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 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 struct ftrace_func_probe {
1046 struct ftrace_probe_ops *probe_ops;
1047 struct ftrace_ops ops;
1048 struct trace_array *tr;
1049 struct list_head list;
1050 void *data;
1051 int ref;
1052 };
1053
1054 /*
1055 * We make these constant because no one should touch them,
1056 * but they are used as the default "empty hash", to avoid allocating
1057 * it all the time. These are in a read only section such that if
1058 * anyone does try to modify it, it will cause an exception.
1059 */
1060 static const struct hlist_head empty_buckets[1];
1061 static const struct ftrace_hash empty_hash = {
1062 .buckets = (struct hlist_head *)empty_buckets,
1063 };
1064 #define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
1065
1066 struct ftrace_ops global_ops = {
1067 .func = ftrace_stub,
1068 .local_hash.notrace_hash = EMPTY_HASH,
1069 .local_hash.filter_hash = EMPTY_HASH,
1070 INIT_OPS_HASH(global_ops)
1071 .flags = FTRACE_OPS_FL_INITIALIZED |
1072 FTRACE_OPS_FL_PID,
1073 };
1074
1075 /*
1076 * Used by the stack unwinder to know about dynamic ftrace trampolines.
1077 */
ftrace_ops_trampoline(unsigned long addr)1078 struct ftrace_ops *ftrace_ops_trampoline(unsigned long addr)
1079 {
1080 struct ftrace_ops *op = NULL;
1081
1082 /*
1083 * Some of the ops may be dynamically allocated,
1084 * they are freed after a synchronize_rcu().
1085 */
1086 preempt_disable_notrace();
1087
1088 do_for_each_ftrace_op(op, ftrace_ops_list) {
1089 /*
1090 * This is to check for dynamically allocated trampolines.
1091 * Trampolines that are in kernel text will have
1092 * core_kernel_text() return true.
1093 */
1094 if (op->trampoline && op->trampoline_size)
1095 if (addr >= op->trampoline &&
1096 addr < op->trampoline + op->trampoline_size) {
1097 preempt_enable_notrace();
1098 return op;
1099 }
1100 } while_for_each_ftrace_op(op);
1101 preempt_enable_notrace();
1102
1103 return NULL;
1104 }
1105
1106 /*
1107 * This is used by __kernel_text_address() to return true if the
1108 * address is on a dynamically allocated trampoline that would
1109 * not return true for either core_kernel_text() or
1110 * is_module_text_address().
1111 */
is_ftrace_trampoline(unsigned long addr)1112 bool is_ftrace_trampoline(unsigned long addr)
1113 {
1114 return ftrace_ops_trampoline(addr) != NULL;
1115 }
1116
1117 struct ftrace_page {
1118 struct ftrace_page *next;
1119 struct dyn_ftrace *records;
1120 int index;
1121 int order;
1122 };
1123
1124 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1125 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1126
1127 static struct ftrace_page *ftrace_pages_start;
1128 static struct ftrace_page *ftrace_pages;
1129
1130 static __always_inline unsigned long
ftrace_hash_key(struct ftrace_hash * hash,unsigned long ip)1131 ftrace_hash_key(struct ftrace_hash *hash, unsigned long ip)
1132 {
1133 if (hash->size_bits > 0)
1134 return hash_long(ip, hash->size_bits);
1135
1136 return 0;
1137 }
1138
1139 /* Only use this function if ftrace_hash_empty() has already been tested */
1140 static __always_inline struct ftrace_func_entry *
__ftrace_lookup_ip(struct ftrace_hash * hash,unsigned long ip)1141 __ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1142 {
1143 unsigned long key;
1144 struct ftrace_func_entry *entry;
1145 struct hlist_head *hhd;
1146
1147 key = ftrace_hash_key(hash, ip);
1148 hhd = &hash->buckets[key];
1149
1150 hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1151 if (entry->ip == ip)
1152 return entry;
1153 }
1154 return NULL;
1155 }
1156
1157 /**
1158 * ftrace_lookup_ip - Test to see if an ip exists in an ftrace_hash
1159 * @hash: The hash to look at
1160 * @ip: The instruction pointer to test
1161 *
1162 * Search a given @hash to see if a given instruction pointer (@ip)
1163 * exists in it.
1164 *
1165 * Returns: the entry that holds the @ip if found. NULL otherwise.
1166 */
1167 struct ftrace_func_entry *
ftrace_lookup_ip(struct ftrace_hash * hash,unsigned long ip)1168 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1169 {
1170 if (ftrace_hash_empty(hash))
1171 return NULL;
1172
1173 return __ftrace_lookup_ip(hash, ip);
1174 }
1175
__add_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1176 static void __add_hash_entry(struct ftrace_hash *hash,
1177 struct ftrace_func_entry *entry)
1178 {
1179 struct hlist_head *hhd;
1180 unsigned long key;
1181
1182 key = ftrace_hash_key(hash, entry->ip);
1183 hhd = &hash->buckets[key];
1184 hlist_add_head(&entry->hlist, hhd);
1185 hash->count++;
1186 }
1187
1188 static struct ftrace_func_entry *
add_hash_entry(struct ftrace_hash * hash,unsigned long ip)1189 add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1190 {
1191 struct ftrace_func_entry *entry;
1192
1193 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1194 if (!entry)
1195 return NULL;
1196
1197 entry->ip = ip;
1198 __add_hash_entry(hash, entry);
1199
1200 return entry;
1201 }
1202
1203 static void
free_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1204 free_hash_entry(struct ftrace_hash *hash,
1205 struct ftrace_func_entry *entry)
1206 {
1207 hlist_del(&entry->hlist);
1208 kfree(entry);
1209 hash->count--;
1210 }
1211
1212 static void
remove_hash_entry(struct ftrace_hash * hash,struct ftrace_func_entry * entry)1213 remove_hash_entry(struct ftrace_hash *hash,
1214 struct ftrace_func_entry *entry)
1215 {
1216 hlist_del_rcu(&entry->hlist);
1217 hash->count--;
1218 }
1219
ftrace_hash_clear(struct ftrace_hash * hash)1220 static void ftrace_hash_clear(struct ftrace_hash *hash)
1221 {
1222 struct hlist_head *hhd;
1223 struct hlist_node *tn;
1224 struct ftrace_func_entry *entry;
1225 int size = 1 << hash->size_bits;
1226 int i;
1227
1228 if (!hash->count)
1229 return;
1230
1231 for (i = 0; i < size; i++) {
1232 hhd = &hash->buckets[i];
1233 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1234 free_hash_entry(hash, entry);
1235 }
1236 FTRACE_WARN_ON(hash->count);
1237 }
1238
free_ftrace_mod(struct ftrace_mod_load * ftrace_mod)1239 static void free_ftrace_mod(struct ftrace_mod_load *ftrace_mod)
1240 {
1241 list_del(&ftrace_mod->list);
1242 kfree(ftrace_mod->module);
1243 kfree(ftrace_mod->func);
1244 kfree(ftrace_mod);
1245 }
1246
clear_ftrace_mod_list(struct list_head * head)1247 static void clear_ftrace_mod_list(struct list_head *head)
1248 {
1249 struct ftrace_mod_load *p, *n;
1250
