static u64		ftrace_update_time;
2863 unsigned long		ftrace_update_tot_cnt;
2864 
ops_traces_mod(struct ftrace_ops * ops)2865 static inline int ops_traces_mod(struct ftrace_ops *ops)
2866 {
2867 	/*
2868 	 * Filter_hash being empty will default to trace module.
2869 	 * But notrace hash requires a test of individual module functions.
2870 	 */
2871 	return ftrace_hash_empty(ops->func_hash->filter_hash) &&
2872 		ftrace_hash_empty(ops->func_hash->notrace_hash);
2873 }
2874 
2875 /*
2876  * Check if the current ops references the record.
2877  *
2878  * If the ops traces all functions, then it was already accounted for.
2879  * If the ops does not trace the current record function, skip it.
2880  * If the ops ignores the function via notrace filter, skip it.
2881  */
2882 static inline bool
ops_references_rec(struct ftrace_ops * ops,struct dyn_ftrace * rec)2883 ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2884 {
2885 	/* If ops isn't enabled, ignore it */
2886 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2887 		return false;
2888 
2889 	/* If ops traces all then it includes this function */
2890 	if (ops_traces_mod(ops))
2891 		return true;
2892 
2893 	/* The function must be in the filter */
2894 	if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
2895 	    !__ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
2896 		return false;
2897 
2898 	/* If in notrace hash, we ignore it too */
2899 	if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
2900 		return false;
2901 
2902 	return true;
2903 }
2904 
ftrace_update_code(struct module * mod,struct ftrace_page * new_pgs)2905 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2906 {
2907 	struct ftrace_page *pg;
2908 	struct dyn_ftrace *p;
2909 	u64 start, stop;
2910 	unsigned long update_cnt = 0;
2911 	unsigned long rec_flags = 0;
2912 	int i;
2913 
2914 	start = ftrace_now(raw_smp_processor_id());
2915 
2916 	/*
2917 	 * When a module is loaded, this function is called to convert
2918 	 * the calls to mcount in its text to nops, and also to create
2919 	 * an entry in the ftrace data. Now, if ftrace is activated
2920 	 * after this call, but before the module sets its text to
2921 	 * read-only, the modification of enabling ftrace can fail if
2922 	 * the read-only is done while ftrace is converting the calls.
2923 	 * To prevent this, the module's records are set as disabled
2924 	 * and will be enabled after the call to set the module's text
2925 	 * to read-only.
2926 	 */
2927 	if (mod)
2928 		rec_flags |= FTRACE_FL_DISABLED;
2929 
2930 	for (pg = new_pgs; pg; pg = pg->next) {
2931 
2932 		for (i = 0; i < pg->index; i++) {
2933 
2934 			/* If something went wrong, bail without enabling anything */
2935 			if (unlikely(ftrace_disabled))
2936 				return -1;
2937 
2938 			p = &pg->records[i];
2939 			p->flags = rec_flags;
2940 
2941 			/*
2942 			 * Do the initial record conversion from mcount jump
2943 			 * to the NOP instructions.
2944 			 */
2945 			if (!__is_defined(CC_USING_NOP_MCOUNT) &&
2946 			    !ftrace_code_disable(mod, p))
2947 				break;
2948 
2949 			update_cnt++;
2950 		}
2951 	}
2952 
2953 	stop = ftrace_now(raw_smp_processor_id());
2954 	ftrace_update_time = stop - start;
2955 	ftrace_update_tot_cnt += update_cnt;
2956 
2957 	return 0;
2958 }
2959 
ftrace_allocate_records(struct ftrace_page * pg,int count)2960 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2961 {
2962 	int order;
2963 	int cnt;
2964 
2965 	if (WARN_ON(!count))
2966 		return -EINVAL;
2967 
2968 	order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2969 
2970 	/*
2971 	 * We want to fill as much as possible. No more than a page
2972 	 * may be empty.
2973 	 */
2974 	while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2975 		order--;
2976 
2977  again:
2978 	pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2979 
2980 	if (!pg->records) {
2981 		/* if we can't allocate this size, try something smaller */
2982 		if (!order)
2983 			return -ENOMEM;
2984 		order >>= 1;
2985 		goto again;
2986 	}
2987 
2988 	cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2989 	pg->size = cnt;
2990 
2991 	if (cnt > count)
2992 		cnt = count;
2993 
2994 	return cnt;
2995 }
2996 
2997 static struct ftrace_page *
ftrace_allocate_pages(unsigned long num_to_init)2998 ftrace_allocate_pages(unsigned long num_to_init)
2999 {
3000 	struct ftrace_page *start_pg;
3001 	struct ftrace_page *pg;
3002 	int order;
3003 	int cnt;
3004 
3005 	if (!num_to_init)
3006 		return NULL;
3007 
3008 	start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3009 	if (!pg)
3010 		return NULL;
3011 
3012 	/*
3013 	 * Try to allocate as much as possible in one continues
3014 	 * location that fills in all of the space. We want to
3015 	 * waste as little space as possible.
3016 	 */
3017 	for (;;) {
3018 		cnt = ftrace_allocate_records(pg, num_to_init);
3019 		if (cnt < 0)
3020 			goto free_pages;
3021 
3022 		num_to_init -= cnt;
3023 		if (!num_to_init)
3024 			break;
3025 
3026 		pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3027 		if (!pg->next)
3028 			goto free_pages;
3029 
3030 		pg = pg->next;
3031 	}
3032 
3033 	return start_pg;
3034 
3035  free_pages:
3036 	pg = start_pg;
3037 	while (pg) {
3038 		order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3039 		free_pages((unsigned long)pg->records, order);
3040 		start_pg = pg->next;
3041 		kfree(pg);
3042 		pg = start_pg;
3043 	}
3044 	pr_info("ftrace: FAILED to allocate memory for functions\n");
3045 	return NULL;
3046 }
3047 
3048 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3049 
3050 struct ftrace_iterator {
3051 	loff_t				pos;
3052 	loff_t				func_pos;
3053 	loff_t				mod_pos;
3054 	struct ftrace_page		*pg;
3055 	struct dyn_ftrace		*func;
3056 	struct ftrace_func_probe	*probe;
3057 	struct ftrace_func_entry	*probe_entry;
3058 	struct trace_parser		parser;
3059 	struct ftrace_hash		*hash;
3060 	struct ftrace_ops		*ops;
3061 	struct trace_array		*tr;
3062 	struct list_head		*mod_list;
3063 	int				pidx;
3064 	int				idx;
3065 	unsigned			flags;
3066 };
3067 
3068 static void *
t_probe_next(struct seq_file * m,loff_t * pos)3069 t_probe_next(struct seq_file *m, loff_t *pos)
3070 {
3071 	struct ftrace_iterator *iter = m->private;
3072 	struct trace_array *tr = iter->ops->private;
3073 	struct list_head *func_probes;
3074 	struct ftrace_hash *hash;
3075 	struct list_head *next;
3076 	struct hlist_node *hnd = NULL;
3077 	struct hlist_head *hhd;
3078 	int size;
3079 
3080 	(*pos)++;
3081 	iter->pos = *pos;
3082 
3083 	if (!tr)
3084 		return NULL;
3085 
3086 	func_probes = &tr->func_probes;
3087 	if (list_empty(func_probes))
3088 		return NULL;
3089 
3090 	if (!iter->probe) {
3091 		next = func_probes->next;
3092 		iter->probe = list_entry(next, struct ftrace_func_probe, list);
3093 	}
3094 
3095 	if (iter->probe_entry)
3096 		hnd = &iter->probe_entry->hlist;
3097 
3098 	hash = iter->probe->ops.func_hash->filter_hash;
3099 
3100 	/*
3101 	 * A probe being registered may temporarily have an empty hash
3102 	 * and it's at the end of the func_probes list.
3103 	 */
3104 	if (!hash || hash == EMPTY_HASH)
3105 		return NULL;
3106 
3107 	size = 1 << hash->size_bits;
3108 
3109  retry:
3110 	if (iter->pidx >= size) {
3111 		if (iter->probe->list.next == func_probes)
3112 			return NULL;
3113 		next = iter->probe->list.next;
3114 		iter->probe = list_entry(next, struct ftrace_func_probe, list);
3115 		hash = iter->probe->ops.func_hash->filter_hash;
3116 		size = 1 << hash->size_bits;
3117 		iter->pidx = 0;
3118 	}
3119 
3120 	hhd = &hash->buckets[iter->pidx];
3121 
3122 	if (hlist_empty(hhd)) {
3123 		iter->pidx++;
3124 		hnd = NULL;
3125 		goto retry;
3126 	}
3127 
3128 	if (!hnd)
3129 		hnd = hhd->first;
3130 	else {
3131 		hnd = hnd->next;
3132 		if (!hnd) {
3133 			iter->pidx++;
3134 			goto retry;
3135 		}
3136 	}
3137 
3138 	if (WARN_ON_ONCE(!hnd))
3139 		return NULL;
3140 
3141 	iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
3142 
3143 	return iter;
3144 }
3145 
t_probe_start(struct seq_file * m,loff_t * pos)3146 static void *t_probe_start(struct seq_file *m, loff_t *pos)
3147 {
3148 	struct ftrace_iterator *iter = m->private;
3149 	void *p = NULL;
3150 	loff_t l;
3151 
3152 	if (!(iter->flags & FTRACE_ITER_DO_PROBES))
3153 		return NULL;
3154 
3155 	if (iter->mod_pos > *pos)
3156 		return NULL;
3157 
3158 	iter->probe = NULL;
3159 	iter->probe_entry = NULL;
3160 	iter->pidx = 0;
3161 	for (l = 0; l <= (*pos - iter->mod_pos); ) {
3162 		p = t_probe_next(m, &l);
3163 		if (!p)
3164 			break;
3165 	}
3166 	if (!p)
3167 		return NULL;
3168 
3169 	/* Only set this if we have an item */
3170 	iter->flags |= FTRACE_ITER_PROBE;
3171 
3172 	return iter;
3173 }
3174 
3175 static int
t_probe_show(struct seq_file * m,struct ftrace_iterator * iter)3176 t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
3177 {
3178 	struct ftrace_func_entry *probe_entry;
3179 	struct ftrace_probe_ops *probe_ops;
3180 	struct ftrace_func_probe *probe;
3181 
3182 	probe = iter->probe;
3183 	probe_entry = iter->probe_entry;
3184 
3185 	if (WARN_ON_ONCE(!probe || !probe_entry))
3186 		return -EIO;
3187 
3188 	probe_ops = probe->probe_ops;
3189 
3190 	if (probe_ops->print)
3191 		return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
3192 
3193 	seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
3194 		   (void *)probe_ops->func);
3195 
3196 	return 0;
3197 }
3198 
3199 static void *
t_mod_next(struct seq_file * m,loff_t * pos)3200 t_mod_next(struct seq_file *m, loff_t *pos)
3201 {
3202 	struct ftrace_iterator *iter = m->private;
3203 	struct trace_array *tr = iter->tr;
3204 
3205 	(*pos)++;
3206 	iter->pos = *pos;
3207 
3208 	iter->mod_list = iter->mod_list->next;
3209 
3210 	if (iter->mod_list == &tr->mod_trace ||
3211 	    iter->mod_list == &tr->mod_notrace) {
3212 		iter->flags &= ~FTRACE_ITER_MOD;
3213 		return NULL;
3214 	}
3215 
3216 	iter->mod_pos = *pos;
3217 
3218 	return iter;
3219 }
3220 
t_mod_start(struct seq_file * m,loff_t * pos)3221 static void *t_mod_start(struct seq_file *m, loff_t *pos)
3222 {
3223 	struct ftrace_iterator *iter = m->private;
3224 	void *p = NULL;
3225 	loff_t l;
3226 
3227 	if (iter->func_pos > *pos)
3228 		return NULL;
3229 
3230 	iter->mod_pos = iter->func_pos;
3231 
3232 	/* probes are only available if tr is set */
3233 	if (!iter->tr)
3234 		return NULL;
3235 
3236 	for (l = 0; l <= (*pos - iter->func_pos); ) {
3237 		p = t_mod_next(m, &l);
3238 		if (!p)
3239 			break;
3240 	}
3241 	if (!p) {
3242 		iter->flags &= ~FTRACE_ITER_MOD;
3243 		return t_probe_start(m, pos);
3244 	}
3245 
3246 	/* Only set this if we have an item */
3247 	iter->flags |= FTRACE_ITER_MOD;
3248 
3249 	return iter;
3250 }
3251 
3252 static int
t_mod_show(struct seq_file * m,struct ftrace_iterator * iter)3253 t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
3254 {
3255 	struct ftrace_mod_load *ftrace_mod;
3256 	struct trace_array *tr = iter->tr;
3257 
3258 	if (WARN_ON_ONCE(!iter->mod_list) ||
3259 			 iter->mod_list == &tr->mod_trace ||
3260 			 iter->mod_list == &tr->mod_notrace)
3261 		return -EIO;
3262 
3263 	ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
3264 
3265 	if (ftrace_mod->func)
3266 		seq_printf(m, "%s", ftrace_mod->func);
3267 	else
3268 		seq_putc(m, '*');
3269 
3270 	seq_printf(m, ":mod:%s\n", ftrace_mod->module);
3271 
3272 	return 0;
3273 }
3274 
3275 static void *
t_func_next(struct seq_file * m,loff_t * pos)3276 t_func_next(struct seq_file *m, loff_t *pos)
3277 {
3278 	struct ftrace_iterator *iter = m->private;
3279 	struct dyn_ftrace *rec = NULL;
3280 
3281 	(*pos)++;
3282 
3283  retry:
3284 	if (iter->idx >= iter->pg->index) {
3285 		if (iter->pg->next) {
3286 			iter->pg = iter->pg->next;
3287 			iter->idx = 0;
3288 			goto retry;
3289 		}
3290 	} else {
3291 		rec = &iter->pg->records[iter->idx++];
3292 		if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3293 		     !ftrace_lookup_ip(iter->hash, rec->ip)) ||
3294 
3295 		    ((iter->flags & FTRACE_ITER_ENABLED) &&
3296 		     !(rec->flags & FTRACE_FL_ENABLED))) {
3297 
3298 			rec = NULL;
3299 			goto retry;
3300 		}
3301 	}
3302 
3303 	if (!rec)
3304 		return NULL;
3305 
3306 	iter->pos = iter->func_pos = *pos;
3307 	iter->func = rec;
3308 
3309 	return iter;
3310 }
3311 
3312 static void *
t_next(struct seq_file * m,void * v,loff_t * pos)3313 t_next(struct seq_file *m, void *v, loff_t *pos)
3314 {
3315 	struct ftrace_iterator *iter = m->private;
3316 	loff_t l = *pos; /* t_probe_start() must use original pos */
3317 	void *ret;
3318 
3319 	if (unlikely(ftrace_disabled))
3320 		return NULL;
3321 
3322 	if (iter->flags & FTRACE_ITER_PROBE)
3323 		return t_probe_next(m, pos);
3324 
3325 	if (iter->flags & FTRACE_ITER_MOD)
3326 		return t_mod_next(m, pos);
3327 
3328 	if (iter->flags & FTRACE_ITER_PRINTALL) {
3329 		/* next must increment pos, and t_probe_start does not */
3330 		(*pos)++;
3331 		return t_mod_start(m, &l);
3332 	}
3333 
3334 	ret = t_func_next(m, pos);
3335 
3336 	if (!ret)
3337 		return t_mod_start(m, &l);
3338 
3339 	return ret;
3340 }
3341 
reset_iter_read(struct ftrace_iterator * iter)3342 static void reset_iter_read(struct ftrace_iterator *iter)
3343 {
3344 	iter->pos = 0;
3345 	iter->func_pos = 0;
3346 	iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
3347 }
3348 
t_start(struct seq_file * m,loff_t * pos)3349 static void *t_start(struct seq_file *m, loff_t *pos)
3350 {
3351 	struct ftrace_iterator *iter = m->private;
3352 	void *p = NULL;
3353 	loff_t l;
3354 
3355 	mutex_lock(&ftrace_lock);
3356 
3357 	if (unlikely(ftrace_disabled))
3358 		return NULL;
3359 
3360 	/*
3361 	 * If an lseek was done, then reset and start from beginning.
3362 	 */
3363 	if (*pos < iter->pos)
3364 		reset_iter_read(iter);
3365 
3366 	/*
3367 	 * For set_ftrace_filter reading, if we have the filter
3368 	 * off, we can short cut and just print out that all
3369 	 * functions are enabled.
3370 	 */
3371 	if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3372 	    ftrace_hash_empty(iter->hash)) {
3373 		iter->func_pos = 1; /* Account for the message */
3374 		if (*pos > 0)
3375 			return t_mod_start(m, pos);
3376 		iter->flags |= FTRACE_ITER_PRINTALL;
3377 		/* reset in case of seek/pread */
3378 		iter->flags &= ~FTRACE_ITER_PROBE;
3379 		return iter;
3380 	}
3381 
3382 	if (iter->flags & FTRACE_ITER_MOD)
3383 		return t_mod_start(m, pos);
3384 
3385 	/*
3386 	 * Unfortunately, we need to restart at ftrace_pages_start
3387 	 * every time we let go of the ftrace_mutex. This is because
3388 	 * those pointers can change without the lock.
