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
1186 static struct ftrace_func_entry *
add_hash_entry(struct ftrace_hash * hash,unsigned long ip)1187 add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1188 {
1189 struct ftrace_func_entry *entry;
1190
1191 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1192 if (!entry)
1193 return NULL;
1194
1195 entry->ip = ip;
1196 __add_hash_entry(hash, entry);
1197
1198 return entry;
1199 }
1200
1201 static void
free_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1202 free_hash_entry(struct ftrace_hash *hash,
1203 struct ftrace_func_entry *entry)
1204 {
1205 hlist_del(&entry->hlist);
1206 kfree(entry);
1207 hash->count--;
1208 }
1209
1210 static void
remove_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1211 remove_hash_entry(struct ftrace_hash *hash,
1212 struct ftrace_func_entry *entry)
1213 {
1214 hlist_del_rcu(&entry->hlist);
1215 hash->count--;
1216 }
1217
ftrace_hash_clear(struct ftrace_hash * hash)1218 static void ftrace_hash_clear(struct ftrace_hash *hash)
1219 {
1220 struct hlist_head *hhd;
1221 struct hlist_node *tn;
1222 struct ftrace_func_entry *entry;
1223 int size = 1 << hash->size_bits;
1224 int i;
1225
1226 if (!hash->count)
1227 return;
1228
1229 for (i = 0; i < size; i++) {
1230 hhd = &hash->buckets[i];
1231 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1232 free_hash_entry(hash, entry);
1233 }
1234 FTRACE_WARN_ON(hash->count);
1235 }
1236
free_ftrace_mod(struct ftrace_mod_load * ftrace_mod)1237 static void free_ftrace_mod(struct ftrace_mod_load *ftrace_mod)
1238 {
1239 list_del(&ftrace_mod->list);
1240 kfree(ftrace_mod->module);
1241 kfree(ftrace_mod->func);
1242 kfree(ftrace_mod);
1243 }
1244
clear_ftrace_mod_list(struct list_head * head)1245 static void clear_ftrace_mod_list(struct list_head *head)
1246 {
1247 struct ftrace_mod_load *p, *n;
1248
1249 /* stack tracer isn't supported yet */
1250 if (!head)
1251 return;
1252
1253 mutex_lock(&ftrace_lock);
1254 list_for_each_entry_safe(p, n, head, list)
1255 free_ftrace_mod(p);
1256 mutex_unlock(&ftrace_lock);
1257 }
1258
free_ftrace_hash(struct ftrace_hash * hash)1259 static void free_ftrace_hash(struct ftrace_hash *hash)
1260 {
1261 if (!hash || hash == EMPTY_HASH)
1262 return;
1263 ftrace_hash_clear(hash);
1264 kfree(hash->buckets);
1265 kfree(hash);
1266 }
1267
__free_ftrace_hash_rcu(struct rcu_head * rcu)1268 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1269 {
1270 struct ftrace_hash *hash;
1271
1272 hash = container_of(rcu, struct ftrace_hash, rcu);
1273 free_ftrace_hash(hash);
1274 }
1275
free_ftrace_hash_rcu(struct ftrace_hash * hash)1276 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1277 {
1278 if (!hash || hash == EMPTY_HASH)
1279 return;
1280 call_rcu(&hash->rcu, __free_ftrace_hash_rcu);
1281 }
1282
1283 /**
1284 * ftrace_free_filter - remove all filters for an ftrace_ops
1285 * @ops: the ops to remove the filters from
1286 */
ftrace_free_filter(struct ftrace_ops * ops)1287 void ftrace_free_filter(struct ftrace_ops *ops)
1288 {
1289 ftrace_ops_init(ops);
1290 free_ftrace_hash(ops->func_hash->filter_hash);
1291 free_ftrace_hash(ops->func_hash->notrace_hash);
1292 }
1293 EXPORT_SYMBOL_GPL(ftrace_free_filter);
1294
alloc_ftrace_hash(int size_bits)1295 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1296 {
1297 struct ftrace_hash *hash;
1298 int size;
1299
1300 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1301 if (!hash)
1302 return NULL;
1303
1304 size = 1 << size_bits;
1305 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1306
1307 if (!hash->buckets) {
1308 kfree(hash);
1309 return NULL;
1310 }
1311
1312 hash->size_bits = size_bits;
1313
1314 return hash;
1315 }
1316
1317
ftrace_add_mod(struct trace_array * tr,const char * func,const char * module,int enable)1318 static int ftrace_add_mod(struct trace_array *tr,
1319 const char *func, const char *module,
1320 int enable)
1321 {
1322 struct ftrace_mod_load *ftrace_mod;
1323 struct list_head *mod_head = enable ? &tr->mod_trace : &tr->mod_notrace;
1324
1325 ftrace_mod = kzalloc(sizeof(*ftrace_mod), GFP_KERNEL);
1326 if (!ftrace_mod)
1327 return -ENOMEM;
1328
1329 INIT_LIST_HEAD(&ftrace_mod->list);
1330 ftrace_mod->func = kstrdup(func, GFP_KERNEL);
1331 ftrace_mod->module = kstrdup(module, GFP_KERNEL);
1332 ftrace_mod->enable = enable;
1333
1334 if (!ftrace_mod->func || !ftrace_mod->module)
1335 goto out_free;
1336
1337 list_add(&ftrace_mod->list, mod_head);
1338
1339 return 0;
1340
1341 out_free:
1342 free_ftrace_mod(ftrace_mod);
1343
1344 return -ENOMEM;
1345 }
1346
1347 static struct ftrace_hash *
alloc_and_copy_ftrace_hash(int size_bits,struct ftrace_hash * hash)1348 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1349 {
1350 struct ftrace_func_entry *entry;
1351 struct ftrace_hash *new_hash;
1352 int size;
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 if (add_hash_entry(new_hash, entry->ip) == NULL)
1370 goto free_hash;
1371 }
1372 }
1373
1374 FTRACE_WARN_ON(new_hash->count != hash->count);
1375
1376 return new_hash;
1377
1378 free_hash:
1379 free_ftrace_hash(new_hash);
1380 return NULL;
1381 }
1382
1383 static void
1384 ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1385 static void
1386 ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1387
1388 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1389 struct ftrace_hash *new_hash);
1390
dup_hash(struct ftrace_hash * src,int size)1391 static struct ftrace_hash *dup_hash(struct ftrace_hash *src, int size)
1392 {
1393 struct ftrace_func_entry *entry;
1394 struct ftrace_hash *new_hash;
1395 struct hlist_head *hhd;
1396 struct hlist_node *tn;
1397 int bits = 0;
1398 int i;
1399
1400 /*
1401 * Use around half the size (max bit of it), but
1402 * a minimum of 2 is fine (as size of 0 or 1 both give 1 for bits).
1403 */
1404 bits = fls(size / 2);
1405
1406 /* Don't allocate too much */
1407 if (bits > FTRACE_HASH_MAX_BITS)
1408 bits = FTRACE_HASH_MAX_BITS;
1409
1410 new_hash = alloc_ftrace_hash(bits);
1411 if (!new_hash)
1412 return NULL;
1413
1414 new_hash->flags = src->flags;
1415
1416 size = 1 << src->size_bits;
1417 for (i = 0; i < size; i++) {
1418 hhd = &src->buckets[i];
1419 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1420 remove_hash_entry(src, entry);
1421 __add_hash_entry(new_hash, entry);
1422 }
1423 }
1424 return new_hash;
1425 }
1426
1427 static struct ftrace_hash *
__ftrace_hash_move(struct ftrace_hash * src)1428 __ftrace_hash_move(struct ftrace_hash *src)
1429 {
1430 int size = src->count;
1431
1432 /*
1433 * If the new source is empty, just return the empty_hash.
1434 */
1435 if (ftrace_hash_empty(src))
1436 return EMPTY_HASH;
1437
1438 return dup_hash(src, size);
1439 }
1440
1441 static int
ftrace_hash_move(struct ftrace_ops * ops,int enable,struct ftrace_hash ** dst,struct ftrace_hash * src)1442 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1443 struct ftrace_hash **dst, struct ftrace_hash *src)
1444 {
1445 struct ftrace_hash *new_hash;
1446 int ret;
1447
1448 /* Reject setting notrace hash on IPMODIFY ftrace_ops */
1449 if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1450 return -EINVAL;
1451
1452 new_hash = __ftrace_hash_move(src);
1453 if (!new_hash)
1454 return -ENOMEM;
1455
1456 /* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1457 if (enable) {
1458 /* IPMODIFY should be updated only when filter_hash updating */
1459 ret = ftrace_hash_ipmodify_update(ops, new_hash);
1460 if (ret < 0) {
1461 free_ftrace_hash(new_hash);
1462 return ret;
1463 }
1464 }
1465
1466 /*
1467 * Remove the current set, update the hash and add
1468 * them back.
1469 */
1470 ftrace_hash_rec_disable_modify(ops, enable);
1471
1472 rcu_assign_pointer(*dst, new_hash);
1473
1474 ftrace_hash_rec_enable_modify(ops, enable);
1475
1476 return 0;
1477 }
1478
hash_contains_ip(unsigned long ip,struct ftrace_ops_hash * hash)1479 static bool hash_contains_ip(unsigned long ip,
1480 struct ftrace_ops_hash *hash)
1481 {
1482 /*
1483 * The function record is a match if it exists in the filter
1484 * hash and not in the notrace hash. Note, an empty hash is
1485 * considered a match for the filter hash, but an empty
1486 * notrace hash is considered not in the notrace hash.
1487 */
1488 return (ftrace_hash_empty(hash->filter_hash) ||
1489 __ftrace_lookup_ip(hash->filter_hash, ip)) &&
1490 (ftrace_hash_empty(hash->notrace_hash) ||
1491 !__ftrace_lookup_ip(hash->notrace_hash, ip));
1492 }
1493
1494 /*
1495 * Test the hashes for this ops to see if we want to call
1496 * the ops->func or not.
1497 *
1498 * It's a match if the ip is in the ops->filter_hash or
1499 * the filter_hash does not exist or is empty,
1500 * AND
1501 * the ip is not in the ops->notrace_hash.
1502 *
1503 * This needs to be called with preemption disabled as
1504 * the hashes are freed with call_rcu().
1505 */
1506 int
ftrace_ops_test(struct ftrace_ops * ops,unsigned long ip,void * regs)1507 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1508 {
1509 struct ftrace_ops_hash hash;
1510 int ret;
1511
1512 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1513 /*
1514 * There's a small race when adding ops that the ftrace handler
1515 * that wants regs, may be called without them. We can not
1516 * allow that handler to be called if regs is NULL.
1517 */
1518 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1519 return 0;
1520 #endif
1521
1522 rcu_assign_pointer(hash.filter_hash, ops->func_hash->filter_hash);
1523 rcu_assign_pointer(hash.notrace_hash, ops->func_hash->notrace_hash);
1524
1525 if (hash_contains_ip(ip, &hash))
1526 ret = 1;
1527 else
1528 ret = 0;
1529
1530 return ret;
1531 }
1532
1533 /*
1534 * This is a double for. Do not use 'break' to break out of the loop,
1535 * you must use a goto.
1536 */
1537 #define do_for_each_ftrace_rec(pg, rec) \
1538 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1539 int _____i; \
1540 for (_____i = 0; _____i < pg->index; _____i++) { \
1541 rec = &pg->records[_____i];
1542
1543 #define while_for_each_ftrace_rec() \
1544 } \
1545 }
1546
1547
ftrace_cmp_recs(const void * a,const void * b)1548 static int ftrace_cmp_recs(const void *a, const void *b)
1549 {
1550 const struct dyn_ftrace *key = a;
1551 const struct dyn_ftrace *rec = b;
1552
1553 if (key->flags < rec->ip)
1554 return -1;
1555 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1556 return 1;
1557 return 0;
1558 }
1559
lookup_rec(unsigned long start,unsigned long end)1560 static struct dyn_ftrace *lookup_rec(unsigned long start, unsigned long end)
1561 {
1562 struct ftrace_page *pg;
1563 struct dyn_ftrace *rec = NULL;
1564 struct dyn_ftrace key;
1565
1566 key.ip = start;
1567 key.flags = end; /* overload flags, as it is unsigned long */
1568
1569 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1570 if (pg->index == 0 ||
1571 end < pg->records[0].ip ||
1572 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1573 continue;
1574 rec = bsearch(&key, pg->records, pg->index,
1575 sizeof(struct dyn_ftrace),
1576 ftrace_cmp_recs);
1577 if (rec)
1578 break;
1579 }
1580 return rec;
1581 }
1582
1583 /**
1584 * ftrace_location_range - return the first address of a traced location
1585 * if it touches the given ip range
1586 * @start: start of range to search.
1587 * @end: end of range to search (inclusive). @end points to the last byte
1588 * to check.
1589 *
1590 * Returns: rec->ip if the related ftrace location is a least partly within
1591 * the given address range. That is, the first address of the instruction
1592 * that is either a NOP or call to the function tracer. It checks the ftrace
1593 * internal tables to determine if the address belongs or not.
1594 */
ftrace_location_range(unsigned long start,unsigned long end)1595 unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1596 {
1597 struct dyn_ftrace *rec;
1598
1599 rec = lookup_rec(start, end);
1600 if (rec)
1601 return rec->ip;
1602
1603 return 0;
1604 }
1605
1606 /**
1607 * ftrace_location - return the ftrace location
1608 * @ip: the instruction pointer to check
1609 *
1610 * Returns:
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 __rcu *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
call_direct_funcs(unsigned long ip,unsigned long pip,struct ftrace_ops * ops,struct ftrace_regs * fregs)2558 static void call_direct_funcs(unsigned long ip, unsigned long pip,
2559 struct ftrace_ops *ops, struct ftrace_regs *fregs)
2560 {
2561 unsigned long addr = READ_ONCE(ops->direct_call);
2562
2563 if (!addr)
2564 return;
2565
2566 arch_ftrace_set_direct_caller(fregs, addr);
2567 }
2568 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
2569
2570 /**
2571 * ftrace_get_addr_new - Get the call address to set to
2572 * @rec: The ftrace record descriptor
2573 *
2574 * If the record has the FTRACE_FL_REGS set, that means that it
2575 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2576 * is not set, then it wants to convert to the normal callback.
2577 *
2578 * Returns: the address of the trampoline to set to
2579 */
ftrace_get_addr_new(struct dyn_ftrace * rec)2580 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2581 {
2582 struct ftrace_ops *ops;
2583 unsigned long addr;
2584
2585 if ((rec->flags & FTRACE_FL_DIRECT) &&
2586 (ftrace_rec_count(rec) == 1)) {
2587 addr = ftrace_find_rec_direct(rec->ip);
2588 if (addr)
2589 return addr;
2590 WARN_ON_ONCE(1);
2591 }
2592
2593 /* Trampolines take precedence over regs */
2594 if (rec->flags & FTRACE_FL_TRAMP) {
2595 ops = ftrace_find_tramp_ops_new(rec);
2596 if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2597 pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2598 (void *)rec->ip, (void *)rec->ip, rec->flags);
2599 /* Ftrace is shutting down, return anything */
2600 return (unsigned long)FTRACE_ADDR;
2601 }
2602 return ops->trampoline;
2603 }
2604
2605 if (rec->flags & FTRACE_FL_REGS)
2606 return (unsigned long)FTRACE_REGS_ADDR;
2607 else
2608 return (unsigned long)FTRACE_ADDR;
2609 }
2610
2611 /**
2612 * ftrace_get_addr_curr - Get the call address that is already there
2613 * @rec: The ftrace record descriptor
2614 *
2615 * The FTRACE_FL_REGS_EN is set when the record already points to
2616 * a function that saves all the regs. Basically the '_EN' version
2617 * represents the current state of the function.
2618 *
2619 * Returns: the address of the trampoline that is currently being called
2620 */
ftrace_get_addr_curr(struct dyn_ftrace * rec)2621 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2622 {
2623 struct ftrace_ops *ops;
2624 unsigned long addr;
2625
2626 /* Direct calls take precedence over trampolines */
2627 if (rec->flags & FTRACE_FL_DIRECT_EN) {
2628 addr = ftrace_find_rec_direct(rec->ip);
2629 if (addr)
2630 return addr;
2631 WARN_ON_ONCE(1);
2632 }
2633
2634 /* Trampolines take precedence over regs */
2635 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2636 ops = ftrace_find_tramp_ops_curr(rec);
2637 if (FTRACE_WARN_ON(!ops)) {
2638 pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2639 (void *)rec->ip, (void *)rec->ip);
2640 /* Ftrace is shutting down, return anything */
2641 return (unsigned long)FTRACE_ADDR;
2642 }
2643 return ops->trampoline;
2644 }
2645
2646 if (rec->flags & FTRACE_FL_REGS_EN)
2647 return (unsigned long)FTRACE_REGS_ADDR;
2648 else
2649 return (unsigned long)FTRACE_ADDR;
2650 }
2651
2652 static int
__ftrace_replace_code(struct dyn_ftrace * rec,bool enable)2653 __ftrace_replace_code(struct dyn_ftrace *rec, bool enable)
2654 {
2655 unsigned long ftrace_old_addr;
2656 unsigned long ftrace_addr;
2657 int ret;
2658
2659 ftrace_addr = ftrace_get_addr_new(rec);
2660
2661 /* This needs to be done before we call ftrace_update_record */
2662 ftrace_old_addr = ftrace_get_addr_curr(rec);
2663
2664 ret = ftrace_update_record(rec, enable);
2665
2666 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2667
2668 switch (ret) {
2669 case FTRACE_UPDATE_IGNORE:
2670 return 0;
2671
2672 case FTRACE_UPDATE_MAKE_CALL:
2673 ftrace_bug_type = FTRACE_BUG_CALL;
2674 return ftrace_make_call(rec, ftrace_addr);
2675
2676 case FTRACE_UPDATE_MAKE_NOP:
2677 ftrace_bug_type = FTRACE_BUG_NOP;
2678 return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2679
2680 case FTRACE_UPDATE_MODIFY_CALL:
2681 ftrace_bug_type = FTRACE_BUG_UPDATE;
2682 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2683 }
2684
2685 return -1; /* unknown ftrace bug */
2686 }
2687
ftrace_replace_code(int mod_flags)2688 void __weak ftrace_replace_code(int mod_flags)
2689 {
2690 struct dyn_ftrace *rec;
2691 struct ftrace_page *pg;
2692 bool enable = mod_flags & FTRACE_MODIFY_ENABLE_FL;
2693 int schedulable = mod_flags & FTRACE_MODIFY_MAY_SLEEP_FL;
2694 int failed;
2695
2696 if (unlikely(ftrace_disabled))
2697 return;
2698
2699 do_for_each_ftrace_rec(pg, rec) {
2700
2701 if (skip_record(rec))
2702 continue;
2703
2704 failed = __ftrace_replace_code(rec, enable);
2705 if (failed) {
2706 ftrace_bug(failed, rec);
2707 /* Stop processing */
2708 return;
2709 }
2710 if (schedulable)
2711 cond_resched();
2712 } while_for_each_ftrace_rec();
2713 }
2714
2715 struct ftrace_rec_iter {
2716 struct ftrace_page *pg;
2717 int index;
2718 };
2719
2720 /**
2721 * ftrace_rec_iter_start - start up iterating over traced functions
2722 *
2723 * Returns: an iterator handle that is used to iterate over all
2724 * the records that represent address locations where functions
2725 * are traced.
2726 *
2727 * May return NULL if no records are available.
2728 */
ftrace_rec_iter_start(void)2729 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2730 {
2731 /*
2732 * We only use a single iterator.
2733 * Protected by the ftrace_lock mutex.
2734 */
2735 static struct ftrace_rec_iter ftrace_rec_iter;
2736 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2737
2738 iter->pg = ftrace_pages_start;
2739 iter->index = 0;
2740
2741 /* Could have empty pages */
2742 while (iter->pg && !iter->pg->index)
2743 iter->pg = iter->pg->next;
2744
2745 if (!iter->pg)
2746 return NULL;
2747
2748 return iter;
2749 }
2750
2751 /**
2752 * ftrace_rec_iter_next - get the next record to process.
2753 * @iter: The handle to the iterator.
2754 *
2755 * Returns: the next iterator after the given iterator @iter.
2756 */
ftrace_rec_iter_next(struct ftrace_rec_iter * iter)2757 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2758 {
2759 iter->index++;
2760
2761 if (iter->index >= iter->pg->index) {
2762 iter->pg = iter->pg->next;
2763 iter->index = 0;
2764
2765 /* Could have empty pages */
2766 while (iter->pg && !iter->pg->index)
2767 iter->pg = iter->pg->next;
2768 }
2769
2770 if (!iter->pg)
2771 return NULL;
2772
2773 return iter;
2774 }
2775
2776 /**
2777 * ftrace_rec_iter_record - get the record at the iterator location
2778 * @iter: The current iterator location
2779 *
2780 * Returns: the record that the current @iter is at.
2781 */
ftrace_rec_iter_record(struct ftrace_rec_iter * iter)2782 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2783 {
2784 return &iter->pg->records[iter->index];
2785 }
2786
2787 static int
ftrace_nop_initialize(struct module * mod,struct dyn_ftrace * rec)2788 ftrace_nop_initialize(struct module *mod, struct dyn_ftrace *rec)
2789 {
2790 int ret;
2791
2792 if (unlikely(ftrace_disabled))
2793 return 0;
2794
2795 ret = ftrace_init_nop(mod, rec);
2796 if (ret) {
2797 ftrace_bug_type = FTRACE_BUG_INIT;
2798 ftrace_bug(ret, rec);
2799 return 0;
2800 }
2801 return 1;
2802 }
2803
2804 /*
2805 * archs can override this function if they must do something
2806 * before the modifying code is performed.
2807 */
ftrace_arch_code_modify_prepare(void)2808 void __weak ftrace_arch_code_modify_prepare(void)
2809 {
2810 }
2811
2812 /*
2813 * archs can override this function if they must do something
2814 * after the modifying code is performed.
2815 */
ftrace_arch_code_modify_post_process(void)2816 void __weak ftrace_arch_code_modify_post_process(void)
2817 {
2818 }
2819
update_ftrace_func(ftrace_func_t func)2820 static int update_ftrace_func(ftrace_func_t func)
2821 {
2822 static ftrace_func_t save_func;
2823
2824 /* Avoid updating if it hasn't changed */
2825 if (func == save_func)
2826 return 0;
2827
2828 save_func = func;
2829
2830 return ftrace_update_ftrace_func(func);
2831 }
2832
ftrace_modify_all_code(int command)2833 void ftrace_modify_all_code(int command)
2834 {
2835 int update = command & FTRACE_UPDATE_TRACE_FUNC;
2836 int mod_flags = 0;
2837 int err = 0;
2838
2839 if (command & FTRACE_MAY_SLEEP)
2840 mod_flags = FTRACE_MODIFY_MAY_SLEEP_FL;
2841
2842 /*
2843 * If the ftrace_caller calls a ftrace_ops func directly,
2844 * we need to make sure that it only traces functions it
2845 * expects to trace. When doing the switch of functions,
2846 * we need to update to the ftrace_ops_list_func first
2847 * before the transition between old and new calls are set,
2848 * as the ftrace_ops_list_func will check the ops hashes
2849 * to make sure the ops are having the right functions
2850 * traced.
2851 */
2852 if (update) {
2853 err = update_ftrace_func(ftrace_ops_list_func);
2854 if (FTRACE_WARN_ON(err))
2855 return;
2856 }
2857
2858 if (command & FTRACE_UPDATE_CALLS)
2859 ftrace_replace_code(mod_flags | FTRACE_MODIFY_ENABLE_FL);
2860 else if (command & FTRACE_DISABLE_CALLS)
2861 ftrace_replace_code(mod_flags);
2862
2863 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2864 function_trace_op = set_function_trace_op;
2865 smp_wmb();
2866 /* If irqs are disabled, we are in stop machine */
2867 if (!irqs_disabled())
2868 smp_call_function(ftrace_sync_ipi, NULL, 1);
2869 err = update_ftrace_func(ftrace_trace_function);
2870 if (FTRACE_WARN_ON(err))
2871 return;
2872 }
2873
2874 if (command & FTRACE_START_FUNC_RET)
2875 err = ftrace_enable_ftrace_graph_caller();
2876 else if (command & FTRACE_STOP_FUNC_RET)
2877 err = ftrace_disable_ftrace_graph_caller();
2878 FTRACE_WARN_ON(err);
2879 }
2880
__ftrace_modify_code(void * data)2881 static int __ftrace_modify_code(void *data)
2882 {
2883 int *command = data;
2884
2885 ftrace_modify_all_code(*command);
2886
2887 return 0;
2888 }
2889
2890 /**
2891 * ftrace_run_stop_machine - go back to the stop machine method
2892 * @command: The command to tell ftrace what to do
2893 *
2894 * If an arch needs to fall back to the stop machine method, the
2895 * it can call this function.
2896 */
ftrace_run_stop_machine(int command)2897 void ftrace_run_stop_machine(int command)
2898 {
2899 stop_machine(__ftrace_modify_code, &command, NULL);
2900 }
2901
2902 /**
2903 * arch_ftrace_update_code - modify the code to trace or not trace
2904 * @command: The command that needs to be done
2905 *
2906 * Archs can override this function if it does not need to
2907 * run stop_machine() to modify code.
2908 */
arch_ftrace_update_code(int command)2909 void __weak arch_ftrace_update_code(int command)
2910 {
2911 ftrace_run_stop_machine(command);
2912 }
2913
ftrace_run_update_code(int command)2914 static void ftrace_run_update_code(int command)
2915 {
2916 ftrace_arch_code_modify_prepare();
2917
2918 /*
2919 * By default we use stop_machine() to modify the code.