1251 /* stack tracer isn't supported yet */
1252 if (!head)
1253 return;
1254
1255 mutex_lock(&ftrace_lock);
1256 list_for_each_entry_safe(p, n, head, list)
1257 free_ftrace_mod(p);
1258 mutex_unlock(&ftrace_lock);
1259 }
1260
free_ftrace_hash(struct ftrace_hash * hash)1261 static void free_ftrace_hash(struct ftrace_hash *hash)
1262 {
1263 if (!hash || hash == EMPTY_HASH)
1264 return;
1265 ftrace_hash_clear(hash);
1266 kfree(hash->buckets);
1267 kfree(hash);
1268 }
1269
__free_ftrace_hash_rcu(struct rcu_head * rcu)1270 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1271 {
1272 struct ftrace_hash *hash;
1273
1274 hash = container_of(rcu, struct ftrace_hash, rcu);
1275 free_ftrace_hash(hash);
1276 }
1277
free_ftrace_hash_rcu(struct ftrace_hash * hash)1278 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1279 {
1280 if (!hash || hash == EMPTY_HASH)
1281 return;
1282 call_rcu(&hash->rcu, __free_ftrace_hash_rcu);
1283 }
1284
1285 /**
1286 * ftrace_free_filter - remove all filters for an ftrace_ops
1287 * @ops: the ops to remove the filters from
1288 */
ftrace_free_filter(struct ftrace_ops * ops)1289 void ftrace_free_filter(struct ftrace_ops *ops)
1290 {
1291 ftrace_ops_init(ops);
1292 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
1293 return;
1294 free_ftrace_hash(ops->func_hash->filter_hash);
1295 free_ftrace_hash(ops->func_hash->notrace_hash);
1296 ops->func_hash->filter_hash = EMPTY_HASH;
1297 ops->func_hash->notrace_hash = EMPTY_HASH;
1298 }
1299 EXPORT_SYMBOL_GPL(ftrace_free_filter);
1300
alloc_ftrace_hash(int size_bits)1301 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1302 {
1303 struct ftrace_hash *hash;
1304 int size;
1305
1306 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1307 if (!hash)
1308 return NULL;
1309
1310 size = 1 << size_bits;
1311 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1312
1313 if (!hash->buckets) {
1314 kfree(hash);
1315 return NULL;
1316 }
1317
1318 hash->size_bits = size_bits;
1319
1320 return hash;
1321 }
1322
1323 /* 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)1324 static int ftrace_add_mod(struct trace_array *tr,
1325 const char *func, const char *module,
1326 int enable)
1327 {
1328 struct ftrace_mod_load *ftrace_mod;
1329 struct list_head *mod_head = enable ? &tr->mod_trace : &tr->mod_notrace;
1330
1331 ftrace_mod = kzalloc(sizeof(*ftrace_mod), GFP_KERNEL);
1332 if (!ftrace_mod)
1333 return -ENOMEM;
1334
1335 INIT_LIST_HEAD(&ftrace_mod->list);
1336 ftrace_mod->func = kstrdup(func, GFP_KERNEL);
1337 ftrace_mod->module = kstrdup(module, GFP_KERNEL);
1338 ftrace_mod->enable = enable;
1339
1340 if (!ftrace_mod->func || !ftrace_mod->module)
1341 goto out_free;
1342
1343 list_add(&ftrace_mod->list, mod_head);
1344
1345 return 0;
1346
1347 out_free:
1348 free_ftrace_mod(ftrace_mod);
1349
1350 return -ENOMEM;
1351 }
1352
1353 static struct ftrace_hash *
alloc_and_copy_ftrace_hash(int size_bits,struct ftrace_hash * hash)1354 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1355 {
1356 struct ftrace_func_entry *entry;
1357 struct ftrace_hash *new_hash;
1358 int size;
1359 int i;
1360
1361 new_hash = alloc_ftrace_hash(size_bits);
1362 if (!new_hash)
1363 return NULL;
1364
1365 if (hash)
1366 new_hash->flags = hash->flags;
1367
1368 /* Empty hash? */
1369 if (ftrace_hash_empty(hash))
1370 return new_hash;
1371
1372 size = 1 << hash->size_bits;
1373 for (i = 0; i < size; i++) {
1374 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1375 if (add_hash_entry(new_hash, entry->ip) == NULL)
1376 goto free_hash;
1377 }
1378 }
1379
1380 FTRACE_WARN_ON(new_hash->count != hash->count);
1381
1382 return new_hash;
1383
1384 free_hash:
1385 free_ftrace_hash(new_hash);
1386 return NULL;
1387 }
1388
1389 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops);
1390 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops);
1391
1392 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1393 struct ftrace_hash *new_hash);
1394
1395 /*
1396 * Allocate a new hash and remove entries from @src and move them to the new hash.
1397 * On success, the @src hash will be empty and should be freed.
1398 */
__move_hash(struct ftrace_hash * src,int size)1399 static struct ftrace_hash *__move_hash(struct ftrace_hash *src, int size)
1400 {
1401 struct ftrace_func_entry *entry;
1402 struct ftrace_hash *new_hash;
1403 struct hlist_head *hhd;
1404 struct hlist_node *tn;
1405 int bits = 0;
1406 int i;
1407
1408 /*
1409 * Use around half the size (max bit of it), but
1410 * a minimum of 2 is fine (as size of 0 or 1 both give 1 for bits).
1411 */
1412 bits = fls(size / 2);
1413
1414 /* Don't allocate too much */
1415 if (bits > FTRACE_HASH_MAX_BITS)
1416 bits = FTRACE_HASH_MAX_BITS;
1417
1418 new_hash = alloc_ftrace_hash(bits);
1419 if (!new_hash)
1420 return NULL;
1421
1422 new_hash->flags = src->flags;
1423
1424 size = 1 << src->size_bits;
1425 for (i = 0; i < size; i++) {
1426 hhd = &src->buckets[i];
1427 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1428 remove_hash_entry(src, entry);
1429 __add_hash_entry(new_hash, entry);
1430 }
1431 }
1432 return new_hash;
1433 }
1434
1435 /* Move the @src entries to a newly allocated hash */
1436 static struct ftrace_hash *
__ftrace_hash_move(struct ftrace_hash * src)1437 __ftrace_hash_move(struct ftrace_hash *src)
1438 {
1439 int size = src->count;
1440
1441 /*
1442 * If the new source is empty, just return the empty_hash.
1443 */
1444 if (ftrace_hash_empty(src))
1445 return EMPTY_HASH;
1446
1447 return __move_hash(src, size);
1448 }
1449
1450 /**
1451 * ftrace_hash_move - move a new hash to a filter and do updates
1452 * @ops: The ops with the hash that @dst points to
1453 * @enable: True if for the filter hash, false for the notrace hash
1454 * @dst: Points to the @ops hash that should be updated
1455 * @src: The hash to update @dst with
1456 *
1457 * This is called when an ftrace_ops hash is being updated and the
1458 * the kernel needs to reflect this. Note, this only updates the kernel
1459 * function callbacks if the @ops is enabled (not to be confused with
1460 * @enable above). If the @ops is enabled, its hash determines what
1461 * callbacks get called. This function gets called when the @ops hash
1462 * is updated and it requires new callbacks.
1463 *
1464 * On success the elements of @src is moved to @dst, and @dst is updated
1465 * properly, as well as the functions determined by the @ops hashes
1466 * are now calling the @ops callback function.
