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