1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Infrastructure for profiling code inserted by 'gcc -pg'.
4 *
5 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
7 *
8 * Originally ported from the -rt patch by:
9 * Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
10 *
11 * Based on code in the latency_tracer, that is:
12 *
13 * Copyright (C) 2004-2006 Ingo Molnar
14 * Copyright (C) 2004 Nadia Yvette Chambers
15 */
16
17 #include <linux/stop_machine.h>
18 #include <linux/clocksource.h>
19 #include <linux/sched/task.h>
20 #include <linux/kallsyms.h>
21 #include <linux/security.h>
22 #include <linux/seq_file.h>
23 #include <linux/tracefs.h>
24 #include <linux/hardirq.h>
25 #include <linux/kthread.h>
26 #include <linux/uaccess.h>
27 #include <linux/bsearch.h>
28 #include <linux/module.h>
29 #include <linux/ftrace.h>
30 #include <linux/sysctl.h>
31 #include <linux/slab.h>
32 #include <linux/ctype.h>
33 #include <linux/sort.h>
34 #include <linux/list.h>
35 #include <linux/hash.h>
36 #include <linux/rcupdate.h>
37 #include <linux/kprobes.h>
38
39 #include <trace/events/sched.h>
40
41 #include <asm/sections.h>
42 #include <asm/setup.h>
43
44 #include "ftrace_internal.h"
45 #include "trace_output.h"
46 #include "trace_stat.h"
47
48 /* Flags that do not get reset */
49 #define FTRACE_NOCLEAR_FLAGS (FTRACE_FL_DISABLED | FTRACE_FL_TOUCHED | \
50 FTRACE_FL_MODIFIED)
51
52 #define FTRACE_INVALID_FUNCTION "__ftrace_invalid_address__"
53
54 #define FTRACE_WARN_ON(cond) \
55 ({ \
56 int ___r = cond; \
57 if (WARN_ON(___r)) \
58 ftrace_kill(); \
59 ___r; \
60 })
61
62 #define FTRACE_WARN_ON_ONCE(cond) \
63 ({ \
64 int ___r = cond; \
65 if (WARN_ON_ONCE(___r)) \
66 ftrace_kill(); \
67 ___r; \
68 })
69
70 /* hash bits for specific function selection */
71 #define FTRACE_HASH_DEFAULT_BITS 10
72 #define FTRACE_HASH_MAX_BITS 12
73
74 #ifdef CONFIG_DYNAMIC_FTRACE
75 #define INIT_OPS_HASH(opsname) \
76 .func_hash = &opsname.local_hash, \
77 .local_hash.regex_lock = __MUTEX_INITIALIZER(opsname.local_hash.regex_lock), \
78 .subop_list = LIST_HEAD_INIT(opsname.subop_list),
79 #else
80 #define INIT_OPS_HASH(opsname)
81 #endif
82
83 enum {
84 FTRACE_MODIFY_ENABLE_FL = (1 << 0),
85 FTRACE_MODIFY_MAY_SLEEP_FL = (1 << 1),
86 };
87
88 struct ftrace_ops ftrace_list_end __read_mostly = {
89 .func = ftrace_stub,
90 .flags = FTRACE_OPS_FL_STUB,
91 INIT_OPS_HASH(ftrace_list_end)
92 };
93
94 /* ftrace_enabled is a method to turn ftrace on or off */
95 int ftrace_enabled __read_mostly;
96 static int __maybe_unused last_ftrace_enabled;
97
98 /* Current function tracing op */
99 struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
100 /* What to set function_trace_op to */
101 static struct ftrace_ops *set_function_trace_op;
102
ftrace_pids_enabled(struct ftrace_ops * ops)103 bool ftrace_pids_enabled(struct ftrace_ops *ops)
104 {
105 struct trace_array *tr;
106
107 if (!(ops->flags & FTRACE_OPS_FL_PID) || !ops->private)
108 return false;
109
110 tr = ops->private;
111
112 return tr->function_pids != NULL || tr->function_no_pids != NULL;
113 }
114
115 static void ftrace_update_trampoline(struct ftrace_ops *ops);
116
117 /*
118 * ftrace_disabled is set when an anomaly is discovered.
119 * ftrace_disabled is much stronger than ftrace_enabled.
120 */
121 static int ftrace_disabled __read_mostly;
122
123 DEFINE_MUTEX(ftrace_lock);
124
125 struct ftrace_ops __rcu *ftrace_ops_list __read_mostly = (struct ftrace_ops __rcu *)&ftrace_list_end;
126 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
127 struct ftrace_ops global_ops;
128
129 /* Defined by vmlinux.lds.h see the comment above arch_ftrace_ops_list_func for details */
130 void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
131 struct ftrace_ops *op, struct ftrace_regs *fregs);
132
133 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_CALL_OPS
134 /*
135 * Stub used to invoke the list ops without requiring a separate trampoline.
136 */
137 const struct ftrace_ops ftrace_list_ops = {
138 .func = ftrace_ops_list_func,
139 .flags = FTRACE_OPS_FL_STUB,
140 };
141
ftrace_ops_nop_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)142 static void ftrace_ops_nop_func(unsigned long ip, unsigned long parent_ip,
143 struct ftrace_ops *op,
144 struct ftrace_regs *fregs)
145 {
146 /* do nothing */
147 }
148
149 /*
150 * Stub used when a call site is disabled. May be called transiently by threads
151 * which have made it into ftrace_caller but haven't yet recovered the ops at
152 * the point the call site is disabled.
153 */
154 const struct ftrace_ops ftrace_nop_ops = {
155 .func = ftrace_ops_nop_func,
156 .flags = FTRACE_OPS_FL_STUB,
157 };
158 #endif
159
ftrace_ops_init(struct ftrace_ops * ops)160 static inline void ftrace_ops_init(struct ftrace_ops *ops)
161 {
162 #ifdef CONFIG_DYNAMIC_FTRACE
163 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
164 mutex_init(&ops->local_hash.regex_lock);
165 INIT_LIST_HEAD(&ops->subop_list);
166 ops->func_hash = &ops->local_hash;
167 ops->flags |= FTRACE_OPS_FL_INITIALIZED;
168 }
169 #endif
170 }
171
172 /* Call this function for when a callback filters on set_ftrace_pid */
ftrace_pid_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)173 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
174 struct ftrace_ops *op, struct ftrace_regs *fregs)
175 {
176 struct trace_array *tr = op->private;
177 int pid;
178
179 if (tr) {
180 pid = this_cpu_read(tr->array_buffer.data->ftrace_ignore_pid);
181 if (pid == FTRACE_PID_IGNORE)
182 return;
183 if (pid != FTRACE_PID_TRACE &&
184 pid != current->pid)
185 return;
186 }
187
188 op->saved_func(ip, parent_ip, op, fregs);
189 }
190
ftrace_sync_ipi(void * data)191 static void ftrace_sync_ipi(void *data)
192 {
193 /* Probably not needed, but do it anyway */
194 smp_rmb();
195 }
196
ftrace_ops_get_list_func(struct ftrace_ops * ops)197 static ftrace_func_t ftrace_ops_get_list_func(struct ftrace_ops *ops)
198 {
199 /*
200 * If this is a dynamic or RCU ops, or we force list func,
201 * then it needs to call the list anyway.
202 */
203 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_RCU) ||
204 FTRACE_FORCE_LIST_FUNC)
205 return ftrace_ops_list_func;
206
207 return ftrace_ops_get_func(ops);
208 }
209
update_ftrace_function(void)210 static void update_ftrace_function(void)
211 {
212 ftrace_func_t func;
213
214 /*
215 * Prepare the ftrace_ops that the arch callback will use.
216 * If there's only one ftrace_ops registered, the ftrace_ops_list
217 * will point to the ops we want.
218 */
219 set_function_trace_op = rcu_dereference_protected(ftrace_ops_list,
220 lockdep_is_held(&ftrace_lock));
221
222 /* If there's no ftrace_ops registered, just call the stub function */
223 if (set_function_trace_op == &ftrace_list_end) {
224 func = ftrace_stub;
225
226 /*
227 * If we are at the end of the list and this ops is
228 * recursion safe and not dynamic and the arch supports passing ops,
229 * then have the mcount trampoline call the function directly.
230 */
231 } else if (rcu_dereference_protected(ftrace_ops_list->next,
232 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
233 func = ftrace_ops_get_list_func(ftrace_ops_list);
234
235 } else {
236 /* Just use the default ftrace_ops */
237 set_function_trace_op = &ftrace_list_end;
238 func = ftrace_ops_list_func;
239 }
240
241 /* If there's no change, then do nothing more here */
242 if (ftrace_trace_function == func)
243 return;
244
245 /*
246 * If we are using the list function, it doesn't care
247 * about the function_trace_ops.
248 */
249 if (func == ftrace_ops_list_func) {
250 ftrace_trace_function = func;
251 /*
252 * Don't even bother setting function_trace_ops,
253 * it would be racy to do so anyway.
254 */
255 return;
256 }
257
258 #ifndef CONFIG_DYNAMIC_FTRACE
259 /*
260 * For static tracing, we need to be a bit more careful.
261 * The function change takes affect immediately. Thus,
262 * we need to coordinate the setting of the function_trace_ops
263 * with the setting of the ftrace_trace_function.
264 *
265 * Set the function to the list ops, which will call the
266 * function we want, albeit indirectly, but it handles the
267 * ftrace_ops and doesn't depend on function_trace_op.
268 */
269 ftrace_trace_function = ftrace_ops_list_func;
270 /*
271 * Make sure all CPUs see this. Yes this is slow, but static
272 * tracing is slow and nasty to have enabled.
273 */
274 synchronize_rcu_tasks_rude();
275 /* Now all cpus are using the list ops. */
276 function_trace_op = set_function_trace_op;
277 /* Make sure the function_trace_op is visible on all CPUs */
278 smp_wmb();
279 /* Nasty way to force a rmb on all cpus */
280 smp_call_function(ftrace_sync_ipi, NULL, 1);
281 /* OK, we are all set to update the ftrace_trace_function now! */
282 #endif /* !CONFIG_DYNAMIC_FTRACE */
283
284 ftrace_trace_function = func;
285 }
286
add_ftrace_ops(struct ftrace_ops __rcu ** list,struct ftrace_ops * ops)287 static void add_ftrace_ops(struct ftrace_ops __rcu **list,
288 struct ftrace_ops *ops)
289 {
290 rcu_assign_pointer(ops->next, *list);
291
292 /*
293 * We are entering ops into the list but another
294 * CPU might be walking that list. We need to make sure
295 * the ops->next pointer is valid before another CPU sees
296 * the ops pointer included into the list.
297 */
298 rcu_assign_pointer(*list, ops);
299 }
300
remove_ftrace_ops(struct ftrace_ops __rcu ** list,struct ftrace_ops * ops)301 static int remove_ftrace_ops(struct ftrace_ops __rcu **list,
302 struct ftrace_ops *ops)
303 {
304 struct ftrace_ops **p;
305
306 /*
307 * If we are removing the last function, then simply point
308 * to the ftrace_stub.
309 */
310 if (rcu_dereference_protected(*list,
311 lockdep_is_held(&ftrace_lock)) == ops &&
312 rcu_dereference_protected(ops->next,
313 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
314 rcu_assign_pointer(*list, &ftrace_list_end);
315 return 0;
316 }
317
318 for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
319 if (*p == ops)
320 break;
321
322 if (*p != ops)
323 return -1;
324
325 *p = (*p)->next;
326 return 0;
327 }
328
329 static void ftrace_update_trampoline(struct ftrace_ops *ops);
330
__register_ftrace_function(struct ftrace_ops * ops)331 int __register_ftrace_function(struct ftrace_ops *ops)
332 {
333 if (ops->flags & FTRACE_OPS_FL_DELETED)
334 return -EINVAL;
335
336 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
337 return -EBUSY;
338
339 #ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
340 /*
341 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
342 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
343 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
344 */
345 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
346 !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
347 return -EINVAL;
348
349 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
350 ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
351 #endif
352 if (!ftrace_enabled && (ops->flags & FTRACE_OPS_FL_PERMANENT))
353 return -EBUSY;
354
355 if (!is_kernel_core_data((unsigned long)ops))
356 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
357
358 add_ftrace_ops(&ftrace_ops_list, ops);
359
360 /* Always save the function, and reset at unregistering */
361 ops->saved_func = ops->func;
362
363 if (ftrace_pids_enabled(ops))
364 ops->func = ftrace_pid_func;
365
366 ftrace_update_trampoline(ops);
367
368 if (ftrace_enabled)
369 update_ftrace_function();
370
371 return 0;
372 }
373
__unregister_ftrace_function(struct ftrace_ops * ops)374 int __unregister_ftrace_function(struct ftrace_ops *ops)
375 {
376 int ret;
377
378 if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
379 return -EBUSY;
380
381 ret = remove_ftrace_ops(&ftrace_ops_list, ops);
382
383 if (ret < 0)
384 return ret;
385
386 if (ftrace_enabled)
387 update_ftrace_function();
388
389 ops->func = ops->saved_func;
390
391 return 0;
392 }
393
ftrace_update_pid_func(void)394 static void ftrace_update_pid_func(void)
395 {
396 struct ftrace_ops *op;
397
398 /* Only do something if we are tracing something */
399 if (ftrace_trace_function == ftrace_stub)
400 return;
401
402 do_for_each_ftrace_op(op, ftrace_ops_list) {
403 if (op->flags & FTRACE_OPS_FL_PID) {
404 op->func = ftrace_pids_enabled(op) ?
405 ftrace_pid_func : op->saved_func;
406 ftrace_update_trampoline(op);
407 }
408 } while_for_each_ftrace_op(op);
409
410 fgraph_update_pid_func();
411
412 update_ftrace_function();
413 }
414
415 #ifdef CONFIG_FUNCTION_PROFILER
416 struct ftrace_profile {
417 struct hlist_node node;
418 unsigned long ip;
419 unsigned long counter;
420 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
421 unsigned long long time;
422 unsigned long long time_squared;
423 #endif
424 };
425
426 struct ftrace_profile_page {
427 struct ftrace_profile_page *next;
428 unsigned long index;
429 struct ftrace_profile records[];
430 };
431
432 struct ftrace_profile_stat {
433 atomic_t disabled;
434 struct hlist_head *hash;
435 struct ftrace_profile_page *pages;
436 struct ftrace_profile_page *start;
437 struct tracer_stat stat;
438 };
439
440 #define PROFILE_RECORDS_SIZE \
441 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
442
443 #define PROFILES_PER_PAGE \
444 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
445
446 static int ftrace_profile_enabled __read_mostly;
447
448 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
449 static DEFINE_MUTEX(ftrace_profile_lock);
450
451 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
452
453 #define FTRACE_PROFILE_HASH_BITS 10
454 #define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
455
456 static void *
function_stat_next(void * v,int idx)457 function_stat_next(void *v, int idx)
458 {
459 struct ftrace_profile *rec = v;
460 struct ftrace_profile_page *pg;
461
462 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
463
464 again:
465 if (idx != 0)
466 rec++;
467
468 if ((void *)rec >= (void *)&pg->records[pg->index]) {
469 pg = pg->next;
470 if (!pg)
471 return NULL;
472 rec = &pg->records[0];
473 if (!rec->counter)
474 goto again;
475 }
476
477 return rec;
478 }
479
function_stat_start(struct tracer_stat * trace)480 static void *function_stat_start(struct tracer_stat *trace)
481 {
482 struct ftrace_profile_stat *stat =
483 container_of(trace, struct ftrace_profile_stat, stat);
484
485 if (!stat || !stat->start)
486 return NULL;
487
488 return function_stat_next(&stat->start->records[0], 0);
489 }
490
491 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
492 /* function graph compares on total time */
function_stat_cmp(const void * p1,const void * p2)493 static int function_stat_cmp(const void *p1, const void *p2)
494 {
495 const struct ftrace_profile *a = p1;
496 const struct ftrace_profile *b = p2;
497
498 if (a->time < b->time)
499 return -1;
500 if (a->time > b->time)
501 return 1;
502 else
503 return 0;
504 }
505 #else
506 /* not function graph compares against hits */
function_stat_cmp(const void * p1,const void * p2)507 static int function_stat_cmp(const void *p1, const void *p2)
508 {
509 const struct ftrace_profile *a = p1;
510 const struct ftrace_profile *b = p2;
511
512 if (a->counter < b->counter)
513 return -1;
514 if (a->counter > b->counter)
515 return 1;
516 else
517 return 0;
518 }
519 #endif
520
function_stat_headers(struct seq_file * m)521 static int function_stat_headers(struct seq_file *m)
522 {
523 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
524 seq_puts(m, " Function "
525 "Hit Time Avg s^2\n"
526 " -------- "
527 "--- ---- --- ---\n");
528 #else
529 seq_puts(m, " Function Hit\n"
530 " -------- ---\n");
531 #endif
532 return 0;
533 }
534
function_stat_show(struct seq_file * m,void * v)535 static int function_stat_show(struct seq_file *m, void *v)
536 {
537 struct ftrace_profile *rec = v;
538 char str[KSYM_SYMBOL_LEN];
539 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
540 static struct trace_seq s;
541 unsigned long long avg;
542 unsigned long long stddev;
543 unsigned long long stddev_denom;
544 #endif
545 guard(mutex)(&ftrace_profile_lock);
546
547 /* we raced with function_profile_reset() */
548 if (unlikely(rec->counter == 0))
549 return -EBUSY;
550
551 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
552 avg = div64_ul(rec->time, rec->counter);
553 if (tracing_thresh && (avg < tracing_thresh))
554 return 0;
555 #endif
556
557 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
558 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
559
560 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
561 seq_puts(m, " ");
562
563 /*
564 * Variance formula:
565 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
566 * Maybe Welford's method is better here?
567 * Divide only by 1000 for ns^2 -> us^2 conversion.
568 * trace_print_graph_duration will divide by 1000 again.
569 */
570 stddev = 0;
571 stddev_denom = rec->counter * (rec->counter - 1) * 1000;
572 if (stddev_denom) {
573 stddev = rec->counter * rec->time_squared -
574 rec->time * rec->time;
575 stddev = div64_ul(stddev, stddev_denom);
576 }
577
578 trace_seq_init(&s);
579 trace_print_graph_duration(rec->time, &s);
580 trace_seq_puts(&s, " ");
581 trace_print_graph_duration(avg, &s);
582 trace_seq_puts(&s, " ");
583 trace_print_graph_duration(stddev, &s);
584 trace_print_seq(m, &s);
585 #endif
586 seq_putc(m, '\n');
587
588 return 0;
589 }
590
ftrace_profile_reset(struct ftrace_profile_stat * stat)591 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
592 {
593 struct ftrace_profile_page *pg;
594
595 pg = stat->pages = stat->start;
596
597 while (pg) {
598 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
599 pg->index = 0;
600 pg = pg->next;
601 }
602
603 memset(stat->hash, 0,
604 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
605 }
606
ftrace_profile_pages_init(struct ftrace_profile_stat * stat)607 static int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
608 {
609 struct ftrace_profile_page *pg;
610 int functions;
611 int pages;
612 int i;
613
614 /* If we already allocated, do nothing */
615 if (stat->pages)
616 return 0;
617
618 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
619 if (!stat->pages)
620 return -ENOMEM;
621
622 #ifdef CONFIG_DYNAMIC_FTRACE
623 functions = ftrace_update_tot_cnt;
624 #else
625 /*
626 * We do not know the number of functions that exist because
627 * dynamic tracing is what counts them. With past experience
628 * we have around 20K functions. That should be more than enough.
629 * It is highly unlikely we will execute every function in
630 * the kernel.
631 */
632 functions = 20000;
633 #endif
634
635 pg = stat->start = stat->pages;
636
637 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
638
639 for (i = 1; i < pages; i++) {
640 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
641 if (!pg->next)
642 goto out_free;
643 pg = pg->next;
644 }
645
646 return 0;
647
648 out_free:
649 pg = stat->start;
650 while (pg) {
651 unsigned long tmp = (unsigned long)pg;
652
653 pg = pg->next;
654 free_page(tmp);
655 }
656
657 stat->pages = NULL;
658 stat->start = NULL;
659
660 return -ENOMEM;
661 }
662
ftrace_profile_init_cpu(int cpu)663 static int ftrace_profile_init_cpu(int cpu)
664 {
665 struct ftrace_profile_stat *stat;
666 int size;
667
668 stat = &per_cpu(ftrace_profile_stats, cpu);
669
670 if (stat->hash) {
671 /* If the profile is already created, simply reset it */
672 ftrace_profile_reset(stat);
673 return 0;
674 }
675
676 /*
677 * We are profiling all functions, but usually only a few thousand
678 * functions are hit. We'll make a hash of 1024 items.
679 */
680 size = FTRACE_PROFILE_HASH_SIZE;
681
682 stat->hash = kcalloc(size, sizeof(struct hlist_head), GFP_KERNEL);
683
684 if (!stat->hash)
685 return -ENOMEM;
686
687 /* Preallocate the function profiling pages */
688 if (ftrace_profile_pages_init(stat) < 0) {
689 kfree(stat->hash);
690 stat->hash = NULL;
691 return -ENOMEM;
692 }
693
694 return 0;
695 }
696
ftrace_profile_init(void)697 static int ftrace_profile_init(void)
698 {
699 int cpu;
700 int ret = 0;
701
702 for_each_possible_cpu(cpu) {
703 ret = ftrace_profile_init_cpu(cpu);
704 if (ret)
705 break;
706 }
707
708 return ret;
709 }
710
711 /* interrupts must be disabled */
712 static struct ftrace_profile *
ftrace_find_profiled_func(struct ftrace_profile_stat * stat,unsigned long ip)713 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
714 {
715 struct ftrace_profile *rec;
716 struct hlist_head *hhd;
717 unsigned long key;
718
719 key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
720 hhd = &stat->hash[key];
721
722 if (hlist_empty(hhd))
723 return NULL;
724
725 hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
726 if (rec->ip == ip)
727 return rec;
728 }
729
730 return NULL;
731 }
732
ftrace_add_profile(struct ftrace_profile_stat * stat,struct ftrace_profile * rec)733 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
734 struct ftrace_profile *rec)
735 {
736 unsigned long key;
737
738 key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
739 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
740 }
741
742 /*
743 * The memory is already allocated, this simply finds a new record to use.
744 */
745 static struct ftrace_profile *
ftrace_profile_alloc(struct ftrace_profile_stat * stat,unsigned long ip)746 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
747 {
748 struct ftrace_profile *rec = NULL;
749
750 /* prevent recursion (from NMIs) */
751 if (atomic_inc_return(&stat->disabled) != 1)
752 goto out;
753
754 /*
755 * Try to find the function again since an NMI
756 * could have added it
757 */
758 rec = ftrace_find_profiled_func(stat, ip);
759 if (rec)
760 goto out;
761
762 if (stat->pages->index == PROFILES_PER_PAGE) {
763 if (!stat->pages->next)
764 goto out;
765 stat->pages = stat->pages->next;
766 }
767
768 rec = &stat->pages->records[stat->pages->index++];
769 rec->ip = ip;
770 ftrace_add_profile(stat, rec);
771
772 out:
773 atomic_dec(&stat->disabled);
774
775 return rec;
776 }
777
778 static void
function_profile_call(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * ops,struct ftrace_regs * fregs)779 function_profile_call(unsigned long ip, unsigned long parent_ip,
780 struct ftrace_ops *ops, struct ftrace_regs *fregs)
781 {
782 struct ftrace_profile_stat *stat;
783 struct ftrace_profile *rec;
784
785 if (!ftrace_profile_enabled)
786 return;
787
788 guard(preempt_notrace)();
789
790 stat = this_cpu_ptr(&ftrace_profile_stats);
791 if (!stat->hash || !ftrace_profile_enabled)
792 return;
793
794 rec = ftrace_find_profiled_func(stat, ip);
795 if (!rec) {
796 rec = ftrace_profile_alloc(stat, ip);
797 if (!rec)
798 return;
799 }
800
801 rec->counter++;
802 }
803
804 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
805 static bool fgraph_graph_time = true;
806
ftrace_graph_graph_time_control(bool enable)807 void ftrace_graph_graph_time_control(bool enable)
808 {
809 fgraph_graph_time = enable;
810 }
811
812 struct profile_fgraph_data {
813 unsigned long long calltime;
814 unsigned long long subtime;
815 unsigned long long sleeptime;
816 };
817
profile_graph_entry(struct ftrace_graph_ent * trace,struct fgraph_ops * gops,struct ftrace_regs * fregs)818 static int profile_graph_entry(struct ftrace_graph_ent *trace,
819 struct fgraph_ops *gops,
820 struct ftrace_regs *fregs)
821 {
822 struct profile_fgraph_data *profile_data;
823
824 function_profile_call(trace->func, 0, NULL, NULL);
825
826 /* If function graph is shutting down, ret_stack can be NULL */
827 if (!current->ret_stack)
828 return 0;
829
830 profile_data = fgraph_reserve_data(gops->idx, sizeof(*profile_data));
831 if (!profile_data)
832 return 0;
833
834 profile_data->subtime = 0;
835 profile_data->sleeptime = current->ftrace_sleeptime;
836 profile_data->calltime = trace_clock_local();
837
838 return 1;
839 }
840
profile_graph_return(struct ftrace_graph_ret * trace,struct fgraph_ops * gops,struct ftrace_regs * fregs)841 static void profile_graph_return(struct ftrace_graph_ret *trace,
842 struct fgraph_ops *gops,
843 struct ftrace_regs *fregs)
844 {
845 struct profile_fgraph_data *profile_data;
846 struct ftrace_profile_stat *stat;
847 unsigned long long calltime;
848 unsigned long long rettime = trace_clock_local();
849 struct ftrace_profile *rec;
850 int size;
851
852 guard(preempt_notrace)();
853
854 stat = this_cpu_ptr(&ftrace_profile_stats);
855 if (!stat->hash || !ftrace_profile_enabled)
856 return;
857
858 profile_data = fgraph_retrieve_data(gops->idx, &size);
859
860 /* If the calltime was zero'd ignore it */
861 if (!profile_data || !profile_data->calltime)
862 return;
863
864 calltime = rettime - profile_data->calltime;
865
866 if (!fgraph_sleep_time) {
867 if (current->ftrace_sleeptime)
868 calltime -= current->ftrace_sleeptime - profile_data->sleeptime;
869 }
870
871 if (!fgraph_graph_time) {
872 struct profile_fgraph_data *parent_data;
873
874 /* Append this call time to the parent time to subtract */
875 parent_data = fgraph_retrieve_parent_data(gops->idx, &size, 1);
876 if (parent_data)
877 parent_data->subtime += calltime;
878
879 if (profile_data->subtime && profile_data->subtime < calltime)
880 calltime -= profile_data->subtime;
881 else
882 calltime = 0;
883 }
884
885 rec = ftrace_find_profiled_func(stat, trace->func);
886 if (rec) {
887 rec->time += calltime;
888 rec->time_squared += calltime * calltime;
889 }
890 }
891
892 static struct fgraph_ops fprofiler_ops = {
893 .entryfunc = &profile_graph_entry,
894 .retfunc = &profile_graph_return,
895 };
896
register_ftrace_profiler(void)897 static int register_ftrace_profiler(void)
898 {
899 ftrace_ops_set_global_filter(&fprofiler_ops.ops);
900 return register_ftrace_graph(&fprofiler_ops);
901 }
902
unregister_ftrace_profiler(void)903 static void unregister_ftrace_profiler(void)
904 {
905 unregister_ftrace_graph(&fprofiler_ops);
906 }
907 #else
908 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
909 .func = function_profile_call,
910 };
911
register_ftrace_profiler(void)912 static int register_ftrace_profiler(void)
913 {
914 ftrace_ops_set_global_filter(&ftrace_profile_ops);
915 return register_ftrace_function(&ftrace_profile_ops);
916 }
917
unregister_ftrace_profiler(void)918 static void unregister_ftrace_profiler(void)
919 {
920 unregister_ftrace_function(&ftrace_profile_ops);
921 }
922 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
923
924 static ssize_t
ftrace_profile_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)925 ftrace_profile_write(struct file *filp, const char __user *ubuf,
926 size_t cnt, loff_t *ppos)
927 {
928 unsigned long val;
929 int ret;
930
931 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
932 if (ret)
933 return ret;
934
935 val = !!val;
936
937 guard(mutex)(&ftrace_profile_lock);
938 if (ftrace_profile_enabled ^ val) {
939 if (val) {
940 ret = ftrace_profile_init();
941 if (ret < 0)
942 return ret;
943
944 ret = register_ftrace_profiler();
945 if (ret < 0)
946 return ret;
947 ftrace_profile_enabled = 1;
948 } else {
949 ftrace_profile_enabled = 0;
950 /*
951 * unregister_ftrace_profiler calls stop_machine
952 * so this acts like an synchronize_rcu.
953 */
954 unregister_ftrace_profiler();
955 }
956 }
957
958 *ppos += cnt;
959
960 return cnt;
961 }
962
963 static ssize_t
ftrace_profile_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)964 ftrace_profile_read(struct file *filp, char __user *ubuf,
965 size_t cnt, loff_t *ppos)
966 {
967 char buf[64]; /* big enough to hold a number */
968 int r;
969
970 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
971 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
972 }
973
974 static const struct file_operations ftrace_profile_fops = {
975 .open = tracing_open_generic,
976 .read = ftrace_profile_read,
977 .write = ftrace_profile_write,
978 .llseek = default_llseek,
979 };
980
981 /* used to initialize the real stat files */
982 static struct tracer_stat function_stats __initdata = {
983 .name = "functions",
984 .stat_start = function_stat_start,
985 .stat_next = function_stat_next,
986 .stat_cmp = function_stat_cmp,
987 .stat_headers = function_stat_headers,
988 .stat_show = function_stat_show
989 };
990
ftrace_profile_tracefs(struct dentry * d_tracer)991 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
992 {
993 struct ftrace_profile_stat *stat;
994 char *name;
995 int ret;
996 int cpu;
997
998 for_each_possible_cpu(cpu) {
999 stat = &per_cpu(ftrace_profile_stats, cpu);
1000
1001 name = kasprintf(GFP_KERNEL, "function%d", cpu);
1002 if (!name) {
1003 /*
1004 * The files created are permanent, if something happens
1005 * we still do not free memory.
1006 */
1007 WARN(1,
1008 "Could not allocate stat file for cpu %d\n",
1009 cpu);
1010 return;
1011 }
1012 stat->stat = function_stats;
1013 stat->stat.name = name;
1014 ret = register_stat_tracer(&stat->stat);
1015 if (ret) {
1016 WARN(1,
1017 "Could not register function stat for cpu %d\n",
1018 cpu);
1019 kfree(name);
1020 return;
1021 }
1022 }
1023
1024 trace_create_file("function_profile_enabled",
1025 TRACE_MODE_WRITE, d_tracer, NULL,
1026 &ftrace_profile_fops);
1027 }
1028
1029 #else /* CONFIG_FUNCTION_PROFILER */
ftrace_profile_tracefs(struct dentry * d_tracer)1030 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1031 {
1032 }
1033 #endif /* CONFIG_FUNCTION_PROFILER */
1034
1035 #ifdef CONFIG_DYNAMIC_FTRACE
1036
1037 static struct ftrace_ops *removed_ops;
1038
1039 /*
1040 * Set when doing a global update, like enabling all recs or disabling them.
1041 * It is not set when just updating a single ftrace_ops.
1042 */
1043 static bool update_all_ops;
1044
1045 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1046 # error Dynamic ftrace depends on MCOUNT_RECORD
1047 #endif
1048
1049 struct ftrace_func_probe {
1050 struct ftrace_probe_ops *probe_ops;
1051 struct ftrace_ops ops;
1052 struct trace_array *tr;
1053 struct list_head list;
1054 void *data;
1055 int ref;
1056 };
1057
1058 /*
1059 * We make these constant because no one should touch them,
1060 * but they are used as the default "empty hash", to avoid allocating
1061 * it all the time. These are in a read only section such that if
1062 * anyone does try to modify it, it will cause an exception.
1063 */
1064 static const struct hlist_head empty_buckets[1];
1065 static const struct ftrace_hash empty_hash = {
1066 .buckets = (struct hlist_head *)empty_buckets,
1067 };
1068 #define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
1069
1070 struct ftrace_ops global_ops = {
1071 .func = ftrace_stub,
1072 .local_hash.notrace_hash = EMPTY_HASH,
1073 .local_hash.filter_hash = EMPTY_HASH,
1074 INIT_OPS_HASH(global_ops)
1075 .flags = FTRACE_OPS_FL_INITIALIZED |
1076 FTRACE_OPS_FL_PID,
1077 };
1078
1079 /*
1080 * Used by the stack unwinder to know about dynamic ftrace trampolines.
1081 */
ftrace_ops_trampoline(unsigned long addr)1082 struct ftrace_ops *ftrace_ops_trampoline(unsigned long addr)
1083 {
1084 struct ftrace_ops *op = NULL;
1085
1086 /*
1087 * Some of the ops may be dynamically allocated,
1088 * they are freed after a synchronize_rcu().
1089 */
1090 preempt_disable_notrace();
1091
1092 do_for_each_ftrace_op(op, ftrace_ops_list) {
1093 /*
1094 * This is to check for dynamically allocated trampolines.
1095 * Trampolines that are in kernel text will have
1096 * core_kernel_text() return true.
1097 */
1098 if (op->trampoline && op->trampoline_size)
1099 if (addr >= op->trampoline &&
1100 addr < op->trampoline + op->trampoline_size) {
1101 preempt_enable_notrace();
1102 return op;
1103 }
1104 } while_for_each_ftrace_op(op);
1105 preempt_enable_notrace();
1106
1107 return NULL;
1108 }
1109
1110 /*
1111 * This is used by __kernel_text_address() to return true if the
1112 * address is on a dynamically allocated trampoline that would
1113 * not return true for either core_kernel_text() or
1114 * is_module_text_address().
1115 */
is_ftrace_trampoline(unsigned long addr)1116 bool is_ftrace_trampoline(unsigned long addr)
1117 {
1118 return ftrace_ops_trampoline(addr) != NULL;
1119 }
1120
1121 struct ftrace_page {
1122 struct ftrace_page *next;
1123 struct dyn_ftrace *records;
1124 int index;
1125 int order;
1126 };
1127
1128 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1129 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1130
1131 static struct ftrace_page *ftrace_pages_start;
1132 static struct ftrace_page *ftrace_pages;
1133
1134 static __always_inline unsigned long
ftrace_hash_key(struct ftrace_hash * hash,unsigned long ip)1135 ftrace_hash_key(struct ftrace_hash *hash, unsigned long ip)
1136 {
1137 if (hash->size_bits > 0)
1138 return hash_long(ip, hash->size_bits);
1139
1140 return 0;
1141 }
1142
1143 /* Only use this function if ftrace_hash_empty() has already been tested */
1144 static __always_inline struct ftrace_func_entry *
__ftrace_lookup_ip(struct ftrace_hash * hash,unsigned long ip)1145 __ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1146 {
1147 unsigned long key;
1148 struct ftrace_func_entry *entry;
1149 struct hlist_head *hhd;
1150
1151 key = ftrace_hash_key(hash, ip);
1152 hhd = &hash->buckets[key];
1153
1154 hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1155 if (entry->ip == ip)
1156 return entry;
1157 }
1158 return NULL;
1159 }
1160
1161 /**
1162 * ftrace_lookup_ip - Test to see if an ip exists in an ftrace_hash
1163 * @hash: The hash to look at
1164 * @ip: The instruction pointer to test
1165 *
1166 * Search a given @hash to see if a given instruction pointer (@ip)
1167 * exists in it.
1168 *
1169 * Returns: the entry that holds the @ip if found. NULL otherwise.