1467 *
1468 * Regardless of return type, @src should be freed with free_ftrace_hash().
1469 */
1470 static int
ftrace_hash_move(struct ftrace_ops * ops,int enable,struct ftrace_hash ** dst,struct ftrace_hash * src)1471 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1472 struct ftrace_hash **dst, struct ftrace_hash *src)
1473 {
1474 struct ftrace_hash *new_hash;
1475 int ret;
1476
1477 /* Reject setting notrace hash on IPMODIFY ftrace_ops */
1478 if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1479 return -EINVAL;
1480
1481 new_hash = __ftrace_hash_move(src);
1482 if (!new_hash)
1483 return -ENOMEM;
1484
1485 /* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1486 if (enable) {
1487 /* IPMODIFY should be updated only when filter_hash updating */
1488 ret = ftrace_hash_ipmodify_update(ops, new_hash);
1489 if (ret < 0) {
1490 free_ftrace_hash(new_hash);
1491 return ret;
1492 }
1493 }
1494
1495 /*
1496 * Remove the current set, update the hash and add
1497 * them back.
1498 */
1499 ftrace_hash_rec_disable_modify(ops);
1500
1501 rcu_assign_pointer(*dst, new_hash);
1502
1503 ftrace_hash_rec_enable_modify(ops);
1504
1505 return 0;
1506 }
1507
hash_contains_ip(unsigned long ip,struct ftrace_ops_hash * hash)1508 static bool hash_contains_ip(unsigned long ip,
1509 struct ftrace_ops_hash *hash)
1510 {
1511 /*
1512 * The function record is a match if it exists in the filter
1513 * hash and not in the notrace hash. Note, an empty hash is
1514 * considered a match for the filter hash, but an empty
1515 * notrace hash is considered not in the notrace hash.
1516 */
1517 return (ftrace_hash_empty(hash->filter_hash) ||
1518 __ftrace_lookup_ip(hash->filter_hash, ip)) &&
1519 (ftrace_hash_empty(hash->notrace_hash) ||
1520 !__ftrace_lookup_ip(hash->notrace_hash, ip));
1521 }
1522
1523 /*
1524 * Test the hashes for this ops to see if we want to call
1525 * the ops->func or not.
1526 *
1527 * It's a match if the ip is in the ops->filter_hash or
1528 * the filter_hash does not exist or is empty,
1529 * AND
1530 * the ip is not in the ops->notrace_hash.
1531 *
1532 * This needs to be called with preemption disabled as
1533 * the hashes are freed with call_rcu().
1534 */
1535 int
ftrace_ops_test(struct ftrace_ops * ops,unsigned long ip,void * regs)1536 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1537 {
1538 struct ftrace_ops_hash hash;
1539 int ret;
1540
1541 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1542 /*
1543 * There's a small race when adding ops that the ftrace handler
1544 * that wants regs, may be called without them. We can not
1545 * allow that handler to be called if regs is NULL.
1546 */
1547 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1548 return 0;
1549 #endif
1550
1551 rcu_assign_pointer(hash.filter_hash, ops->func_hash->filter_hash);
1552 rcu_assign_pointer(hash.notrace_hash, ops->func_hash->notrace_hash);
1553
1554 if (hash_contains_ip(ip, &hash))
1555 ret = 1;
1556 else
1557 ret = 0;
1558
1559 return ret;
1560 }
1561
1562 /*
1563 * This is a double for. Do not use 'break' to break out of the loop,
1564 * you must use a goto.
1565 */
1566 #define do_for_each_ftrace_rec(pg, rec) \
1567 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1568 int _____i; \
1569 for (_____i = 0; _____i < pg->index; _____i++) { \
1570 rec = &pg->records[_____i];
1571
1572 #define while_for_each_ftrace_rec() \
1573 } \
1574 }
1575
1576
ftrace_cmp_recs(const void * a,const void * b)1577 static int ftrace_cmp_recs(const void *a, const void *b)
1578 {
1579 const struct dyn_ftrace *key = a;
1580 const struct dyn_ftrace *rec = b;
1581
1582 if (key->flags < rec->ip)
1583 return -1;
1584 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1585 return 1;
1586 return 0;
1587 }
1588
lookup_rec(unsigned long start,unsigned long end)1589 static struct dyn_ftrace *lookup_rec(unsigned long start, unsigned long end)
1590 {
1591 struct ftrace_page *pg;
1592 struct dyn_ftrace *rec = NULL;
1593 struct dyn_ftrace key;
1594
1595 key.ip = start;
1596 key.flags = end; /* overload flags, as it is unsigned long */
1597
1598 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1599 if (pg->index == 0 ||
1600 end < pg->records[0].ip ||
1601 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1602 continue;
1603 rec = bsearch(&key, pg->records, pg->index,
1604 sizeof(struct dyn_ftrace),
1605 ftrace_cmp_recs);
1606 if (rec)
1607 break;
1608 }
1609 return rec;
1610 }
1611
1612 /**
1613 * ftrace_location_range - return the first address of a traced location
1614 * if it touches the given ip range
1615 * @start: start of range to search.
1616 * @end: end of range to search (inclusive). @end points to the last byte
1617 * to check.
1618 *
1619 * Returns: rec->ip if the related ftrace location is a least partly within
1620 * the given address range. That is, the first address of the instruction
1621 * that is either a NOP or call to the function tracer. It checks the ftrace
1622 * internal tables to determine if the address belongs or not.
1623 */
ftrace_location_range(unsigned long start,unsigned long end)1624 unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1625 {
1626 struct dyn_ftrace *rec;
1627 unsigned long ip = 0;
1628
1629 rcu_read_lock();
1630 rec = lookup_rec(start, end);
1631 if (rec)
1632 ip = rec->ip;
1633 rcu_read_unlock();
1634
1635 return ip;
1636 }
1637
1638 /**
1639 * ftrace_location - return the ftrace location
1640 * @ip: the instruction pointer to check
1641 *
1642 * Returns:
1643 * * If @ip matches the ftrace location, return @ip.
1644 * * If @ip matches sym+0, return sym's ftrace location.
1645 * * Otherwise, return 0.
1646 */
ftrace_location(unsigned long ip)1647 unsigned long ftrace_location(unsigned long ip)
1648 {
1649 unsigned long loc;
1650 unsigned long offset;
1651 unsigned long size;
1652
1653 loc = ftrace_location_range(ip, ip);
1654 if (!loc) {
1655 if (!kallsyms_lookup_size_offset(ip, &size, &offset))
1656 return 0;
1657
1658 /* map sym+0 to __fentry__ */
1659 if (!offset)
1660 loc = ftrace_location_range(ip, ip + size - 1);
1661 }
1662 return loc;
1663 }
1664
1665 /**
1666 * ftrace_text_reserved - return true if range contains an ftrace location
1667 * @start: start of range to search
1668 * @end: end of range to search (inclusive). @end points to the last byte to check.
1669 *
1670 * Returns: 1 if @start and @end contains a ftrace location.
1671 * That is, the instruction that is either a NOP or call to
1672 * the function tracer. It checks the ftrace internal tables to
1673 * determine if the address belongs or not.