2920 * But archs can do what ever they want as long as it
2921 * is safe. The stop_machine() is the safest, but also
2922 * produces the most overhead.
2923 */
2924 arch_ftrace_update_code(command);
2925
2926 ftrace_arch_code_modify_post_process();
2927 }
2928
ftrace_run_modify_code(struct ftrace_ops * ops,int command,struct ftrace_ops_hash * old_hash)2929 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2930 struct ftrace_ops_hash *old_hash)
2931 {
2932 ops->flags |= FTRACE_OPS_FL_MODIFYING;
2933 ops->old_hash.filter_hash = old_hash->filter_hash;
2934 ops->old_hash.notrace_hash = old_hash->notrace_hash;
2935 ftrace_run_update_code(command);
2936 ops->old_hash.filter_hash = NULL;
2937 ops->old_hash.notrace_hash = NULL;
2938 ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2939 }
2940
2941 static ftrace_func_t saved_ftrace_func;
2942 static int ftrace_start_up;
2943
arch_ftrace_trampoline_free(struct ftrace_ops * ops)2944 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2945 {
2946 }
2947
2948 /* List of trace_ops that have allocated trampolines */
2949 static LIST_HEAD(ftrace_ops_trampoline_list);
2950
ftrace_add_trampoline_to_kallsyms(struct ftrace_ops * ops)2951 static void ftrace_add_trampoline_to_kallsyms(struct ftrace_ops *ops)
2952 {
2953 lockdep_assert_held(&ftrace_lock);
2954 list_add_rcu(&ops->list, &ftrace_ops_trampoline_list);
2955 }
2956
ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops * ops)2957 static void ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops *ops)
2958 {
2959 lockdep_assert_held(&ftrace_lock);
2960 list_del_rcu(&ops->list);
2961 synchronize_rcu();
2962 }
2963
2964 /*
2965 * "__builtin__ftrace" is used as a module name in /proc/kallsyms for symbols
2966 * for pages allocated for ftrace purposes, even though "__builtin__ftrace" is
2967 * not a module.
2968 */
2969 #define FTRACE_TRAMPOLINE_MOD "__builtin__ftrace"
2970 #define FTRACE_TRAMPOLINE_SYM "ftrace_trampoline"
2971
ftrace_trampoline_free(struct ftrace_ops * ops)2972 static void ftrace_trampoline_free(struct ftrace_ops *ops)
2973 {
2974 if (ops && (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP) &&
2975 ops->trampoline) {
2976 /*
2977 * Record the text poke event before the ksymbol unregister
2978 * event.
2979 */
2980 perf_event_text_poke((void *)ops->trampoline,
2981 (void *)ops->trampoline,
2982 ops->trampoline_size, NULL, 0);
2983 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
2984 ops->trampoline, ops->trampoline_size,
2985 true, FTRACE_TRAMPOLINE_SYM);
2986 /* Remove from kallsyms after the perf events */
2987 ftrace_remove_trampoline_from_kallsyms(ops);
2988 }
2989
2990 arch_ftrace_trampoline_free(ops);
2991 }
2992
ftrace_startup_enable(int command)2993 static void ftrace_startup_enable(int command)
2994 {
2995 if (saved_ftrace_func != ftrace_trace_function) {
2996 saved_ftrace_func = ftrace_trace_function;
2997 command |= FTRACE_UPDATE_TRACE_FUNC;
2998 }
2999
3000 if (!command || !ftrace_enabled)
3001 return;
3002
3003 ftrace_run_update_code(command);
3004 }
3005
ftrace_startup_all(int command)3006 static void ftrace_startup_all(int command)
3007 {
3008 update_all_ops = true;
3009 ftrace_startup_enable(command);
3010 update_all_ops = false;
3011 }
3012
ftrace_startup(struct ftrace_ops * ops,int command)3013 int ftrace_startup(struct ftrace_ops *ops, int command)
3014 {
3015 int ret;
3016
3017 if (unlikely(ftrace_disabled))
3018 return -ENODEV;
3019
3020 ret = __register_ftrace_function(ops);
3021 if (ret)
3022 return ret;
3023
3024 ftrace_start_up++;
3025
3026 /*
3027 * Note that ftrace probes uses this to start up
3028 * and modify functions it will probe. But we still
3029 * set the ADDING flag for modification, as probes
3030 * do not have trampolines. If they add them in the
3031 * future, then the probes will need to distinguish
3032 * between adding and updating probes.
3033 */
3034 ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
3035
3036 ret = ftrace_hash_ipmodify_enable(ops);
3037 if (ret < 0) {
3038 /* Rollback registration process */
3039 __unregister_ftrace_function(ops);
3040 ftrace_start_up--;
3041 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
3042 if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
3043 ftrace_trampoline_free(ops);
3044 return ret;
3045 }
3046
3047 if (ftrace_hash_rec_enable(ops, 1))
3048 command |= FTRACE_UPDATE_CALLS;
3049
3050 ftrace_startup_enable(command);
3051
3052 /*
3053 * If ftrace is in an undefined state, we just remove ops from list
3054 * to prevent the NULL pointer, instead of totally rolling it back and
3055 * free trampoline, because those actions could cause further damage.
3056 */
3057 if (unlikely(ftrace_disabled)) {
3058 __unregister_ftrace_function(ops);
3059 return -ENODEV;
3060 }
3061
3062 ops->flags &= ~FTRACE_OPS_FL_ADDING;
3063
3064 return 0;
3065 }
3066
ftrace_shutdown(struct ftrace_ops * ops,int command)3067 int ftrace_shutdown(struct ftrace_ops *ops, int command)
3068 {
3069 int ret;
3070
3071 if (unlikely(ftrace_disabled))
3072 return -ENODEV;
3073
3074 ret = __unregister_ftrace_function(ops);
3075 if (ret)
3076 return ret;
3077
3078 ftrace_start_up--;
3079 /*
3080 * Just warn in case of unbalance, no need to kill ftrace, it's not
3081 * critical but the ftrace_call callers may be never nopped again after
3082 * further ftrace uses.
3083 */
3084 WARN_ON_ONCE(ftrace_start_up < 0);
3085
3086 /* Disabling ipmodify never fails */
3087 ftrace_hash_ipmodify_disable(ops);
3088
3089 if (ftrace_hash_rec_disable(ops, 1))
3090 command |= FTRACE_UPDATE_CALLS;
3091
3092 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
3093
3094 if (saved_ftrace_func != ftrace_trace_function) {
3095 saved_ftrace_func = ftrace_trace_function;
3096 command |= FTRACE_UPDATE_TRACE_FUNC;
3097 }
3098
3099 if (!command || !ftrace_enabled)
3100 goto out;
3101
3102 /*
3103 * If the ops uses a trampoline, then it needs to be
3104 * tested first on update.
3105 */
3106 ops->flags |= FTRACE_OPS_FL_REMOVING;
3107 removed_ops = ops;
3108
3109 /* The trampoline logic checks the old hashes */
3110 ops->old_hash.filter_hash = ops->func_hash->filter_hash;
3111 ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
3112
3113 ftrace_run_update_code(command);
3114
3115 /*
3116 * If there's no more ops registered with ftrace, run a
3117 * sanity check to make sure all rec flags are cleared.
3118 */
3119 if (rcu_dereference_protected(ftrace_ops_list,
3120 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
3121 struct ftrace_page *pg;
3122 struct dyn_ftrace *rec;
3123
3124 do_for_each_ftrace_rec(pg, rec) {
3125 if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_NOCLEAR_FLAGS))
3126 pr_warn(" %pS flags:%lx\n",
3127 (void *)rec->ip, rec->flags);
3128 } while_for_each_ftrace_rec();
3129 }
3130
3131 ops->old_hash.filter_hash = NULL;
3132 ops->old_hash.notrace_hash = NULL;
3133
3134 removed_ops = NULL;
3135 ops->flags &= ~FTRACE_OPS_FL_REMOVING;
3136
3137 out:
3138 /*
3139 * Dynamic ops may be freed, we must make sure that all
3140 * callers are done before leaving this function.
3141 */
3142 if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
3143 /*
3144 * We need to do a hard force of sched synchronization.
3145 * This is because we use preempt_disable() to do RCU, but
3146 * the function tracers can be called where RCU is not watching
3147 * (like before user_exit()). We can not rely on the RCU
3148 * infrastructure to do the synchronization, thus we must do it
3149 * ourselves.
3150 */
3151 synchronize_rcu_tasks_rude();
3152
3153 /*
3154 * When the kernel is preemptive, tasks can be preempted
3155 * while on a ftrace trampoline. Just scheduling a task on
3156 * a CPU is not good enough to flush them. Calling
3157 * synchronize_rcu_tasks() will wait for those tasks to
3158 * execute and either schedule voluntarily or enter user space.
3159 */
3160 synchronize_rcu_tasks();
3161
3162 ftrace_trampoline_free(ops);
3163 }
3164
3165 return 0;
3166 }
3167
3168 static u64 ftrace_update_time;
3169 unsigned long ftrace_update_tot_cnt;
3170 unsigned long ftrace_number_of_pages;
3171 unsigned long ftrace_number_of_groups;
3172
ops_traces_mod(struct ftrace_ops * ops)3173 static inline int ops_traces_mod(struct ftrace_ops *ops)
3174 {
3175 /*
3176 * Filter_hash being empty will default to trace module.
3177 * But notrace hash requires a test of individual module functions.
3178 */
3179 return ftrace_hash_empty(ops->func_hash->filter_hash) &&
3180 ftrace_hash_empty(ops->func_hash->notrace_hash);
3181 }
3182
ftrace_update_code(struct module * mod,struct ftrace_page * new_pgs)3183 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
3184 {
3185 bool init_nop = ftrace_need_init_nop();
3186 struct ftrace_page *pg;
3187 struct dyn_ftrace *p;
3188 u64 start, stop;
3189 unsigned long update_cnt = 0;
3190 unsigned long rec_flags = 0;
3191 int i;
3192
3193 start = ftrace_now(raw_smp_processor_id());
3194
3195 /*
3196 * When a module is loaded, this function is called to convert
3197 * the calls to mcount in its text to nops, and also to create
3198 * an entry in the ftrace data. Now, if ftrace is activated
3199 * after this call, but before the module sets its text to
3200 * read-only, the modification of enabling ftrace can fail if
3201 * the read-only is done while ftrace is converting the calls.
3202 * To prevent this, the module's records are set as disabled
3203 * and will be enabled after the call to set the module's text
3204 * to read-only.
3205 */
3206 if (mod)
3207 rec_flags |= FTRACE_FL_DISABLED;
3208
3209 for (pg = new_pgs; pg; pg = pg->next) {
3210
3211 for (i = 0; i < pg->index; i++) {
3212
3213 /* If something went wrong, bail without enabling anything */
3214 if (unlikely(ftrace_disabled))
3215 return -1;
3216
3217 p = &pg->records[i];
3218 p->flags = rec_flags;
3219
3220 /*
3221 * Do the initial record conversion from mcount jump
3222 * to the NOP instructions.
3223 */
3224 if (init_nop && !ftrace_nop_initialize(mod, p))
3225 break;
3226
3227 update_cnt++;
3228 }
3229 }
3230
3231 stop = ftrace_now(raw_smp_processor_id());
3232 ftrace_update_time = stop - start;
3233 ftrace_update_tot_cnt += update_cnt;
3234
3235 return 0;
3236 }
3237
ftrace_allocate_records(struct ftrace_page * pg,int count)3238 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
3239 {
3240 int order;
3241 int pages;
3242 int cnt;
3243
3244 if (WARN_ON(!count))
3245 return -EINVAL;
3246
3247 /* We want to fill as much as possible, with no empty pages */
3248 pages = DIV_ROUND_UP(count, ENTRIES_PER_PAGE);
3249 order = fls(pages) - 1;
3250
3251 again:
3252 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
3253
3254 if (!pg->records) {
3255 /* if we can't allocate this size, try something smaller */
3256 if (!order)
3257 return -ENOMEM;
3258 order--;
3259 goto again;
3260 }
3261
3262 ftrace_number_of_pages += 1 << order;
3263 ftrace_number_of_groups++;
3264
3265 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
3266 pg->order = order;
3267
3268 if (cnt > count)
3269 cnt = count;
3270
3271 return cnt;
3272 }
3273
ftrace_free_pages(struct ftrace_page * pages)3274 static void ftrace_free_pages(struct ftrace_page *pages)
3275 {
3276 struct ftrace_page *pg = pages;
3277
3278 while (pg) {
3279 if (pg->records) {
3280 free_pages((unsigned long)pg->records, pg->order);
3281 ftrace_number_of_pages -= 1 << pg->order;
3282 }
3283 pages = pg->next;
3284 kfree(pg);
3285 pg = pages;
3286 ftrace_number_of_groups--;
3287 }
3288 }
3289
3290 static struct ftrace_page *
ftrace_allocate_pages(unsigned long num_to_init)3291 ftrace_allocate_pages(unsigned long num_to_init)
3292 {
3293 struct ftrace_page *start_pg;
3294 struct ftrace_page *pg;
3295 int cnt;
3296
3297 if (!num_to_init)
3298 return NULL;
3299
3300 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3301 if (!pg)
3302 return NULL;
3303
3304 /*
3305 * Try to allocate as much as possible in one continues
3306 * location that fills in all of the space. We want to
3307 * waste as little space as possible.
3308 */
3309 for (;;) {
3310 cnt = ftrace_allocate_records(pg, num_to_init);
3311 if (cnt < 0)
3312 goto free_pages;
3313
3314 num_to_init -= cnt;
3315 if (!num_to_init)
3316 break;
3317
3318 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3319 if (!pg->next)
3320 goto free_pages;
3321
3322 pg = pg->next;
3323 }
3324
3325 return start_pg;
3326
3327 free_pages:
3328 ftrace_free_pages(start_pg);
3329 pr_info("ftrace: FAILED to allocate memory for functions\n");
3330 return NULL;
3331 }
3332
3333 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3334
3335 struct ftrace_iterator {
3336 loff_t pos;
3337 loff_t func_pos;
3338 loff_t mod_pos;
3339 struct ftrace_page *pg;
3340 struct dyn_ftrace *func;
3341 struct ftrace_func_probe *probe;
3342 struct ftrace_func_entry *probe_entry;
3343 struct trace_parser parser;
3344 struct ftrace_hash *hash;
3345 struct ftrace_ops *ops;
3346 struct trace_array *tr;
3347 struct list_head *mod_list;
3348 int pidx;
3349 int idx;
3350 unsigned flags;
3351 };
3352
3353 static void *
t_probe_next(struct seq_file * m,loff_t * pos)3354 t_probe_next(struct seq_file *m, loff_t *pos)
3355 {
3356 struct ftrace_iterator *iter = m->private;
3357 struct trace_array *tr = iter->ops->private;
3358 struct list_head *func_probes;
3359 struct ftrace_hash *hash;
3360 struct list_head *next;
3361 struct hlist_node *hnd = NULL;
3362 struct hlist_head *hhd;
3363 int size;
3364
3365 (*pos)++;
3366 iter->pos = *pos;
3367
3368 if (!tr)
3369 return NULL;
3370
3371 func_probes = &tr->func_probes;
3372 if (list_empty(func_probes))
3373 return NULL;
3374
3375 if (!iter->probe) {
3376 next = func_probes->next;
3377 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3378 }
3379
3380 if (iter->probe_entry)
3381 hnd = &iter->probe_entry->hlist;
3382
3383 hash = iter->probe->ops.func_hash->filter_hash;
3384
3385 /*
3386 * A probe being registered may temporarily have an empty hash
3387 * and it's at the end of the func_probes list.
3388 */
3389 if (!hash || hash == EMPTY_HASH)
3390 return NULL;
3391
3392 size = 1 << hash->size_bits;
3393
3394 retry:
3395 if (iter->pidx >= size) {
3396 if (iter->probe->list.next == func_probes)
3397 return NULL;
3398 next = iter->probe->list.next;
3399 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3400 hash = iter->probe->ops.func_hash->filter_hash;
3401 size = 1 << hash->size_bits;
3402 iter->pidx = 0;
3403 }
3404
3405 hhd = &hash->buckets[iter->pidx];
3406
3407 if (hlist_empty(hhd)) {
3408 iter->pidx++;
3409 hnd = NULL;
3410 goto retry;
3411 }
3412
3413 if (!hnd)
3414 hnd = hhd->first;
3415 else {
3416 hnd = hnd->next;
3417 if (!hnd) {
3418 iter->pidx++;
3419 goto retry;
3420 }
3421 }
3422
3423 if (WARN_ON_ONCE(!hnd))
3424 return NULL;
3425
3426 iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
3427
3428 return iter;
3429 }
3430
t_probe_start(struct seq_file * m,loff_t * pos)3431 static void *t_probe_start(struct seq_file *m, loff_t *pos)
3432 {
3433 struct ftrace_iterator *iter = m->private;
3434 void *p = NULL;
3435 loff_t l;
3436
3437 if (!(iter->flags & FTRACE_ITER_DO_PROBES))
3438 return NULL;
3439
3440 if (iter->mod_pos > *pos)
3441 return NULL;
3442
3443 iter->probe = NULL;
3444 iter->probe_entry = NULL;
3445 iter->pidx = 0;
3446 for (l = 0; l <= (*pos - iter->mod_pos); ) {
3447 p = t_probe_next(m, &l);
3448 if (!p)
3449 break;
3450 }
3451 if (!p)
3452 return NULL;
3453
3454 /* Only set this if we have an item */
3455 iter->flags |= FTRACE_ITER_PROBE;
3456
3457 return iter;
3458 }
3459
3460 static int
t_probe_show(struct seq_file * m,struct ftrace_iterator * iter)3461 t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
3462 {
3463 struct ftrace_func_entry *probe_entry;
3464 struct ftrace_probe_ops *probe_ops;
3465 struct ftrace_func_probe *probe;
3466
3467 probe = iter->probe;
3468 probe_entry = iter->probe_entry;
3469
3470 if (WARN_ON_ONCE(!probe || !probe_entry))
3471 return -EIO;
3472
3473 probe_ops = probe->probe_ops;
3474
3475 if (probe_ops->print)
3476 return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
3477
3478 seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
3479 (void *)probe_ops->func);
3480
3481 return 0;
3482 }
3483
3484 static void *
t_mod_next(struct seq_file * m,loff_t * pos)3485 t_mod_next(struct seq_file *m, loff_t *pos)
3486 {
3487 struct ftrace_iterator *iter = m->private;
3488 struct trace_array *tr = iter->tr;
3489
3490 (*pos)++;
3491 iter->pos = *pos;
3492
3493 iter->mod_list = iter->mod_list->next;
3494
3495 if (iter->mod_list == &tr->mod_trace ||
3496 iter->mod_list == &tr->mod_notrace) {
3497 iter->flags &= ~FTRACE_ITER_MOD;
3498 return NULL;
3499 }
3500
3501 iter->mod_pos = *pos;
3502
3503 return iter;
3504 }
3505
t_mod_start(struct seq_file * m,loff_t * pos)3506 static void *t_mod_start(struct seq_file *m, loff_t *pos)
3507 {
3508 struct ftrace_iterator *iter = m->private;
3509 void *p = NULL;
3510 loff_t l;
3511
3512 if (iter->func_pos > *pos)
3513 return NULL;
3514
3515 iter->mod_pos = iter->func_pos;
3516
3517 /* probes are only available if tr is set */
3518 if (!iter->tr)
3519 return NULL;
3520
3521 for (l = 0; l <= (*pos - iter->func_pos); ) {
3522 p = t_mod_next(m, &l);
3523 if (!p)
3524 break;
3525 }
3526 if (!p) {
3527 iter->flags &= ~FTRACE_ITER_MOD;
3528 return t_probe_start(m, pos);
3529 }
3530
3531 /* Only set this if we have an item */
3532 iter->flags |= FTRACE_ITER_MOD;
3533
3534 return iter;
3535 }
3536
3537 static int
t_mod_show(struct seq_file * m,struct ftrace_iterator * iter)3538 t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
3539 {
3540 struct ftrace_mod_load *ftrace_mod;
3541 struct trace_array *tr = iter->tr;
3542
3543 if (WARN_ON_ONCE(!iter->mod_list) ||
3544 iter->mod_list == &tr->mod_trace ||
3545 iter->mod_list == &tr->mod_notrace)
3546 return -EIO;
3547
3548 ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
3549
3550 if (ftrace_mod->func)
3551 seq_printf(m, "%s", ftrace_mod->func);
3552 else
3553 seq_putc(m, '*');
3554
3555 seq_printf(m, ":mod:%s\n", ftrace_mod->module);
3556
3557 return 0;
3558 }
3559
3560 static void *
t_func_next(struct seq_file * m,loff_t * pos)3561 t_func_next(struct seq_file *m, loff_t *pos)
3562 {
3563 struct ftrace_iterator *iter = m->private;
3564 struct dyn_ftrace *rec = NULL;
3565
3566 (*pos)++;
3567
3568 retry:
3569 if (iter->idx >= iter->pg->index) {
3570 if (iter->pg->next) {
3571 iter->pg = iter->pg->next;
3572 iter->idx = 0;
3573 goto retry;
3574 }
3575 } else {
3576 rec = &iter->pg->records[iter->idx++];
3577 if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3578 !ftrace_lookup_ip(iter->hash, rec->ip)) ||
3579
3580 ((iter->flags & FTRACE_ITER_ENABLED) &&
3581 !(rec->flags & FTRACE_FL_ENABLED)) ||
3582
3583 ((iter->flags & FTRACE_ITER_TOUCHED) &&
3584 !(rec->flags & FTRACE_FL_TOUCHED))) {
3585
3586 rec = NULL;
3587 goto retry;
3588 }
3589 }
3590
3591 if (!rec)
3592 return NULL;
3593
3594 iter->pos = iter->func_pos = *pos;
3595 iter->func = rec;
3596
3597 return iter;
3598 }
3599
3600 static void *
t_next(struct seq_file * m,void * v,loff_t * pos)3601 t_next(struct seq_file *m, void *v, loff_t *pos)
3602 {
3603 struct ftrace_iterator *iter = m->private;
3604 loff_t l = *pos; /* t_probe_start() must use original pos */
3605 void *ret;
3606
3607 if (unlikely(ftrace_disabled))
3608 return NULL;
3609
3610 if (iter->flags & FTRACE_ITER_PROBE)
3611 return t_probe_next(m, pos);
3612
3613 if (iter->flags & FTRACE_ITER_MOD)
3614 return t_mod_next(m, pos);
3615
3616 if (iter->flags & FTRACE_ITER_PRINTALL) {
3617 /* next must increment pos, and t_probe_start does not */
3618 (*pos)++;
3619 return t_mod_start(m, &l);
3620 }
3621
3622 ret = t_func_next(m, pos);
3623
3624 if (!ret)
3625 return t_mod_start(m, &l);
3626
3627 return ret;
3628 }
3629
reset_iter_read(struct ftrace_iterator * iter)3630 static void reset_iter_read(struct ftrace_iterator *iter)
3631 {
3632 iter->pos = 0;
3633 iter->func_pos = 0;
3634 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
3635 }
3636
t_start(struct seq_file * m,loff_t * pos)3637 static void *t_start(struct seq_file *m, loff_t *pos)
3638 {
3639 struct ftrace_iterator *iter = m->private;
3640 void *p = NULL;
3641 loff_t l;
3642
3643 mutex_lock(&ftrace_lock);
3644
3645 if (unlikely(ftrace_disabled))
3646 return NULL;
3647
3648 /*
3649 * If an lseek was done, then reset and start from beginning.
3650 */
3651 if (*pos < iter->pos)
3652 reset_iter_read(iter);
3653
3654 /*
3655 * For set_ftrace_filter reading, if we have the filter
3656 * off, we can short cut and just print out that all
3657 * functions are enabled.
3658 */
3659 if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3660 ftrace_hash_empty(iter->hash)) {
3661 iter->func_pos = 1; /* Account for the message */
3662 if (*pos > 0)
3663 return t_mod_start(m, pos);
3664 iter->flags |= FTRACE_ITER_PRINTALL;
3665 /* reset in case of seek/pread */
3666 iter->flags &= ~FTRACE_ITER_PROBE;
3667 return iter;
3668 }
3669
3670 if (iter->flags & FTRACE_ITER_MOD)
3671 return t_mod_start(m, pos);
3672
3673 /*
3674 * Unfortunately, we need to restart at ftrace_pages_start
3675 * every time we let go of the ftrace_mutex. This is because
3676 * those pointers can change without the lock.
3677 */
3678 iter->pg = ftrace_pages_start;
3679 iter->idx = 0;
3680 for (l = 0; l <= *pos; ) {
3681 p = t_func_next(m, &l);
3682 if (!p)
3683 break;
3684 }
3685
3686 if (!p)
3687 return t_mod_start(m, pos);
3688
3689 return iter;
3690 }
3691
t_stop(struct seq_file * m,void * p)3692 static void t_stop(struct seq_file *m, void *p)
3693 {
3694 mutex_unlock(&ftrace_lock);
3695 }
3696
3697 void * __weak
arch_ftrace_trampoline_func(struct ftrace_ops * ops,struct dyn_ftrace * rec)3698 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3699 {
3700 return NULL;
3701 }
3702
add_trampoline_func(struct seq_file * m,struct ftrace_ops * ops,struct dyn_ftrace * rec)3703 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3704 struct dyn_ftrace *rec)
3705 {
3706 void *ptr;
3707
3708 ptr = arch_ftrace_trampoline_func(ops, rec);
3709 if (ptr)
3710 seq_printf(m, " ->%pS", ptr);
3711 }
3712
3713 #ifdef FTRACE_MCOUNT_MAX_OFFSET
3714 /*
3715 * Weak functions can still have an mcount/fentry that is saved in
3716 * the __mcount_loc section. These can be detected by having a
3717 * symbol offset of greater than FTRACE_MCOUNT_MAX_OFFSET, as the
3718 * symbol found by kallsyms is not the function that the mcount/fentry
3719 * is part of. The offset is much greater in these cases.