1170 */
1171 struct ftrace_func_entry *
ftrace_lookup_ip(struct ftrace_hash * hash,unsigned long ip)1172 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1173 {
1174 if (ftrace_hash_empty(hash))
1175 return NULL;
1176
1177 return __ftrace_lookup_ip(hash, ip);
1178 }
1179
__add_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1180 static void __add_hash_entry(struct ftrace_hash *hash,
1181 struct ftrace_func_entry *entry)
1182 {
1183 struct hlist_head *hhd;
1184 unsigned long key;
1185
1186 key = ftrace_hash_key(hash, entry->ip);
1187 hhd = &hash->buckets[key];
1188 hlist_add_head(&entry->hlist, hhd);
1189 hash->count++;
1190 }
1191
1192 static struct ftrace_func_entry *
add_hash_entry(struct ftrace_hash * hash,unsigned long ip)1193 add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1194 {
1195 struct ftrace_func_entry *entry;
1196
1197 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1198 if (!entry)
1199 return NULL;
1200
1201 entry->ip = ip;
1202 __add_hash_entry(hash, entry);
1203
1204 return entry;
1205 }
1206
1207 static void
free_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1208 free_hash_entry(struct ftrace_hash *hash,
1209 struct ftrace_func_entry *entry)
1210 {
1211 hlist_del(&entry->hlist);
1212 kfree(entry);
1213 hash->count--;
1214 }
1215
1216 static void
remove_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1217 remove_hash_entry(struct ftrace_hash *hash,
1218 struct ftrace_func_entry *entry)
1219 {
1220 hlist_del_rcu(&entry->hlist);
1221 hash->count--;
1222 }
1223
ftrace_hash_clear(struct ftrace_hash * hash)1224 static void ftrace_hash_clear(struct ftrace_hash *hash)
1225 {
1226 struct hlist_head *hhd;
1227 struct hlist_node *tn;
1228 struct ftrace_func_entry *entry;
1229 int size = 1 << hash->size_bits;
1230 int i;
1231
1232 if (!hash->count)
1233 return;
1234
1235 for (i = 0; i < size; i++) {
1236 hhd = &hash->buckets[i];
1237 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1238 free_hash_entry(hash, entry);
1239 }
1240 FTRACE_WARN_ON(hash->count);
1241 }
1242
free_ftrace_mod(struct ftrace_mod_load * ftrace_mod)1243 static void free_ftrace_mod(struct ftrace_mod_load *ftrace_mod)
1244 {
1245 list_del(&ftrace_mod->list);
1246 kfree(ftrace_mod->module);
1247 kfree(ftrace_mod->func);
1248 kfree(ftrace_mod);
1249 }
1250
clear_ftrace_mod_list(struct list_head * head)1251 static void clear_ftrace_mod_list(struct list_head *head)
1252 {
1253 struct ftrace_mod_load *p, *n;
1254
1255 /* stack tracer isn't supported yet */
1256 if (!head)
1257 return;
1258
1259 mutex_lock(&ftrace_lock);
1260 list_for_each_entry_safe(p, n, head, list)
1261 free_ftrace_mod(p);
1262 mutex_unlock(&ftrace_lock);
1263 }
1264
free_ftrace_hash(struct ftrace_hash * hash)1265 static void free_ftrace_hash(struct ftrace_hash *hash)
1266 {
1267 if (!hash || hash == EMPTY_HASH)
1268 return;
1269 ftrace_hash_clear(hash);
1270 kfree(hash->buckets);
1271 kfree(hash);
1272 }
1273
__free_ftrace_hash_rcu(struct rcu_head * rcu)1274 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1275 {
1276 struct ftrace_hash *hash;
1277
1278 hash = container_of(rcu, struct ftrace_hash, rcu);
1279 free_ftrace_hash(hash);
1280 }
1281
free_ftrace_hash_rcu(struct ftrace_hash * hash)1282 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1283 {
1284 if (!hash || hash == EMPTY_HASH)
1285 return;
1286 call_rcu(&hash->rcu, __free_ftrace_hash_rcu);
1287 }
1288
1289 /**
1290 * ftrace_free_filter - remove all filters for an ftrace_ops
1291 * @ops: the ops to remove the filters from
1292 */
ftrace_free_filter(struct ftrace_ops * ops)1293 void ftrace_free_filter(struct ftrace_ops *ops)
1294 {
1295 ftrace_ops_init(ops);
1296 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
1297 return;
1298 free_ftrace_hash(ops->func_hash->filter_hash);
1299 free_ftrace_hash(ops->func_hash->notrace_hash);
1300 ops->func_hash->filter_hash = EMPTY_HASH;
1301 ops->func_hash->notrace_hash = EMPTY_HASH;
1302 }
1303 EXPORT_SYMBOL_GPL(ftrace_free_filter);
1304
alloc_ftrace_hash(int size_bits)1305 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1306 {
1307 struct ftrace_hash *hash;
1308 int size;
1309
1310 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1311 if (!hash)
1312 return NULL;
1313
1314 size = 1 << size_bits;
1315 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1316
1317 if (!hash->buckets) {
1318 kfree(hash);
1319 return NULL;
1320 }
1321
1322 hash->size_bits = size_bits;
1323
1324 return hash;
1325 }
1326
1327 /* Used to save filters on functions for modules not loaded yet */
ftrace_add_mod(struct trace_array * tr,const char * func,const char * module,int enable)1328 static int ftrace_add_mod(struct trace_array *tr,
1329 const char *func, const char *module,
1330 int enable)
1331 {
1332 struct ftrace_mod_load *ftrace_mod;
1333 struct list_head *mod_head = enable ? &tr->mod_trace : &tr->mod_notrace;
1334
1335 ftrace_mod = kzalloc(sizeof(*ftrace_mod), GFP_KERNEL);
1336 if (!ftrace_mod)
1337 return -ENOMEM;
1338
1339 INIT_LIST_HEAD(&ftrace_mod->list);
1340 ftrace_mod->func = kstrdup(func, GFP_KERNEL);
1341 ftrace_mod->module = kstrdup(module, GFP_KERNEL);
1342 ftrace_mod->enable = enable;
1343
1344 if (!ftrace_mod->func || !ftrace_mod->module)
1345 goto out_free;
1346
1347 list_add(&ftrace_mod->list, mod_head);
1348
1349 return 0;
1350
1351 out_free:
1352 free_ftrace_mod(ftrace_mod);
1353
1354 return -ENOMEM;
1355 }
1356
1357 static struct ftrace_hash *
alloc_and_copy_ftrace_hash(int size_bits,struct ftrace_hash * hash)1358 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1359 {
1360 struct ftrace_func_entry *entry;
1361 struct ftrace_hash *new_hash;
1362 int size;
1363 int i;
1364
1365 new_hash = alloc_ftrace_hash(size_bits);
1366 if (!new_hash)
1367 return NULL;
1368
1369 if (hash)
1370 new_hash->flags = hash->flags;
1371
1372 /* Empty hash? */
1373 if (ftrace_hash_empty(hash))
1374 return new_hash;
1375
1376 size = 1 << hash->size_bits;
1377 for (i = 0; i < size; i++) {
1378 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1379 if (add_hash_entry(new_hash, entry->ip) == NULL)
1380 goto free_hash;
1381 }
1382 }
1383
1384 FTRACE_WARN_ON(new_hash->count != hash->count);
1385
1386 return new_hash;
1387
1388 free_hash:
1389 free_ftrace_hash(new_hash);
1390 return NULL;
1391 }
1392
1393 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops);
1394 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops);
1395
1396 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1397 struct ftrace_hash *new_hash);
1398
1399 /*
1400 * Allocate a new hash and remove entries from @src and move them to the new hash.
1401 * On success, the @src hash will be empty and should be freed.
1402 */
__move_hash(struct ftrace_hash * src,int size)1403 static struct ftrace_hash *__move_hash(struct ftrace_hash *src, int size)
1404 {
1405 struct ftrace_func_entry *entry;
1406 struct ftrace_hash *new_hash;
1407 struct hlist_head *hhd;
1408 struct hlist_node *tn;
1409 int bits = 0;
1410 int i;
1411
1412 /*
1413 * Use around half the size (max bit of it), but
1414 * a minimum of 2 is fine (as size of 0 or 1 both give 1 for bits).
1415 */
1416 bits = fls(size / 2);
1417
1418 /* Don't allocate too much */
1419 if (bits > FTRACE_HASH_MAX_BITS)
1420 bits = FTRACE_HASH_MAX_BITS;
1421
1422 new_hash = alloc_ftrace_hash(bits);
1423 if (!new_hash)
1424 return NULL;
1425
1426 new_hash->flags = src->flags;
1427
1428 size = 1 << src->size_bits;
1429 for (i = 0; i < size; i++) {
1430 hhd = &src->buckets[i];
1431 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1432 remove_hash_entry(src, entry);
1433 __add_hash_entry(new_hash, entry);
1434 }
1435 }
1436 return new_hash;
1437 }
1438
1439 /* Move the @src entries to a newly allocated hash */
1440 static struct ftrace_hash *
__ftrace_hash_move(struct ftrace_hash * src)1441 __ftrace_hash_move(struct ftrace_hash *src)
1442 {
1443 int size = src->count;
1444
1445 /*
1446 * If the new source is empty, just return the empty_hash.
1447 */
1448 if (ftrace_hash_empty(src))
1449 return EMPTY_HASH;
1450
1451 return __move_hash(src, size);
1452 }
1453
1454 /**
1455 * ftrace_hash_move - move a new hash to a filter and do updates
1456 * @ops: The ops with the hash that @dst points to
1457 * @enable: True if for the filter hash, false for the notrace hash
1458 * @dst: Points to the @ops hash that should be updated
1459 * @src: The hash to update @dst with
1460 *
1461 * This is called when an ftrace_ops hash is being updated and the
1462 * the kernel needs to reflect this. Note, this only updates the kernel
1463 * function callbacks if the @ops is enabled (not to be confused with
1464 * @enable above). If the @ops is enabled, its hash determines what
1465 * callbacks get called. This function gets called when the @ops hash
1466 * is updated and it requires new callbacks.
1467 *
1468 * On success the elements of @src is moved to @dst, and @dst is updated
1469 * properly, as well as the functions determined by the @ops hashes
1470 * are now calling the @ops callback function.
1471 *
1472 * Regardless of return type, @src should be freed with free_ftrace_hash().
1473 */
1474 static int
ftrace_hash_move(struct ftrace_ops * ops,int enable,struct ftrace_hash ** dst,struct ftrace_hash * src)1475 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1476 struct ftrace_hash **dst, struct ftrace_hash *src)
1477 {
1478 struct ftrace_hash *new_hash;
1479 int ret;
1480
1481 /* Reject setting notrace hash on IPMODIFY ftrace_ops */
1482 if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1483 return -EINVAL;
1484
1485 new_hash = __ftrace_hash_move(src);
1486 if (!new_hash)
1487 return -ENOMEM;
1488
1489 /* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1490 if (enable) {
1491 /* IPMODIFY should be updated only when filter_hash updating */
1492 ret = ftrace_hash_ipmodify_update(ops, new_hash);
1493 if (ret < 0) {
1494 free_ftrace_hash(new_hash);
1495 return ret;
1496 }
1497 }
1498
1499 /*
1500 * Remove the current set, update the hash and add
1501 * them back.
1502 */
1503 ftrace_hash_rec_disable_modify(ops);
1504
1505 rcu_assign_pointer(*dst, new_hash);
1506
1507 ftrace_hash_rec_enable_modify(ops);
1508
1509 return 0;
1510 }
1511
hash_contains_ip(unsigned long ip,struct ftrace_ops_hash * hash)1512 static bool hash_contains_ip(unsigned long ip,
1513 struct ftrace_ops_hash *hash)
1514 {
1515 /*
1516 * The function record is a match if it exists in the filter
1517 * hash and not in the notrace hash. Note, an empty hash is
1518 * considered a match for the filter hash, but an empty
1519 * notrace hash is considered not in the notrace hash.
1520 */
1521 return (ftrace_hash_empty(hash->filter_hash) ||
1522 __ftrace_lookup_ip(hash->filter_hash, ip)) &&
1523 (ftrace_hash_empty(hash->notrace_hash) ||
1524 !__ftrace_lookup_ip(hash->notrace_hash, ip));
1525 }
1526
1527 /*
1528 * Test the hashes for this ops to see if we want to call
1529 * the ops->func or not.
1530 *
1531 * It's a match if the ip is in the ops->filter_hash or
1532 * the filter_hash does not exist or is empty,
1533 * AND
1534 * the ip is not in the ops->notrace_hash.
1535 *
1536 * This needs to be called with preemption disabled as
1537 * the hashes are freed with call_rcu().
1538 */
1539 int
ftrace_ops_test(struct ftrace_ops * ops,unsigned long ip,void * regs)1540 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1541 {
1542 struct ftrace_ops_hash hash;
1543 int ret;
1544
1545 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1546 /*
1547 * There's a small race when adding ops that the ftrace handler
1548 * that wants regs, may be called without them. We can not
1549 * allow that handler to be called if regs is NULL.
1550 */
1551 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1552 return 0;
1553 #endif
1554
1555 rcu_assign_pointer(hash.filter_hash, ops->func_hash->filter_hash);
1556 rcu_assign_pointer(hash.notrace_hash, ops->func_hash->notrace_hash);
1557
1558 if (hash_contains_ip(ip, &hash))
1559 ret = 1;
1560 else
1561 ret = 0;
1562
1563 return ret;
1564 }
1565
1566 /*
1567 * This is a double for. Do not use 'break' to break out of the loop,
1568 * you must use a goto.
1569 */
1570 #define do_for_each_ftrace_rec(pg, rec) \
1571 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1572 int _____i; \
1573 for (_____i = 0; _____i < pg->index; _____i++) { \
1574 rec = &pg->records[_____i];
1575
1576 #define while_for_each_ftrace_rec() \
1577 } \
1578 }
1579
1580
ftrace_cmp_recs(const void * a,const void * b)1581 static int ftrace_cmp_recs(const void *a, const void *b)
1582 {
1583 const struct dyn_ftrace *key = a;
1584 const struct dyn_ftrace *rec = b;
1585
1586 if (key->flags < rec->ip)
1587 return -1;
1588 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1589 return 1;
1590 return 0;
1591 }
1592
lookup_rec(unsigned long start,unsigned long end)1593 static struct dyn_ftrace *lookup_rec(unsigned long start, unsigned long end)
1594 {
1595 struct ftrace_page *pg;
1596 struct dyn_ftrace *rec = NULL;
1597 struct dyn_ftrace key;
1598
1599 key.ip = start;
1600 key.flags = end; /* overload flags, as it is unsigned long */
1601
1602 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1603 if (pg->index == 0 ||
1604 end < pg->records[0].ip ||
1605 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1606 continue;
1607 rec = bsearch(&key, pg->records, pg->index,
1608 sizeof(struct dyn_ftrace),
1609 ftrace_cmp_recs);
1610 if (rec)
1611 break;
1612 }
1613 return rec;
1614 }
1615
1616 /**
1617 * ftrace_location_range - return the first address of a traced location
1618 * if it touches the given ip range
1619 * @start: start of range to search.
1620 * @end: end of range to search (inclusive). @end points to the last byte
1621 * to check.
1622 *
1623 * Returns: rec->ip if the related ftrace location is a least partly within
1624 * the given address range. That is, the first address of the instruction
1625 * that is either a NOP or call to the function tracer. It checks the ftrace
1626 * internal tables to determine if the address belongs or not.
1627 */
ftrace_location_range(unsigned long start,unsigned long end)1628 unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1629 {
1630 struct dyn_ftrace *rec;
1631 unsigned long ip = 0;
1632
1633 rcu_read_lock();
1634 rec = lookup_rec(start, end);
1635 if (rec)
1636 ip = rec->ip;
1637 rcu_read_unlock();
1638
1639 return ip;
1640 }
1641
1642 /**
1643 * ftrace_location - return the ftrace location
1644 * @ip: the instruction pointer to check
1645 *
1646 * Returns:
1647 * * If @ip matches the ftrace location, return @ip.
1648 * * If @ip matches sym+0, return sym's ftrace location.
1649 * * Otherwise, return 0.
1650 */
ftrace_location(unsigned long ip)1651 unsigned long ftrace_location(unsigned long ip)
1652 {
1653 unsigned long loc;
1654 unsigned long offset;
1655 unsigned long size;
1656
1657 loc = ftrace_location_range(ip, ip);
1658 if (!loc) {
1659 if (!kallsyms_lookup_size_offset(ip, &size, &offset))
1660 return 0;
1661
1662 /* map sym+0 to __fentry__ */
1663 if (!offset)
1664 loc = ftrace_location_range(ip, ip + size - 1);
1665 }
1666 return loc;
1667 }
1668
1669 /**
1670 * ftrace_text_reserved - return true if range contains an ftrace location
1671 * @start: start of range to search
1672 * @end: end of range to search (inclusive). @end points to the last byte to check.
1673 *
1674 * Returns: 1 if @start and @end contains a ftrace location.
1675 * That is, the instruction that is either a NOP or call to
1676 * the function tracer. It checks the ftrace internal tables to
1677 * determine if the address belongs or not.
1678 */
ftrace_text_reserved(const void * start,const void * end)1679 int ftrace_text_reserved(const void *start, const void *end)
1680 {
1681 unsigned long ret;
1682
1683 ret = ftrace_location_range((unsigned long)start,
1684 (unsigned long)end);
1685
1686 return (int)!!ret;
1687 }
1688
1689 /* Test if ops registered to this rec needs regs */
test_rec_ops_needs_regs(struct dyn_ftrace * rec)1690 static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1691 {
1692 struct ftrace_ops *ops;
1693 bool keep_regs = false;
1694
1695 for (ops = ftrace_ops_list;
1696 ops != &ftrace_list_end; ops = ops->next) {
1697 /* pass rec in as regs to have non-NULL val */
1698 if (ftrace_ops_test(ops, rec->ip, rec)) {
1699 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1700 keep_regs = true;
1701 break;
1702 }
1703 }
1704 }
1705
1706 return keep_regs;
1707 }
1708
1709 static struct ftrace_ops *
1710 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
1711 static struct ftrace_ops *
1712 ftrace_find_tramp_ops_any_other(struct dyn_ftrace *rec, struct ftrace_ops *op_exclude);
1713 static struct ftrace_ops *
1714 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
1715
skip_record(struct dyn_ftrace * rec)1716 static bool skip_record(struct dyn_ftrace *rec)
1717 {
1718 /*
1719 * At boot up, weak functions are set to disable. Function tracing
1720 * can be enabled before they are, and they still need to be disabled now.
1721 * If the record is disabled, still continue if it is marked as already
1722 * enabled (this is needed to keep the accounting working).
1723 */
1724 return rec->flags & FTRACE_FL_DISABLED &&
1725 !(rec->flags & FTRACE_FL_ENABLED);
1726 }
1727
1728 /*
1729 * This is the main engine to the ftrace updates to the dyn_ftrace records.
1730 *
1731 * It will iterate through all the available ftrace functions
1732 * (the ones that ftrace can have callbacks to) and set the flags
1733 * in the associated dyn_ftrace records.
1734 *
1735 * @inc: If true, the functions associated to @ops are added to
1736 * the dyn_ftrace records, otherwise they are removed.
1737 */
__ftrace_hash_rec_update(struct ftrace_ops * ops,bool inc)1738 static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1739 bool inc)
1740 {
1741 struct ftrace_hash *hash;
1742 struct ftrace_hash *notrace_hash;
1743 struct ftrace_page *pg;
1744 struct dyn_ftrace *rec;
1745 bool update = false;
1746 int count = 0;
1747 int all = false;
1748
1749 /* Only update if the ops has been registered */
1750 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1751 return false;
1752
1753 /*
1754 * If the count is zero, we update all records.
1755 * Otherwise we just update the items in the hash.
1756 */
1757 hash = ops->func_hash->filter_hash;
1758 notrace_hash = ops->func_hash->notrace_hash;
1759 if (ftrace_hash_empty(hash))
1760 all = true;
1761
1762 do_for_each_ftrace_rec(pg, rec) {
1763 int in_notrace_hash = 0;
1764 int in_hash = 0;
1765 int match = 0;
1766
1767 if (skip_record(rec))
1768 continue;
1769
1770 if (all) {
1771 /*
1772 * Only the filter_hash affects all records.
1773 * Update if the record is not in the notrace hash.
1774 */
1775 if (!notrace_hash || !ftrace_lookup_ip(notrace_hash, rec->ip))
1776 match = 1;
1777 } else {
1778 in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1779 in_notrace_hash = !!ftrace_lookup_ip(notrace_hash, rec->ip);
1780
1781 /*
1782 * We want to match all functions that are in the hash but
1783 * not in the other hash.
1784 */
1785 if (in_hash && !in_notrace_hash)
1786 match = 1;
1787 }
1788 if (!match)
1789 continue;
1790
1791 if (inc) {
1792 rec->flags++;
1793 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1794 return false;
1795
1796 if (ops->flags & FTRACE_OPS_FL_DIRECT)
1797 rec->flags |= FTRACE_FL_DIRECT;
1798
1799 /*
1800 * If there's only a single callback registered to a
1801 * function, and the ops has a trampoline registered
1802 * for it, then we can call it directly.
1803 */
1804 if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1805 rec->flags |= FTRACE_FL_TRAMP;
1806 else
1807 /*
1808 * If we are adding another function callback
1809 * to this function, and the previous had a
1810 * custom trampoline in use, then we need to go
1811 * back to the default trampoline.
1812 */
1813 rec->flags &= ~FTRACE_FL_TRAMP;
1814
1815 /*
1816 * If any ops wants regs saved for this function
1817 * then all ops will get saved regs.
1818 */
1819 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1820 rec->flags |= FTRACE_FL_REGS;
1821 } else {
1822 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1823 return false;
1824 rec->flags--;
1825
1826 /*
1827 * Only the internal direct_ops should have the
1828 * DIRECT flag set. Thus, if it is removing a
1829 * function, then that function should no longer
1830 * be direct.
1831 */
1832 if (ops->flags & FTRACE_OPS_FL_DIRECT)
1833 rec->flags &= ~FTRACE_FL_DIRECT;
1834
1835 /*
1836 * If the rec had REGS enabled and the ops that is
1837 * being removed had REGS set, then see if there is
1838 * still any ops for this record that wants regs.
1839 * If not, we can stop recording them.
1840 */
1841 if (ftrace_rec_count(rec) > 0 &&
1842 rec->flags & FTRACE_FL_REGS &&
1843 ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1844 if (!test_rec_ops_needs_regs(rec))
1845 rec->flags &= ~FTRACE_FL_REGS;
1846 }
1847
1848 /*
1849 * The TRAMP needs to be set only if rec count
1850 * is decremented to one, and the ops that is
1851 * left has a trampoline. As TRAMP can only be
1852 * enabled if there is only a single ops attached
1853 * to it.
1854 */
1855 if (ftrace_rec_count(rec) == 1 &&
1856 ftrace_find_tramp_ops_any_other(rec, ops))
1857 rec->flags |= FTRACE_FL_TRAMP;
1858 else
1859 rec->flags &= ~FTRACE_FL_TRAMP;
1860
1861 /*
1862 * flags will be cleared in ftrace_check_record()
1863 * if rec count is zero.
1864 */
1865 }
1866
1867 /*
1868 * If the rec has a single associated ops, and ops->func can be
1869 * called directly, allow the call site to call via the ops.
1870 */
1871 if (IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_CALL_OPS) &&
1872 ftrace_rec_count(rec) == 1 &&
1873 ftrace_ops_get_func(ops) == ops->func)
1874 rec->flags |= FTRACE_FL_CALL_OPS;
1875 else
1876 rec->flags &= ~FTRACE_FL_CALL_OPS;
1877
1878 count++;
1879
1880 /* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1881 update |= ftrace_test_record(rec, true) != FTRACE_UPDATE_IGNORE;
1882
1883 /* Shortcut, if we handled all records, we are done. */
1884 if (!all && count == hash->count)
1885 return update;
1886 } while_for_each_ftrace_rec();
1887
1888 return update;
1889 }
1890
1891 /*
1892 * This is called when an ops is removed from tracing. It will decrement
1893 * the counters of the dyn_ftrace records for all the functions that
1894 * the @ops attached to.
1895 */
ftrace_hash_rec_disable(struct ftrace_ops * ops)1896 static bool ftrace_hash_rec_disable(struct ftrace_ops *ops)
1897 {
1898 return __ftrace_hash_rec_update(ops, false);
1899 }
1900
1901 /*
1902 * This is called when an ops is added to tracing. It will increment
1903 * the counters of the dyn_ftrace records for all the functions that
1904 * the @ops attached to.
1905 */
ftrace_hash_rec_enable(struct ftrace_ops * ops)1906 static bool ftrace_hash_rec_enable(struct ftrace_ops *ops)
1907 {
1908 return __ftrace_hash_rec_update(ops, true);
1909 }
1910
1911 /*
1912 * This function will update what functions @ops traces when its filter
1913 * changes.
1914 *
1915 * The @inc states if the @ops callbacks are going to be added or removed.
1916 * When one of the @ops hashes are updated to a "new_hash" the dyn_ftrace
1917 * records are update via:
1918 *
1919 * ftrace_hash_rec_disable_modify(ops);
1920 * ops->hash = new_hash
1921 * ftrace_hash_rec_enable_modify(ops);
1922 *
1923 * Where the @ops is removed from all the records it is tracing using
1924 * its old hash. The @ops hash is updated to the new hash, and then
1925 * the @ops is added back to the records so that it is tracing all
1926 * the new functions.
1927 */
ftrace_hash_rec_update_modify(struct ftrace_ops * ops,bool inc)1928 static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops, bool inc)
1929 {
1930 struct ftrace_ops *op;
1931
1932 __ftrace_hash_rec_update(ops, inc);
1933
1934 if (ops->func_hash != &global_ops.local_hash)
1935 return;
1936
1937 /*
1938 * If the ops shares the global_ops hash, then we need to update
1939 * all ops that are enabled and use this hash.
1940 */
1941 do_for_each_ftrace_op(op, ftrace_ops_list) {
1942 /* Already done */
1943 if (op == ops)
1944 continue;
1945 if (op->func_hash == &global_ops.local_hash)
1946 __ftrace_hash_rec_update(op, inc);
1947 } while_for_each_ftrace_op(op);
1948 }
1949
ftrace_hash_rec_disable_modify(struct ftrace_ops * ops)1950 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops)
1951 {
1952 ftrace_hash_rec_update_modify(ops, false);
1953 }
1954
ftrace_hash_rec_enable_modify(struct ftrace_ops * ops)1955 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops)
1956 {
1957 ftrace_hash_rec_update_modify(ops, true);
1958 }
1959
1960 /*
1961 * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1962 * or no-needed to update, -EBUSY if it detects a conflict of the flag
1963 * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1964 * Note that old_hash and new_hash has below meanings
1965 * - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1966 * - If the hash is EMPTY_HASH, it hits nothing
1967 * - Anything else hits the recs which match the hash entries.
1968 *
1969 * DIRECT ops does not have IPMODIFY flag, but we still need to check it
1970 * against functions with FTRACE_FL_IPMODIFY. If there is any overlap, call
1971 * ops_func(SHARE_IPMODIFY_SELF) to make sure current ops can share with
1972 * IPMODIFY. If ops_func(SHARE_IPMODIFY_SELF) returns non-zero, propagate
1973 * the return value to the caller and eventually to the owner of the DIRECT
1974 * ops.
1975 */
__ftrace_hash_update_ipmodify(struct ftrace_ops * ops,struct ftrace_hash * old_hash,struct ftrace_hash * new_hash)1976 static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1977 struct ftrace_hash *old_hash,
1978 struct ftrace_hash *new_hash)
1979 {
1980 struct ftrace_page *pg;
1981 struct dyn_ftrace *rec, *end = NULL;
1982 int in_old, in_new;
1983 bool is_ipmodify, is_direct;
1984
1985 /* Only update if the ops has been registered */
1986 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1987 return 0;
1988
1989 is_ipmodify = ops->flags & FTRACE_OPS_FL_IPMODIFY;
1990 is_direct = ops->flags & FTRACE_OPS_FL_DIRECT;
1991
1992 /* neither IPMODIFY nor DIRECT, skip */
1993 if (!is_ipmodify && !is_direct)
1994 return 0;
1995
1996 if (WARN_ON_ONCE(is_ipmodify && is_direct))
1997 return 0;
1998
1999 /*
2000 * Since the IPMODIFY and DIRECT are very address sensitive
2001 * actions, we do not allow ftrace_ops to set all functions to new
2002 * hash.
2003 */
2004 if (!new_hash || !old_hash)
2005 return -EINVAL;
2006
2007 /* Update rec->flags */
2008 do_for_each_ftrace_rec(pg, rec) {
2009
2010 if (rec->flags & FTRACE_FL_DISABLED)
2011 continue;
2012
2013 /* We need to update only differences of filter_hash */
2014 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
2015 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
2016 if (in_old == in_new)
2017 continue;
2018
2019 if (in_new) {
2020 if (rec->flags & FTRACE_FL_IPMODIFY) {
2021 int ret;
2022
2023 /* Cannot have two ipmodify on same rec */
2024 if (is_ipmodify)
2025 goto rollback;
2026
2027 FTRACE_WARN_ON(rec->flags & FTRACE_FL_DIRECT);
2028
2029 /*
2030 * Another ops with IPMODIFY is already
2031 * attached. We are now attaching a direct
2032 * ops. Run SHARE_IPMODIFY_SELF, to check
2033 * whether sharing is supported.
2034 */
2035 if (!ops->ops_func)
2036 return -EBUSY;
2037 ret = ops->ops_func(ops, FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_SELF);
2038 if (ret)
2039 return ret;
2040 } else if (is_ipmodify) {
2041 rec->flags |= FTRACE_FL_IPMODIFY;
2042 }
2043 } else if (is_ipmodify) {
2044 rec->flags &= ~FTRACE_FL_IPMODIFY;
2045 }
2046 } while_for_each_ftrace_rec();
2047
2048 return 0;
2049
2050 rollback:
2051 end = rec;
2052
2053 /* Roll back what we did above */
2054 do_for_each_ftrace_rec(pg, rec) {
2055
2056 if (rec->flags & FTRACE_FL_DISABLED)
2057 continue;
2058
2059 if (rec == end)
2060 return -EBUSY;
2061
2062 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
2063 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
2064 if (in_old == in_new)
2065 continue;
2066
2067 if (in_new)
2068 rec->flags &= ~FTRACE_FL_IPMODIFY;
2069 else
2070 rec->flags |= FTRACE_FL_IPMODIFY;
2071 } while_for_each_ftrace_rec();
2072
2073 return -EBUSY;
2074 }
2075
ftrace_hash_ipmodify_enable(struct ftrace_ops * ops)2076 static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
2077 {
2078 struct ftrace_hash *hash = ops->func_hash->filter_hash;
2079
2080 if (ftrace_hash_empty(hash))
2081 hash = NULL;
2082
2083 return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
2084 }
2085
2086 /* Disabling always succeeds */
ftrace_hash_ipmodify_disable(struct ftrace_ops * ops)2087 static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
2088 {
2089 struct ftrace_hash *hash = ops->func_hash->filter_hash;
2090
2091 if (ftrace_hash_empty(hash))
2092 hash = NULL;
2093
2094 __ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
2095 }
2096
ftrace_hash_ipmodify_update(struct ftrace_ops * ops,struct ftrace_hash * new_hash)2097 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
2098 struct ftrace_hash *new_hash)
2099 {
2100 struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
2101
2102 if (ftrace_hash_empty(old_hash))
2103 old_hash = NULL;
2104
2105 if (ftrace_hash_empty(new_hash))
2106 new_hash = NULL;
2107
2108 return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
2109 }
2110
print_ip_ins(const char * fmt,const unsigned char * p)2111 static void print_ip_ins(const char *fmt, const unsigned char *p)
2112 {
2113 char ins[MCOUNT_INSN_SIZE];
2114
2115 if (copy_from_kernel_nofault(ins, p, MCOUNT_INSN_SIZE)) {
2116 printk(KERN_CONT "%s[FAULT] %px\n", fmt, p);
2117 return;
2118 }
2119
2120 printk(KERN_CONT "%s", fmt);
2121 pr_cont("%*phC", MCOUNT_INSN_SIZE, ins);
2122 }
2123
2124 enum ftrace_bug_type ftrace_bug_type;
2125 const void *ftrace_expected;
2126
print_bug_type(void)2127 static void print_bug_type(void)
2128 {
2129 switch (ftrace_bug_type) {
2130 case FTRACE_BUG_UNKNOWN:
2131 break;
2132 case FTRACE_BUG_INIT:
2133 pr_info("Initializing ftrace call sites\n");
2134 break;
2135 case FTRACE_BUG_NOP:
2136 pr_info("Setting ftrace call site to NOP\n");
2137 break;
2138 case FTRACE_BUG_CALL:
2139 pr_info("Setting ftrace call site to call ftrace function\n");
2140 break;
2141 case FTRACE_BUG_UPDATE:
2142 pr_info("Updating ftrace call site to call a different ftrace function\n");
2143 break;
2144 }
2145 }
2146
2147 /**
2148 * ftrace_bug - report and shutdown function tracer
2149 * @failed: The failed type (EFAULT, EINVAL, EPERM)
2150 * @rec: The record that failed
2151 *
2152 * The arch code that enables or disables the function tracing
2153 * can call ftrace_bug() when it has detected a problem in
2154 * modifying the code. @failed should be one of either:
2155 * EFAULT - if the problem happens on reading the @ip address
2156 * EINVAL - if what is read at @ip is not what was expected
2157 * EPERM - if the problem happens on writing to the @ip address
2158 */
ftrace_bug(int failed,struct dyn_ftrace * rec)2159 void ftrace_bug(int failed, struct dyn_ftrace *rec)
2160 {
2161 unsigned long ip = rec ? rec->ip : 0;
2162
2163 pr_info("------------[ ftrace bug ]------------\n");
2164
2165 switch (failed) {
2166 case -EFAULT:
2167 pr_info("ftrace faulted on modifying ");
2168 print_ip_sym(KERN_INFO, ip);
2169 break;
2170 case -EINVAL:
2171 pr_info("ftrace failed to modify ");
2172 print_ip_sym(KERN_INFO, ip);
2173 print_ip_ins(" actual: ", (unsigned char *)ip);
2174 pr_cont("\n");
2175 if (ftrace_expected) {
2176 print_ip_ins(" expected: ", ftrace_expected);
2177 pr_cont("\n");
2178 }
2179 break;
2180 case -EPERM:
2181 pr_info("ftrace faulted on writing ");
2182 print_ip_sym(KERN_INFO, ip);
2183 break;
2184 default:
2185 pr_info("ftrace faulted on unknown error ");
2186 print_ip_sym(KERN_INFO, ip);
2187 }
2188 print_bug_type();
2189 if (rec) {
2190 struct ftrace_ops *ops = NULL;
2191
2192 pr_info("ftrace record flags: %lx\n", rec->flags);
2193 pr_cont(" (%ld)%s%s", ftrace_rec_count(rec),
2194 rec->flags & FTRACE_FL_REGS ? " R" : " ",
2195 rec->flags & FTRACE_FL_CALL_OPS ? " O" : " ");
2196 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2197 ops = ftrace_find_tramp_ops_any(rec);
2198 if (ops) {
2199 do {
2200 pr_cont("\ttramp: %pS (%pS)",
2201 (void *)ops->trampoline,
2202 (void *)ops->func);
2203 ops = ftrace_find_tramp_ops_next(rec, ops);
2204 } while (ops);
2205 } else
2206 pr_cont("\ttramp: ERROR!");
2207
2208 }
2209 ip = ftrace_get_addr_curr(rec);
2210 pr_cont("\n expected tramp: %lx\n", ip);
2211 }
2212
2213 FTRACE_WARN_ON_ONCE(1);
2214 }
2215
ftrace_check_record(struct dyn_ftrace * rec,bool enable,bool update)2216 static int ftrace_check_record(struct dyn_ftrace *rec, bool enable, bool update)
2217 {
2218 unsigned long flag = 0UL;
2219
2220 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2221
2222 if (skip_record(rec))
2223 return FTRACE_UPDATE_IGNORE;
2224
2225 /*
2226 * If we are updating calls:
2227 *
2228 * If the record has a ref count, then we need to enable it
2229 * because someone is using it.
2230 *
2231 * Otherwise we make sure its disabled.
2232 *
2233 * If we are disabling calls, then disable all records that
2234 * are enabled.
2235 */
2236 if (enable && ftrace_rec_count(rec))
2237 flag = FTRACE_FL_ENABLED;
2238
2239 /*
2240 * If enabling and the REGS flag does not match the REGS_EN, or
2241 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2242 * this record. Set flags to fail the compare against ENABLED.
2243 * Same for direct calls.
2244 */
2245 if (flag) {
2246 if (!(rec->flags & FTRACE_FL_REGS) !=
2247 !(rec->flags & FTRACE_FL_REGS_EN))
2248 flag |= FTRACE_FL_REGS;
2249
2250 if (!(rec->flags & FTRACE_FL_TRAMP) !=
2251 !(rec->flags & FTRACE_FL_TRAMP_EN))
2252 flag |= FTRACE_FL_TRAMP;
2253
2254 /*
2255 * Direct calls are special, as count matters.
2256 * We must test the record for direct, if the
2257 * DIRECT and DIRECT_EN do not match, but only
2258 * if the count is 1. That's because, if the
2259 * count is something other than one, we do not
2260 * want the direct enabled (it will be done via the
2261 * direct helper). But if DIRECT_EN is set, and
2262 * the count is not one, we need to clear it.
2263 *
2264 */
2265 if (ftrace_rec_count(rec) == 1) {
2266 if (!(rec->flags & FTRACE_FL_DIRECT) !=
2267 !(rec->flags & FTRACE_FL_DIRECT_EN))
2268 flag |= FTRACE_FL_DIRECT;
2269 } else if (rec->flags & FTRACE_FL_DIRECT_EN) {
2270 flag |= FTRACE_FL_DIRECT;
2271 }
2272
2273 /*
2274 * Ops calls are special, as count matters.
2275 * As with direct calls, they must only be enabled when count
2276 * is one, otherwise they'll be handled via the list ops.
2277 */
2278 if (ftrace_rec_count(rec) == 1) {
2279 if (!(rec->flags & FTRACE_FL_CALL_OPS) !=
2280 !(rec->flags & FTRACE_FL_CALL_OPS_EN))
2281 flag |= FTRACE_FL_CALL_OPS;
2282 } else if (rec->flags & FTRACE_FL_CALL_OPS_EN) {
2283 flag |= FTRACE_FL_CALL_OPS;
2284 }
2285 }
2286
2287 /* If the state of this record hasn't changed, then do nothing */
2288 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2289 return FTRACE_UPDATE_IGNORE;
2290
2291 if (flag) {
2292 /* Save off if rec is being enabled (for return value) */
2293 flag ^= rec->flags & FTRACE_FL_ENABLED;
2294
2295 if (update) {
2296 rec->flags |= FTRACE_FL_ENABLED | FTRACE_FL_TOUCHED;
2297 if (flag & FTRACE_FL_REGS) {
2298 if (rec->flags & FTRACE_FL_REGS)
2299 rec->flags |= FTRACE_FL_REGS_EN;
2300 else
2301 rec->flags &= ~FTRACE_FL_REGS_EN;
2302 }
2303 if (flag & FTRACE_FL_TRAMP) {
2304 if (rec->flags & FTRACE_FL_TRAMP)
2305 rec->flags |= FTRACE_FL_TRAMP_EN;
2306 else
2307 rec->flags &= ~FTRACE_FL_TRAMP_EN;
2308 }
2309
2310 /* Keep track of anything that modifies the function */
2311 if (rec->flags & (FTRACE_FL_DIRECT | FTRACE_FL_IPMODIFY))
2312 rec->flags |= FTRACE_FL_MODIFIED;
2313
2314 if (flag & FTRACE_FL_DIRECT) {
2315 /*
2316 * If there's only one user (direct_ops helper)
2317 * then we can call the direct function
2318 * directly (no ftrace trampoline).