1674 */
ftrace_text_reserved(const void * start,const void * end)1675 int ftrace_text_reserved(const void *start, const void *end)
1676 {
1677 unsigned long ret;
1678
1679 ret = ftrace_location_range((unsigned long)start,
1680 (unsigned long)end);
1681
1682 return (int)!!ret;
1683 }
1684
1685 /* Test if ops registered to this rec needs regs */
test_rec_ops_needs_regs(struct dyn_ftrace * rec)1686 static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1687 {
1688 struct ftrace_ops *ops;
1689 bool keep_regs = false;
1690
1691 for (ops = ftrace_ops_list;
1692 ops != &ftrace_list_end; ops = ops->next) {
1693 /* pass rec in as regs to have non-NULL val */
1694 if (ftrace_ops_test(ops, rec->ip, rec)) {
1695 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1696 keep_regs = true;
1697 break;
1698 }
1699 }
1700 }
1701
1702 return keep_regs;
1703 }
1704
1705 static struct ftrace_ops *
1706 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
1707 static struct ftrace_ops *
1708 ftrace_find_tramp_ops_any_other(struct dyn_ftrace *rec, struct ftrace_ops *op_exclude);
1709 static struct ftrace_ops *
1710 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
1711
skip_record(struct dyn_ftrace * rec)1712 static bool skip_record(struct dyn_ftrace *rec)
1713 {
1714 /*
1715 * At boot up, weak functions are set to disable. Function tracing
1716 * can be enabled before they are, and they still need to be disabled now.
1717 * If the record is disabled, still continue if it is marked as already
1718 * enabled (this is needed to keep the accounting working).
1719 */
1720 return rec->flags & FTRACE_FL_DISABLED &&
1721 !(rec->flags & FTRACE_FL_ENABLED);
1722 }
1723
1724 /*
1725 * This is the main engine to the ftrace updates to the dyn_ftrace records.
1726 *
1727 * It will iterate through all the available ftrace functions
1728 * (the ones that ftrace can have callbacks to) and set the flags
1729 * in the associated dyn_ftrace records.
1730 *
1731 * @inc: If true, the functions associated to @ops are added to
1732 * the dyn_ftrace records, otherwise they are removed.
1733 */
__ftrace_hash_rec_update(struct ftrace_ops * ops,bool inc)1734 static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1735 bool inc)
1736 {
1737 struct ftrace_hash *hash;
1738 struct ftrace_hash *notrace_hash;
1739 struct ftrace_page *pg;
1740 struct dyn_ftrace *rec;
1741 bool update = false;
1742 int count = 0;
1743 int all = false;
1744
1745 /* Only update if the ops has been registered */
1746 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1747 return false;
1748
1749 /*
1750 * If the count is zero, we update all records.
1751 * Otherwise we just update the items in the hash.
1752 */
1753 hash = ops->func_hash->filter_hash;
1754 notrace_hash = ops->func_hash->notrace_hash;
1755 if (ftrace_hash_empty(hash))
1756 all = true;
1757
1758 do_for_each_ftrace_rec(pg, rec) {
1759 int in_notrace_hash = 0;
1760 int in_hash = 0;
1761 int match = 0;
1762
1763 if (skip_record(rec))
1764 continue;
1765
1766 if (all) {
1767 /*
1768 * Only the filter_hash affects all records.
1769 * Update if the record is not in the notrace hash.
1770 */
1771 if (!notrace_hash || !ftrace_lookup_ip(notrace_hash, rec->ip))
1772 match = 1;
1773 } else {
1774 in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1775 in_notrace_hash = !!ftrace_lookup_ip(notrace_hash, rec->ip);
1776
1777 /*
1778 * We want to match all functions that are in the hash but
1779 * not in the other hash.
1780 */
1781 if (in_hash && !in_notrace_hash)
1782 match = 1;
1783 }
1784 if (!match)
1785 continue;
1786
1787 if (inc) {
1788 rec->flags++;
1789 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1790 return false;
1791
1792 if (ops->flags & FTRACE_OPS_FL_DIRECT)
1793 rec->flags |= FTRACE_FL_DIRECT;
1794
1795 /*
1796 * If there's only a single callback registered to a
1797 * function, and the ops has a trampoline registered
1798 * for it, then we can call it directly.
1799 */
1800 if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1801 rec->flags |= FTRACE_FL_TRAMP;
1802 else
1803 /*
1804 * If we are adding another function callback
1805 * to this function, and the previous had a
1806 * custom trampoline in use, then we need to go
1807 * back to the default trampoline.
1808 */
1809 rec->flags &= ~FTRACE_FL_TRAMP;
1810
1811 /*
1812 * If any ops wants regs saved for this function
1813 * then all ops will get saved regs.
1814 */
1815 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1816 rec->flags |= FTRACE_FL_REGS;
1817 } else {
1818 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1819 return false;
1820 rec->flags--;
1821
1822 /*
1823 * Only the internal direct_ops should have the
1824 * DIRECT flag set. Thus, if it is removing a
1825 * function, then that function should no longer
1826 * be direct.
1827 */
1828 if (ops->flags & FTRACE_OPS_FL_DIRECT)
1829 rec->flags &= ~FTRACE_FL_DIRECT;
1830
1831 /*
1832 * If the rec had REGS enabled and the ops that is
1833 * being removed had REGS set, then see if there is
1834 * still any ops for this record that wants regs.
1835 * If not, we can stop recording them.
1836 */
1837 if (ftrace_rec_count(rec) > 0 &&
1838 rec->flags & FTRACE_FL_REGS &&
1839 ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1840 if (!test_rec_ops_needs_regs(rec))
1841 rec->flags &= ~FTRACE_FL_REGS;
1842 }
1843
1844 /*
1845 * The TRAMP needs to be set only if rec count
1846 * is decremented to one, and the ops that is
1847 * left has a trampoline. As TRAMP can only be
1848 * enabled if there is only a single ops attached
1849 * to it.
1850 */
1851 if (ftrace_rec_count(rec) == 1 &&
1852 ftrace_find_tramp_ops_any_other(rec, ops))
1853 rec->flags |= FTRACE_FL_TRAMP;
1854 else
1855 rec->flags &= ~FTRACE_FL_TRAMP;
1856
1857 /*
1858 * flags will be cleared in ftrace_check_record()
1859 * if rec count is zero.
1860 */
1861 }
1862
1863 /*
1864 * If the rec has a single associated ops, and ops->func can be
1865 * called directly, allow the call site to call via the ops.
1866 */
1867 if (IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_CALL_OPS) &&
1868 ftrace_rec_count(rec) == 1 &&
1869 ftrace_ops_get_func(ops) == ops->func)
1870 rec->flags |= FTRACE_FL_CALL_OPS;
1871 else
1872 rec->flags &= ~FTRACE_FL_CALL_OPS;
1873
1874 count++;
1875
1876 /* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1877 update |= ftrace_test_record(rec, true) != FTRACE_UPDATE_IGNORE;
1878
1879 /* Shortcut, if we handled all records, we are done. */
1880 if (!all && count == hash->count)
1881 return update;
1882 } while_for_each_ftrace_rec();
1883
1884 return update;
1885 }
1886
1887 /*
1888 * This is called when an ops is removed from tracing. It will decrement
1889 * the counters of the dyn_ftrace records for all the functions that
1890 * the @ops attached to.