3389 	 */
3390 	iter->pg = ftrace_pages_start;
3391 	iter->idx = 0;
3392 	for (l = 0; l <= *pos; ) {
3393 		p = t_func_next(m, &l);
3394 		if (!p)
3395 			break;
3396 	}
3397 
3398 	if (!p)
3399 		return t_mod_start(m, pos);
3400 
3401 	return iter;
3402 }
3403 
t_stop(struct seq_file * m,void * p)3404 static void t_stop(struct seq_file *m, void *p)
3405 {
3406 	mutex_unlock(&ftrace_lock);
3407 }
3408 
3409 void * __weak
arch_ftrace_trampoline_func(struct ftrace_ops * ops,struct dyn_ftrace * rec)3410 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3411 {
3412 	return NULL;
3413 }
3414 
add_trampoline_func(struct seq_file * m,struct ftrace_ops * ops,struct dyn_ftrace * rec)3415 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3416 				struct dyn_ftrace *rec)
3417 {
3418 	void *ptr;
3419 
3420 	ptr = arch_ftrace_trampoline_func(ops, rec);
3421 	if (ptr)
3422 		seq_printf(m, " ->%pS", ptr);
3423 }
3424 
t_show(struct seq_file * m,void * v)3425 static int t_show(struct seq_file *m, void *v)
3426 {
3427 	struct ftrace_iterator *iter = m->private;
3428 	struct dyn_ftrace *rec;
3429 
3430 	if (iter->flags & FTRACE_ITER_PROBE)
3431 		return t_probe_show(m, iter);
3432 
3433 	if (iter->flags & FTRACE_ITER_MOD)
3434 		return t_mod_show(m, iter);
3435 
3436 	if (iter->flags & FTRACE_ITER_PRINTALL) {
3437 		if (iter->flags & FTRACE_ITER_NOTRACE)
3438 			seq_puts(m, "#### no functions disabled ####\n");
3439 		else
3440 			seq_puts(m, "#### all functions enabled ####\n");
3441 		return 0;
3442 	}
3443 
3444 	rec = iter->func;
3445 
3446 	if (!rec)
3447 		return 0;
3448 
3449 	seq_printf(m, "%ps", (void *)rec->ip);
3450 	if (iter->flags & FTRACE_ITER_ENABLED) {
3451 		struct ftrace_ops *ops;
3452 
3453 		seq_printf(m, " (%ld)%s%s",
3454 			   ftrace_rec_count(rec),
3455 			   rec->flags & FTRACE_FL_REGS ? " R" : "  ",
3456 			   rec->flags & FTRACE_FL_IPMODIFY ? " I" : "  ");
3457 		if (rec->flags & FTRACE_FL_TRAMP_EN) {
3458 			ops = ftrace_find_tramp_ops_any(rec);
3459 			if (ops) {
3460 				do {
3461 					seq_printf(m, "\ttramp: %pS (%pS)",
3462 						   (void *)ops->trampoline,
3463 						   (void *)ops->func);
3464 					add_trampoline_func(m, ops, rec);
3465 					ops = ftrace_find_tramp_ops_next(rec, ops);
3466 				} while (ops);
3467 			} else
3468 				seq_puts(m, "\ttramp: ERROR!");
3469 		} else {
3470 			add_trampoline_func(m, NULL, rec);
3471 		}
3472 	}
3473 
3474 	seq_putc(m, '\n');
3475 
3476 	return 0;
3477 }
3478 
3479 static const struct seq_operations show_ftrace_seq_ops = {
3480 	.start = t_start,
3481 	.next = t_next,
3482 	.stop = t_stop,
3483 	.show = t_show,
3484 };
3485 
3486 static int
ftrace_avail_open(struct inode * inode,struct file * file)3487 ftrace_avail_open(struct inode *inode, struct file *file)
3488 {
3489 	struct ftrace_iterator *iter;
3490 	int ret;
3491 
3492 	ret = security_locked_down(LOCKDOWN_TRACEFS);
3493 	if (ret)
3494 		return ret;
3495 
3496 	if (unlikely(ftrace_disabled))
3497 		return -ENODEV;
3498 
3499 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3500 	if (!iter)
3501 		return -ENOMEM;
3502 
3503 	iter->pg = ftrace_pages_start;
3504 	iter->ops = &global_ops;
3505 
3506 	return 0;
3507 }
3508 
3509 static int
ftrace_enabled_open(struct inode * inode,struct file * file)3510 ftrace_enabled_open(struct inode *inode, struct file *file)
3511 {
3512 	struct ftrace_iterator *iter;
3513 
3514 	/*
3515 	 * This shows us what functions are currently being
3516 	 * traced and by what. Not sure if we want lockdown
3517 	 * to hide such critical information for an admin.
3518 	 * Although, perhaps it can show information we don't
3519 	 * want people to see, but if something is tracing
3520 	 * something, we probably want to know about it.
3521 	 */
3522 
3523 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3524 	if (!iter)
3525 		return -ENOMEM;
3526 
3527 	iter->pg = ftrace_pages_start;
3528 	iter->flags = FTRACE_ITER_ENABLED;
3529 	iter->ops = &global_ops;
3530 
3531 	return 0;
3532 }
3533 
3534 /**
3535  * ftrace_regex_open - initialize function tracer filter files
3536  * @ops: The ftrace_ops that hold the hash filters
3537  * @flag: The type of filter to process
3538  * @inode: The inode, usually passed in to your open routine
3539  * @file: The file, usually passed in to your open routine
3540  *
3541  * ftrace_regex_open() initializes the filter files for the
3542  * @ops. Depending on @flag it may process the filter hash or
3543  * the notrace hash of @ops. With this called from the open
3544  * routine, you can use ftrace_filter_write() for the write
3545  * routine if @flag has FTRACE_ITER_FILTER set, or
3546  * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3547  * tracing_lseek() should be used as the lseek routine, and
3548  * release must call ftrace_regex_release().
3549  */
3550 int
ftrace_regex_open(struct ftrace_ops * ops,int flag,struct inode * inode,struct file * file)3551 ftrace_regex_open(struct ftrace_ops *ops, int flag,
3552 		  struct inode *inode, struct file *file)
3553 {
3554 	struct ftrace_iterator *iter;
3555 	struct ftrace_hash *hash;
3556 	struct list_head *mod_head;
3557 	struct trace_array *tr = ops->private;
3558 	int ret = -ENOMEM;
3559 
3560 	ftrace_ops_init(ops);
3561 
3562 	if (unlikely(ftrace_disabled))
3563 		return -ENODEV;
3564 
3565 	if (tracing_check_open_get_tr(tr))
3566 		return -ENODEV;
3567 
3568 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3569 	if (!iter)
3570 		goto out;
3571 
3572 	if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX))
3573 		goto out;
3574 
3575 	iter->ops = ops;
3576 	iter->flags = flag;
3577 	iter->tr = tr;
3578 
3579 	mutex_lock(&ops->func_hash->regex_lock);
3580 
3581 	if (flag & FTRACE_ITER_NOTRACE) {
3582 		hash = ops->func_hash->notrace_hash;
3583 		mod_head = tr ? &tr->mod_notrace : NULL;
3584 	} else {
3585 		hash = ops->func_hash->filter_hash;
3586 		mod_head = tr ? &tr->mod_trace : NULL;
3587 	}
3588 
3589 	iter->mod_list = mod_head;
3590 
3591 	if (file->f_mode & FMODE_WRITE) {
3592 		const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3593 
3594 		if (file->f_flags & O_TRUNC) {
3595 			iter->hash = alloc_ftrace_hash(size_bits);
3596 			clear_ftrace_mod_list(mod_head);
3597 	        } else {
3598 			iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3599 		}
3600 
3601 		if (!iter->hash) {
3602 			trace_parser_put(&iter->parser);
3603 			goto out_unlock;
3604 		}
3605 	} else
3606 		iter->hash = hash;
3607 
3608 	ret = 0;
3609 
3610 	if (file->f_mode & FMODE_READ) {
3611 		iter->pg = ftrace_pages_start;
3612 
3613 		ret = seq_open(file, &show_ftrace_seq_ops);
3614 		if (!ret) {
3615 			struct seq_file *m = file->private_data;
3616 			m->private = iter;
3617 		} else {
3618 			/* Failed */
3619 			free_ftrace_hash(iter->hash);
3620 			trace_parser_put(&iter->parser);
3621 		}
3622 	} else
3623 		file->private_data = iter;
3624 
3625  out_unlock:
3626 	mutex_unlock(&ops->func_hash->regex_lock);
3627 
3628  out:
3629 	if (ret) {
3630 		kfree(iter);
3631 		if (tr)
3632 			trace_array_put(tr);
3633 	}
3634 
3635 	return ret;
3636 }
3637 
3638 static int
ftrace_filter_open(struct inode * inode,struct file * file)3639 ftrace_filter_open(struct inode *inode, struct file *file)
3640 {
3641 	struct ftrace_ops *ops = inode->i_private;
3642 
3643 	/* Checks for tracefs lockdown */
3644 	return ftrace_regex_open(ops,
3645 			FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
3646 			inode, file);
3647 }
3648 
3649 static int
ftrace_notrace_open(struct inode * inode,struct file * file)3650 ftrace_notrace_open(struct inode *inode, struct file *file)
3651 {
3652 	struct ftrace_ops *ops = inode->i_private;
3653 
3654 	/* Checks for tracefs lockdown */
3655 	return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3656 				 inode, file);
3657 }
3658 
3659 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
3660 struct ftrace_glob {
3661 	char *search;
3662 	unsigned len;
3663 	int type;
3664 };
3665 
3666 /*
3667  * If symbols in an architecture don't correspond exactly to the user-visible
3668  * name of what they represent, it is possible to define this function to
3669  * perform the necessary adjustments.
3670 */
arch_ftrace_match_adjust(char * str,const char * search)3671 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
3672 {
3673 	return str;
3674 }
3675 
ftrace_match(char * str,struct ftrace_glob * g)3676 static int ftrace_match(char *str, struct ftrace_glob *g)
3677 {
3678 	int matched = 0;
3679 	int slen;
3680 
3681 	str = arch_ftrace_match_adjust(str, g->search);
3682 
3683 	switch (g->type) {
3684 	case MATCH_FULL:
3685 		if (strcmp(str, g->search) == 0)
3686 			matched = 1;
3687 		break;
3688 	case MATCH_FRONT_ONLY:
3689 		if (strncmp(str, g->search, g->len) == 0)
3690 			matched = 1;
3691 		break;
3692 	case MATCH_MIDDLE_ONLY:
3693 		if (strstr(str, g->search))
3694 			matched = 1;
3695 		break;
3696 	case MATCH_END_ONLY:
3697 		slen = strlen(str);
3698 		if (slen >= g->len &&
3699 		    memcmp(str + slen - g->len, g->search, g->len) == 0)
3700 			matched = 1;
3701 		break;
3702 	case MATCH_GLOB:
3703 		if (glob_match(g->search, str))
3704 			matched = 1;
3705 		break;
3706 	}
3707 
3708 	return matched;
3709 }
3710 
3711 static int
enter_record(struct ftrace_hash * hash,struct dyn_ftrace * rec,int clear_filter)3712 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
3713 {
3714 	struct ftrace_func_entry *entry;
3715 	int ret = 0;
3716 
3717 	entry = ftrace_lookup_ip(hash, rec->ip);
3718 	if (clear_filter) {
3719 		/* Do nothing if it doesn't exist */
3720 		if (!entry)
3721 			return 0;
3722 
3723 		free_hash_entry(hash, entry);
3724 	} else {
3725 		/* Do nothing if it exists */
3726 		if (entry)
3727 			return 0;
3728 
3729 		ret = add_hash_entry(hash, rec->ip);
3730 	}
3731 	return ret;
3732 }
3733 
3734 static int
add_rec_by_index(struct ftrace_hash * hash,struct ftrace_glob * func_g,int clear_filter)3735 add_rec_by_index(struct ftrace_hash *hash, struct ftrace_glob *func_g,
3736 		 int clear_filter)
3737 {
3738 	long index = simple_strtoul(func_g->search, NULL, 0);
3739 	struct ftrace_page *pg;
3740 	struct dyn_ftrace *rec;
3741 
3742 	/* The index starts at 1 */
3743 	if (--index < 0)
3744 		return 0;
3745 
3746 	do_for_each_ftrace_rec(pg, rec) {
3747 		if (pg->index <= index) {
3748 			index -= pg->index;
3749 			/* this is a double loop, break goes to the next page */
3750 			break;
3751 		}
3752 		rec = &pg->records[index];
3753 		enter_record(hash, rec, clear_filter);
3754 		return 1;
3755 	} while_for_each_ftrace_rec();
3756 	return 0;
3757 }
3758 
3759 static int
ftrace_match_record(struct dyn_ftrace * rec,struct ftrace_glob * func_g,struct ftrace_glob * mod_g,int exclude_mod)3760 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
3761 		struct ftrace_glob *mod_g, int exclude_mod)
3762 {
3763 	char str[KSYM_SYMBOL_LEN];
3764 	char *modname;
3765 
3766 	kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3767 
3768 	if (mod_g) {
3769 		int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
3770 
3771 		/* blank module name to match all modules */
3772 		if (!mod_g->len) {
3773 			/* blank module globbing: modname xor exclude_mod */
3774 			if (!exclude_mod != !modname)
3775 				goto func_match;
3776 			return 0;
3777 		}
3778 
3779 		/*
3780 		 * exclude_mod is set to trace everything but the given
3781 		 * module. If it is set and the module matches, then
3782 		 * return 0. If it is not set, and the module doesn't match
3783 		 * also return 0. Otherwise, check the function to see if
3784 		 * that matches.
3785 		 */
3786 		if (!mod_matches == !exclude_mod)
3787 			return 0;
3788 func_match:
3789 		/* blank search means to match all funcs in the mod */
3790 		if (!func_g->len)
3791 			return 1;
3792 	}
3793 
3794 	return ftrace_match(str, func_g);
3795 }
3796 
3797 static int
match_records(struct ftrace_hash * hash,char * func,int len,char * mod)3798 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
3799 {
3800 	struct ftrace_page *pg;
3801 	struct dyn_ftrace *rec;
3802 	struct ftrace_glob func_g = { .type = MATCH_FULL };
3803 	struct ftrace_glob mod_g = { .type = MATCH_FULL };
3804 	struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
3805 	int exclude_mod = 0;
3806 	int found = 0;
3807 	int ret;
3808 	int clear_filter = 0;
3809 
3810 	if (func) {
3811 		func_g.type = filter_parse_regex(func, len, &func_g.search,
3812 						 &clear_filter);
3813 		func_g.len = strlen(func_g.search);
3814 	}
3815 
3816 	if (mod) {
3817 		mod_g.type = filter_parse_regex(mod, strlen(mod),
3818 				&mod_g.search, &exclude_mod);
3819 		mod_g.len = strlen(mod_g.search);
3820 	}
3821 
3822 	mutex_lock(&ftrace_lock);
3823 
3824 	if (unlikely(ftrace_disabled))
3825 		goto out_unlock;
3826 
3827 	if (func_g.type == MATCH_INDEX) {
3828 		found = add_rec_by_index(hash, &func_g, clear_filter);
3829 		goto out_unlock;
3830 	}
3831 
3832 	do_for_each_ftrace_rec(pg, rec) {
3833 
3834 		if (rec->flags & FTRACE_FL_DISABLED)
3835 			continue;
3836 
3837 		if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
3838 			ret = enter_record(hash, rec, clear_filter);
3839 			if (ret < 0) {
3840 				found = ret;
3841 				goto out_unlock;
3842 			}
3843 			found = 1;
3844 		}
3845 	} while_for_each_ftrace_rec();
3846  out_unlock:
3847 	mutex_unlock(&ftrace_lock);
3848 
3849 	return found;
3850 }
3851 
3852 static int
ftrace_match_records(struct ftrace_hash * hash,char * buff,int len)3853 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
3854 {
3855 	return match_records(hash, buff, len, NULL);
3856 }
3857 
ftrace_ops_update_code(struct ftrace_ops * ops,struct ftrace_ops_hash * old_hash)3858 static void ftrace_ops_update_code(struct ftrace_ops *ops,
3859 				   struct ftrace_ops_hash *old_hash)
3860 {
3861 	struct ftrace_ops *op;
3862 
3863 	if (!ftrace_enabled)
3864 		return;
3865 
3866 	if (ops->flags & FTRACE_OPS_FL_ENABLED) {
3867 		ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
3868 		return;
3869 	}
3870 
3871 	/*
3872 	 * If this is the shared global_ops filter, then we need to
3873 	 * check if there is another ops that shares it, is enabled.
3874 	 * If so, we still need to run the modify code.