3720 *
3721 * Test the record to make sure that the ip points to a valid kallsyms
3722 * and if not, mark it disabled.
3723 */
test_for_valid_rec(struct dyn_ftrace * rec)3724 static int test_for_valid_rec(struct dyn_ftrace *rec)
3725 {
3726 char str[KSYM_SYMBOL_LEN];
3727 unsigned long offset;
3728 const char *ret;
3729
3730 ret = kallsyms_lookup(rec->ip, NULL, &offset, NULL, str);
3731
3732 /* Weak functions can cause invalid addresses */
3733 if (!ret || offset > FTRACE_MCOUNT_MAX_OFFSET) {
3734 rec->flags |= FTRACE_FL_DISABLED;
3735 return 0;
3736 }
3737 return 1;
3738 }
3739
3740 static struct workqueue_struct *ftrace_check_wq __initdata;
3741 static struct work_struct ftrace_check_work __initdata;
3742
3743 /*
3744 * Scan all the mcount/fentry entries to make sure they are valid.
3745 */
ftrace_check_work_func(struct work_struct * work)3746 static __init void ftrace_check_work_func(struct work_struct *work)
3747 {
3748 struct ftrace_page *pg;
3749 struct dyn_ftrace *rec;
3750
3751 mutex_lock(&ftrace_lock);
3752 do_for_each_ftrace_rec(pg, rec) {
3753 test_for_valid_rec(rec);
3754 } while_for_each_ftrace_rec();
3755 mutex_unlock(&ftrace_lock);
3756 }
3757
ftrace_check_for_weak_functions(void)3758 static int __init ftrace_check_for_weak_functions(void)
3759 {
3760 INIT_WORK(&ftrace_check_work, ftrace_check_work_func);
3761
3762 ftrace_check_wq = alloc_workqueue("ftrace_check_wq", WQ_UNBOUND, 0);
3763
3764 queue_work(ftrace_check_wq, &ftrace_check_work);
3765 return 0;
3766 }
3767
ftrace_check_sync(void)3768 static int __init ftrace_check_sync(void)
3769 {
3770 /* Make sure the ftrace_check updates are finished */
3771 if (ftrace_check_wq)
3772 destroy_workqueue(ftrace_check_wq);
3773 return 0;
3774 }
3775
3776 late_initcall_sync(ftrace_check_sync);
3777 subsys_initcall(ftrace_check_for_weak_functions);
3778
print_rec(struct seq_file * m,unsigned long ip)3779 static int print_rec(struct seq_file *m, unsigned long ip)
3780 {
3781 unsigned long offset;
3782 char str[KSYM_SYMBOL_LEN];
3783 char *modname;
3784 const char *ret;
3785
3786 ret = kallsyms_lookup(ip, NULL, &offset, &modname, str);
3787 /* Weak functions can cause invalid addresses */
3788 if (!ret || offset > FTRACE_MCOUNT_MAX_OFFSET) {
3789 snprintf(str, KSYM_SYMBOL_LEN, "%s_%ld",
3790 FTRACE_INVALID_FUNCTION, offset);
3791 ret = NULL;
3792 }
3793
3794 seq_puts(m, str);
3795 if (modname)
3796 seq_printf(m, " [%s]", modname);
3797 return ret == NULL ? -1 : 0;
3798 }
3799 #else
test_for_valid_rec(struct dyn_ftrace * rec)3800 static inline int test_for_valid_rec(struct dyn_ftrace *rec)
3801 {
3802 return 1;
3803 }
3804
print_rec(struct seq_file * m,unsigned long ip)3805 static inline int print_rec(struct seq_file *m, unsigned long ip)
3806 {
3807 seq_printf(m, "%ps", (void *)ip);
3808 return 0;
3809 }
3810 #endif
3811
t_show(struct seq_file * m,void * v)3812 static int t_show(struct seq_file *m, void *v)
3813 {
3814 struct ftrace_iterator *iter = m->private;
3815 struct dyn_ftrace *rec;
3816
3817 if (iter->flags & FTRACE_ITER_PROBE)
3818 return t_probe_show(m, iter);
3819
3820 if (iter->flags & FTRACE_ITER_MOD)
3821 return t_mod_show(m, iter);
3822
3823 if (iter->flags & FTRACE_ITER_PRINTALL) {
3824 if (iter->flags & FTRACE_ITER_NOTRACE)
3825 seq_puts(m, "#### no functions disabled ####\n");
3826 else
3827 seq_puts(m, "#### all functions enabled ####\n");
3828 return 0;
3829 }
3830
3831 rec = iter->func;
3832
3833 if (!rec)
3834 return 0;
3835
3836 if (iter->flags & FTRACE_ITER_ADDRS)
3837 seq_printf(m, "%lx ", rec->ip);
3838
3839 if (print_rec(m, rec->ip)) {
3840 /* This should only happen when a rec is disabled */
3841 WARN_ON_ONCE(!(rec->flags & FTRACE_FL_DISABLED));
3842 seq_putc(m, '\n');
3843 return 0;
3844 }
3845
3846 if (iter->flags & (FTRACE_ITER_ENABLED | FTRACE_ITER_TOUCHED)) {
3847 struct ftrace_ops *ops;
3848
3849 seq_printf(m, " (%ld)%s%s%s%s%s",
3850 ftrace_rec_count(rec),
3851 rec->flags & FTRACE_FL_REGS ? " R" : " ",
3852 rec->flags & FTRACE_FL_IPMODIFY ? " I" : " ",
3853 rec->flags & FTRACE_FL_DIRECT ? " D" : " ",
3854 rec->flags & FTRACE_FL_CALL_OPS ? " O" : " ",
3855 rec->flags & FTRACE_FL_MODIFIED ? " M " : " ");
3856 if (rec->flags & FTRACE_FL_TRAMP_EN) {
3857 ops = ftrace_find_tramp_ops_any(rec);
3858 if (ops) {
3859 do {
3860 seq_printf(m, "\ttramp: %pS (%pS)",
3861 (void *)ops->trampoline,
3862 (void *)ops->func);
3863 add_trampoline_func(m, ops, rec);
3864 ops = ftrace_find_tramp_ops_next(rec, ops);
3865 } while (ops);
3866 } else
3867 seq_puts(m, "\ttramp: ERROR!");
3868 } else {
3869 add_trampoline_func(m, NULL, rec);
3870 }
3871 if (rec->flags & FTRACE_FL_CALL_OPS_EN) {
3872 ops = ftrace_find_unique_ops(rec);
3873 if (ops) {
3874 seq_printf(m, "\tops: %pS (%pS)",
3875 ops, ops->func);
3876 } else {
3877 seq_puts(m, "\tops: ERROR!");
3878 }
3879 }
3880 if (rec->flags & FTRACE_FL_DIRECT) {
3881 unsigned long direct;
3882
3883 direct = ftrace_find_rec_direct(rec->ip);
3884 if (direct)
3885 seq_printf(m, "\n\tdirect-->%pS", (void *)direct);
3886 }
3887 }
3888
3889 seq_putc(m, '\n');
3890
3891 return 0;
3892 }
3893
3894 static const struct seq_operations show_ftrace_seq_ops = {
3895 .start = t_start,
3896 .next = t_next,
3897 .stop = t_stop,
3898 .show = t_show,
3899 };
3900
3901 static int
ftrace_avail_open(struct inode * inode,struct file * file)3902 ftrace_avail_open(struct inode *inode, struct file *file)
3903 {
3904 struct ftrace_iterator *iter;
3905 int ret;
3906
3907 ret = security_locked_down(LOCKDOWN_TRACEFS);
3908 if (ret)
3909 return ret;
3910
3911 if (unlikely(ftrace_disabled))
3912 return -ENODEV;
3913
3914 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3915 if (!iter)
3916 return -ENOMEM;
3917
3918 iter->pg = ftrace_pages_start;
3919 iter->ops = &global_ops;
3920
3921 return 0;
3922 }
3923
3924 static int
ftrace_enabled_open(struct inode * inode,struct file * file)3925 ftrace_enabled_open(struct inode *inode, struct file *file)
3926 {
3927 struct ftrace_iterator *iter;
3928
3929 /*
3930 * This shows us what functions are currently being
3931 * traced and by what. Not sure if we want lockdown
3932 * to hide such critical information for an admin.
3933 * Although, perhaps it can show information we don't
3934 * want people to see, but if something is tracing
3935 * something, we probably want to know about it.
3936 */
3937
3938 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3939 if (!iter)
3940 return -ENOMEM;
3941
3942 iter->pg = ftrace_pages_start;
3943 iter->flags = FTRACE_ITER_ENABLED;
3944 iter->ops = &global_ops;
3945
3946 return 0;
3947 }
3948
3949 static int
ftrace_touched_open(struct inode * inode,struct file * file)3950 ftrace_touched_open(struct inode *inode, struct file *file)
3951 {
3952 struct ftrace_iterator *iter;
3953
3954 /*
3955 * This shows us what functions have ever been enabled
3956 * (traced, direct, patched, etc). Not sure if we want lockdown
3957 * to hide such critical information for an admin.
3958 * Although, perhaps it can show information we don't
3959 * want people to see, but if something had traced
3960 * something, we probably want to know about it.
3961 */
3962
3963 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3964 if (!iter)
3965 return -ENOMEM;
3966
3967 iter->pg = ftrace_pages_start;
3968 iter->flags = FTRACE_ITER_TOUCHED;
3969 iter->ops = &global_ops;
3970
3971 return 0;
3972 }
3973
3974 static int
ftrace_avail_addrs_open(struct inode * inode,struct file * file)3975 ftrace_avail_addrs_open(struct inode *inode, struct file *file)
3976 {
3977 struct ftrace_iterator *iter;
3978 int ret;
3979
3980 ret = security_locked_down(LOCKDOWN_TRACEFS);
3981 if (ret)
3982 return ret;
3983
3984 if (unlikely(ftrace_disabled))
3985 return -ENODEV;
3986
3987 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3988 if (!iter)
3989 return -ENOMEM;
3990
3991 iter->pg = ftrace_pages_start;
3992 iter->flags = FTRACE_ITER_ADDRS;
3993 iter->ops = &global_ops;
3994
3995 return 0;
3996 }
3997
3998 /**
3999 * ftrace_regex_open - initialize function tracer filter files
4000 * @ops: The ftrace_ops that hold the hash filters
4001 * @flag: The type of filter to process
4002 * @inode: The inode, usually passed in to your open routine
4003 * @file: The file, usually passed in to your open routine
4004 *
4005 * ftrace_regex_open() initializes the filter files for the
4006 * @ops. Depending on @flag it may process the filter hash or
4007 * the notrace hash of @ops. With this called from the open
4008 * routine, you can use ftrace_filter_write() for the write
4009 * routine if @flag has FTRACE_ITER_FILTER set, or
4010 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
4011 * tracing_lseek() should be used as the lseek routine, and
4012 * release must call ftrace_regex_release().
4013 *
4014 * Returns: 0 on success or a negative errno value on failure
4015 */
4016 int
ftrace_regex_open(struct ftrace_ops * ops,int flag,struct inode * inode,struct file * file)4017 ftrace_regex_open(struct ftrace_ops *ops, int flag,
4018 struct inode *inode, struct file *file)
4019 {
4020 struct ftrace_iterator *iter;
4021 struct ftrace_hash *hash;
4022 struct list_head *mod_head;
4023 struct trace_array *tr = ops->private;
4024 int ret = -ENOMEM;
4025
4026 ftrace_ops_init(ops);
4027
4028 if (unlikely(ftrace_disabled))
4029 return -ENODEV;
4030
4031 if (tracing_check_open_get_tr(tr))
4032 return -ENODEV;
4033
4034 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4035 if (!iter)
4036 goto out;
4037
4038 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX))
4039 goto out;
4040
4041 iter->ops = ops;
4042 iter->flags = flag;
4043 iter->tr = tr;
4044
4045 mutex_lock(&ops->func_hash->regex_lock);
4046
4047 if (flag & FTRACE_ITER_NOTRACE) {
4048 hash = ops->func_hash->notrace_hash;
4049 mod_head = tr ? &tr->mod_notrace : NULL;
4050 } else {
4051 hash = ops->func_hash->filter_hash;
4052 mod_head = tr ? &tr->mod_trace : NULL;
4053 }
4054
4055 iter->mod_list = mod_head;
4056
4057 if (file->f_mode & FMODE_WRITE) {
4058 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
4059
4060 if (file->f_flags & O_TRUNC) {
4061 iter->hash = alloc_ftrace_hash(size_bits);
4062 clear_ftrace_mod_list(mod_head);
4063 } else {
4064 iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
4065 }
4066
4067 if (!iter->hash) {
4068 trace_parser_put(&iter->parser);
4069 goto out_unlock;
4070 }
4071 } else
4072 iter->hash = hash;
4073
4074 ret = 0;
4075
4076 if (file->f_mode & FMODE_READ) {
4077 iter->pg = ftrace_pages_start;
4078
4079 ret = seq_open(file, &show_ftrace_seq_ops);
4080 if (!ret) {
4081 struct seq_file *m = file->private_data;
4082 m->private = iter;
4083 } else {
4084 /* Failed */
4085 free_ftrace_hash(iter->hash);
4086 trace_parser_put(&iter->parser);
4087 }
4088 } else
4089 file->private_data = iter;
4090
4091 out_unlock:
4092 mutex_unlock(&ops->func_hash->regex_lock);
4093
4094 out:
4095 if (ret) {
4096 kfree(iter);
4097 if (tr)
4098 trace_array_put(tr);
4099 }
4100
4101 return ret;
4102 }
4103
4104 static int
ftrace_filter_open(struct inode * inode,struct file * file)4105 ftrace_filter_open(struct inode *inode, struct file *file)
4106 {
4107 struct ftrace_ops *ops = inode->i_private;
4108
4109 /* Checks for tracefs lockdown */
4110 return ftrace_regex_open(ops,
4111 FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
4112 inode, file);
4113 }
4114
4115 static int
ftrace_notrace_open(struct inode * inode,struct file * file)4116 ftrace_notrace_open(struct inode *inode, struct file *file)
4117 {
4118 struct ftrace_ops *ops = inode->i_private;
4119
4120 /* Checks for tracefs lockdown */
4121 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
4122 inode, file);
4123 }
4124
4125 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
4126 struct ftrace_glob {
4127 char *search;
4128 unsigned len;
4129 int type;
4130 };
4131
4132 /*
4133 * If symbols in an architecture don't correspond exactly to the user-visible
4134 * name of what they represent, it is possible to define this function to
4135 * perform the necessary adjustments.
4136 */
arch_ftrace_match_adjust(char * str,const char * search)4137 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
4138 {
4139 return str;
4140 }
4141
ftrace_match(char * str,struct ftrace_glob * g)4142 static int ftrace_match(char *str, struct ftrace_glob *g)
4143 {
4144 int matched = 0;
4145 int slen;
4146
4147 str = arch_ftrace_match_adjust(str, g->search);
4148
4149 switch (g->type) {
4150 case MATCH_FULL:
4151 if (strcmp(str, g->search) == 0)
4152 matched = 1;
4153 break;
4154 case MATCH_FRONT_ONLY:
4155 if (strncmp(str, g->search, g->len) == 0)
4156 matched = 1;
4157 break;
4158 case MATCH_MIDDLE_ONLY:
4159 if (strstr(str, g->search))
4160 matched = 1;
4161 break;
4162 case MATCH_END_ONLY:
4163 slen = strlen(str);
4164 if (slen >= g->len &&
4165 memcmp(str + slen - g->len, g->search, g->len) == 0)
4166 matched = 1;
4167 break;
4168 case MATCH_GLOB:
4169 if (glob_match(g->search, str))
4170 matched = 1;
4171 break;
4172 }
4173
4174 return matched;
4175 }
4176
4177 static int
enter_record(struct ftrace_hash * hash,struct dyn_ftrace * rec,int clear_filter)4178 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
4179 {
4180 struct ftrace_func_entry *entry;
4181 int ret = 0;
4182
4183 entry = ftrace_lookup_ip(hash, rec->ip);
4184 if (clear_filter) {
4185 /* Do nothing if it doesn't exist */
4186 if (!entry)
4187 return 0;
4188
4189 free_hash_entry(hash, entry);
4190 } else {
4191 /* Do nothing if it exists */
4192 if (entry)
4193 return 0;
4194 if (add_hash_entry(hash, rec->ip) == NULL)
4195 ret = -ENOMEM;
4196 }
4197 return ret;
4198 }
4199
4200 static int
add_rec_by_index(struct ftrace_hash * hash,struct ftrace_glob * func_g,int clear_filter)4201 add_rec_by_index(struct ftrace_hash *hash, struct ftrace_glob *func_g,
4202 int clear_filter)
4203 {
4204 long index = simple_strtoul(func_g->search, NULL, 0);
4205 struct ftrace_page *pg;
4206 struct dyn_ftrace *rec;
4207
4208 /* The index starts at 1 */
4209 if (--index < 0)
4210 return 0;
4211
4212 do_for_each_ftrace_rec(pg, rec) {
4213 if (pg->index <= index) {
4214 index -= pg->index;
4215 /* this is a double loop, break goes to the next page */
4216 break;
4217 }
4218 rec = &pg->records[index];
4219 enter_record(hash, rec, clear_filter);
4220 return 1;
4221 } while_for_each_ftrace_rec();
4222 return 0;
4223 }
4224
4225 #ifdef FTRACE_MCOUNT_MAX_OFFSET
lookup_ip(unsigned long ip,char ** modname,char * str)4226 static int lookup_ip(unsigned long ip, char **modname, char *str)
4227 {
4228 unsigned long offset;
4229
4230 kallsyms_lookup(ip, NULL, &offset, modname, str);
4231 if (offset > FTRACE_MCOUNT_MAX_OFFSET)
4232 return -1;
4233 return 0;
4234 }
4235 #else
lookup_ip(unsigned long ip,char ** modname,char * str)4236 static int lookup_ip(unsigned long ip, char **modname, char *str)
4237 {
4238 kallsyms_lookup(ip, NULL, NULL, modname, str);
4239 return 0;
4240 }
4241 #endif
4242
4243 static int
ftrace_match_record(struct dyn_ftrace * rec,struct ftrace_glob * func_g,struct ftrace_glob * mod_g,int exclude_mod)4244 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
4245 struct ftrace_glob *mod_g, int exclude_mod)
4246 {
4247 char str[KSYM_SYMBOL_LEN];
4248 char *modname;
4249
4250 if (lookup_ip(rec->ip, &modname, str)) {
4251 /* This should only happen when a rec is disabled */
4252 WARN_ON_ONCE(system_state == SYSTEM_RUNNING &&
4253 !(rec->flags & FTRACE_FL_DISABLED));
4254 return 0;
4255 }
4256
4257 if (mod_g) {
4258 int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
4259
4260 /* blank module name to match all modules */
4261 if (!mod_g->len) {
4262 /* blank module globbing: modname xor exclude_mod */
4263 if (!exclude_mod != !modname)
4264 goto func_match;
4265 return 0;
4266 }
4267
4268 /*
4269 * exclude_mod is set to trace everything but the given
4270 * module. If it is set and the module matches, then
4271 * return 0. If it is not set, and the module doesn't match
4272 * also return 0. Otherwise, check the function to see if
4273 * that matches.
4274 */
4275 if (!mod_matches == !exclude_mod)
4276 return 0;
4277 func_match:
4278 /* blank search means to match all funcs in the mod */
4279 if (!func_g->len)
4280 return 1;
4281 }
4282
4283 return ftrace_match(str, func_g);
4284 }
4285
4286 static int
match_records(struct ftrace_hash * hash,char * func,int len,char * mod)4287 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
4288 {
4289 struct ftrace_page *pg;
4290 struct dyn_ftrace *rec;
4291 struct ftrace_glob func_g = { .type = MATCH_FULL };
4292 struct ftrace_glob mod_g = { .type = MATCH_FULL };
4293 struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
4294 int exclude_mod = 0;
4295 int found = 0;
4296 int ret;
4297 int clear_filter = 0;
4298
4299 if (func) {
4300 func_g.type = filter_parse_regex(func, len, &func_g.search,
4301 &clear_filter);
4302 func_g.len = strlen(func_g.search);
4303 }
4304
4305 if (mod) {
4306 mod_g.type = filter_parse_regex(mod, strlen(mod),
4307 &mod_g.search, &exclude_mod);
4308 mod_g.len = strlen(mod_g.search);
4309 }
4310
4311 mutex_lock(&ftrace_lock);
4312
4313 if (unlikely(ftrace_disabled))
4314 goto out_unlock;
4315
4316 if (func_g.type == MATCH_INDEX) {
4317 found = add_rec_by_index(hash, &func_g, clear_filter);
4318 goto out_unlock;
4319 }
4320
4321 do_for_each_ftrace_rec(pg, rec) {
4322
4323 if (rec->flags & FTRACE_FL_DISABLED)
4324 continue;
4325
4326 if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
4327 ret = enter_record(hash, rec, clear_filter);
4328 if (ret < 0) {
4329 found = ret;
4330 goto out_unlock;
4331 }
4332 found = 1;
4333 }
4334 cond_resched();
4335 } while_for_each_ftrace_rec();
4336 out_unlock:
4337 mutex_unlock(&ftrace_lock);
4338
4339 return found;
4340 }
4341
4342 static int
ftrace_match_records(struct ftrace_hash * hash,char * buff,int len)4343 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
4344 {
4345 return match_records(hash, buff, len, NULL);
4346 }
4347
ftrace_ops_update_code(struct ftrace_ops * ops,struct ftrace_ops_hash * old_hash)4348 static void ftrace_ops_update_code(struct ftrace_ops *ops,
4349 struct ftrace_ops_hash *old_hash)
4350 {
4351 struct ftrace_ops *op;
4352
4353 if (!ftrace_enabled)
4354 return;
4355
4356 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
4357 ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
4358 return;
4359 }
4360
4361 /*
4362 * If this is the shared global_ops filter, then we need to
4363 * check if there is another ops that shares it, is enabled.
4364 * If so, we still need to run the modify code.
4365 */
4366 if (ops->func_hash != &global_ops.local_hash)
4367 return;
4368
4369 do_for_each_ftrace_op(op, ftrace_ops_list) {
4370 if (op->func_hash == &global_ops.local_hash &&
4371 op->flags & FTRACE_OPS_FL_ENABLED) {
4372 ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
4373 /* Only need to do this once */
4374 return;
4375 }
4376 } while_for_each_ftrace_op(op);
4377 }
4378
ftrace_hash_move_and_update_ops(struct ftrace_ops * ops,struct ftrace_hash ** orig_hash,struct ftrace_hash * hash,int enable)4379 static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
4380 struct ftrace_hash **orig_hash,
4381 struct ftrace_hash *hash,
4382 int enable)
4383 {
4384 struct ftrace_ops_hash old_hash_ops;
4385 struct ftrace_hash *old_hash;
4386 int ret;
4387
4388 old_hash = *orig_hash;
4389 old_hash_ops.filter_hash = ops->func_hash->filter_hash;
4390 old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
4391 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
4392 if (!ret) {
4393 ftrace_ops_update_code(ops, &old_hash_ops);
4394 free_ftrace_hash_rcu(old_hash);
4395 }
4396 return ret;
4397 }
4398
module_exists(const char * module)4399 static bool module_exists(const char *module)
4400 {
4401 /* All modules have the symbol __this_module */
4402 static const char this_mod[] = "__this_module";
4403 char modname[MAX_PARAM_PREFIX_LEN + sizeof(this_mod) + 2];
4404 unsigned long val;
4405 int n;
4406
4407 n = snprintf(modname, sizeof(modname), "%s:%s", module, this_mod);
4408
4409 if (n > sizeof(modname) - 1)
4410 return false;
4411
4412 val = module_kallsyms_lookup_name(modname);
4413 return val != 0;
4414 }
4415
cache_mod(struct trace_array * tr,const char * func,char * module,int enable)4416 static int cache_mod(struct trace_array *tr,
4417 const char *func, char *module, int enable)
4418 {
4419 struct ftrace_mod_load *ftrace_mod, *n;
4420 struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
4421 int ret;
4422
4423 mutex_lock(&ftrace_lock);
4424
4425 /* We do not cache inverse filters */
4426 if (func[0] == '!') {
4427 func++;
4428 ret = -EINVAL;
4429
4430 /* Look to remove this hash */
4431 list_for_each_entry_safe(ftrace_mod, n, head, list) {
4432 if (strcmp(ftrace_mod->module, module) != 0)
4433 continue;
4434
4435 /* no func matches all */
4436 if (strcmp(func, "*") == 0 ||
4437 (ftrace_mod->func &&
4438 strcmp(ftrace_mod->func, func) == 0)) {
4439 ret = 0;
4440 free_ftrace_mod(ftrace_mod);
4441 continue;
4442 }
4443 }
4444 goto out;
4445 }
4446
4447 ret = -EINVAL;
4448 /* We only care about modules that have not been loaded yet */
4449 if (module_exists(module))
4450 goto out;
4451
4452 /* Save this string off, and execute it when the module is loaded */
4453 ret = ftrace_add_mod(tr, func, module, enable);
4454 out:
4455 mutex_unlock(&ftrace_lock);
4456
4457 return ret;
4458 }
4459
4460 static int
4461 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4462 int reset, int enable);
4463
4464 #ifdef CONFIG_MODULES
process_mod_list(struct list_head * head,struct ftrace_ops * ops,char * mod,bool enable)4465 static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
4466 char *mod, bool enable)
4467 {
4468 struct ftrace_mod_load *ftrace_mod, *n;
4469 struct ftrace_hash **orig_hash, *new_hash;
4470 LIST_HEAD(process_mods);
4471 char *func;
4472
4473 mutex_lock(&ops->func_hash->regex_lock);
4474
4475 if (enable)
4476 orig_hash = &ops->func_hash->filter_hash;
4477 else
4478 orig_hash = &ops->func_hash->notrace_hash;
4479
4480 new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
4481 *orig_hash);
4482 if (!new_hash)
4483 goto out; /* warn? */
4484
4485 mutex_lock(&ftrace_lock);
4486
4487 list_for_each_entry_safe(ftrace_mod, n, head, list) {
4488
4489 if (strcmp(ftrace_mod->module, mod) != 0)
4490 continue;
4491
4492 if (ftrace_mod->func)
4493 func = kstrdup(ftrace_mod->func, GFP_KERNEL);
4494 else
4495 func = kstrdup("*", GFP_KERNEL);
4496
4497 if (!func) /* warn? */
4498 continue;
4499
4500 list_move(&ftrace_mod->list, &process_mods);
4501
4502 /* Use the newly allocated func, as it may be "*" */
4503 kfree(ftrace_mod->func);
4504 ftrace_mod->func = func;
4505 }
4506
4507 mutex_unlock(&ftrace_lock);
4508
4509 list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
4510
4511 func = ftrace_mod->func;
4512
4513 /* Grabs ftrace_lock, which is why we have this extra step */
4514 match_records(new_hash, func, strlen(func), mod);
4515 free_ftrace_mod(ftrace_mod);
4516 }
4517
4518 if (enable && list_empty(head))
4519 new_hash->flags &= ~FTRACE_HASH_FL_MOD;
4520
4521 mutex_lock(&ftrace_lock);
4522
4523 ftrace_hash_move_and_update_ops(ops, orig_hash,
4524 new_hash, enable);
4525 mutex_unlock(&ftrace_lock);
4526
4527 out:
4528 mutex_unlock(&ops->func_hash->regex_lock);
4529
4530 free_ftrace_hash(new_hash);
4531 }
4532
process_cached_mods(const char * mod_name)4533 static void process_cached_mods(const char *mod_name)
4534 {
4535 struct trace_array *tr;
4536 char *mod;
4537
4538 mod = kstrdup(mod_name, GFP_KERNEL);
4539 if (!mod)
4540 return;
4541
4542 mutex_lock(&trace_types_lock);
4543 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
4544 if (!list_empty(&tr->mod_trace))
4545 process_mod_list(&tr->mod_trace, tr->ops, mod, true);
4546 if (!list_empty(&tr->mod_notrace))
4547 process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
4548 }
4549 mutex_unlock(&trace_types_lock);
4550
4551 kfree(mod);
4552 }
4553 #endif
4554
4555 /*
4556 * We register the module command as a template to show others how
4557 * to register the a command as well.