1891 */
ftrace_hash_rec_disable(struct ftrace_ops * ops)1892 static bool ftrace_hash_rec_disable(struct ftrace_ops *ops)
1893 {
1894 return __ftrace_hash_rec_update(ops, false);
1895 }
1896
1897 /*
1898 * This is called when an ops is added to tracing. It will increment
1899 * the counters of the dyn_ftrace records for all the functions that
1900 * the @ops attached to.
1901 */
ftrace_hash_rec_enable(struct ftrace_ops * ops)1902 static bool ftrace_hash_rec_enable(struct ftrace_ops *ops)
1903 {
1904 return __ftrace_hash_rec_update(ops, true);
1905 }
1906
1907 /*
1908 * This function will update what functions @ops traces when its filter
1909 * changes.
1910 *
1911 * The @inc states if the @ops callbacks are going to be added or removed.
1912 * When one of the @ops hashes are updated to a "new_hash" the dyn_ftrace
1913 * records are update via:
1914 *
1915 * ftrace_hash_rec_disable_modify(ops);
1916 * ops->hash = new_hash
1917 * ftrace_hash_rec_enable_modify(ops);
1918 *
1919 * Where the @ops is removed from all the records it is tracing using
1920 * its old hash. The @ops hash is updated to the new hash, and then
1921 * the @ops is added back to the records so that it is tracing all
1922 * the new functions.
1923 */
ftrace_hash_rec_update_modify(struct ftrace_ops * ops,bool inc)1924 static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops, bool inc)
1925 {
1926 struct ftrace_ops *op;
1927
1928 __ftrace_hash_rec_update(ops, inc);
1929
1930 if (ops->func_hash != &global_ops.local_hash)
1931 return;
1932
1933 /*
1934 * If the ops shares the global_ops hash, then we need to update
1935 * all ops that are enabled and use this hash.
1936 */
1937 do_for_each_ftrace_op(op, ftrace_ops_list) {
1938 /* Already done */
1939 if (op == ops)
1940 continue;
1941 if (op->func_hash == &global_ops.local_hash)
1942 __ftrace_hash_rec_update(op, inc);
1943 } while_for_each_ftrace_op(op);
1944 }
1945
ftrace_hash_rec_disable_modify(struct ftrace_ops * ops)1946 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops)
1947 {
1948 ftrace_hash_rec_update_modify(ops, false);
1949 }
1950
ftrace_hash_rec_enable_modify(struct ftrace_ops * ops)1951 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops)
1952 {
1953 ftrace_hash_rec_update_modify(ops, true);
1954 }
1955
1956 /*
1957 * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1958 * or no-needed to update, -EBUSY if it detects a conflict of the flag
1959 * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1960 * Note that old_hash and new_hash has below meanings
1961 * - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1962 * - If the hash is EMPTY_HASH, it hits nothing
1963 * - Anything else hits the recs which match the hash entries.
1964 *
1965 * DIRECT ops does not have IPMODIFY flag, but we still need to check it
1966 * against functions with FTRACE_FL_IPMODIFY. If there is any overlap, call
1967 * ops_func(SHARE_IPMODIFY_SELF) to make sure current ops can share with
1968 * IPMODIFY. If ops_func(SHARE_IPMODIFY_SELF) returns non-zero, propagate
1969 * the return value to the caller and eventually to the owner of the DIRECT
1970 * ops.
1971 */
__ftrace_hash_update_ipmodify(struct ftrace_ops * ops,struct ftrace_hash * old_hash,struct ftrace_hash * new_hash)1972 static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1973 struct ftrace_hash *old_hash,
1974 struct ftrace_hash *new_hash)
1975 {
1976 struct ftrace_page *pg;
1977 struct dyn_ftrace *rec, *end = NULL;
1978 int in_old, in_new;
1979 bool is_ipmodify, is_direct;
1980
1981 /* Only update if the ops has been registered */
1982 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1983 return 0;
1984
1985 is_ipmodify = ops->flags & FTRACE_OPS_FL_IPMODIFY;
1986 is_direct = ops->flags & FTRACE_OPS_FL_DIRECT;
1987
1988 /* neither IPMODIFY nor DIRECT, skip */
1989 if (!is_ipmodify && !is_direct)
1990 return 0;
1991
1992 if (WARN_ON_ONCE(is_ipmodify && is_direct))
1993 return 0;
1994
1995 /*
1996 * Since the IPMODIFY and DIRECT are very address sensitive
1997 * actions, we do not allow ftrace_ops to set all functions to new
1998 * hash.
1999 */
2000 if (!new_hash || !old_hash)
2001 return -EINVAL;
2002
2003 /* Update rec->flags */
2004 do_for_each_ftrace_rec(pg, rec) {
2005
2006 if (rec->flags & FTRACE_FL_DISABLED)
2007 continue;
2008
2009 /* We need to update only differences of filter_hash */
2010 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
2011 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
2012 if (in_old == in_new)
2013 continue;
2014
2015 if (in_new) {
2016 if (rec->flags & FTRACE_FL_IPMODIFY) {
2017 int ret;
2018
2019 /* Cannot have two ipmodify on same rec */
2020 if (is_ipmodify)
2021 goto rollback;
2022
2023 FTRACE_WARN_ON(rec->flags & FTRACE_FL_DIRECT);
2024
2025 /*
2026 * Another ops with IPMODIFY is already
2027 * attached. We are now attaching a direct
2028 * ops. Run SHARE_IPMODIFY_SELF, to check
2029 * whether sharing is supported.