3875 	 */
3876 	if (ops->func_hash != &global_ops.local_hash)
3877 		return;
3878 
3879 	do_for_each_ftrace_op(op, ftrace_ops_list) {
3880 		if (op->func_hash == &global_ops.local_hash &&
3881 		    op->flags & FTRACE_OPS_FL_ENABLED) {
3882 			ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
3883 			/* Only need to do this once */
3884 			return;
3885 		}
3886 	} while_for_each_ftrace_op(op);
3887 }
3888 
ftrace_hash_move_and_update_ops(struct ftrace_ops * ops,struct ftrace_hash ** orig_hash,struct ftrace_hash * hash,int enable)3889 static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
3890 					   struct ftrace_hash **orig_hash,
3891 					   struct ftrace_hash *hash,
3892 					   int enable)
3893 {
3894 	struct ftrace_ops_hash old_hash_ops;
3895 	struct ftrace_hash *old_hash;
3896 	int ret;
3897 
3898 	old_hash = *orig_hash;
3899 	old_hash_ops.filter_hash = ops->func_hash->filter_hash;
3900 	old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
3901 	ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3902 	if (!ret) {
3903 		ftrace_ops_update_code(ops, &old_hash_ops);
3904 		free_ftrace_hash_rcu(old_hash);
3905 	}
3906 	return ret;
3907 }
3908 
module_exists(const char * module)3909 static bool module_exists(const char *module)
3910 {
3911 	/* All modules have the symbol __this_module */
3912 	static const char this_mod[] = "__this_module";
3913 	char modname[MAX_PARAM_PREFIX_LEN + sizeof(this_mod) + 2];
3914 	unsigned long val;
3915 	int n;
3916 
3917 	n = snprintf(modname, sizeof(modname), "%s:%s", module, this_mod);
3918 
3919 	if (n > sizeof(modname) - 1)
3920 		return false;
3921 
3922 	val = module_kallsyms_lookup_name(modname);
3923 	return val != 0;
3924 }
3925 
cache_mod(struct trace_array * tr,const char * func,char * module,int enable)3926 static int cache_mod(struct trace_array *tr,
3927 		     const char *func, char *module, int enable)
3928 {
3929 	struct ftrace_mod_load *ftrace_mod, *n;
3930 	struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
3931 	int ret;
3932 
3933 	mutex_lock(&ftrace_lock);
3934 
3935 	/* We do not cache inverse filters */
3936 	if (func[0] == '!') {
3937 		func++;
3938 		ret = -EINVAL;
3939 
3940 		/* Look to remove this hash */
3941 		list_for_each_entry_safe(ftrace_mod, n, head, list) {
3942 			if (strcmp(ftrace_mod->module, module) != 0)
3943 				continue;
3944 
3945 			/* no func matches all */
3946 			if (strcmp(func, "*") == 0 ||
3947 			    (ftrace_mod->func &&
3948 			     strcmp(ftrace_mod->func, func) == 0)) {
3949 				ret = 0;
3950 				free_ftrace_mod(ftrace_mod);
3951 				continue;
3952 			}
3953 		}
3954 		goto out;
3955 	}
3956 
3957 	ret = -EINVAL;
3958 	/* We only care about modules that have not been loaded yet */
3959 	if (module_exists(module))
3960 		goto out;
3961 
3962 	/* Save this string off, and execute it when the module is loaded */
3963 	ret = ftrace_add_mod(tr, func, module, enable);
3964  out:
3965 	mutex_unlock(&ftrace_lock);
3966 
3967 	return ret;
3968 }
3969 
3970 static int
3971 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3972 		 int reset, int enable);
3973 
3974 #ifdef CONFIG_MODULES
process_mod_list(struct list_head * head,struct ftrace_ops * ops,char * mod,bool enable)3975 static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
3976 			     char *mod, bool enable)
3977 {
3978 	struct ftrace_mod_load *ftrace_mod, *n;
3979 	struct ftrace_hash **orig_hash, *new_hash;
3980 	LIST_HEAD(process_mods);
3981 	char *func;
3982 	int ret;
3983 
3984 	mutex_lock(&ops->func_hash->regex_lock);
3985 
3986 	if (enable)
3987 		orig_hash = &ops->func_hash->filter_hash;
3988 	else
3989 		orig_hash = &ops->func_hash->notrace_hash;
3990 
3991 	new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
3992 					      *orig_hash);
3993 	if (!new_hash)
3994 		goto out; /* warn? */
3995 
3996 	mutex_lock(&ftrace_lock);
3997 
3998 	list_for_each_entry_safe(ftrace_mod, n, head, list) {
3999 
4000 		if (strcmp(ftrace_mod->module, mod) != 0)
4001 			continue;
4002 
4003 		if (ftrace_mod->func)
4004 			func = kstrdup(ftrace_mod->func, GFP_KERNEL);
4005 		else
4006 			func = kstrdup("*", GFP_KERNEL);
4007 
4008 		if (!func) /* warn? */
4009 			continue;
4010 
4011 		list_del(&ftrace_mod->list);
4012 		list_add(&ftrace_mod->list, &process_mods);
4013 
4014 		/* Use the newly allocated func, as it may be "*" */
4015 		kfree(ftrace_mod->func);
4016 		ftrace_mod->func = func;
4017 	}
4018 
4019 	mutex_unlock(&ftrace_lock);
4020 
4021 	list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
4022 
4023 		func = ftrace_mod->func;
4024 
4025 		/* Grabs ftrace_lock, which is why we have this extra step */
4026 		match_records(new_hash, func, strlen(func), mod);
4027 		free_ftrace_mod(ftrace_mod);
4028 	}
4029 
4030 	if (enable && list_empty(head))
4031 		new_hash->flags &= ~FTRACE_HASH_FL_MOD;
4032 
4033 	mutex_lock(&ftrace_lock);
4034 
4035 	ret = ftrace_hash_move_and_update_ops(ops, orig_hash,
4036 					      new_hash, enable);
4037 	mutex_unlock(&ftrace_lock);
4038 
4039  out:
4040 	mutex_unlock(&ops->func_hash->regex_lock);
4041 
4042 	free_ftrace_hash(new_hash);
4043 }
4044 
process_cached_mods(const char * mod_name)4045 static void process_cached_mods(const char *mod_name)
4046 {
4047 	struct trace_array *tr;
4048 	char *mod;
4049 
4050 	mod = kstrdup(mod_name, GFP_KERNEL);
4051 	if (!mod)
4052 		return;
4053 
4054 	mutex_lock(&trace_types_lock);
4055 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
4056 		if (!list_empty(&tr->mod_trace))
4057 			process_mod_list(&tr->mod_trace, tr->ops, mod, true);
4058 		if (!list_empty(&tr->mod_notrace))
4059 			process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
4060 	}
4061 	mutex_unlock(&trace_types_lock);
4062 
4063 	kfree(mod);
4064 }
4065 #endif
4066 
4067 /*
4068  * We register the module command as a template to show others how
4069  * to register the a command as well.
4070  */
4071 
4072 static int
ftrace_mod_callback(struct trace_array * tr,struct ftrace_hash * hash,char * func_orig,char * cmd,char * module,int enable)4073 ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
4074 		    char *func_orig, char *cmd, char *module, int enable)
4075 {
4076 	char *func;
4077 	int ret;
4078 
4079 	/* match_records() modifies func, and we need the original */
4080 	func = kstrdup(func_orig, GFP_KERNEL);
4081 	if (!func)
4082 		return -ENOMEM;
4083 
4084 	/*
4085 	 * cmd == 'mod' because we only registered this func
4086 	 * for the 'mod' ftrace_func_command.
4087 	 * But if you register one func with multiple commands,
4088 	 * you can tell which command was used by the cmd
4089 	 * parameter.
4090 	 */
4091 	ret = match_records(hash, func, strlen(func), module);
4092 	kfree(func);
4093 
4094 	if (!ret)
4095 		return cache_mod(tr, func_orig, module, enable);
4096 	if (ret < 0)
4097 		return ret;
4098 	return 0;
4099 }
4100 
4101 static struct ftrace_func_command ftrace_mod_cmd = {
4102 	.name			= "mod",
4103 	.func			= ftrace_mod_callback,
4104 };
4105 
ftrace_mod_cmd_init(void)4106 static int __init ftrace_mod_cmd_init(void)
4107 {
4108 	return register_ftrace_command(&ftrace_mod_cmd);
4109 }
4110 core_initcall(ftrace_mod_cmd_init);
4111 
function_trace_probe_call(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct pt_regs * pt_regs)4112 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
4113 				      struct ftrace_ops *op, struct pt_regs *pt_regs)
4114 {
4115 	struct ftrace_probe_ops *probe_ops;
4116 	struct ftrace_func_probe *probe;
4117 
4118 	probe = container_of(op, struct ftrace_func_probe, ops);
4119 	probe_ops = probe->probe_ops;
4120 
4121 	/*
4122 	 * Disable preemption for these calls to prevent a RCU grace
4123 	 * period. This syncs the hash iteration and freeing of items
4124 	 * on the hash. rcu_read_lock is too dangerous here.
4125 	 */
4126 	preempt_disable_notrace();
4127 	probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
4128 	preempt_enable_notrace();
4129 }
4130 
4131 struct ftrace_func_map {
4132 	struct ftrace_func_entry	entry;
4133 	void				*data;
4134 };
4135 
4136 struct ftrace_func_mapper {
4137 	struct ftrace_hash		hash;
4138 };
4139 
4140 /**
4141  * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
4142  *
4143  * Returns a ftrace_func_mapper descriptor that can be used to map ips to data.
4144  */
allocate_ftrace_func_mapper(void)4145 struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
4146 {
4147 	struct ftrace_hash *hash;
4148 
4149 	/*
4150 	 * The mapper is simply a ftrace_hash, but since the entries
4151 	 * in the hash are not ftrace_func_entry type, we define it
4152 	 * as a separate structure.
4153 	 */
4154 	hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4155 	return (struct ftrace_func_mapper *)hash;
4156 }
4157 
4158 /**
4159  * ftrace_func_mapper_find_ip - Find some data mapped to an ip
4160  * @mapper: The mapper that has the ip maps
4161  * @ip: the instruction pointer to find the data for
4162  *
4163  * Returns the data mapped to @ip if found otherwise NULL. The return
4164  * is actually the address of the mapper data pointer. The address is
4165  * returned for use cases where the data is no bigger than a long, and
4166  * the user can use the data pointer as its data instead of having to
4167  * allocate more memory for the reference.
4168  */
ftrace_func_mapper_find_ip(struct ftrace_func_mapper * mapper,unsigned long ip)4169 void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
4170 				  unsigned long ip)
4171 {
4172 	struct ftrace_func_entry *entry;
4173 	struct ftrace_func_map *map;
4174 
4175 	entry = ftrace_lookup_ip(&mapper->hash, ip);
4176 	if (!entry)
4177 		return NULL;
4178 
4179 	map = (struct ftrace_func_map *)entry;
4180 	return &map->data;
4181 }
4182 
4183 /**
4184  * ftrace_func_mapper_add_ip - Map some data to an ip
4185  * @mapper: The mapper that has the ip maps
4186  * @ip: The instruction pointer address to map @data to
4187  * @data: The data to map to @ip
4188  *
4189  * Returns 0 on succes otherwise an error.
4190  */
ftrace_func_mapper_add_ip(struct ftrace_func_mapper * mapper,unsigned long ip,void * data)4191 int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
4192 			      unsigned long ip, void *data)
4193 {
4194 	struct ftrace_func_entry *entry;
4195 	struct ftrace_func_map *map;
4196 
4197 	entry = ftrace_lookup_ip(&mapper->hash, ip);
4198 	if (entry)
4199 		return -EBUSY;
4200 
4201 	map = kmalloc(sizeof(*map), GFP_KERNEL);
4202 	if (!map)
4203 		return -ENOMEM;
4204 
4205 	map->entry.ip = ip;
4206 	map->data = data;
4207 
4208 	__add_hash_entry(&mapper->hash, &map->entry);
4209 
4210 	return 0;
4211 }
4212 
4213 /**
4214  * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
4215  * @mapper: The mapper that has the ip maps
4216  * @ip: The instruction pointer address to remove the data from
4217  *
4218  * Returns the data if it is found, otherwise NULL.
4219  * Note, if the data pointer is used as the data itself, (see
4220  * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
4221  * if the data pointer was set to zero.
4222  */
ftrace_func_mapper_remove_ip(struct ftrace_func_mapper * mapper,unsigned long ip)4223 void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
4224 				   unsigned long ip)
4225 {
4226 	struct ftrace_func_entry *entry;
4227 	struct ftrace_func_map *map;
4228 	void *data;
4229 
4230 	entry = ftrace_lookup_ip(&mapper->hash, ip);
4231 	if (!entry)
4232 		return NULL;
4233 
4234 	map = (struct ftrace_func_map *)entry;
4235 	data = map->data;
4236 
4237 	remove_hash_entry(&mapper->hash, entry);
4238 	kfree(entry);
4239 
4240 	return data;
4241 }
4242 
4243 /**
4244  * free_ftrace_func_mapper - free a mapping of ips and data
4245  * @mapper: The mapper that has the ip maps
4246  * @free_func: A function to be called on each data item.
4247  *
4248  * This is used to free the function mapper. The @free_func is optional
4249  * and can be used if the data needs to be freed as well.
4250  */
free_ftrace_func_mapper(struct ftrace_func_mapper * mapper,ftrace_mapper_func free_func)4251 void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
4252 			     ftrace_mapper_func free_func)
4253 {
4254 	struct ftrace_func_entry *entry;
4255 	struct ftrace_func_map *map;
4256 	struct hlist_head *hhd;
4257 	int size, i;
4258 
4259 	if (!mapper)
4260 		return;
4261 
4262 	if (free_func && mapper->hash.count) {
4263 		size = 1 << mapper->hash.size_bits;
4264 		for (i = 0; i < size; i++) {
4265 			hhd = &mapper->hash.buckets[i];
4266 			hlist_for_each_entry(entry, hhd, hlist) {
4267 				map = (struct ftrace_func_map *)entry;
4268 				free_func(map);
4269 			}
4270 		}
4271 	}
4272 	free_ftrace_hash(&mapper->hash);
4273 }
4274 
release_probe(struct ftrace_func_probe * probe)4275 static void release_probe(struct ftrace_func_probe *probe)
4276 {
4277 	struct ftrace_probe_ops *probe_ops;
4278 
4279 	mutex_lock(&ftrace_lock);
4280 
4281 	WARN_ON(probe->ref <= 0);
4282 
4283 	/* Subtract the ref that was used to protect this instance */
4284 	probe->ref--;
4285 
4286 	if (!probe->ref) {
4287 		probe_ops = probe->probe_ops;
4288 		/*
4289 		 * Sending zero as ip tells probe_ops to free
4290 		 * the probe->data itself
4291 		 */
4292 		if (probe_ops->free)
4293 			probe_ops->free(probe_ops, probe->tr, 0, probe->data);
4294 		list_del(&probe->list);
4295 		kfree(probe);
4296 	}
4297 	mutex_unlock(&ftrace_lock);
4298 }
4299 
acquire_probe_locked(struct ftrace_func_probe * probe)4300 static void acquire_probe_locked(struct ftrace_func_probe *probe)
4301 {
4302 	/*
4303 	 * Add one ref to keep it from being freed when releasing the
4304 	 * ftrace_lock mutex.
4305 	 */
4306 	probe->ref++;
4307 }
4308 
4309 int
register_ftrace_function_probe(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops,void * data)4310 register_ftrace_function_probe(char *glob, struct trace_array *tr,
4311 			       struct ftrace_probe_ops *probe_ops,
4312 			       void *data)
4313 {
4314 	struct ftrace_func_entry *entry;
4315 	struct ftrace_func_probe *probe;
4316 	struct ftrace_hash **orig_hash;
4317 	struct ftrace_hash *old_hash;
4318 	struct ftrace_hash *hash;
4319 	int count = 0;
4320 	int size;
4321 	int ret;
4322 	int i;
4323 
4324 	if (WARN_ON(!tr))
4325 		return -EINVAL;
4326 
4327 	/* We do not support '!' for function probes */
4328 	if (WARN_ON(glob[0] == '!'))
4329 		return -EINVAL;
4330 
4331 
4332 	mutex_lock(&ftrace_lock);
4333 	/* Check if the probe_ops is already registered */
4334 	list_for_each_entry(probe, &tr->func_probes, list) {
4335 		if (probe->probe_ops == probe_ops)
4336 			break;
4337 	}
4338 	if (&probe->list == &tr->func_probes) {
4339 		probe = kzalloc(sizeof(*probe), GFP_KERNEL);
4340 		if (!probe) {
4341 			mutex_unlock(&ftrace_lock);
4342 			return -ENOMEM;
4343 		}
4344 		probe->probe_ops = probe_ops;
4345 		probe->ops.func = function_trace_probe_call;
4346 		probe->tr = tr;
4347 		ftrace_ops_init(&probe->ops);
4348 		list_add(&probe->list, &tr->func_probes);
4349 	}
4350 
4351 	acquire_probe_locked(probe);
4352 
4353 	mutex_unlock(&ftrace_lock);
4354 
4355 	/*
4356 	 * Note, there's a small window here that the func_hash->filter_hash
4357 	 * may be NULL or empty. Need to be carefule when reading the loop.