2319 */
2320 if (ftrace_rec_count(rec) == 1) {
2321 if (rec->flags & FTRACE_FL_DIRECT)
2322 rec->flags |= FTRACE_FL_DIRECT_EN;
2323 else
2324 rec->flags &= ~FTRACE_FL_DIRECT_EN;
2325 } else {
2326 /*
2327 * Can only call directly if there's
2328 * only one callback to the function.
2329 */
2330 rec->flags &= ~FTRACE_FL_DIRECT_EN;
2331 }
2332 }
2333
2334 if (flag & FTRACE_FL_CALL_OPS) {
2335 if (ftrace_rec_count(rec) == 1) {
2336 if (rec->flags & FTRACE_FL_CALL_OPS)
2337 rec->flags |= FTRACE_FL_CALL_OPS_EN;
2338 else
2339 rec->flags &= ~FTRACE_FL_CALL_OPS_EN;
2340 } else {
2341 /*
2342 * Can only call directly if there's
2343 * only one set of associated ops.
2344 */
2345 rec->flags &= ~FTRACE_FL_CALL_OPS_EN;
2346 }
2347 }
2348 }
2349
2350 /*
2351 * If this record is being updated from a nop, then
2352 * return UPDATE_MAKE_CALL.
2353 * Otherwise,
2354 * return UPDATE_MODIFY_CALL to tell the caller to convert
2355 * from the save regs, to a non-save regs function or
2356 * vice versa, or from a trampoline call.
2357 */
2358 if (flag & FTRACE_FL_ENABLED) {
2359 ftrace_bug_type = FTRACE_BUG_CALL;
2360 return FTRACE_UPDATE_MAKE_CALL;
2361 }
2362
2363 ftrace_bug_type = FTRACE_BUG_UPDATE;
2364 return FTRACE_UPDATE_MODIFY_CALL;
2365 }
2366
2367 if (update) {
2368 /* If there's no more users, clear all flags */
2369 if (!ftrace_rec_count(rec))
2370 rec->flags &= FTRACE_NOCLEAR_FLAGS;
2371 else
2372 /*
2373 * Just disable the record, but keep the ops TRAMP
2374 * and REGS states. The _EN flags must be disabled though.
2375 */
2376 rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2377 FTRACE_FL_REGS_EN | FTRACE_FL_DIRECT_EN |
2378 FTRACE_FL_CALL_OPS_EN);
2379 }
2380
2381 ftrace_bug_type = FTRACE_BUG_NOP;
2382 return FTRACE_UPDATE_MAKE_NOP;
2383 }
2384
2385 /**
2386 * ftrace_update_record - set a record that now is tracing or not
2387 * @rec: the record to update
2388 * @enable: set to true if the record is tracing, false to force disable
2389 *
2390 * The records that represent all functions that can be traced need
2391 * to be updated when tracing has been enabled.
2392 */
ftrace_update_record(struct dyn_ftrace * rec,bool enable)2393 int ftrace_update_record(struct dyn_ftrace *rec, bool enable)
2394 {
2395 return ftrace_check_record(rec, enable, true);
2396 }
2397
2398 /**
2399 * ftrace_test_record - check if the record has been enabled or not
2400 * @rec: the record to test
2401 * @enable: set to true to check if enabled, false if it is disabled
2402 *
2403 * The arch code may need to test if a record is already set to
2404 * tracing to determine how to modify the function code that it
2405 * represents.
2406 */
ftrace_test_record(struct dyn_ftrace * rec,bool enable)2407 int ftrace_test_record(struct dyn_ftrace *rec, bool enable)
2408 {
2409 return ftrace_check_record(rec, enable, false);
2410 }
2411
2412 static struct ftrace_ops *
ftrace_find_tramp_ops_any(struct dyn_ftrace * rec)2413 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2414 {
2415 struct ftrace_ops *op;
2416 unsigned long ip = rec->ip;
2417
2418 do_for_each_ftrace_op(op, ftrace_ops_list) {
2419
2420 if (!op->trampoline)
2421 continue;
2422
2423 if (hash_contains_ip(ip, op->func_hash))
2424 return op;
2425 } while_for_each_ftrace_op(op);
2426
2427 return NULL;
2428 }
2429
2430 static struct ftrace_ops *
ftrace_find_tramp_ops_any_other(struct dyn_ftrace * rec,struct ftrace_ops * op_exclude)2431 ftrace_find_tramp_ops_any_other(struct dyn_ftrace *rec, struct ftrace_ops *op_exclude)
2432 {
2433 struct ftrace_ops *op;
2434 unsigned long ip = rec->ip;
2435
2436 do_for_each_ftrace_op(op, ftrace_ops_list) {
2437
2438 if (op == op_exclude || !op->trampoline)
2439 continue;
2440
2441 if (hash_contains_ip(ip, op->func_hash))
2442 return op;
2443 } while_for_each_ftrace_op(op);
2444
2445 return NULL;
2446 }
2447
2448 static struct ftrace_ops *
ftrace_find_tramp_ops_next(struct dyn_ftrace * rec,struct ftrace_ops * op)2449 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2450 struct ftrace_ops *op)
2451 {
2452 unsigned long ip = rec->ip;
2453
2454 while_for_each_ftrace_op(op) {
2455
2456 if (!op->trampoline)
2457 continue;
2458
2459 if (hash_contains_ip(ip, op->func_hash))
2460 return op;
2461 }
2462
2463 return NULL;
2464 }
2465
2466 static struct ftrace_ops *
ftrace_find_tramp_ops_curr(struct dyn_ftrace * rec)2467 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2468 {
2469 struct ftrace_ops *op;
2470 unsigned long ip = rec->ip;
2471
2472 /*
2473 * Need to check removed ops first.
2474 * If they are being removed, and this rec has a tramp,
2475 * and this rec is in the ops list, then it would be the
2476 * one with the tramp.
2477 */
2478 if (removed_ops) {
2479 if (hash_contains_ip(ip, &removed_ops->old_hash))
2480 return removed_ops;
2481 }
2482
2483 /*
2484 * Need to find the current trampoline for a rec.
2485 * Now, a trampoline is only attached to a rec if there
2486 * was a single 'ops' attached to it. But this can be called
2487 * when we are adding another op to the rec or removing the
2488 * current one. Thus, if the op is being added, we can
2489 * ignore it because it hasn't attached itself to the rec
2490 * yet.
2491 *
2492 * If an ops is being modified (hooking to different functions)
2493 * then we don't care about the new functions that are being
2494 * added, just the old ones (that are probably being removed).
2495 *
2496 * If we are adding an ops to a function that already is using
2497 * a trampoline, it needs to be removed (trampolines are only
2498 * for single ops connected), then an ops that is not being
2499 * modified also needs to be checked.
2500 */
2501 do_for_each_ftrace_op(op, ftrace_ops_list) {
2502
2503 if (!op->trampoline)
2504 continue;
2505
2506 /*
2507 * If the ops is being added, it hasn't gotten to
2508 * the point to be removed from this tree yet.
2509 */
2510 if (op->flags & FTRACE_OPS_FL_ADDING)
2511 continue;
2512
2513
2514 /*
2515 * If the ops is being modified and is in the old
2516 * hash, then it is probably being removed from this
2517 * function.
2518 */
2519 if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2520 hash_contains_ip(ip, &op->old_hash))
2521 return op;
2522 /*
2523 * If the ops is not being added or modified, and it's
2524 * in its normal filter hash, then this must be the one
2525 * we want!
2526 */
2527 if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2528 hash_contains_ip(ip, op->func_hash))
2529 return op;
2530
2531 } while_for_each_ftrace_op(op);
2532
2533 return NULL;
2534 }
2535
2536 static struct ftrace_ops *
ftrace_find_tramp_ops_new(struct dyn_ftrace * rec)2537 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2538 {
2539 struct ftrace_ops *op;
2540 unsigned long ip = rec->ip;
2541
2542 do_for_each_ftrace_op(op, ftrace_ops_list) {
2543 /* pass rec in as regs to have non-NULL val */
2544 if (hash_contains_ip(ip, op->func_hash))
2545 return op;
2546 } while_for_each_ftrace_op(op);
2547
2548 return NULL;
2549 }
2550
2551 struct ftrace_ops *
ftrace_find_unique_ops(struct dyn_ftrace * rec)2552 ftrace_find_unique_ops(struct dyn_ftrace *rec)
2553 {
2554 struct ftrace_ops *op, *found = NULL;
2555 unsigned long ip = rec->ip;
2556
2557 do_for_each_ftrace_op(op, ftrace_ops_list) {
2558
2559 if (hash_contains_ip(ip, op->func_hash)) {
2560 if (found)
2561 return NULL;
2562 found = op;
2563 }
2564
2565 } while_for_each_ftrace_op(op);
2566
2567 return found;
2568 }
2569
2570 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
2571 /* Protected by rcu_tasks for reading, and direct_mutex for writing */
2572 static struct ftrace_hash __rcu *direct_functions = EMPTY_HASH;
2573 static DEFINE_MUTEX(direct_mutex);
2574
2575 /*
2576 * Search the direct_functions hash to see if the given instruction pointer
2577 * has a direct caller attached to it.
2578 */
ftrace_find_rec_direct(unsigned long ip)2579 unsigned long ftrace_find_rec_direct(unsigned long ip)
2580 {
2581 struct ftrace_func_entry *entry;
2582
2583 entry = __ftrace_lookup_ip(direct_functions, ip);
2584 if (!entry)
2585 return 0;
2586
2587 return entry->direct;
2588 }
2589
call_direct_funcs(unsigned long ip,unsigned long pip,struct ftrace_ops * ops,struct ftrace_regs * fregs)2590 static void call_direct_funcs(unsigned long ip, unsigned long pip,
2591 struct ftrace_ops *ops, struct ftrace_regs *fregs)
2592 {
2593 unsigned long addr = READ_ONCE(ops->direct_call);
2594
2595 if (!addr)
2596 return;
2597
2598 arch_ftrace_set_direct_caller(fregs, addr);
2599 }
2600 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
2601
2602 /**
2603 * ftrace_get_addr_new - Get the call address to set to
2604 * @rec: The ftrace record descriptor
2605 *
2606 * If the record has the FTRACE_FL_REGS set, that means that it
2607 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2608 * is not set, then it wants to convert to the normal callback.
2609 *
2610 * Returns: the address of the trampoline to set to
2611 */
ftrace_get_addr_new(struct dyn_ftrace * rec)2612 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2613 {
2614 struct ftrace_ops *ops;
2615 unsigned long addr;
2616
2617 if ((rec->flags & FTRACE_FL_DIRECT) &&
2618 (ftrace_rec_count(rec) == 1)) {
2619 addr = ftrace_find_rec_direct(rec->ip);
2620 if (addr)
2621 return addr;
2622 WARN_ON_ONCE(1);
2623 }
2624
2625 /* Trampolines take precedence over regs */
2626 if (rec->flags & FTRACE_FL_TRAMP) {
2627 ops = ftrace_find_tramp_ops_new(rec);
2628 if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2629 pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2630 (void *)rec->ip, (void *)rec->ip, rec->flags);
2631 /* Ftrace is shutting down, return anything */
2632 return (unsigned long)FTRACE_ADDR;
2633 }
2634 return ops->trampoline;
2635 }
2636
2637 if (rec->flags & FTRACE_FL_REGS)
2638 return (unsigned long)FTRACE_REGS_ADDR;
2639 else
2640 return (unsigned long)FTRACE_ADDR;
2641 }
2642
2643 /**
2644 * ftrace_get_addr_curr - Get the call address that is already there
2645 * @rec: The ftrace record descriptor
2646 *
2647 * The FTRACE_FL_REGS_EN is set when the record already points to
2648 * a function that saves all the regs. Basically the '_EN' version
2649 * represents the current state of the function.
2650 *
2651 * Returns: the address of the trampoline that is currently being called
2652 */
ftrace_get_addr_curr(struct dyn_ftrace * rec)2653 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2654 {
2655 struct ftrace_ops *ops;
2656 unsigned long addr;
2657
2658 /* Direct calls take precedence over trampolines */
2659 if (rec->flags & FTRACE_FL_DIRECT_EN) {
2660 addr = ftrace_find_rec_direct(rec->ip);
2661 if (addr)
2662 return addr;
2663 WARN_ON_ONCE(1);
2664 }
2665
2666 /* Trampolines take precedence over regs */
2667 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2668 ops = ftrace_find_tramp_ops_curr(rec);
2669 if (FTRACE_WARN_ON(!ops)) {
2670 pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2671 (void *)rec->ip, (void *)rec->ip);
2672 /* Ftrace is shutting down, return anything */
2673 return (unsigned long)FTRACE_ADDR;
2674 }
2675 return ops->trampoline;
2676 }
2677
2678 if (rec->flags & FTRACE_FL_REGS_EN)
2679 return (unsigned long)FTRACE_REGS_ADDR;
2680 else
2681 return (unsigned long)FTRACE_ADDR;
2682 }
2683
2684 static int
__ftrace_replace_code(struct dyn_ftrace * rec,bool enable)2685 __ftrace_replace_code(struct dyn_ftrace *rec, bool enable)
2686 {
2687 unsigned long ftrace_old_addr;
2688 unsigned long ftrace_addr;
2689 int ret;
2690
2691 ftrace_addr = ftrace_get_addr_new(rec);
2692
2693 /* This needs to be done before we call ftrace_update_record */
2694 ftrace_old_addr = ftrace_get_addr_curr(rec);
2695
2696 ret = ftrace_update_record(rec, enable);
2697
2698 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2699
2700 switch (ret) {
2701 case FTRACE_UPDATE_IGNORE:
2702 return 0;
2703
2704 case FTRACE_UPDATE_MAKE_CALL:
2705 ftrace_bug_type = FTRACE_BUG_CALL;
2706 return ftrace_make_call(rec, ftrace_addr);
2707
2708 case FTRACE_UPDATE_MAKE_NOP:
2709 ftrace_bug_type = FTRACE_BUG_NOP;
2710 return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2711
2712 case FTRACE_UPDATE_MODIFY_CALL:
2713 ftrace_bug_type = FTRACE_BUG_UPDATE;
2714 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2715 }
2716
2717 return -1; /* unknown ftrace bug */
2718 }
2719
ftrace_replace_code(int mod_flags)2720 void __weak ftrace_replace_code(int mod_flags)
2721 {
2722 struct dyn_ftrace *rec;
2723 struct ftrace_page *pg;
2724 bool enable = mod_flags & FTRACE_MODIFY_ENABLE_FL;
2725 int schedulable = mod_flags & FTRACE_MODIFY_MAY_SLEEP_FL;
2726 int failed;
2727
2728 if (unlikely(ftrace_disabled))
2729 return;
2730
2731 do_for_each_ftrace_rec(pg, rec) {
2732
2733 if (skip_record(rec))
2734 continue;
2735
2736 failed = __ftrace_replace_code(rec, enable);
2737 if (failed) {
2738 ftrace_bug(failed, rec);
2739 /* Stop processing */
2740 return;
2741 }
2742 if (schedulable)
2743 cond_resched();
2744 } while_for_each_ftrace_rec();
2745 }
2746
2747 struct ftrace_rec_iter {
2748 struct ftrace_page *pg;
2749 int index;
2750 };
2751
2752 /**
2753 * ftrace_rec_iter_start - start up iterating over traced functions
2754 *
2755 * Returns: an iterator handle that is used to iterate over all
2756 * the records that represent address locations where functions
2757 * are traced.
2758 *
2759 * May return NULL if no records are available.
2760 */
ftrace_rec_iter_start(void)2761 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2762 {
2763 /*
2764 * We only use a single iterator.
2765 * Protected by the ftrace_lock mutex.
2766 */
2767 static struct ftrace_rec_iter ftrace_rec_iter;
2768 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2769
2770 iter->pg = ftrace_pages_start;
2771 iter->index = 0;
2772
2773 /* Could have empty pages */
2774 while (iter->pg && !iter->pg->index)
2775 iter->pg = iter->pg->next;
2776
2777 if (!iter->pg)
2778 return NULL;
2779
2780 return iter;
2781 }
2782
2783 /**
2784 * ftrace_rec_iter_next - get the next record to process.
2785 * @iter: The handle to the iterator.
2786 *
2787 * Returns: the next iterator after the given iterator @iter.
2788 */
ftrace_rec_iter_next(struct ftrace_rec_iter * iter)2789 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2790 {
2791 iter->index++;
2792
2793 if (iter->index >= iter->pg->index) {
2794 iter->pg = iter->pg->next;
2795 iter->index = 0;
2796
2797 /* Could have empty pages */
2798 while (iter->pg && !iter->pg->index)
2799 iter->pg = iter->pg->next;
2800 }
2801
2802 if (!iter->pg)
2803 return NULL;
2804
2805 return iter;
2806 }
2807
2808 /**
2809 * ftrace_rec_iter_record - get the record at the iterator location
2810 * @iter: The current iterator location
2811 *
2812 * Returns: the record that the current @iter is at.
2813 */
ftrace_rec_iter_record(struct ftrace_rec_iter * iter)2814 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2815 {
2816 return &iter->pg->records[iter->index];
2817 }
2818
2819 static int
ftrace_nop_initialize(struct module * mod,struct dyn_ftrace * rec)2820 ftrace_nop_initialize(struct module *mod, struct dyn_ftrace *rec)
2821 {
2822 int ret;
2823
2824 if (unlikely(ftrace_disabled))
2825 return 0;
2826
2827 ret = ftrace_init_nop(mod, rec);
2828 if (ret) {
2829 ftrace_bug_type = FTRACE_BUG_INIT;
2830 ftrace_bug(ret, rec);
2831 return 0;
2832 }
2833 return 1;
2834 }
2835
2836 /*
2837 * archs can override this function if they must do something
2838 * before the modifying code is performed.
2839 */
ftrace_arch_code_modify_prepare(void)2840 void __weak ftrace_arch_code_modify_prepare(void)
2841 {
2842 }
2843
2844 /*
2845 * archs can override this function if they must do something
2846 * after the modifying code is performed.
2847 */
ftrace_arch_code_modify_post_process(void)2848 void __weak ftrace_arch_code_modify_post_process(void)
2849 {
2850 }
2851
update_ftrace_func(ftrace_func_t func)2852 static int update_ftrace_func(ftrace_func_t func)
2853 {
2854 static ftrace_func_t save_func;
2855
2856 /* Avoid updating if it hasn't changed */
2857 if (func == save_func)
2858 return 0;
2859
2860 save_func = func;
2861
2862 return ftrace_update_ftrace_func(func);
2863 }
2864
ftrace_modify_all_code(int command)2865 void ftrace_modify_all_code(int command)
2866 {
2867 int update = command & FTRACE_UPDATE_TRACE_FUNC;
2868 int mod_flags = 0;
2869 int err = 0;
2870
2871 if (command & FTRACE_MAY_SLEEP)
2872 mod_flags = FTRACE_MODIFY_MAY_SLEEP_FL;
2873
2874 /*
2875 * If the ftrace_caller calls a ftrace_ops func directly,
2876 * we need to make sure that it only traces functions it
2877 * expects to trace. When doing the switch of functions,
2878 * we need to update to the ftrace_ops_list_func first
2879 * before the transition between old and new calls are set,
2880 * as the ftrace_ops_list_func will check the ops hashes
2881 * to make sure the ops are having the right functions
2882 * traced.
2883 */
2884 if (update) {
2885 err = update_ftrace_func(ftrace_ops_list_func);
2886 if (FTRACE_WARN_ON(err))
2887 return;
2888 }
2889
2890 if (command & FTRACE_UPDATE_CALLS)
2891 ftrace_replace_code(mod_flags | FTRACE_MODIFY_ENABLE_FL);
2892 else if (command & FTRACE_DISABLE_CALLS)
2893 ftrace_replace_code(mod_flags);
2894
2895 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2896 function_trace_op = set_function_trace_op;
2897 smp_wmb();
2898 /* If irqs are disabled, we are in stop machine */
2899 if (!irqs_disabled())
2900 smp_call_function(ftrace_sync_ipi, NULL, 1);
2901 err = update_ftrace_func(ftrace_trace_function);
2902 if (FTRACE_WARN_ON(err))
2903 return;
2904 }
2905
2906 if (command & FTRACE_START_FUNC_RET)
2907 err = ftrace_enable_ftrace_graph_caller();
2908 else if (command & FTRACE_STOP_FUNC_RET)
2909 err = ftrace_disable_ftrace_graph_caller();
2910 FTRACE_WARN_ON(err);
2911 }
2912
__ftrace_modify_code(void * data)2913 static int __ftrace_modify_code(void *data)
2914 {
2915 int *command = data;
2916
2917 ftrace_modify_all_code(*command);
2918
2919 return 0;
2920 }
2921
2922 /**
2923 * ftrace_run_stop_machine - go back to the stop machine method
2924 * @command: The command to tell ftrace what to do
2925 *
2926 * If an arch needs to fall back to the stop machine method, the
2927 * it can call this function.
2928 */
ftrace_run_stop_machine(int command)2929 void ftrace_run_stop_machine(int command)
2930 {
2931 stop_machine(__ftrace_modify_code, &command, NULL);
2932 }
2933
2934 /**
2935 * arch_ftrace_update_code - modify the code to trace or not trace
2936 * @command: The command that needs to be done
2937 *
2938 * Archs can override this function if it does not need to
2939 * run stop_machine() to modify code.
2940 */
arch_ftrace_update_code(int command)2941 void __weak arch_ftrace_update_code(int command)
2942 {
2943 ftrace_run_stop_machine(command);
2944 }
2945
ftrace_run_update_code(int command)2946 static void ftrace_run_update_code(int command)
2947 {
2948 ftrace_arch_code_modify_prepare();
2949
2950 /*
2951 * By default we use stop_machine() to modify the code.
2952 * But archs can do what ever they want as long as it
2953 * is safe. The stop_machine() is the safest, but also
2954 * produces the most overhead.
2955 */
2956 arch_ftrace_update_code(command);
2957
2958 ftrace_arch_code_modify_post_process();
2959 }
2960
ftrace_run_modify_code(struct ftrace_ops * ops,int command,struct ftrace_ops_hash * old_hash)2961 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2962 struct ftrace_ops_hash *old_hash)
2963 {
2964 ops->flags |= FTRACE_OPS_FL_MODIFYING;
2965 ops->old_hash.filter_hash = old_hash->filter_hash;
2966 ops->old_hash.notrace_hash = old_hash->notrace_hash;
2967 ftrace_run_update_code(command);
2968 ops->old_hash.filter_hash = NULL;
2969 ops->old_hash.notrace_hash = NULL;
2970 ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2971 }
2972
2973 static ftrace_func_t saved_ftrace_func;
2974 static int ftrace_start_up;
2975
arch_ftrace_trampoline_free(struct ftrace_ops * ops)2976 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2977 {
2978 }
2979
2980 /* List of trace_ops that have allocated trampolines */
2981 static LIST_HEAD(ftrace_ops_trampoline_list);
2982
ftrace_add_trampoline_to_kallsyms(struct ftrace_ops * ops)2983 static void ftrace_add_trampoline_to_kallsyms(struct ftrace_ops *ops)
2984 {
2985 lockdep_assert_held(&ftrace_lock);
2986 list_add_rcu(&ops->list, &ftrace_ops_trampoline_list);
2987 }
2988
ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops * ops)2989 static void ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops *ops)
2990 {
2991 lockdep_assert_held(&ftrace_lock);
2992 list_del_rcu(&ops->list);
2993 synchronize_rcu();
2994 }
2995
2996 /*
2997 * "__builtin__ftrace" is used as a module name in /proc/kallsyms for symbols
2998 * for pages allocated for ftrace purposes, even though "__builtin__ftrace" is
2999 * not a module.
3000 */
3001 #define FTRACE_TRAMPOLINE_MOD "__builtin__ftrace"
3002 #define FTRACE_TRAMPOLINE_SYM "ftrace_trampoline"
3003
ftrace_trampoline_free(struct ftrace_ops * ops)3004 static void ftrace_trampoline_free(struct ftrace_ops *ops)
3005 {
3006 if (ops && (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP) &&
3007 ops->trampoline) {
3008 /*
3009 * Record the text poke event before the ksymbol unregister
3010 * event.
3011 */
3012 perf_event_text_poke((void *)ops->trampoline,
3013 (void *)ops->trampoline,
3014 ops->trampoline_size, NULL, 0);
3015 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
3016 ops->trampoline, ops->trampoline_size,
3017 true, FTRACE_TRAMPOLINE_SYM);
3018 /* Remove from kallsyms after the perf events */
3019 ftrace_remove_trampoline_from_kallsyms(ops);
3020 }
3021
3022 arch_ftrace_trampoline_free(ops);
3023 }
3024
ftrace_startup_enable(int command)3025 static void ftrace_startup_enable(int command)
3026 {
3027 if (saved_ftrace_func != ftrace_trace_function) {
3028 saved_ftrace_func = ftrace_trace_function;
3029 command |= FTRACE_UPDATE_TRACE_FUNC;
3030 }
3031
3032 if (!command || !ftrace_enabled)
3033 return;
3034
3035 ftrace_run_update_code(command);
3036 }
3037
ftrace_startup_all(int command)3038 static void ftrace_startup_all(int command)
3039 {
3040 update_all_ops = true;
3041 ftrace_startup_enable(command);
3042 update_all_ops = false;
3043 }
3044
ftrace_startup(struct ftrace_ops * ops,int command)3045 int ftrace_startup(struct ftrace_ops *ops, int command)
3046 {
3047 int ret;
3048
3049 if (unlikely(ftrace_disabled))
3050 return -ENODEV;
3051
3052 ret = __register_ftrace_function(ops);
3053 if (ret)
3054 return ret;
3055
3056 ftrace_start_up++;
3057
3058 /*
3059 * Note that ftrace probes uses this to start up
3060 * and modify functions it will probe. But we still
3061 * set the ADDING flag for modification, as probes
3062 * do not have trampolines. If they add them in the
3063 * future, then the probes will need to distinguish
3064 * between adding and updating probes.
3065 */
3066 ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
3067
3068 ret = ftrace_hash_ipmodify_enable(ops);
3069 if (ret < 0) {
3070 /* Rollback registration process */
3071 __unregister_ftrace_function(ops);
3072 ftrace_start_up--;
3073 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
3074 if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
3075 ftrace_trampoline_free(ops);
3076 return ret;
3077 }
3078
3079 if (ftrace_hash_rec_enable(ops))
3080 command |= FTRACE_UPDATE_CALLS;
3081
3082 ftrace_startup_enable(command);
3083
3084 /*
3085 * If ftrace is in an undefined state, we just remove ops from list
3086 * to prevent the NULL pointer, instead of totally rolling it back and
3087 * free trampoline, because those actions could cause further damage.
3088 */
3089 if (unlikely(ftrace_disabled)) {
3090 __unregister_ftrace_function(ops);
3091 return -ENODEV;
3092 }
3093
3094 ops->flags &= ~FTRACE_OPS_FL_ADDING;
3095
3096 return 0;
3097 }
3098
ftrace_shutdown(struct ftrace_ops * ops,int command)3099 int ftrace_shutdown(struct ftrace_ops *ops, int command)
3100 {
3101 int ret;
3102
3103 if (unlikely(ftrace_disabled))
3104 return -ENODEV;
3105
3106 ret = __unregister_ftrace_function(ops);
3107 if (ret)
3108 return ret;
3109
3110 ftrace_start_up--;
3111 /*
3112 * Just warn in case of unbalance, no need to kill ftrace, it's not
3113 * critical but the ftrace_call callers may be never nopped again after
3114 * further ftrace uses.
3115 */
3116 WARN_ON_ONCE(ftrace_start_up < 0);
3117
3118 /* Disabling ipmodify never fails */
3119 ftrace_hash_ipmodify_disable(ops);
3120
3121 if (ftrace_hash_rec_disable(ops))
3122 command |= FTRACE_UPDATE_CALLS;
3123
3124 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
3125
3126 if (saved_ftrace_func != ftrace_trace_function) {
3127 saved_ftrace_func = ftrace_trace_function;
3128 command |= FTRACE_UPDATE_TRACE_FUNC;
3129 }
3130
3131 if (!command || !ftrace_enabled)
3132 goto out;
3133
3134 /*
3135 * If the ops uses a trampoline, then it needs to be
3136 * tested first on update.
3137 */
3138 ops->flags |= FTRACE_OPS_FL_REMOVING;
3139 removed_ops = ops;
3140
3141 /* The trampoline logic checks the old hashes */
3142 ops->old_hash.filter_hash = ops->func_hash->filter_hash;
3143 ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
3144
3145 ftrace_run_update_code(command);
3146
3147 /*
3148 * If there's no more ops registered with ftrace, run a
3149 * sanity check to make sure all rec flags are cleared.
3150 */
3151 if (rcu_dereference_protected(ftrace_ops_list,
3152 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
3153 struct ftrace_page *pg;
3154 struct dyn_ftrace *rec;
3155
3156 do_for_each_ftrace_rec(pg, rec) {
3157 if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_NOCLEAR_FLAGS))
3158 pr_warn(" %pS flags:%lx\n",
3159 (void *)rec->ip, rec->flags);
3160 } while_for_each_ftrace_rec();
3161 }
3162
3163 ops->old_hash.filter_hash = NULL;
3164 ops->old_hash.notrace_hash = NULL;
3165
3166 removed_ops = NULL;
3167 ops->flags &= ~FTRACE_OPS_FL_REMOVING;
3168
3169 out:
3170 /*
3171 * Dynamic ops may be freed, we must make sure that all
3172 * callers are done before leaving this function.
3173 */
3174 if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
3175 /*
3176 * We need to do a hard force of sched synchronization.
3177 * This is because we use preempt_disable() to do RCU, but
3178 * the function tracers can be called where RCU is not watching
3179 * (like before user_exit()). We can not rely on the RCU
3180 * infrastructure to do the synchronization, thus we must do it
3181 * ourselves.
3182 */
3183 synchronize_rcu_tasks_rude();
3184
3185 /*
3186 * When the kernel is preemptive, tasks can be preempted
3187 * while on a ftrace trampoline. Just scheduling a task on
3188 * a CPU is not good enough to flush them. Calling
3189 * synchronize_rcu_tasks() will wait for those tasks to
3190 * execute and either schedule voluntarily or enter user space.
3191 */
3192 synchronize_rcu_tasks();
3193
3194 ftrace_trampoline_free(ops);
3195 }
3196
3197 return 0;
3198 }
3199
3200 /* Simply make a copy of @src and return it */
copy_hash(struct ftrace_hash * src)3201 static struct ftrace_hash *copy_hash(struct ftrace_hash *src)
3202 {
3203 if (ftrace_hash_empty(src))
3204 return EMPTY_HASH;
3205
3206 return alloc_and_copy_ftrace_hash(src->size_bits, src);
3207 }
3208
3209 /*
3210 * Append @new_hash entries to @hash:
3211 *
3212 * If @hash is the EMPTY_HASH then it traces all functions and nothing
3213 * needs to be done.
3214 *
3215 * If @new_hash is the EMPTY_HASH, then make *hash the EMPTY_HASH so
3216 * that it traces everything.
3217 *
3218 * Otherwise, go through all of @new_hash and add anything that @hash
3219 * doesn't already have, to @hash.
3220 *
3221 * The filter_hash updates uses just the append_hash() function
3222 * and the notrace_hash does not.
3223 */
append_hash(struct ftrace_hash ** hash,struct ftrace_hash * new_hash,int size_bits)3224 static int append_hash(struct ftrace_hash **hash, struct ftrace_hash *new_hash,
3225 int size_bits)
3226 {
3227 struct ftrace_func_entry *entry;
3228 int size;
3229 int i;
3230
3231 if (*hash) {
3232 /* An empty hash does everything */
3233 if (ftrace_hash_empty(*hash))
3234 return 0;
3235 } else {
3236 *hash = alloc_ftrace_hash(size_bits);
3237 if (!*hash)
3238 return -ENOMEM;
3239 }
3240
3241 /* If new_hash has everything make hash have everything */
3242 if (ftrace_hash_empty(new_hash)) {
3243 free_ftrace_hash(*hash);
3244 *hash = EMPTY_HASH;
3245 return 0;
3246 }
3247
3248 size = 1 << new_hash->size_bits;
3249 for (i = 0; i < size; i++) {
3250 hlist_for_each_entry(entry, &new_hash->buckets[i], hlist) {
3251 /* Only add if not already in hash */
3252 if (!__ftrace_lookup_ip(*hash, entry->ip) &&
3253 add_hash_entry(*hash, entry->ip) == NULL)
3254 return -ENOMEM;
3255 }
3256 }
3257 return 0;
3258 }
3259
3260 /*
3261 * Remove functions from @hash that are in @notrace_hash
3262 */
remove_hash(struct ftrace_hash * hash,struct ftrace_hash * notrace_hash)3263 static void remove_hash(struct ftrace_hash *hash, struct ftrace_hash *notrace_hash)
3264 {
3265 struct ftrace_func_entry *entry;
3266 struct hlist_node *tmp;
3267 int size;
3268 int i;
3269
3270 /* If the notrace hash is empty, there's nothing to do */
3271 if (ftrace_hash_empty(notrace_hash))
3272 return;
3273
3274 size = 1 << hash->size_bits;
3275 for (i = 0; i < size; i++) {
3276 hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
3277 if (!__ftrace_lookup_ip(notrace_hash, entry->ip))
3278 continue;
3279 remove_hash_entry(hash, entry);
3280 kfree(entry);
3281 }
3282 }
3283 }
3284
3285 /*
3286 * Add to @hash only those that are in both @new_hash1 and @new_hash2
3287 *
3288 * The notrace_hash updates uses just the intersect_hash() function
3289 * and the filter_hash does not.
3290 */
intersect_hash(struct ftrace_hash ** hash,struct ftrace_hash * new_hash1,struct ftrace_hash * new_hash2)3291 static int intersect_hash(struct ftrace_hash **hash, struct ftrace_hash *new_hash1,
3292 struct ftrace_hash *new_hash2)
3293 {
3294 struct ftrace_func_entry *entry;
3295 int size;
3296 int i;
3297
3298 /*
3299 * If new_hash1 or new_hash2 is the EMPTY_HASH then make the hash
3300 * empty as well as empty for notrace means none are notraced.
3301 */
3302 if (ftrace_hash_empty(new_hash1) || ftrace_hash_empty(new_hash2)) {
3303 free_ftrace_hash(*hash);
3304 *hash = EMPTY_HASH;
3305 return 0;
3306 }
3307
3308 size = 1 << new_hash1->size_bits;
3309 for (i = 0; i < size; i++) {
3310 hlist_for_each_entry(entry, &new_hash1->buckets[i], hlist) {
3311 /* Only add if in both @new_hash1 and @new_hash2 */
3312 if (__ftrace_lookup_ip(new_hash2, entry->ip) &&
3313 add_hash_entry(*hash, entry->ip) == NULL)
3314 return -ENOMEM;
3315 }
3316 }
3317 /* If nothing intersects, make it the empty set */
3318 if (ftrace_hash_empty(*hash)) {
3319 free_ftrace_hash(*hash);
3320 *hash = EMPTY_HASH;
3321 }
3322 return 0;
3323 }
3324
ops_equal(struct ftrace_hash * A,struct ftrace_hash * B)3325 static bool ops_equal(struct ftrace_hash *A, struct ftrace_hash *B)
3326 {
3327 struct ftrace_func_entry *entry;
3328 int size;
3329 int i;
3330
3331 if (ftrace_hash_empty(A))
3332 return ftrace_hash_empty(B);
3333
3334 if (ftrace_hash_empty(B))
3335 return ftrace_hash_empty(A);
3336
3337 if (A->count != B->count)
3338 return false;
3339
3340 size = 1 << A->size_bits;
3341 for (i = 0; i < size; i++) {
3342 hlist_for_each_entry(entry, &A->buckets[i], hlist) {
3343 if (!__ftrace_lookup_ip(B, entry->ip))
3344 return false;
3345 }
3346 }
3347
3348 return true;
3349 }
3350
3351 static void ftrace_ops_update_code(struct ftrace_ops *ops,
3352 struct ftrace_ops_hash *old_hash);
3353
__ftrace_hash_move_and_update_ops(struct ftrace_ops * ops,struct ftrace_hash ** orig_hash,struct ftrace_hash * hash,int enable)3354 static int __ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
3355 struct ftrace_hash **orig_hash,
3356 struct ftrace_hash *hash,
3357 int enable)
3358 {
3359 struct ftrace_ops_hash old_hash_ops;
3360 struct ftrace_hash *old_hash;
3361 int ret;
3362
3363 old_hash = *orig_hash;
3364 old_hash_ops.filter_hash = ops->func_hash->filter_hash;
3365 old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
3366 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3367 if (!ret) {
3368 ftrace_ops_update_code(ops, &old_hash_ops);
3369 free_ftrace_hash_rcu(old_hash);
3370 }
3371 return ret;
3372 }
3373
ftrace_update_ops(struct ftrace_ops * ops,struct ftrace_hash * filter_hash,struct ftrace_hash * notrace_hash)3374 static int ftrace_update_ops(struct ftrace_ops *ops, struct ftrace_hash *filter_hash,
3375 struct ftrace_hash *notrace_hash)
3376 {
3377 int ret;
3378
3379 if (!ops_equal(filter_hash, ops->func_hash->filter_hash)) {
3380 ret = __ftrace_hash_move_and_update_ops(ops, &ops->func_hash->filter_hash,
3381 filter_hash, 1);
3382 if (ret < 0)
3383 return ret;
3384 }
3385
3386 if (!ops_equal(notrace_hash, ops->func_hash->notrace_hash)) {
3387 ret = __ftrace_hash_move_and_update_ops(ops, &ops->func_hash->notrace_hash,
3388 notrace_hash, 0);
3389 if (ret < 0)
3390 return ret;
3391 }
3392
3393 return 0;
3394 }
3395
add_first_hash(struct ftrace_hash ** filter_hash,struct ftrace_hash ** notrace_hash,struct ftrace_ops_hash * func_hash)3396 static int add_first_hash(struct ftrace_hash **filter_hash, struct ftrace_hash **notrace_hash,
3397 struct ftrace_ops_hash *func_hash)
3398 {
3399 /* If the filter hash is not empty, simply remove the nohash from it */
3400 if (!ftrace_hash_empty(func_hash->filter_hash)) {
3401 *filter_hash = copy_hash(func_hash->filter_hash);
3402 if (!*filter_hash)
3403 return -ENOMEM;
3404 remove_hash(*filter_hash, func_hash->notrace_hash);
3405 *notrace_hash = EMPTY_HASH;
3406
3407 } else {
3408 *notrace_hash = copy_hash(func_hash->notrace_hash);
3409 if (!*notrace_hash)
3410 return -ENOMEM;
3411 *filter_hash = EMPTY_HASH;
3412 }
3413 return 0;
3414 }
3415
add_next_hash(struct ftrace_hash ** filter_hash,struct ftrace_hash ** notrace_hash,struct ftrace_ops_hash * ops_hash,struct ftrace_ops_hash * subops_hash)3416 static int add_next_hash(struct ftrace_hash **filter_hash, struct ftrace_hash **notrace_hash,
3417 struct ftrace_ops_hash *ops_hash, struct ftrace_ops_hash *subops_hash)
3418 {
3419 int size_bits;
3420 int ret;
3421
3422 /* If the subops trace all functions so must the main ops */
3423 if (ftrace_hash_empty(ops_hash->filter_hash) ||
3424 ftrace_hash_empty(subops_hash->filter_hash)) {
3425 *filter_hash = EMPTY_HASH;
3426 } else {
3427 /*
3428 * The main ops filter hash is not empty, so its
3429 * notrace_hash had better be, as the notrace hash
3430 * is only used for empty main filter hashes.