2030 */
2031 if (!ops->ops_func)
2032 return -EBUSY;
2033 ret = ops->ops_func(ops, FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_SELF);
2034 if (ret)
2035 return ret;
2036 } else if (is_ipmodify) {
2037 rec->flags |= FTRACE_FL_IPMODIFY;
2038 }
2039 } else if (is_ipmodify) {
2040 rec->flags &= ~FTRACE_FL_IPMODIFY;
2041 }
2042 } while_for_each_ftrace_rec();
2043
2044 return 0;
2045
2046 rollback:
2047 end = rec;
2048
2049 /* Roll back what we did above */
2050 do_for_each_ftrace_rec(pg, rec) {
2051
2052 if (rec->flags & FTRACE_FL_DISABLED)
2053 continue;
2054
2055 if (rec == end)
2056 return -EBUSY;
2057
2058 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
2059 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
2060 if (in_old == in_new)
2061 continue;
2062
2063 if (in_new)
2064 rec->flags &= ~FTRACE_FL_IPMODIFY;
2065 else
2066 rec->flags |= FTRACE_FL_IPMODIFY;
2067 } while_for_each_ftrace_rec();
2068
2069 return -EBUSY;
2070 }
2071
ftrace_hash_ipmodify_enable(struct ftrace_ops * ops)2072 static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
2073 {
2074 struct ftrace_hash *hash = ops->func_hash->filter_hash;
2075
2076 if (ftrace_hash_empty(hash))
2077 hash = NULL;
2078
2079 return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
2080 }
2081
2082 /* Disabling always succeeds */
ftrace_hash_ipmodify_disable(struct ftrace_ops * ops)2083 static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
2084 {
2085 struct ftrace_hash *hash = ops->func_hash->filter_hash;
2086
2087 if (ftrace_hash_empty(hash))
2088 hash = NULL;
2089
2090 __ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
2091 }
2092
ftrace_hash_ipmodify_update(struct ftrace_ops * ops,struct ftrace_hash * new_hash)2093 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
2094 struct ftrace_hash *new_hash)
2095 {
2096 struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
2097
2098 if (ftrace_hash_empty(old_hash))
2099 old_hash = NULL;
2100
2101 if (ftrace_hash_empty(new_hash))
2102 new_hash = NULL;
2103
2104 return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
2105 }
2106
print_ip_ins(const char * fmt,const unsigned char * p)2107 static void print_ip_ins(const char *fmt, const unsigned char *p)
2108 {
2109 char ins[MCOUNT_INSN_SIZE];
2110
2111 if (copy_from_kernel_nofault(ins, p, MCOUNT_INSN_SIZE)) {
2112 printk(KERN_CONT "%s[FAULT] %px\n", fmt, p);
2113 return;
2114 }
2115
2116 printk(KERN_CONT "%s", fmt);
2117 pr_cont("%*phC", MCOUNT_INSN_SIZE, ins);
2118 }
2119
2120 enum ftrace_bug_type ftrace_bug_type;
2121 const void *ftrace_expected;
2122
print_bug_type(void)2123 static void print_bug_type(void)
2124 {
2125 switch (ftrace_bug_type) {
2126 case FTRACE_BUG_UNKNOWN:
2127 break;
2128 case FTRACE_BUG_INIT:
2129 pr_info("Initializing ftrace call sites\n");
2130 break;
2131 case FTRACE_BUG_NOP:
2132 pr_info("Setting ftrace call site to NOP\n");
2133 break;
2134 case FTRACE_BUG_CALL:
2135 pr_info("Setting ftrace call site to call ftrace function\n");
2136 break;
2137 case FTRACE_BUG_UPDATE:
2138 pr_info("Updating ftrace call site to call a different ftrace function\n");
2139 break;
2140 }
2141 }
2142
2143 /**
2144 * ftrace_bug - report and shutdown function tracer
2145 * @failed: The failed type (EFAULT, EINVAL, EPERM)
2146 * @rec: The record that failed
2147 *
2148 * The arch code that enables or disables the function tracing
2149 * can call ftrace_bug() when it has detected a problem in
2150 * modifying the code. @failed should be one of either:
2151 * EFAULT - if the problem happens on reading the @ip address
2152 * EINVAL - if what is read at @ip is not what was expected
2153 * EPERM - if the problem happens on writing to the @ip address
2154 */
ftrace_bug(int failed,struct dyn_ftrace * rec)2155 void ftrace_bug(int failed, struct dyn_ftrace *rec)
2156 {
2157 unsigned long ip = rec ? rec->ip : 0;
2158
2159 pr_info("------------[ ftrace bug ]------------\n");
2160
2161 switch (failed) {
2162 case -EFAULT:
2163 pr_info("ftrace faulted on modifying ");
2164 print_ip_sym(KERN_INFO, ip);
2165 break;
2166 case -EINVAL:
2167 pr_info("ftrace failed to modify ");
2168 print_ip_sym(KERN_INFO, ip);
2169 print_ip_ins(" actual: ", (unsigned char *)ip);
2170 pr_cont("\n");
2171 if (ftrace_expected) {
2172 print_ip_ins(" expected: ", ftrace_expected);
2173 pr_cont("\n");
2174 }
2175 break;
2176 case -EPERM:
2177 pr_info("ftrace faulted on writing ");
2178 print_ip_sym(KERN_INFO, ip);
2179 break;
2180 default:
2181 pr_info("ftrace faulted on unknown error ");
2182 print_ip_sym(KERN_INFO, ip);
2183 }
2184 print_bug_type();
2185 if (rec) {
2186 struct ftrace_ops *ops = NULL;
2187
2188 pr_info("ftrace record flags: %lx\n", rec->flags);
2189 pr_cont(" (%ld)%s%s", ftrace_rec_count(rec),
2190 rec->flags & FTRACE_FL_REGS ? " R" : " ",
2191 rec->flags & FTRACE_FL_CALL_OPS ? " O" : " ");
2192 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2193 ops = ftrace_find_tramp_ops_any(rec);
2194 if (ops) {
2195 do {
2196 pr_cont("\ttramp: %pS (%pS)",
2197 (void *)ops->trampoline,
2198 (void *)ops->func);
2199 ops = ftrace_find_tramp_ops_next(rec, ops);
2200 } while (ops);
2201 } else
2202 pr_cont("\ttramp: ERROR!");
2203
2204 }
2205 ip = ftrace_get_addr_curr(rec);
2206 pr_cont("\n expected tramp: %lx\n", ip);
2207 }
2208
2209 FTRACE_WARN_ON_ONCE(1);
2210 }
2211
ftrace_check_record(struct dyn_ftrace * rec,bool enable,bool update)2212 static int ftrace_check_record(struct dyn_ftrace *rec, bool enable, bool update)
2213 {
2214 unsigned long flag = 0UL;
2215
2216 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2217
2218 if (skip_record(rec))
2219 return FTRACE_UPDATE_IGNORE;
2220
2221 /*
2222 * If we are updating calls:
2223 *
2224 * If the record has a ref count, then we need to enable it
2225 * because someone is using it.
2226 *
2227 * Otherwise we make sure its disabled.
2228 *
2229 * If we are disabling calls, then disable all records that
2230 * are enabled.
2231 */
2232 if (enable && ftrace_rec_count(rec))
2233 flag = FTRACE_FL_ENABLED;
2234
2235 /*
2236 * If enabling and the REGS flag does not match the REGS_EN, or
2237 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2238 * this record. Set flags to fail the compare against ENABLED.
2239 * Same for direct calls.
2240 */
2241 if (flag) {
2242 if (!(rec->flags & FTRACE_FL_REGS) !=
2243 !(rec->flags & FTRACE_FL_REGS_EN))
2244 flag |= FTRACE_FL_REGS;
2245
2246 if (!(rec->flags & FTRACE_FL_TRAMP) !=
2247 !(rec->flags & FTRACE_FL_TRAMP_EN))
2248 flag |= FTRACE_FL_TRAMP;
2249
2250 /*
2251 * Direct calls are special, as count matters.
2252 * We must test the record for direct, if the
2253 * DIRECT and DIRECT_EN do not match, but only
2254 * if the count is 1. That's because, if the
2255 * count is something other than one, we do not
2256 * want the direct enabled (it will be done via the
2257 * direct helper). But if DIRECT_EN is set, and
2258 * the count is not one, we need to clear it.
2259 *
2260 */
2261 if (ftrace_rec_count(rec) == 1) {
2262 if (!(rec->flags & FTRACE_FL_DIRECT) !=
2263 !(rec->flags & FTRACE_FL_DIRECT_EN))
2264 flag |= FTRACE_FL_DIRECT;
2265 } else if (rec->flags & FTRACE_FL_DIRECT_EN) {
2266 flag |= FTRACE_FL_DIRECT;
2267 }
2268
2269 /*
2270 * Ops calls are special, as count matters.
2271 * As with direct calls, they must only be enabled when count
2272 * is one, otherwise they'll be handled via the list ops.