4358 	 */
4359 	mutex_lock(&probe->ops.func_hash->regex_lock);
4360 
4361 	orig_hash = &probe->ops.func_hash->filter_hash;
4362 	old_hash = *orig_hash;
4363 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4364 
4365 	if (!hash) {
4366 		ret = -ENOMEM;
4367 		goto out;
4368 	}
4369 
4370 	ret = ftrace_match_records(hash, glob, strlen(glob));
4371 
4372 	/* Nothing found? */
4373 	if (!ret)
4374 		ret = -EINVAL;
4375 
4376 	if (ret < 0)
4377 		goto out;
4378 
4379 	size = 1 << hash->size_bits;
4380 	for (i = 0; i < size; i++) {
4381 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4382 			if (ftrace_lookup_ip(old_hash, entry->ip))
4383 				continue;
4384 			/*
4385 			 * The caller might want to do something special
4386 			 * for each function we find. We call the callback
4387 			 * to give the caller an opportunity to do so.
4388 			 */
4389 			if (probe_ops->init) {
4390 				ret = probe_ops->init(probe_ops, tr,
4391 						      entry->ip, data,
4392 						      &probe->data);
4393 				if (ret < 0) {
4394 					if (probe_ops->free && count)
4395 						probe_ops->free(probe_ops, tr,
4396 								0, probe->data);
4397 					probe->data = NULL;
4398 					goto out;
4399 				}
4400 			}
4401 			count++;
4402 		}
4403 	}
4404 
4405 	mutex_lock(&ftrace_lock);
4406 
4407 	if (!count) {
4408 		/* Nothing was added? */
4409 		ret = -EINVAL;
4410 		goto out_unlock;
4411 	}
4412 
4413 	ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4414 					      hash, 1);
4415 	if (ret < 0)
4416 		goto err_unlock;
4417 
4418 	/* One ref for each new function traced */
4419 	probe->ref += count;
4420 
4421 	if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
4422 		ret = ftrace_startup(&probe->ops, 0);
4423 
4424  out_unlock:
4425 	mutex_unlock(&ftrace_lock);
4426 
4427 	if (!ret)
4428 		ret = count;
4429  out:
4430 	mutex_unlock(&probe->ops.func_hash->regex_lock);
4431 	free_ftrace_hash(hash);
4432 
4433 	release_probe(probe);
4434 
4435 	return ret;
4436 
4437  err_unlock:
4438 	if (!probe_ops->free || !count)
4439 		goto out_unlock;
4440 
4441 	/* Failed to do the move, need to call the free functions */
4442 	for (i = 0; i < size; i++) {
4443 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4444 			if (ftrace_lookup_ip(old_hash, entry->ip))
4445 				continue;
4446 			probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4447 		}
4448 	}
4449 	goto out_unlock;
4450 }
4451 
4452 int
unregister_ftrace_function_probe_func(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops)4453 unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
4454 				      struct ftrace_probe_ops *probe_ops)
4455 {
4456 	struct ftrace_ops_hash old_hash_ops;
4457 	struct ftrace_func_entry *entry;
4458 	struct ftrace_func_probe *probe;
4459 	struct ftrace_glob func_g;
4460 	struct ftrace_hash **orig_hash;
4461 	struct ftrace_hash *old_hash;
4462 	struct ftrace_hash *hash = NULL;
4463 	struct hlist_node *tmp;
4464 	struct hlist_head hhd;
4465 	char str[KSYM_SYMBOL_LEN];
4466 	int count = 0;
4467 	int i, ret = -ENODEV;
4468 	int size;
4469 
4470 	if (!glob || !strlen(glob) || !strcmp(glob, "*"))
4471 		func_g.search = NULL;
4472 	else {
4473 		int not;
4474 
4475 		func_g.type = filter_parse_regex(glob, strlen(glob),
4476 						 &func_g.search, &not);
4477 		func_g.len = strlen(func_g.search);
4478 
4479 		/* we do not support '!' for function probes */
4480 		if (WARN_ON(not))
4481 			return -EINVAL;
4482 	}
4483 
4484 	mutex_lock(&ftrace_lock);
4485 	/* Check if the probe_ops is already registered */
4486 	list_for_each_entry(probe, &tr->func_probes, list) {
4487 		if (probe->probe_ops == probe_ops)
4488 			break;
4489 	}
4490 	if (&probe->list == &tr->func_probes)
4491 		goto err_unlock_ftrace;
4492 
4493 	ret = -EINVAL;
4494 	if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
4495 		goto err_unlock_ftrace;
4496 
4497 	acquire_probe_locked(probe);
4498 
4499 	mutex_unlock(&ftrace_lock);
4500 
4501 	mutex_lock(&probe->ops.func_hash->regex_lock);
4502 
4503 	orig_hash = &probe->ops.func_hash->filter_hash;
4504 	old_hash = *orig_hash;
4505 
4506 	if (ftrace_hash_empty(old_hash))
4507 		goto out_unlock;
4508 
4509 	old_hash_ops.filter_hash = old_hash;
4510 	/* Probes only have filters */
4511 	old_hash_ops.notrace_hash = NULL;
4512 
4513 	ret = -ENOMEM;
4514 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4515 	if (!hash)
4516 		goto out_unlock;
4517 
4518 	INIT_HLIST_HEAD(&hhd);
4519 
4520 	size = 1 << hash->size_bits;
4521 	for (i = 0; i < size; i++) {
4522 		hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
4523 
4524 			if (func_g.search) {
4525 				kallsyms_lookup(entry->ip, NULL, NULL,
4526 						NULL, str);
4527 				if (!ftrace_match(str, &func_g))
4528 					continue;
4529 			}
4530 			count++;
4531 			remove_hash_entry(hash, entry);
4532 			hlist_add_head(&entry->hlist, &hhd);
4533 		}
4534 	}
4535 
4536 	/* Nothing found? */
4537 	if (!count) {
4538 		ret = -EINVAL;
4539 		goto out_unlock;
4540 	}
4541 
4542 	mutex_lock(&ftrace_lock);
4543 
4544 	WARN_ON(probe->ref < count);
4545 
4546 	probe->ref -= count;
4547 
4548 	if (ftrace_hash_empty(hash))
4549 		ftrace_shutdown(&probe->ops, 0);
4550 
4551 	ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4552 					      hash, 1);
4553 
4554 	/* still need to update the function call sites */
4555 	if (ftrace_enabled && !ftrace_hash_empty(hash))
4556 		ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
4557 				       &old_hash_ops);
4558 	synchronize_rcu();
4559 
4560 	hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
4561 		hlist_del(&entry->hlist);
4562 		if (probe_ops->free)
4563 			probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4564 		kfree(entry);
4565 	}
4566 	mutex_unlock(&ftrace_lock);
4567 
4568  out_unlock:
4569 	mutex_unlock(&probe->ops.func_hash->regex_lock);
4570 	free_ftrace_hash(hash);
4571 
4572 	release_probe(probe);
4573 
4574 	return ret;
4575 
4576  err_unlock_ftrace:
4577 	mutex_unlock(&ftrace_lock);
4578 	return ret;
4579 }
4580 
clear_ftrace_function_probes(struct trace_array * tr)4581 void clear_ftrace_function_probes(struct trace_array *tr)
4582 {
4583 	struct ftrace_func_probe *probe, *n;
4584 
4585 	list_for_each_entry_safe(probe, n, &tr->func_probes, list)
4586 		unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
4587 }
4588 
4589 static LIST_HEAD(ftrace_commands);
4590 static DEFINE_MUTEX(ftrace_cmd_mutex);
4591 
4592 /*
4593  * Currently we only register ftrace commands from __init, so mark this
4594  * __init too.
4595  */
register_ftrace_command(struct ftrace_func_command * cmd)4596 __init int register_ftrace_command(struct ftrace_func_command *cmd)
4597 {
4598 	struct ftrace_func_command *p;
4599 	int ret = 0;
4600 
4601 	mutex_lock(&ftrace_cmd_mutex);
4602 	list_for_each_entry(p, &ftrace_commands, list) {
4603 		if (strcmp(cmd->name, p->name) == 0) {
4604 			ret = -EBUSY;
4605 			goto out_unlock;
4606 		}
4607 	}
4608 	list_add(&cmd->list, &ftrace_commands);
4609  out_unlock:
4610 	mutex_unlock(&ftrace_cmd_mutex);
4611 
4612 	return ret;
4613 }
4614 
4615 /*
4616  * Currently we only unregister ftrace commands from __init, so mark
4617  * this __init too.
4618  */
unregister_ftrace_command(struct ftrace_func_command * cmd)4619 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
4620 {
4621 	struct ftrace_func_command *p, *n;
4622 	int ret = -ENODEV;
4623 
4624 	mutex_lock(&ftrace_cmd_mutex);
4625 	list_for_each_entry_safe(p, n, &ftrace_commands, list) {
4626 		if (strcmp(cmd->name, p->name) == 0) {
4627 			ret = 0;
4628 			list_del_init(&p->list);
4629 			goto out_unlock;
4630 		}
4631 	}
4632  out_unlock:
4633 	mutex_unlock(&ftrace_cmd_mutex);
4634 
4635 	return ret;
4636 }
4637 
ftrace_process_regex(struct ftrace_iterator * iter,char * buff,int len,int enable)4638 static int ftrace_process_regex(struct ftrace_iterator *iter,
4639 				char *buff, int len, int enable)
4640 {
4641 	struct ftrace_hash *hash = iter->hash;
4642 	struct trace_array *tr = iter->ops->private;
4643 	char *func, *command, *next = buff;
4644 	struct ftrace_func_command *p;
4645 	int ret = -EINVAL;
4646 
4647 	func = strsep(&next, ":");
4648 
4649 	if (!next) {
4650 		ret = ftrace_match_records(hash, func, len);
4651 		if (!ret)
4652 			ret = -EINVAL;
4653 		if (ret < 0)
4654 			return ret;
4655 		return 0;
4656 	}
4657 
4658 	/* command found */
4659 
4660 	command = strsep(&next, ":");
4661 
4662 	mutex_lock(&ftrace_cmd_mutex);
4663 	list_for_each_entry(p, &ftrace_commands, list) {
4664 		if (strcmp(p->name, command) == 0) {
4665 			ret = p->func(tr, hash, func, command, next, enable);
4666 			goto out_unlock;
4667 		}
4668 	}
4669  out_unlock:
4670 	mutex_unlock(&ftrace_cmd_mutex);
4671 
4672 	return ret;
4673 }
4674 
4675 static ssize_t
ftrace_regex_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos,int enable)4676 ftrace_regex_write(struct file *file, const char __user *ubuf,
4677 		   size_t cnt, loff_t *ppos, int enable)
4678 {
4679 	struct ftrace_iterator *iter;
4680 	struct trace_parser *parser;
4681 	ssize_t ret, read;
4682 
4683 	if (!cnt)
4684 		return 0;
4685 
4686 	if (file->f_mode & FMODE_READ) {
4687 		struct seq_file *m = file->private_data;
4688 		iter = m->private;
4689 	} else
4690 		iter = file->private_data;
4691 
4692 	if (unlikely(ftrace_disabled))
4693 		return -ENODEV;
4694 
4695 	/* iter->hash is a local copy, so we don't need regex_lock */
4696 
4697 	parser = &iter->parser;
4698 	read = trace_get_user(parser, ubuf, cnt, ppos);
4699 
4700 	if (read >= 0 && trace_parser_loaded(parser) &&
4701 	    !trace_parser_cont(parser)) {
4702 		ret = ftrace_process_regex(iter, parser->buffer,
4703 					   parser->idx, enable);
4704 		trace_parser_clear(parser);
4705 		if (ret < 0)
4706 			goto out;
4707 	}
4708 
4709 	ret = read;
4710  out:
4711 	return ret;
4712 }
4713 
4714 ssize_t
ftrace_filter_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)4715 ftrace_filter_write(struct file *file, const char __user *ubuf,
4716 		    size_t cnt, loff_t *ppos)
4717 {
4718 	return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
4719 }
4720 
4721 ssize_t
ftrace_notrace_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)4722 ftrace_notrace_write(struct file *file, const char __user *ubuf,
4723 		     size_t cnt, loff_t *ppos)
4724 {
4725 	return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
4726 }
4727 
4728 static int
ftrace_match_addr(struct ftrace_hash * hash,unsigned long ip,int remove)4729 ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
4730 {
4731 	struct ftrace_func_entry *entry;
4732 
4733 	if (!ftrace_location(ip))
4734 		return -EINVAL;
4735 
4736 	if (remove) {
4737 		entry = ftrace_lookup_ip(hash, ip);
4738 		if (!entry)
4739 			return -ENOENT;
4740 		free_hash_entry(hash, entry);
4741 		return 0;
4742 	}
4743 
4744 	return add_hash_entry(hash, ip);
4745 }
4746 
4747 static int
ftrace_set_hash(struct ftrace_ops * ops,unsigned char * buf,int len,unsigned long ip,int remove,int reset,int enable)4748 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
4749 		unsigned long ip, int remove, int reset, int enable)
4750 {
4751 	struct ftrace_hash **orig_hash;
4752 	struct ftrace_hash *hash;
4753 	int ret;
4754 
4755 	if (unlikely(ftrace_disabled))
4756 		return -ENODEV;
4757 
4758 	mutex_lock(&ops->func_hash->regex_lock);
4759 
4760 	if (enable)
4761 		orig_hash = &ops->func_hash->filter_hash;
4762 	else
4763 		orig_hash = &ops->func_hash->notrace_hash;
4764 
4765 	if (reset)
4766 		hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4767 	else
4768 		hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
4769 
4770 	if (!hash) {
4771 		ret = -ENOMEM;
4772 		goto out_regex_unlock;
4773 	}
4774 
4775 	if (buf && !ftrace_match_records(hash, buf, len)) {
4776 		ret = -EINVAL;
4777 		goto out_regex_unlock;
4778 	}
4779 	if (ip) {
4780 		ret = ftrace_match_addr(hash, ip, remove);
4781 		if (ret < 0)
4782 			goto out_regex_unlock;
4783 	}
4784 
4785 	mutex_lock(&ftrace_lock);
4786 	ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
4787 	mutex_unlock(&ftrace_lock);
4788 
4789  out_regex_unlock:
4790 	mutex_unlock(&ops->func_hash->regex_lock);
4791 
4792 	free_ftrace_hash(hash);
4793 	return ret;
4794 }
4795 
4796 static int
ftrace_set_addr(struct ftrace_ops * ops,unsigned long ip,int remove,int reset,int enable)4797 ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
4798 		int reset, int enable)
4799 {
4800 	return ftrace_set_hash(ops, NULL, 0, ip, remove, reset, enable);
4801 }
4802 
4803 /**
4804  * ftrace_set_filter_ip - set a function to filter on in ftrace by address
4805  * @ops - the ops to set the filter with
4806  * @ip - the address to add to or remove from the filter.
4807  * @remove - non zero to remove the ip from the filter
4808  * @reset - non zero to reset all filters before applying this filter.
4809  *
4810  * Filters denote which functions should be enabled when tracing is enabled
4811  * If @ip is NULL, it failes to update filter.
4812  */
ftrace_set_filter_ip(struct ftrace_ops * ops,unsigned long ip,int remove,int reset)4813 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
4814 			 int remove, int reset)
4815 {
4816 	ftrace_ops_init(ops);
4817 	return ftrace_set_addr(ops, ip, remove, reset, 1);
4818 }
4819 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
4820 
4821 /**
4822  * ftrace_ops_set_global_filter - setup ops to use global filters
4823  * @ops - the ops which will use the global filters
4824  *
4825  * ftrace users who need global function trace filtering should call this.
4826  * It can set the global filter only if ops were not initialized before.
4827  */
ftrace_ops_set_global_filter(struct ftrace_ops * ops)4828 void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
4829 {
4830 	if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
4831 		return;
4832 
4833 	ftrace_ops_init(ops);
4834 	ops->func_hash = &global_ops.local_hash;
4835 }
4836 EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
4837 
4838 static int
ftrace_set_regex(struct ftrace_ops * ops,unsigned char * buf,int len,int reset,int enable)4839 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4840 		 int reset, int enable)
4841 {
4842 	return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
4843 }
4844 
4845 /**
4846  * ftrace_set_filter - set a function to filter on in ftrace
4847  * @ops - the ops to set the filter with
4848  * @buf - the string that holds the function filter text.
4849  * @len - the length of the string.
4850  * @reset - non zero to reset all filters before applying this filter.
4851  *
4852  * Filters denote which functions should be enabled when tracing is enabled.
4853  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4854  */
ftrace_set_filter(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)4855 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
4856 		       int len, int reset)
4857 {
4858 	ftrace_ops_init(ops);
4859 	return ftrace_set_regex(ops, buf, len, reset, 1);
4860 }
4861 EXPORT_SYMBOL_GPL(ftrace_set_filter);
4862 
4863 /**
4864  * ftrace_set_notrace - set a function to not trace in ftrace
4865  * @ops - the ops to set the notrace filter with
4866  * @buf - the string that holds the function notrace text.
4867  * @len - the length of the string.
4868  * @reset - non zero to reset all filters before applying this filter.