3431 */
3432 WARN_ON_ONCE(!ftrace_hash_empty(ops_hash->notrace_hash));
3433
3434 size_bits = max(ops_hash->filter_hash->size_bits,
3435 subops_hash->filter_hash->size_bits);
3436
3437 /* Copy the subops hash */
3438 *filter_hash = alloc_and_copy_ftrace_hash(size_bits, subops_hash->filter_hash);
3439 if (!filter_hash)
3440 return -ENOMEM;
3441 /* Remove any notrace functions from the copy */
3442 remove_hash(*filter_hash, subops_hash->notrace_hash);
3443
3444 ret = append_hash(filter_hash, ops_hash->filter_hash,
3445 size_bits);
3446 if (ret < 0) {
3447 free_ftrace_hash(*filter_hash);
3448 *filter_hash = EMPTY_HASH;
3449 return ret;
3450 }
3451 }
3452
3453 /*
3454 * Only process notrace hashes if the main filter hash is empty
3455 * (tracing all functions), otherwise the filter hash will just
3456 * remove the notrace hash functions, and the notrace hash is
3457 * not needed.
3458 */
3459 if (ftrace_hash_empty(*filter_hash)) {
3460 /*
3461 * Intersect the notrace functions. That is, if two
3462 * subops are not tracing a set of functions, the
3463 * main ops will only not trace the functions that are
3464 * in both subops, but has to trace the functions that
3465 * are only notrace in one of the subops, for the other
3466 * subops to be able to trace them.
3467 */
3468 size_bits = max(ops_hash->notrace_hash->size_bits,
3469 subops_hash->notrace_hash->size_bits);
3470 *notrace_hash = alloc_ftrace_hash(size_bits);
3471 if (!*notrace_hash)
3472 return -ENOMEM;
3473
3474 ret = intersect_hash(notrace_hash, ops_hash->notrace_hash,
3475 subops_hash->notrace_hash);
3476 if (ret < 0) {
3477 free_ftrace_hash(*notrace_hash);
3478 *notrace_hash = EMPTY_HASH;
3479 return ret;
3480 }
3481 }
3482 return 0;
3483 }
3484
3485 /**
3486 * ftrace_startup_subops - enable tracing for subops of an ops
3487 * @ops: Manager ops (used to pick all the functions of its subops)
3488 * @subops: A new ops to add to @ops
3489 * @command: Extra commands to use to enable tracing
3490 *
3491 * The @ops is a manager @ops that has the filter that includes all the functions
3492 * that its list of subops are tracing. Adding a new @subops will add the
3493 * functions of @subops to @ops.
3494 */
ftrace_startup_subops(struct ftrace_ops * ops,struct ftrace_ops * subops,int command)3495 int ftrace_startup_subops(struct ftrace_ops *ops, struct ftrace_ops *subops, int command)
3496 {
3497 struct ftrace_hash *filter_hash = EMPTY_HASH;
3498 struct ftrace_hash *notrace_hash = EMPTY_HASH;
3499 struct ftrace_hash *save_filter_hash;
3500 struct ftrace_hash *save_notrace_hash;
3501 int ret;
3502
3503 if (unlikely(ftrace_disabled))
3504 return -ENODEV;
3505
3506 ftrace_ops_init(ops);
3507 ftrace_ops_init(subops);
3508
3509 if (WARN_ON_ONCE(subops->flags & FTRACE_OPS_FL_ENABLED))
3510 return -EBUSY;
3511
3512 /* Make everything canonical (Just in case!) */
3513 if (!ops->func_hash->filter_hash)
3514 ops->func_hash->filter_hash = EMPTY_HASH;
3515 if (!ops->func_hash->notrace_hash)
3516 ops->func_hash->notrace_hash = EMPTY_HASH;
3517 if (!subops->func_hash->filter_hash)
3518 subops->func_hash->filter_hash = EMPTY_HASH;
3519 if (!subops->func_hash->notrace_hash)
3520 subops->func_hash->notrace_hash = EMPTY_HASH;
3521
3522 /* For the first subops to ops just enable it normally */
3523 if (list_empty(&ops->subop_list)) {
3524
3525 /* The ops was empty, should have empty hashes */
3526 WARN_ON_ONCE(!ftrace_hash_empty(ops->func_hash->filter_hash));
3527 WARN_ON_ONCE(!ftrace_hash_empty(ops->func_hash->notrace_hash));
3528
3529 ret = add_first_hash(&filter_hash, ¬race_hash, subops->func_hash);
3530 if (ret < 0)
3531 return ret;
3532
3533 save_filter_hash = ops->func_hash->filter_hash;
3534 save_notrace_hash = ops->func_hash->notrace_hash;
3535
3536 ops->func_hash->filter_hash = filter_hash;
3537 ops->func_hash->notrace_hash = notrace_hash;
3538 list_add(&subops->list, &ops->subop_list);
3539 ret = ftrace_startup(ops, command);
3540 if (ret < 0) {
3541 list_del(&subops->list);
3542 ops->func_hash->filter_hash = save_filter_hash;
3543 ops->func_hash->notrace_hash = save_notrace_hash;
3544 free_ftrace_hash(filter_hash);
3545 free_ftrace_hash(notrace_hash);
3546 } else {
3547 free_ftrace_hash(save_filter_hash);
3548 free_ftrace_hash(save_notrace_hash);
3549 subops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP;
3550 subops->managed = ops;
3551 }
3552 return ret;
3553 }
3554
3555 /*
3556 * Here there's already something attached. Here are the rules:
3557 * If the new subops and main ops filter hashes are not empty:
3558 * o Make a copy of the subops filter hash
3559 * o Remove all functions in the nohash from it.
3560 * o Add in the main hash filter functions
3561 * o Remove any of these functions from the main notrace hash
3562 */
3563
3564 ret = add_next_hash(&filter_hash, ¬race_hash, ops->func_hash, subops->func_hash);
3565 if (ret < 0)
3566 return ret;
3567
3568 list_add(&subops->list, &ops->subop_list);
3569
3570 ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3571 free_ftrace_hash(filter_hash);
3572 free_ftrace_hash(notrace_hash);
3573 if (ret < 0) {
3574 list_del(&subops->list);
3575 } else {
3576 subops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP;
3577 subops->managed = ops;
3578 }
3579 return ret;
3580 }
3581
rebuild_hashes(struct ftrace_hash ** filter_hash,struct ftrace_hash ** notrace_hash,struct ftrace_ops * ops)3582 static int rebuild_hashes(struct ftrace_hash **filter_hash, struct ftrace_hash **notrace_hash,
3583 struct ftrace_ops *ops)
3584 {
3585 struct ftrace_ops_hash temp_hash;
3586 struct ftrace_ops *subops;
3587 bool first = true;
3588 int ret;
3589
3590 temp_hash.filter_hash = EMPTY_HASH;
3591 temp_hash.notrace_hash = EMPTY_HASH;
3592
3593 list_for_each_entry(subops, &ops->subop_list, list) {
3594 *filter_hash = EMPTY_HASH;
3595 *notrace_hash = EMPTY_HASH;
3596
3597 if (first) {
3598 ret = add_first_hash(filter_hash, notrace_hash, subops->func_hash);
3599 if (ret < 0)
3600 return ret;
3601 first = false;
3602 } else {
3603 ret = add_next_hash(filter_hash, notrace_hash,
3604 &temp_hash, subops->func_hash);
3605 if (ret < 0) {
3606 free_ftrace_hash(temp_hash.filter_hash);
3607 free_ftrace_hash(temp_hash.notrace_hash);
3608 return ret;
3609 }
3610 }
3611
3612 free_ftrace_hash(temp_hash.filter_hash);
3613 free_ftrace_hash(temp_hash.notrace_hash);
3614
3615 temp_hash.filter_hash = *filter_hash;
3616 temp_hash.notrace_hash = *notrace_hash;
3617 }
3618 return 0;
3619 }
3620
3621 /**
3622 * ftrace_shutdown_subops - Remove a subops from a manager ops
3623 * @ops: A manager ops to remove @subops from
3624 * @subops: The subops to remove from @ops
3625 * @command: Any extra command flags to add to modifying the text
3626 *
3627 * Removes the functions being traced by the @subops from @ops. Note, it
3628 * will not affect functions that are being traced by other subops that
3629 * still exist in @ops.
3630 *
3631 * If the last subops is removed from @ops, then @ops is shutdown normally.
3632 */
ftrace_shutdown_subops(struct ftrace_ops * ops,struct ftrace_ops * subops,int command)3633 int ftrace_shutdown_subops(struct ftrace_ops *ops, struct ftrace_ops *subops, int command)
3634 {
3635 struct ftrace_hash *filter_hash = EMPTY_HASH;
3636 struct ftrace_hash *notrace_hash = EMPTY_HASH;
3637 int ret;
3638
3639 if (unlikely(ftrace_disabled))
3640 return -ENODEV;
3641
3642 if (WARN_ON_ONCE(!(subops->flags & FTRACE_OPS_FL_ENABLED)))
3643 return -EINVAL;
3644
3645 list_del(&subops->list);
3646
3647 if (list_empty(&ops->subop_list)) {
3648 /* Last one, just disable the current ops */
3649
3650 ret = ftrace_shutdown(ops, command);
3651 if (ret < 0) {
3652 list_add(&subops->list, &ops->subop_list);
3653 return ret;
3654 }
3655
3656 subops->flags &= ~FTRACE_OPS_FL_ENABLED;
3657
3658 free_ftrace_hash(ops->func_hash->filter_hash);
3659 free_ftrace_hash(ops->func_hash->notrace_hash);
3660 ops->func_hash->filter_hash = EMPTY_HASH;
3661 ops->func_hash->notrace_hash = EMPTY_HASH;
3662 subops->flags &= ~(FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP);
3663 subops->managed = NULL;
3664
3665 return 0;
3666 }
3667
3668 /* Rebuild the hashes without subops */
3669 ret = rebuild_hashes(&filter_hash, ¬race_hash, ops);
3670 if (ret < 0)
3671 return ret;
3672
3673 ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3674 if (ret < 0) {
3675 list_add(&subops->list, &ops->subop_list);
3676 } else {
3677 subops->flags &= ~(FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP);
3678 subops->managed = NULL;
3679 }
3680 free_ftrace_hash(filter_hash);
3681 free_ftrace_hash(notrace_hash);
3682 return ret;
3683 }
3684
ftrace_hash_move_and_update_subops(struct ftrace_ops * subops,struct ftrace_hash ** orig_subhash,struct ftrace_hash * hash)3685 static int ftrace_hash_move_and_update_subops(struct ftrace_ops *subops,
3686 struct ftrace_hash **orig_subhash,
3687 struct ftrace_hash *hash)
3688 {
3689 struct ftrace_ops *ops = subops->managed;
3690 struct ftrace_hash *notrace_hash;
3691 struct ftrace_hash *filter_hash;
3692 struct ftrace_hash *save_hash;
3693 struct ftrace_hash *new_hash;
3694 int ret;
3695
3696 /* Manager ops can not be subops (yet) */
3697 if (WARN_ON_ONCE(!ops || ops->flags & FTRACE_OPS_FL_SUBOP))
3698 return -EINVAL;
3699
3700 /* Move the new hash over to the subops hash */
3701 save_hash = *orig_subhash;
3702 *orig_subhash = __ftrace_hash_move(hash);
3703 if (!*orig_subhash) {
3704 *orig_subhash = save_hash;
3705 return -ENOMEM;
3706 }
3707
3708 ret = rebuild_hashes(&filter_hash, ¬race_hash, ops);
3709 if (!ret) {
3710 ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3711 free_ftrace_hash(filter_hash);
3712 free_ftrace_hash(notrace_hash);
3713 }
3714
3715 if (ret) {
3716 /* Put back the original hash */
3717 new_hash = *orig_subhash;
3718 *orig_subhash = save_hash;
3719 free_ftrace_hash_rcu(new_hash);
3720 } else {
3721 free_ftrace_hash_rcu(save_hash);
3722 }
3723 return ret;
3724 }
3725
3726
3727 u64 ftrace_update_time;
3728 u64 ftrace_total_mod_time;
3729 unsigned long ftrace_update_tot_cnt;
3730 unsigned long ftrace_number_of_pages;
3731 unsigned long ftrace_number_of_groups;
3732
ops_traces_mod(struct ftrace_ops * ops)3733 static inline int ops_traces_mod(struct ftrace_ops *ops)
3734 {
3735 /*
3736 * Filter_hash being empty will default to trace module.
3737 * But notrace hash requires a test of individual module functions.
3738 */
3739 return ftrace_hash_empty(ops->func_hash->filter_hash) &&
3740 ftrace_hash_empty(ops->func_hash->notrace_hash);
3741 }
3742
ftrace_update_code(struct module * mod,struct ftrace_page * new_pgs)3743 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
3744 {
3745 bool init_nop = ftrace_need_init_nop();
3746 struct ftrace_page *pg;
3747 struct dyn_ftrace *p;
3748 u64 start, stop, update_time;
3749 unsigned long update_cnt = 0;
3750 unsigned long rec_flags = 0;
3751 int i;
3752
3753 start = ftrace_now(raw_smp_processor_id());
3754
3755 /*
3756 * When a module is loaded, this function is called to convert
3757 * the calls to mcount in its text to nops, and also to create
3758 * an entry in the ftrace data. Now, if ftrace is activated
3759 * after this call, but before the module sets its text to
3760 * read-only, the modification of enabling ftrace can fail if
3761 * the read-only is done while ftrace is converting the calls.
3762 * To prevent this, the module's records are set as disabled
3763 * and will be enabled after the call to set the module's text
3764 * to read-only.
3765 */
3766 if (mod)
3767 rec_flags |= FTRACE_FL_DISABLED;
3768
3769 for (pg = new_pgs; pg; pg = pg->next) {
3770
3771 for (i = 0; i < pg->index; i++) {
3772
3773 /* If something went wrong, bail without enabling anything */
3774 if (unlikely(ftrace_disabled))
3775 return -1;
3776
3777 p = &pg->records[i];
3778 p->flags = rec_flags;
3779
3780 /*
3781 * Do the initial record conversion from mcount jump
3782 * to the NOP instructions.
3783 */
3784 if (init_nop && !ftrace_nop_initialize(mod, p))
3785 break;
3786
3787 update_cnt++;
3788 }
3789 }
3790
3791 stop = ftrace_now(raw_smp_processor_id());
3792 update_time = stop - start;
3793 if (mod)
3794 ftrace_total_mod_time += update_time;
3795 else
3796 ftrace_update_time = update_time;
3797 ftrace_update_tot_cnt += update_cnt;
3798
3799 return 0;
3800 }
3801
ftrace_allocate_records(struct ftrace_page * pg,int count)3802 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
3803 {
3804 int order;
3805 int pages;
3806 int cnt;
3807
3808 if (WARN_ON(!count))
3809 return -EINVAL;
3810
3811 /* We want to fill as much as possible, with no empty pages */
3812 pages = DIV_ROUND_UP(count, ENTRIES_PER_PAGE);
3813 order = fls(pages) - 1;
3814
3815 again:
3816 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
3817
3818 if (!pg->records) {
3819 /* if we can't allocate this size, try something smaller */
3820 if (!order)
3821 return -ENOMEM;
3822 order--;
3823 goto again;
3824 }
3825
3826 ftrace_number_of_pages += 1 << order;
3827 ftrace_number_of_groups++;
3828
3829 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
3830 pg->order = order;
3831
3832 if (cnt > count)
3833 cnt = count;
3834
3835 return cnt;
3836 }
3837
ftrace_free_pages(struct ftrace_page * pages)3838 static void ftrace_free_pages(struct ftrace_page *pages)
3839 {
3840 struct ftrace_page *pg = pages;
3841
3842 while (pg) {
3843 if (pg->records) {
3844 free_pages((unsigned long)pg->records, pg->order);
3845 ftrace_number_of_pages -= 1 << pg->order;
3846 }
3847 pages = pg->next;
3848 kfree(pg);
3849 pg = pages;
3850 ftrace_number_of_groups--;
3851 }
3852 }
3853
3854 static struct ftrace_page *
ftrace_allocate_pages(unsigned long num_to_init)3855 ftrace_allocate_pages(unsigned long num_to_init)
3856 {
3857 struct ftrace_page *start_pg;
3858 struct ftrace_page *pg;
3859 int cnt;
3860
3861 if (!num_to_init)
3862 return NULL;
3863
3864 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3865 if (!pg)
3866 return NULL;
3867
3868 /*
3869 * Try to allocate as much as possible in one continues
3870 * location that fills in all of the space. We want to
3871 * waste as little space as possible.
3872 */
3873 for (;;) {
3874 cnt = ftrace_allocate_records(pg, num_to_init);
3875 if (cnt < 0)
3876 goto free_pages;
3877
3878 num_to_init -= cnt;
3879 if (!num_to_init)
3880 break;
3881
3882 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3883 if (!pg->next)
3884 goto free_pages;
3885
3886 pg = pg->next;
3887 }
3888
3889 return start_pg;
3890
3891 free_pages:
3892 ftrace_free_pages(start_pg);
3893 pr_info("ftrace: FAILED to allocate memory for functions\n");
3894 return NULL;
3895 }
3896
3897 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3898
3899 struct ftrace_iterator {
3900 loff_t pos;
3901 loff_t func_pos;
3902 loff_t mod_pos;
3903 struct ftrace_page *pg;
3904 struct dyn_ftrace *func;
3905 struct ftrace_func_probe *probe;
3906 struct ftrace_func_entry *probe_entry;
3907 struct trace_parser parser;
3908 struct ftrace_hash *hash;
3909 struct ftrace_ops *ops;
3910 struct trace_array *tr;
3911 struct list_head *mod_list;
3912 int pidx;
3913 int idx;
3914 unsigned flags;
3915 };
3916
3917 static void *
t_probe_next(struct seq_file * m,loff_t * pos)3918 t_probe_next(struct seq_file *m, loff_t *pos)
3919 {
3920 struct ftrace_iterator *iter = m->private;
3921 struct trace_array *tr = iter->ops->private;
3922 struct list_head *func_probes;
3923 struct ftrace_hash *hash;
3924 struct list_head *next;
3925 struct hlist_node *hnd = NULL;
3926 struct hlist_head *hhd;
3927 int size;
3928
3929 (*pos)++;
3930 iter->pos = *pos;
3931
3932 if (!tr)
3933 return NULL;
3934
3935 func_probes = &tr->func_probes;
3936 if (list_empty(func_probes))
3937 return NULL;
3938
3939 if (!iter->probe) {
3940 next = func_probes->next;
3941 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3942 }
3943
3944 if (iter->probe_entry)
3945 hnd = &iter->probe_entry->hlist;
3946
3947 hash = iter->probe->ops.func_hash->filter_hash;
3948
3949 /*
3950 * A probe being registered may temporarily have an empty hash
3951 * and it's at the end of the func_probes list.
3952 */
3953 if (!hash || hash == EMPTY_HASH)
3954 return NULL;
3955
3956 size = 1 << hash->size_bits;
3957
3958 retry:
3959 if (iter->pidx >= size) {
3960 if (iter->probe->list.next == func_probes)
3961 return NULL;
3962 next = iter->probe->list.next;
3963 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3964 hash = iter->probe->ops.func_hash->filter_hash;
3965 size = 1 << hash->size_bits;
3966 iter->pidx = 0;
3967 }
3968
3969 hhd = &hash->buckets[iter->pidx];
3970
3971 if (hlist_empty(hhd)) {
3972 iter->pidx++;
3973 hnd = NULL;
3974 goto retry;
3975 }
3976
3977 if (!hnd)
3978 hnd = hhd->first;
3979 else {
3980 hnd = hnd->next;
3981 if (!hnd) {
3982 iter->pidx++;
3983 goto retry;
3984 }
3985 }
3986
3987 if (WARN_ON_ONCE(!hnd))
3988 return NULL;
3989
3990 iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
3991
3992 return iter;
3993 }
3994
t_probe_start(struct seq_file * m,loff_t * pos)3995 static void *t_probe_start(struct seq_file *m, loff_t *pos)
3996 {
3997 struct ftrace_iterator *iter = m->private;
3998 void *p = NULL;
3999 loff_t l;
4000
4001 if (!(iter->flags & FTRACE_ITER_DO_PROBES))
4002 return NULL;
4003
4004 if (iter->mod_pos > *pos)
4005 return NULL;
4006
4007 iter->probe = NULL;
4008 iter->probe_entry = NULL;
4009 iter->pidx = 0;
4010 for (l = 0; l <= (*pos - iter->mod_pos); ) {
4011 p = t_probe_next(m, &l);
4012 if (!p)
4013 break;
4014 }
4015 if (!p)
4016 return NULL;
4017
4018 /* Only set this if we have an item */
4019 iter->flags |= FTRACE_ITER_PROBE;
4020
4021 return iter;
4022 }
4023
4024 static int
t_probe_show(struct seq_file * m,struct ftrace_iterator * iter)4025 t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
4026 {
4027 struct ftrace_func_entry *probe_entry;
4028 struct ftrace_probe_ops *probe_ops;
4029 struct ftrace_func_probe *probe;
4030
4031 probe = iter->probe;
4032 probe_entry = iter->probe_entry;
4033
4034 if (WARN_ON_ONCE(!probe || !probe_entry))
4035 return -EIO;
4036
4037 probe_ops = probe->probe_ops;
4038
4039 if (probe_ops->print)
4040 return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
4041
4042 seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
4043 (void *)probe_ops->func);
4044
4045 return 0;
4046 }
4047
4048 static void *
t_mod_next(struct seq_file * m,loff_t * pos)4049 t_mod_next(struct seq_file *m, loff_t *pos)
4050 {
4051 struct ftrace_iterator *iter = m->private;
4052 struct trace_array *tr = iter->tr;
4053
4054 (*pos)++;
4055 iter->pos = *pos;
4056
4057 iter->mod_list = iter->mod_list->next;
4058
4059 if (iter->mod_list == &tr->mod_trace ||
4060 iter->mod_list == &tr->mod_notrace) {
4061 iter->flags &= ~FTRACE_ITER_MOD;
4062 return NULL;
4063 }
4064
4065 iter->mod_pos = *pos;
4066
4067 return iter;
4068 }
4069
t_mod_start(struct seq_file * m,loff_t * pos)4070 static void *t_mod_start(struct seq_file *m, loff_t *pos)
4071 {
4072 struct ftrace_iterator *iter = m->private;
4073 void *p = NULL;
4074 loff_t l;
4075
4076 if (iter->func_pos > *pos)
4077 return NULL;
4078
4079 iter->mod_pos = iter->func_pos;
4080
4081 /* probes are only available if tr is set */
4082 if (!iter->tr)
4083 return NULL;
4084
4085 for (l = 0; l <= (*pos - iter->func_pos); ) {
4086 p = t_mod_next(m, &l);
4087 if (!p)
4088 break;
4089 }
4090 if (!p) {
4091 iter->flags &= ~FTRACE_ITER_MOD;
4092 return t_probe_start(m, pos);
4093 }
4094
4095 /* Only set this if we have an item */
4096 iter->flags |= FTRACE_ITER_MOD;
4097
4098 return iter;
4099 }
4100
4101 static int
t_mod_show(struct seq_file * m,struct ftrace_iterator * iter)4102 t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
4103 {
4104 struct ftrace_mod_load *ftrace_mod;
4105 struct trace_array *tr = iter->tr;
4106
4107 if (WARN_ON_ONCE(!iter->mod_list) ||
4108 iter->mod_list == &tr->mod_trace ||
4109 iter->mod_list == &tr->mod_notrace)
4110 return -EIO;
4111
4112 ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
4113
4114 if (ftrace_mod->func)
4115 seq_printf(m, "%s", ftrace_mod->func);
4116 else
4117 seq_putc(m, '*');
4118
4119 seq_printf(m, ":mod:%s\n", ftrace_mod->module);
4120
4121 return 0;
4122 }
4123
4124 static void *
t_func_next(struct seq_file * m,loff_t * pos)4125 t_func_next(struct seq_file *m, loff_t *pos)
4126 {
4127 struct ftrace_iterator *iter = m->private;
4128 struct dyn_ftrace *rec = NULL;
4129
4130 (*pos)++;
4131
4132 retry:
4133 if (iter->idx >= iter->pg->index) {
4134 if (iter->pg->next) {
4135 iter->pg = iter->pg->next;
4136 iter->idx = 0;
4137 goto retry;
4138 }
4139 } else {
4140 rec = &iter->pg->records[iter->idx++];
4141 if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
4142 !ftrace_lookup_ip(iter->hash, rec->ip)) ||
4143
4144 ((iter->flags & FTRACE_ITER_ENABLED) &&
4145 !(rec->flags & FTRACE_FL_ENABLED)) ||
4146
4147 ((iter->flags & FTRACE_ITER_TOUCHED) &&
4148 !(rec->flags & FTRACE_FL_TOUCHED))) {
4149
4150 rec = NULL;
4151 goto retry;
4152 }
4153 }
4154
4155 if (!rec)
4156 return NULL;
4157
4158 iter->pos = iter->func_pos = *pos;
4159 iter->func = rec;
4160
4161 return iter;
4162 }
4163
4164 static void *
t_next(struct seq_file * m,void * v,loff_t * pos)4165 t_next(struct seq_file *m, void *v, loff_t *pos)
4166 {
4167 struct ftrace_iterator *iter = m->private;
4168 loff_t l = *pos; /* t_probe_start() must use original pos */
4169 void *ret;
4170
4171 if (unlikely(ftrace_disabled))
4172 return NULL;
4173
4174 if (iter->flags & FTRACE_ITER_PROBE)
4175 return t_probe_next(m, pos);
4176
4177 if (iter->flags & FTRACE_ITER_MOD)
4178 return t_mod_next(m, pos);
4179
4180 if (iter->flags & FTRACE_ITER_PRINTALL) {
4181 /* next must increment pos, and t_probe_start does not */
4182 (*pos)++;
4183 return t_mod_start(m, &l);
4184 }
4185
4186 ret = t_func_next(m, pos);
4187
4188 if (!ret)
4189 return t_mod_start(m, &l);
4190
4191 return ret;
4192 }
4193
reset_iter_read(struct ftrace_iterator * iter)4194 static void reset_iter_read(struct ftrace_iterator *iter)
4195 {
4196 iter->pos = 0;
4197 iter->func_pos = 0;
4198 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
4199 }
4200
t_start(struct seq_file * m,loff_t * pos)4201 static void *t_start(struct seq_file *m, loff_t *pos)
4202 {
4203 struct ftrace_iterator *iter = m->private;
4204 void *p = NULL;
4205 loff_t l;
4206
4207 mutex_lock(&ftrace_lock);
4208
4209 if (unlikely(ftrace_disabled))
4210 return NULL;
4211
4212 /*
4213 * If an lseek was done, then reset and start from beginning.
4214 */
4215 if (*pos < iter->pos)
4216 reset_iter_read(iter);
4217
4218 /*
4219 * For set_ftrace_filter reading, if we have the filter
4220 * off, we can short cut and just print out that all
4221 * functions are enabled.
4222 */
4223 if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
4224 ftrace_hash_empty(iter->hash)) {
4225 iter->func_pos = 1; /* Account for the message */
4226 if (*pos > 0)
4227 return t_mod_start(m, pos);
4228 iter->flags |= FTRACE_ITER_PRINTALL;
4229 /* reset in case of seek/pread */
4230 iter->flags &= ~FTRACE_ITER_PROBE;
4231 return iter;
4232 }
4233
4234 if (iter->flags & FTRACE_ITER_MOD)
4235 return t_mod_start(m, pos);
4236
4237 /*
4238 * Unfortunately, we need to restart at ftrace_pages_start
4239 * every time we let go of the ftrace_mutex. This is because
4240 * those pointers can change without the lock.
4241 */
4242 iter->pg = ftrace_pages_start;
4243 iter->idx = 0;
4244 for (l = 0; l <= *pos; ) {
4245 p = t_func_next(m, &l);
4246 if (!p)
4247 break;
4248 }
4249
4250 if (!p)
4251 return t_mod_start(m, pos);
4252
4253 return iter;
4254 }
4255
t_stop(struct seq_file * m,void * p)4256 static void t_stop(struct seq_file *m, void *p)
4257 {
4258 mutex_unlock(&ftrace_lock);
4259 }
4260
4261 void * __weak
arch_ftrace_trampoline_func(struct ftrace_ops * ops,struct dyn_ftrace * rec)4262 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
4263 {
4264 return NULL;
4265 }
4266
add_trampoline_func(struct seq_file * m,struct ftrace_ops * ops,struct dyn_ftrace * rec)4267 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
4268 struct dyn_ftrace *rec)
4269 {
4270 void *ptr;
4271
4272 ptr = arch_ftrace_trampoline_func(ops, rec);
4273 if (ptr)
4274 seq_printf(m, " ->%pS", ptr);
4275 }
4276
4277 #ifdef FTRACE_MCOUNT_MAX_OFFSET
4278 /*
4279 * Weak functions can still have an mcount/fentry that is saved in
4280 * the __mcount_loc section. These can be detected by having a
4281 * symbol offset of greater than FTRACE_MCOUNT_MAX_OFFSET, as the
4282 * symbol found by kallsyms is not the function that the mcount/fentry
4283 * is part of. The offset is much greater in these cases.
4284 *
4285 * Test the record to make sure that the ip points to a valid kallsyms
4286 * and if not, mark it disabled.
4287 */
test_for_valid_rec(struct dyn_ftrace * rec)4288 static int test_for_valid_rec(struct dyn_ftrace *rec)
4289 {
4290 char str[KSYM_SYMBOL_LEN];
4291 unsigned long offset;
4292 const char *ret;
4293
4294 ret = kallsyms_lookup(rec->ip, NULL, &offset, NULL, str);
4295
4296 /* Weak functions can cause invalid addresses */
4297 if (!ret || offset > FTRACE_MCOUNT_MAX_OFFSET) {
4298 rec->flags |= FTRACE_FL_DISABLED;
4299 return 0;
4300 }
4301 return 1;
4302 }
4303
4304 static struct workqueue_struct *ftrace_check_wq __initdata;
4305 static struct work_struct ftrace_check_work __initdata;
4306
4307 /*
4308 * Scan all the mcount/fentry entries to make sure they are valid.
4309 */
ftrace_check_work_func(struct work_struct * work)4310 static __init void ftrace_check_work_func(struct work_struct *work)
4311 {
4312 struct ftrace_page *pg;
4313 struct dyn_ftrace *rec;
4314
4315 mutex_lock(&ftrace_lock);
4316 do_for_each_ftrace_rec(pg, rec) {
4317 test_for_valid_rec(rec);
4318 } while_for_each_ftrace_rec();
4319 mutex_unlock(&ftrace_lock);
4320 }
4321
ftrace_check_for_weak_functions(void)4322 static int __init ftrace_check_for_weak_functions(void)
4323 {
4324 INIT_WORK(&ftrace_check_work, ftrace_check_work_func);
4325
4326 ftrace_check_wq = alloc_workqueue("ftrace_check_wq", WQ_UNBOUND, 0);
4327
4328 queue_work(ftrace_check_wq, &ftrace_check_work);
4329 return 0;
4330 }
4331
ftrace_check_sync(void)4332 static int __init ftrace_check_sync(void)
4333 {
4334 /* Make sure the ftrace_check updates are finished */
4335 if (ftrace_check_wq)
4336 destroy_workqueue(ftrace_check_wq);
4337 return 0;
4338 }
4339
4340 late_initcall_sync(ftrace_check_sync);
4341 subsys_initcall(ftrace_check_for_weak_functions);
4342
print_rec(struct seq_file * m,unsigned long ip)4343 static int print_rec(struct seq_file *m, unsigned long ip)
4344 {
4345 unsigned long offset;
4346 char str[KSYM_SYMBOL_LEN];
4347 char *modname;
4348 const char *ret;
4349
4350 ret = kallsyms_lookup(ip, NULL, &offset, &modname, str);
4351 /* Weak functions can cause invalid addresses */
4352 if (!ret || offset > FTRACE_MCOUNT_MAX_OFFSET) {
4353 snprintf(str, KSYM_SYMBOL_LEN, "%s_%ld",
4354 FTRACE_INVALID_FUNCTION, offset);
4355 ret = NULL;
4356 }
4357
4358 seq_puts(m, str);
4359 if (modname)
4360 seq_printf(m, " [%s]", modname);
4361 return ret == NULL ? -1 : 0;
4362 }
4363 #else
test_for_valid_rec(struct dyn_ftrace * rec)4364 static inline int test_for_valid_rec(struct dyn_ftrace *rec)
4365 {
4366 return 1;
4367 }
4368
print_rec(struct seq_file * m,unsigned long ip)4369 static inline int print_rec(struct seq_file *m, unsigned long ip)
4370 {
4371 seq_printf(m, "%ps", (void *)ip);
4372 return 0;
4373 }
4374 #endif
4375
t_show(struct seq_file * m,void * v)4376 static int t_show(struct seq_file *m, void *v)
4377 {
4378 struct ftrace_iterator *iter = m->private;
4379 struct dyn_ftrace *rec;
4380
4381 if (iter->flags & FTRACE_ITER_PROBE)
4382 return t_probe_show(m, iter);
4383
4384 if (iter->flags & FTRACE_ITER_MOD)
4385 return t_mod_show(m, iter);
4386
4387 if (iter->flags & FTRACE_ITER_PRINTALL) {
4388 if (iter->flags & FTRACE_ITER_NOTRACE)
4389 seq_puts(m, "#### no functions disabled ####\n");
4390 else
4391 seq_puts(m, "#### all functions enabled ####\n");
4392 return 0;
4393 }
4394
4395 rec = iter->func;
4396
4397 if (!rec)
4398 return 0;
4399
4400 if (iter->flags & FTRACE_ITER_ADDRS)
4401 seq_printf(m, "%lx ", rec->ip);
4402
4403 if (print_rec(m, rec->ip)) {
4404 /* This should only happen when a rec is disabled */
4405 WARN_ON_ONCE(!(rec->flags & FTRACE_FL_DISABLED));
4406 seq_putc(m, '\n');
4407 return 0;
4408 }
4409
4410 if (iter->flags & (FTRACE_ITER_ENABLED | FTRACE_ITER_TOUCHED)) {
4411 struct ftrace_ops *ops;
4412
4413 seq_printf(m, " (%ld)%s%s%s%s%s",
4414 ftrace_rec_count(rec),
4415 rec->flags & FTRACE_FL_REGS ? " R" : " ",
4416 rec->flags & FTRACE_FL_IPMODIFY ? " I" : " ",
4417 rec->flags & FTRACE_FL_DIRECT ? " D" : " ",
4418 rec->flags & FTRACE_FL_CALL_OPS ? " O" : " ",
4419 rec->flags & FTRACE_FL_MODIFIED ? " M " : " ");
4420 if (rec->flags & FTRACE_FL_TRAMP_EN) {
4421 ops = ftrace_find_tramp_ops_any(rec);
4422 if (ops) {
4423 do {
4424 seq_printf(m, "\ttramp: %pS (%pS)",
4425 (void *)ops->trampoline,
4426 (void *)ops->func);
4427 add_trampoline_func(m, ops, rec);
4428 ops = ftrace_find_tramp_ops_next(rec, ops);
4429 } while (ops);
4430 } else
4431 seq_puts(m, "\ttramp: ERROR!");
4432 } else {
4433 add_trampoline_func(m, NULL, rec);
4434 }
4435 if (rec->flags & FTRACE_FL_CALL_OPS_EN) {
4436 ops = ftrace_find_unique_ops(rec);
4437 if (ops) {
4438 seq_printf(m, "\tops: %pS (%pS)",
4439 ops, ops->func);
4440 } else {
4441 seq_puts(m, "\tops: ERROR!");
4442 }
4443 }
4444 if (rec->flags & FTRACE_FL_DIRECT) {
4445 unsigned long direct;
4446
4447 direct = ftrace_find_rec_direct(rec->ip);
4448 if (direct)
4449 seq_printf(m, "\n\tdirect-->%pS", (void *)direct);
4450 }
4451 }
4452
4453 seq_putc(m, '\n');
4454
4455 return 0;
4456 }
4457
4458 static const struct seq_operations show_ftrace_seq_ops = {
4459 .start = t_start,
4460 .next = t_next,
4461 .stop = t_stop,
4462 .show = t_show,
4463 };
4464
4465 static int
ftrace_avail_open(struct inode * inode,struct file * file)4466 ftrace_avail_open(struct inode *inode, struct file *file)
4467 {
4468 struct ftrace_iterator *iter;
4469 int ret;
4470
4471 ret = security_locked_down(LOCKDOWN_TRACEFS);
4472 if (ret)
4473 return ret;
4474
4475 if (unlikely(ftrace_disabled))
4476 return -ENODEV;
4477
4478 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4479 if (!iter)
4480 return -ENOMEM;
4481
4482 iter->pg = ftrace_pages_start;
4483 iter->ops = &global_ops;
4484
4485 return 0;
4486 }
4487
4488 static int
ftrace_enabled_open(struct inode * inode,struct file * file)4489 ftrace_enabled_open(struct inode *inode, struct file *file)
4490 {
4491 struct ftrace_iterator *iter;
4492
4493 /*
4494 * This shows us what functions are currently being
4495 * traced and by what. Not sure if we want lockdown
4496 * to hide such critical information for an admin.