2273 */
2274 if (ftrace_rec_count(rec) == 1) {
2275 if (!(rec->flags & FTRACE_FL_CALL_OPS) !=
2276 !(rec->flags & FTRACE_FL_CALL_OPS_EN))
2277 flag |= FTRACE_FL_CALL_OPS;
2278 } else if (rec->flags & FTRACE_FL_CALL_OPS_EN) {
2279 flag |= FTRACE_FL_CALL_OPS;
2280 }
2281 }
2282
2283 /* If the state of this record hasn't changed, then do nothing */
2284 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2285 return FTRACE_UPDATE_IGNORE;
2286
2287 if (flag) {
2288 /* Save off if rec is being enabled (for return value) */
2289 flag ^= rec->flags & FTRACE_FL_ENABLED;
2290
2291 if (update) {
2292 rec->flags |= FTRACE_FL_ENABLED | FTRACE_FL_TOUCHED;
2293 if (flag & FTRACE_FL_REGS) {
2294 if (rec->flags & FTRACE_FL_REGS)
2295 rec->flags |= FTRACE_FL_REGS_EN;
2296 else
2297 rec->flags &= ~FTRACE_FL_REGS_EN;
2298 }
2299 if (flag & FTRACE_FL_TRAMP) {
2300 if (rec->flags & FTRACE_FL_TRAMP)
2301 rec->flags |= FTRACE_FL_TRAMP_EN;
2302 else
2303 rec->flags &= ~FTRACE_FL_TRAMP_EN;
2304 }
2305
2306 /* Keep track of anything that modifies the function */
2307 if (rec->flags & (FTRACE_FL_DIRECT | FTRACE_FL_IPMODIFY))
2308 rec->flags |= FTRACE_FL_MODIFIED;
2309
2310 if (flag & FTRACE_FL_DIRECT) {
2311 /*
2312 * If there's only one user (direct_ops helper)
2313 * then we can call the direct function
2314 * directly (no ftrace trampoline).
2315 */
2316 if (ftrace_rec_count(rec) == 1) {
2317 if (rec->flags & FTRACE_FL_DIRECT)
2318 rec->flags |= FTRACE_FL_DIRECT_EN;
2319 else
2320 rec->flags &= ~FTRACE_FL_DIRECT_EN;
2321 } else {
2322 /*
2323 * Can only call directly if there's
2324 * only one callback to the function.
2325 */
2326 rec->flags &= ~FTRACE_FL_DIRECT_EN;
2327 }
2328 }
2329
2330 if (flag & FTRACE_FL_CALL_OPS) {
2331 if (ftrace_rec_count(rec) == 1) {
2332 if (rec->flags & FTRACE_FL_CALL_OPS)
2333 rec->flags |= FTRACE_FL_CALL_OPS_EN;
2334 else
2335 rec->flags &= ~FTRACE_FL_CALL_OPS_EN;
2336 } else {
2337 /*
2338 * Can only call directly if there's
2339 * only one set of associated ops.
2340 */
2341 rec->flags &= ~FTRACE_FL_CALL_OPS_EN;
2342 }
2343 }
2344 }
2345
2346 /*
2347 * If this record is being updated from a nop, then
2348 * return UPDATE_MAKE_CALL.
2349 * Otherwise,
2350 * return UPDATE_MODIFY_CALL to tell the caller to convert
2351 * from the save regs, to a non-save regs function or
2352 * vice versa, or from a trampoline call.
2353 */
2354 if (flag & FTRACE_FL_ENABLED) {
2355 ftrace_bug_type = FTRACE_BUG_CALL;
2356 return FTRACE_UPDATE_MAKE_CALL;
2357 }
2358
2359 ftrace_bug_type = FTRACE_BUG_UPDATE;
2360 return FTRACE_UPDATE_MODIFY_CALL;
2361 }
2362
2363 if (update) {
2364 /* If there's no more users, clear all flags */
2365 if (!ftrace_rec_count(rec))
2366 rec->flags &= FTRACE_NOCLEAR_FLAGS;
2367 else
2368 /*
2369 * Just disable the record, but keep the ops TRAMP
2370 * and REGS states. The _EN flags must be disabled though.
2371 */
2372 rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2373 FTRACE_FL_REGS_EN | FTRACE_FL_DIRECT_EN |
2374 FTRACE_FL_CALL_OPS_EN);
2375 }
2376
2377 ftrace_bug_type = FTRACE_BUG_NOP;
2378 return FTRACE_UPDATE_MAKE_NOP;
2379 }
2380
2381 /**
2382 * ftrace_update_record - set a record that now is tracing or not
2383 * @rec: the record to update
2384 * @enable: set to true if the record is tracing, false to force disable
2385 *
2386 * The records that represent all functions that can be traced need
2387 * to be updated when tracing has been enabled.
2388 */
ftrace_update_record(struct dyn_ftrace * rec,bool enable)2389 int ftrace_update_record(struct dyn_ftrace *rec, bool enable)
2390 {
2391 return ftrace_check_record(rec, enable, true);
2392 }
2393
2394 /**
2395 * ftrace_test_record - check if the record has been enabled or not
2396 * @rec: the record to test
2397 * @enable: set to true to check if enabled, false if it is disabled
2398 *
2399 * The arch code may need to test if a record is already set to
2400 * tracing to determine how to modify the function code that it
2401 * represents.
2402 */
ftrace_test_record(struct dyn_ftrace * rec,bool enable)2403 int ftrace_test_record(struct dyn_ftrace *rec, bool enable)
2404 {
2405 return ftrace_check_record(rec, enable, false);
2406 }
2407
2408 static struct ftrace_ops *
ftrace_find_tramp_ops_any(struct dyn_ftrace * rec)2409 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2410 {
2411 struct ftrace_ops *op;
2412 unsigned long ip = rec->ip;
2413
2414 do_for_each_ftrace_op(op, ftrace_ops_list) {
2415
2416 if (!op->trampoline)
2417 continue;
2418
2419 if (hash_contains_ip(ip, op->func_hash))
2420 return op;
2421 } while_for_each_ftrace_op(op);
2422
2423 return NULL;
2424 }
2425
2426 static struct ftrace_ops *
ftrace_find_tramp_ops_any_other(struct dyn_ftrace * rec,struct ftrace_ops * op_exclude)2427 ftrace_find_tramp_ops_any_other(struct dyn_ftrace *rec, struct ftrace_ops *op_exclude)
2428 {
2429 struct ftrace_ops *op;
2430 unsigned long ip = rec->ip;
2431
2432 do_for_each_ftrace_op(op, ftrace_ops_list) {
2433
2434 if (op == op_exclude || !op->trampoline)
2435 continue;
2436
2437 if (hash_contains_ip(ip, op->func_hash))
2438 return op;
2439 } while_for_each_ftrace_op(op);
2440
2441 return NULL;
2442 }
2443
2444 static struct ftrace_ops *
ftrace_find_tramp_ops_next(struct dyn_ftrace * rec,struct ftrace_ops * op)2445 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2446 struct ftrace_ops *op)
2447 {
2448 unsigned long ip = rec->ip;
2449
2450 while_for_each_ftrace_op(op) {
2451
2452 if (!op->trampoline)
2453 continue;
2454
2455 if (hash_contains_ip(ip, op->func_hash))
2456 return op;
2457 }
2458
2459 return NULL;
2460 }
2461
2462 static struct ftrace_ops *
ftrace_find_tramp_ops_curr(struct dyn_ftrace * rec)2463 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2464 {
2465 struct ftrace_ops *op;
2466 unsigned long ip = rec->ip;
2467
2468 /*
2469 * Need to check removed ops first.