4869  *
4870  * Notrace Filters denote which functions should not be enabled when tracing
4871  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4872  * for tracing.
4873  */
ftrace_set_notrace(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)4874 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
4875 			int len, int reset)
4876 {
4877 	ftrace_ops_init(ops);
4878 	return ftrace_set_regex(ops, buf, len, reset, 0);
4879 }
4880 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
4881 /**
4882  * ftrace_set_global_filter - set a function to filter on with global tracers
4883  * @buf - the string that holds the function filter text.
4884  * @len - the length of the string.
4885  * @reset - non zero to reset all filters before applying this filter.
4886  *
4887  * Filters denote which functions should be enabled when tracing is enabled.
4888  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4889  */
ftrace_set_global_filter(unsigned char * buf,int len,int reset)4890 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
4891 {
4892 	ftrace_set_regex(&global_ops, buf, len, reset, 1);
4893 }
4894 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
4895 
4896 /**
4897  * ftrace_set_global_notrace - set a function to not trace with global tracers
4898  * @buf - the string that holds the function notrace text.
4899  * @len - the length of the string.
4900  * @reset - non zero to reset all filters before applying this filter.
4901  *
4902  * Notrace Filters denote which functions should not be enabled when tracing
4903  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4904  * for tracing.
4905  */
ftrace_set_global_notrace(unsigned char * buf,int len,int reset)4906 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
4907 {
4908 	ftrace_set_regex(&global_ops, buf, len, reset, 0);
4909 }
4910 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
4911 
4912 /*
4913  * command line interface to allow users to set filters on boot up.
4914  */
4915 #define FTRACE_FILTER_SIZE		COMMAND_LINE_SIZE
4916 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4917 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
4918 
4919 /* Used by function selftest to not test if filter is set */
4920 bool ftrace_filter_param __initdata;
4921 
set_ftrace_notrace(char * str)4922 static int __init set_ftrace_notrace(char *str)
4923 {
4924 	ftrace_filter_param = true;
4925 	strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
4926 	return 1;
4927 }
4928 __setup("ftrace_notrace=", set_ftrace_notrace);
4929 
set_ftrace_filter(char * str)4930 static int __init set_ftrace_filter(char *str)
4931 {
4932 	ftrace_filter_param = true;
4933 	strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
4934 	return 1;
4935 }
4936 __setup("ftrace_filter=", set_ftrace_filter);
4937 
4938 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4939 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
4940 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4941 static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
4942 
set_graph_function(char * str)4943 static int __init set_graph_function(char *str)
4944 {
4945 	strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
4946 	return 1;
4947 }
4948 __setup("ftrace_graph_filter=", set_graph_function);
4949 
set_graph_notrace_function(char * str)4950 static int __init set_graph_notrace_function(char *str)
4951 {
4952 	strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
4953 	return 1;
4954 }
4955 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
4956 
set_graph_max_depth_function(char * str)4957 static int __init set_graph_max_depth_function(char *str)
4958 {
4959 	if (!str)
4960 		return 0;
4961 	fgraph_max_depth = simple_strtoul(str, NULL, 0);
4962 	return 1;
4963 }
4964 __setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
4965 
set_ftrace_early_graph(char * buf,int enable)4966 static void __init set_ftrace_early_graph(char *buf, int enable)
4967 {
4968 	int ret;
4969 	char *func;
4970 	struct ftrace_hash *hash;
4971 
4972 	hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4973 	if (WARN_ON(!hash))
4974 		return;
4975 
4976 	while (buf) {
4977 		func = strsep(&buf, ",");
4978 		/* we allow only one expression at a time */
4979 		ret = ftrace_graph_set_hash(hash, func);
4980 		if (ret)
4981 			printk(KERN_DEBUG "ftrace: function %s not "
4982 					  "traceable\n", func);
4983 	}
4984 
4985 	if (enable)
4986 		ftrace_graph_hash = hash;
4987 	else
4988 		ftrace_graph_notrace_hash = hash;
4989 }
4990 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4991 
4992 void __init
ftrace_set_early_filter(struct ftrace_ops * ops,char * buf,int enable)4993 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
4994 {
4995 	char *func;
4996 
4997 	ftrace_ops_init(ops);
4998 
4999 	while (buf) {
5000 		func = strsep(&buf, ",");
5001 		ftrace_set_regex(ops, func, strlen(func), 0, enable);
5002 	}
5003 }
5004 
set_ftrace_early_filters(void)5005 static void __init set_ftrace_early_filters(void)
5006 {
5007 	if (ftrace_filter_buf[0])
5008 		ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
5009 	if (ftrace_notrace_buf[0])
5010 		ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
5011 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5012 	if (ftrace_graph_buf[0])
5013 		set_ftrace_early_graph(ftrace_graph_buf, 1);
5014 	if (ftrace_graph_notrace_buf[0])
5015 		set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
5016 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5017 }
5018 
ftrace_regex_release(struct inode * inode,struct file * file)5019 int ftrace_regex_release(struct inode *inode, struct file *file)
5020 {
5021 	struct seq_file *m = (struct seq_file *)file->private_data;
5022 	struct ftrace_iterator *iter;
5023 	struct ftrace_hash **orig_hash;
5024 	struct trace_parser *parser;
5025 	int filter_hash;
5026 	int ret;
5027 
5028 	if (file->f_mode & FMODE_READ) {
5029 		iter = m->private;
5030 		seq_release(inode, file);
5031 	} else
5032 		iter = file->private_data;
5033 
5034 	parser = &iter->parser;
5035 	if (trace_parser_loaded(parser)) {
5036 		ftrace_match_records(iter->hash, parser->buffer, parser->idx);
5037 	}
5038 
5039 	trace_parser_put(parser);
5040 
5041 	mutex_lock(&iter->ops->func_hash->regex_lock);
5042 
5043 	if (file->f_mode & FMODE_WRITE) {
5044 		filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
5045 
5046 		if (filter_hash) {
5047 			orig_hash = &iter->ops->func_hash->filter_hash;
5048 			if (iter->tr && !list_empty(&iter->tr->mod_trace))
5049 				iter->hash->flags |= FTRACE_HASH_FL_MOD;
5050 		} else
5051 			orig_hash = &iter->ops->func_hash->notrace_hash;
5052 
5053 		mutex_lock(&ftrace_lock);
5054 		ret = ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
5055 						      iter->hash, filter_hash);
5056 		mutex_unlock(&ftrace_lock);
5057 	} else {
5058 		/* For read only, the hash is the ops hash */
5059 		iter->hash = NULL;
5060 	}
5061 
5062 	mutex_unlock(&iter->ops->func_hash->regex_lock);
5063 	free_ftrace_hash(iter->hash);
5064 	if (iter->tr)
5065 		trace_array_put(iter->tr);
5066 	kfree(iter);
5067 
5068 	return 0;
5069 }
5070 
5071 static const struct file_operations ftrace_avail_fops = {
5072 	.open = ftrace_avail_open,
5073 	.read = seq_read,
5074 	.llseek = seq_lseek,
5075 	.release = seq_release_private,
5076 };
5077 
5078 static const struct file_operations ftrace_enabled_fops = {
5079 	.open = ftrace_enabled_open,
5080 	.read = seq_read,
5081 	.llseek = seq_lseek,
5082 	.release = seq_release_private,
5083 };
5084 
5085 static const struct file_operations ftrace_filter_fops = {
5086 	.open = ftrace_filter_open,
5087 	.read = seq_read,
5088 	.write = ftrace_filter_write,
5089 	.llseek = tracing_lseek,
5090 	.release = ftrace_regex_release,
5091 };
5092 
5093 static const struct file_operations ftrace_notrace_fops = {
5094 	.open = ftrace_notrace_open,
5095 	.read = seq_read,
5096 	.write = ftrace_notrace_write,
5097 	.llseek = tracing_lseek,
5098 	.release = ftrace_regex_release,
5099 };
5100 
5101 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5102 
5103 static DEFINE_MUTEX(graph_lock);
5104 
5105 struct ftrace_hash *ftrace_graph_hash = EMPTY_HASH;
5106 struct ftrace_hash *ftrace_graph_notrace_hash = EMPTY_HASH;
5107 
5108 enum graph_filter_type {
5109 	GRAPH_FILTER_NOTRACE	= 0,
5110 	GRAPH_FILTER_FUNCTION,
5111 };
5112 
5113 #define FTRACE_GRAPH_EMPTY	((void *)1)
5114 
5115 struct ftrace_graph_data {
5116 	struct ftrace_hash		*hash;
5117 	struct ftrace_func_entry	*entry;
5118 	int				idx;   /* for hash table iteration */
5119 	enum graph_filter_type		type;
5120 	struct ftrace_hash		*new_hash;
5121 	const struct seq_operations	*seq_ops;
5122 	struct trace_parser		parser;
5123 };
5124 
5125 static void *
__g_next(struct seq_file * m,loff_t * pos)5126 __g_next(struct seq_file *m, loff_t *pos)
5127 {
5128 	struct ftrace_graph_data *fgd = m->private;
5129 	struct ftrace_func_entry *entry = fgd->entry;
5130 	struct hlist_head *head;
5131 	int i, idx = fgd->idx;
5132 
5133 	if (*pos >= fgd->hash->count)
5134 		return NULL;
5135 
5136 	if (entry) {
5137 		hlist_for_each_entry_continue(entry, hlist) {
5138 			fgd->entry = entry;
5139 			return entry;
5140 		}
5141 
5142 		idx++;
5143 	}
5144 
5145 	for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
5146 		head = &fgd->hash->buckets[i];
5147 		hlist_for_each_entry(entry, head, hlist) {
5148 			fgd->entry = entry;
5149 			fgd->idx = i;
5150 			return entry;
5151 		}
5152 	}
5153 	return NULL;
5154 }
5155 
5156 static void *
g_next(struct seq_file * m,void * v,loff_t * pos)5157 g_next(struct seq_file *m, void *v, loff_t *pos)
5158 {
5159 	(*pos)++;
5160 	return __g_next(m, pos);
5161 }
5162 
g_start(struct seq_file * m,loff_t * pos)5163 static void *g_start(struct seq_file *m, loff_t *pos)
5164 {
5165 	struct ftrace_graph_data *fgd = m->private;
5166 
5167 	mutex_lock(&graph_lock);
5168 
5169 	if (fgd->type == GRAPH_FILTER_FUNCTION)
5170 		fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5171 					lockdep_is_held(&graph_lock));
5172 	else
5173 		fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5174 					lockdep_is_held(&graph_lock));
5175 
5176 	/* Nothing, tell g_show to print all functions are enabled */
5177 	if (ftrace_hash_empty(fgd->hash) && !*pos)
5178 		return FTRACE_GRAPH_EMPTY;
5179 
5180 	fgd->idx = 0;
5181 	fgd->entry = NULL;
5182 	return __g_next(m, pos);
5183 }
5184 
g_stop(struct seq_file * m,void * p)5185 static void g_stop(struct seq_file *m, void *p)
5186 {
5187 	mutex_unlock(&graph_lock);
5188 }
5189 
g_show(struct seq_file * m,void * v)5190 static int g_show(struct seq_file *m, void *v)
5191 {
5192 	struct ftrace_func_entry *entry = v;
5193 
5194 	if (!entry)
5195 		return 0;
5196 
5197 	if (entry == FTRACE_GRAPH_EMPTY) {
5198 		struct ftrace_graph_data *fgd = m->private;
5199 
5200 		if (fgd->type == GRAPH_FILTER_FUNCTION)
5201 			seq_puts(m, "#### all functions enabled ####\n");
5202 		else
5203 			seq_puts(m, "#### no functions disabled ####\n");
5204 		return 0;
5205 	}
5206 
5207 	seq_printf(m, "%ps\n", (void *)entry->ip);
5208 
5209 	return 0;
5210 }
5211 
5212 static const struct seq_operations ftrace_graph_seq_ops = {
5213 	.start = g_start,
5214 	.next = g_next,
5215 	.stop = g_stop,
5216 	.show = g_show,
5217 };
5218 
5219 static int
__ftrace_graph_open(struct inode * inode,struct file * file,struct ftrace_graph_data * fgd)5220 __ftrace_graph_open(struct inode *inode, struct file *file,
5221 		    struct ftrace_graph_data *fgd)
5222 {
5223 	int ret;
5224 	struct ftrace_hash *new_hash = NULL;
5225 
5226 	ret = security_locked_down(LOCKDOWN_TRACEFS);
5227 	if (ret)
5228 		return ret;
5229 
5230 	if (file->f_mode & FMODE_WRITE) {
5231 		const int size_bits = FTRACE_HASH_DEFAULT_BITS;
5232 
5233 		if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
5234 			return -ENOMEM;
5235 
5236 		if (file->f_flags & O_TRUNC)
5237 			new_hash = alloc_ftrace_hash(size_bits);
5238 		else
5239 			new_hash = alloc_and_copy_ftrace_hash(size_bits,
5240 							      fgd->hash);
5241 		if (!new_hash) {
5242 			ret = -ENOMEM;
5243 			goto out;
5244 		}
5245 	}
5246 
5247 	if (file->f_mode & FMODE_READ) {
5248 		ret = seq_open(file, &ftrace_graph_seq_ops);
5249 		if (!ret) {
5250 			struct seq_file *m = file->private_data;
5251 			m->private = fgd;
5252 		} else {
5253 			/* Failed */
5254 			free_ftrace_hash(new_hash);
5255 			new_hash = NULL;
5256 		}
5257 	} else
5258 		file->private_data = fgd;
5259 
5260 out:
5261 	if (ret < 0 && file->f_mode & FMODE_WRITE)
5262 		trace_parser_put(&fgd->parser);
5263 
5264 	fgd->new_hash = new_hash;
5265 
5266 	/*
5267 	 * All uses of fgd->hash must be taken with the graph_lock
5268 	 * held. The graph_lock is going to be released, so force
5269 	 * fgd->hash to be reinitialized when it is taken again.