4558 */
4559
4560 static int
ftrace_mod_callback(struct trace_array * tr,struct ftrace_hash * hash,char * func_orig,char * cmd,char * module,int enable)4561 ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
4562 char *func_orig, char *cmd, char *module, int enable)
4563 {
4564 char *func;
4565 int ret;
4566
4567 /* match_records() modifies func, and we need the original */
4568 func = kstrdup(func_orig, GFP_KERNEL);
4569 if (!func)
4570 return -ENOMEM;
4571
4572 /*
4573 * cmd == 'mod' because we only registered this func
4574 * for the 'mod' ftrace_func_command.
4575 * But if you register one func with multiple commands,
4576 * you can tell which command was used by the cmd
4577 * parameter.
4578 */
4579 ret = match_records(hash, func, strlen(func), module);
4580 kfree(func);
4581
4582 if (!ret)
4583 return cache_mod(tr, func_orig, module, enable);
4584 if (ret < 0)
4585 return ret;
4586 return 0;
4587 }
4588
4589 static struct ftrace_func_command ftrace_mod_cmd = {
4590 .name = "mod",
4591 .func = ftrace_mod_callback,
4592 };
4593
ftrace_mod_cmd_init(void)4594 static int __init ftrace_mod_cmd_init(void)
4595 {
4596 return register_ftrace_command(&ftrace_mod_cmd);
4597 }
4598 core_initcall(ftrace_mod_cmd_init);
4599
function_trace_probe_call(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)4600 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
4601 struct ftrace_ops *op, struct ftrace_regs *fregs)
4602 {
4603 struct ftrace_probe_ops *probe_ops;
4604 struct ftrace_func_probe *probe;
4605
4606 probe = container_of(op, struct ftrace_func_probe, ops);
4607 probe_ops = probe->probe_ops;
4608
4609 /*
4610 * Disable preemption for these calls to prevent a RCU grace
4611 * period. This syncs the hash iteration and freeing of items
4612 * on the hash. rcu_read_lock is too dangerous here.
4613 */
4614 preempt_disable_notrace();
4615 probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
4616 preempt_enable_notrace();
4617 }
4618
4619 struct ftrace_func_map {
4620 struct ftrace_func_entry entry;
4621 void *data;
4622 };
4623
4624 struct ftrace_func_mapper {
4625 struct ftrace_hash hash;
4626 };
4627
4628 /**
4629 * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
4630 *
4631 * Returns: a ftrace_func_mapper descriptor that can be used to map ips to data.
4632 */
allocate_ftrace_func_mapper(void)4633 struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
4634 {
4635 struct ftrace_hash *hash;
4636
4637 /*
4638 * The mapper is simply a ftrace_hash, but since the entries
4639 * in the hash are not ftrace_func_entry type, we define it
4640 * as a separate structure.
4641 */
4642 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4643 return (struct ftrace_func_mapper *)hash;
4644 }
4645
4646 /**
4647 * ftrace_func_mapper_find_ip - Find some data mapped to an ip
4648 * @mapper: The mapper that has the ip maps
4649 * @ip: the instruction pointer to find the data for
4650 *
4651 * Returns: the data mapped to @ip if found otherwise NULL. The return
4652 * is actually the address of the mapper data pointer. The address is
4653 * returned for use cases where the data is no bigger than a long, and
4654 * the user can use the data pointer as its data instead of having to
4655 * allocate more memory for the reference.
4656 */
ftrace_func_mapper_find_ip(struct ftrace_func_mapper * mapper,unsigned long ip)4657 void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
4658 unsigned long ip)
4659 {
4660 struct ftrace_func_entry *entry;
4661 struct ftrace_func_map *map;
4662
4663 entry = ftrace_lookup_ip(&mapper->hash, ip);
4664 if (!entry)
4665 return NULL;
4666
4667 map = (struct ftrace_func_map *)entry;
4668 return &map->data;
4669 }
4670
4671 /**
4672 * ftrace_func_mapper_add_ip - Map some data to an ip
4673 * @mapper: The mapper that has the ip maps
4674 * @ip: The instruction pointer address to map @data to
4675 * @data: The data to map to @ip
4676 *
4677 * Returns: 0 on success otherwise an error.
4678 */
ftrace_func_mapper_add_ip(struct ftrace_func_mapper * mapper,unsigned long ip,void * data)4679 int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
4680 unsigned long ip, void *data)
4681 {
4682 struct ftrace_func_entry *entry;
4683 struct ftrace_func_map *map;
4684
4685 entry = ftrace_lookup_ip(&mapper->hash, ip);
4686 if (entry)
4687 return -EBUSY;
4688
4689 map = kmalloc(sizeof(*map), GFP_KERNEL);
4690 if (!map)
4691 return -ENOMEM;
4692
4693 map->entry.ip = ip;
4694 map->data = data;
4695
4696 __add_hash_entry(&mapper->hash, &map->entry);
4697
4698 return 0;
4699 }
4700
4701 /**
4702 * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
4703 * @mapper: The mapper that has the ip maps
4704 * @ip: The instruction pointer address to remove the data from
4705 *
4706 * Returns: the data if it is found, otherwise NULL.
4707 * Note, if the data pointer is used as the data itself, (see
4708 * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
4709 * if the data pointer was set to zero.
4710 */
ftrace_func_mapper_remove_ip(struct ftrace_func_mapper * mapper,unsigned long ip)4711 void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
4712 unsigned long ip)
4713 {
4714 struct ftrace_func_entry *entry;
4715 struct ftrace_func_map *map;
4716 void *data;
4717
4718 entry = ftrace_lookup_ip(&mapper->hash, ip);
4719 if (!entry)
4720 return NULL;
4721
4722 map = (struct ftrace_func_map *)entry;
4723 data = map->data;
4724
4725 remove_hash_entry(&mapper->hash, entry);
4726 kfree(entry);
4727
4728 return data;
4729 }
4730
4731 /**
4732 * free_ftrace_func_mapper - free a mapping of ips and data
4733 * @mapper: The mapper that has the ip maps
4734 * @free_func: A function to be called on each data item.
4735 *
4736 * This is used to free the function mapper. The @free_func is optional
4737 * and can be used if the data needs to be freed as well.
4738 */
free_ftrace_func_mapper(struct ftrace_func_mapper * mapper,ftrace_mapper_func free_func)4739 void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
4740 ftrace_mapper_func free_func)
4741 {
4742 struct ftrace_func_entry *entry;
4743 struct ftrace_func_map *map;
4744 struct hlist_head *hhd;
4745 int size, i;
4746
4747 if (!mapper)
4748 return;
4749
4750 if (free_func && mapper->hash.count) {
4751 size = 1 << mapper->hash.size_bits;
4752 for (i = 0; i < size; i++) {
4753 hhd = &mapper->hash.buckets[i];
4754 hlist_for_each_entry(entry, hhd, hlist) {
4755 map = (struct ftrace_func_map *)entry;
4756 free_func(map);
4757 }
4758 }
4759 }
4760 free_ftrace_hash(&mapper->hash);
4761 }
4762
release_probe(struct ftrace_func_probe * probe)4763 static void release_probe(struct ftrace_func_probe *probe)
4764 {
4765 struct ftrace_probe_ops *probe_ops;
4766
4767 mutex_lock(&ftrace_lock);
4768
4769 WARN_ON(probe->ref <= 0);
4770
4771 /* Subtract the ref that was used to protect this instance */
4772 probe->ref--;
4773
4774 if (!probe->ref) {
4775 probe_ops = probe->probe_ops;
4776 /*
4777 * Sending zero as ip tells probe_ops to free
4778 * the probe->data itself
4779 */
4780 if (probe_ops->free)
4781 probe_ops->free(probe_ops, probe->tr, 0, probe->data);
4782 list_del(&probe->list);
4783 kfree(probe);
4784 }
4785 mutex_unlock(&ftrace_lock);
4786 }
4787
acquire_probe_locked(struct ftrace_func_probe * probe)4788 static void acquire_probe_locked(struct ftrace_func_probe *probe)
4789 {
4790 /*
4791 * Add one ref to keep it from being freed when releasing the
4792 * ftrace_lock mutex.
4793 */
4794 probe->ref++;
4795 }
4796
4797 int
register_ftrace_function_probe(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops,void * data)4798 register_ftrace_function_probe(char *glob, struct trace_array *tr,
4799 struct ftrace_probe_ops *probe_ops,
4800 void *data)
4801 {
4802 struct ftrace_func_probe *probe = NULL, *iter;
4803 struct ftrace_func_entry *entry;
4804 struct ftrace_hash **orig_hash;
4805 struct ftrace_hash *old_hash;
4806 struct ftrace_hash *hash;
4807 int count = 0;
4808 int size;
4809 int ret;
4810 int i;
4811
4812 if (WARN_ON(!tr))
4813 return -EINVAL;
4814
4815 /* We do not support '!' for function probes */
4816 if (WARN_ON(glob[0] == '!'))
4817 return -EINVAL;
4818
4819
4820 mutex_lock(&ftrace_lock);
4821 /* Check if the probe_ops is already registered */
4822 list_for_each_entry(iter, &tr->func_probes, list) {
4823 if (iter->probe_ops == probe_ops) {
4824 probe = iter;
4825 break;
4826 }
4827 }
4828 if (!probe) {
4829 probe = kzalloc(sizeof(*probe), GFP_KERNEL);
4830 if (!probe) {
4831 mutex_unlock(&ftrace_lock);
4832 return -ENOMEM;
4833 }
4834 probe->probe_ops = probe_ops;
4835 probe->ops.func = function_trace_probe_call;
4836 probe->tr = tr;
4837 ftrace_ops_init(&probe->ops);
4838 list_add(&probe->list, &tr->func_probes);
4839 }
4840
4841 acquire_probe_locked(probe);
4842
4843 mutex_unlock(&ftrace_lock);
4844
4845 /*
4846 * Note, there's a small window here that the func_hash->filter_hash
4847 * may be NULL or empty. Need to be careful when reading the loop.
4848 */
4849 mutex_lock(&probe->ops.func_hash->regex_lock);
4850
4851 orig_hash = &probe->ops.func_hash->filter_hash;
4852 old_hash = *orig_hash;
4853 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4854
4855 if (!hash) {
4856 ret = -ENOMEM;
4857 goto out;
4858 }
4859
4860 ret = ftrace_match_records(hash, glob, strlen(glob));
4861
4862 /* Nothing found? */
4863 if (!ret)
4864 ret = -EINVAL;
4865
4866 if (ret < 0)
4867 goto out;
4868
4869 size = 1 << hash->size_bits;
4870 for (i = 0; i < size; i++) {
4871 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4872 if (ftrace_lookup_ip(old_hash, entry->ip))
4873 continue;
4874 /*
4875 * The caller might want to do something special
4876 * for each function we find. We call the callback
4877 * to give the caller an opportunity to do so.
4878 */
4879 if (probe_ops->init) {
4880 ret = probe_ops->init(probe_ops, tr,
4881 entry->ip, data,
4882 &probe->data);
4883 if (ret < 0) {
4884 if (probe_ops->free && count)
4885 probe_ops->free(probe_ops, tr,
4886 0, probe->data);
4887 probe->data = NULL;
4888 goto out;
4889 }
4890 }
4891 count++;
4892 }
4893 }
4894
4895 mutex_lock(&ftrace_lock);
4896
4897 if (!count) {
4898 /* Nothing was added? */
4899 ret = -EINVAL;
4900 goto out_unlock;
4901 }
4902
4903 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4904 hash, 1);
4905 if (ret < 0)
4906 goto err_unlock;
4907
4908 /* One ref for each new function traced */
4909 probe->ref += count;
4910
4911 if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
4912 ret = ftrace_startup(&probe->ops, 0);
4913
4914 out_unlock:
4915 mutex_unlock(&ftrace_lock);
4916
4917 if (!ret)
4918 ret = count;
4919 out:
4920 mutex_unlock(&probe->ops.func_hash->regex_lock);
4921 free_ftrace_hash(hash);
4922
4923 release_probe(probe);
4924
4925 return ret;
4926
4927 err_unlock:
4928 if (!probe_ops->free || !count)
4929 goto out_unlock;
4930
4931 /* Failed to do the move, need to call the free functions */
4932 for (i = 0; i < size; i++) {
4933 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4934 if (ftrace_lookup_ip(old_hash, entry->ip))
4935 continue;
4936 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4937 }
4938 }
4939 goto out_unlock;
4940 }
4941
4942 int
unregister_ftrace_function_probe_func(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops)4943 unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
4944 struct ftrace_probe_ops *probe_ops)
4945 {
4946 struct ftrace_func_probe *probe = NULL, *iter;
4947 struct ftrace_ops_hash old_hash_ops;
4948 struct ftrace_func_entry *entry;
4949 struct ftrace_glob func_g;
4950 struct ftrace_hash **orig_hash;
4951 struct ftrace_hash *old_hash;
4952 struct ftrace_hash *hash = NULL;
4953 struct hlist_node *tmp;
4954 struct hlist_head hhd;
4955 char str[KSYM_SYMBOL_LEN];
4956 int count = 0;
4957 int i, ret = -ENODEV;
4958 int size;
4959
4960 if (!glob || !strlen(glob) || !strcmp(glob, "*"))
4961 func_g.search = NULL;
4962 else {
4963 int not;
4964
4965 func_g.type = filter_parse_regex(glob, strlen(glob),
4966 &func_g.search, ¬);
4967 func_g.len = strlen(func_g.search);
4968
4969 /* we do not support '!' for function probes */
4970 if (WARN_ON(not))
4971 return -EINVAL;
4972 }
4973
4974 mutex_lock(&ftrace_lock);
4975 /* Check if the probe_ops is already registered */
4976 list_for_each_entry(iter, &tr->func_probes, list) {
4977 if (iter->probe_ops == probe_ops) {
4978 probe = iter;
4979 break;
4980 }
4981 }
4982 if (!probe)
4983 goto err_unlock_ftrace;
4984
4985 ret = -EINVAL;
4986 if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
4987 goto err_unlock_ftrace;
4988
4989 acquire_probe_locked(probe);
4990
4991 mutex_unlock(&ftrace_lock);
4992
4993 mutex_lock(&probe->ops.func_hash->regex_lock);
4994
4995 orig_hash = &probe->ops.func_hash->filter_hash;
4996 old_hash = *orig_hash;
4997
4998 if (ftrace_hash_empty(old_hash))
4999 goto out_unlock;
5000
5001 old_hash_ops.filter_hash = old_hash;
5002 /* Probes only have filters */
5003 old_hash_ops.notrace_hash = NULL;
5004
5005 ret = -ENOMEM;
5006 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
5007 if (!hash)
5008 goto out_unlock;
5009
5010 INIT_HLIST_HEAD(&hhd);
5011
5012 size = 1 << hash->size_bits;
5013 for (i = 0; i < size; i++) {
5014 hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
5015
5016 if (func_g.search) {
5017 kallsyms_lookup(entry->ip, NULL, NULL,
5018 NULL, str);
5019 if (!ftrace_match(str, &func_g))
5020 continue;
5021 }
5022 count++;
5023 remove_hash_entry(hash, entry);
5024 hlist_add_head(&entry->hlist, &hhd);
5025 }
5026 }
5027
5028 /* Nothing found? */
5029 if (!count) {
5030 ret = -EINVAL;
5031 goto out_unlock;
5032 }
5033
5034 mutex_lock(&ftrace_lock);
5035
5036 WARN_ON(probe->ref < count);
5037
5038 probe->ref -= count;
5039
5040 if (ftrace_hash_empty(hash))
5041 ftrace_shutdown(&probe->ops, 0);
5042
5043 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
5044 hash, 1);
5045
5046 /* still need to update the function call sites */
5047 if (ftrace_enabled && !ftrace_hash_empty(hash))
5048 ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
5049 &old_hash_ops);
5050 synchronize_rcu();
5051
5052 hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
5053 hlist_del(&entry->hlist);
5054 if (probe_ops->free)
5055 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
5056 kfree(entry);
5057 }
5058 mutex_unlock(&ftrace_lock);
5059
5060 out_unlock:
5061 mutex_unlock(&probe->ops.func_hash->regex_lock);
5062 free_ftrace_hash(hash);
5063
5064 release_probe(probe);
5065
5066 return ret;
5067
5068 err_unlock_ftrace:
5069 mutex_unlock(&ftrace_lock);
5070 return ret;
5071 }
5072
clear_ftrace_function_probes(struct trace_array * tr)5073 void clear_ftrace_function_probes(struct trace_array *tr)
5074 {
5075 struct ftrace_func_probe *probe, *n;
5076
5077 list_for_each_entry_safe(probe, n, &tr->func_probes, list)
5078 unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
5079 }
5080
5081 static LIST_HEAD(ftrace_commands);
5082 static DEFINE_MUTEX(ftrace_cmd_mutex);
5083
5084 /*
5085 * Currently we only register ftrace commands from __init, so mark this
5086 * __init too.
5087 */
register_ftrace_command(struct ftrace_func_command * cmd)5088 __init int register_ftrace_command(struct ftrace_func_command *cmd)
5089 {
5090 struct ftrace_func_command *p;
5091 int ret = 0;
5092
5093 mutex_lock(&ftrace_cmd_mutex);
5094 list_for_each_entry(p, &ftrace_commands, list) {
5095 if (strcmp(cmd->name, p->name) == 0) {
5096 ret = -EBUSY;
5097 goto out_unlock;
5098 }
5099 }
5100 list_add(&cmd->list, &ftrace_commands);
5101 out_unlock:
5102 mutex_unlock(&ftrace_cmd_mutex);
5103
5104 return ret;
5105 }
5106
5107 /*
5108 * Currently we only unregister ftrace commands from __init, so mark
5109 * this __init too.
5110 */
unregister_ftrace_command(struct ftrace_func_command * cmd)5111 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
5112 {
5113 struct ftrace_func_command *p, *n;
5114 int ret = -ENODEV;
5115
5116 mutex_lock(&ftrace_cmd_mutex);
5117 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
5118 if (strcmp(cmd->name, p->name) == 0) {
5119 ret = 0;
5120 list_del_init(&p->list);
5121 goto out_unlock;
5122 }
5123 }
5124 out_unlock:
5125 mutex_unlock(&ftrace_cmd_mutex);
5126
5127 return ret;
5128 }
5129
ftrace_process_regex(struct ftrace_iterator * iter,char * buff,int len,int enable)5130 static int ftrace_process_regex(struct ftrace_iterator *iter,
5131 char *buff, int len, int enable)
5132 {
5133 struct ftrace_hash *hash = iter->hash;
5134 struct trace_array *tr = iter->ops->private;
5135 char *func, *command, *next = buff;
5136 struct ftrace_func_command *p;
5137 int ret = -EINVAL;
5138
5139 func = strsep(&next, ":");
5140
5141 if (!next) {
5142 ret = ftrace_match_records(hash, func, len);
5143 if (!ret)
5144 ret = -EINVAL;
5145 if (ret < 0)
5146 return ret;
5147 return 0;
5148 }
5149
5150 /* command found */
5151
5152 command = strsep(&next, ":");
5153
5154 mutex_lock(&ftrace_cmd_mutex);
5155 list_for_each_entry(p, &ftrace_commands, list) {
5156 if (strcmp(p->name, command) == 0) {
5157 ret = p->func(tr, hash, func, command, next, enable);
5158 goto out_unlock;
5159 }
5160 }
5161 out_unlock:
5162 mutex_unlock(&ftrace_cmd_mutex);
5163
5164 return ret;
5165 }
5166
5167 static ssize_t
ftrace_regex_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos,int enable)5168 ftrace_regex_write(struct file *file, const char __user *ubuf,
5169 size_t cnt, loff_t *ppos, int enable)
5170 {
5171 struct ftrace_iterator *iter;
5172 struct trace_parser *parser;
5173 ssize_t ret, read;
5174
5175 if (!cnt)
5176 return 0;
5177
5178 if (file->f_mode & FMODE_READ) {
5179 struct seq_file *m = file->private_data;
5180 iter = m->private;
5181 } else
5182 iter = file->private_data;
5183
5184 if (unlikely(ftrace_disabled))
5185 return -ENODEV;
5186
5187 /* iter->hash is a local copy, so we don't need regex_lock */
5188
5189 parser = &iter->parser;
5190 read = trace_get_user(parser, ubuf, cnt, ppos);
5191
5192 if (read >= 0 && trace_parser_loaded(parser) &&
5193 !trace_parser_cont(parser)) {
5194 ret = ftrace_process_regex(iter, parser->buffer,
5195 parser->idx, enable);
5196 trace_parser_clear(parser);
5197 if (ret < 0)
5198 goto out;
5199 }
5200
5201 ret = read;
5202 out:
5203 return ret;
5204 }
5205
5206 ssize_t
ftrace_filter_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5207 ftrace_filter_write(struct file *file, const char __user *ubuf,
5208 size_t cnt, loff_t *ppos)
5209 {
5210 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
5211 }
5212
5213 ssize_t
ftrace_notrace_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5214 ftrace_notrace_write(struct file *file, const char __user *ubuf,
5215 size_t cnt, loff_t *ppos)
5216 {
5217 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
5218 }
5219
5220 static int
__ftrace_match_addr(struct ftrace_hash * hash,unsigned long ip,int remove)5221 __ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
5222 {
5223 struct ftrace_func_entry *entry;
5224
5225 ip = ftrace_location(ip);
5226 if (!ip)
5227 return -EINVAL;
5228
5229 if (remove) {
5230 entry = ftrace_lookup_ip(hash, ip);
5231 if (!entry)
5232 return -ENOENT;
5233 free_hash_entry(hash, entry);
5234 return 0;
5235 }
5236
5237 entry = add_hash_entry(hash, ip);
5238 return entry ? 0 : -ENOMEM;
5239 }
5240
5241 static int
ftrace_match_addr(struct ftrace_hash * hash,unsigned long * ips,unsigned int cnt,int remove)5242 ftrace_match_addr(struct ftrace_hash *hash, unsigned long *ips,
5243 unsigned int cnt, int remove)
5244 {
5245 unsigned int i;
5246 int err;
5247
5248 for (i = 0; i < cnt; i++) {
5249 err = __ftrace_match_addr(hash, ips[i], remove);
5250 if (err) {
5251 /*
5252 * This expects the @hash is a temporary hash and if this
5253 * fails the caller must free the @hash.