4497 * Although, perhaps it can show information we don't
4498 * want people to see, but if something is tracing
4499 * something, we probably want to know about it.
4500 */
4501
4502 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4503 if (!iter)
4504 return -ENOMEM;
4505
4506 iter->pg = ftrace_pages_start;
4507 iter->flags = FTRACE_ITER_ENABLED;
4508 iter->ops = &global_ops;
4509
4510 return 0;
4511 }
4512
4513 static int
ftrace_touched_open(struct inode * inode,struct file * file)4514 ftrace_touched_open(struct inode *inode, struct file *file)
4515 {
4516 struct ftrace_iterator *iter;
4517
4518 /*
4519 * This shows us what functions have ever been enabled
4520 * (traced, direct, patched, etc). Not sure if we want lockdown
4521 * to hide such critical information for an admin.
4522 * Although, perhaps it can show information we don't
4523 * want people to see, but if something had traced
4524 * something, we probably want to know about it.
4525 */
4526
4527 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4528 if (!iter)
4529 return -ENOMEM;
4530
4531 iter->pg = ftrace_pages_start;
4532 iter->flags = FTRACE_ITER_TOUCHED;
4533 iter->ops = &global_ops;
4534
4535 return 0;
4536 }
4537
4538 static int
ftrace_avail_addrs_open(struct inode * inode,struct file * file)4539 ftrace_avail_addrs_open(struct inode *inode, struct file *file)
4540 {
4541 struct ftrace_iterator *iter;
4542 int ret;
4543
4544 ret = security_locked_down(LOCKDOWN_TRACEFS);
4545 if (ret)
4546 return ret;
4547
4548 if (unlikely(ftrace_disabled))
4549 return -ENODEV;
4550
4551 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4552 if (!iter)
4553 return -ENOMEM;
4554
4555 iter->pg = ftrace_pages_start;
4556 iter->flags = FTRACE_ITER_ADDRS;
4557 iter->ops = &global_ops;
4558
4559 return 0;
4560 }
4561
4562 /**
4563 * ftrace_regex_open - initialize function tracer filter files
4564 * @ops: The ftrace_ops that hold the hash filters
4565 * @flag: The type of filter to process
4566 * @inode: The inode, usually passed in to your open routine
4567 * @file: The file, usually passed in to your open routine
4568 *
4569 * ftrace_regex_open() initializes the filter files for the
4570 * @ops. Depending on @flag it may process the filter hash or
4571 * the notrace hash of @ops. With this called from the open
4572 * routine, you can use ftrace_filter_write() for the write
4573 * routine if @flag has FTRACE_ITER_FILTER set, or
4574 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
4575 * tracing_lseek() should be used as the lseek routine, and
4576 * release must call ftrace_regex_release().
4577 *
4578 * Returns: 0 on success or a negative errno value on failure
4579 */
4580 int
ftrace_regex_open(struct ftrace_ops * ops,int flag,struct inode * inode,struct file * file)4581 ftrace_regex_open(struct ftrace_ops *ops, int flag,
4582 struct inode *inode, struct file *file)
4583 {
4584 struct ftrace_iterator *iter;
4585 struct ftrace_hash *hash;
4586 struct list_head *mod_head;
4587 struct trace_array *tr = ops->private;
4588 int ret = -ENOMEM;
4589
4590 ftrace_ops_init(ops);
4591
4592 if (unlikely(ftrace_disabled))
4593 return -ENODEV;
4594
4595 if (tracing_check_open_get_tr(tr))
4596 return -ENODEV;
4597
4598 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4599 if (!iter)
4600 goto out;
4601
4602 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX))
4603 goto out;
4604
4605 iter->ops = ops;
4606 iter->flags = flag;
4607 iter->tr = tr;
4608
4609 mutex_lock(&ops->func_hash->regex_lock);
4610
4611 if (flag & FTRACE_ITER_NOTRACE) {
4612 hash = ops->func_hash->notrace_hash;
4613 mod_head = tr ? &tr->mod_notrace : NULL;
4614 } else {
4615 hash = ops->func_hash->filter_hash;
4616 mod_head = tr ? &tr->mod_trace : NULL;
4617 }
4618
4619 iter->mod_list = mod_head;
4620
4621 if (file->f_mode & FMODE_WRITE) {
4622 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
4623
4624 if (file->f_flags & O_TRUNC) {
4625 iter->hash = alloc_ftrace_hash(size_bits);
4626 clear_ftrace_mod_list(mod_head);
4627 } else {
4628 iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
4629 }
4630
4631 if (!iter->hash) {
4632 trace_parser_put(&iter->parser);
4633 goto out_unlock;
4634 }
4635 } else
4636 iter->hash = hash;
4637
4638 ret = 0;
4639
4640 if (file->f_mode & FMODE_READ) {
4641 iter->pg = ftrace_pages_start;
4642
4643 ret = seq_open(file, &show_ftrace_seq_ops);
4644 if (!ret) {
4645 struct seq_file *m = file->private_data;
4646 m->private = iter;
4647 } else {
4648 /* Failed */
4649 free_ftrace_hash(iter->hash);
4650 trace_parser_put(&iter->parser);
4651 }
4652 } else
4653 file->private_data = iter;
4654
4655 out_unlock:
4656 mutex_unlock(&ops->func_hash->regex_lock);
4657
4658 out:
4659 if (ret) {
4660 kfree(iter);
4661 if (tr)
4662 trace_array_put(tr);
4663 }
4664
4665 return ret;
4666 }
4667
4668 static int
ftrace_filter_open(struct inode * inode,struct file * file)4669 ftrace_filter_open(struct inode *inode, struct file *file)
4670 {
4671 struct ftrace_ops *ops = inode->i_private;
4672
4673 /* Checks for tracefs lockdown */
4674 return ftrace_regex_open(ops,
4675 FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
4676 inode, file);
4677 }
4678
4679 static int
ftrace_notrace_open(struct inode * inode,struct file * file)4680 ftrace_notrace_open(struct inode *inode, struct file *file)
4681 {
4682 struct ftrace_ops *ops = inode->i_private;
4683
4684 /* Checks for tracefs lockdown */
4685 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
4686 inode, file);
4687 }
4688
4689 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
4690 struct ftrace_glob {
4691 char *search;
4692 unsigned len;
4693 int type;
4694 };
4695
4696 /*
4697 * If symbols in an architecture don't correspond exactly to the user-visible
4698 * name of what they represent, it is possible to define this function to
4699 * perform the necessary adjustments.
4700 */
arch_ftrace_match_adjust(char * str,const char * search)4701 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
4702 {
4703 return str;
4704 }
4705
ftrace_match(char * str,struct ftrace_glob * g)4706 static int ftrace_match(char *str, struct ftrace_glob *g)
4707 {
4708 int matched = 0;
4709 int slen;
4710
4711 str = arch_ftrace_match_adjust(str, g->search);
4712
4713 switch (g->type) {
4714 case MATCH_FULL:
4715 if (strcmp(str, g->search) == 0)
4716 matched = 1;
4717 break;
4718 case MATCH_FRONT_ONLY:
4719 if (strncmp(str, g->search, g->len) == 0)
4720 matched = 1;
4721 break;
4722 case MATCH_MIDDLE_ONLY:
4723 if (strstr(str, g->search))
4724 matched = 1;
4725 break;
4726 case MATCH_END_ONLY:
4727 slen = strlen(str);
4728 if (slen >= g->len &&
4729 memcmp(str + slen - g->len, g->search, g->len) == 0)
4730 matched = 1;
4731 break;
4732 case MATCH_GLOB:
4733 if (glob_match(g->search, str))
4734 matched = 1;
4735 break;
4736 }
4737
4738 return matched;
4739 }
4740
4741 static int
enter_record(struct ftrace_hash * hash,struct dyn_ftrace * rec,int clear_filter)4742 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
4743 {
4744 struct ftrace_func_entry *entry;
4745 int ret = 0;
4746
4747 entry = ftrace_lookup_ip(hash, rec->ip);
4748 if (clear_filter) {
4749 /* Do nothing if it doesn't exist */
4750 if (!entry)
4751 return 0;
4752
4753 free_hash_entry(hash, entry);
4754 } else {
4755 /* Do nothing if it exists */
4756 if (entry)
4757 return 0;
4758 if (add_hash_entry(hash, rec->ip) == NULL)
4759 ret = -ENOMEM;
4760 }
4761 return ret;
4762 }
4763
4764 static int
add_rec_by_index(struct ftrace_hash * hash,struct ftrace_glob * func_g,int clear_filter)4765 add_rec_by_index(struct ftrace_hash *hash, struct ftrace_glob *func_g,
4766 int clear_filter)
4767 {
4768 long index;
4769 struct ftrace_page *pg;
4770 struct dyn_ftrace *rec;
4771
4772 /* The index starts at 1 */
4773 if (kstrtoul(func_g->search, 0, &index) || --index < 0)
4774 return 0;
4775
4776 do_for_each_ftrace_rec(pg, rec) {
4777 if (pg->index <= index) {
4778 index -= pg->index;
4779 /* this is a double loop, break goes to the next page */
4780 break;
4781 }
4782 rec = &pg->records[index];
4783 enter_record(hash, rec, clear_filter);
4784 return 1;
4785 } while_for_each_ftrace_rec();
4786 return 0;
4787 }
4788
4789 #ifdef FTRACE_MCOUNT_MAX_OFFSET
lookup_ip(unsigned long ip,char ** modname,char * str)4790 static int lookup_ip(unsigned long ip, char **modname, char *str)
4791 {
4792 unsigned long offset;
4793
4794 kallsyms_lookup(ip, NULL, &offset, modname, str);
4795 if (offset > FTRACE_MCOUNT_MAX_OFFSET)
4796 return -1;
4797 return 0;
4798 }
4799 #else
lookup_ip(unsigned long ip,char ** modname,char * str)4800 static int lookup_ip(unsigned long ip, char **modname, char *str)
4801 {
4802 kallsyms_lookup(ip, NULL, NULL, modname, str);
4803 return 0;
4804 }
4805 #endif
4806
4807 static int
ftrace_match_record(struct dyn_ftrace * rec,struct ftrace_glob * func_g,struct ftrace_glob * mod_g,int exclude_mod)4808 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
4809 struct ftrace_glob *mod_g, int exclude_mod)
4810 {
4811 char str[KSYM_SYMBOL_LEN];
4812 char *modname;
4813
4814 if (lookup_ip(rec->ip, &modname, str)) {
4815 /* This should only happen when a rec is disabled */
4816 WARN_ON_ONCE(system_state == SYSTEM_RUNNING &&
4817 !(rec->flags & FTRACE_FL_DISABLED));
4818 return 0;
4819 }
4820
4821 if (mod_g) {
4822 int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
4823
4824 /* blank module name to match all modules */
4825 if (!mod_g->len) {
4826 /* blank module globbing: modname xor exclude_mod */
4827 if (!exclude_mod != !modname)
4828 goto func_match;
4829 return 0;
4830 }
4831
4832 /*
4833 * exclude_mod is set to trace everything but the given
4834 * module. If it is set and the module matches, then
4835 * return 0. If it is not set, and the module doesn't match
4836 * also return 0. Otherwise, check the function to see if
4837 * that matches.
4838 */
4839 if (!mod_matches == !exclude_mod)
4840 return 0;
4841 func_match:
4842 /* blank search means to match all funcs in the mod */
4843 if (!func_g->len)
4844 return 1;
4845 }
4846
4847 return ftrace_match(str, func_g);
4848 }
4849
4850 static int
match_records(struct ftrace_hash * hash,char * func,int len,char * mod)4851 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
4852 {
4853 struct ftrace_page *pg;
4854 struct dyn_ftrace *rec;
4855 struct ftrace_glob func_g = { .type = MATCH_FULL };
4856 struct ftrace_glob mod_g = { .type = MATCH_FULL };
4857 struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
4858 int exclude_mod = 0;
4859 int found = 0;
4860 int ret;
4861 int clear_filter = 0;
4862
4863 if (func) {
4864 func_g.type = filter_parse_regex(func, len, &func_g.search,
4865 &clear_filter);
4866 func_g.len = strlen(func_g.search);
4867 }
4868
4869 if (mod) {
4870 mod_g.type = filter_parse_regex(mod, strlen(mod),
4871 &mod_g.search, &exclude_mod);
4872 mod_g.len = strlen(mod_g.search);
4873 }
4874
4875 guard(mutex)(&ftrace_lock);
4876
4877 if (unlikely(ftrace_disabled))
4878 return 0;
4879
4880 if (func_g.type == MATCH_INDEX)
4881 return add_rec_by_index(hash, &func_g, clear_filter);
4882
4883 do_for_each_ftrace_rec(pg, rec) {
4884
4885 if (rec->flags & FTRACE_FL_DISABLED)
4886 continue;
4887
4888 if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
4889 ret = enter_record(hash, rec, clear_filter);
4890 if (ret < 0)
4891 return ret;
4892 found = 1;
4893 }
4894 cond_resched();
4895 } while_for_each_ftrace_rec();
4896
4897 return found;
4898 }
4899
4900 static int
ftrace_match_records(struct ftrace_hash * hash,char * buff,int len)4901 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
4902 {
4903 return match_records(hash, buff, len, NULL);
4904 }
4905
ftrace_ops_update_code(struct ftrace_ops * ops,struct ftrace_ops_hash * old_hash)4906 static void ftrace_ops_update_code(struct ftrace_ops *ops,
4907 struct ftrace_ops_hash *old_hash)
4908 {
4909 struct ftrace_ops *op;
4910
4911 if (!ftrace_enabled)
4912 return;
4913
4914 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
4915 ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
4916 return;
4917 }
4918
4919 /*
4920 * If this is the shared global_ops filter, then we need to
4921 * check if there is another ops that shares it, is enabled.
4922 * If so, we still need to run the modify code.
4923 */
4924 if (ops->func_hash != &global_ops.local_hash)
4925 return;
4926
4927 do_for_each_ftrace_op(op, ftrace_ops_list) {
4928 if (op->func_hash == &global_ops.local_hash &&
4929 op->flags & FTRACE_OPS_FL_ENABLED) {
4930 ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
4931 /* Only need to do this once */
4932 return;
4933 }
4934 } while_for_each_ftrace_op(op);
4935 }
4936
ftrace_hash_move_and_update_ops(struct ftrace_ops * ops,struct ftrace_hash ** orig_hash,struct ftrace_hash * hash,int enable)4937 static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
4938 struct ftrace_hash **orig_hash,
4939 struct ftrace_hash *hash,
4940 int enable)
4941 {
4942 if (ops->flags & FTRACE_OPS_FL_SUBOP)
4943 return ftrace_hash_move_and_update_subops(ops, orig_hash, hash);
4944
4945 /*
4946 * If this ops is not enabled, it could be sharing its filters
4947 * with a subop. If that's the case, update the subop instead of
4948 * this ops. Shared filters are only allowed to have one ops set
4949 * at a time, and if we update the ops that is not enabled,
4950 * it will not affect subops that share it.
4951 */
4952 if (!(ops->flags & FTRACE_OPS_FL_ENABLED)) {
4953 struct ftrace_ops *op;
4954
4955 /* Check if any other manager subops maps to this hash */
4956 do_for_each_ftrace_op(op, ftrace_ops_list) {
4957 struct ftrace_ops *subops;
4958
4959 list_for_each_entry(subops, &op->subop_list, list) {
4960 if ((subops->flags & FTRACE_OPS_FL_ENABLED) &&
4961 subops->func_hash == ops->func_hash) {
4962 return ftrace_hash_move_and_update_subops(subops, orig_hash, hash);
4963 }
4964 }
4965 } while_for_each_ftrace_op(op);
4966 }
4967
4968 return __ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
4969 }
4970
cache_mod(struct trace_array * tr,const char * func,char * module,int enable)4971 static int cache_mod(struct trace_array *tr,
4972 const char *func, char *module, int enable)
4973 {
4974 struct ftrace_mod_load *ftrace_mod, *n;
4975 struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
4976
4977 guard(mutex)(&ftrace_lock);
4978
4979 /* We do not cache inverse filters */
4980 if (func[0] == '!') {
4981 int ret = -EINVAL;
4982
4983 func++;
4984
4985 /* Look to remove this hash */
4986 list_for_each_entry_safe(ftrace_mod, n, head, list) {
4987 if (strcmp(ftrace_mod->module, module) != 0)
4988 continue;
4989
4990 /* no func matches all */
4991 if (strcmp(func, "*") == 0 ||
4992 (ftrace_mod->func &&
4993 strcmp(ftrace_mod->func, func) == 0)) {
4994 ret = 0;
4995 free_ftrace_mod(ftrace_mod);
4996 continue;
4997 }
4998 }
4999 return ret;
5000 }
5001
5002 /* We only care about modules that have not been loaded yet */
5003 if (module_exists(module))
5004 return -EINVAL;
5005
5006 /* Save this string off, and execute it when the module is loaded */
5007 return ftrace_add_mod(tr, func, module, enable);
5008 }
5009
5010 #ifdef CONFIG_MODULES
process_mod_list(struct list_head * head,struct ftrace_ops * ops,char * mod,bool enable)5011 static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
5012 char *mod, bool enable)
5013 {
5014 struct ftrace_mod_load *ftrace_mod, *n;
5015 struct ftrace_hash **orig_hash, *new_hash;
5016 LIST_HEAD(process_mods);
5017 char *func;
5018
5019 mutex_lock(&ops->func_hash->regex_lock);
5020
5021 if (enable)
5022 orig_hash = &ops->func_hash->filter_hash;
5023 else
5024 orig_hash = &ops->func_hash->notrace_hash;
5025
5026 new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
5027 *orig_hash);
5028 if (!new_hash)
5029 goto out; /* warn? */
5030
5031 mutex_lock(&ftrace_lock);
5032
5033 list_for_each_entry_safe(ftrace_mod, n, head, list) {
5034
5035 if (strcmp(ftrace_mod->module, mod) != 0)
5036 continue;
5037
5038 if (ftrace_mod->func)
5039 func = kstrdup(ftrace_mod->func, GFP_KERNEL);
5040 else
5041 func = kstrdup("*", GFP_KERNEL);
5042
5043 if (!func) /* warn? */
5044 continue;
5045
5046 list_move(&ftrace_mod->list, &process_mods);
5047
5048 /* Use the newly allocated func, as it may be "*" */
5049 kfree(ftrace_mod->func);
5050 ftrace_mod->func = func;
5051 }
5052
5053 mutex_unlock(&ftrace_lock);
5054
5055 list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
5056
5057 func = ftrace_mod->func;
5058
5059 /* Grabs ftrace_lock, which is why we have this extra step */
5060 match_records(new_hash, func, strlen(func), mod);
5061 free_ftrace_mod(ftrace_mod);
5062 }
5063
5064 if (enable && list_empty(head))
5065 new_hash->flags &= ~FTRACE_HASH_FL_MOD;
5066
5067 mutex_lock(&ftrace_lock);
5068
5069 ftrace_hash_move_and_update_ops(ops, orig_hash,
5070 new_hash, enable);
5071 mutex_unlock(&ftrace_lock);
5072
5073 out:
5074 mutex_unlock(&ops->func_hash->regex_lock);
5075
5076 free_ftrace_hash(new_hash);
5077 }
5078
process_cached_mods(const char * mod_name)5079 static void process_cached_mods(const char *mod_name)
5080 {
5081 struct trace_array *tr;
5082 char *mod;
5083
5084 mod = kstrdup(mod_name, GFP_KERNEL);
5085 if (!mod)
5086 return;
5087
5088 mutex_lock(&trace_types_lock);
5089 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5090 if (!list_empty(&tr->mod_trace))
5091 process_mod_list(&tr->mod_trace, tr->ops, mod, true);
5092 if (!list_empty(&tr->mod_notrace))
5093 process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
5094 }
5095 mutex_unlock(&trace_types_lock);
5096
5097 kfree(mod);
5098 }
5099 #endif
5100
5101 /*
5102 * We register the module command as a template to show others how
5103 * to register the a command as well.
5104 */
5105
5106 static int
ftrace_mod_callback(struct trace_array * tr,struct ftrace_hash * hash,char * func_orig,char * cmd,char * module,int enable)5107 ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
5108 char *func_orig, char *cmd, char *module, int enable)
5109 {
5110 char *func;
5111 int ret;
5112
5113 if (!tr)
5114 return -ENODEV;
5115
5116 /* match_records() modifies func, and we need the original */
5117 func = kstrdup(func_orig, GFP_KERNEL);
5118 if (!func)
5119 return -ENOMEM;
5120
5121 /*
5122 * cmd == 'mod' because we only registered this func
5123 * for the 'mod' ftrace_func_command.
5124 * But if you register one func with multiple commands,
5125 * you can tell which command was used by the cmd
5126 * parameter.
5127 */
5128 ret = match_records(hash, func, strlen(func), module);
5129 kfree(func);
5130
5131 if (!ret)
5132 return cache_mod(tr, func_orig, module, enable);
5133 if (ret < 0)
5134 return ret;
5135 return 0;
5136 }
5137
5138 static struct ftrace_func_command ftrace_mod_cmd = {
5139 .name = "mod",
5140 .func = ftrace_mod_callback,
5141 };
5142
ftrace_mod_cmd_init(void)5143 static int __init ftrace_mod_cmd_init(void)
5144 {
5145 return register_ftrace_command(&ftrace_mod_cmd);
5146 }
5147 core_initcall(ftrace_mod_cmd_init);
5148
function_trace_probe_call(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)5149 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
5150 struct ftrace_ops *op, struct ftrace_regs *fregs)
5151 {
5152 struct ftrace_probe_ops *probe_ops;
5153 struct ftrace_func_probe *probe;
5154
5155 probe = container_of(op, struct ftrace_func_probe, ops);
5156 probe_ops = probe->probe_ops;
5157
5158 /*
5159 * Disable preemption for these calls to prevent a RCU grace
5160 * period. This syncs the hash iteration and freeing of items
5161 * on the hash. rcu_read_lock is too dangerous here.
5162 */
5163 preempt_disable_notrace();
5164 probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
5165 preempt_enable_notrace();
5166 }
5167
5168 struct ftrace_func_map {
5169 struct ftrace_func_entry entry;
5170 void *data;
5171 };
5172
5173 struct ftrace_func_mapper {
5174 struct ftrace_hash hash;
5175 };
5176
5177 /**
5178 * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
5179 *
5180 * Returns: a ftrace_func_mapper descriptor that can be used to map ips to data.
5181 */
allocate_ftrace_func_mapper(void)5182 struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
5183 {
5184 struct ftrace_hash *hash;
5185
5186 /*
5187 * The mapper is simply a ftrace_hash, but since the entries
5188 * in the hash are not ftrace_func_entry type, we define it
5189 * as a separate structure.
5190 */
5191 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5192 return (struct ftrace_func_mapper *)hash;
5193 }
5194
5195 /**
5196 * ftrace_func_mapper_find_ip - Find some data mapped to an ip
5197 * @mapper: The mapper that has the ip maps
5198 * @ip: the instruction pointer to find the data for
5199 *
5200 * Returns: the data mapped to @ip if found otherwise NULL. The return
5201 * is actually the address of the mapper data pointer. The address is
5202 * returned for use cases where the data is no bigger than a long, and
5203 * the user can use the data pointer as its data instead of having to
5204 * allocate more memory for the reference.
5205 */
ftrace_func_mapper_find_ip(struct ftrace_func_mapper * mapper,unsigned long ip)5206 void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
5207 unsigned long ip)
5208 {
5209 struct ftrace_func_entry *entry;
5210 struct ftrace_func_map *map;
5211
5212 entry = ftrace_lookup_ip(&mapper->hash, ip);
5213 if (!entry)
5214 return NULL;
5215
5216 map = (struct ftrace_func_map *)entry;
5217 return &map->data;
5218 }
5219
5220 /**
5221 * ftrace_func_mapper_add_ip - Map some data to an ip
5222 * @mapper: The mapper that has the ip maps
5223 * @ip: The instruction pointer address to map @data to
5224 * @data: The data to map to @ip
5225 *
5226 * Returns: 0 on success otherwise an error.
5227 */
ftrace_func_mapper_add_ip(struct ftrace_func_mapper * mapper,unsigned long ip,void * data)5228 int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
5229 unsigned long ip, void *data)
5230 {
5231 struct ftrace_func_entry *entry;
5232 struct ftrace_func_map *map;
5233
5234 entry = ftrace_lookup_ip(&mapper->hash, ip);
5235 if (entry)
5236 return -EBUSY;
5237
5238 map = kmalloc(sizeof(*map), GFP_KERNEL);
5239 if (!map)
5240 return -ENOMEM;
5241
5242 map->entry.ip = ip;
5243 map->data = data;
5244
5245 __add_hash_entry(&mapper->hash, &map->entry);
5246
5247 return 0;
5248 }
5249
5250 /**
5251 * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
5252 * @mapper: The mapper that has the ip maps
5253 * @ip: The instruction pointer address to remove the data from
5254 *
5255 * Returns: the data if it is found, otherwise NULL.
5256 * Note, if the data pointer is used as the data itself, (see
5257 * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
5258 * if the data pointer was set to zero.
5259 */
ftrace_func_mapper_remove_ip(struct ftrace_func_mapper * mapper,unsigned long ip)5260 void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
5261 unsigned long ip)
5262 {
5263 struct ftrace_func_entry *entry;
5264 struct ftrace_func_map *map;
5265 void *data;
5266
5267 entry = ftrace_lookup_ip(&mapper->hash, ip);
5268 if (!entry)
5269 return NULL;
5270
5271 map = (struct ftrace_func_map *)entry;
5272 data = map->data;
5273
5274 remove_hash_entry(&mapper->hash, entry);
5275 kfree(entry);
5276
5277 return data;
5278 }
5279
5280 /**
5281 * free_ftrace_func_mapper - free a mapping of ips and data
5282 * @mapper: The mapper that has the ip maps
5283 * @free_func: A function to be called on each data item.
5284 *
5285 * This is used to free the function mapper. The @free_func is optional
5286 * and can be used if the data needs to be freed as well.
5287 */
free_ftrace_func_mapper(struct ftrace_func_mapper * mapper,ftrace_mapper_func free_func)5288 void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
5289 ftrace_mapper_func free_func)
5290 {
5291 struct ftrace_func_entry *entry;
5292 struct ftrace_func_map *map;
5293 struct hlist_head *hhd;
5294 int size, i;
5295
5296 if (!mapper)
5297 return;
5298
5299 if (free_func && mapper->hash.count) {
5300 size = 1 << mapper->hash.size_bits;
5301 for (i = 0; i < size; i++) {
5302 hhd = &mapper->hash.buckets[i];
5303 hlist_for_each_entry(entry, hhd, hlist) {
5304 map = (struct ftrace_func_map *)entry;
5305 free_func(map);
5306 }
5307 }
5308 }
5309 free_ftrace_hash(&mapper->hash);
5310 }
5311
release_probe(struct ftrace_func_probe * probe)5312 static void release_probe(struct ftrace_func_probe *probe)
5313 {
5314 struct ftrace_probe_ops *probe_ops;
5315
5316 guard(mutex)(&ftrace_lock);
5317
5318 WARN_ON(probe->ref <= 0);
5319
5320 /* Subtract the ref that was used to protect this instance */
5321 probe->ref--;
5322
5323 if (!probe->ref) {
5324 probe_ops = probe->probe_ops;
5325 /*
5326 * Sending zero as ip tells probe_ops to free
5327 * the probe->data itself
5328 */
5329 if (probe_ops->free)
5330 probe_ops->free(probe_ops, probe->tr, 0, probe->data);
5331 list_del(&probe->list);
5332 kfree(probe);
5333 }
5334 }
5335
acquire_probe_locked(struct ftrace_func_probe * probe)5336 static void acquire_probe_locked(struct ftrace_func_probe *probe)
5337 {
5338 /*
5339 * Add one ref to keep it from being freed when releasing the
5340 * ftrace_lock mutex.
5341 */
5342 probe->ref++;
5343 }
5344
5345 int
register_ftrace_function_probe(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops,void * data)5346 register_ftrace_function_probe(char *glob, struct trace_array *tr,
5347 struct ftrace_probe_ops *probe_ops,
5348 void *data)
5349 {
5350 struct ftrace_func_probe *probe = NULL, *iter;
5351 struct ftrace_func_entry *entry;
5352 struct ftrace_hash **orig_hash;
5353 struct ftrace_hash *old_hash;
5354 struct ftrace_hash *hash;
5355 int count = 0;
5356 int size;
5357 int ret;
5358 int i;
5359
5360 if (WARN_ON(!tr))
5361 return -EINVAL;
5362
5363 /* We do not support '!' for function probes */
5364 if (WARN_ON(glob[0] == '!'))
5365 return -EINVAL;
5366
5367
5368 mutex_lock(&ftrace_lock);
5369 /* Check if the probe_ops is already registered */
5370 list_for_each_entry(iter, &tr->func_probes, list) {
5371 if (iter->probe_ops == probe_ops) {
5372 probe = iter;
5373 break;
5374 }
5375 }
5376 if (!probe) {
5377 probe = kzalloc(sizeof(*probe), GFP_KERNEL);
5378 if (!probe) {
5379 mutex_unlock(&ftrace_lock);
5380 return -ENOMEM;
5381 }
5382 probe->probe_ops = probe_ops;
5383 probe->ops.func = function_trace_probe_call;
5384 probe->tr = tr;
5385 ftrace_ops_init(&probe->ops);
5386 list_add(&probe->list, &tr->func_probes);
5387 }
5388
5389 acquire_probe_locked(probe);
5390
5391 mutex_unlock(&ftrace_lock);
5392
5393 /*
5394 * Note, there's a small window here that the func_hash->filter_hash
5395 * may be NULL or empty. Need to be careful when reading the loop.
5396 */
5397 mutex_lock(&probe->ops.func_hash->regex_lock);
5398
5399 orig_hash = &probe->ops.func_hash->filter_hash;
5400 old_hash = *orig_hash;
5401 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
5402
5403 if (!hash) {
5404 ret = -ENOMEM;
5405 goto out;
5406 }
5407
5408 ret = ftrace_match_records(hash, glob, strlen(glob));
5409
5410 /* Nothing found? */
5411 if (!ret)
5412 ret = -EINVAL;
5413
5414 if (ret < 0)
5415 goto out;
5416
5417 size = 1 << hash->size_bits;
5418 for (i = 0; i < size; i++) {
5419 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5420 if (ftrace_lookup_ip(old_hash, entry->ip))
5421 continue;
5422 /*
5423 * The caller might want to do something special
5424 * for each function we find. We call the callback
5425 * to give the caller an opportunity to do so.
5426 */
5427 if (probe_ops->init) {
5428 ret = probe_ops->init(probe_ops, tr,
5429 entry->ip, data,
5430 &probe->data);
5431 if (ret < 0) {
5432 if (probe_ops->free && count)
5433 probe_ops->free(probe_ops, tr,
5434 0, probe->data);
5435 probe->data = NULL;
5436 goto out;
5437 }
5438 }
5439 count++;
5440 }
5441 }
5442
5443 mutex_lock(&ftrace_lock);
5444
5445 if (!count) {
5446 /* Nothing was added? */
5447 ret = -EINVAL;
5448 goto out_unlock;
5449 }
5450
5451 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
5452 hash, 1);
5453 if (ret < 0)
5454 goto err_unlock;
5455
5456 /* One ref for each new function traced */
5457 probe->ref += count;
5458
5459 if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
5460 ret = ftrace_startup(&probe->ops, 0);
5461
5462 out_unlock:
5463 mutex_unlock(&ftrace_lock);
5464
5465 if (!ret)
5466 ret = count;
5467 out:
5468 mutex_unlock(&probe->ops.func_hash->regex_lock);
5469 free_ftrace_hash(hash);
5470
5471 release_probe(probe);
5472
5473 return ret;
5474
5475 err_unlock:
5476 if (!probe_ops->free || !count)
5477 goto out_unlock;
5478
5479 /* Failed to do the move, need to call the free functions */
5480 for (i = 0; i < size; i++) {
5481 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5482 if (ftrace_lookup_ip(old_hash, entry->ip))
5483 continue;
5484 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
5485 }
5486 }
5487 goto out_unlock;
5488 }
5489
5490 int
unregister_ftrace_function_probe_func(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops)5491 unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
5492 struct ftrace_probe_ops *probe_ops)
5493 {
5494 struct ftrace_func_probe *probe = NULL, *iter;
5495 struct ftrace_ops_hash old_hash_ops;
5496 struct ftrace_func_entry *entry;
5497 struct ftrace_glob func_g;
5498 struct ftrace_hash **orig_hash;
5499 struct ftrace_hash *old_hash;
5500 struct ftrace_hash *hash = NULL;
5501 struct hlist_node *tmp;
5502 struct hlist_head hhd;
5503 char str[KSYM_SYMBOL_LEN];
5504 int count = 0;
5505 int i, ret = -ENODEV;
5506 int size;
5507
5508 if (!glob || !strlen(glob) || !strcmp(glob, "*"))
5509 func_g.search = NULL;
5510 else {
5511 int not;
5512
5513 func_g.type = filter_parse_regex(glob, strlen(glob),
5514 &func_g.search, ¬);
5515 func_g.len = strlen(func_g.search);
5516
5517 /* we do not support '!' for function probes */
5518 if (WARN_ON(not))
5519 return -EINVAL;
5520 }
5521
5522 mutex_lock(&ftrace_lock);
5523 /* Check if the probe_ops is already registered */
5524 list_for_each_entry(iter, &tr->func_probes, list) {
5525 if (iter->probe_ops == probe_ops) {
5526 probe = iter;
5527 break;
5528 }
5529 }
5530 if (!probe)
5531 goto err_unlock_ftrace;
5532
5533 ret = -EINVAL;
5534 if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
5535 goto err_unlock_ftrace;
5536
5537 acquire_probe_locked(probe);
5538
5539 mutex_unlock(&ftrace_lock);
5540
5541 mutex_lock(&probe->ops.func_hash->regex_lock);
5542
5543 orig_hash = &probe->ops.func_hash->filter_hash;
5544 old_hash = *orig_hash;
5545
5546 if (ftrace_hash_empty(old_hash))
5547 goto out_unlock;
5548
5549 old_hash_ops.filter_hash = old_hash;
5550 /* Probes only have filters */
5551 old_hash_ops.notrace_hash = NULL;
5552
5553 ret = -ENOMEM;
5554 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
5555 if (!hash)
5556 goto out_unlock;
5557
5558 INIT_HLIST_HEAD(&hhd);
5559
5560 size = 1 << hash->size_bits;
5561 for (i = 0; i < size; i++) {
5562 hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
5563
5564 if (func_g.search) {
5565 kallsyms_lookup(entry->ip, NULL, NULL,
5566 NULL, str);
5567 if (!ftrace_match(str, &func_g))
5568 continue;
5569 }
5570 count++;
5571 remove_hash_entry(hash, entry);
5572 hlist_add_head(&entry->hlist, &hhd);
5573 }
5574 }
5575
5576 /* Nothing found? */
5577 if (!count) {
5578 ret = -EINVAL;
5579 goto out_unlock;
5580 }
5581
5582 mutex_lock(&ftrace_lock);
5583
5584 WARN_ON(probe->ref < count);
5585
5586 probe->ref -= count;
5587
5588 if (ftrace_hash_empty(hash))
5589 ftrace_shutdown(&probe->ops, 0);
5590
5591 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
5592 hash, 1);
5593
5594 /* still need to update the function call sites */
5595 if (ftrace_enabled && !ftrace_hash_empty(hash))
5596 ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
5597 &old_hash_ops);
5598 synchronize_rcu();
5599
5600 hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
5601 hlist_del(&entry->hlist);
5602 if (probe_ops->free)
5603 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
5604 kfree(entry);
5605 }
5606 mutex_unlock(&ftrace_lock);
5607
5608 out_unlock:
5609 mutex_unlock(&probe->ops.func_hash->regex_lock);
5610 free_ftrace_hash(hash);
5611
5612 release_probe(probe);
5613
5614 return ret;
5615
5616 err_unlock_ftrace:
5617 mutex_unlock(&ftrace_lock);
5618 return ret;
5619 }
5620
clear_ftrace_function_probes(struct trace_array * tr)5621 void clear_ftrace_function_probes(struct trace_array *tr)
5622 {
5623 struct ftrace_func_probe *probe, *n;
5624
5625 list_for_each_entry_safe(probe, n, &tr->func_probes, list)
5626 unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
5627 }
5628
5629 static LIST_HEAD(ftrace_commands);
5630 static DEFINE_MUTEX(ftrace_cmd_mutex);
5631
5632 /*
5633 * Currently we only register ftrace commands from __init, so mark this
5634 * __init too.