5270 	 */
5271 	fgd->hash = NULL;
5272 
5273 	return ret;
5274 }
5275 
5276 static int
ftrace_graph_open(struct inode * inode,struct file * file)5277 ftrace_graph_open(struct inode *inode, struct file *file)
5278 {
5279 	struct ftrace_graph_data *fgd;
5280 	int ret;
5281 
5282 	if (unlikely(ftrace_disabled))
5283 		return -ENODEV;
5284 
5285 	fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5286 	if (fgd == NULL)
5287 		return -ENOMEM;
5288 
5289 	mutex_lock(&graph_lock);
5290 
5291 	fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5292 					lockdep_is_held(&graph_lock));
5293 	fgd->type = GRAPH_FILTER_FUNCTION;
5294 	fgd->seq_ops = &ftrace_graph_seq_ops;
5295 
5296 	ret = __ftrace_graph_open(inode, file, fgd);
5297 	if (ret < 0)
5298 		kfree(fgd);
5299 
5300 	mutex_unlock(&graph_lock);
5301 	return ret;
5302 }
5303 
5304 static int
ftrace_graph_notrace_open(struct inode * inode,struct file * file)5305 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
5306 {
5307 	struct ftrace_graph_data *fgd;
5308 	int ret;
5309 
5310 	if (unlikely(ftrace_disabled))
5311 		return -ENODEV;
5312 
5313 	fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5314 	if (fgd == NULL)
5315 		return -ENOMEM;
5316 
5317 	mutex_lock(&graph_lock);
5318 
5319 	fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5320 					lockdep_is_held(&graph_lock));
5321 	fgd->type = GRAPH_FILTER_NOTRACE;
5322 	fgd->seq_ops = &ftrace_graph_seq_ops;
5323 
5324 	ret = __ftrace_graph_open(inode, file, fgd);
5325 	if (ret < 0)
5326 		kfree(fgd);
5327 
5328 	mutex_unlock(&graph_lock);
5329 	return ret;
5330 }
5331 
5332 static int
ftrace_graph_release(struct inode * inode,struct file * file)5333 ftrace_graph_release(struct inode *inode, struct file *file)
5334 {
5335 	struct ftrace_graph_data *fgd;
5336 	struct ftrace_hash *old_hash, *new_hash;
5337 	struct trace_parser *parser;
5338 	int ret = 0;
5339 
5340 	if (file->f_mode & FMODE_READ) {
5341 		struct seq_file *m = file->private_data;
5342 
5343 		fgd = m->private;
5344 		seq_release(inode, file);
5345 	} else {
5346 		fgd = file->private_data;
5347 	}
5348 
5349 
5350 	if (file->f_mode & FMODE_WRITE) {
5351 
5352 		parser = &fgd->parser;
5353 
5354 		if (trace_parser_loaded((parser))) {
5355 			ret = ftrace_graph_set_hash(fgd->new_hash,
5356 						    parser->buffer);
5357 		}
5358 
5359 		trace_parser_put(parser);
5360 
5361 		new_hash = __ftrace_hash_move(fgd->new_hash);
5362 		if (!new_hash) {
5363 			ret = -ENOMEM;
5364 			goto out;
5365 		}
5366 
5367 		mutex_lock(&graph_lock);
5368 
5369 		if (fgd->type == GRAPH_FILTER_FUNCTION) {
5370 			old_hash = rcu_dereference_protected(ftrace_graph_hash,
5371 					lockdep_is_held(&graph_lock));
5372 			rcu_assign_pointer(ftrace_graph_hash, new_hash);
5373 		} else {
5374 			old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5375 					lockdep_is_held(&graph_lock));
5376 			rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
5377 		}
5378 
5379 		mutex_unlock(&graph_lock);
5380 
5381 		/* Wait till all users are no longer using the old hash */
5382 		synchronize_rcu();
5383 
5384 		free_ftrace_hash(old_hash);
5385 	}
5386 
5387  out:
5388 	free_ftrace_hash(fgd->new_hash);
5389 	kfree(fgd);
5390 
5391 	return ret;
5392 }
5393 
5394 static int
ftrace_graph_set_hash(struct ftrace_hash * hash,char * buffer)5395 ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
5396 {
5397 	struct ftrace_glob func_g;
5398 	struct dyn_ftrace *rec;
5399 	struct ftrace_page *pg;
5400 	struct ftrace_func_entry *entry;
5401 	int fail = 1;
5402 	int not;
5403 
5404 	/* decode regex */
5405 	func_g.type = filter_parse_regex(buffer, strlen(buffer),
5406 					 &func_g.search, &not);
5407 
5408 	func_g.len = strlen(func_g.search);
5409 
5410 	mutex_lock(&ftrace_lock);
5411 
5412 	if (unlikely(ftrace_disabled)) {
5413 		mutex_unlock(&ftrace_lock);
5414 		return -ENODEV;
5415 	}
5416 
5417 	do_for_each_ftrace_rec(pg, rec) {
5418 
5419 		if (rec->flags & FTRACE_FL_DISABLED)
5420 			continue;
5421 
5422 		if (ftrace_match_record(rec, &func_g, NULL, 0)) {
5423 			entry = ftrace_lookup_ip(hash, rec->ip);
5424 
5425 			if (!not) {
5426 				fail = 0;
5427 
5428 				if (entry)
5429 					continue;
5430 				if (add_hash_entry(hash, rec->ip) < 0)
5431 					goto out;
5432 			} else {
5433 				if (entry) {
5434 					free_hash_entry(hash, entry);
5435 					fail = 0;
5436 				}
5437 			}
5438 		}
5439 	} while_for_each_ftrace_rec();
5440 out:
5441 	mutex_unlock(&ftrace_lock);
5442 
5443 	if (fail)
5444 		return -EINVAL;
5445 
5446 	return 0;
5447 }
5448 
5449 static ssize_t
ftrace_graph_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5450 ftrace_graph_write(struct file *file, const char __user *ubuf,
5451 		   size_t cnt, loff_t *ppos)
5452 {
5453 	ssize_t read, ret = 0;
5454 	struct ftrace_graph_data *fgd = file->private_data;
5455 	struct trace_parser *parser;
5456 
5457 	if (!cnt)
5458 		return 0;
5459 
5460 	/* Read mode uses seq functions */
5461 	if (file->f_mode & FMODE_READ) {
5462 		struct seq_file *m = file->private_data;
5463 		fgd = m->private;
5464 	}
5465 
5466 	parser = &fgd->parser;
5467 
5468 	read = trace_get_user(parser, ubuf, cnt, ppos);
5469 
5470 	if (read >= 0 && trace_parser_loaded(parser) &&
5471 	    !trace_parser_cont(parser)) {
5472 
5473 		ret = ftrace_graph_set_hash(fgd->new_hash,
5474 					    parser->buffer);
5475 		trace_parser_clear(parser);
5476 	}
5477 
5478 	if (!ret)
5479 		ret = read;
5480 
5481 	return ret;
5482 }
5483 
5484 static const struct file_operations ftrace_graph_fops = {
5485 	.open		= ftrace_graph_open,
5486 	.read		= seq_read,
5487 	.write		= ftrace_graph_write,
5488 	.llseek		= tracing_lseek,
5489 	.release	= ftrace_graph_release,
5490 };
5491 
5492 static const struct file_operations ftrace_graph_notrace_fops = {
5493 	.open		= ftrace_graph_notrace_open,
5494 	.read		= seq_read,
5495 	.write		= ftrace_graph_write,
5496 	.llseek		= tracing_lseek,
5497 	.release	= ftrace_graph_release,
5498 };
5499 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5500 
ftrace_create_filter_files(struct ftrace_ops * ops,struct dentry * parent)5501 void ftrace_create_filter_files(struct ftrace_ops *ops,
5502 				struct dentry *parent)
5503 {
5504 
5505 	trace_create_file("set_ftrace_filter", 0644, parent,
5506 			  ops, &ftrace_filter_fops);
5507 
5508 	trace_create_file("set_ftrace_notrace", 0644, parent,
5509 			  ops, &ftrace_notrace_fops);
5510 }
5511 
5512 /*
5513  * The name "destroy_filter_files" is really a misnomer. Although
5514  * in the future, it may actually delete the files, but this is
5515  * really intended to make sure the ops passed in are disabled
5516  * and that when this function returns, the caller is free to
5517  * free the ops.
5518  *
5519  * The "destroy" name is only to match the "create" name that this
5520  * should be paired with.
5521  */
ftrace_destroy_filter_files(struct ftrace_ops * ops)5522 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
5523 {
5524 	mutex_lock(&ftrace_lock);
5525 	if (ops->flags & FTRACE_OPS_FL_ENABLED)
5526 		ftrace_shutdown(ops, 0);
5527 	ops->flags |= FTRACE_OPS_FL_DELETED;
5528 	ftrace_free_filter(ops);
5529 	mutex_unlock(&ftrace_lock);
5530 }
5531 
ftrace_init_dyn_tracefs(struct dentry * d_tracer)5532 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
5533 {
5534 
5535 	trace_create_file("available_filter_functions", 0444,
5536 			d_tracer, NULL, &ftrace_avail_fops);
5537 
5538 	trace_create_file("enabled_functions", 0444,
5539 			d_tracer, NULL, &ftrace_enabled_fops);
5540 
5541 	ftrace_create_filter_files(&global_ops, d_tracer);
5542 
5543 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5544 	trace_create_file("set_graph_function", 0644, d_tracer,
5545 				    NULL,
5546 				    &ftrace_graph_fops);
5547 	trace_create_file("set_graph_notrace", 0644, d_tracer,
5548 				    NULL,
5549 				    &ftrace_graph_notrace_fops);
5550 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5551 
5552 	return 0;
5553 }
5554 
ftrace_cmp_ips(const void * a,const void * b)5555 static int ftrace_cmp_ips(const void *a, const void *b)
5556 {
5557 	const unsigned long *ipa = a;
5558 	const unsigned long *ipb = b;
5559 
5560 	if (*ipa > *ipb)
5561 		return 1;
5562 	if (*ipa < *ipb)
5563 		return -1;
5564 	return 0;
5565 }
5566 
ftrace_process_locs(struct module * mod,unsigned long * start,unsigned long * end)5567 static int __norecordmcount ftrace_process_locs(struct module *mod,
5568 						unsigned long *start,
5569 						unsigned long *end)
5570 {
5571 	struct ftrace_page *start_pg;
5572 	struct ftrace_page *pg;
5573 	struct dyn_ftrace *rec;
5574 	unsigned long count;
5575 	unsigned long *p;
5576 	unsigned long addr;
5577 	unsigned long flags = 0; /* Shut up gcc */
5578 	int ret = -ENOMEM;
5579 
5580 	count = end - start;
5581 
5582 	if (!count)
5583 		return 0;
5584 
5585 	sort(start, count, sizeof(*start),
5586 	     ftrace_cmp_ips, NULL);
5587 
5588 	start_pg = ftrace_allocate_pages(count);
5589 	if (!start_pg)
5590 		return -ENOMEM;
5591 
5592 	mutex_lock(&ftrace_lock);
5593 
5594 	/*
5595 	 * Core and each module needs their own pages, as
5596 	 * modules will free them when they are removed.
5597 	 * Force a new page to be allocated for modules.
5598 	 */
5599 	if (!mod) {
5600 		WARN_ON(ftrace_pages || ftrace_pages_start);
5601 		/* First initialization */
5602 		ftrace_pages = ftrace_pages_start = start_pg;
5603 	} else {
5604 		if (!ftrace_pages)
5605 			goto out;
5606 
5607 		if (WARN_ON(ftrace_pages->next)) {
5608 			/* Hmm, we have free pages? */
5609 			while (ftrace_pages->next)
5610 				ftrace_pages = ftrace_pages->next;
5611 		}
5612 
5613 		ftrace_pages->next = start_pg;
5614 	}
5615 
5616 	p = start;
5617 	pg = start_pg;
5618 	while (p < end) {
5619 		addr = ftrace_call_adjust(*p++);
5620 		/*
5621 		 * Some architecture linkers will pad between
5622 		 * the different mcount_loc sections of different
5623 		 * object files to satisfy alignments.
5624 		 * Skip any NULL pointers.
5625 		 */
5626 		if (!addr)
5627 			continue;
5628 
5629 		if (pg->index == pg->size) {
5630 			/* We should have allocated enough */
5631 			if (WARN_ON(!pg->next))
5632 				break;
5633 			pg = pg->next;
5634 		}
5635 
5636 		rec = &pg->records[pg->index++];
5637 		rec->ip = addr;
5638 	}
5639 
5640 	/* We should have used all pages */
5641 	WARN_ON(pg->next);
5642 
5643 	/* Assign the last page to ftrace_pages */
5644 	ftrace_pages = pg;
5645 
5646 	/*
5647 	 * We only need to disable interrupts on start up
5648 	 * because we are modifying code that an interrupt
5649 	 * may execute, and the modification is not atomic.
5650 	 * But for modules, nothing runs the code we modify
5651 	 * until we are finished with it, and there's no
5652 	 * reason to cause large interrupt latencies while we do it.
5653 	 */
5654 	if (!mod)
5655 		local_irq_save(flags);
5656 	ftrace_update_code(mod, start_pg);
5657 	if (!mod)
5658 		local_irq_restore(flags);
5659 	ret = 0;
5660  out:
5661 	mutex_unlock(&ftrace_lock);
5662 
5663 	return ret;
5664 }
5665 
5666 struct ftrace_mod_func {
5667 	struct list_head	list;
5668 	char			*name;
5669 	unsigned long		ip;
5670 	unsigned int		size;
5671 };
5672 
5673 struct ftrace_mod_map {
5674 	struct rcu_head		rcu;
5675 	struct list_head	list;
5676 	struct module		*mod;
5677 	unsigned long		start_addr;
5678 	unsigned long		end_addr;
5679 	struct list_head	funcs;
5680 	unsigned int		num_funcs;
5681 };
5682 
5683 #ifdef CONFIG_MODULES
5684 
5685 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
5686 
5687 static LIST_HEAD(ftrace_mod_maps);
5688 
referenced_filters(struct dyn_ftrace * rec)5689 static int referenced_filters(struct dyn_ftrace *rec)
5690 {
5691 	struct ftrace_ops *ops;
5692 	int cnt = 0;
5693 
5694 	for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
5695 		if (ops_references_rec(ops, rec))
5696 		    cnt++;
5697 	}
5698 
5699 	return cnt;
5700 }
5701 
5702 static void
clear_mod_from_hash(struct ftrace_page * pg,struct ftrace_hash * hash)5703 clear_mod_from_hash(struct ftrace_page *pg, struct ftrace_hash *hash)
5704 {
5705 	struct ftrace_func_entry *entry;
5706 	struct dyn_ftrace *rec;
5707 	int i;
5708 
5709 	if (ftrace_hash_empty(hash))
5710 		return;
5711 
5712 	for (i = 0; i < pg->index; i++) {
5713 		rec = &pg->records[i];
5714 		entry = __ftrace_lookup_ip(hash, rec->ip);
5715 		/*
5716 		 * Do not allow this rec to match again.
5717 		 * Yeah, it may waste some memory, but will be removed
5718 		 * if/when the hash is modified again.
5719 		 */
5720 		if (entry)
5721 			entry->ip = 0;
5722 	}
5723 }
5724 
5725 /* Clear any records from hashs */
clear_mod_from_hashes(struct ftrace_page * pg)5726 static void clear_mod_from_hashes(struct ftrace_page *pg)
5727 {
5728 	struct trace_array *tr;
5729 
5730 	mutex_lock(&trace_types_lock);
5731 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5732 		if (!tr->ops || !tr->ops->func_hash)
5733 			continue;
5734 		mutex_lock(&tr->ops->func_hash->regex_lock);
5735 		clear_mod_from_hash(pg, tr->ops->func_hash->filter_hash);
5736 		clear_mod_from_hash(pg, tr->ops->func_hash->notrace_hash);
5737 		mutex_unlock(&tr->ops->func_hash->regex_lock);
5738 	}
5739 	mutex_unlock(&trace_types_lock);
5740 }
5741 
ftrace_free_mod_map(struct rcu_head * rcu)5742 static void ftrace_free_mod_map(struct rcu_head *rcu)
5743 {
5744 	struct ftrace_mod_map *mod_map = container_of(rcu, struct ftrace_mod_map, rcu);
5745 	struct ftrace_mod_func *mod_func;
5746 	struct ftrace_mod_func *n;
5747 
5748 	/* All the contents of mod_map are now not visible to readers */
5749 	list_for_each_entry_safe(mod_func, n, &mod_map->funcs, list) {
5750 		kfree(mod_func->name);
5751 		list_del(&mod_func->list);
5752 		kfree(mod_func);
5753 	}
5754 
5755 	kfree(mod_map);
5756 }
5757 
ftrace_release_mod(struct module * mod)5758 void ftrace_release_mod(struct module *mod)
5759 {
5760 	struct ftrace_mod_map *mod_map;
5761 	struct ftrace_mod_map *n;
5762 	struct dyn_ftrace *rec;
5763 	struct ftrace_page **last_pg;
5764 	struct ftrace_page *tmp_page = NULL;
5765 	struct ftrace_page *pg;
5766 	int order;
5767 
5768 	mutex_lock(&ftrace_lock);
5769 
5770 	if (ftrace_disabled)
5771 		goto out_unlock;
5772 
5773 	list_for_each_entry_safe(mod_map, n, &ftrace_mod_maps, list) {
5774 		if (mod_map->mod == mod) {
5775 			list_del_rcu(&mod_map->list);
5776 			call_rcu(&mod_map->rcu, ftrace_free_mod_map);
5777 			break;
5778 		}
5779 	}
5780 
5781 	/*
5782 	 * Each module has its own ftrace_pages, remove
5783 	 * them from the list.
5784 	 */
5785 	last_pg = &ftrace_pages_start;
5786 	for (pg = ftrace_pages_start; pg; pg = *last_pg) {
5787 		rec = &pg->records[0];
5788 		if (within_module_core(rec->ip, mod) ||
5789 		    within_module_init(rec->ip, mod)) {
5790 			/*
5791 			 * As core pages are first, the first
5792 			 * page should never be a module page.
5793 			 */
5794 			if (WARN_ON(pg == ftrace_pages_start))
5795 				goto out_unlock;
5796 
5797 			/* Check if we are deleting the last page */
5798 			if (pg == ftrace_pages)
5799 				ftrace_pages = next_to_ftrace_page(last_pg);
5800 
5801 			ftrace_update_tot_cnt -= pg->index;
5802 			*last_pg = pg->next;
5803 
5804 			pg->next = tmp_page;
5805 			tmp_page = pg;
5806 		} else
5807 			last_pg = &pg->next;
5808 	}
5809  out_unlock:
5810 	mutex_unlock(&ftrace_lock);
5811 
5812 	for (pg = tmp_page; pg; pg = tmp_page) {
5813 
5814 		/* Needs to be called outside of ftrace_lock */
5815 		clear_mod_from_hashes(pg);
5816 
5817 		order = get_count_order(pg->size / ENTRIES_PER_PAGE);
5818 		free_pages((unsigned long)pg->records, order);
5819 		tmp_page = pg->next;
5820 		kfree(pg);
5821 	}
5822 }
5823 
ftrace_module_enable(struct module * mod)5824 void ftrace_module_enable(struct module *mod)
5825 {
5826 	struct dyn_ftrace *rec;
5827 	struct ftrace_page *pg;
5828 
5829 	mutex_lock(&ftrace_lock);
5830 
5831 	if (ftrace_disabled)
5832 		goto out_unlock;
5833 
5834 	/*
5835 	 * If the tracing is enabled, go ahead and enable the record.
5836 	 *
5837 	 * The reason not to enable the record immediately is the
5838 	 * inherent check of ftrace_make_nop/ftrace_make_call for
5839 	 * correct previous instructions.  Making first the NOP
5840 	 * conversion puts the module to the correct state, thus
5841 	 * passing the ftrace_make_call check.
5842 	 *
5843 	 * We also delay this to after the module code already set the
5844 	 * text to read-only, as we now need to set it back to read-write
5845 	 * so that we can modify the text.
5846 	 */
5847 	if (ftrace_start_up)
5848 		ftrace_arch_code_modify_prepare();
5849 
5850 	do_for_each_ftrace_rec(pg, rec) {
5851 		int cnt;
5852 		/*
5853 		 * do_for_each_ftrace_rec() is a double loop.