5254 */
5255 return err;
5256 }
5257 }
5258 return 0;
5259 }
5260
5261 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)5262 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
5263 unsigned long *ips, unsigned int cnt,
5264 int remove, int reset, int enable)
5265 {
5266 struct ftrace_hash **orig_hash;
5267 struct ftrace_hash *hash;
5268 int ret;
5269
5270 if (unlikely(ftrace_disabled))
5271 return -ENODEV;
5272
5273 mutex_lock(&ops->func_hash->regex_lock);
5274
5275 if (enable)
5276 orig_hash = &ops->func_hash->filter_hash;
5277 else
5278 orig_hash = &ops->func_hash->notrace_hash;
5279
5280 if (reset)
5281 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5282 else
5283 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
5284
5285 if (!hash) {
5286 ret = -ENOMEM;
5287 goto out_regex_unlock;
5288 }
5289
5290 if (buf && !ftrace_match_records(hash, buf, len)) {
5291 ret = -EINVAL;
5292 goto out_regex_unlock;
5293 }
5294 if (ips) {
5295 ret = ftrace_match_addr(hash, ips, cnt, remove);
5296 if (ret < 0)
5297 goto out_regex_unlock;
5298 }
5299
5300 mutex_lock(&ftrace_lock);
5301 ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
5302 mutex_unlock(&ftrace_lock);
5303
5304 out_regex_unlock:
5305 mutex_unlock(&ops->func_hash->regex_lock);
5306
5307 free_ftrace_hash(hash);
5308 return ret;
5309 }
5310
5311 static int
ftrace_set_addr(struct ftrace_ops * ops,unsigned long * ips,unsigned int cnt,int remove,int reset,int enable)5312 ftrace_set_addr(struct ftrace_ops *ops, unsigned long *ips, unsigned int cnt,
5313 int remove, int reset, int enable)
5314 {
5315 return ftrace_set_hash(ops, NULL, 0, ips, cnt, remove, reset, enable);
5316 }
5317
5318 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
5319
5320 struct ftrace_direct_func {
5321 struct list_head next;
5322 unsigned long addr;
5323 int count;
5324 };
5325
5326 static LIST_HEAD(ftrace_direct_funcs);
5327
5328 static int register_ftrace_function_nolock(struct ftrace_ops *ops);
5329
5330 /*
5331 * If there are multiple ftrace_ops, use SAVE_REGS by default, so that direct
5332 * call will be jumped from ftrace_regs_caller. Only if the architecture does
5333 * not support ftrace_regs_caller but direct_call, use SAVE_ARGS so that it
5334 * jumps from ftrace_caller for multiple ftrace_ops.
5335 */
5336 #ifndef CONFIG_HAVE_DYNAMIC_FTRACE_WITH_REGS
5337 #define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_ARGS)
5338 #else
5339 #define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_REGS)
5340 #endif
5341
check_direct_multi(struct ftrace_ops * ops)5342 static int check_direct_multi(struct ftrace_ops *ops)
5343 {
5344 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED))
5345 return -EINVAL;
5346 if ((ops->flags & MULTI_FLAGS) != MULTI_FLAGS)
5347 return -EINVAL;
5348 return 0;
5349 }
5350
remove_direct_functions_hash(struct ftrace_hash * hash,unsigned long addr)5351 static void remove_direct_functions_hash(struct ftrace_hash *hash, unsigned long addr)
5352 {
5353 struct ftrace_func_entry *entry, *del;
5354 int size, i;
5355
5356 size = 1 << hash->size_bits;
5357 for (i = 0; i < size; i++) {
5358 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5359 del = __ftrace_lookup_ip(direct_functions, entry->ip);
5360 if (del && del->direct == addr) {
5361 remove_hash_entry(direct_functions, del);
5362 kfree(del);
5363 }
5364 }
5365 }
5366 }
5367
5368 /**
5369 * register_ftrace_direct - Call a custom trampoline directly
5370 * for multiple functions registered in @ops
5371 * @ops: The address of the struct ftrace_ops object
5372 * @addr: The address of the trampoline to call at @ops functions
5373 *
5374 * This is used to connect a direct calls to @addr from the nop locations
5375 * of the functions registered in @ops (with by ftrace_set_filter_ip
5376 * function).
5377 *
5378 * The location that it calls (@addr) must be able to handle a direct call,
5379 * and save the parameters of the function being traced, and restore them
5380 * (or inject new ones if needed), before returning.
5381 *
5382 * Returns:
5383 * 0 on success
5384 * -EINVAL - The @ops object was already registered with this call or
5385 * when there are no functions in @ops object.
5386 * -EBUSY - Another direct function is already attached (there can be only one)
5387 * -ENODEV - @ip does not point to a ftrace nop location (or not supported)
5388 * -ENOMEM - There was an allocation failure.
5389 */
register_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)5390 int register_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
5391 {
5392 struct ftrace_hash *hash, *new_hash = NULL, *free_hash = NULL;
5393 struct ftrace_func_entry *entry, *new;
5394 int err = -EBUSY, size, i;
5395
5396 if (ops->func || ops->trampoline)
5397 return -EINVAL;
5398 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED))
5399 return -EINVAL;
5400 if (ops->flags & FTRACE_OPS_FL_ENABLED)
5401 return -EINVAL;
5402
5403 hash = ops->func_hash->filter_hash;
5404 if (ftrace_hash_empty(hash))
5405 return -EINVAL;
5406
5407 mutex_lock(&direct_mutex);
5408
5409 /* Make sure requested entries are not already registered.. */
5410 size = 1 << hash->size_bits;
5411 for (i = 0; i < size; i++) {
5412 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5413 if (ftrace_find_rec_direct(entry->ip))
5414 goto out_unlock;
5415 }
5416 }
5417
5418 err = -ENOMEM;
5419
5420 /* Make a copy hash to place the new and the old entries in */
5421 size = hash->count + direct_functions->count;
5422 if (size > 32)
5423 size = 32;
5424 new_hash = alloc_ftrace_hash(fls(size));
5425 if (!new_hash)
5426 goto out_unlock;
5427
5428 /* Now copy over the existing direct entries */
5429 size = 1 << direct_functions->size_bits;
5430 for (i = 0; i < size; i++) {
5431 hlist_for_each_entry(entry, &direct_functions->buckets[i], hlist) {
5432 new = add_hash_entry(new_hash, entry->ip);
5433 if (!new)
5434 goto out_unlock;
5435 new->direct = entry->direct;
5436 }
5437 }
5438
5439 /* ... and add the new entries */
5440 size = 1 << hash->size_bits;
5441 for (i = 0; i < size; i++) {
5442 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5443 new = add_hash_entry(new_hash, entry->ip);
5444 if (!new)
5445 goto out_unlock;
5446 /* Update both the copy and the hash entry */
5447 new->direct = addr;
5448 entry->direct = addr;
5449 }
5450 }
5451
5452 free_hash = direct_functions;
5453 rcu_assign_pointer(direct_functions, new_hash);
5454 new_hash = NULL;
5455
5456 ops->func = call_direct_funcs;
5457 ops->flags = MULTI_FLAGS;
5458 ops->trampoline = FTRACE_REGS_ADDR;
5459 ops->direct_call = addr;
5460
5461 err = register_ftrace_function_nolock(ops);
5462
5463 out_unlock:
5464 mutex_unlock(&direct_mutex);
5465
5466 if (free_hash && free_hash != EMPTY_HASH) {
5467 synchronize_rcu_tasks();
5468 free_ftrace_hash(free_hash);
5469 }
5470
5471 if (new_hash)
5472 free_ftrace_hash(new_hash);
5473
5474 return err;
5475 }
5476 EXPORT_SYMBOL_GPL(register_ftrace_direct);
5477
5478 /**
5479 * unregister_ftrace_direct - Remove calls to custom trampoline
5480 * previously registered by register_ftrace_direct for @ops object.
5481 * @ops: The address of the struct ftrace_ops object
5482 *
5483 * This is used to remove a direct calls to @addr from the nop locations
5484 * of the functions registered in @ops (with by ftrace_set_filter_ip
5485 * function).
5486 *
5487 * Returns:
5488 * 0 on success
5489 * -EINVAL - The @ops object was not properly registered.
5490 */
unregister_ftrace_direct(struct ftrace_ops * ops,unsigned long addr,bool free_filters)5491 int unregister_ftrace_direct(struct ftrace_ops *ops, unsigned long addr,
5492 bool free_filters)
5493 {
5494 struct ftrace_hash *hash = ops->func_hash->filter_hash;
5495 int err;
5496
5497 if (check_direct_multi(ops))
5498 return -EINVAL;
5499 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
5500 return -EINVAL;
5501
5502 mutex_lock(&direct_mutex);
5503 err = unregister_ftrace_function(ops);
5504 remove_direct_functions_hash(hash, addr);
5505 mutex_unlock(&direct_mutex);
5506
5507 /* cleanup for possible another register call */
5508 ops->func = NULL;
5509 ops->trampoline = 0;
5510
5511 if (free_filters)
5512 ftrace_free_filter(ops);
5513 return err;
5514 }
5515 EXPORT_SYMBOL_GPL(unregister_ftrace_direct);
5516
5517 static int
__modify_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)5518 __modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
5519 {
5520 struct ftrace_hash *hash;
5521 struct ftrace_func_entry *entry, *iter;
5522 static struct ftrace_ops tmp_ops = {
5523 .func = ftrace_stub,
5524 .flags = FTRACE_OPS_FL_STUB,
5525 };
5526 int i, size;
5527 int err;
5528
5529 lockdep_assert_held_once(&direct_mutex);
5530
5531 /* Enable the tmp_ops to have the same functions as the direct ops */
5532 ftrace_ops_init(&tmp_ops);
5533 tmp_ops.func_hash = ops->func_hash;
5534 tmp_ops.direct_call = addr;
5535
5536 err = register_ftrace_function_nolock(&tmp_ops);
5537 if (err)
5538 return err;
5539
5540 /*
5541 * Now the ftrace_ops_list_func() is called to do the direct callers.
5542 * We can safely change the direct functions attached to each entry.
5543 */
5544 mutex_lock(&ftrace_lock);
5545
5546 hash = ops->func_hash->filter_hash;
5547 size = 1 << hash->size_bits;
5548 for (i = 0; i < size; i++) {
5549 hlist_for_each_entry(iter, &hash->buckets[i], hlist) {
5550 entry = __ftrace_lookup_ip(direct_functions, iter->ip);
5551 if (!entry)
5552 continue;
5553 entry->direct = addr;
5554 }
5555 }
5556 /* Prevent store tearing if a trampoline concurrently accesses the value */
5557 WRITE_ONCE(ops->direct_call, addr);
5558
5559 mutex_unlock(&ftrace_lock);
5560
5561 /* Removing the tmp_ops will add the updated direct callers to the functions */
5562 unregister_ftrace_function(&tmp_ops);
5563
5564 return err;
5565 }
5566
5567 /**
5568 * modify_ftrace_direct_nolock - Modify an existing direct 'multi' call
5569 * to call something else
5570 * @ops: The address of the struct ftrace_ops object
5571 * @addr: The address of the new trampoline to call at @ops functions
5572 *
5573 * This is used to unregister currently registered direct caller and
5574 * register new one @addr on functions registered in @ops object.
5575 *
5576 * Note there's window between ftrace_shutdown and ftrace_startup calls
5577 * where there will be no callbacks called.
5578 *
5579 * Caller should already have direct_mutex locked, so we don't lock
5580 * direct_mutex here.
5581 *
5582 * Returns: zero on success. Non zero on error, which includes:
5583 * -EINVAL - The @ops object was not properly registered.
5584 */
modify_ftrace_direct_nolock(struct ftrace_ops * ops,unsigned long addr)5585 int modify_ftrace_direct_nolock(struct ftrace_ops *ops, unsigned long addr)
5586 {
5587 if (check_direct_multi(ops))
5588 return -EINVAL;
5589 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
5590 return -EINVAL;
5591
5592 return __modify_ftrace_direct(ops, addr);
5593 }
5594 EXPORT_SYMBOL_GPL(modify_ftrace_direct_nolock);
5595
5596 /**
5597 * modify_ftrace_direct - Modify an existing direct 'multi' call
5598 * to call something else
5599 * @ops: The address of the struct ftrace_ops object
5600 * @addr: The address of the new trampoline to call at @ops functions
5601 *
5602 * This is used to unregister currently registered direct caller and
5603 * register new one @addr on functions registered in @ops object.
5604 *
5605 * Note there's window between ftrace_shutdown and ftrace_startup calls
5606 * where there will be no callbacks called.
5607 *
5608 * Returns: zero on success. Non zero on error, which includes:
5609 * -EINVAL - The @ops object was not properly registered.
5610 */
modify_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)5611 int modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
5612 {
5613 int err;
5614
5615 if (check_direct_multi(ops))
5616 return -EINVAL;
5617 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
5618 return -EINVAL;
5619
5620 mutex_lock(&direct_mutex);
5621 err = __modify_ftrace_direct(ops, addr);
5622 mutex_unlock(&direct_mutex);
5623 return err;
5624 }
5625 EXPORT_SYMBOL_GPL(modify_ftrace_direct);
5626 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
5627
5628 /**
5629 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
5630 * @ops: the ops to set the filter with
5631 * @ip: the address to add to or remove from the filter.
5632 * @remove: non zero to remove the ip from the filter
5633 * @reset: non zero to reset all filters before applying this filter.
5634 *
5635 * Filters denote which functions should be enabled when tracing is enabled
5636 * If @ip is NULL, it fails to update filter.
5637 *
5638 * This can allocate memory which must be freed before @ops can be freed,
5639 * either by removing each filtered addr or by using
5640 * ftrace_free_filter(@ops).
5641 */
ftrace_set_filter_ip(struct ftrace_ops * ops,unsigned long ip,int remove,int reset)5642 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
5643 int remove, int reset)
5644 {
5645 ftrace_ops_init(ops);
5646 return ftrace_set_addr(ops, &ip, 1, remove, reset, 1);
5647 }
5648 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
5649
5650 /**
5651 * ftrace_set_filter_ips - set functions to filter on in ftrace by addresses
5652 * @ops: the ops to set the filter with
5653 * @ips: the array of addresses to add to or remove from the filter.
5654 * @cnt: the number of addresses in @ips
5655 * @remove: non zero to remove ips from the filter
5656 * @reset: non zero to reset all filters before applying this filter.
5657 *
5658 * Filters denote which functions should be enabled when tracing is enabled
5659 * If @ips array or any ip specified within is NULL , it fails to update filter.
5660 *
5661 * This can allocate memory which must be freed before @ops can be freed,
5662 * either by removing each filtered addr or by using
5663 * ftrace_free_filter(@ops).
5664 */
ftrace_set_filter_ips(struct ftrace_ops * ops,unsigned long * ips,unsigned int cnt,int remove,int reset)5665 int ftrace_set_filter_ips(struct ftrace_ops *ops, unsigned long *ips,
5666 unsigned int cnt, int remove, int reset)
5667 {
5668 ftrace_ops_init(ops);
5669 return ftrace_set_addr(ops, ips, cnt, remove, reset, 1);
5670 }
5671 EXPORT_SYMBOL_GPL(ftrace_set_filter_ips);
5672
5673 /**
5674 * ftrace_ops_set_global_filter - setup ops to use global filters
5675 * @ops: the ops which will use the global filters
5676 *
5677 * ftrace users who need global function trace filtering should call this.
5678 * It can set the global filter only if ops were not initialized before.
5679 */
ftrace_ops_set_global_filter(struct ftrace_ops * ops)5680 void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
5681 {
5682 if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
5683 return;
5684
5685 ftrace_ops_init(ops);
5686 ops->func_hash = &global_ops.local_hash;
5687 }
5688 EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
5689
5690 static int
ftrace_set_regex(struct ftrace_ops * ops,unsigned char * buf,int len,int reset,int enable)5691 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
5692 int reset, int enable)
5693 {
5694 return ftrace_set_hash(ops, buf, len, NULL, 0, 0, reset, enable);
5695 }
5696
5697 /**
5698 * ftrace_set_filter - set a function to filter on in ftrace
5699 * @ops: the ops to set the filter with
5700 * @buf: the string that holds the function filter text.
5701 * @len: the length of the string.
5702 * @reset: non-zero to reset all filters before applying this filter.
5703 *
5704 * Filters denote which functions should be enabled when tracing is enabled.
5705 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
5706 *
5707 * This can allocate memory which must be freed before @ops can be freed,
5708 * either by removing each filtered addr or by using
5709 * ftrace_free_filter(@ops).
5710 */
ftrace_set_filter(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)5711 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
5712 int len, int reset)
5713 {
5714 ftrace_ops_init(ops);
5715 return ftrace_set_regex(ops, buf, len, reset, 1);
5716 }
5717 EXPORT_SYMBOL_GPL(ftrace_set_filter);
5718
5719 /**
5720 * ftrace_set_notrace - set a function to not trace in ftrace
5721 * @ops: the ops to set the notrace filter with
5722 * @buf: the string that holds the function notrace text.
5723 * @len: the length of the string.
5724 * @reset: non-zero to reset all filters before applying this filter.
5725 *
5726 * Notrace Filters denote which functions should not be enabled when tracing
5727 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
5728 * for tracing.
5729 *
5730 * This can allocate memory which must be freed before @ops can be freed,
5731 * either by removing each filtered addr or by using
5732 * ftrace_free_filter(@ops).
5733 */
ftrace_set_notrace(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)5734 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
5735 int len, int reset)
5736 {
5737 ftrace_ops_init(ops);
5738 return ftrace_set_regex(ops, buf, len, reset, 0);
5739 }
5740 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
5741 /**
5742 * ftrace_set_global_filter - set a function to filter on with global tracers
5743 * @buf: the string that holds the function filter text.
5744 * @len: the length of the string.
5745 * @reset: non-zero to reset all filters before applying this filter.
5746 *
5747 * Filters denote which functions should be enabled when tracing is enabled.
5748 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
5749 */
ftrace_set_global_filter(unsigned char * buf,int len,int reset)5750 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
5751 {
5752 ftrace_set_regex(&global_ops, buf, len, reset, 1);
5753 }
5754 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
5755
5756 /**
5757 * ftrace_set_global_notrace - set a function to not trace with global tracers
5758 * @buf: the string that holds the function notrace text.
5759 * @len: the length of the string.
5760 * @reset: non-zero to reset all filters before applying this filter.
5761 *
5762 * Notrace Filters denote which functions should not be enabled when tracing
5763 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
5764 * for tracing.
5765 */
ftrace_set_global_notrace(unsigned char * buf,int len,int reset)5766 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
5767 {
5768 ftrace_set_regex(&global_ops, buf, len, reset, 0);
5769 }
5770 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
5771
5772 /*
5773 * command line interface to allow users to set filters on boot up.
5774 */
5775 #define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
5776 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
5777 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
5778
5779 /* Used by function selftest to not test if filter is set */
5780 bool ftrace_filter_param __initdata;
5781
set_ftrace_notrace(char * str)5782 static int __init set_ftrace_notrace(char *str)
5783 {
5784 ftrace_filter_param = true;
5785 strscpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
5786 return 1;
5787 }
5788 __setup("ftrace_notrace=", set_ftrace_notrace);
5789
set_ftrace_filter(char * str)5790 static int __init set_ftrace_filter(char *str)
5791 {
5792 ftrace_filter_param = true;
5793 strscpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
5794 return 1;
5795 }
5796 __setup("ftrace_filter=", set_ftrace_filter);
5797
5798 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5799 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
5800 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
5801 static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
5802
set_graph_function(char * str)5803 static int __init set_graph_function(char *str)
5804 {
5805 strscpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
5806 return 1;
5807 }
5808 __setup("ftrace_graph_filter=", set_graph_function);
5809
set_graph_notrace_function(char * str)5810 static int __init set_graph_notrace_function(char *str)
5811 {
5812 strscpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
5813 return 1;
5814 }
5815 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
5816
set_graph_max_depth_function(char * str)5817 static int __init set_graph_max_depth_function(char *str)
5818 {
5819 if (!str)
5820 return 0;
5821 fgraph_max_depth = simple_strtoul(str, NULL, 0);
5822 return 1;
5823 }
5824 __setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
5825
set_ftrace_early_graph(char * buf,int enable)5826 static void __init set_ftrace_early_graph(char *buf, int enable)
5827 {
5828 int ret;
5829 char *func;
5830 struct ftrace_hash *hash;
5831
5832 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5833 if (MEM_FAIL(!hash, "Failed to allocate hash\n"))
5834 return;
5835
5836 while (buf) {
5837 func = strsep(&buf, ",");
5838 /* we allow only one expression at a time */
5839 ret = ftrace_graph_set_hash(hash, func);
5840 if (ret)
5841 printk(KERN_DEBUG "ftrace: function %s not "
5842 "traceable\n", func);
5843 }
5844
5845 if (enable)
5846 ftrace_graph_hash = hash;
5847 else
5848 ftrace_graph_notrace_hash = hash;
5849 }
5850 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5851
5852 void __init
ftrace_set_early_filter(struct ftrace_ops * ops,char * buf,int enable)5853 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
5854 {
5855 char *func;
5856
5857 ftrace_ops_init(ops);
5858
5859 while (buf) {
5860 func = strsep(&buf, ",");
5861 ftrace_set_regex(ops, func, strlen(func), 0, enable);
5862 }
5863 }
5864
set_ftrace_early_filters(void)5865 static void __init set_ftrace_early_filters(void)
5866 {
5867 if (ftrace_filter_buf[0])
5868 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
5869 if (ftrace_notrace_buf[0])
5870 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
5871 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5872 if (ftrace_graph_buf[0])
5873 set_ftrace_early_graph(ftrace_graph_buf, 1);
5874 if (ftrace_graph_notrace_buf[0])
5875 set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
5876 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5877 }
5878
ftrace_regex_release(struct inode * inode,struct file * file)5879 int ftrace_regex_release(struct inode *inode, struct file *file)
5880 {
5881 struct seq_file *m = (struct seq_file *)file->private_data;
5882 struct ftrace_iterator *iter;
5883 struct ftrace_hash **orig_hash;
5884 struct trace_parser *parser;
5885 int filter_hash;
5886
5887 if (file->f_mode & FMODE_READ) {
5888 iter = m->private;
5889 seq_release(inode, file);
5890 } else
5891 iter = file->private_data;
5892
5893 parser = &iter->parser;
5894 if (trace_parser_loaded(parser)) {
5895 int enable = !(iter->flags & FTRACE_ITER_NOTRACE);
5896
5897 ftrace_process_regex(iter, parser->buffer,
5898 parser->idx, enable);
5899 }
5900
5901 trace_parser_put(parser);
5902
5903 mutex_lock(&iter->ops->func_hash->regex_lock);
5904
5905 if (file->f_mode & FMODE_WRITE) {
5906 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
5907
5908 if (filter_hash) {
5909 orig_hash = &iter->ops->func_hash->filter_hash;
5910 if (iter->tr) {
5911 if (list_empty(&iter->tr->mod_trace))
5912 iter->hash->flags &= ~FTRACE_HASH_FL_MOD;
5913 else
5914 iter->hash->flags |= FTRACE_HASH_FL_MOD;
5915 }
5916 } else
5917 orig_hash = &iter->ops->func_hash->notrace_hash;
5918
5919 mutex_lock(&ftrace_lock);
5920 ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
5921 iter->hash, filter_hash);
5922 mutex_unlock(&ftrace_lock);
5923 } else {
5924 /* For read only, the hash is the ops hash */
5925 iter->hash = NULL;
5926 }
5927
5928 mutex_unlock(&iter->ops->func_hash->regex_lock);
5929 free_ftrace_hash(iter->hash);
5930 if (iter->tr)
5931 trace_array_put(iter->tr);
5932 kfree(iter);
5933
5934 return 0;
5935 }
5936
5937 static const struct file_operations ftrace_avail_fops = {
5938 .open = ftrace_avail_open,
5939 .read = seq_read,
5940 .llseek = seq_lseek,
5941 .release = seq_release_private,
5942 };
5943
5944 static const struct file_operations ftrace_enabled_fops = {
5945 .open = ftrace_enabled_open,
5946 .read = seq_read,
5947 .llseek = seq_lseek,
5948 .release = seq_release_private,
5949 };
5950
5951 static const struct file_operations ftrace_touched_fops = {
5952 .open = ftrace_touched_open,
5953 .read = seq_read,
5954 .llseek = seq_lseek,
5955 .release = seq_release_private,
5956 };
5957
5958 static const struct file_operations ftrace_avail_addrs_fops = {
5959 .open = ftrace_avail_addrs_open,
5960 .read = seq_read,
5961 .llseek = seq_lseek,
5962 .release = seq_release_private,
5963 };
5964
5965 static const struct file_operations ftrace_filter_fops = {
5966 .open = ftrace_filter_open,
5967 .read = seq_read,
5968 .write = ftrace_filter_write,
5969 .llseek = tracing_lseek,
5970 .release = ftrace_regex_release,
5971 };
5972
5973 static const struct file_operations ftrace_notrace_fops = {
5974 .open = ftrace_notrace_open,
5975 .read = seq_read,
5976 .write = ftrace_notrace_write,
5977 .llseek = tracing_lseek,
5978 .release = ftrace_regex_release,
5979 };
5980
5981 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5982
5983 static DEFINE_MUTEX(graph_lock);
5984
5985 struct ftrace_hash __rcu *ftrace_graph_hash = EMPTY_HASH;
5986 struct ftrace_hash __rcu *ftrace_graph_notrace_hash = EMPTY_HASH;
5987
5988 enum graph_filter_type {
5989 GRAPH_FILTER_NOTRACE = 0,
5990 GRAPH_FILTER_FUNCTION,
5991 };
5992
5993 #define FTRACE_GRAPH_EMPTY ((void *)1)
5994
5995 struct ftrace_graph_data {
5996 struct ftrace_hash *hash;
5997 struct ftrace_func_entry *entry;
5998 int idx; /* for hash table iteration */
5999 enum graph_filter_type type;
6000 struct ftrace_hash *new_hash;
6001 const struct seq_operations *seq_ops;
6002 struct trace_parser parser;
6003 };
6004
6005 static void *
__g_next(struct seq_file * m,loff_t * pos)6006 __g_next(struct seq_file *m, loff_t *pos)
6007 {
6008 struct ftrace_graph_data *fgd = m->private;
6009 struct ftrace_func_entry *entry = fgd->entry;
6010 struct hlist_head *head;
6011 int i, idx = fgd->idx;
6012
6013 if (*pos >= fgd->hash->count)
6014 return NULL;
6015
6016 if (entry) {
6017 hlist_for_each_entry_continue(entry, hlist) {
6018 fgd->entry = entry;
6019 return entry;
6020 }
6021
6022 idx++;
6023 }
6024
6025 for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
6026 head = &fgd->hash->buckets[i];
6027 hlist_for_each_entry(entry, head, hlist) {
6028 fgd->entry = entry;
6029 fgd->idx = i;
6030 return entry;
6031 }
6032 }
6033 return NULL;
6034 }
6035
6036 static void *
g_next(struct seq_file * m,void * v,loff_t * pos)6037 g_next(struct seq_file *m, void *v, loff_t *pos)
6038 {
6039 (*pos)++;
6040 return __g_next(m, pos);
6041 }
6042
g_start(struct seq_file * m,loff_t * pos)6043 static void *g_start(struct seq_file *m, loff_t *pos)
6044 {
6045 struct ftrace_graph_data *fgd = m->private;
6046
6047 mutex_lock(&graph_lock);
6048
6049 if (fgd->type == GRAPH_FILTER_FUNCTION)
6050 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
6051 lockdep_is_held(&graph_lock));
6052 else
6053 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6054 lockdep_is_held(&graph_lock));
6055
6056 /* Nothing, tell g_show to print all functions are enabled */
6057 if (ftrace_hash_empty(fgd->hash) && !*pos)
6058 return FTRACE_GRAPH_EMPTY;
6059
6060 fgd->idx = 0;
6061 fgd->entry = NULL;
6062 return __g_next(m, pos);
6063 }
6064
g_stop(struct seq_file * m,void * p)6065 static void g_stop(struct seq_file *m, void *p)
6066 {
6067 mutex_unlock(&graph_lock);
6068 }
6069
g_show(struct seq_file * m,void * v)6070 static int g_show(struct seq_file *m, void *v)
6071 {
6072 struct ftrace_func_entry *entry = v;
6073
6074 if (!entry)
6075 return 0;
6076
6077 if (entry == FTRACE_GRAPH_EMPTY) {
6078 struct ftrace_graph_data *fgd = m->private;
6079
6080 if (fgd->type == GRAPH_FILTER_FUNCTION)
6081 seq_puts(m, "#### all functions enabled ####\n");
6082 else
6083 seq_puts(m, "#### no functions disabled ####\n");
6084 return 0;
6085 }
6086
6087 seq_printf(m, "%ps\n", (void *)entry->ip);
6088
6089 return 0;
6090 }
6091
6092 static const struct seq_operations ftrace_graph_seq_ops = {
6093 .start = g_start,
6094 .next = g_next,
6095 .stop = g_stop,
6096 .show = g_show,
6097 };
6098
6099 static int
__ftrace_graph_open(struct inode * inode,struct file * file,struct ftrace_graph_data * fgd)6100 __ftrace_graph_open(struct inode *inode, struct file *file,
6101 struct ftrace_graph_data *fgd)
6102 {
6103 int ret;
6104 struct ftrace_hash *new_hash = NULL;
6105
6106 ret = security_locked_down(LOCKDOWN_TRACEFS);
6107 if (ret)
6108 return ret;
6109
6110 if (file->f_mode & FMODE_WRITE) {
6111 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
6112
6113 if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
6114 return -ENOMEM;
6115
6116 if (file->f_flags & O_TRUNC)
6117 new_hash = alloc_ftrace_hash(size_bits);
6118 else
6119 new_hash = alloc_and_copy_ftrace_hash(size_bits,
6120 fgd->hash);
6121 if (!new_hash) {
6122 ret = -ENOMEM;
6123 goto out;
6124 }
6125 }
6126
6127 if (file->f_mode & FMODE_READ) {
6128 ret = seq_open(file, &ftrace_graph_seq_ops);
6129 if (!ret) {
6130 struct seq_file *m = file->private_data;
6131 m->private = fgd;
6132 } else {
6133 /* Failed */
6134 free_ftrace_hash(new_hash);
6135 new_hash = NULL;
6136 }
6137 } else
6138 file->private_data = fgd;
6139
6140 out:
6141 if (ret < 0 && file->f_mode & FMODE_WRITE)
6142 trace_parser_put(&fgd->parser);
6143
6144 fgd->new_hash = new_hash;
6145
6146 /*
6147 * All uses of fgd->hash must be taken with the graph_lock
6148 * held. The graph_lock is going to be released, so force
6149 * fgd->hash to be reinitialized when it is taken again.
