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