5854 		 * module text shares the pg. If a record is
5855 		 * not part of this module, then skip this pg,
5856 		 * which the "break" will do.
5857 		 */
5858 		if (!within_module_core(rec->ip, mod) &&
5859 		    !within_module_init(rec->ip, mod))
5860 			break;
5861 
5862 		cnt = 0;
5863 
5864 		/*
5865 		 * When adding a module, we need to check if tracers are
5866 		 * currently enabled and if they are, and can trace this record,
5867 		 * we need to enable the module functions as well as update the
5868 		 * reference counts for those function records.
5869 		 */
5870 		if (ftrace_start_up)
5871 			cnt += referenced_filters(rec);
5872 
5873 		/* This clears FTRACE_FL_DISABLED */
5874 		rec->flags = cnt;
5875 
5876 		if (ftrace_start_up && cnt) {
5877 			int failed = __ftrace_replace_code(rec, 1);
5878 			if (failed) {
5879 				ftrace_bug(failed, rec);
5880 				goto out_loop;
5881 			}
5882 		}
5883 
5884 	} while_for_each_ftrace_rec();
5885 
5886  out_loop:
5887 	if (ftrace_start_up)
5888 		ftrace_arch_code_modify_post_process();
5889 
5890  out_unlock:
5891 	mutex_unlock(&ftrace_lock);
5892 
5893 	process_cached_mods(mod->name);
5894 }
5895 
ftrace_module_init(struct module * mod)5896 void ftrace_module_init(struct module *mod)
5897 {
5898 	if (ftrace_disabled || !mod->num_ftrace_callsites)
5899 		return;
5900 
5901 	ftrace_process_locs(mod, mod->ftrace_callsites,
5902 			    mod->ftrace_callsites + mod->num_ftrace_callsites);
5903 }
5904 
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)5905 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
5906 				struct dyn_ftrace *rec)
5907 {
5908 	struct ftrace_mod_func *mod_func;
5909 	unsigned long symsize;
5910 	unsigned long offset;
5911 	char str[KSYM_SYMBOL_LEN];
5912 	char *modname;
5913 	const char *ret;
5914 
5915 	ret = kallsyms_lookup(rec->ip, &symsize, &offset, &modname, str);
5916 	if (!ret)
5917 		return;
5918 
5919 	mod_func = kmalloc(sizeof(*mod_func), GFP_KERNEL);
5920 	if (!mod_func)
5921 		return;
5922 
5923 	mod_func->name = kstrdup(str, GFP_KERNEL);
5924 	if (!mod_func->name) {
5925 		kfree(mod_func);
5926 		return;
5927 	}
5928 
5929 	mod_func->ip = rec->ip - offset;
5930 	mod_func->size = symsize;
5931 
5932 	mod_map->num_funcs++;
5933 
5934 	list_add_rcu(&mod_func->list, &mod_map->funcs);
5935 }
5936 
5937 static struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)5938 allocate_ftrace_mod_map(struct module *mod,
5939 			unsigned long start, unsigned long end)
5940 {
5941 	struct ftrace_mod_map *mod_map;
5942 
5943 	mod_map = kmalloc(sizeof(*mod_map), GFP_KERNEL);
5944 	if (!mod_map)
5945 		return NULL;
5946 
5947 	mod_map->mod = mod;
5948 	mod_map->start_addr = start;
5949 	mod_map->end_addr = end;
5950 	mod_map->num_funcs = 0;
5951 
5952 	INIT_LIST_HEAD_RCU(&mod_map->funcs);
5953 
5954 	list_add_rcu(&mod_map->list, &ftrace_mod_maps);
5955 
5956 	return mod_map;
5957 }
5958 
5959 static const char *
ftrace_func_address_lookup(struct ftrace_mod_map * mod_map,unsigned long addr,unsigned long * size,unsigned long * off,char * sym)5960 ftrace_func_address_lookup(struct ftrace_mod_map *mod_map,
5961 			   unsigned long addr, unsigned long *size,
5962 			   unsigned long *off, char *sym)
5963 {
5964 	struct ftrace_mod_func *found_func =  NULL;
5965 	struct ftrace_mod_func *mod_func;
5966 
5967 	list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
5968 		if (addr >= mod_func->ip &&
5969 		    addr < mod_func->ip + mod_func->size) {
5970 			found_func = mod_func;
5971 			break;
5972 		}
5973 	}
5974 
5975 	if (found_func) {
5976 		if (size)
5977 			*size = found_func->size;
5978 		if (off)
5979 			*off = addr - found_func->ip;
5980 		if (sym)
5981 			strlcpy(sym, found_func->name, KSYM_NAME_LEN);
5982 
5983 		return found_func->name;
5984 	}
5985 
5986 	return NULL;
5987 }
5988 
5989 const char *
ftrace_mod_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)5990 ftrace_mod_address_lookup(unsigned long addr, unsigned long *size,
5991 		   unsigned long *off, char **modname, char *sym)
5992 {
5993 	struct ftrace_mod_map *mod_map;
5994 	const char *ret = NULL;
5995 
5996 	/* mod_map is freed via call_rcu() */
5997 	preempt_disable();
5998 	list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
5999 		ret = ftrace_func_address_lookup(mod_map, addr, size, off, sym);
6000 		if (ret) {
6001 			if (modname)
6002 				*modname = mod_map->mod->name;
6003 			break;
6004 		}
6005 	}
6006 	preempt_enable();
6007 
6008 	return ret;
6009 }
6010 
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)6011 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
6012 			   char *type, char *name,
6013 			   char *module_name, int *exported)
6014 {
6015 	struct ftrace_mod_map *mod_map;
6016 	struct ftrace_mod_func *mod_func;
6017 
6018 	preempt_disable();
6019 	list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
6020 
6021 		if (symnum >= mod_map->num_funcs) {
6022 			symnum -= mod_map->num_funcs;
6023 			continue;
6024 		}
6025 
6026 		list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
6027 			if (symnum > 1) {
6028 				symnum--;
6029 				continue;
6030 			}
6031 
6032 			*value = mod_func->ip;
6033 			*type = 'T';
6034 			strlcpy(name, mod_func->name, KSYM_NAME_LEN);
6035 			strlcpy(module_name, mod_map->mod->name, MODULE_NAME_LEN);
6036 			*exported = 1;
6037 			preempt_enable();
6038 			return 0;
6039 		}
6040 		WARN_ON(1);
6041 		break;
6042 	}
6043 	preempt_enable();
6044 	return -ERANGE;
6045 }
6046 
6047 #else
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)6048 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
6049 				struct dyn_ftrace *rec) { }
6050 static inline struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)6051 allocate_ftrace_mod_map(struct module *mod,
6052 			unsigned long start, unsigned long end)
6053 {
6054 	return NULL;
6055 }
6056 #endif /* CONFIG_MODULES */
6057 
6058 struct ftrace_init_func {
6059 	struct list_head list;
6060 	unsigned long ip;
6061 };
6062 
6063 /* Clear any init ips from hashes */
6064 static void
clear_func_from_hash(struct ftrace_init_func * func,struct ftrace_hash * hash)6065 clear_func_from_hash(struct ftrace_init_func *func, struct ftrace_hash *hash)
6066 {
6067 	struct ftrace_func_entry *entry;
6068 
6069 	entry = ftrace_lookup_ip(hash, func->ip);
6070 	/*
6071 	 * Do not allow this rec to match again.
6072 	 * Yeah, it may waste some memory, but will be removed
6073 	 * if/when the hash is modified again.
6074 	 */
6075 	if (entry)
6076 		entry->ip = 0;
6077 }
6078 
6079 static void
clear_func_from_hashes(struct ftrace_init_func * func)6080 clear_func_from_hashes(struct ftrace_init_func *func)
6081 {
6082 	struct trace_array *tr;
6083 
6084 	mutex_lock(&trace_types_lock);
6085 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6086 		if (!tr->ops || !tr->ops->func_hash)
6087 			continue;
6088 		mutex_lock(&tr->ops->func_hash->regex_lock);
6089 		clear_func_from_hash(func, tr->ops->func_hash->filter_hash);
6090 		clear_func_from_hash(func, tr->ops->func_hash->notrace_hash);
6091 		mutex_unlock(&tr->ops->func_hash->regex_lock);
6092 	}
6093 	mutex_unlock(&trace_types_lock);
6094 }
6095 
add_to_clear_hash_list(struct list_head * clear_list,struct dyn_ftrace * rec)6096 static void add_to_clear_hash_list(struct list_head *clear_list,
6097 				   struct dyn_ftrace *rec)
6098 {
6099 	struct ftrace_init_func *func;
6100 
6101 	func = kmalloc(sizeof(*func), GFP_KERNEL);
6102 	if (!func) {
6103 		WARN_ONCE(1, "alloc failure, ftrace filter could be stale\n");
6104 		return;
6105 	}
6106 
6107 	func->ip = rec->ip;
6108 	list_add(&func->list, clear_list);
6109 }
6110 
ftrace_free_mem(struct module * mod,void * start_ptr,void * end_ptr)6111 void ftrace_free_mem(struct module *mod, void *start_ptr, void *end_ptr)
6112 {
6113 	unsigned long start = (unsigned long)(start_ptr);
6114 	unsigned long end = (unsigned long)(end_ptr);
6115 	struct ftrace_page **last_pg = &ftrace_pages_start;
6116 	struct ftrace_page *pg;
6117 	struct dyn_ftrace *rec;
6118 	struct dyn_ftrace key;
6119 	struct ftrace_mod_map *mod_map = NULL;
6120 	struct ftrace_init_func *func, *func_next;
6121 	struct list_head clear_hash;
6122 	int order;
6123 
6124 	INIT_LIST_HEAD(&clear_hash);
6125 
6126 	key.ip = start;
6127 	key.flags = end;	/* overload flags, as it is unsigned long */
6128 
6129 	mutex_lock(&ftrace_lock);
6130 
6131 	/*
6132 	 * If we are freeing module init memory, then check if
6133 	 * any tracer is active. If so, we need to save a mapping of
6134 	 * the module functions being freed with the address.
6135 	 */
6136 	if (mod && ftrace_ops_list != &ftrace_list_end)
6137 		mod_map = allocate_ftrace_mod_map(mod, start, end);
6138 
6139 	for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
6140 		if (end < pg->records[0].ip ||
6141 		    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
6142 			continue;
6143  again:
6144 		rec = bsearch(&key, pg->records, pg->index,
6145 			      sizeof(struct dyn_ftrace),
6146 			      ftrace_cmp_recs);
6147 		if (!rec)
6148 			continue;
6149 
6150 		/* rec will be cleared from hashes after ftrace_lock unlock */
6151 		add_to_clear_hash_list(&clear_hash, rec);
6152 
6153 		if (mod_map)
6154 			save_ftrace_mod_rec(mod_map, rec);
6155 
6156 		pg->index--;
6157 		ftrace_update_tot_cnt--;
6158 		if (!pg->index) {
6159 			*last_pg = pg->next;
6160 			order = get_count_order(pg->size / ENTRIES_PER_PAGE);
6161 			free_pages((unsigned long)pg->records, order);
6162 			kfree(pg);
6163 			pg = container_of(last_pg, struct ftrace_page, next);
6164 			if (!(*last_pg))
6165 				ftrace_pages = pg;
6166 			continue;
6167 		}
6168 		memmove(rec, rec + 1,
6169 			(pg->index - (rec - pg->records)) * sizeof(*rec));
6170 		/* More than one function may be in this block */
6171 		goto again;
6172 	}
6173 	mutex_unlock(&ftrace_lock);
6174 
6175 	list_for_each_entry_safe(func, func_next, &clear_hash, list) {
6176 		clear_func_from_hashes(func);
6177 		kfree(func);
6178 	}
6179 }
6180 
ftrace_free_init_mem(void)6181 void __init ftrace_free_init_mem(void)
6182 {
6183 	void *start = (void *)(&__init_begin);
6184 	void *end = (void *)(&__init_end);
6185 
6186 	ftrace_free_mem(NULL, start, end);
6187 }
6188 
ftrace_init(void)6189 void __init ftrace_init(void)
6190 {
6191 	extern unsigned long __start_mcount_loc[];
6192 	extern unsigned long __stop_mcount_loc[];
6193 	unsigned long count, flags;
6194 	int ret;
6195 
6196 	local_irq_save(flags);
6197 	ret = ftrace_dyn_arch_init();
6198 	local_irq_restore(flags);
6199 	if (ret)
6200 		goto failed;
6201 
6202 	count = __stop_mcount_loc - __start_mcount_loc;
6203 	if (!count) {
6204 		pr_info("ftrace: No functions to be traced?\n");
6205 		goto failed;
6206 	}
6207 
6208 	pr_info("ftrace: allocating %ld entries in %ld pages\n",
6209 		count, count / ENTRIES_PER_PAGE + 1);
6210 
6211 	last_ftrace_enabled = ftrace_enabled = 1;
6212 
6213 	ret = ftrace_process_locs(NULL,
6214 				  __start_mcount_loc,
6215 				  __stop_mcount_loc);
6216 
6217 	set_ftrace_early_filters();
6218 
6219 	return;
6220  failed:
6221 	ftrace_disabled = 1;
6222 }
6223 
6224 /* Do nothing if arch does not support this */
arch_ftrace_update_trampoline(struct ftrace_ops * ops)6225 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
6226 {
6227 }
6228 
ftrace_update_trampoline(struct ftrace_ops * ops)6229 static void ftrace_update_trampoline(struct ftrace_ops *ops)
6230 {
6231 	arch_ftrace_update_trampoline(ops);
6232 }
6233 
ftrace_init_trace_array(struct trace_array * tr)6234 void ftrace_init_trace_array(struct trace_array *tr)
6235 {
6236 	INIT_LIST_HEAD(&tr->func_probes);
6237 	INIT_LIST_HEAD(&tr->mod_trace);
6238 	INIT_LIST_HEAD(&tr->mod_notrace);
6239 }
6240 #else
6241 
6242 struct ftrace_ops global_ops = {
6243 	.func			= ftrace_stub,
6244 	.flags			= FTRACE_OPS_FL_RECURSION_SAFE |
6245 				  FTRACE_OPS_FL_INITIALIZED |
6246 				  FTRACE_OPS_FL_PID,
6247 };
6248 
ftrace_nodyn_init(void)6249 static int __init ftrace_nodyn_init(void)
6250 {
6251 	ftrace_enabled = 1;
6252 	return 0;
6253 }
6254 core_initcall(ftrace_nodyn_init);
6255 
ftrace_init_dyn_tracefs(struct dentry * d_tracer)6256 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
ftrace_startup_enable(int command)6257 static inline void ftrace_startup_enable(int command) { }
ftrace_startup_all(int command)6258 static inline void ftrace_startup_all(int command) { }
6259 
6260 # define ftrace_startup_sysctl()	do { } while (0)
6261 # define ftrace_shutdown_sysctl()	do { } while (0)
6262 
ftrace_update_trampoline(struct ftrace_ops * ops)6263 static void ftrace_update_trampoline(struct ftrace_ops *ops)
6264 {
6265 }
6266 
6267 #endif /* CONFIG_DYNAMIC_FTRACE */
6268 
ftrace_init_global_array_ops(struct trace_array * tr)6269 __init void ftrace_init_global_array_ops(struct trace_array *tr)
6270 {
6271 	tr->ops = &global_ops;
6272 	tr->ops->private = tr;
6273 	ftrace_init_trace_array(tr);
6274 }
6275 
ftrace_init_array_ops(struct trace_array * tr,ftrace_func_t func)6276 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
6277 {
6278 	/* If we filter on pids, update to use the pid function */
6279 	if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
6280 		if (WARN_ON(tr->ops->func != ftrace_stub))
6281 			printk("ftrace ops had %pS for function\n",
6282 			       tr->ops->func);
6283 	}
6284 	tr->ops->func = func;
6285 	tr->ops->private = tr;
6286 }
6287 
ftrace_reset_array_ops(struct trace_array * tr)6288 void ftrace_reset_array_ops(struct trace_array *tr)
6289 {
6290 	tr->ops->func = ftrace_stub;
6291 }
6292 
6293 static nokprobe_inline void
__ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * ignored,struct pt_regs * regs)6294 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
6295 		       struct ftrace_ops *ignored, struct pt_regs *regs)
6296 {
6297 	struct ftrace_ops *op;
6298 	int bit;
6299 
6300 	bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
6301 	if (bit < 0)
6302 		return;
6303 
6304 	/*
6305 	 * Some of the ops may be dynamically allocated,
6306 	 * they must be freed after a synchronize_rcu().
6307 	 */
6308 	preempt_disable_notrace();
6309 
6310 	do_for_each_ftrace_op(op, ftrace_ops_list) {
6311 		/* Stub functions don't need to be called nor tested */
6312 		if (op->flags & FTRACE_OPS_FL_STUB)
6313 			continue;
6314 		/*
6315 		 * Check the following for each ops before calling their func:
6316 		 *  if RCU flag is set, then rcu_is_watching() must be true
6317 		 *  if PER_CPU is set, then ftrace_function_local_disable()
6318 		 *                          must be false
6319 		 *  Otherwise test if the ip matches the ops filter
6320 		 *
6321 		 * If any of the above fails then the op->func() is not executed.