6150 */
6151 fgd->hash = NULL;
6152
6153 return ret;
6154 }
6155
6156 static int
ftrace_graph_open(struct inode * inode,struct file * file)6157 ftrace_graph_open(struct inode *inode, struct file *file)
6158 {
6159 struct ftrace_graph_data *fgd;
6160 int ret;
6161
6162 if (unlikely(ftrace_disabled))
6163 return -ENODEV;
6164
6165 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
6166 if (fgd == NULL)
6167 return -ENOMEM;
6168
6169 mutex_lock(&graph_lock);
6170
6171 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
6172 lockdep_is_held(&graph_lock));
6173 fgd->type = GRAPH_FILTER_FUNCTION;
6174 fgd->seq_ops = &ftrace_graph_seq_ops;
6175
6176 ret = __ftrace_graph_open(inode, file, fgd);
6177 if (ret < 0)
6178 kfree(fgd);
6179
6180 mutex_unlock(&graph_lock);
6181 return ret;
6182 }
6183
6184 static int
ftrace_graph_notrace_open(struct inode * inode,struct file * file)6185 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
6186 {
6187 struct ftrace_graph_data *fgd;
6188 int ret;
6189
6190 if (unlikely(ftrace_disabled))
6191 return -ENODEV;
6192
6193 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
6194 if (fgd == NULL)
6195 return -ENOMEM;
6196
6197 mutex_lock(&graph_lock);
6198
6199 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6200 lockdep_is_held(&graph_lock));
6201 fgd->type = GRAPH_FILTER_NOTRACE;
6202 fgd->seq_ops = &ftrace_graph_seq_ops;
6203
6204 ret = __ftrace_graph_open(inode, file, fgd);
6205 if (ret < 0)
6206 kfree(fgd);
6207
6208 mutex_unlock(&graph_lock);
6209 return ret;
6210 }
6211
6212 static int
ftrace_graph_release(struct inode * inode,struct file * file)6213 ftrace_graph_release(struct inode *inode, struct file *file)
6214 {
6215 struct ftrace_graph_data *fgd;
6216 struct ftrace_hash *old_hash, *new_hash;
6217 struct trace_parser *parser;
6218 int ret = 0;
6219
6220 if (file->f_mode & FMODE_READ) {
6221 struct seq_file *m = file->private_data;
6222
6223 fgd = m->private;
6224 seq_release(inode, file);
6225 } else {
6226 fgd = file->private_data;
6227 }
6228
6229
6230 if (file->f_mode & FMODE_WRITE) {
6231
6232 parser = &fgd->parser;
6233
6234 if (trace_parser_loaded((parser))) {
6235 ret = ftrace_graph_set_hash(fgd->new_hash,
6236 parser->buffer);
6237 }
6238
6239 trace_parser_put(parser);
6240
6241 new_hash = __ftrace_hash_move(fgd->new_hash);
6242 if (!new_hash) {
6243 ret = -ENOMEM;
6244 goto out;
6245 }
6246
6247 mutex_lock(&graph_lock);
6248
6249 if (fgd->type == GRAPH_FILTER_FUNCTION) {
6250 old_hash = rcu_dereference_protected(ftrace_graph_hash,
6251 lockdep_is_held(&graph_lock));
6252 rcu_assign_pointer(ftrace_graph_hash, new_hash);
6253 } else {
6254 old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6255 lockdep_is_held(&graph_lock));
6256 rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
6257 }
6258
6259 mutex_unlock(&graph_lock);
6260
6261 /*
6262 * We need to do a hard force of sched synchronization.
6263 * This is because we use preempt_disable() to do RCU, but
6264 * the function tracers can be called where RCU is not watching
6265 * (like before user_exit()). We can not rely on the RCU
6266 * infrastructure to do the synchronization, thus we must do it
6267 * ourselves.
6268 */
6269 if (old_hash != EMPTY_HASH)
6270 synchronize_rcu_tasks_rude();
6271
6272 free_ftrace_hash(old_hash);
6273 }
6274
6275 out:
6276 free_ftrace_hash(fgd->new_hash);
6277 kfree(fgd);
6278
6279 return ret;
6280 }
6281
6282 static int
ftrace_graph_set_hash(struct ftrace_hash * hash,char * buffer)6283 ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
6284 {
6285 struct ftrace_glob func_g;
6286 struct dyn_ftrace *rec;
6287 struct ftrace_page *pg;
6288 struct ftrace_func_entry *entry;
6289 int fail = 1;
6290 int not;
6291
6292 /* decode regex */
6293 func_g.type = filter_parse_regex(buffer, strlen(buffer),
6294 &func_g.search, ¬);
6295
6296 func_g.len = strlen(func_g.search);
6297
6298 mutex_lock(&ftrace_lock);
6299
6300 if (unlikely(ftrace_disabled)) {
6301 mutex_unlock(&ftrace_lock);
6302 return -ENODEV;
6303 }
6304
6305 do_for_each_ftrace_rec(pg, rec) {
6306
6307 if (rec->flags & FTRACE_FL_DISABLED)
6308 continue;
6309
6310 if (ftrace_match_record(rec, &func_g, NULL, 0)) {
6311 entry = ftrace_lookup_ip(hash, rec->ip);
6312
6313 if (!not) {
6314 fail = 0;
6315
6316 if (entry)
6317 continue;
6318 if (add_hash_entry(hash, rec->ip) == NULL)
6319 goto out;
6320 } else {
6321 if (entry) {
6322 free_hash_entry(hash, entry);
6323 fail = 0;
6324 }
6325 }
6326 }
6327 } while_for_each_ftrace_rec();
6328 out:
6329 mutex_unlock(&ftrace_lock);
6330
6331 if (fail)
6332 return -EINVAL;
6333
6334 return 0;
6335 }
6336
6337 static ssize_t
ftrace_graph_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)6338 ftrace_graph_write(struct file *file, const char __user *ubuf,
6339 size_t cnt, loff_t *ppos)
6340 {
6341 ssize_t read, ret = 0;
6342 struct ftrace_graph_data *fgd = file->private_data;
6343 struct trace_parser *parser;
6344
6345 if (!cnt)
6346 return 0;
6347
6348 /* Read mode uses seq functions */
6349 if (file->f_mode & FMODE_READ) {
6350 struct seq_file *m = file->private_data;
6351 fgd = m->private;
6352 }
6353
6354 parser = &fgd->parser;
6355
6356 read = trace_get_user(parser, ubuf, cnt, ppos);
6357
6358 if (read >= 0 && trace_parser_loaded(parser) &&
6359 !trace_parser_cont(parser)) {
6360
6361 ret = ftrace_graph_set_hash(fgd->new_hash,
6362 parser->buffer);
6363 trace_parser_clear(parser);
6364 }
6365
6366 if (!ret)
6367 ret = read;
6368
6369 return ret;
6370 }
6371
6372 static const struct file_operations ftrace_graph_fops = {
6373 .open = ftrace_graph_open,
6374 .read = seq_read,
6375 .write = ftrace_graph_write,
6376 .llseek = tracing_lseek,
6377 .release = ftrace_graph_release,
6378 };
6379
6380 static const struct file_operations ftrace_graph_notrace_fops = {
6381 .open = ftrace_graph_notrace_open,
6382 .read = seq_read,
6383 .write = ftrace_graph_write,
6384 .llseek = tracing_lseek,
6385 .release = ftrace_graph_release,
6386 };
6387 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6388
ftrace_create_filter_files(struct ftrace_ops * ops,struct dentry * parent)6389 void ftrace_create_filter_files(struct ftrace_ops *ops,
6390 struct dentry *parent)
6391 {
6392
6393 trace_create_file("set_ftrace_filter", TRACE_MODE_WRITE, parent,
6394 ops, &ftrace_filter_fops);
6395
6396 trace_create_file("set_ftrace_notrace", TRACE_MODE_WRITE, parent,
6397 ops, &ftrace_notrace_fops);
6398 }
6399
6400 /*
6401 * The name "destroy_filter_files" is really a misnomer. Although
6402 * in the future, it may actually delete the files, but this is
6403 * really intended to make sure the ops passed in are disabled
6404 * and that when this function returns, the caller is free to
6405 * free the ops.
6406 *
6407 * The "destroy" name is only to match the "create" name that this
6408 * should be paired with.
6409 */
ftrace_destroy_filter_files(struct ftrace_ops * ops)6410 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
6411 {
6412 mutex_lock(&ftrace_lock);
6413 if (ops->flags & FTRACE_OPS_FL_ENABLED)
6414 ftrace_shutdown(ops, 0);
6415 ops->flags |= FTRACE_OPS_FL_DELETED;
6416 ftrace_free_filter(ops);
6417 mutex_unlock(&ftrace_lock);
6418 }
6419
ftrace_init_dyn_tracefs(struct dentry * d_tracer)6420 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
6421 {
6422
6423 trace_create_file("available_filter_functions", TRACE_MODE_READ,
6424 d_tracer, NULL, &ftrace_avail_fops);
6425
6426 trace_create_file("available_filter_functions_addrs", TRACE_MODE_READ,
6427 d_tracer, NULL, &ftrace_avail_addrs_fops);
6428
6429 trace_create_file("enabled_functions", TRACE_MODE_READ,
6430 d_tracer, NULL, &ftrace_enabled_fops);
6431
6432 trace_create_file("touched_functions", TRACE_MODE_READ,
6433 d_tracer, NULL, &ftrace_touched_fops);
6434
6435 ftrace_create_filter_files(&global_ops, d_tracer);
6436
6437 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6438 trace_create_file("set_graph_function", TRACE_MODE_WRITE, d_tracer,
6439 NULL,
6440 &ftrace_graph_fops);
6441 trace_create_file("set_graph_notrace", TRACE_MODE_WRITE, d_tracer,
6442 NULL,
6443 &ftrace_graph_notrace_fops);
6444 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6445
6446 return 0;
6447 }
6448
ftrace_cmp_ips(const void * a,const void * b)6449 static int ftrace_cmp_ips(const void *a, const void *b)
6450 {
6451 const unsigned long *ipa = a;
6452 const unsigned long *ipb = b;
6453
6454 if (*ipa > *ipb)
6455 return 1;
6456 if (*ipa < *ipb)
6457 return -1;
6458 return 0;
6459 }
6460
6461 #ifdef CONFIG_FTRACE_SORT_STARTUP_TEST
test_is_sorted(unsigned long * start,unsigned long count)6462 static void test_is_sorted(unsigned long *start, unsigned long count)
6463 {
6464 int i;
6465
6466 for (i = 1; i < count; i++) {
6467 if (WARN(start[i - 1] > start[i],
6468 "[%d] %pS at %lx is not sorted with %pS at %lx\n", i,
6469 (void *)start[i - 1], start[i - 1],
6470 (void *)start[i], start[i]))
6471 break;
6472 }
6473 if (i == count)
6474 pr_info("ftrace section at %px sorted properly\n", start);
6475 }
6476 #else
test_is_sorted(unsigned long * start,unsigned long count)6477 static void test_is_sorted(unsigned long *start, unsigned long count)
6478 {
6479 }
6480 #endif
6481
ftrace_process_locs(struct module * mod,unsigned long * start,unsigned long * end)6482 static int ftrace_process_locs(struct module *mod,
6483 unsigned long *start,
6484 unsigned long *end)
6485 {
6486 struct ftrace_page *pg_unuse = NULL;
6487 struct ftrace_page *start_pg;
6488 struct ftrace_page *pg;
6489 struct dyn_ftrace *rec;
6490 unsigned long skipped = 0;
6491 unsigned long count;
6492 unsigned long *p;
6493 unsigned long addr;
6494 unsigned long flags = 0; /* Shut up gcc */
6495 int ret = -ENOMEM;
6496
6497 count = end - start;
6498
6499 if (!count)
6500 return 0;
6501
6502 /*
6503 * Sorting mcount in vmlinux at build time depend on
6504 * CONFIG_BUILDTIME_MCOUNT_SORT, while mcount loc in
6505 * modules can not be sorted at build time.
6506 */
6507 if (!IS_ENABLED(CONFIG_BUILDTIME_MCOUNT_SORT) || mod) {
6508 sort(start, count, sizeof(*start),
6509 ftrace_cmp_ips, NULL);
6510 } else {
6511 test_is_sorted(start, count);
6512 }
6513
6514 start_pg = ftrace_allocate_pages(count);
6515 if (!start_pg)
6516 return -ENOMEM;
6517
6518 mutex_lock(&ftrace_lock);
6519
6520 /*
6521 * Core and each module needs their own pages, as
6522 * modules will free them when they are removed.
6523 * Force a new page to be allocated for modules.
6524 */
6525 if (!mod) {
6526 WARN_ON(ftrace_pages || ftrace_pages_start);
6527 /* First initialization */
6528 ftrace_pages = ftrace_pages_start = start_pg;
6529 } else {
6530 if (!ftrace_pages)
6531 goto out;
6532
6533 if (WARN_ON(ftrace_pages->next)) {
6534 /* Hmm, we have free pages? */
6535 while (ftrace_pages->next)
6536 ftrace_pages = ftrace_pages->next;
6537 }
6538
6539 ftrace_pages->next = start_pg;
6540 }
6541
6542 p = start;
6543 pg = start_pg;
6544 while (p < end) {
6545 unsigned long end_offset;
6546 addr = ftrace_call_adjust(*p++);
6547 /*
6548 * Some architecture linkers will pad between
6549 * the different mcount_loc sections of different
6550 * object files to satisfy alignments.
6551 * Skip any NULL pointers.
6552 */
6553 if (!addr) {
6554 skipped++;
6555 continue;
6556 }
6557
6558 end_offset = (pg->index+1) * sizeof(pg->records[0]);
6559 if (end_offset > PAGE_SIZE << pg->order) {
6560 /* We should have allocated enough */
6561 if (WARN_ON(!pg->next))
6562 break;
6563 pg = pg->next;
6564 }
6565
6566 rec = &pg->records[pg->index++];
6567 rec->ip = addr;
6568 }
6569
6570 if (pg->next) {
6571 pg_unuse = pg->next;
6572 pg->next = NULL;
6573 }
6574
6575 /* Assign the last page to ftrace_pages */
6576 ftrace_pages = pg;
6577
6578 /*
6579 * We only need to disable interrupts on start up
6580 * because we are modifying code that an interrupt
6581 * may execute, and the modification is not atomic.
6582 * But for modules, nothing runs the code we modify
6583 * until we are finished with it, and there's no
6584 * reason to cause large interrupt latencies while we do it.
6585 */
6586 if (!mod)
6587 local_irq_save(flags);
6588 ftrace_update_code(mod, start_pg);
6589 if (!mod)
6590 local_irq_restore(flags);
6591 ret = 0;
6592 out:
6593 mutex_unlock(&ftrace_lock);
6594
6595 /* We should have used all pages unless we skipped some */
6596 if (pg_unuse) {
6597 WARN_ON(!skipped);
6598 ftrace_free_pages(pg_unuse);
6599 }
6600 return ret;
6601 }
6602
6603 struct ftrace_mod_func {
6604 struct list_head list;
6605 char *name;
6606 unsigned long ip;
6607 unsigned int size;
6608 };
6609
6610 struct ftrace_mod_map {
6611 struct rcu_head rcu;
6612 struct list_head list;
6613 struct module *mod;
6614 unsigned long start_addr;
6615 unsigned long end_addr;
6616 struct list_head funcs;
6617 unsigned int num_funcs;
6618 };
6619
ftrace_get_trampoline_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)6620 static int ftrace_get_trampoline_kallsym(unsigned int symnum,
6621 unsigned long *value, char *type,
6622 char *name, char *module_name,
6623 int *exported)
6624 {
6625 struct ftrace_ops *op;
6626
6627 list_for_each_entry_rcu(op, &ftrace_ops_trampoline_list, list) {
6628 if (!op->trampoline || symnum--)
6629 continue;
6630 *value = op->trampoline;
6631 *type = 't';
6632 strscpy(name, FTRACE_TRAMPOLINE_SYM, KSYM_NAME_LEN);
6633 strscpy(module_name, FTRACE_TRAMPOLINE_MOD, MODULE_NAME_LEN);
6634 *exported = 0;
6635 return 0;
6636 }
6637
6638 return -ERANGE;
6639 }
6640
6641 #if defined(CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS) || defined(CONFIG_MODULES)
6642 /*
6643 * Check if the current ops references the given ip.
6644 *
6645 * If the ops traces all functions, then it was already accounted for.
6646 * If the ops does not trace the current record function, skip it.
6647 * If the ops ignores the function via notrace filter, skip it.
6648 */
6649 static bool
ops_references_ip(struct ftrace_ops * ops,unsigned long ip)6650 ops_references_ip(struct ftrace_ops *ops, unsigned long ip)
6651 {
6652 /* If ops isn't enabled, ignore it */
6653 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6654 return false;
6655
6656 /* If ops traces all then it includes this function */
6657 if (ops_traces_mod(ops))
6658 return true;
6659
6660 /* The function must be in the filter */
6661 if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
6662 !__ftrace_lookup_ip(ops->func_hash->filter_hash, ip))
6663 return false;
6664
6665 /* If in notrace hash, we ignore it too */
6666 if (ftrace_lookup_ip(ops->func_hash->notrace_hash, ip))
6667 return false;
6668
6669 return true;
6670 }
6671 #endif
6672
6673 #ifdef CONFIG_MODULES
6674
6675 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
6676
6677 static LIST_HEAD(ftrace_mod_maps);
6678
referenced_filters(struct dyn_ftrace * rec)6679 static int referenced_filters(struct dyn_ftrace *rec)
6680 {
6681 struct ftrace_ops *ops;
6682 int cnt = 0;
6683
6684 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
6685 if (ops_references_ip(ops, rec->ip)) {
6686 if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_DIRECT))
6687 continue;
6688 if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_IPMODIFY))
6689 continue;
6690 cnt++;
6691 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
6692 rec->flags |= FTRACE_FL_REGS;
6693 if (cnt == 1 && ops->trampoline)
6694 rec->flags |= FTRACE_FL_TRAMP;
6695 else
6696 rec->flags &= ~FTRACE_FL_TRAMP;
6697 }
6698 }
6699
6700 return cnt;
6701 }
6702
6703 static void
clear_mod_from_hash(struct ftrace_page * pg,struct ftrace_hash * hash)6704 clear_mod_from_hash(struct ftrace_page *pg, struct ftrace_hash *hash)
6705 {
6706 struct ftrace_func_entry *entry;
6707 struct dyn_ftrace *rec;
6708 int i;
6709
6710 if (ftrace_hash_empty(hash))
6711 return;
6712
6713 for (i = 0; i < pg->index; i++) {
6714 rec = &pg->records[i];
6715 entry = __ftrace_lookup_ip(hash, rec->ip);
6716 /*
6717 * Do not allow this rec to match again.
6718 * Yeah, it may waste some memory, but will be removed
6719 * if/when the hash is modified again.
6720 */
6721 if (entry)
6722 entry->ip = 0;
6723 }
6724 }
6725
6726 /* Clear any records from hashes */
clear_mod_from_hashes(struct ftrace_page * pg)6727 static void clear_mod_from_hashes(struct ftrace_page *pg)
6728 {
6729 struct trace_array *tr;
6730
6731 mutex_lock(&trace_types_lock);
6732 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6733 if (!tr->ops || !tr->ops->func_hash)
6734 continue;
6735 mutex_lock(&tr->ops->func_hash->regex_lock);
6736 clear_mod_from_hash(pg, tr->ops->func_hash->filter_hash);
6737 clear_mod_from_hash(pg, tr->ops->func_hash->notrace_hash);
6738 mutex_unlock(&tr->ops->func_hash->regex_lock);
6739 }
6740 mutex_unlock(&trace_types_lock);
6741 }
6742
ftrace_free_mod_map(struct rcu_head * rcu)6743 static void ftrace_free_mod_map(struct rcu_head *rcu)
6744 {
6745 struct ftrace_mod_map *mod_map = container_of(rcu, struct ftrace_mod_map, rcu);
6746 struct ftrace_mod_func *mod_func;
6747 struct ftrace_mod_func *n;
6748
6749 /* All the contents of mod_map are now not visible to readers */
6750 list_for_each_entry_safe(mod_func, n, &mod_map->funcs, list) {
6751 kfree(mod_func->name);
6752 list_del(&mod_func->list);
6753 kfree(mod_func);
6754 }
6755
6756 kfree(mod_map);
6757 }
6758
ftrace_release_mod(struct module * mod)6759 void ftrace_release_mod(struct module *mod)
6760 {
6761 struct ftrace_mod_map *mod_map;
6762 struct ftrace_mod_map *n;
6763 struct dyn_ftrace *rec;
6764 struct ftrace_page **last_pg;
6765 struct ftrace_page *tmp_page = NULL;
6766 struct ftrace_page *pg;
6767
6768 mutex_lock(&ftrace_lock);
6769
6770 if (ftrace_disabled)
6771 goto out_unlock;
6772
6773 list_for_each_entry_safe(mod_map, n, &ftrace_mod_maps, list) {
6774 if (mod_map->mod == mod) {
6775 list_del_rcu(&mod_map->list);
6776 call_rcu(&mod_map->rcu, ftrace_free_mod_map);
6777 break;
6778 }
6779 }
6780
6781 /*
6782 * Each module has its own ftrace_pages, remove
6783 * them from the list.