5635 */
register_ftrace_command(struct ftrace_func_command * cmd)5636 __init int register_ftrace_command(struct ftrace_func_command *cmd)
5637 {
5638 struct ftrace_func_command *p;
5639
5640 guard(mutex)(&ftrace_cmd_mutex);
5641 list_for_each_entry(p, &ftrace_commands, list) {
5642 if (strcmp(cmd->name, p->name) == 0)
5643 return -EBUSY;
5644 }
5645 list_add(&cmd->list, &ftrace_commands);
5646
5647 return 0;
5648 }
5649
5650 /*
5651 * Currently we only unregister ftrace commands from __init, so mark
5652 * this __init too.
5653 */
unregister_ftrace_command(struct ftrace_func_command * cmd)5654 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
5655 {
5656 struct ftrace_func_command *p, *n;
5657
5658 guard(mutex)(&ftrace_cmd_mutex);
5659
5660 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
5661 if (strcmp(cmd->name, p->name) == 0) {
5662 list_del_init(&p->list);
5663 return 0;
5664 }
5665 }
5666
5667 return -ENODEV;
5668 }
5669
ftrace_process_regex(struct ftrace_iterator * iter,char * buff,int len,int enable)5670 static int ftrace_process_regex(struct ftrace_iterator *iter,
5671 char *buff, int len, int enable)
5672 {
5673 struct ftrace_hash *hash = iter->hash;
5674 struct trace_array *tr = iter->ops->private;
5675 char *func, *command, *next = buff;
5676 struct ftrace_func_command *p;
5677 int ret;
5678
5679 func = strsep(&next, ":");
5680
5681 if (!next) {
5682 ret = ftrace_match_records(hash, func, len);
5683 if (!ret)
5684 ret = -EINVAL;
5685 if (ret < 0)
5686 return ret;
5687 return 0;
5688 }
5689
5690 /* command found */
5691
5692 command = strsep(&next, ":");
5693
5694 guard(mutex)(&ftrace_cmd_mutex);
5695
5696 list_for_each_entry(p, &ftrace_commands, list) {
5697 if (strcmp(p->name, command) == 0)
5698 return p->func(tr, hash, func, command, next, enable);
5699 }
5700
5701 return -EINVAL;
5702 }
5703
5704 static ssize_t
ftrace_regex_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos,int enable)5705 ftrace_regex_write(struct file *file, const char __user *ubuf,
5706 size_t cnt, loff_t *ppos, int enable)
5707 {
5708 struct ftrace_iterator *iter;
5709 struct trace_parser *parser;
5710 ssize_t ret, read;
5711
5712 if (!cnt)
5713 return 0;
5714
5715 if (file->f_mode & FMODE_READ) {
5716 struct seq_file *m = file->private_data;
5717 iter = m->private;
5718 } else
5719 iter = file->private_data;
5720
5721 if (unlikely(ftrace_disabled))
5722 return -ENODEV;
5723
5724 /* iter->hash is a local copy, so we don't need regex_lock */
5725
5726 parser = &iter->parser;
5727 read = trace_get_user(parser, ubuf, cnt, ppos);
5728
5729 if (read >= 0 && trace_parser_loaded(parser) &&
5730 !trace_parser_cont(parser)) {
5731 ret = ftrace_process_regex(iter, parser->buffer,
5732 parser->idx, enable);
5733 trace_parser_clear(parser);
5734 if (ret < 0)
5735 return ret;
5736 }
5737
5738 return read;
5739 }
5740
5741 ssize_t
ftrace_filter_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5742 ftrace_filter_write(struct file *file, const char __user *ubuf,
5743 size_t cnt, loff_t *ppos)
5744 {
5745 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
5746 }
5747
5748 ssize_t
ftrace_notrace_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5749 ftrace_notrace_write(struct file *file, const char __user *ubuf,
5750 size_t cnt, loff_t *ppos)
5751 {
5752 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
5753 }
5754
5755 static int
__ftrace_match_addr(struct ftrace_hash * hash,unsigned long ip,int remove)5756 __ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
5757 {
5758 struct ftrace_func_entry *entry;
5759
5760 ip = ftrace_location(ip);
5761 if (!ip)
5762 return -EINVAL;
5763
5764 if (remove) {
5765 entry = ftrace_lookup_ip(hash, ip);
5766 if (!entry)
5767 return -ENOENT;
5768 free_hash_entry(hash, entry);
5769 return 0;
5770 } else if (__ftrace_lookup_ip(hash, ip) != NULL) {
5771 /* Already exists */
5772 return 0;
5773 }
5774
5775 entry = add_hash_entry(hash, ip);
5776 return entry ? 0 : -ENOMEM;
5777 }
5778
5779 static int
ftrace_match_addr(struct ftrace_hash * hash,unsigned long * ips,unsigned int cnt,int remove)5780 ftrace_match_addr(struct ftrace_hash *hash, unsigned long *ips,
5781 unsigned int cnt, int remove)
5782 {
5783 unsigned int i;
5784 int err;
5785
5786 for (i = 0; i < cnt; i++) {
5787 err = __ftrace_match_addr(hash, ips[i], remove);
5788 if (err) {
5789 /*
5790 * This expects the @hash is a temporary hash and if this
5791 * fails the caller must free the @hash.
5792 */
5793 return err;
5794 }
5795 }
5796 return 0;
5797 }
5798
5799 static int
ftrace_set_hash(struct ftrace_ops * ops,unsigned char * buf,int len,unsigned long * ips,unsigned int cnt,int remove,int reset,int enable,char * mod)5800 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
5801 unsigned long *ips, unsigned int cnt,
5802 int remove, int reset, int enable, char *mod)
5803 {
5804 struct ftrace_hash **orig_hash;
5805 struct ftrace_hash *hash;
5806 int ret;
5807
5808 if (unlikely(ftrace_disabled))
5809 return -ENODEV;
5810
5811 mutex_lock(&ops->func_hash->regex_lock);
5812
5813 if (enable)
5814 orig_hash = &ops->func_hash->filter_hash;
5815 else
5816 orig_hash = &ops->func_hash->notrace_hash;
5817
5818 if (reset)
5819 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5820 else
5821 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
5822
5823 if (!hash) {
5824 ret = -ENOMEM;
5825 goto out_regex_unlock;
5826 }
5827
5828 if (buf && !match_records(hash, buf, len, mod)) {
5829 /* If this was for a module and nothing was enabled, flag it */
5830 if (mod)
5831 (*orig_hash)->flags |= FTRACE_HASH_FL_MOD;
5832
5833 /*
5834 * Even if it is a mod, return error to let caller know
5835 * nothing was added
5836 */
5837 ret = -EINVAL;
5838 goto out_regex_unlock;
5839 }
5840 if (ips) {
5841 ret = ftrace_match_addr(hash, ips, cnt, remove);
5842 if (ret < 0)
5843 goto out_regex_unlock;
5844 }
5845
5846 mutex_lock(&ftrace_lock);
5847 ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
5848 mutex_unlock(&ftrace_lock);
5849
5850 out_regex_unlock:
5851 mutex_unlock(&ops->func_hash->regex_lock);
5852
5853 free_ftrace_hash(hash);
5854 return ret;
5855 }
5856
5857 static int
ftrace_set_addr(struct ftrace_ops * ops,unsigned long * ips,unsigned int cnt,int remove,int reset,int enable)5858 ftrace_set_addr(struct ftrace_ops *ops, unsigned long *ips, unsigned int cnt,
5859 int remove, int reset, int enable)
5860 {
5861 return ftrace_set_hash(ops, NULL, 0, ips, cnt, remove, reset, enable, NULL);
5862 }
5863
5864 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
5865
5866 static int register_ftrace_function_nolock(struct ftrace_ops *ops);
5867
5868 /*
5869 * If there are multiple ftrace_ops, use SAVE_REGS by default, so that direct
5870 * call will be jumped from ftrace_regs_caller. Only if the architecture does
5871 * not support ftrace_regs_caller but direct_call, use SAVE_ARGS so that it
5872 * jumps from ftrace_caller for multiple ftrace_ops.
5873 */
5874 #ifndef CONFIG_HAVE_DYNAMIC_FTRACE_WITH_REGS
5875 #define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_ARGS)
5876 #else
5877 #define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_REGS)
5878 #endif
5879
check_direct_multi(struct ftrace_ops * ops)5880 static int check_direct_multi(struct ftrace_ops *ops)
5881 {
5882 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED))
5883 return -EINVAL;
5884 if ((ops->flags & MULTI_FLAGS) != MULTI_FLAGS)
5885 return -EINVAL;
5886 return 0;
5887 }
5888
remove_direct_functions_hash(struct ftrace_hash * hash,unsigned long addr)5889 static void remove_direct_functions_hash(struct ftrace_hash *hash, unsigned long addr)
5890 {
5891 struct ftrace_func_entry *entry, *del;
5892 int size, i;
5893
5894 size = 1 << hash->size_bits;
5895 for (i = 0; i < size; i++) {
5896 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5897 del = __ftrace_lookup_ip(direct_functions, entry->ip);
5898 if (del && del->direct == addr) {
5899 remove_hash_entry(direct_functions, del);
5900 kfree(del);
5901 }
5902 }
5903 }
5904 }
5905
register_ftrace_direct_cb(struct rcu_head * rhp)5906 static void register_ftrace_direct_cb(struct rcu_head *rhp)
5907 {
5908 struct ftrace_hash *fhp = container_of(rhp, struct ftrace_hash, rcu);
5909
5910 free_ftrace_hash(fhp);
5911 }
5912
5913 /**
5914 * register_ftrace_direct - Call a custom trampoline directly
5915 * for multiple functions registered in @ops
5916 * @ops: The address of the struct ftrace_ops object
5917 * @addr: The address of the trampoline to call at @ops functions
5918 *
5919 * This is used to connect a direct calls to @addr from the nop locations
5920 * of the functions registered in @ops (with by ftrace_set_filter_ip
5921 * function).
5922 *
5923 * The location that it calls (@addr) must be able to handle a direct call,
5924 * and save the parameters of the function being traced, and restore them
5925 * (or inject new ones if needed), before returning.
5926 *
5927 * Returns:
5928 * 0 on success
5929 * -EINVAL - The @ops object was already registered with this call or
5930 * when there are no functions in @ops object.
5931 * -EBUSY - Another direct function is already attached (there can be only one)
5932 * -ENODEV - @ip does not point to a ftrace nop location (or not supported)
5933 * -ENOMEM - There was an allocation failure.
5934 */
register_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)5935 int register_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
5936 {
5937 struct ftrace_hash *hash, *new_hash = NULL, *free_hash = NULL;
5938 struct ftrace_func_entry *entry, *new;
5939 int err = -EBUSY, size, i;
5940
5941 if (ops->func || ops->trampoline)
5942 return -EINVAL;
5943 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED))
5944 return -EINVAL;
5945 if (ops->flags & FTRACE_OPS_FL_ENABLED)
5946 return -EINVAL;
5947
5948 hash = ops->func_hash->filter_hash;
5949 if (ftrace_hash_empty(hash))
5950 return -EINVAL;
5951
5952 mutex_lock(&direct_mutex);
5953
5954 /* Make sure requested entries are not already registered.. */
5955 size = 1 << hash->size_bits;
5956 for (i = 0; i < size; i++) {
5957 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5958 if (ftrace_find_rec_direct(entry->ip))
5959 goto out_unlock;
5960 }
5961 }
5962
5963 err = -ENOMEM;
5964
5965 /* Make a copy hash to place the new and the old entries in */
5966 size = hash->count + direct_functions->count;
5967 size = fls(size);
5968 if (size > FTRACE_HASH_MAX_BITS)
5969 size = FTRACE_HASH_MAX_BITS;
5970 new_hash = alloc_ftrace_hash(size);
5971 if (!new_hash)
5972 goto out_unlock;
5973
5974 /* Now copy over the existing direct entries */
5975 size = 1 << direct_functions->size_bits;
5976 for (i = 0; i < size; i++) {
5977 hlist_for_each_entry(entry, &direct_functions->buckets[i], hlist) {
5978 new = add_hash_entry(new_hash, entry->ip);
5979 if (!new)
5980 goto out_unlock;
5981 new->direct = entry->direct;
5982 }
5983 }
5984
5985 /* ... and add the new entries */
5986 size = 1 << hash->size_bits;
5987 for (i = 0; i < size; i++) {
5988 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5989 new = add_hash_entry(new_hash, entry->ip);
5990 if (!new)
5991 goto out_unlock;
5992 /* Update both the copy and the hash entry */
5993 new->direct = addr;
5994 entry->direct = addr;
5995 }
5996 }
5997
5998 free_hash = direct_functions;
5999 rcu_assign_pointer(direct_functions, new_hash);
6000 new_hash = NULL;
6001
6002 ops->func = call_direct_funcs;
6003 ops->flags = MULTI_FLAGS;
6004 ops->trampoline = FTRACE_REGS_ADDR;
6005 ops->direct_call = addr;
6006
6007 err = register_ftrace_function_nolock(ops);
6008
6009 out_unlock:
6010 mutex_unlock(&direct_mutex);
6011
6012 if (free_hash && free_hash != EMPTY_HASH)
6013 call_rcu_tasks(&free_hash->rcu, register_ftrace_direct_cb);
6014
6015 if (new_hash)
6016 free_ftrace_hash(new_hash);
6017
6018 return err;
6019 }
6020 EXPORT_SYMBOL_GPL(register_ftrace_direct);
6021
6022 /**
6023 * unregister_ftrace_direct - Remove calls to custom trampoline
6024 * previously registered by register_ftrace_direct for @ops object.
6025 * @ops: The address of the struct ftrace_ops object
6026 * @addr: The address of the direct function that is called by the @ops functions
6027 * @free_filters: Set to true to remove all filters for the ftrace_ops, false otherwise
6028 *
6029 * This is used to remove a direct calls to @addr from the nop locations
6030 * of the functions registered in @ops (with by ftrace_set_filter_ip
6031 * function).
6032 *
6033 * Returns:
6034 * 0 on success
6035 * -EINVAL - The @ops object was not properly registered.
6036 */
unregister_ftrace_direct(struct ftrace_ops * ops,unsigned long addr,bool free_filters)6037 int unregister_ftrace_direct(struct ftrace_ops *ops, unsigned long addr,
6038 bool free_filters)
6039 {
6040 struct ftrace_hash *hash = ops->func_hash->filter_hash;
6041 int err;
6042
6043 if (check_direct_multi(ops))
6044 return -EINVAL;
6045 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6046 return -EINVAL;
6047
6048 mutex_lock(&direct_mutex);
6049 err = unregister_ftrace_function(ops);
6050 remove_direct_functions_hash(hash, addr);
6051 mutex_unlock(&direct_mutex);
6052
6053 /* cleanup for possible another register call */
6054 ops->func = NULL;
6055 ops->trampoline = 0;
6056
6057 if (free_filters)
6058 ftrace_free_filter(ops);
6059 return err;
6060 }
6061 EXPORT_SYMBOL_GPL(unregister_ftrace_direct);
6062
6063 static int
__modify_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)6064 __modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
6065 {
6066 struct ftrace_hash *hash;
6067 struct ftrace_func_entry *entry, *iter;
6068 static struct ftrace_ops tmp_ops = {
6069 .func = ftrace_stub,
6070 .flags = FTRACE_OPS_FL_STUB,
6071 };
6072 int i, size;
6073 int err;
6074
6075 lockdep_assert_held_once(&direct_mutex);
6076
6077 /* Enable the tmp_ops to have the same functions as the direct ops */
6078 ftrace_ops_init(&tmp_ops);
6079 tmp_ops.func_hash = ops->func_hash;
6080 tmp_ops.direct_call = addr;
6081
6082 err = register_ftrace_function_nolock(&tmp_ops);
6083 if (err)
6084 return err;
6085
6086 /*
6087 * Now the ftrace_ops_list_func() is called to do the direct callers.
6088 * We can safely change the direct functions attached to each entry.
6089 */
6090 mutex_lock(&ftrace_lock);
6091
6092 hash = ops->func_hash->filter_hash;
6093 size = 1 << hash->size_bits;
6094 for (i = 0; i < size; i++) {
6095 hlist_for_each_entry(iter, &hash->buckets[i], hlist) {
6096 entry = __ftrace_lookup_ip(direct_functions, iter->ip);
6097 if (!entry)
6098 continue;
6099 entry->direct = addr;
6100 }
6101 }
6102 /* Prevent store tearing if a trampoline concurrently accesses the value */
6103 WRITE_ONCE(ops->direct_call, addr);
6104
6105 mutex_unlock(&ftrace_lock);
6106
6107 /* Removing the tmp_ops will add the updated direct callers to the functions */
6108 unregister_ftrace_function(&tmp_ops);
6109
6110 return err;
6111 }
6112
6113 /**
6114 * modify_ftrace_direct_nolock - Modify an existing direct 'multi' call
6115 * to call something else
6116 * @ops: The address of the struct ftrace_ops object
6117 * @addr: The address of the new trampoline to call at @ops functions
6118 *
6119 * This is used to unregister currently registered direct caller and
6120 * register new one @addr on functions registered in @ops object.
6121 *
6122 * Note there's window between ftrace_shutdown and ftrace_startup calls
6123 * where there will be no callbacks called.
6124 *
6125 * Caller should already have direct_mutex locked, so we don't lock
6126 * direct_mutex here.
6127 *
6128 * Returns: zero on success. Non zero on error, which includes:
6129 * -EINVAL - The @ops object was not properly registered.
6130 */
modify_ftrace_direct_nolock(struct ftrace_ops * ops,unsigned long addr)6131 int modify_ftrace_direct_nolock(struct ftrace_ops *ops, unsigned long addr)
6132 {
6133 if (check_direct_multi(ops))
6134 return -EINVAL;
6135 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6136 return -EINVAL;
6137
6138 return __modify_ftrace_direct(ops, addr);
6139 }
6140 EXPORT_SYMBOL_GPL(modify_ftrace_direct_nolock);
6141
6142 /**
6143 * modify_ftrace_direct - Modify an existing direct 'multi' call
6144 * to call something else
6145 * @ops: The address of the struct ftrace_ops object
6146 * @addr: The address of the new trampoline to call at @ops functions
6147 *
6148 * This is used to unregister currently registered direct caller and
6149 * register new one @addr on functions registered in @ops object.
6150 *
6151 * Note there's window between ftrace_shutdown and ftrace_startup calls
6152 * where there will be no callbacks called.
6153 *
6154 * Returns: zero on success. Non zero on error, which includes:
6155 * -EINVAL - The @ops object was not properly registered.
6156 */
modify_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)6157 int modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
6158 {
6159 int err;
6160
6161 if (check_direct_multi(ops))
6162 return -EINVAL;
6163 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6164 return -EINVAL;
6165
6166 mutex_lock(&direct_mutex);
6167 err = __modify_ftrace_direct(ops, addr);
6168 mutex_unlock(&direct_mutex);
6169 return err;
6170 }
6171 EXPORT_SYMBOL_GPL(modify_ftrace_direct);
6172 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
6173
6174 /**
6175 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
6176 * @ops: the ops to set the filter with
6177 * @ip: the address to add to or remove from the filter.
6178 * @remove: non zero to remove the ip from the filter
6179 * @reset: non zero to reset all filters before applying this filter.
6180 *
6181 * Filters denote which functions should be enabled when tracing is enabled
6182 * If @ip is NULL, it fails to update filter.
6183 *
6184 * This can allocate memory which must be freed before @ops can be freed,
6185 * either by removing each filtered addr or by using
6186 * ftrace_free_filter(@ops).
6187 */
ftrace_set_filter_ip(struct ftrace_ops * ops,unsigned long ip,int remove,int reset)6188 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
6189 int remove, int reset)
6190 {
6191 ftrace_ops_init(ops);
6192 return ftrace_set_addr(ops, &ip, 1, remove, reset, 1);
6193 }
6194 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
6195
6196 /**
6197 * ftrace_set_filter_ips - set functions to filter on in ftrace by addresses
6198 * @ops: the ops to set the filter with
6199 * @ips: the array of addresses to add to or remove from the filter.
6200 * @cnt: the number of addresses in @ips
6201 * @remove: non zero to remove ips from the filter
6202 * @reset: non zero to reset all filters before applying this filter.
6203 *
6204 * Filters denote which functions should be enabled when tracing is enabled
6205 * If @ips array or any ip specified within is NULL , it fails to update filter.
6206 *
6207 * This can allocate memory which must be freed before @ops can be freed,
6208 * either by removing each filtered addr or by using
6209 * ftrace_free_filter(@ops).
6210 */
ftrace_set_filter_ips(struct ftrace_ops * ops,unsigned long * ips,unsigned int cnt,int remove,int reset)6211 int ftrace_set_filter_ips(struct ftrace_ops *ops, unsigned long *ips,
6212 unsigned int cnt, int remove, int reset)
6213 {
6214 ftrace_ops_init(ops);
6215 return ftrace_set_addr(ops, ips, cnt, remove, reset, 1);
6216 }
6217 EXPORT_SYMBOL_GPL(ftrace_set_filter_ips);
6218
6219 /**
6220 * ftrace_ops_set_global_filter - setup ops to use global filters
6221 * @ops: the ops which will use the global filters
6222 *
6223 * ftrace users who need global function trace filtering should call this.
6224 * It can set the global filter only if ops were not initialized before.
6225 */
ftrace_ops_set_global_filter(struct ftrace_ops * ops)6226 void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
6227 {
6228 if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
6229 return;
6230
6231 ftrace_ops_init(ops);
6232 ops->func_hash = &global_ops.local_hash;
6233 }
6234 EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
6235
6236 static int
ftrace_set_regex(struct ftrace_ops * ops,unsigned char * buf,int len,int reset,int enable)6237 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
6238 int reset, int enable)
6239 {
6240 char *mod = NULL, *func, *command, *next = buf;
6241 char *tmp __free(kfree) = NULL;
6242 struct trace_array *tr = ops->private;
6243 int ret;
6244
6245 func = strsep(&next, ":");
6246
6247 /* This can also handle :mod: parsing */
6248 if (next) {
6249 if (!tr)
6250 return -EINVAL;
6251
6252 command = strsep(&next, ":");
6253 if (strcmp(command, "mod") != 0)
6254 return -EINVAL;
6255
6256 mod = next;
6257 len = command - func;
6258 /* Save the original func as ftrace_set_hash() can modify it */
6259 tmp = kstrdup(func, GFP_KERNEL);
6260 }
6261
6262 ret = ftrace_set_hash(ops, func, len, NULL, 0, 0, reset, enable, mod);
6263
6264 if (tr && mod && ret < 0) {
6265 /* Did tmp fail to allocate? */
6266 if (!tmp)
6267 return -ENOMEM;
6268 ret = cache_mod(tr, tmp, mod, enable);
6269 }
6270
6271 return ret;
6272 }
6273
6274 /**
6275 * ftrace_set_filter - set a function to filter on in ftrace
6276 * @ops: the ops to set the filter with
6277 * @buf: the string that holds the function filter text.
6278 * @len: the length of the string.
6279 * @reset: non-zero to reset all filters before applying this filter.
6280 *
6281 * Filters denote which functions should be enabled when tracing is enabled.
6282 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
6283 *
6284 * This can allocate memory which must be freed before @ops can be freed,
6285 * either by removing each filtered addr or by using
6286 * ftrace_free_filter(@ops).
6287 */
ftrace_set_filter(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)6288 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
6289 int len, int reset)
6290 {
6291 ftrace_ops_init(ops);
6292 return ftrace_set_regex(ops, buf, len, reset, 1);
6293 }
6294 EXPORT_SYMBOL_GPL(ftrace_set_filter);
6295
6296 /**
6297 * ftrace_set_notrace - set a function to not trace in ftrace
6298 * @ops: the ops to set the notrace filter with
6299 * @buf: the string that holds the function notrace text.
6300 * @len: the length of the string.
6301 * @reset: non-zero to reset all filters before applying this filter.
6302 *
6303 * Notrace Filters denote which functions should not be enabled when tracing
6304 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
6305 * for tracing.
6306 *
6307 * This can allocate memory which must be freed before @ops can be freed,
6308 * either by removing each filtered addr or by using
6309 * ftrace_free_filter(@ops).
6310 */
ftrace_set_notrace(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)6311 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
6312 int len, int reset)
6313 {
6314 ftrace_ops_init(ops);
6315 return ftrace_set_regex(ops, buf, len, reset, 0);
6316 }
6317 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
6318 /**
6319 * ftrace_set_global_filter - set a function to filter on with global tracers
6320 * @buf: the string that holds the function filter text.
6321 * @len: the length of the string.
6322 * @reset: non-zero to reset all filters before applying this filter.
6323 *
6324 * Filters denote which functions should be enabled when tracing is enabled.
6325 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
6326 */
ftrace_set_global_filter(unsigned char * buf,int len,int reset)6327 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
6328 {
6329 ftrace_set_regex(&global_ops, buf, len, reset, 1);
6330 }
6331 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
6332
6333 /**
6334 * ftrace_set_global_notrace - set a function to not trace with global tracers
6335 * @buf: the string that holds the function notrace text.
6336 * @len: the length of the string.
6337 * @reset: non-zero to reset all filters before applying this filter.
6338 *
6339 * Notrace Filters denote which functions should not be enabled when tracing
6340 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
6341 * for tracing.
6342 */
ftrace_set_global_notrace(unsigned char * buf,int len,int reset)6343 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
6344 {
6345 ftrace_set_regex(&global_ops, buf, len, reset, 0);
6346 }
6347 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
6348
6349 /*
6350 * command line interface to allow users to set filters on boot up.
6351 */
6352 #define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
6353 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
6354 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
6355
6356 /* Used by function selftest to not test if filter is set */
6357 bool ftrace_filter_param __initdata;
6358
set_ftrace_notrace(char * str)6359 static int __init set_ftrace_notrace(char *str)
6360 {
6361 ftrace_filter_param = true;
6362 strscpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
6363 return 1;
6364 }
6365 __setup("ftrace_notrace=", set_ftrace_notrace);
6366
set_ftrace_filter(char * str)6367 static int __init set_ftrace_filter(char *str)
6368 {
6369 ftrace_filter_param = true;
6370 strscpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
6371 return 1;
6372 }
6373 __setup("ftrace_filter=", set_ftrace_filter);
6374
6375 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6376 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
6377 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
6378 static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
6379
set_graph_function(char * str)6380 static int __init set_graph_function(char *str)
6381 {
6382 strscpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
6383 return 1;
6384 }
6385 __setup("ftrace_graph_filter=", set_graph_function);
6386
set_graph_notrace_function(char * str)6387 static int __init set_graph_notrace_function(char *str)
6388 {
6389 strscpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
6390 return 1;
6391 }
6392 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
6393
set_graph_max_depth_function(char * str)6394 static int __init set_graph_max_depth_function(char *str)
6395 {
6396 if (!str || kstrtouint(str, 0, &fgraph_max_depth))
6397 return 0;
6398 return 1;
6399 }
6400 __setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
6401
set_ftrace_early_graph(char * buf,int enable)6402 static void __init set_ftrace_early_graph(char *buf, int enable)
6403 {
6404 int ret;
6405 char *func;
6406 struct ftrace_hash *hash;
6407
6408 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
6409 if (MEM_FAIL(!hash, "Failed to allocate hash\n"))
6410 return;
6411
6412 while (buf) {
6413 func = strsep(&buf, ",");
6414 /* we allow only one expression at a time */
6415 ret = ftrace_graph_set_hash(hash, func);
6416 if (ret)
6417 printk(KERN_DEBUG "ftrace: function %s not "
6418 "traceable\n", func);
6419 }
6420
6421 if (enable)
6422 ftrace_graph_hash = hash;
6423 else
6424 ftrace_graph_notrace_hash = hash;
6425 }
6426 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6427
6428 void __init
ftrace_set_early_filter(struct ftrace_ops * ops,char * buf,int enable)6429 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
6430 {
6431 char *func;
6432
6433 ftrace_ops_init(ops);
6434
6435 /* The trace_array is needed for caching module function filters */
6436 if (!ops->private) {
6437 struct trace_array *tr = trace_get_global_array();
6438
6439 ops->private = tr;
6440 ftrace_init_trace_array(tr);
6441 }
6442
6443 while (buf) {
6444 func = strsep(&buf, ",");
6445 ftrace_set_regex(ops, func, strlen(func), 0, enable);
6446 }
6447 }
6448
set_ftrace_early_filters(void)6449 static void __init set_ftrace_early_filters(void)
6450 {
6451 if (ftrace_filter_buf[0])
6452 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
6453 if (ftrace_notrace_buf[0])
6454 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
6455 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6456 if (ftrace_graph_buf[0])
6457 set_ftrace_early_graph(ftrace_graph_buf, 1);
6458 if (ftrace_graph_notrace_buf[0])
6459 set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
6460 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6461 }
6462
ftrace_regex_release(struct inode * inode,struct file * file)6463 int ftrace_regex_release(struct inode *inode, struct file *file)
6464 {
6465 struct seq_file *m = (struct seq_file *)file->private_data;
6466 struct ftrace_iterator *iter;
6467 struct ftrace_hash **orig_hash;
6468 struct trace_parser *parser;
6469 int filter_hash;
6470
6471 if (file->f_mode & FMODE_READ) {
6472 iter = m->private;
6473 seq_release(inode, file);
6474 } else
6475 iter = file->private_data;
6476
6477 parser = &iter->parser;
6478 if (trace_parser_loaded(parser)) {
6479 int enable = !(iter->flags & FTRACE_ITER_NOTRACE);
6480
6481 ftrace_process_regex(iter, parser->buffer,
6482 parser->idx, enable);
6483 }
6484
6485 trace_parser_put(parser);
6486
6487 mutex_lock(&iter->ops->func_hash->regex_lock);
6488
6489 if (file->f_mode & FMODE_WRITE) {
6490 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
6491
6492 if (filter_hash) {
6493 orig_hash = &iter->ops->func_hash->filter_hash;
6494 if (iter->tr) {
6495 if (list_empty(&iter->tr->mod_trace))
6496 iter->hash->flags &= ~FTRACE_HASH_FL_MOD;
6497 else
6498 iter->hash->flags |= FTRACE_HASH_FL_MOD;
6499 }
6500 } else
6501 orig_hash = &iter->ops->func_hash->notrace_hash;
6502
6503 mutex_lock(&ftrace_lock);
6504 ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
6505 iter->hash, filter_hash);
6506 mutex_unlock(&ftrace_lock);
6507 } else {
6508 /* For read only, the hash is the ops hash */
6509 iter->hash = NULL;
6510 }
6511
6512 mutex_unlock(&iter->ops->func_hash->regex_lock);
6513 free_ftrace_hash(iter->hash);
6514 if (iter->tr)
6515 trace_array_put(iter->tr);
6516 kfree(iter);
6517
6518 return 0;
6519 }
6520
6521 static const struct file_operations ftrace_avail_fops = {
6522 .open = ftrace_avail_open,
6523 .read = seq_read,
6524 .llseek = seq_lseek,
6525 .release = seq_release_private,
6526 };
6527
6528 static const struct file_operations ftrace_enabled_fops = {
6529 .open = ftrace_enabled_open,
6530 .read = seq_read,
6531 .llseek = seq_lseek,
6532 .release = seq_release_private,
6533 };
6534
6535 static const struct file_operations ftrace_touched_fops = {
6536 .open = ftrace_touched_open,
6537 .read = seq_read,
6538 .llseek = seq_lseek,
6539 .release = seq_release_private,
6540 };
6541
6542 static const struct file_operations ftrace_avail_addrs_fops = {
6543 .open = ftrace_avail_addrs_open,
6544 .read = seq_read,
6545 .llseek = seq_lseek,
6546 .release = seq_release_private,
6547 };
6548
6549 static const struct file_operations ftrace_filter_fops = {
6550 .open = ftrace_filter_open,
6551 .read = seq_read,
6552 .write = ftrace_filter_write,
6553 .llseek = tracing_lseek,
6554 .release = ftrace_regex_release,
6555 };
6556
6557 static const struct file_operations ftrace_notrace_fops = {
6558 .open = ftrace_notrace_open,
6559 .read = seq_read,
6560 .write = ftrace_notrace_write,
6561 .llseek = tracing_lseek,
6562 .release = ftrace_regex_release,
6563 };
6564
6565 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6566
6567 static DEFINE_MUTEX(graph_lock);
6568
6569 struct ftrace_hash __rcu *ftrace_graph_hash = EMPTY_HASH;
6570 struct ftrace_hash __rcu *ftrace_graph_notrace_hash = EMPTY_HASH;
6571
6572 enum graph_filter_type {
6573 GRAPH_FILTER_NOTRACE = 0,
6574 GRAPH_FILTER_FUNCTION,
6575 };
6576
6577 #define FTRACE_GRAPH_EMPTY ((void *)1)
6578
6579 struct ftrace_graph_data {
6580 struct ftrace_hash *hash;
6581 struct ftrace_func_entry *entry;
6582 int idx; /* for hash table iteration */
6583 enum graph_filter_type type;
6584 struct ftrace_hash *new_hash;
6585 const struct seq_operations *seq_ops;
6586 struct trace_parser parser;
6587 };
6588
6589 static void *
__g_next(struct seq_file * m,loff_t * pos)6590 __g_next(struct seq_file *m, loff_t *pos)
6591 {
6592 struct ftrace_graph_data *fgd = m->private;
6593 struct ftrace_func_entry *entry = fgd->entry;
6594 struct hlist_head *head;
6595 int i, idx = fgd->idx;
6596
6597 if (*pos >= fgd->hash->count)
6598 return NULL;
6599
6600 if (entry) {
6601 hlist_for_each_entry_continue(entry, hlist) {
6602 fgd->entry = entry;
6603 return entry;
6604 }
6605
6606 idx++;
6607 }
6608
6609 for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
6610 head = &fgd->hash->buckets[i];
6611 hlist_for_each_entry(entry, head, hlist) {
6612 fgd->entry = entry;
6613 fgd->idx = i;
6614 return entry;
6615 }
6616 }
6617 return NULL;
6618 }
6619
6620 static void *
g_next(struct seq_file * m,void * v,loff_t * pos)6621 g_next(struct seq_file *m, void *v, loff_t *pos)
6622 {
6623 (*pos)++;
6624 return __g_next(m, pos);
6625 }
6626
g_start(struct seq_file * m,loff_t * pos)6627 static void *g_start(struct seq_file *m, loff_t *pos)
6628 {
6629 struct ftrace_graph_data *fgd = m->private;
6630
6631 mutex_lock(&graph_lock);
6632
6633 if (fgd->type == GRAPH_FILTER_FUNCTION)
6634 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
6635 lockdep_is_held(&graph_lock));
6636 else
6637 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6638 lockdep_is_held(&graph_lock));
6639
6640 /* Nothing, tell g_show to print all functions are enabled */
6641 if (ftrace_hash_empty(fgd->hash) && !*pos)
6642 return FTRACE_GRAPH_EMPTY;
6643
6644 fgd->idx = 0;
6645 fgd->entry = NULL;
6646 return __g_next(m, pos);
6647 }
6648
g_stop(struct seq_file * m,void * p)6649 static void g_stop(struct seq_file *m, void *p)
6650 {
6651 mutex_unlock(&graph_lock);
6652 }
6653
g_show(struct seq_file * m,void * v)6654 static int g_show(struct seq_file *m, void *v)
6655 {
6656 struct ftrace_func_entry *entry = v;
6657
6658 if (!entry)
6659 return 0;
6660
6661 if (entry == FTRACE_GRAPH_EMPTY) {
6662 struct ftrace_graph_data *fgd = m->private;
6663
6664 if (fgd->type == GRAPH_FILTER_FUNCTION)
6665 seq_puts(m, "#### all functions enabled ####\n");
6666 else
6667 seq_puts(m, "#### no functions disabled ####\n");
6668 return 0;
6669 }
6670
6671 seq_printf(m, "%ps\n", (void *)entry->ip);
6672
6673 return 0;
6674 }
6675
6676 static const struct seq_operations ftrace_graph_seq_ops = {
6677 .start = g_start,
6678 .next = g_next,
6679 .stop = g_stop,
6680 .show = g_show,
6681 };
6682
6683 static int
__ftrace_graph_open(struct inode * inode,struct file * file,struct ftrace_graph_data * fgd)6684 __ftrace_graph_open(struct inode *inode, struct file *file,
6685 struct ftrace_graph_data *fgd)
6686 {
6687 int ret;
6688 struct ftrace_hash *new_hash = NULL;
6689
6690 ret = security_locked_down(LOCKDOWN_TRACEFS);
6691 if (ret)
6692 return ret;
6693
6694 if (file->f_mode & FMODE_WRITE) {
6695 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
6696
6697 if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
6698 return -ENOMEM;
6699
6700 if (file->f_flags & O_TRUNC)
6701 new_hash = alloc_ftrace_hash(size_bits);
6702 else
6703 new_hash = alloc_and_copy_ftrace_hash(size_bits,
6704 fgd->hash);
6705 if (!new_hash) {
6706 ret = -ENOMEM;
6707 goto out;
6708 }
6709 }
6710
6711 if (file->f_mode & FMODE_READ) {
6712 ret = seq_open(file, &ftrace_graph_seq_ops);
6713 if (!ret) {
6714 struct seq_file *m = file->private_data;
6715 m->private = fgd;
6716 } else {
6717 /* Failed */
6718 free_ftrace_hash(new_hash);
6719 new_hash = NULL;
6720 }
6721 } else
6722 file->private_data = fgd;
6723
6724 out:
6725 if (ret < 0 && file->f_mode & FMODE_WRITE)
6726 trace_parser_put(&fgd->parser);
6727
6728 fgd->new_hash = new_hash;
6729
6730 /*
6731 * All uses of fgd->hash must be taken with the graph_lock
6732 * held. The graph_lock is going to be released, so force
6733 * fgd->hash to be reinitialized when it is taken again.