6322 		 */
6323 		if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
6324 		    ftrace_ops_test(op, ip, regs)) {
6325 			if (FTRACE_WARN_ON(!op->func)) {
6326 				pr_warn("op=%p %pS\n", op, op);
6327 				goto out;
6328 			}
6329 			op->func(ip, parent_ip, op, regs);
6330 		}
6331 	} while_for_each_ftrace_op(op);
6332 out:
6333 	preempt_enable_notrace();
6334 	trace_clear_recursion(bit);
6335 }
6336 
6337 /*
6338  * Some archs only support passing ip and parent_ip. Even though
6339  * the list function ignores the op parameter, we do not want any
6340  * C side effects, where a function is called without the caller
6341  * sending a third parameter.
6342  * Archs are to support both the regs and ftrace_ops at the same time.
6343  * If they support ftrace_ops, it is assumed they support regs.
6344  * If call backs want to use regs, they must either check for regs
6345  * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
6346  * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
6347  * An architecture can pass partial regs with ftrace_ops and still
6348  * set the ARCH_SUPPORTS_FTRACE_OPS.
6349  */
6350 #if ARCH_SUPPORTS_FTRACE_OPS
ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct pt_regs * regs)6351 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
6352 				 struct ftrace_ops *op, struct pt_regs *regs)
6353 {
6354 	__ftrace_ops_list_func(ip, parent_ip, NULL, regs);
6355 }
6356 NOKPROBE_SYMBOL(ftrace_ops_list_func);
6357 #else
ftrace_ops_no_ops(unsigned long ip,unsigned long parent_ip)6358 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
6359 {
6360 	__ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
6361 }
6362 NOKPROBE_SYMBOL(ftrace_ops_no_ops);
6363 #endif
6364 
6365 /*
6366  * If there's only one function registered but it does not support
6367  * recursion, needs RCU protection and/or requires per cpu handling, then
6368  * this function will be called by the mcount trampoline.
6369  */
ftrace_ops_assist_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct pt_regs * regs)6370 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
6371 				   struct ftrace_ops *op, struct pt_regs *regs)
6372 {
6373 	int bit;
6374 
6375 	if ((op->flags & FTRACE_OPS_FL_RCU) && !rcu_is_watching())
6376 		return;
6377 
6378 	bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
6379 	if (bit < 0)
6380 		return;
6381 
6382 	preempt_disable_notrace();
6383 
6384 	op->func(ip, parent_ip, op, regs);
6385 
6386 	preempt_enable_notrace();
6387 	trace_clear_recursion(bit);
6388 }
6389 NOKPROBE_SYMBOL(ftrace_ops_assist_func);
6390 
6391 /**
6392  * ftrace_ops_get_func - get the function a trampoline should call
6393  * @ops: the ops to get the function for
6394  *
6395  * Normally the mcount trampoline will call the ops->func, but there
6396  * are times that it should not. For example, if the ops does not
6397  * have its own recursion protection, then it should call the
6398  * ftrace_ops_assist_func() instead.
6399  *
6400  * Returns the function that the trampoline should call for @ops.
6401  */
ftrace_ops_get_func(struct ftrace_ops * ops)6402 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
6403 {
6404 	/*
6405 	 * If the function does not handle recursion, needs to be RCU safe,
6406 	 * or does per cpu logic, then we need to call the assist handler.
6407 	 */
6408 	if (!(ops->flags & FTRACE_OPS_FL_RECURSION_SAFE) ||
6409 	    ops->flags & FTRACE_OPS_FL_RCU)
6410 		return ftrace_ops_assist_func;
6411 
6412 	return ops->func;
6413 }
6414 
6415 static void
ftrace_filter_pid_sched_switch_probe(void * data,bool preempt,struct task_struct * prev,struct task_struct * next)6416 ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
6417 		    struct task_struct *prev, struct task_struct *next)
6418 {
6419 	struct trace_array *tr = data;
6420 	struct trace_pid_list *pid_list;
6421 
6422 	pid_list = rcu_dereference_sched(tr->function_pids);
6423 
6424 	this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
6425 		       trace_ignore_this_task(pid_list, next));
6426 }
6427 
6428 static void
ftrace_pid_follow_sched_process_fork(void * data,struct task_struct * self,struct task_struct * task)6429 ftrace_pid_follow_sched_process_fork(void *data,
6430 				     struct task_struct *self,
6431 				     struct task_struct *task)
6432 {
6433 	struct trace_pid_list *pid_list;
6434 	struct trace_array *tr = data;
6435 
6436 	pid_list = rcu_dereference_sched(tr->function_pids);
6437 	trace_filter_add_remove_task(pid_list, self, task);
6438 }
6439 
6440 static void
ftrace_pid_follow_sched_process_exit(void * data,struct task_struct * task)6441 ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
6442 {
6443 	struct trace_pid_list *pid_list;
6444 	struct trace_array *tr = data;
6445 
6446 	pid_list = rcu_dereference_sched(tr->function_pids);
6447 	trace_filter_add_remove_task(pid_list, NULL, task);
6448 }
6449 
ftrace_pid_follow_fork(struct trace_array * tr,bool enable)6450 void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
6451 {
6452 	if (enable) {
6453 		register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
6454 						  tr);
6455 		register_trace_sched_process_exit(ftrace_pid_follow_sched_process_exit,
6456 						  tr);
6457 	} else {
6458 		unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
6459 						    tr);
6460 		unregister_trace_sched_process_exit(ftrace_pid_follow_sched_process_exit,
6461 						    tr);
6462 	}
6463 }
6464 
clear_ftrace_pids(struct trace_array * tr)6465 static void clear_ftrace_pids(struct trace_array *tr)
6466 {
6467 	struct trace_pid_list *pid_list;
6468 	int cpu;
6469 
6470 	pid_list = rcu_dereference_protected(tr->function_pids,
6471 					     lockdep_is_held(&ftrace_lock));
6472 	if (!pid_list)
6473 		return;
6474 
6475 	unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
6476 
6477 	for_each_possible_cpu(cpu)
6478 		per_cpu_ptr(tr->trace_buffer.data, cpu)->ftrace_ignore_pid = false;
6479 
6480 	rcu_assign_pointer(tr->function_pids, NULL);
6481 
6482 	/* Wait till all users are no longer using pid filtering */
6483 	synchronize_rcu();
6484 
6485 	trace_free_pid_list(pid_list);
6486 }
6487 
ftrace_clear_pids(struct trace_array * tr)6488 void ftrace_clear_pids(struct trace_array *tr)
6489 {
6490 	mutex_lock(&ftrace_lock);
6491 
6492 	clear_ftrace_pids(tr);
6493 
6494 	mutex_unlock(&ftrace_lock);
6495 }
6496 
ftrace_pid_reset(struct trace_array * tr)6497 static void ftrace_pid_reset(struct trace_array *tr)
6498 {
6499 	mutex_lock(&ftrace_lock);
6500 	clear_ftrace_pids(tr);
6501 
6502 	ftrace_update_pid_func();
6503 	ftrace_startup_all(0);
6504 
6505 	mutex_unlock(&ftrace_lock);
6506 }
6507 
6508 /* Greater than any max PID */
6509 #define FTRACE_NO_PIDS		(void *)(PID_MAX_LIMIT + 1)
6510 
fpid_start(struct seq_file * m,loff_t * pos)6511 static void *fpid_start(struct seq_file *m, loff_t *pos)
6512 	__acquires(RCU)
6513 {
6514 	struct trace_pid_list *pid_list;
6515 	struct trace_array *tr = m->private;
6516 
6517 	mutex_lock(&ftrace_lock);
6518 	rcu_read_lock_sched();
6519 
6520 	pid_list = rcu_dereference_sched(tr->function_pids);
6521 
6522 	if (!pid_list)
6523 		return !(*pos) ? FTRACE_NO_PIDS : NULL;
6524 
6525 	return trace_pid_start(pid_list, pos);
6526 }
6527 
fpid_next(struct seq_file * m,void * v,loff_t * pos)6528 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
6529 {
6530 	struct trace_array *tr = m->private;
6531 	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
6532 
6533 	if (v == FTRACE_NO_PIDS)
6534 		return NULL;
6535 
6536 	return trace_pid_next(pid_list, v, pos);
6537 }
6538 
fpid_stop(struct seq_file * m,void * p)6539 static void fpid_stop(struct seq_file *m, void *p)
6540 	__releases(RCU)
6541 {
6542 	rcu_read_unlock_sched();
6543 	mutex_unlock(&ftrace_lock);
6544 }
6545 
fpid_show(struct seq_file * m,void * v)6546 static int fpid_show(struct seq_file *m, void *v)
6547 {
6548 	if (v == FTRACE_NO_PIDS) {
6549 		seq_puts(m, "no pid\n");
6550 		return 0;
6551 	}
6552 
6553 	return trace_pid_show(m, v);
6554 }
6555 
6556 static const struct seq_operations ftrace_pid_sops = {
6557 	.start = fpid_start,
6558 	.next = fpid_next,
6559 	.stop = fpid_stop,
6560 	.show = fpid_show,
6561 };
6562 
6563 static int
ftrace_pid_open(struct inode * inode,struct file * file)6564 ftrace_pid_open(struct inode *inode, struct file *file)
6565 {
6566 	struct trace_array *tr = inode->i_private;
6567 	struct seq_file *m;
6568 	int ret = 0;
6569 
6570 	ret = tracing_check_open_get_tr(tr);
6571 	if (ret)
6572 		return ret;
6573 
6574 	if ((file->f_mode & FMODE_WRITE) &&
6575 	    (file->f_flags & O_TRUNC))
6576 		ftrace_pid_reset(tr);
6577 
6578 	ret = seq_open(file, &ftrace_pid_sops);
6579 	if (ret < 0) {
6580 		trace_array_put(tr);
6581 	} else {
6582 		m = file->private_data;
6583 		/* copy tr over to seq ops */
6584 		m->private = tr;
6585 	}
6586 
6587 	return ret;
6588 }
6589 
ignore_task_cpu(void * data)6590 static void ignore_task_cpu(void *data)
6591 {
6592 	struct trace_array *tr = data;
6593 	struct trace_pid_list *pid_list;
6594 
6595 	/*
6596 	 * This function is called by on_each_cpu() while the
6597 	 * event_mutex is held.
6598 	 */
6599 	pid_list = rcu_dereference_protected(tr->function_pids,
6600 					     mutex_is_locked(&ftrace_lock));
6601 
6602 	this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
6603 		       trace_ignore_this_task(pid_list, current));
6604 }
6605 
6606 static ssize_t
ftrace_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)6607 ftrace_pid_write(struct file *filp, const char __user *ubuf,
6608 		   size_t cnt, loff_t *ppos)
6609 {
6610 	struct seq_file *m = filp->private_data;
6611 	struct trace_array *tr = m->private;
6612 	struct trace_pid_list *filtered_pids = NULL;
6613 	struct trace_pid_list *pid_list;
6614 	ssize_t ret;
6615 
6616 	if (!cnt)
6617 		return 0;
6618 
6619 	mutex_lock(&ftrace_lock);
6620 
6621 	filtered_pids = rcu_dereference_protected(tr->function_pids,
6622 					     lockdep_is_held(&ftrace_lock));
6623 
6624 	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
6625 	if (ret < 0)
6626 		goto out;
6627 
6628 	rcu_assign_pointer(tr->function_pids, pid_list);
6629 
6630 	if (filtered_pids) {
6631 		synchronize_rcu();
6632 		trace_free_pid_list(filtered_pids);
6633 	} else if (pid_list) {
6634 		/* Register a probe to set whether to ignore the tracing of a task */
6635 		register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
6636 	}
6637 
6638 	/*
6639 	 * Ignoring of pids is done at task switch. But we have to
6640 	 * check for those tasks that are currently running.
6641 	 * Always do this in case a pid was appended or removed.
6642 	 */
6643 	on_each_cpu(ignore_task_cpu, tr, 1);
6644 
6645 	ftrace_update_pid_func();
6646 	ftrace_startup_all(0);
6647  out:
6648 	mutex_unlock(&ftrace_lock);
6649 
6650 	if (ret > 0)
6651 		*ppos += ret;
6652 
6653 	return ret;
6654 }
6655 
6656 static int
ftrace_pid_release(struct inode * inode,struct file * file)6657 ftrace_pid_release(struct inode *inode, struct file *file)
6658 {
6659 	struct trace_array *tr = inode->i_private;
6660 
6661 	trace_array_put(tr);
6662 
6663 	return seq_release(inode, file);
6664 }
6665 
6666 static const struct file_operations ftrace_pid_fops = {
6667 	.open		= ftrace_pid_open,
6668 	.write		= ftrace_pid_write,
6669 	.read		= seq_read,
6670 	.llseek		= tracing_lseek,
6671 	.release	= ftrace_pid_release,
6672 };
6673 
ftrace_init_tracefs(struct trace_array * tr,struct dentry * d_tracer)6674 void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
6675 {
6676 	trace_create_file("set_ftrace_pid", 0644, d_tracer,
6677 			    tr, &ftrace_pid_fops);
6678 }
6679 
ftrace_init_tracefs_toplevel(struct trace_array * tr,struct dentry * d_tracer)6680 void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
6681 					 struct dentry *d_tracer)
6682 {
6683 	/* Only the top level directory has the dyn_tracefs and profile */
6684 	WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
6685 
6686 	ftrace_init_dyn_tracefs(d_tracer);
6687 	ftrace_profile_tracefs(d_tracer);
6688 }
6689 
6690 /**
6691  * ftrace_kill - kill ftrace
6692  *
6693  * This function should be used by panic code. It stops ftrace
6694  * but in a not so nice way. If you need to simply kill ftrace
6695  * from a non-atomic section, use ftrace_kill.
6696  */
ftrace_kill(void)6697 void ftrace_kill(void)
6698 {
6699 	ftrace_disabled = 1;
6700 	ftrace_enabled = 0;
6701 	ftrace_trace_function = ftrace_stub;
6702 }
6703 
6704 /**
6705  * Test if ftrace is dead or not.
6706  */
ftrace_is_dead(void)6707 int ftrace_is_dead(void)
6708 {
6709 	return ftrace_disabled;
6710 }
6711 
6712 /**
6713  * register_ftrace_function - register a function for profiling
6714  * @ops - ops structure that holds the function for profiling.
6715  *
6716  * Register a function to be called by all functions in the
6717  * kernel.
6718  *
6719  * Note: @ops->func and all the functions it calls must be labeled
6720  *       with "notrace", otherwise it will go into a
6721  *       recursive loop.
6722  */
register_ftrace_function(struct ftrace_ops * ops)6723 int register_ftrace_function(struct ftrace_ops *ops)
6724 {
6725 	int ret = -1;
6726 
6727 	ftrace_ops_init(ops);
6728 
6729 	mutex_lock(&ftrace_lock);
6730 
6731 	ret = ftrace_startup(ops, 0);
6732 
6733 	mutex_unlock(&ftrace_lock);
6734 
6735 	return ret;
6736 }
6737 EXPORT_SYMBOL_GPL(register_ftrace_function);
6738 
6739 /**
6740  * unregister_ftrace_function - unregister a function for profiling.
6741  * @ops - ops structure that holds the function to unregister
6742  *
6743  * Unregister a function that was added to be called by ftrace profiling.
6744  */
unregister_ftrace_function(struct ftrace_ops * ops)6745 int unregister_ftrace_function(struct ftrace_ops *ops)
6746 {
6747 	int ret;
6748 
6749 	mutex_lock(&ftrace_lock);
6750 	ret = ftrace_shutdown(ops, 0);
6751 	mutex_unlock(&ftrace_lock);
6752 
6753 	return ret;
6754 }
6755 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
6756 
6757 int
ftrace_enable_sysctl(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)6758 ftrace_enable_sysctl(struct ctl_table *table, int write,
6759 		     void __user *buffer, size_t *lenp,
6760 		     loff_t *ppos)
6761 {
6762 	int ret = -ENODEV;
6763 
6764 	mutex_lock(&ftrace_lock);
6765 
6766 	if (unlikely(ftrace_disabled))
6767 		goto out;
6768 
6769 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
6770 
6771 	if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
6772 		goto out;
6773 
6774 	last_ftrace_enabled = !!ftrace_enabled;
6775 
6776 	if (ftrace_enabled) {
6777 
6778 		/* we are starting ftrace again */
6779 		if (rcu_dereference_protected(ftrace_ops_list,
6780 			lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
6781 			update_ftrace_function();
6782 
6783 		ftrace_startup_sysctl();
6784 
6785 	} else {
6786 		/* stopping ftrace calls (just send to ftrace_stub) */
6787 		ftrace_trace_function = ftrace_stub;
6788 
6789 		ftrace_shutdown_sysctl();
6790 	}
6791 
6792  out:
6793 	mutex_unlock(&ftrace_lock);
6794 	return ret;
6795 }
6796