6784 */
6785 last_pg = &ftrace_pages_start;
6786 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
6787 rec = &pg->records[0];
6788 if (within_module(rec->ip, mod)) {
6789 /*
6790 * As core pages are first, the first
6791 * page should never be a module page.
6792 */
6793 if (WARN_ON(pg == ftrace_pages_start))
6794 goto out_unlock;
6795
6796 /* Check if we are deleting the last page */
6797 if (pg == ftrace_pages)
6798 ftrace_pages = next_to_ftrace_page(last_pg);
6799
6800 ftrace_update_tot_cnt -= pg->index;
6801 *last_pg = pg->next;
6802
6803 pg->next = tmp_page;
6804 tmp_page = pg;
6805 } else
6806 last_pg = &pg->next;
6807 }
6808 out_unlock:
6809 mutex_unlock(&ftrace_lock);
6810
6811 for (pg = tmp_page; pg; pg = tmp_page) {
6812
6813 /* Needs to be called outside of ftrace_lock */
6814 clear_mod_from_hashes(pg);
6815
6816 if (pg->records) {
6817 free_pages((unsigned long)pg->records, pg->order);
6818 ftrace_number_of_pages -= 1 << pg->order;
6819 }
6820 tmp_page = pg->next;
6821 kfree(pg);
6822 ftrace_number_of_groups--;
6823 }
6824 }
6825
ftrace_module_enable(struct module * mod)6826 void ftrace_module_enable(struct module *mod)
6827 {
6828 struct dyn_ftrace *rec;
6829 struct ftrace_page *pg;
6830
6831 mutex_lock(&ftrace_lock);
6832
6833 if (ftrace_disabled)
6834 goto out_unlock;
6835
6836 /*
6837 * If the tracing is enabled, go ahead and enable the record.
6838 *
6839 * The reason not to enable the record immediately is the
6840 * inherent check of ftrace_make_nop/ftrace_make_call for
6841 * correct previous instructions. Making first the NOP
6842 * conversion puts the module to the correct state, thus
6843 * passing the ftrace_make_call check.
6844 *
6845 * We also delay this to after the module code already set the
6846 * text to read-only, as we now need to set it back to read-write
6847 * so that we can modify the text.
6848 */
6849 if (ftrace_start_up)
6850 ftrace_arch_code_modify_prepare();
6851
6852 do_for_each_ftrace_rec(pg, rec) {
6853 int cnt;
6854 /*
6855 * do_for_each_ftrace_rec() is a double loop.
6856 * module text shares the pg. If a record is
6857 * not part of this module, then skip this pg,
6858 * which the "break" will do.
6859 */
6860 if (!within_module(rec->ip, mod))
6861 break;
6862
6863 /* Weak functions should still be ignored */
6864 if (!test_for_valid_rec(rec)) {
6865 /* Clear all other flags. Should not be enabled anyway */
6866 rec->flags = FTRACE_FL_DISABLED;
6867 continue;
6868 }
6869
6870 cnt = 0;
6871
6872 /*
6873 * When adding a module, we need to check if tracers are
6874 * currently enabled and if they are, and can trace this record,
6875 * we need to enable the module functions as well as update the
6876 * reference counts for those function records.
6877 */
6878 if (ftrace_start_up)
6879 cnt += referenced_filters(rec);
6880
6881 rec->flags &= ~FTRACE_FL_DISABLED;
6882 rec->flags += cnt;
6883
6884 if (ftrace_start_up && cnt) {
6885 int failed = __ftrace_replace_code(rec, 1);
6886 if (failed) {
6887 ftrace_bug(failed, rec);
6888 goto out_loop;
6889 }
6890 }
6891
6892 } while_for_each_ftrace_rec();
6893
6894 out_loop:
6895 if (ftrace_start_up)
6896 ftrace_arch_code_modify_post_process();
6897
6898 out_unlock:
6899 mutex_unlock(&ftrace_lock);
6900
6901 process_cached_mods(mod->name);
6902 }
6903
ftrace_module_init(struct module * mod)6904 void ftrace_module_init(struct module *mod)
6905 {
6906 int ret;
6907
6908 if (ftrace_disabled || !mod->num_ftrace_callsites)
6909 return;
6910
6911 ret = ftrace_process_locs(mod, mod->ftrace_callsites,
6912 mod->ftrace_callsites + mod->num_ftrace_callsites);
6913 if (ret)
6914 pr_warn("ftrace: failed to allocate entries for module '%s' functions\n",
6915 mod->name);
6916 }
6917
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)6918 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
6919 struct dyn_ftrace *rec)
6920 {
6921 struct ftrace_mod_func *mod_func;
6922 unsigned long symsize;
6923 unsigned long offset;
6924 char str[KSYM_SYMBOL_LEN];
6925 char *modname;
6926 const char *ret;
6927
6928 ret = kallsyms_lookup(rec->ip, &symsize, &offset, &modname, str);
6929 if (!ret)
6930 return;
6931
6932 mod_func = kmalloc(sizeof(*mod_func), GFP_KERNEL);
6933 if (!mod_func)
6934 return;
6935
6936 mod_func->name = kstrdup(str, GFP_KERNEL);
6937 if (!mod_func->name) {
6938 kfree(mod_func);
6939 return;
6940 }
6941
6942 mod_func->ip = rec->ip - offset;
6943 mod_func->size = symsize;
6944
6945 mod_map->num_funcs++;
6946
6947 list_add_rcu(&mod_func->list, &mod_map->funcs);
6948 }
6949
6950 static struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)6951 allocate_ftrace_mod_map(struct module *mod,
6952 unsigned long start, unsigned long end)
6953 {
6954 struct ftrace_mod_map *mod_map;
6955
6956 mod_map = kmalloc(sizeof(*mod_map), GFP_KERNEL);
6957 if (!mod_map)
6958 return NULL;
6959
6960 mod_map->mod = mod;
6961 mod_map->start_addr = start;
6962 mod_map->end_addr = end;
6963 mod_map->num_funcs = 0;
6964
6965 INIT_LIST_HEAD_RCU(&mod_map->funcs);
6966
6967 list_add_rcu(&mod_map->list, &ftrace_mod_maps);
6968
6969 return mod_map;
6970 }
6971
6972 static const char *
ftrace_func_address_lookup(struct ftrace_mod_map * mod_map,unsigned long addr,unsigned long * size,unsigned long * off,char * sym)6973 ftrace_func_address_lookup(struct ftrace_mod_map *mod_map,
6974 unsigned long addr, unsigned long *size,
6975 unsigned long *off, char *sym)
6976 {
6977 struct ftrace_mod_func *found_func = NULL;
6978 struct ftrace_mod_func *mod_func;
6979
6980 list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
6981 if (addr >= mod_func->ip &&
6982 addr < mod_func->ip + mod_func->size) {
6983 found_func = mod_func;
6984 break;
6985 }
6986 }
6987
6988 if (found_func) {
6989 if (size)
6990 *size = found_func->size;
6991 if (off)
6992 *off = addr - found_func->ip;
6993 if (sym)
6994 strscpy(sym, found_func->name, KSYM_NAME_LEN);
6995
6996 return found_func->name;
6997 }
6998
6999 return NULL;
7000 }
7001
7002 const char *
ftrace_mod_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)7003 ftrace_mod_address_lookup(unsigned long addr, unsigned long *size,
7004 unsigned long *off, char **modname, char *sym)
7005 {
7006 struct ftrace_mod_map *mod_map;
7007 const char *ret = NULL;
7008
7009 /* mod_map is freed via call_rcu() */
7010 preempt_disable();
7011 list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
7012 ret = ftrace_func_address_lookup(mod_map, addr, size, off, sym);
7013 if (ret) {
7014 if (modname)
7015 *modname = mod_map->mod->name;
7016 break;
7017 }
7018 }
7019 preempt_enable();
7020
7021 return ret;
7022 }
7023
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7024 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
7025 char *type, char *name,
7026 char *module_name, int *exported)
7027 {
7028 struct ftrace_mod_map *mod_map;
7029 struct ftrace_mod_func *mod_func;
7030 int ret;
7031
7032 preempt_disable();
7033 list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
7034
7035 if (symnum >= mod_map->num_funcs) {
7036 symnum -= mod_map->num_funcs;
7037 continue;
7038 }
7039
7040 list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
7041 if (symnum > 1) {
7042 symnum--;
7043 continue;
7044 }
7045
7046 *value = mod_func->ip;
7047 *type = 'T';
7048 strscpy(name, mod_func->name, KSYM_NAME_LEN);
7049 strscpy(module_name, mod_map->mod->name, MODULE_NAME_LEN);
7050 *exported = 1;
7051 preempt_enable();
7052 return 0;
7053 }
7054 WARN_ON(1);
7055 break;
7056 }
7057 ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
7058 module_name, exported);
7059 preempt_enable();
7060 return ret;
7061 }
7062
7063 #else
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)7064 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
7065 struct dyn_ftrace *rec) { }
7066 static inline struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)7067 allocate_ftrace_mod_map(struct module *mod,
7068 unsigned long start, unsigned long end)
7069 {
7070 return NULL;
7071 }
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7072 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
7073 char *type, char *name, char *module_name,
7074 int *exported)
7075 {
7076 int ret;
7077
7078 preempt_disable();
7079 ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
7080 module_name, exported);
7081 preempt_enable();
7082 return ret;
7083 }
7084 #endif /* CONFIG_MODULES */
7085
7086 struct ftrace_init_func {
7087 struct list_head list;
7088 unsigned long ip;
7089 };
7090
7091 /* Clear any init ips from hashes */
7092 static void
clear_func_from_hash(struct ftrace_init_func * func,struct ftrace_hash * hash)7093 clear_func_from_hash(struct ftrace_init_func *func, struct ftrace_hash *hash)
7094 {
7095 struct ftrace_func_entry *entry;
7096
7097 entry = ftrace_lookup_ip(hash, func->ip);
7098 /*
7099 * Do not allow this rec to match again.
7100 * Yeah, it may waste some memory, but will be removed
7101 * if/when the hash is modified again.
7102 */
7103 if (entry)
7104 entry->ip = 0;
7105 }
7106
7107 static void
clear_func_from_hashes(struct ftrace_init_func * func)7108 clear_func_from_hashes(struct ftrace_init_func *func)
7109 {
7110 struct trace_array *tr;
7111
7112 mutex_lock(&trace_types_lock);
7113 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7114 if (!tr->ops || !tr->ops->func_hash)
7115 continue;
7116 mutex_lock(&tr->ops->func_hash->regex_lock);
7117 clear_func_from_hash(func, tr->ops->func_hash->filter_hash);
7118 clear_func_from_hash(func, tr->ops->func_hash->notrace_hash);
7119 mutex_unlock(&tr->ops->func_hash->regex_lock);
7120 }
7121 mutex_unlock(&trace_types_lock);
7122 }
7123
add_to_clear_hash_list(struct list_head * clear_list,struct dyn_ftrace * rec)7124 static void add_to_clear_hash_list(struct list_head *clear_list,
7125 struct dyn_ftrace *rec)
7126 {
7127 struct ftrace_init_func *func;
7128
7129 func = kmalloc(sizeof(*func), GFP_KERNEL);
7130 if (!func) {
7131 MEM_FAIL(1, "alloc failure, ftrace filter could be stale\n");
7132 return;
7133 }
7134
7135 func->ip = rec->ip;
7136 list_add(&func->list, clear_list);
7137 }
7138
ftrace_free_mem(struct module * mod,void * start_ptr,void * end_ptr)7139 void ftrace_free_mem(struct module *mod, void *start_ptr, void *end_ptr)
7140 {
7141 unsigned long start = (unsigned long)(start_ptr);
7142 unsigned long end = (unsigned long)(end_ptr);
7143 struct ftrace_page **last_pg = &ftrace_pages_start;
7144 struct ftrace_page *pg;
7145 struct dyn_ftrace *rec;
7146 struct dyn_ftrace key;
7147 struct ftrace_mod_map *mod_map = NULL;
7148 struct ftrace_init_func *func, *func_next;
7149 LIST_HEAD(clear_hash);
7150
7151 key.ip = start;
7152 key.flags = end; /* overload flags, as it is unsigned long */
7153
7154 mutex_lock(&ftrace_lock);
7155
7156 /*
7157 * If we are freeing module init memory, then check if
7158 * any tracer is active. If so, we need to save a mapping of
7159 * the module functions being freed with the address.
7160 */
7161 if (mod && ftrace_ops_list != &ftrace_list_end)
7162 mod_map = allocate_ftrace_mod_map(mod, start, end);
7163
7164 for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
7165 if (end < pg->records[0].ip ||
7166 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
7167 continue;
7168 again:
7169 rec = bsearch(&key, pg->records, pg->index,
7170 sizeof(struct dyn_ftrace),
7171 ftrace_cmp_recs);
7172 if (!rec)
7173 continue;
7174
7175 /* rec will be cleared from hashes after ftrace_lock unlock */
7176 add_to_clear_hash_list(&clear_hash, rec);
7177
7178 if (mod_map)
7179 save_ftrace_mod_rec(mod_map, rec);
7180
7181 pg->index--;
7182 ftrace_update_tot_cnt--;
7183 if (!pg->index) {
7184 *last_pg = pg->next;
7185 if (pg->records) {
7186 free_pages((unsigned long)pg->records, pg->order);
7187 ftrace_number_of_pages -= 1 << pg->order;
7188 }
7189 ftrace_number_of_groups--;
7190 kfree(pg);
7191 pg = container_of(last_pg, struct ftrace_page, next);
7192 if (!(*last_pg))
7193 ftrace_pages = pg;
7194 continue;
7195 }
7196 memmove(rec, rec + 1,
7197 (pg->index - (rec - pg->records)) * sizeof(*rec));
7198 /* More than one function may be in this block */
7199 goto again;
7200 }
7201 mutex_unlock(&ftrace_lock);
7202
7203 list_for_each_entry_safe(func, func_next, &clear_hash, list) {
7204 clear_func_from_hashes(func);
7205 kfree(func);
7206 }
7207 }
7208
ftrace_free_init_mem(void)7209 void __init ftrace_free_init_mem(void)
7210 {
7211 void *start = (void *)(&__init_begin);
7212 void *end = (void *)(&__init_end);
7213
7214 ftrace_boot_snapshot();
7215
7216 ftrace_free_mem(NULL, start, end);
7217 }
7218
ftrace_dyn_arch_init(void)7219 int __init __weak ftrace_dyn_arch_init(void)
7220 {
7221 return 0;
7222 }
7223
ftrace_init(void)7224 void __init ftrace_init(void)
7225 {
7226 extern unsigned long __start_mcount_loc[];
7227 extern unsigned long __stop_mcount_loc[];
7228 unsigned long count, flags;
7229 int ret;
7230
7231 local_irq_save(flags);
7232 ret = ftrace_dyn_arch_init();
7233 local_irq_restore(flags);
7234 if (ret)
7235 goto failed;
7236
7237 count = __stop_mcount_loc - __start_mcount_loc;
7238 if (!count) {
7239 pr_info("ftrace: No functions to be traced?\n");
7240 goto failed;
7241 }
7242
7243 pr_info("ftrace: allocating %ld entries in %ld pages\n",
7244 count, DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
7245
7246 ret = ftrace_process_locs(NULL,
7247 __start_mcount_loc,
7248 __stop_mcount_loc);
7249 if (ret) {
7250 pr_warn("ftrace: failed to allocate entries for functions\n");
7251 goto failed;
7252 }
7253
7254 pr_info("ftrace: allocated %ld pages with %ld groups\n",
7255 ftrace_number_of_pages, ftrace_number_of_groups);
7256
7257 last_ftrace_enabled = ftrace_enabled = 1;
7258
7259 set_ftrace_early_filters();
7260
7261 return;
7262 failed:
7263 ftrace_disabled = 1;
7264 }
7265
7266 /* Do nothing if arch does not support this */
arch_ftrace_update_trampoline(struct ftrace_ops * ops)7267 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
7268 {
7269 }
7270
ftrace_update_trampoline(struct ftrace_ops * ops)7271 static void ftrace_update_trampoline(struct ftrace_ops *ops)
7272 {
7273 unsigned long trampoline = ops->trampoline;
7274
7275 arch_ftrace_update_trampoline(ops);
7276 if (ops->trampoline && ops->trampoline != trampoline &&
7277 (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP)) {
7278 /* Add to kallsyms before the perf events */
7279 ftrace_add_trampoline_to_kallsyms(ops);
7280 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
7281 ops->trampoline, ops->trampoline_size, false,
7282 FTRACE_TRAMPOLINE_SYM);
7283 /*
7284 * Record the perf text poke event after the ksymbol register
7285 * event.
7286 */
7287 perf_event_text_poke((void *)ops->trampoline, NULL, 0,
7288 (void *)ops->trampoline,
7289 ops->trampoline_size);
7290 }
7291 }
7292
ftrace_init_trace_array(struct trace_array * tr)7293 void ftrace_init_trace_array(struct trace_array *tr)
7294 {
7295 INIT_LIST_HEAD(&tr->func_probes);
7296 INIT_LIST_HEAD(&tr->mod_trace);
7297 INIT_LIST_HEAD(&tr->mod_notrace);
7298 }
7299 #else
7300
7301 struct ftrace_ops global_ops = {
7302 .func = ftrace_stub,
7303 .flags = FTRACE_OPS_FL_INITIALIZED |
7304 FTRACE_OPS_FL_PID,
7305 };
7306
ftrace_nodyn_init(void)7307 static int __init ftrace_nodyn_init(void)
7308 {
7309 ftrace_enabled = 1;
7310 return 0;
7311 }
7312 core_initcall(ftrace_nodyn_init);
7313
ftrace_init_dyn_tracefs(struct dentry * d_tracer)7314 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
ftrace_startup_all(int command)7315 static inline void ftrace_startup_all(int command) { }
7316
ftrace_update_trampoline(struct ftrace_ops * ops)7317 static void ftrace_update_trampoline(struct ftrace_ops *ops)
7318 {
7319 }
7320
7321 #endif /* CONFIG_DYNAMIC_FTRACE */
7322
ftrace_init_global_array_ops(struct trace_array * tr)7323 __init void ftrace_init_global_array_ops(struct trace_array *tr)
7324 {
7325 tr->ops = &global_ops;
7326 tr->ops->private = tr;
7327 ftrace_init_trace_array(tr);
7328 }
7329
ftrace_init_array_ops(struct trace_array * tr,ftrace_func_t func)7330 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
7331 {
7332 /* If we filter on pids, update to use the pid function */
7333 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
7334 if (WARN_ON(tr->ops->func != ftrace_stub))
7335 printk("ftrace ops had %pS for function\n",
7336 tr->ops->func);
7337 }
7338 tr->ops->func = func;
7339 tr->ops->private = tr;
7340 }
7341
ftrace_reset_array_ops(struct trace_array * tr)7342 void ftrace_reset_array_ops(struct trace_array *tr)
7343 {
7344 tr->ops->func = ftrace_stub;
7345 }
7346
7347 static nokprobe_inline void
__ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * ignored,struct ftrace_regs * fregs)7348 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
7349 struct ftrace_ops *ignored, struct ftrace_regs *fregs)
7350 {
7351 struct pt_regs *regs = ftrace_get_regs(fregs);
7352 struct ftrace_ops *op;
7353 int bit;
7354
7355 /*
7356 * The ftrace_test_and_set_recursion() will disable preemption,
7357 * which is required since some of the ops may be dynamically
7358 * allocated, they must be freed after a synchronize_rcu().
7359 */
7360 bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
7361 if (bit < 0)
7362 return;
7363
7364 do_for_each_ftrace_op(op, ftrace_ops_list) {
7365 /* Stub functions don't need to be called nor tested */
7366 if (op->flags & FTRACE_OPS_FL_STUB)
7367 continue;
7368 /*
7369 * Check the following for each ops before calling their func:
7370 * if RCU flag is set, then rcu_is_watching() must be true
7371 * Otherwise test if the ip matches the ops filter
7372 *
7373 * If any of the above fails then the op->func() is not executed.
7374 */
7375 if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
7376 ftrace_ops_test(op, ip, regs)) {
7377 if (FTRACE_WARN_ON(!op->func)) {
7378 pr_warn("op=%p %pS\n", op, op);
7379 goto out;
7380 }
7381 op->func(ip, parent_ip, op, fregs);
7382 }
7383 } while_for_each_ftrace_op(op);
7384 out:
7385 trace_clear_recursion(bit);
7386 }
7387
7388 /*
7389 * Some archs only support passing ip and parent_ip. Even though
7390 * the list function ignores the op parameter, we do not want any
7391 * C side effects, where a function is called without the caller
7392 * sending a third parameter.
7393 * Archs are to support both the regs and ftrace_ops at the same time.
7394 * If they support ftrace_ops, it is assumed they support regs.
7395 * If call backs want to use regs, they must either check for regs
7396 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
7397 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
7398 * An architecture can pass partial regs with ftrace_ops and still
7399 * set the ARCH_SUPPORTS_FTRACE_OPS.
7400 *
7401 * In vmlinux.lds.h, ftrace_ops_list_func() is defined to be
7402 * arch_ftrace_ops_list_func.
7403 */
7404 #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)7405 void arch_ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
7406 struct ftrace_ops *op, struct ftrace_regs *fregs)
7407 {
7408 __ftrace_ops_list_func(ip, parent_ip, NULL, fregs);
7409 }
7410 #else
arch_ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip)7411 void arch_ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
7412 {
7413 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
7414 }
7415 #endif
7416 NOKPROBE_SYMBOL(arch_ftrace_ops_list_func);
7417
7418 /*
7419 * If there's only one function registered but it does not support
7420 * recursion, needs RCU protection, then this function will be called
7421 * by the mcount trampoline.
7422 */
ftrace_ops_assist_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)7423 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
7424 struct ftrace_ops *op, struct ftrace_regs *fregs)
7425 {
7426 int bit;
7427
7428 bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
7429 if (bit < 0)
7430 return;
7431
7432 if (!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching())
7433 op->func(ip, parent_ip, op, fregs);
7434
7435 trace_clear_recursion(bit);
7436 }
7437 NOKPROBE_SYMBOL(ftrace_ops_assist_func);
7438
7439 /**
7440 * ftrace_ops_get_func - get the function a trampoline should call
7441 * @ops: the ops to get the function for
7442 *
7443 * Normally the mcount trampoline will call the ops->func, but there
7444 * are times that it should not. For example, if the ops does not
7445 * have its own recursion protection, then it should call the
7446 * ftrace_ops_assist_func() instead.
7447 *
7448 * Returns: the function that the trampoline should call for @ops.
7449 */
ftrace_ops_get_func(struct ftrace_ops * ops)7450 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
7451 {
7452 /*
7453 * If the function does not handle recursion or needs to be RCU safe,
7454 * then we need to call the assist handler.