6734 */
6735 fgd->hash = NULL;
6736
6737 return ret;
6738 }
6739
6740 static int
ftrace_graph_open(struct inode * inode,struct file * file)6741 ftrace_graph_open(struct inode *inode, struct file *file)
6742 {
6743 struct ftrace_graph_data *fgd;
6744 int ret;
6745
6746 if (unlikely(ftrace_disabled))
6747 return -ENODEV;
6748
6749 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
6750 if (fgd == NULL)
6751 return -ENOMEM;
6752
6753 mutex_lock(&graph_lock);
6754
6755 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
6756 lockdep_is_held(&graph_lock));
6757 fgd->type = GRAPH_FILTER_FUNCTION;
6758 fgd->seq_ops = &ftrace_graph_seq_ops;
6759
6760 ret = __ftrace_graph_open(inode, file, fgd);
6761 if (ret < 0)
6762 kfree(fgd);
6763
6764 mutex_unlock(&graph_lock);
6765 return ret;
6766 }
6767
6768 static int
ftrace_graph_notrace_open(struct inode * inode,struct file * file)6769 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
6770 {
6771 struct ftrace_graph_data *fgd;
6772 int ret;
6773
6774 if (unlikely(ftrace_disabled))
6775 return -ENODEV;
6776
6777 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
6778 if (fgd == NULL)
6779 return -ENOMEM;
6780
6781 mutex_lock(&graph_lock);
6782
6783 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6784 lockdep_is_held(&graph_lock));
6785 fgd->type = GRAPH_FILTER_NOTRACE;
6786 fgd->seq_ops = &ftrace_graph_seq_ops;
6787
6788 ret = __ftrace_graph_open(inode, file, fgd);
6789 if (ret < 0)
6790 kfree(fgd);
6791
6792 mutex_unlock(&graph_lock);
6793 return ret;
6794 }
6795
6796 static int
ftrace_graph_release(struct inode * inode,struct file * file)6797 ftrace_graph_release(struct inode *inode, struct file *file)
6798 {
6799 struct ftrace_graph_data *fgd;
6800 struct ftrace_hash *old_hash, *new_hash;
6801 struct trace_parser *parser;
6802 int ret = 0;
6803
6804 if (file->f_mode & FMODE_READ) {
6805 struct seq_file *m = file->private_data;
6806
6807 fgd = m->private;
6808 seq_release(inode, file);
6809 } else {
6810 fgd = file->private_data;
6811 }
6812
6813
6814 if (file->f_mode & FMODE_WRITE) {
6815
6816 parser = &fgd->parser;
6817
6818 if (trace_parser_loaded((parser))) {
6819 ret = ftrace_graph_set_hash(fgd->new_hash,
6820 parser->buffer);
6821 }
6822
6823 trace_parser_put(parser);
6824
6825 new_hash = __ftrace_hash_move(fgd->new_hash);
6826 if (!new_hash) {
6827 ret = -ENOMEM;
6828 goto out;
6829 }
6830
6831 mutex_lock(&graph_lock);
6832
6833 if (fgd->type == GRAPH_FILTER_FUNCTION) {
6834 old_hash = rcu_dereference_protected(ftrace_graph_hash,
6835 lockdep_is_held(&graph_lock));
6836 rcu_assign_pointer(ftrace_graph_hash, new_hash);
6837 } else {
6838 old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6839 lockdep_is_held(&graph_lock));
6840 rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
6841 }
6842
6843 mutex_unlock(&graph_lock);
6844
6845 /*
6846 * We need to do a hard force of sched synchronization.
6847 * This is because we use preempt_disable() to do RCU, but
6848 * the function tracers can be called where RCU is not watching
6849 * (like before user_exit()). We can not rely on the RCU
6850 * infrastructure to do the synchronization, thus we must do it
6851 * ourselves.
6852 */
6853 if (old_hash != EMPTY_HASH)
6854 synchronize_rcu_tasks_rude();
6855
6856 free_ftrace_hash(old_hash);
6857 }
6858
6859 out:
6860 free_ftrace_hash(fgd->new_hash);
6861 kfree(fgd);
6862
6863 return ret;
6864 }
6865
6866 static int
ftrace_graph_set_hash(struct ftrace_hash * hash,char * buffer)6867 ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
6868 {
6869 struct ftrace_glob func_g;
6870 struct dyn_ftrace *rec;
6871 struct ftrace_page *pg;
6872 struct ftrace_func_entry *entry;
6873 int fail = 1;
6874 int not;
6875
6876 /* decode regex */
6877 func_g.type = filter_parse_regex(buffer, strlen(buffer),
6878 &func_g.search, ¬);
6879
6880 func_g.len = strlen(func_g.search);
6881
6882 guard(mutex)(&ftrace_lock);
6883
6884 if (unlikely(ftrace_disabled))
6885 return -ENODEV;
6886
6887 do_for_each_ftrace_rec(pg, rec) {
6888
6889 if (rec->flags & FTRACE_FL_DISABLED)
6890 continue;
6891
6892 if (ftrace_match_record(rec, &func_g, NULL, 0)) {
6893 entry = ftrace_lookup_ip(hash, rec->ip);
6894
6895 if (!not) {
6896 fail = 0;
6897
6898 if (entry)
6899 continue;
6900 if (add_hash_entry(hash, rec->ip) == NULL)
6901 return 0;
6902 } else {
6903 if (entry) {
6904 free_hash_entry(hash, entry);
6905 fail = 0;
6906 }
6907 }
6908 }
6909 cond_resched();
6910 } while_for_each_ftrace_rec();
6911
6912 return fail ? -EINVAL : 0;
6913 }
6914
6915 static ssize_t
ftrace_graph_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)6916 ftrace_graph_write(struct file *file, const char __user *ubuf,
6917 size_t cnt, loff_t *ppos)
6918 {
6919 ssize_t read, ret = 0;
6920 struct ftrace_graph_data *fgd = file->private_data;
6921 struct trace_parser *parser;
6922
6923 if (!cnt)
6924 return 0;
6925
6926 /* Read mode uses seq functions */
6927 if (file->f_mode & FMODE_READ) {
6928 struct seq_file *m = file->private_data;
6929 fgd = m->private;
6930 }
6931
6932 parser = &fgd->parser;
6933
6934 read = trace_get_user(parser, ubuf, cnt, ppos);
6935
6936 if (read >= 0 && trace_parser_loaded(parser) &&
6937 !trace_parser_cont(parser)) {
6938
6939 ret = ftrace_graph_set_hash(fgd->new_hash,
6940 parser->buffer);
6941 trace_parser_clear(parser);
6942 }
6943
6944 if (!ret)
6945 ret = read;
6946
6947 return ret;
6948 }
6949
6950 static const struct file_operations ftrace_graph_fops = {
6951 .open = ftrace_graph_open,
6952 .read = seq_read,
6953 .write = ftrace_graph_write,
6954 .llseek = tracing_lseek,
6955 .release = ftrace_graph_release,
6956 };
6957
6958 static const struct file_operations ftrace_graph_notrace_fops = {
6959 .open = ftrace_graph_notrace_open,
6960 .read = seq_read,
6961 .write = ftrace_graph_write,
6962 .llseek = tracing_lseek,
6963 .release = ftrace_graph_release,
6964 };
6965 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6966
ftrace_create_filter_files(struct ftrace_ops * ops,struct dentry * parent)6967 void ftrace_create_filter_files(struct ftrace_ops *ops,
6968 struct dentry *parent)
6969 {
6970
6971 trace_create_file("set_ftrace_filter", TRACE_MODE_WRITE, parent,
6972 ops, &ftrace_filter_fops);
6973
6974 trace_create_file("set_ftrace_notrace", TRACE_MODE_WRITE, parent,
6975 ops, &ftrace_notrace_fops);
6976 }
6977
6978 /*
6979 * The name "destroy_filter_files" is really a misnomer. Although
6980 * in the future, it may actually delete the files, but this is
6981 * really intended to make sure the ops passed in are disabled
6982 * and that when this function returns, the caller is free to
6983 * free the ops.
6984 *
6985 * The "destroy" name is only to match the "create" name that this
6986 * should be paired with.
6987 */
ftrace_destroy_filter_files(struct ftrace_ops * ops)6988 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
6989 {
6990 mutex_lock(&ftrace_lock);
6991 if (ops->flags & FTRACE_OPS_FL_ENABLED)
6992 ftrace_shutdown(ops, 0);
6993 ops->flags |= FTRACE_OPS_FL_DELETED;
6994 ftrace_free_filter(ops);
6995 mutex_unlock(&ftrace_lock);
6996 }
6997
ftrace_init_dyn_tracefs(struct dentry * d_tracer)6998 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
6999 {
7000
7001 trace_create_file("available_filter_functions", TRACE_MODE_READ,
7002 d_tracer, NULL, &ftrace_avail_fops);
7003
7004 trace_create_file("available_filter_functions_addrs", TRACE_MODE_READ,
7005 d_tracer, NULL, &ftrace_avail_addrs_fops);
7006
7007 trace_create_file("enabled_functions", TRACE_MODE_READ,
7008 d_tracer, NULL, &ftrace_enabled_fops);
7009
7010 trace_create_file("touched_functions", TRACE_MODE_READ,
7011 d_tracer, NULL, &ftrace_touched_fops);
7012
7013 ftrace_create_filter_files(&global_ops, d_tracer);
7014
7015 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
7016 trace_create_file("set_graph_function", TRACE_MODE_WRITE, d_tracer,
7017 NULL,
7018 &ftrace_graph_fops);
7019 trace_create_file("set_graph_notrace", TRACE_MODE_WRITE, d_tracer,
7020 NULL,
7021 &ftrace_graph_notrace_fops);
7022 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
7023
7024 return 0;
7025 }
7026
ftrace_cmp_ips(const void * a,const void * b)7027 static int ftrace_cmp_ips(const void *a, const void *b)
7028 {
7029 const unsigned long *ipa = a;
7030 const unsigned long *ipb = b;
7031
7032 if (*ipa > *ipb)
7033 return 1;
7034 if (*ipa < *ipb)
7035 return -1;
7036 return 0;
7037 }
7038
7039 #ifdef CONFIG_FTRACE_SORT_STARTUP_TEST
test_is_sorted(unsigned long * start,unsigned long count)7040 static void test_is_sorted(unsigned long *start, unsigned long count)
7041 {
7042 int i;
7043
7044 for (i = 1; i < count; i++) {
7045 if (WARN(start[i - 1] > start[i],
7046 "[%d] %pS at %lx is not sorted with %pS at %lx\n", i,
7047 (void *)start[i - 1], start[i - 1],
7048 (void *)start[i], start[i]))
7049 break;
7050 }
7051 if (i == count)
7052 pr_info("ftrace section at %px sorted properly\n", start);
7053 }
7054 #else
test_is_sorted(unsigned long * start,unsigned long count)7055 static void test_is_sorted(unsigned long *start, unsigned long count)
7056 {
7057 }
7058 #endif
7059
ftrace_process_locs(struct module * mod,unsigned long * start,unsigned long * end)7060 static int ftrace_process_locs(struct module *mod,
7061 unsigned long *start,
7062 unsigned long *end)
7063 {
7064 struct ftrace_page *pg_unuse = NULL;
7065 struct ftrace_page *start_pg;
7066 struct ftrace_page *pg;
7067 struct dyn_ftrace *rec;
7068 unsigned long skipped = 0;
7069 unsigned long count;
7070 unsigned long *p;
7071 unsigned long addr;
7072 unsigned long flags = 0; /* Shut up gcc */
7073 unsigned long pages;
7074 int ret = -ENOMEM;
7075
7076 count = end - start;
7077
7078 if (!count)
7079 return 0;
7080
7081 pages = DIV_ROUND_UP(count, ENTRIES_PER_PAGE);
7082
7083 /*
7084 * Sorting mcount in vmlinux at build time depend on
7085 * CONFIG_BUILDTIME_MCOUNT_SORT, while mcount loc in
7086 * modules can not be sorted at build time.
7087 */
7088 if (!IS_ENABLED(CONFIG_BUILDTIME_MCOUNT_SORT) || mod) {
7089 sort(start, count, sizeof(*start),
7090 ftrace_cmp_ips, NULL);
7091 } else {
7092 test_is_sorted(start, count);
7093 }
7094
7095 start_pg = ftrace_allocate_pages(count);
7096 if (!start_pg)
7097 return -ENOMEM;
7098
7099 mutex_lock(&ftrace_lock);
7100
7101 /*
7102 * Core and each module needs their own pages, as
7103 * modules will free them when they are removed.
7104 * Force a new page to be allocated for modules.
7105 */
7106 if (!mod) {
7107 WARN_ON(ftrace_pages || ftrace_pages_start);
7108 /* First initialization */
7109 ftrace_pages = ftrace_pages_start = start_pg;
7110 } else {
7111 if (!ftrace_pages)
7112 goto out;
7113
7114 if (WARN_ON(ftrace_pages->next)) {
7115 /* Hmm, we have free pages? */
7116 while (ftrace_pages->next)
7117 ftrace_pages = ftrace_pages->next;
7118 }
7119
7120 ftrace_pages->next = start_pg;
7121 }
7122
7123 p = start;
7124 pg = start_pg;
7125 while (p < end) {
7126 unsigned long end_offset;
7127
7128 addr = *p++;
7129
7130 /*
7131 * Some architecture linkers will pad between
7132 * the different mcount_loc sections of different
7133 * object files to satisfy alignments.
7134 * Skip any NULL pointers.
7135 */
7136 if (!addr) {
7137 skipped++;
7138 continue;
7139 }
7140
7141 /*
7142 * If this is core kernel, make sure the address is in core
7143 * or inittext, as weak functions get zeroed and KASLR can
7144 * move them to something other than zero. It just will not
7145 * move it to an area where kernel text is.
7146 */
7147 if (!mod && !(is_kernel_text(addr) || is_kernel_inittext(addr))) {
7148 skipped++;
7149 continue;
7150 }
7151
7152 addr = ftrace_call_adjust(addr);
7153
7154 end_offset = (pg->index+1) * sizeof(pg->records[0]);
7155 if (end_offset > PAGE_SIZE << pg->order) {
7156 /* We should have allocated enough */
7157 if (WARN_ON(!pg->next))
7158 break;
7159 pg = pg->next;
7160 }
7161
7162 rec = &pg->records[pg->index++];
7163 rec->ip = addr;
7164 }
7165
7166 if (pg->next) {
7167 pg_unuse = pg->next;
7168 pg->next = NULL;
7169 }
7170
7171 /* Assign the last page to ftrace_pages */
7172 ftrace_pages = pg;
7173
7174 /*
7175 * We only need to disable interrupts on start up
7176 * because we are modifying code that an interrupt
7177 * may execute, and the modification is not atomic.
7178 * But for modules, nothing runs the code we modify
7179 * until we are finished with it, and there's no
7180 * reason to cause large interrupt latencies while we do it.
7181 */
7182 if (!mod)
7183 local_irq_save(flags);
7184 ftrace_update_code(mod, start_pg);
7185 if (!mod)
7186 local_irq_restore(flags);
7187 ret = 0;
7188 out:
7189 mutex_unlock(&ftrace_lock);
7190
7191 /* We should have used all pages unless we skipped some */
7192 if (pg_unuse) {
7193 unsigned long pg_remaining, remaining = 0;
7194 unsigned long skip;
7195
7196 /* Count the number of entries unused and compare it to skipped. */
7197 pg_remaining = (ENTRIES_PER_PAGE << pg->order) - pg->index;
7198
7199 if (!WARN(skipped < pg_remaining, "Extra allocated pages for ftrace")) {
7200
7201 skip = skipped - pg_remaining;
7202
7203 for (pg = pg_unuse; pg; pg = pg->next)
7204 remaining += 1 << pg->order;
7205
7206 pages -= remaining;
7207
7208 skip = DIV_ROUND_UP(skip, ENTRIES_PER_PAGE);
7209
7210 /*
7211 * Check to see if the number of pages remaining would
7212 * just fit the number of entries skipped.
7213 */
7214 WARN(skip != remaining, "Extra allocated pages for ftrace: %lu with %lu skipped",
7215 remaining, skipped);
7216 }
7217 /* Need to synchronize with ftrace_location_range() */
7218 synchronize_rcu();
7219 ftrace_free_pages(pg_unuse);
7220 }
7221
7222 if (!mod) {
7223 count -= skipped;
7224 pr_info("ftrace: allocating %ld entries in %ld pages\n",
7225 count, pages);
7226 }
7227
7228 return ret;
7229 }
7230
7231 struct ftrace_mod_func {
7232 struct list_head list;
7233 char *name;
7234 unsigned long ip;
7235 unsigned int size;
7236 };
7237
7238 struct ftrace_mod_map {
7239 struct rcu_head rcu;
7240 struct list_head list;
7241 struct module *mod;
7242 unsigned long start_addr;
7243 unsigned long end_addr;
7244 struct list_head funcs;
7245 unsigned int num_funcs;
7246 };
7247
ftrace_get_trampoline_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7248 static int ftrace_get_trampoline_kallsym(unsigned int symnum,
7249 unsigned long *value, char *type,
7250 char *name, char *module_name,
7251 int *exported)
7252 {
7253 struct ftrace_ops *op;
7254
7255 list_for_each_entry_rcu(op, &ftrace_ops_trampoline_list, list) {
7256 if (!op->trampoline || symnum--)
7257 continue;
7258 *value = op->trampoline;
7259 *type = 't';
7260 strscpy(name, FTRACE_TRAMPOLINE_SYM, KSYM_NAME_LEN);
7261 strscpy(module_name, FTRACE_TRAMPOLINE_MOD, MODULE_NAME_LEN);
7262 *exported = 0;
7263 return 0;
7264 }
7265
7266 return -ERANGE;
7267 }
7268
7269 #if defined(CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS) || defined(CONFIG_MODULES)
7270 /*
7271 * Check if the current ops references the given ip.
7272 *
7273 * If the ops traces all functions, then it was already accounted for.
7274 * If the ops does not trace the current record function, skip it.
7275 * If the ops ignores the function via notrace filter, skip it.
7276 */
7277 static bool
ops_references_ip(struct ftrace_ops * ops,unsigned long ip)7278 ops_references_ip(struct ftrace_ops *ops, unsigned long ip)
7279 {
7280 /* If ops isn't enabled, ignore it */
7281 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
7282 return false;
7283
7284 /* If ops traces all then it includes this function */
7285 if (ops_traces_mod(ops))
7286 return true;
7287
7288 /* The function must be in the filter */
7289 if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
7290 !__ftrace_lookup_ip(ops->func_hash->filter_hash, ip))
7291 return false;
7292
7293 /* If in notrace hash, we ignore it too */
7294 if (ftrace_lookup_ip(ops->func_hash->notrace_hash, ip))
7295 return false;
7296
7297 return true;
7298 }
7299 #endif
7300
7301 #ifdef CONFIG_MODULES
7302
7303 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
7304
7305 static LIST_HEAD(ftrace_mod_maps);
7306
referenced_filters(struct dyn_ftrace * rec)7307 static int referenced_filters(struct dyn_ftrace *rec)
7308 {
7309 struct ftrace_ops *ops;
7310 int cnt = 0;
7311
7312 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
7313 if (ops_references_ip(ops, rec->ip)) {
7314 if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_DIRECT))
7315 continue;
7316 if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_IPMODIFY))
7317 continue;
7318 cnt++;
7319 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
7320 rec->flags |= FTRACE_FL_REGS;
7321 if (cnt == 1 && ops->trampoline)
7322 rec->flags |= FTRACE_FL_TRAMP;
7323 else
7324 rec->flags &= ~FTRACE_FL_TRAMP;
7325 }
7326 }
7327
7328 return cnt;
7329 }
7330
7331 static void
clear_mod_from_hash(struct ftrace_page * pg,struct ftrace_hash * hash)7332 clear_mod_from_hash(struct ftrace_page *pg, struct ftrace_hash *hash)
7333 {
7334 struct ftrace_func_entry *entry;
7335 struct dyn_ftrace *rec;
7336 int i;
7337
7338 if (ftrace_hash_empty(hash))
7339 return;
7340
7341 for (i = 0; i < pg->index; i++) {
7342 rec = &pg->records[i];
7343 entry = __ftrace_lookup_ip(hash, rec->ip);
7344 /*
7345 * Do not allow this rec to match again.
7346 * Yeah, it may waste some memory, but will be removed
7347 * if/when the hash is modified again.
7348 */
7349 if (entry)
7350 entry->ip = 0;
7351 }
7352 }
7353
7354 /* Clear any records from hashes */
clear_mod_from_hashes(struct ftrace_page * pg)7355 static void clear_mod_from_hashes(struct ftrace_page *pg)
7356 {
7357 struct trace_array *tr;
7358
7359 mutex_lock(&trace_types_lock);
7360 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7361 if (!tr->ops || !tr->ops->func_hash)
7362 continue;
7363 mutex_lock(&tr->ops->func_hash->regex_lock);
7364 clear_mod_from_hash(pg, tr->ops->func_hash->filter_hash);
7365 clear_mod_from_hash(pg, tr->ops->func_hash->notrace_hash);
7366 mutex_unlock(&tr->ops->func_hash->regex_lock);
7367 }
7368 mutex_unlock(&trace_types_lock);
7369 }
7370
ftrace_free_mod_map(struct rcu_head * rcu)7371 static void ftrace_free_mod_map(struct rcu_head *rcu)
7372 {
7373 struct ftrace_mod_map *mod_map = container_of(rcu, struct ftrace_mod_map, rcu);
7374 struct ftrace_mod_func *mod_func;
7375 struct ftrace_mod_func *n;
7376
7377 /* All the contents of mod_map are now not visible to readers */
7378 list_for_each_entry_safe(mod_func, n, &mod_map->funcs, list) {
7379 kfree(mod_func->name);
7380 list_del(&mod_func->list);
7381 kfree(mod_func);
7382 }
7383
7384 kfree(mod_map);
7385 }
7386
ftrace_release_mod(struct module * mod)7387 void ftrace_release_mod(struct module *mod)
7388 {
7389 struct ftrace_mod_map *mod_map;
7390 struct ftrace_mod_map *n;
7391 struct dyn_ftrace *rec;
7392 struct ftrace_page **last_pg;
7393 struct ftrace_page *tmp_page = NULL;
7394 struct ftrace_page *pg;
7395
7396 mutex_lock(&ftrace_lock);
7397
7398 if (ftrace_disabled)
7399 goto out_unlock;
7400
7401 list_for_each_entry_safe(mod_map, n, &ftrace_mod_maps, list) {
7402 if (mod_map->mod == mod) {
7403 list_del_rcu(&mod_map->list);
7404 call_rcu(&mod_map->rcu, ftrace_free_mod_map);
7405 break;
7406 }
7407 }
7408
7409 /*
7410 * Each module has its own ftrace_pages, remove
7411 * them from the list.
7412 */
7413 last_pg = &ftrace_pages_start;
7414 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
7415 rec = &pg->records[0];
7416 if (within_module(rec->ip, mod)) {
7417 /*
7418 * As core pages are first, the first
7419 * page should never be a module page.
7420 */
7421 if (WARN_ON(pg == ftrace_pages_start))
7422 goto out_unlock;
7423
7424 /* Check if we are deleting the last page */
7425 if (pg == ftrace_pages)
7426 ftrace_pages = next_to_ftrace_page(last_pg);
7427
7428 ftrace_update_tot_cnt -= pg->index;
7429 *last_pg = pg->next;
7430
7431 pg->next = tmp_page;
7432 tmp_page = pg;
7433 } else
7434 last_pg = &pg->next;
7435 }
7436 out_unlock:
7437 mutex_unlock(&ftrace_lock);
7438
7439 /* Need to synchronize with ftrace_location_range() */
7440 if (tmp_page)
7441 synchronize_rcu();
7442 for (pg = tmp_page; pg; pg = tmp_page) {
7443
7444 /* Needs to be called outside of ftrace_lock */
7445 clear_mod_from_hashes(pg);
7446
7447 if (pg->records) {
7448 free_pages((unsigned long)pg->records, pg->order);
7449 ftrace_number_of_pages -= 1 << pg->order;
7450 }
7451 tmp_page = pg->next;
7452 kfree(pg);
7453 ftrace_number_of_groups--;
7454 }
7455 }
7456
ftrace_module_enable(struct module * mod)7457 void ftrace_module_enable(struct module *mod)
7458 {
7459 struct dyn_ftrace *rec;
7460 struct ftrace_page *pg;
7461
7462 mutex_lock(&ftrace_lock);
7463
7464 if (ftrace_disabled)
7465 goto out_unlock;
7466
7467 /*
7468 * If the tracing is enabled, go ahead and enable the record.
7469 *
7470 * The reason not to enable the record immediately is the
7471 * inherent check of ftrace_make_nop/ftrace_make_call for
7472 * correct previous instructions. Making first the NOP
7473 * conversion puts the module to the correct state, thus
7474 * passing the ftrace_make_call check.
7475 *
7476 * We also delay this to after the module code already set the
7477 * text to read-only, as we now need to set it back to read-write
7478 * so that we can modify the text.
7479 */
7480 if (ftrace_start_up)
7481 ftrace_arch_code_modify_prepare();
7482
7483 do_for_each_ftrace_rec(pg, rec) {
7484 int cnt;
7485 /*
7486 * do_for_each_ftrace_rec() is a double loop.
7487 * module text shares the pg. If a record is
7488 * not part of this module, then skip this pg,
7489 * which the "break" will do.
7490 */
7491 if (!within_module(rec->ip, mod))
7492 break;
7493
7494 /* Weak functions should still be ignored */
7495 if (!test_for_valid_rec(rec)) {
7496 /* Clear all other flags. Should not be enabled anyway */
7497 rec->flags = FTRACE_FL_DISABLED;
7498 continue;
7499 }
7500
7501 cnt = 0;
7502
7503 /*
7504 * When adding a module, we need to check if tracers are
7505 * currently enabled and if they are, and can trace this record,
7506 * we need to enable the module functions as well as update the
7507 * reference counts for those function records.
7508 */
7509 if (ftrace_start_up)
7510 cnt += referenced_filters(rec);
7511
7512 rec->flags &= ~FTRACE_FL_DISABLED;
7513 rec->flags += cnt;
7514
7515 if (ftrace_start_up && cnt) {
7516 int failed = __ftrace_replace_code(rec, 1);
7517 if (failed) {
7518 ftrace_bug(failed, rec);
7519 goto out_loop;
7520 }
7521 }
7522
7523 } while_for_each_ftrace_rec();
7524
7525 out_loop:
7526 if (ftrace_start_up)
7527 ftrace_arch_code_modify_post_process();
7528
7529 out_unlock:
7530 mutex_unlock(&ftrace_lock);
7531
7532 process_cached_mods(mod->name);
7533 }
7534
ftrace_module_init(struct module * mod)7535 void ftrace_module_init(struct module *mod)
7536 {
7537 int ret;
7538
7539 if (ftrace_disabled || !mod->num_ftrace_callsites)
7540 return;
7541
7542 ret = ftrace_process_locs(mod, mod->ftrace_callsites,
7543 mod->ftrace_callsites + mod->num_ftrace_callsites);
7544 if (ret)
7545 pr_warn("ftrace: failed to allocate entries for module '%s' functions\n",
7546 mod->name);
7547 }
7548
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)7549 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
7550 struct dyn_ftrace *rec)
7551 {
7552 struct ftrace_mod_func *mod_func;
7553 unsigned long symsize;
7554 unsigned long offset;
7555 char str[KSYM_SYMBOL_LEN];
7556 char *modname;
7557 const char *ret;
7558
7559 ret = kallsyms_lookup(rec->ip, &symsize, &offset, &modname, str);
7560 if (!ret)
7561 return;
7562
7563 mod_func = kmalloc(sizeof(*mod_func), GFP_KERNEL);
7564 if (!mod_func)
7565 return;
7566
7567 mod_func->name = kstrdup(str, GFP_KERNEL);
7568 if (!mod_func->name) {
7569 kfree(mod_func);
7570 return;
7571 }
7572
7573 mod_func->ip = rec->ip - offset;
7574 mod_func->size = symsize;
7575
7576 mod_map->num_funcs++;
7577
7578 list_add_rcu(&mod_func->list, &mod_map->funcs);
7579 }
7580
7581 static struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)7582 allocate_ftrace_mod_map(struct module *mod,
7583 unsigned long start, unsigned long end)
7584 {
7585 struct ftrace_mod_map *mod_map;
7586
7587 mod_map = kmalloc(sizeof(*mod_map), GFP_KERNEL);
7588 if (!mod_map)
7589 return NULL;
7590
7591 mod_map->mod = mod;
7592 mod_map->start_addr = start;
7593 mod_map->end_addr = end;
7594 mod_map->num_funcs = 0;
7595
7596 INIT_LIST_HEAD_RCU(&mod_map->funcs);
7597
7598 list_add_rcu(&mod_map->list, &ftrace_mod_maps);
7599
7600 return mod_map;
7601 }
7602
7603 static int
ftrace_func_address_lookup(struct ftrace_mod_map * mod_map,unsigned long addr,unsigned long * size,unsigned long * off,char * sym)7604 ftrace_func_address_lookup(struct ftrace_mod_map *mod_map,
7605 unsigned long addr, unsigned long *size,
7606 unsigned long *off, char *sym)
7607 {
7608 struct ftrace_mod_func *found_func = NULL;
7609 struct ftrace_mod_func *mod_func;
7610
7611 list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
7612 if (addr >= mod_func->ip &&
7613 addr < mod_func->ip + mod_func->size) {
7614 found_func = mod_func;
7615 break;
7616 }
7617 }
7618
7619 if (found_func) {
7620 if (size)
7621 *size = found_func->size;
7622 if (off)
7623 *off = addr - found_func->ip;
7624 return strscpy(sym, found_func->name, KSYM_NAME_LEN);
7625 }
7626
7627 return 0;
7628 }
7629
7630 int
ftrace_mod_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)7631 ftrace_mod_address_lookup(unsigned long addr, unsigned long *size,
7632 unsigned long *off, char **modname, char *sym)
7633 {
7634 struct ftrace_mod_map *mod_map;
7635 int ret = 0;
7636
7637 /* mod_map is freed via call_rcu() */
7638 preempt_disable();
7639 list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
7640 ret = ftrace_func_address_lookup(mod_map, addr, size, off, sym);
7641 if (ret) {
7642 if (modname)
7643 *modname = mod_map->mod->name;
7644 break;
7645 }
7646 }
7647 preempt_enable();
7648
7649 return ret;
7650 }
7651
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7652 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
7653 char *type, char *name,
7654 char *module_name, int *exported)
7655 {
7656 struct ftrace_mod_map *mod_map;
7657 struct ftrace_mod_func *mod_func;
7658 int ret;
7659
7660 preempt_disable();
7661 list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
7662
7663 if (symnum >= mod_map->num_funcs) {
7664 symnum -= mod_map->num_funcs;
7665 continue;
7666 }
7667
7668 list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
7669 if (symnum > 1) {
7670 symnum--;
7671 continue;
7672 }
7673
7674 *value = mod_func->ip;
7675 *type = 'T';
7676 strscpy(name, mod_func->name, KSYM_NAME_LEN);
7677 strscpy(module_name, mod_map->mod->name, MODULE_NAME_LEN);
7678 *exported = 1;
7679 preempt_enable();
7680 return 0;
7681 }
7682 WARN_ON(1);
7683 break;
7684 }
7685 ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
7686 module_name, exported);
7687 preempt_enable();
7688 return ret;
7689 }
7690
7691 #else
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)7692 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
7693 struct dyn_ftrace *rec) { }
7694 static inline struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)7695 allocate_ftrace_mod_map(struct module *mod,
7696 unsigned long start, unsigned long end)
7697 {
7698 return NULL;
7699 }
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7700 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
7701 char *type, char *name, char *module_name,
7702 int *exported)
7703 {
7704 int ret;
7705
7706 preempt_disable();
7707 ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
7708 module_name, exported);
7709 preempt_enable();
7710 return ret;
7711 }
7712 #endif /* CONFIG_MODULES */
7713
7714 struct ftrace_init_func {
7715 struct list_head list;
7716 unsigned long ip;
7717 };
7718
7719 /* Clear any init ips from hashes */
7720 static void
clear_func_from_hash(struct ftrace_init_func * func,struct ftrace_hash * hash)7721 clear_func_from_hash(struct ftrace_init_func *func, struct ftrace_hash *hash)
7722 {
7723 struct ftrace_func_entry *entry;
7724
7725 entry = ftrace_lookup_ip(hash, func->ip);
7726 /*
7727 * Do not allow this rec to match again.
7728 * Yeah, it may waste some memory, but will be removed
7729 * if/when the hash is modified again.
7730 */
7731 if (entry)
7732 entry->ip = 0;
7733 }
7734
7735 static void
clear_func_from_hashes(struct ftrace_init_func * func)7736 clear_func_from_hashes(struct ftrace_init_func *func)
7737 {
7738 struct trace_array *tr;
7739
7740 mutex_lock(&trace_types_lock);
7741 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7742 if (!tr->ops || !tr->ops->func_hash)
7743 continue;
7744 mutex_lock(&tr->ops->func_hash->regex_lock);
7745 clear_func_from_hash(func, tr->ops->func_hash->filter_hash);
7746 clear_func_from_hash(func, tr->ops->func_hash->notrace_hash);
7747 mutex_unlock(&tr->ops->func_hash->regex_lock);
7748 }
7749 mutex_unlock(&trace_types_lock);
7750 }
7751
add_to_clear_hash_list(struct list_head * clear_list,struct dyn_ftrace * rec)7752 static void add_to_clear_hash_list(struct list_head *clear_list,
7753 struct dyn_ftrace *rec)
7754 {
7755 struct ftrace_init_func *func;
7756
7757 func = kmalloc(sizeof(*func), GFP_KERNEL);
7758 if (!func) {
7759 MEM_FAIL(1, "alloc failure, ftrace filter could be stale\n");
7760 return;
7761 }
7762
7763 func->ip = rec->ip;
7764 list_add(&func->list, clear_list);
7765 }
7766
ftrace_free_mem(struct module * mod,void * start_ptr,void * end_ptr)7767 void ftrace_free_mem(struct module *mod, void *start_ptr, void *end_ptr)
7768 {
7769 unsigned long start = (unsigned long)(start_ptr);
7770 unsigned long end = (unsigned long)(end_ptr);
7771 struct ftrace_page **last_pg = &ftrace_pages_start;
7772 struct ftrace_page *tmp_page = NULL;
7773 struct ftrace_page *pg;
7774 struct dyn_ftrace *rec;
7775 struct dyn_ftrace key;
7776 struct ftrace_mod_map *mod_map = NULL;
7777 struct ftrace_init_func *func, *func_next;
7778 LIST_HEAD(clear_hash);
7779
7780 key.ip = start;
7781 key.flags = end; /* overload flags, as it is unsigned long */
7782
7783 mutex_lock(&ftrace_lock);
7784
7785 /*
7786 * If we are freeing module init memory, then check if
7787 * any tracer is active. If so, we need to save a mapping of
7788 * the module functions being freed with the address.