7455 */
7456 if (ops->flags & (FTRACE_OPS_FL_RECURSION |
7457 FTRACE_OPS_FL_RCU))
7458 return ftrace_ops_assist_func;
7459
7460 return ops->func;
7461 }
7462
7463 static void
ftrace_filter_pid_sched_switch_probe(void * data,bool preempt,struct task_struct * prev,struct task_struct * next,unsigned int prev_state)7464 ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
7465 struct task_struct *prev,
7466 struct task_struct *next,
7467 unsigned int prev_state)
7468 {
7469 struct trace_array *tr = data;
7470 struct trace_pid_list *pid_list;
7471 struct trace_pid_list *no_pid_list;
7472
7473 pid_list = rcu_dereference_sched(tr->function_pids);
7474 no_pid_list = rcu_dereference_sched(tr->function_no_pids);
7475
7476 if (trace_ignore_this_task(pid_list, no_pid_list, next))
7477 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
7478 FTRACE_PID_IGNORE);
7479 else
7480 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
7481 next->pid);
7482 }
7483
7484 static void
ftrace_pid_follow_sched_process_fork(void * data,struct task_struct * self,struct task_struct * task)7485 ftrace_pid_follow_sched_process_fork(void *data,
7486 struct task_struct *self,
7487 struct task_struct *task)
7488 {
7489 struct trace_pid_list *pid_list;
7490 struct trace_array *tr = data;
7491
7492 pid_list = rcu_dereference_sched(tr->function_pids);
7493 trace_filter_add_remove_task(pid_list, self, task);
7494
7495 pid_list = rcu_dereference_sched(tr->function_no_pids);
7496 trace_filter_add_remove_task(pid_list, self, task);
7497 }
7498
7499 static void
ftrace_pid_follow_sched_process_exit(void * data,struct task_struct * task)7500 ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
7501 {
7502 struct trace_pid_list *pid_list;
7503 struct trace_array *tr = data;
7504
7505 pid_list = rcu_dereference_sched(tr->function_pids);
7506 trace_filter_add_remove_task(pid_list, NULL, task);
7507
7508 pid_list = rcu_dereference_sched(tr->function_no_pids);
7509 trace_filter_add_remove_task(pid_list, NULL, task);
7510 }
7511
ftrace_pid_follow_fork(struct trace_array * tr,bool enable)7512 void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
7513 {
7514 if (enable) {
7515 register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
7516 tr);
7517 register_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
7518 tr);
7519 } else {
7520 unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
7521 tr);
7522 unregister_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
7523 tr);
7524 }
7525 }
7526
clear_ftrace_pids(struct trace_array * tr,int type)7527 static void clear_ftrace_pids(struct trace_array *tr, int type)
7528 {
7529 struct trace_pid_list *pid_list;
7530 struct trace_pid_list *no_pid_list;
7531 int cpu;
7532
7533 pid_list = rcu_dereference_protected(tr->function_pids,
7534 lockdep_is_held(&ftrace_lock));
7535 no_pid_list = rcu_dereference_protected(tr->function_no_pids,
7536 lockdep_is_held(&ftrace_lock));
7537
7538 /* Make sure there's something to do */
7539 if (!pid_type_enabled(type, pid_list, no_pid_list))
7540 return;
7541
7542 /* See if the pids still need to be checked after this */
7543 if (!still_need_pid_events(type, pid_list, no_pid_list)) {
7544 unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
7545 for_each_possible_cpu(cpu)
7546 per_cpu_ptr(tr->array_buffer.data, cpu)->ftrace_ignore_pid = FTRACE_PID_TRACE;
7547 }
7548
7549 if (type & TRACE_PIDS)
7550 rcu_assign_pointer(tr->function_pids, NULL);
7551
7552 if (type & TRACE_NO_PIDS)
7553 rcu_assign_pointer(tr->function_no_pids, NULL);
7554
7555 /* Wait till all users are no longer using pid filtering */
7556 synchronize_rcu();
7557
7558 if ((type & TRACE_PIDS) && pid_list)
7559 trace_pid_list_free(pid_list);
7560
7561 if ((type & TRACE_NO_PIDS) && no_pid_list)
7562 trace_pid_list_free(no_pid_list);
7563 }
7564
ftrace_clear_pids(struct trace_array * tr)7565 void ftrace_clear_pids(struct trace_array *tr)
7566 {
7567 mutex_lock(&ftrace_lock);
7568
7569 clear_ftrace_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
7570
7571 mutex_unlock(&ftrace_lock);
7572 }
7573
ftrace_pid_reset(struct trace_array * tr,int type)7574 static void ftrace_pid_reset(struct trace_array *tr, int type)
7575 {
7576 mutex_lock(&ftrace_lock);
7577 clear_ftrace_pids(tr, type);
7578
7579 ftrace_update_pid_func();
7580 ftrace_startup_all(0);
7581
7582 mutex_unlock(&ftrace_lock);
7583 }
7584
7585 /* Greater than any max PID */
7586 #define FTRACE_NO_PIDS (void *)(PID_MAX_LIMIT + 1)
7587
fpid_start(struct seq_file * m,loff_t * pos)7588 static void *fpid_start(struct seq_file *m, loff_t *pos)
7589 __acquires(RCU)
7590 {
7591 struct trace_pid_list *pid_list;
7592 struct trace_array *tr = m->private;
7593
7594 mutex_lock(&ftrace_lock);
7595 rcu_read_lock_sched();
7596
7597 pid_list = rcu_dereference_sched(tr->function_pids);
7598
7599 if (!pid_list)
7600 return !(*pos) ? FTRACE_NO_PIDS : NULL;
7601
7602 return trace_pid_start(pid_list, pos);
7603 }
7604
fpid_next(struct seq_file * m,void * v,loff_t * pos)7605 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
7606 {
7607 struct trace_array *tr = m->private;
7608 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
7609
7610 if (v == FTRACE_NO_PIDS) {
7611 (*pos)++;
7612 return NULL;
7613 }
7614 return trace_pid_next(pid_list, v, pos);
7615 }
7616
fpid_stop(struct seq_file * m,void * p)7617 static void fpid_stop(struct seq_file *m, void *p)
7618 __releases(RCU)
7619 {
7620 rcu_read_unlock_sched();
7621 mutex_unlock(&ftrace_lock);
7622 }
7623
fpid_show(struct seq_file * m,void * v)7624 static int fpid_show(struct seq_file *m, void *v)
7625 {
7626 if (v == FTRACE_NO_PIDS) {
7627 seq_puts(m, "no pid\n");
7628 return 0;
7629 }
7630
7631 return trace_pid_show(m, v);
7632 }
7633
7634 static const struct seq_operations ftrace_pid_sops = {
7635 .start = fpid_start,
7636 .next = fpid_next,
7637 .stop = fpid_stop,
7638 .show = fpid_show,
7639 };
7640
fnpid_start(struct seq_file * m,loff_t * pos)7641 static void *fnpid_start(struct seq_file *m, loff_t *pos)
7642 __acquires(RCU)
7643 {
7644 struct trace_pid_list *pid_list;
7645 struct trace_array *tr = m->private;
7646
7647 mutex_lock(&ftrace_lock);
7648 rcu_read_lock_sched();
7649
7650 pid_list = rcu_dereference_sched(tr->function_no_pids);
7651
7652 if (!pid_list)
7653 return !(*pos) ? FTRACE_NO_PIDS : NULL;
7654
7655 return trace_pid_start(pid_list, pos);
7656 }
7657
fnpid_next(struct seq_file * m,void * v,loff_t * pos)7658 static void *fnpid_next(struct seq_file *m, void *v, loff_t *pos)
7659 {
7660 struct trace_array *tr = m->private;
7661 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_no_pids);
7662
7663 if (v == FTRACE_NO_PIDS) {
7664 (*pos)++;
7665 return NULL;
7666 }
7667 return trace_pid_next(pid_list, v, pos);
7668 }
7669
7670 static const struct seq_operations ftrace_no_pid_sops = {
7671 .start = fnpid_start,
7672 .next = fnpid_next,
7673 .stop = fpid_stop,
7674 .show = fpid_show,
7675 };
7676
pid_open(struct inode * inode,struct file * file,int type)7677 static int pid_open(struct inode *inode, struct file *file, int type)
7678 {
7679 const struct seq_operations *seq_ops;
7680 struct trace_array *tr = inode->i_private;
7681 struct seq_file *m;
7682 int ret = 0;
7683
7684 ret = tracing_check_open_get_tr(tr);
7685 if (ret)
7686 return ret;
7687
7688 if ((file->f_mode & FMODE_WRITE) &&
7689 (file->f_flags & O_TRUNC))
7690 ftrace_pid_reset(tr, type);
7691
7692 switch (type) {
7693 case TRACE_PIDS:
7694 seq_ops = &ftrace_pid_sops;
7695 break;
7696 case TRACE_NO_PIDS:
7697 seq_ops = &ftrace_no_pid_sops;
7698 break;
7699 default:
7700 trace_array_put(tr);
7701 WARN_ON_ONCE(1);
7702 return -EINVAL;
7703 }
7704
7705 ret = seq_open(file, seq_ops);
7706 if (ret < 0) {
7707 trace_array_put(tr);
7708 } else {
7709 m = file->private_data;
7710 /* copy tr over to seq ops */
7711 m->private = tr;
7712 }
7713
7714 return ret;
7715 }
7716
7717 static int
ftrace_pid_open(struct inode * inode,struct file * file)7718 ftrace_pid_open(struct inode *inode, struct file *file)
7719 {
7720 return pid_open(inode, file, TRACE_PIDS);
7721 }
7722
7723 static int
ftrace_no_pid_open(struct inode * inode,struct file * file)7724 ftrace_no_pid_open(struct inode *inode, struct file *file)
7725 {
7726 return pid_open(inode, file, TRACE_NO_PIDS);
7727 }
7728
ignore_task_cpu(void * data)7729 static void ignore_task_cpu(void *data)
7730 {
7731 struct trace_array *tr = data;
7732 struct trace_pid_list *pid_list;
7733 struct trace_pid_list *no_pid_list;
7734
7735 /*
7736 * This function is called by on_each_cpu() while the
7737 * event_mutex is held.
7738 */
7739 pid_list = rcu_dereference_protected(tr->function_pids,
7740 mutex_is_locked(&ftrace_lock));
7741 no_pid_list = rcu_dereference_protected(tr->function_no_pids,
7742 mutex_is_locked(&ftrace_lock));
7743
7744 if (trace_ignore_this_task(pid_list, no_pid_list, current))
7745 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
7746 FTRACE_PID_IGNORE);
7747 else
7748 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
7749 current->pid);
7750 }
7751
7752 static ssize_t
pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos,int type)7753 pid_write(struct file *filp, const char __user *ubuf,
7754 size_t cnt, loff_t *ppos, int type)
7755 {
7756 struct seq_file *m = filp->private_data;
7757 struct trace_array *tr = m->private;
7758 struct trace_pid_list *filtered_pids;
7759 struct trace_pid_list *other_pids;
7760 struct trace_pid_list *pid_list;
7761 ssize_t ret;
7762
7763 if (!cnt)
7764 return 0;
7765
7766 mutex_lock(&ftrace_lock);
7767
7768 switch (type) {
7769 case TRACE_PIDS:
7770 filtered_pids = rcu_dereference_protected(tr->function_pids,
7771 lockdep_is_held(&ftrace_lock));
7772 other_pids = rcu_dereference_protected(tr->function_no_pids,
7773 lockdep_is_held(&ftrace_lock));
7774 break;
7775 case TRACE_NO_PIDS:
7776 filtered_pids = rcu_dereference_protected(tr->function_no_pids,
7777 lockdep_is_held(&ftrace_lock));
7778 other_pids = rcu_dereference_protected(tr->function_pids,
7779 lockdep_is_held(&ftrace_lock));
7780 break;
7781 default:
7782 ret = -EINVAL;
7783 WARN_ON_ONCE(1);
7784 goto out;
7785 }
7786
7787 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
7788 if (ret < 0)
7789 goto out;
7790
7791 switch (type) {
7792 case TRACE_PIDS:
7793 rcu_assign_pointer(tr->function_pids, pid_list);
7794 break;
7795 case TRACE_NO_PIDS:
7796 rcu_assign_pointer(tr->function_no_pids, pid_list);
7797 break;
7798 }
7799
7800
7801 if (filtered_pids) {
7802 synchronize_rcu();
7803 trace_pid_list_free(filtered_pids);
7804 } else if (pid_list && !other_pids) {
7805 /* Register a probe to set whether to ignore the tracing of a task */
7806 register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
7807 }
7808
7809 /*
7810 * Ignoring of pids is done at task switch. But we have to
7811 * check for those tasks that are currently running.
7812 * Always do this in case a pid was appended or removed.
7813 */
7814 on_each_cpu(ignore_task_cpu, tr, 1);
7815
7816 ftrace_update_pid_func();
7817 ftrace_startup_all(0);
7818 out:
7819 mutex_unlock(&ftrace_lock);
7820
7821 if (ret > 0)
7822 *ppos += ret;
7823
7824 return ret;
7825 }
7826
7827 static ssize_t
ftrace_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)7828 ftrace_pid_write(struct file *filp, const char __user *ubuf,
7829 size_t cnt, loff_t *ppos)
7830 {
7831 return pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
7832 }
7833
7834 static ssize_t
ftrace_no_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)7835 ftrace_no_pid_write(struct file *filp, const char __user *ubuf,
7836 size_t cnt, loff_t *ppos)
7837 {
7838 return pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
7839 }
7840
7841 static int
ftrace_pid_release(struct inode * inode,struct file * file)7842 ftrace_pid_release(struct inode *inode, struct file *file)
7843 {
7844 struct trace_array *tr = inode->i_private;
7845
7846 trace_array_put(tr);
7847
7848 return seq_release(inode, file);
7849 }
7850
7851 static const struct file_operations ftrace_pid_fops = {
7852 .open = ftrace_pid_open,
7853 .write = ftrace_pid_write,
7854 .read = seq_read,
7855 .llseek = tracing_lseek,
7856 .release = ftrace_pid_release,
7857 };
7858
7859 static const struct file_operations ftrace_no_pid_fops = {
7860 .open = ftrace_no_pid_open,
7861 .write = ftrace_no_pid_write,
7862 .read = seq_read,
7863 .llseek = tracing_lseek,
7864 .release = ftrace_pid_release,
7865 };
7866
ftrace_init_tracefs(struct trace_array * tr,struct dentry * d_tracer)7867 void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
7868 {
7869 trace_create_file("set_ftrace_pid", TRACE_MODE_WRITE, d_tracer,
7870 tr, &ftrace_pid_fops);
7871 trace_create_file("set_ftrace_notrace_pid", TRACE_MODE_WRITE,
7872 d_tracer, tr, &ftrace_no_pid_fops);
7873 }
7874
ftrace_init_tracefs_toplevel(struct trace_array * tr,struct dentry * d_tracer)7875 void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
7876 struct dentry *d_tracer)
7877 {
7878 /* Only the top level directory has the dyn_tracefs and profile */
7879 WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
7880
7881 ftrace_init_dyn_tracefs(d_tracer);
7882 ftrace_profile_tracefs(d_tracer);
7883 }
7884
7885 /**
7886 * ftrace_kill - kill ftrace
7887 *
7888 * This function should be used by panic code. It stops ftrace
7889 * but in a not so nice way. If you need to simply kill ftrace
7890 * from a non-atomic section, use ftrace_kill.
7891 */
ftrace_kill(void)7892 void ftrace_kill(void)
7893 {
7894 ftrace_disabled = 1;
7895 ftrace_enabled = 0;
7896 ftrace_trace_function = ftrace_stub;
7897 }
7898
7899 /**
7900 * ftrace_is_dead - Test if ftrace is dead or not.
7901 *
7902 * Returns: 1 if ftrace is "dead", zero otherwise.
7903 */
ftrace_is_dead(void)7904 int ftrace_is_dead(void)
7905 {
7906 return ftrace_disabled;
7907 }
7908
7909 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
7910 /*
7911 * When registering ftrace_ops with IPMODIFY, it is necessary to make sure
7912 * it doesn't conflict with any direct ftrace_ops. If there is existing
7913 * direct ftrace_ops on a kernel function being patched, call
7914 * FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_PEER on it to enable sharing.
7915 *
7916 * @ops: ftrace_ops being registered.
7917 *
7918 * Returns:
7919 * 0 on success;
7920 * Negative on failure.
7921 */
prepare_direct_functions_for_ipmodify(struct ftrace_ops * ops)7922 static int prepare_direct_functions_for_ipmodify(struct ftrace_ops *ops)
7923 {
7924 struct ftrace_func_entry *entry;
7925 struct ftrace_hash *hash;
7926 struct ftrace_ops *op;
7927 int size, i, ret;
7928
7929 lockdep_assert_held_once(&direct_mutex);
7930
7931 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
7932 return 0;
7933
7934 hash = ops->func_hash->filter_hash;
7935 size = 1 << hash->size_bits;
7936 for (i = 0; i < size; i++) {
7937 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
7938 unsigned long ip = entry->ip;
7939 bool found_op = false;
7940
7941 mutex_lock(&ftrace_lock);
7942 do_for_each_ftrace_op(op, ftrace_ops_list) {
7943 if (!(op->flags & FTRACE_OPS_FL_DIRECT))
7944 continue;
7945 if (ops_references_ip(op, ip)) {
7946 found_op = true;
7947 break;
7948 }
7949 } while_for_each_ftrace_op(op);
7950 mutex_unlock(&ftrace_lock);
7951
7952 if (found_op) {
7953 if (!op->ops_func)
7954 return -EBUSY;
7955
7956 ret = op->ops_func(op, FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_PEER);
7957 if (ret)
7958 return ret;
7959 }
7960 }
7961 }
7962
7963 return 0;
7964 }
7965
7966 /*
7967 * Similar to prepare_direct_functions_for_ipmodify, clean up after ops
7968 * with IPMODIFY is unregistered. The cleanup is optional for most DIRECT
7969 * ops.
7970 */
cleanup_direct_functions_after_ipmodify(struct ftrace_ops * ops)7971 static void cleanup_direct_functions_after_ipmodify(struct ftrace_ops *ops)
7972 {
7973 struct ftrace_func_entry *entry;
7974 struct ftrace_hash *hash;
7975 struct ftrace_ops *op;
7976 int size, i;
7977
7978 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
7979 return;
7980
7981 mutex_lock(&direct_mutex);
7982
7983 hash = ops->func_hash->filter_hash;
7984 size = 1 << hash->size_bits;
7985 for (i = 0; i < size; i++) {
7986 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
7987 unsigned long ip = entry->ip;
7988 bool found_op = false;
7989
7990 mutex_lock(&ftrace_lock);
7991 do_for_each_ftrace_op(op, ftrace_ops_list) {
7992 if (!(op->flags & FTRACE_OPS_FL_DIRECT))
7993 continue;
7994 if (ops_references_ip(op, ip)) {
7995 found_op = true;
7996 break;
7997 }
7998 } while_for_each_ftrace_op(op);
7999 mutex_unlock(&ftrace_lock);
8000
8001 /* The cleanup is optional, ignore any errors */
8002 if (found_op && op->ops_func)
8003 op->ops_func(op, FTRACE_OPS_CMD_DISABLE_SHARE_IPMODIFY_PEER);
8004 }
8005 }
8006 mutex_unlock(&direct_mutex);
8007 }
8008
8009 #define lock_direct_mutex() mutex_lock(&direct_mutex)
8010 #define unlock_direct_mutex() mutex_unlock(&direct_mutex)
8011
8012 #else /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
8013
prepare_direct_functions_for_ipmodify(struct ftrace_ops * ops)8014 static int prepare_direct_functions_for_ipmodify(struct ftrace_ops *ops)
8015 {
8016 return 0;
8017 }
8018
cleanup_direct_functions_after_ipmodify(struct ftrace_ops * ops)8019 static void cleanup_direct_functions_after_ipmodify(struct ftrace_ops *ops)
8020 {
8021 }
8022
8023 #define lock_direct_mutex() do { } while (0)
8024 #define unlock_direct_mutex() do { } while (0)
8025
8026 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
8027
8028 /*
8029 * Similar to register_ftrace_function, except we don't lock direct_mutex.
8030 */
register_ftrace_function_nolock(struct ftrace_ops * ops)8031 static int register_ftrace_function_nolock(struct ftrace_ops *ops)
8032 {
8033 int ret;
8034
8035 ftrace_ops_init(ops);
8036
8037 mutex_lock(&ftrace_lock);
8038
8039 ret = ftrace_startup(ops, 0);
8040
8041 mutex_unlock(&ftrace_lock);
8042
8043 return ret;
8044 }
8045
8046 /**
8047 * register_ftrace_function - register a function for profiling
8048 * @ops: ops structure that holds the function for profiling.
8049 *
8050 * Register a function to be called by all functions in the
8051 * kernel.
8052 *
8053 * Note: @ops->func and all the functions it calls must be labeled
8054 * with "notrace", otherwise it will go into a
8055 * recursive loop.
8056 */
register_ftrace_function(struct ftrace_ops * ops)8057 int register_ftrace_function(struct ftrace_ops *ops)
8058 {
8059 int ret;
8060
8061 lock_direct_mutex();
8062 ret = prepare_direct_functions_for_ipmodify(ops);
8063 if (ret < 0)
8064 goto out_unlock;
8065
8066 ret = register_ftrace_function_nolock(ops);
8067
8068 out_unlock:
8069 unlock_direct_mutex();
8070 return ret;
8071 }
8072 EXPORT_SYMBOL_GPL(register_ftrace_function);
8073
8074 /**
8075 * unregister_ftrace_function - unregister a function for profiling.
8076 * @ops: ops structure that holds the function to unregister
8077 *
8078 * Unregister a function that was added to be called by ftrace profiling.
8079 */
unregister_ftrace_function(struct ftrace_ops * ops)8080 int unregister_ftrace_function(struct ftrace_ops *ops)
8081 {
8082 int ret;
8083
8084 mutex_lock(&ftrace_lock);
8085 ret = ftrace_shutdown(ops, 0);
8086 mutex_unlock(&ftrace_lock);
8087
8088 cleanup_direct_functions_after_ipmodify(ops);
8089 return ret;
8090 }
8091 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
8092
symbols_cmp(const void * a,const void * b)8093 static int symbols_cmp(const void *a, const void *b)
8094 {
8095 const char **str_a = (const char **) a;
8096 const char **str_b = (const char **) b;
8097
8098 return strcmp(*str_a, *str_b);
8099 }
8100
8101 struct kallsyms_data {
8102 unsigned long *addrs;
8103 const char **syms;
8104 size_t cnt;
8105 size_t found;
8106 };
8107
8108 /* This function gets called for all kernel and module symbols
8109 * and returns 1 in case we resolved all the requested symbols,
8110 * 0 otherwise.
8111 */
kallsyms_callback(void * data,const char * name,unsigned long addr)8112 static int kallsyms_callback(void *data, const char *name, unsigned long addr)
8113 {
8114 struct kallsyms_data *args = data;
8115 const char **sym;
8116 int idx;
8117
8118 sym = bsearch(&name, args->syms, args->cnt, sizeof(*args->syms), symbols_cmp);
8119 if (!sym)
8120 return 0;
8121
8122 idx = sym - args->syms;
8123 if (args->addrs[idx])
8124 return 0;
8125
8126 if (!ftrace_location(addr))
8127 return 0;
8128
8129 args->addrs[idx] = addr;
8130 args->found++;
8131 return args->found == args->cnt ? 1 : 0;
8132 }
8133
8134 /**
8135 * ftrace_lookup_symbols - Lookup addresses for array of symbols
8136 *
8137 * @sorted_syms: array of symbols pointers symbols to resolve,
8138 * must be alphabetically sorted
8139 * @cnt: number of symbols/addresses in @syms/@addrs arrays
8140 * @addrs: array for storing resulting addresses
8141 *
8142 * This function looks up addresses for array of symbols provided in
8143 * @syms array (must be alphabetically sorted) and stores them in
8144 * @addrs array, which needs to be big enough to store at least @cnt
8145 * addresses.
8146 *
8147 * Returns: 0 if all provided symbols are found, -ESRCH otherwise.
8148 */
ftrace_lookup_symbols(const char ** sorted_syms,size_t cnt,unsigned long * addrs)8149 int ftrace_lookup_symbols(const char **sorted_syms, size_t cnt, unsigned long *addrs)
8150 {
8151 struct kallsyms_data args;
8152 int found_all;
8153
8154 memset(addrs, 0, sizeof(*addrs) * cnt);
8155 args.addrs = addrs;
8156 args.syms = sorted_syms;
8157 args.cnt = cnt;
8158 args.found = 0;
8159
8160 found_all = kallsyms_on_each_symbol(kallsyms_callback, &args);
8161 if (found_all)
8162 return 0;
8163 found_all = module_kallsyms_on_each_symbol(NULL, kallsyms_callback, &args);
8164 return found_all ? 0 : -ESRCH;
8165 }
8166
8167 #ifdef CONFIG_SYSCTL
8168
8169 #ifdef CONFIG_DYNAMIC_FTRACE
ftrace_startup_sysctl(void)8170 static void ftrace_startup_sysctl(void)
8171 {
8172 int command;
8173
8174 if (unlikely(ftrace_disabled))
8175 return;
8176
8177 /* Force update next time */
8178 saved_ftrace_func = NULL;
8179 /* ftrace_start_up is true if we want ftrace running */
8180 if (ftrace_start_up) {
8181 command = FTRACE_UPDATE_CALLS;
8182 if (ftrace_graph_active)
8183 command |= FTRACE_START_FUNC_RET;
8184 ftrace_startup_enable(command);
8185 }
8186 }
8187
ftrace_shutdown_sysctl(void)8188 static void ftrace_shutdown_sysctl(void)
8189 {
8190 int command;
8191
8192 if (unlikely(ftrace_disabled))
8193 return;
8194
8195 /* ftrace_start_up is true if ftrace is running */
8196 if (ftrace_start_up) {
8197 command = FTRACE_DISABLE_CALLS;
8198 if (ftrace_graph_active)
8199 command |= FTRACE_STOP_FUNC_RET;
8200 ftrace_run_update_code(command);
8201 }
8202 }
8203 #else
8204 # define ftrace_startup_sysctl() do { } while (0)
8205 # define ftrace_shutdown_sysctl() do { } while (0)
8206 #endif /* CONFIG_DYNAMIC_FTRACE */
8207
is_permanent_ops_registered(void)8208 static bool is_permanent_ops_registered(void)
8209 {
8210 struct ftrace_ops *op;
8211
8212 do_for_each_ftrace_op(op, ftrace_ops_list) {
8213 if (op->flags & FTRACE_OPS_FL_PERMANENT)
8214 return true;
8215 } while_for_each_ftrace_op(op);
8216
8217 return false;
8218 }
8219
8220 static int
ftrace_enable_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)8221 ftrace_enable_sysctl(struct ctl_table *table, int write,
8222 void *buffer, size_t *lenp, loff_t *ppos)
8223 {
8224 int ret = -ENODEV;
8225
8226 mutex_lock(&ftrace_lock);
8227
8228 if (unlikely(ftrace_disabled))
8229 goto out;
8230
8231 ret = proc_dointvec(table, write, buffer, lenp, ppos);
8232
8233 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
8234 goto out;
8235
8236 if (ftrace_enabled) {
8237
8238 /* we are starting ftrace again */
8239 if (rcu_dereference_protected(ftrace_ops_list,
8240 lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
8241 update_ftrace_function();
8242
8243 ftrace_startup_sysctl();
8244
8245 } else {
8246 if (is_permanent_ops_registered()) {
8247 ftrace_enabled = true;
8248 ret = -EBUSY;
8249 goto out;
8250 }
8251
8252 /* stopping ftrace calls (just send to ftrace_stub) */
8253 ftrace_trace_function = ftrace_stub;
8254
8255 ftrace_shutdown_sysctl();
8256 }
8257
8258 last_ftrace_enabled = !!ftrace_enabled;
8259 out:
8260 mutex_unlock(&ftrace_lock);
8261 return ret;
8262 }
8263
8264 static struct ctl_table ftrace_sysctls[] = {
8265 {
8266 .procname = "ftrace_enabled",
8267 .data = &ftrace_enabled,
8268 .maxlen = sizeof(int),
8269 .mode = 0644,
8270 .proc_handler = ftrace_enable_sysctl,
8271 },
8272 {}
8273 };
8274
ftrace_sysctl_init(void)8275 static int __init ftrace_sysctl_init(void)
8276 {
8277 register_sysctl_init("kernel", ftrace_sysctls);
8278 return 0;
8279 }
8280 late_initcall(ftrace_sysctl_init);
8281 #endif
8282