7789 */
7790 if (mod && ftrace_ops_list != &ftrace_list_end)
7791 mod_map = allocate_ftrace_mod_map(mod, start, end);
7792
7793 for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
7794 if (end < pg->records[0].ip ||
7795 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
7796 continue;
7797 again:
7798 rec = bsearch(&key, pg->records, pg->index,
7799 sizeof(struct dyn_ftrace),
7800 ftrace_cmp_recs);
7801 if (!rec)
7802 continue;
7803
7804 /* rec will be cleared from hashes after ftrace_lock unlock */
7805 add_to_clear_hash_list(&clear_hash, rec);
7806
7807 if (mod_map)
7808 save_ftrace_mod_rec(mod_map, rec);
7809
7810 pg->index--;
7811 ftrace_update_tot_cnt--;
7812 if (!pg->index) {
7813 *last_pg = pg->next;
7814 pg->next = tmp_page;
7815 tmp_page = pg;
7816 pg = container_of(last_pg, struct ftrace_page, next);
7817 if (!(*last_pg))
7818 ftrace_pages = pg;
7819 continue;
7820 }
7821 memmove(rec, rec + 1,
7822 (pg->index - (rec - pg->records)) * sizeof(*rec));
7823 /* More than one function may be in this block */
7824 goto again;
7825 }
7826 mutex_unlock(&ftrace_lock);
7827
7828 list_for_each_entry_safe(func, func_next, &clear_hash, list) {
7829 clear_func_from_hashes(func);
7830 kfree(func);
7831 }
7832 /* Need to synchronize with ftrace_location_range() */
7833 if (tmp_page) {
7834 synchronize_rcu();
7835 ftrace_free_pages(tmp_page);
7836 }
7837 }
7838
ftrace_free_init_mem(void)7839 void __init ftrace_free_init_mem(void)
7840 {
7841 void *start = (void *)(&__init_begin);
7842 void *end = (void *)(&__init_end);
7843
7844 ftrace_boot_snapshot();
7845
7846 ftrace_free_mem(NULL, start, end);
7847 }
7848
ftrace_dyn_arch_init(void)7849 int __init __weak ftrace_dyn_arch_init(void)
7850 {
7851 return 0;
7852 }
7853
ftrace_init(void)7854 void __init ftrace_init(void)
7855 {
7856 extern unsigned long __start_mcount_loc[];
7857 extern unsigned long __stop_mcount_loc[];
7858 unsigned long count, flags;
7859 int ret;
7860
7861 local_irq_save(flags);
7862 ret = ftrace_dyn_arch_init();
7863 local_irq_restore(flags);
7864 if (ret)
7865 goto failed;
7866
7867 count = __stop_mcount_loc - __start_mcount_loc;
7868 if (!count) {
7869 pr_info("ftrace: No functions to be traced?\n");
7870 goto failed;
7871 }
7872
7873 ret = ftrace_process_locs(NULL,
7874 __start_mcount_loc,
7875 __stop_mcount_loc);
7876 if (ret) {
7877 pr_warn("ftrace: failed to allocate entries for functions\n");
7878 goto failed;
7879 }
7880
7881 pr_info("ftrace: allocated %ld pages with %ld groups\n",
7882 ftrace_number_of_pages, ftrace_number_of_groups);
7883
7884 last_ftrace_enabled = ftrace_enabled = 1;
7885
7886 set_ftrace_early_filters();
7887
7888 return;
7889 failed:
7890 ftrace_disabled = 1;
7891 }
7892
7893 /* Do nothing if arch does not support this */
arch_ftrace_update_trampoline(struct ftrace_ops * ops)7894 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
7895 {
7896 }
7897
ftrace_update_trampoline(struct ftrace_ops * ops)7898 static void ftrace_update_trampoline(struct ftrace_ops *ops)
7899 {
7900 unsigned long trampoline = ops->trampoline;
7901
7902 arch_ftrace_update_trampoline(ops);
7903 if (ops->trampoline && ops->trampoline != trampoline &&
7904 (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP)) {
7905 /* Add to kallsyms before the perf events */
7906 ftrace_add_trampoline_to_kallsyms(ops);
7907 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
7908 ops->trampoline, ops->trampoline_size, false,
7909 FTRACE_TRAMPOLINE_SYM);
7910 /*
7911 * Record the perf text poke event after the ksymbol register
7912 * event.
7913 */
7914 perf_event_text_poke((void *)ops->trampoline, NULL, 0,
7915 (void *)ops->trampoline,
7916 ops->trampoline_size);
7917 }
7918 }
7919
ftrace_init_trace_array(struct trace_array * tr)7920 void ftrace_init_trace_array(struct trace_array *tr)
7921 {
7922 if (tr->flags & TRACE_ARRAY_FL_MOD_INIT)
7923 return;
7924
7925 INIT_LIST_HEAD(&tr->func_probes);
7926 INIT_LIST_HEAD(&tr->mod_trace);
7927 INIT_LIST_HEAD(&tr->mod_notrace);
7928
7929 tr->flags |= TRACE_ARRAY_FL_MOD_INIT;
7930 }
7931 #else
7932
7933 struct ftrace_ops global_ops = {
7934 .func = ftrace_stub,
7935 .flags = FTRACE_OPS_FL_INITIALIZED |
7936 FTRACE_OPS_FL_PID,
7937 };
7938
ftrace_nodyn_init(void)7939 static int __init ftrace_nodyn_init(void)
7940 {
7941 ftrace_enabled = 1;
7942 return 0;
7943 }
7944 core_initcall(ftrace_nodyn_init);
7945
ftrace_init_dyn_tracefs(struct dentry * d_tracer)7946 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
ftrace_startup_all(int command)7947 static inline void ftrace_startup_all(int command) { }
7948
ftrace_update_trampoline(struct ftrace_ops * ops)7949 static void ftrace_update_trampoline(struct ftrace_ops *ops)
7950 {
7951 }
7952
7953 #endif /* CONFIG_DYNAMIC_FTRACE */
7954
ftrace_init_global_array_ops(struct trace_array * tr)7955 __init void ftrace_init_global_array_ops(struct trace_array *tr)
7956 {
7957 tr->ops = &global_ops;
7958 if (!global_ops.private)
7959 global_ops.private = tr;
7960 ftrace_init_trace_array(tr);
7961 init_array_fgraph_ops(tr, tr->ops);
7962 }
7963
ftrace_init_array_ops(struct trace_array * tr,ftrace_func_t func)7964 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
7965 {
7966 /* If we filter on pids, update to use the pid function */
7967 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
7968 if (WARN_ON(tr->ops->func != ftrace_stub))
7969 printk("ftrace ops had %pS for function\n",
7970 tr->ops->func);
7971 }
7972 tr->ops->func = func;
7973 tr->ops->private = tr;
7974 }
7975
ftrace_reset_array_ops(struct trace_array * tr)7976 void ftrace_reset_array_ops(struct trace_array *tr)
7977 {
7978 tr->ops->func = ftrace_stub;
7979 }
7980
7981 static nokprobe_inline void
__ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * ignored,struct ftrace_regs * fregs)7982 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
7983 struct ftrace_ops *ignored, struct ftrace_regs *fregs)
7984 {
7985 struct pt_regs *regs = ftrace_get_regs(fregs);
7986 struct ftrace_ops *op;
7987 int bit;
7988
7989 /*
7990 * The ftrace_test_and_set_recursion() will disable preemption,
7991 * which is required since some of the ops may be dynamically
7992 * allocated, they must be freed after a synchronize_rcu().
7993 */
7994 bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
7995 if (bit < 0)
7996 return;
7997
7998 do_for_each_ftrace_op(op, ftrace_ops_list) {
7999 /* Stub functions don't need to be called nor tested */
8000 if (op->flags & FTRACE_OPS_FL_STUB)
8001 continue;
8002 /*
8003 * Check the following for each ops before calling their func:
8004 * if RCU flag is set, then rcu_is_watching() must be true
8005 * Otherwise test if the ip matches the ops filter
8006 *
8007 * If any of the above fails then the op->func() is not executed.
8008 */
8009 if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
8010 ftrace_ops_test(op, ip, regs)) {
8011 if (FTRACE_WARN_ON(!op->func)) {
8012 pr_warn("op=%p %pS\n", op, op);
8013 goto out;
8014 }
8015 op->func(ip, parent_ip, op, fregs);
8016 }
8017 } while_for_each_ftrace_op(op);
8018 out:
8019 trace_clear_recursion(bit);
8020 }
8021
8022 /*
8023 * Some archs only support passing ip and parent_ip. Even though
8024 * the list function ignores the op parameter, we do not want any
8025 * C side effects, where a function is called without the caller
8026 * sending a third parameter.
8027 * Archs are to support both the regs and ftrace_ops at the same time.
8028 * If they support ftrace_ops, it is assumed they support regs.
8029 * If call backs want to use regs, they must either check for regs
8030 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
8031 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
8032 * An architecture can pass partial regs with ftrace_ops and still
8033 * set the ARCH_SUPPORTS_FTRACE_OPS.
8034 *
8035 * In vmlinux.lds.h, ftrace_ops_list_func() is defined to be
8036 * arch_ftrace_ops_list_func.
8037 */
8038 #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)8039 void arch_ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
8040 struct ftrace_ops *op, struct ftrace_regs *fregs)
8041 {
8042 kmsan_unpoison_memory(fregs, ftrace_regs_size());
8043 __ftrace_ops_list_func(ip, parent_ip, NULL, fregs);
8044 }
8045 #else
arch_ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip)8046 void arch_ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
8047 {
8048 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
8049 }
8050 #endif
8051 NOKPROBE_SYMBOL(arch_ftrace_ops_list_func);
8052
8053 /*
8054 * If there's only one function registered but it does not support
8055 * recursion, needs RCU protection, then this function will be called
8056 * by the mcount trampoline.
8057 */
ftrace_ops_assist_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)8058 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
8059 struct ftrace_ops *op, struct ftrace_regs *fregs)
8060 {
8061 int bit;
8062
8063 bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
8064 if (bit < 0)
8065 return;
8066
8067 if (!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching())
8068 op->func(ip, parent_ip, op, fregs);
8069
8070 trace_clear_recursion(bit);
8071 }
8072 NOKPROBE_SYMBOL(ftrace_ops_assist_func);
8073
8074 /**
8075 * ftrace_ops_get_func - get the function a trampoline should call
8076 * @ops: the ops to get the function for
8077 *
8078 * Normally the mcount trampoline will call the ops->func, but there
8079 * are times that it should not. For example, if the ops does not
8080 * have its own recursion protection, then it should call the
8081 * ftrace_ops_assist_func() instead.
8082 *
8083 * Returns: the function that the trampoline should call for @ops.
8084 */
ftrace_ops_get_func(struct ftrace_ops * ops)8085 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
8086 {
8087 /*
8088 * If the function does not handle recursion or needs to be RCU safe,
8089 * then we need to call the assist handler.
8090 */
8091 if (ops->flags & (FTRACE_OPS_FL_RECURSION |
8092 FTRACE_OPS_FL_RCU))
8093 return ftrace_ops_assist_func;
8094
8095 return ops->func;
8096 }
8097
8098 static void
ftrace_filter_pid_sched_switch_probe(void * data,bool preempt,struct task_struct * prev,struct task_struct * next,unsigned int prev_state)8099 ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
8100 struct task_struct *prev,
8101 struct task_struct *next,
8102 unsigned int prev_state)
8103 {
8104 struct trace_array *tr = data;
8105 struct trace_pid_list *pid_list;
8106 struct trace_pid_list *no_pid_list;
8107
8108 pid_list = rcu_dereference_sched(tr->function_pids);
8109 no_pid_list = rcu_dereference_sched(tr->function_no_pids);
8110
8111 if (trace_ignore_this_task(pid_list, no_pid_list, next))
8112 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8113 FTRACE_PID_IGNORE);
8114 else
8115 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8116 next->pid);
8117 }
8118
8119 static void
ftrace_pid_follow_sched_process_fork(void * data,struct task_struct * self,struct task_struct * task)8120 ftrace_pid_follow_sched_process_fork(void *data,
8121 struct task_struct *self,
8122 struct task_struct *task)
8123 {
8124 struct trace_pid_list *pid_list;
8125 struct trace_array *tr = data;
8126
8127 pid_list = rcu_dereference_sched(tr->function_pids);
8128 trace_filter_add_remove_task(pid_list, self, task);
8129
8130 pid_list = rcu_dereference_sched(tr->function_no_pids);
8131 trace_filter_add_remove_task(pid_list, self, task);
8132 }
8133
8134 static void
ftrace_pid_follow_sched_process_exit(void * data,struct task_struct * task)8135 ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
8136 {
8137 struct trace_pid_list *pid_list;
8138 struct trace_array *tr = data;
8139
8140 pid_list = rcu_dereference_sched(tr->function_pids);
8141 trace_filter_add_remove_task(pid_list, NULL, task);
8142
8143 pid_list = rcu_dereference_sched(tr->function_no_pids);
8144 trace_filter_add_remove_task(pid_list, NULL, task);
8145 }
8146
ftrace_pid_follow_fork(struct trace_array * tr,bool enable)8147 void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
8148 {
8149 if (enable) {
8150 register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
8151 tr);
8152 register_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
8153 tr);
8154 } else {
8155 unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
8156 tr);
8157 unregister_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
8158 tr);
8159 }
8160 }
8161
clear_ftrace_pids(struct trace_array * tr,int type)8162 static void clear_ftrace_pids(struct trace_array *tr, int type)
8163 {
8164 struct trace_pid_list *pid_list;
8165 struct trace_pid_list *no_pid_list;
8166 int cpu;
8167
8168 pid_list = rcu_dereference_protected(tr->function_pids,
8169 lockdep_is_held(&ftrace_lock));
8170 no_pid_list = rcu_dereference_protected(tr->function_no_pids,
8171 lockdep_is_held(&ftrace_lock));
8172
8173 /* Make sure there's something to do */
8174 if (!pid_type_enabled(type, pid_list, no_pid_list))
8175 return;
8176
8177 /* See if the pids still need to be checked after this */
8178 if (!still_need_pid_events(type, pid_list, no_pid_list)) {
8179 unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
8180 for_each_possible_cpu(cpu)
8181 per_cpu_ptr(tr->array_buffer.data, cpu)->ftrace_ignore_pid = FTRACE_PID_TRACE;
8182 }
8183
8184 if (type & TRACE_PIDS)
8185 rcu_assign_pointer(tr->function_pids, NULL);
8186
8187 if (type & TRACE_NO_PIDS)
8188 rcu_assign_pointer(tr->function_no_pids, NULL);
8189
8190 /* Wait till all users are no longer using pid filtering */
8191 synchronize_rcu();
8192
8193 if ((type & TRACE_PIDS) && pid_list)
8194 trace_pid_list_free(pid_list);
8195
8196 if ((type & TRACE_NO_PIDS) && no_pid_list)
8197 trace_pid_list_free(no_pid_list);
8198 }
8199
ftrace_clear_pids(struct trace_array * tr)8200 void ftrace_clear_pids(struct trace_array *tr)
8201 {
8202 mutex_lock(&ftrace_lock);
8203
8204 clear_ftrace_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
8205
8206 mutex_unlock(&ftrace_lock);
8207 }
8208
ftrace_pid_reset(struct trace_array * tr,int type)8209 static void ftrace_pid_reset(struct trace_array *tr, int type)
8210 {
8211 mutex_lock(&ftrace_lock);
8212 clear_ftrace_pids(tr, type);
8213
8214 ftrace_update_pid_func();
8215 ftrace_startup_all(0);
8216
8217 mutex_unlock(&ftrace_lock);
8218 }
8219
8220 /* Greater than any max PID */
8221 #define FTRACE_NO_PIDS (void *)(PID_MAX_LIMIT + 1)
8222
fpid_start(struct seq_file * m,loff_t * pos)8223 static void *fpid_start(struct seq_file *m, loff_t *pos)
8224 __acquires(RCU)
8225 {
8226 struct trace_pid_list *pid_list;
8227 struct trace_array *tr = m->private;
8228
8229 mutex_lock(&ftrace_lock);
8230 rcu_read_lock_sched();
8231
8232 pid_list = rcu_dereference_sched(tr->function_pids);
8233
8234 if (!pid_list)
8235 return !(*pos) ? FTRACE_NO_PIDS : NULL;
8236
8237 return trace_pid_start(pid_list, pos);
8238 }
8239
fpid_next(struct seq_file * m,void * v,loff_t * pos)8240 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
8241 {
8242 struct trace_array *tr = m->private;
8243 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
8244
8245 if (v == FTRACE_NO_PIDS) {
8246 (*pos)++;
8247 return NULL;
8248 }
8249 return trace_pid_next(pid_list, v, pos);
8250 }
8251
fpid_stop(struct seq_file * m,void * p)8252 static void fpid_stop(struct seq_file *m, void *p)
8253 __releases(RCU)
8254 {
8255 rcu_read_unlock_sched();
8256 mutex_unlock(&ftrace_lock);
8257 }
8258
fpid_show(struct seq_file * m,void * v)8259 static int fpid_show(struct seq_file *m, void *v)
8260 {
8261 if (v == FTRACE_NO_PIDS) {
8262 seq_puts(m, "no pid\n");
8263 return 0;
8264 }
8265
8266 return trace_pid_show(m, v);
8267 }
8268
8269 static const struct seq_operations ftrace_pid_sops = {
8270 .start = fpid_start,
8271 .next = fpid_next,
8272 .stop = fpid_stop,
8273 .show = fpid_show,
8274 };
8275
fnpid_start(struct seq_file * m,loff_t * pos)8276 static void *fnpid_start(struct seq_file *m, loff_t *pos)
8277 __acquires(RCU)
8278 {
8279 struct trace_pid_list *pid_list;
8280 struct trace_array *tr = m->private;
8281
8282 mutex_lock(&ftrace_lock);
8283 rcu_read_lock_sched();
8284
8285 pid_list = rcu_dereference_sched(tr->function_no_pids);
8286
8287 if (!pid_list)
8288 return !(*pos) ? FTRACE_NO_PIDS : NULL;
8289
8290 return trace_pid_start(pid_list, pos);
8291 }
8292
fnpid_next(struct seq_file * m,void * v,loff_t * pos)8293 static void *fnpid_next(struct seq_file *m, void *v, loff_t *pos)
8294 {
8295 struct trace_array *tr = m->private;
8296 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_no_pids);
8297
8298 if (v == FTRACE_NO_PIDS) {
8299 (*pos)++;
8300 return NULL;
8301 }
8302 return trace_pid_next(pid_list, v, pos);
8303 }
8304
8305 static const struct seq_operations ftrace_no_pid_sops = {
8306 .start = fnpid_start,
8307 .next = fnpid_next,
8308 .stop = fpid_stop,
8309 .show = fpid_show,
8310 };
8311
pid_open(struct inode * inode,struct file * file,int type)8312 static int pid_open(struct inode *inode, struct file *file, int type)
8313 {
8314 const struct seq_operations *seq_ops;
8315 struct trace_array *tr = inode->i_private;
8316 struct seq_file *m;
8317 int ret = 0;
8318
8319 ret = tracing_check_open_get_tr(tr);
8320 if (ret)
8321 return ret;
8322
8323 if ((file->f_mode & FMODE_WRITE) &&
8324 (file->f_flags & O_TRUNC))
8325 ftrace_pid_reset(tr, type);
8326
8327 switch (type) {
8328 case TRACE_PIDS:
8329 seq_ops = &ftrace_pid_sops;
8330 break;
8331 case TRACE_NO_PIDS:
8332 seq_ops = &ftrace_no_pid_sops;
8333 break;
8334 default:
8335 trace_array_put(tr);
8336 WARN_ON_ONCE(1);
8337 return -EINVAL;
8338 }
8339
8340 ret = seq_open(file, seq_ops);
8341 if (ret < 0) {
8342 trace_array_put(tr);
8343 } else {
8344 m = file->private_data;
8345 /* copy tr over to seq ops */
8346 m->private = tr;
8347 }
8348
8349 return ret;
8350 }
8351
8352 static int
ftrace_pid_open(struct inode * inode,struct file * file)8353 ftrace_pid_open(struct inode *inode, struct file *file)
8354 {
8355 return pid_open(inode, file, TRACE_PIDS);
8356 }
8357
8358 static int
ftrace_no_pid_open(struct inode * inode,struct file * file)8359 ftrace_no_pid_open(struct inode *inode, struct file *file)
8360 {
8361 return pid_open(inode, file, TRACE_NO_PIDS);
8362 }
8363
ignore_task_cpu(void * data)8364 static void ignore_task_cpu(void *data)
8365 {
8366 struct trace_array *tr = data;
8367 struct trace_pid_list *pid_list;
8368 struct trace_pid_list *no_pid_list;
8369
8370 /*
8371 * This function is called by on_each_cpu() while the
8372 * event_mutex is held.
8373 */
8374 pid_list = rcu_dereference_protected(tr->function_pids,
8375 mutex_is_locked(&ftrace_lock));
8376 no_pid_list = rcu_dereference_protected(tr->function_no_pids,
8377 mutex_is_locked(&ftrace_lock));
8378
8379 if (trace_ignore_this_task(pid_list, no_pid_list, current))
8380 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8381 FTRACE_PID_IGNORE);
8382 else
8383 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8384 current->pid);
8385 }
8386
8387 static ssize_t
pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos,int type)8388 pid_write(struct file *filp, const char __user *ubuf,
8389 size_t cnt, loff_t *ppos, int type)
8390 {
8391 struct seq_file *m = filp->private_data;
8392 struct trace_array *tr = m->private;
8393 struct trace_pid_list *filtered_pids;
8394 struct trace_pid_list *other_pids;
8395 struct trace_pid_list *pid_list;
8396 ssize_t ret;
8397
8398 if (!cnt)
8399 return 0;
8400
8401 guard(mutex)(&ftrace_lock);
8402
8403 switch (type) {
8404 case TRACE_PIDS:
8405 filtered_pids = rcu_dereference_protected(tr->function_pids,
8406 lockdep_is_held(&ftrace_lock));
8407 other_pids = rcu_dereference_protected(tr->function_no_pids,
8408 lockdep_is_held(&ftrace_lock));
8409 break;
8410 case TRACE_NO_PIDS:
8411 filtered_pids = rcu_dereference_protected(tr->function_no_pids,
8412 lockdep_is_held(&ftrace_lock));
8413 other_pids = rcu_dereference_protected(tr->function_pids,
8414 lockdep_is_held(&ftrace_lock));
8415 break;
8416 default:
8417 WARN_ON_ONCE(1);
8418 return -EINVAL;
8419 }
8420
8421 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
8422 if (ret < 0)
8423 return ret;
8424
8425 switch (type) {
8426 case TRACE_PIDS:
8427 rcu_assign_pointer(tr->function_pids, pid_list);
8428 break;
8429 case TRACE_NO_PIDS:
8430 rcu_assign_pointer(tr->function_no_pids, pid_list);
8431 break;
8432 }
8433
8434
8435 if (filtered_pids) {
8436 synchronize_rcu();
8437 trace_pid_list_free(filtered_pids);
8438 } else if (pid_list && !other_pids) {
8439 /* Register a probe to set whether to ignore the tracing of a task */
8440 register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
8441 }
8442
8443 /*
8444 * Ignoring of pids is done at task switch. But we have to
8445 * check for those tasks that are currently running.
8446 * Always do this in case a pid was appended or removed.
8447 */
8448 on_each_cpu(ignore_task_cpu, tr, 1);
8449
8450 ftrace_update_pid_func();
8451 ftrace_startup_all(0);
8452
8453 *ppos += ret;
8454
8455 return ret;
8456 }
8457
8458 static ssize_t
ftrace_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)8459 ftrace_pid_write(struct file *filp, const char __user *ubuf,
8460 size_t cnt, loff_t *ppos)
8461 {
8462 return pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
8463 }
8464
8465 static ssize_t
ftrace_no_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)8466 ftrace_no_pid_write(struct file *filp, const char __user *ubuf,
8467 size_t cnt, loff_t *ppos)
8468 {
8469 return pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
8470 }
8471
8472 static int
ftrace_pid_release(struct inode * inode,struct file * file)8473 ftrace_pid_release(struct inode *inode, struct file *file)
8474 {
8475 struct trace_array *tr = inode->i_private;
8476
8477 trace_array_put(tr);
8478
8479 return seq_release(inode, file);
8480 }
8481
8482 static const struct file_operations ftrace_pid_fops = {
8483 .open = ftrace_pid_open,
8484 .write = ftrace_pid_write,
8485 .read = seq_read,
8486 .llseek = tracing_lseek,
8487 .release = ftrace_pid_release,
8488 };
8489
8490 static const struct file_operations ftrace_no_pid_fops = {
8491 .open = ftrace_no_pid_open,
8492 .write = ftrace_no_pid_write,
8493 .read = seq_read,
8494 .llseek = tracing_lseek,
8495 .release = ftrace_pid_release,
8496 };
8497
ftrace_init_tracefs(struct trace_array * tr,struct dentry * d_tracer)8498 void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
8499 {
8500 trace_create_file("set_ftrace_pid", TRACE_MODE_WRITE, d_tracer,
8501 tr, &ftrace_pid_fops);
8502 trace_create_file("set_ftrace_notrace_pid", TRACE_MODE_WRITE,
8503 d_tracer, tr, &ftrace_no_pid_fops);
8504 }
8505
ftrace_init_tracefs_toplevel(struct trace_array * tr,struct dentry * d_tracer)8506 void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
8507 struct dentry *d_tracer)
8508 {
8509 /* Only the top level directory has the dyn_tracefs and profile */
8510 WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
8511
8512 ftrace_init_dyn_tracefs(d_tracer);
8513 ftrace_profile_tracefs(d_tracer);
8514 }
8515
8516 /**
8517 * ftrace_kill - kill ftrace
8518 *
8519 * This function should be used by panic code. It stops ftrace
8520 * but in a not so nice way. If you need to simply kill ftrace
8521 * from a non-atomic section, use ftrace_kill.
8522 */
ftrace_kill(void)8523 void ftrace_kill(void)
8524 {
8525 ftrace_disabled = 1;
8526 ftrace_enabled = 0;
8527 ftrace_trace_function = ftrace_stub;
8528 kprobe_ftrace_kill();
8529 }
8530
8531 /**
8532 * ftrace_is_dead - Test if ftrace is dead or not.
8533 *
8534 * Returns: 1 if ftrace is "dead", zero otherwise.
8535 */
ftrace_is_dead(void)8536 int ftrace_is_dead(void)
8537 {
8538 return ftrace_disabled;
8539 }
8540
8541 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
8542 /*
8543 * When registering ftrace_ops with IPMODIFY, it is necessary to make sure
8544 * it doesn't conflict with any direct ftrace_ops. If there is existing
8545 * direct ftrace_ops on a kernel function being patched, call
8546 * FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_PEER on it to enable sharing.
8547 *
8548 * @ops: ftrace_ops being registered.
8549 *
8550 * Returns:
8551 * 0 on success;
8552 * Negative on failure.
8553 */
prepare_direct_functions_for_ipmodify(struct ftrace_ops * ops)8554 static int prepare_direct_functions_for_ipmodify(struct ftrace_ops *ops)
8555 {
8556 struct ftrace_func_entry *entry;
8557 struct ftrace_hash *hash;
8558 struct ftrace_ops *op;
8559 int size, i, ret;
8560
8561 lockdep_assert_held_once(&direct_mutex);
8562
8563 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
8564 return 0;
8565
8566 hash = ops->func_hash->filter_hash;
8567 size = 1 << hash->size_bits;
8568 for (i = 0; i < size; i++) {
8569 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
8570 unsigned long ip = entry->ip;
8571 bool found_op = false;
8572
8573 mutex_lock(&ftrace_lock);
8574 do_for_each_ftrace_op(op, ftrace_ops_list) {
8575 if (!(op->flags & FTRACE_OPS_FL_DIRECT))
8576 continue;
8577 if (ops_references_ip(op, ip)) {
8578 found_op = true;
8579 break;
8580 }
8581 } while_for_each_ftrace_op(op);
8582 mutex_unlock(&ftrace_lock);
8583
8584 if (found_op) {
8585 if (!op->ops_func)
8586 return -EBUSY;
8587
8588 ret = op->ops_func(op, FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_PEER);
8589 if (ret)
8590 return ret;
8591 }
8592 }
8593 }
8594
8595 return 0;
8596 }
8597
8598 /*
8599 * Similar to prepare_direct_functions_for_ipmodify, clean up after ops
8600 * with IPMODIFY is unregistered. The cleanup is optional for most DIRECT
8601 * ops.
8602 */
cleanup_direct_functions_after_ipmodify(struct ftrace_ops * ops)8603 static void cleanup_direct_functions_after_ipmodify(struct ftrace_ops *ops)
8604 {
8605 struct ftrace_func_entry *entry;
8606 struct ftrace_hash *hash;
8607 struct ftrace_ops *op;
8608 int size, i;
8609
8610 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
8611 return;
8612
8613 mutex_lock(&direct_mutex);
8614
8615 hash = ops->func_hash->filter_hash;
8616 size = 1 << hash->size_bits;
8617 for (i = 0; i < size; i++) {
8618 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
8619 unsigned long ip = entry->ip;
8620 bool found_op = false;
8621
8622 mutex_lock(&ftrace_lock);
8623 do_for_each_ftrace_op(op, ftrace_ops_list) {
8624 if (!(op->flags & FTRACE_OPS_FL_DIRECT))
8625 continue;
8626 if (ops_references_ip(op, ip)) {
8627 found_op = true;
8628 break;
8629 }
8630 } while_for_each_ftrace_op(op);
8631 mutex_unlock(&ftrace_lock);
8632
8633 /* The cleanup is optional, ignore any errors */
8634 if (found_op && op->ops_func)
8635 op->ops_func(op, FTRACE_OPS_CMD_DISABLE_SHARE_IPMODIFY_PEER);
8636 }
8637 }
8638 mutex_unlock(&direct_mutex);
8639 }
8640
8641 #define lock_direct_mutex() mutex_lock(&direct_mutex)
8642 #define unlock_direct_mutex() mutex_unlock(&direct_mutex)
8643
8644 #else /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
8645
prepare_direct_functions_for_ipmodify(struct ftrace_ops * ops)8646 static int prepare_direct_functions_for_ipmodify(struct ftrace_ops *ops)
8647 {
8648 return 0;
8649 }
8650
cleanup_direct_functions_after_ipmodify(struct ftrace_ops * ops)8651 static void cleanup_direct_functions_after_ipmodify(struct ftrace_ops *ops)
8652 {
8653 }
8654
8655 #define lock_direct_mutex() do { } while (0)
8656 #define unlock_direct_mutex() do { } while (0)
8657
8658 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
8659
8660 /*
8661 * Similar to register_ftrace_function, except we don't lock direct_mutex.
8662 */
register_ftrace_function_nolock(struct ftrace_ops * ops)8663 static int register_ftrace_function_nolock(struct ftrace_ops *ops)
8664 {
8665 int ret;
8666
8667 ftrace_ops_init(ops);
8668
8669 mutex_lock(&ftrace_lock);
8670
8671 ret = ftrace_startup(ops, 0);
8672
8673 mutex_unlock(&ftrace_lock);
8674
8675 return ret;
8676 }
8677
8678 /**
8679 * register_ftrace_function - register a function for profiling
8680 * @ops: ops structure that holds the function for profiling.
8681 *
8682 * Register a function to be called by all functions in the
8683 * kernel.
8684 *
8685 * Note: @ops->func and all the functions it calls must be labeled
8686 * with "notrace", otherwise it will go into a
8687 * recursive loop.
8688 */
register_ftrace_function(struct ftrace_ops * ops)8689 int register_ftrace_function(struct ftrace_ops *ops)
8690 {
8691 int ret;
8692
8693 lock_direct_mutex();
8694 ret = prepare_direct_functions_for_ipmodify(ops);
8695 if (ret < 0)
8696 goto out_unlock;
8697
8698 ret = register_ftrace_function_nolock(ops);
8699
8700 out_unlock:
8701 unlock_direct_mutex();
8702 return ret;
8703 }
8704 EXPORT_SYMBOL_GPL(register_ftrace_function);
8705
8706 /**
8707 * unregister_ftrace_function - unregister a function for profiling.
8708 * @ops: ops structure that holds the function to unregister
8709 *
8710 * Unregister a function that was added to be called by ftrace profiling.
8711 */
unregister_ftrace_function(struct ftrace_ops * ops)8712 int unregister_ftrace_function(struct ftrace_ops *ops)
8713 {
8714 int ret;
8715
8716 mutex_lock(&ftrace_lock);
8717 ret = ftrace_shutdown(ops, 0);
8718 mutex_unlock(&ftrace_lock);
8719
8720 cleanup_direct_functions_after_ipmodify(ops);
8721 return ret;
8722 }
8723 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
8724
symbols_cmp(const void * a,const void * b)8725 static int symbols_cmp(const void *a, const void *b)
8726 {
8727 const char **str_a = (const char **) a;
8728 const char **str_b = (const char **) b;
8729
8730 return strcmp(*str_a, *str_b);
8731 }
8732
8733 struct kallsyms_data {
8734 unsigned long *addrs;
8735 const char **syms;
8736 size_t cnt;
8737 size_t found;
8738 };
8739
8740 /* This function gets called for all kernel and module symbols
8741 * and returns 1 in case we resolved all the requested symbols,
8742 * 0 otherwise.
8743 */
kallsyms_callback(void * data,const char * name,unsigned long addr)8744 static int kallsyms_callback(void *data, const char *name, unsigned long addr)
8745 {
8746 struct kallsyms_data *args = data;
8747 const char **sym;
8748 int idx;
8749
8750 sym = bsearch(&name, args->syms, args->cnt, sizeof(*args->syms), symbols_cmp);
8751 if (!sym)
8752 return 0;
8753
8754 idx = sym - args->syms;
8755 if (args->addrs[idx])
8756 return 0;
8757
8758 if (!ftrace_location(addr))
8759 return 0;
8760
8761 args->addrs[idx] = addr;
8762 args->found++;
8763 return args->found == args->cnt ? 1 : 0;
8764 }
8765
8766 /**
8767 * ftrace_lookup_symbols - Lookup addresses for array of symbols
8768 *
8769 * @sorted_syms: array of symbols pointers symbols to resolve,
8770 * must be alphabetically sorted
8771 * @cnt: number of symbols/addresses in @syms/@addrs arrays
8772 * @addrs: array for storing resulting addresses
8773 *
8774 * This function looks up addresses for array of symbols provided in
8775 * @syms array (must be alphabetically sorted) and stores them in
8776 * @addrs array, which needs to be big enough to store at least @cnt
8777 * addresses.
8778 *
8779 * Returns: 0 if all provided symbols are found, -ESRCH otherwise.
8780 */
ftrace_lookup_symbols(const char ** sorted_syms,size_t cnt,unsigned long * addrs)8781 int ftrace_lookup_symbols(const char **sorted_syms, size_t cnt, unsigned long *addrs)
8782 {
8783 struct kallsyms_data args;
8784 int found_all;
8785
8786 memset(addrs, 0, sizeof(*addrs) * cnt);
8787 args.addrs = addrs;
8788 args.syms = sorted_syms;
8789 args.cnt = cnt;
8790 args.found = 0;
8791
8792 found_all = kallsyms_on_each_symbol(kallsyms_callback, &args);
8793 if (found_all)
8794 return 0;
8795 found_all = module_kallsyms_on_each_symbol(NULL, kallsyms_callback, &args);
8796 return found_all ? 0 : -ESRCH;
8797 }
8798
8799 #ifdef CONFIG_SYSCTL
8800
8801 #ifdef CONFIG_DYNAMIC_FTRACE
ftrace_startup_sysctl(void)8802 static void ftrace_startup_sysctl(void)
8803 {
8804 int command;
8805
8806 if (unlikely(ftrace_disabled))
8807 return;
8808
8809 /* Force update next time */
8810 saved_ftrace_func = NULL;
8811 /* ftrace_start_up is true if we want ftrace running */
8812 if (ftrace_start_up) {
8813 command = FTRACE_UPDATE_CALLS;
8814 if (ftrace_graph_active)
8815 command |= FTRACE_START_FUNC_RET;
8816 ftrace_startup_enable(command);
8817 }
8818 }
8819
ftrace_shutdown_sysctl(void)8820 static void ftrace_shutdown_sysctl(void)
8821 {
8822 int command;
8823
8824 if (unlikely(ftrace_disabled))
8825 return;
8826
8827 /* ftrace_start_up is true if ftrace is running */
8828 if (ftrace_start_up) {
8829 command = FTRACE_DISABLE_CALLS;
8830 if (ftrace_graph_active)
8831 command |= FTRACE_STOP_FUNC_RET;
8832 ftrace_run_update_code(command);
8833 }
8834 }
8835 #else
8836 # define ftrace_startup_sysctl() do { } while (0)
8837 # define ftrace_shutdown_sysctl() do { } while (0)
8838 #endif /* CONFIG_DYNAMIC_FTRACE */
8839
is_permanent_ops_registered(void)8840 static bool is_permanent_ops_registered(void)
8841 {
8842 struct ftrace_ops *op;
8843
8844 do_for_each_ftrace_op(op, ftrace_ops_list) {
8845 if (op->flags & FTRACE_OPS_FL_PERMANENT)
8846 return true;
8847 } while_for_each_ftrace_op(op);
8848
8849 return false;
8850 }
8851
8852 static int
ftrace_enable_sysctl(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)8853 ftrace_enable_sysctl(const struct ctl_table *table, int write,
8854 void *buffer, size_t *lenp, loff_t *ppos)
8855 {
8856 int ret;
8857
8858 guard(mutex)(&ftrace_lock);
8859
8860 if (unlikely(ftrace_disabled))
8861 return -ENODEV;
8862
8863 ret = proc_dointvec(table, write, buffer, lenp, ppos);
8864
8865 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
8866 return ret;
8867
8868 if (ftrace_enabled) {
8869
8870 /* we are starting ftrace again */
8871 if (rcu_dereference_protected(ftrace_ops_list,
8872 lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
8873 update_ftrace_function();
8874
8875 ftrace_startup_sysctl();
8876
8877 } else {
8878 if (is_permanent_ops_registered()) {
8879 ftrace_enabled = true;
8880 return -EBUSY;
8881 }
8882
8883 /* stopping ftrace calls (just send to ftrace_stub) */
8884 ftrace_trace_function = ftrace_stub;
8885
8886 ftrace_shutdown_sysctl();
8887 }
8888
8889 last_ftrace_enabled = !!ftrace_enabled;
8890 return 0;
8891 }
8892
8893 static const struct ctl_table ftrace_sysctls[] = {
8894 {
8895 .procname = "ftrace_enabled",
8896 .data = &ftrace_enabled,
8897 .maxlen = sizeof(int),
8898 .mode = 0644,
8899 .proc_handler = ftrace_enable_sysctl,
8900 },
8901 };
8902
ftrace_sysctl_init(void)8903 static int __init ftrace_sysctl_init(void)
8904 {
8905 register_sysctl_init("kernel", ftrace_sysctls);
8906 return 0;
8907 }
8908 late_initcall(ftrace_sysctl_init);
8909 #endif
8910