2470 * If they are being removed, and this rec has a tramp,
2471 * and this rec is in the ops list, then it would be the
2472 * one with the tramp.
2473 */
2474 if (removed_ops) {
2475 if (hash_contains_ip(ip, &removed_ops->old_hash))
2476 return removed_ops;
2477 }
2478
2479 /*
2480 * Need to find the current trampoline for a rec.
2481 * Now, a trampoline is only attached to a rec if there
2482 * was a single 'ops' attached to it. But this can be called
2483 * when we are adding another op to the rec or removing the
2484 * current one. Thus, if the op is being added, we can
2485 * ignore it because it hasn't attached itself to the rec
2486 * yet.
2487 *
2488 * If an ops is being modified (hooking to different functions)
2489 * then we don't care about the new functions that are being
2490 * added, just the old ones (that are probably being removed).
2491 *
2492 * If we are adding an ops to a function that already is using
2493 * a trampoline, it needs to be removed (trampolines are only
2494 * for single ops connected), then an ops that is not being
2495 * modified also needs to be checked.
2496 */
2497 do_for_each_ftrace_op(op, ftrace_ops_list) {
2498
2499 if (!op->trampoline)
2500 continue;
2501
2502 /*
2503 * If the ops is being added, it hasn't gotten to
2504 * the point to be removed from this tree yet.
2505 */
2506 if (op->flags & FTRACE_OPS_FL_ADDING)
2507 continue;
2508
2509
2510 /*
2511 * If the ops is being modified and is in the old
2512 * hash, then it is probably being removed from this
2513 * function.
2514 */
2515 if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2516 hash_contains_ip(ip, &op->old_hash))
2517 return op;
2518 /*
2519 * If the ops is not being added or modified, and it's
2520 * in its normal filter hash, then this must be the one
2521 * we want!
2522 */
2523 if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2524 hash_contains_ip(ip, op->func_hash))
2525 return op;
2526
2527 } while_for_each_ftrace_op(op);
2528
2529 return NULL;
2530 }
2531
2532 static struct ftrace_ops *
ftrace_find_tramp_ops_new(struct dyn_ftrace * rec)2533 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2534 {
2535 struct ftrace_ops *op;
2536 unsigned long ip = rec->ip;
2537
2538 do_for_each_ftrace_op(op, ftrace_ops_list) {
2539 /* pass rec in as regs to have non-NULL val */
2540 if (hash_contains_ip(ip, op->func_hash))
2541 return op;
2542 } while_for_each_ftrace_op(op);
2543
2544 return NULL;
2545 }
2546
2547 struct ftrace_ops *
ftrace_find_unique_ops(struct dyn_ftrace * rec)2548 ftrace_find_unique_ops(struct dyn_ftrace *rec)
2549 {
2550 struct ftrace_ops *op, *found = NULL;
2551 unsigned long ip = rec->ip;
2552
2553 do_for_each_ftrace_op(op, ftrace_ops_list) {
2554
2555 if (hash_contains_ip(ip, op->func_hash)) {
2556 if (found)
2557 return NULL;
2558 found = op;
2559 }
2560
2561 } while_for_each_ftrace_op(op);
2562
2563 return found;
2564 }
2565
2566 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
2567 /* Protected by rcu_tasks for reading, and direct_mutex for writing */
2568 static struct ftrace_hash __rcu *direct_functions = EMPTY_HASH;
2569 static DEFINE_MUTEX(direct_mutex);
2570
2571 /*
2572 * Search the direct_functions hash to see if the given instruction pointer
2573 * has a direct caller attached to it.
2574 */
ftrace_find_rec_direct(unsigned long ip)2575 unsigned long ftrace_find_rec_direct(unsigned long ip)
2576 {
2577 struct ftrace_func_entry *entry;
2578
2579 entry = __ftrace_lookup_ip(direct_functions, ip);
2580 if (!entry)
2581 return 0;
2582
2583 return entry->direct;
2584 }
2585
call_direct_funcs(unsigned long ip,unsigned long pip,struct ftrace_ops * ops,struct ftrace_regs * fregs)2586 static void call_direct_funcs(unsigned long ip, unsigned long pip,
2587 struct ftrace_ops *ops, struct ftrace_regs *fregs)
2588 {
2589 unsigned long addr = READ_ONCE(ops->direct_call);
2590
2591 if (!addr)
2592 return;
2593
2594 arch_ftrace_set_direct_caller(fregs, addr);
2595 }
2596 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
2597
2598 /**
2599 * ftrace_get_addr_new - Get the call address to set to
2600 * @rec: The ftrace record descriptor
2601 *
2602 * If the record has the FTRACE_FL_REGS set, that means that it
2603 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2604 * is not set, then it wants to convert to the normal callback.
2605 *
2606 * Returns: the address of the trampoline to set to
2607 */
ftrace_get_addr_new(struct dyn_ftrace * rec)2608 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2609 {
2610 struct ftrace_ops *ops;
2611 unsigned long addr;
2612
2613 if ((rec->flags & FTRACE_FL_DIRECT) &&
2614 (ftrace_rec_count(rec) == 1)) {
2615 addr = ftrace_find_rec_direct(rec->ip);
2616 if (addr)
2617 return addr;
2618 WARN_ON_ONCE(1);
2619 }
2620
2621 /* Trampolines take precedence over regs */
2622 if (rec->flags & FTRACE_FL_TRAMP) {
2623 ops = ftrace_find_tramp_ops_new(rec);
2624 if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2625 pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2626 (void *)rec->ip, (void *)rec->ip, rec->flags);
2627 /* Ftrace is shutting down, return anything */
2628 return (unsigned long)FTRACE_ADDR;
2629 }
2630 return ops->trampoline;
2631 }
2632
2633 if (rec->flags & FTRACE_FL_REGS)
2634 return (unsigned long)FTRACE_REGS_ADDR;
2635 else
2636 return (unsigned long)FTRACE_ADDR;
2637 }
2638
2639 /**
2640 * ftrace_get_addr_curr - Get the call address that is already there
2641 * @rec: The ftrace record descriptor
2642 *
2643 * The FTRACE_FL_REGS_EN is set when the record already points to
2644 * a function that saves all the regs. Basically the '_EN' version
2645 * represents the current state of the function.
2646 *
2647 * Returns: the address of the trampoline that is currently being called
2648 */
ftrace_get_addr_curr(struct dyn_ftrace * rec)2649 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2650 {
2651 struct ftrace_ops *ops;
2652 unsigned long addr;
2653
2654 /* Direct calls take precedence over trampolines */
2655 if (rec->flags & FTRACE_FL_DIRECT_EN) {
2656 addr = ftrace_find_rec_direct(rec->ip);
2657 if (addr)
2658 return addr;
2659 WARN_ON_ONCE(1);
2660 }
2661
2662 /* Trampolines take precedence over regs */
2663 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2664 ops = ftrace_find_tramp_ops_curr(rec);
2665 if (FTRACE_WARN_ON(!ops)) {
2666 pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2667 (void *)rec->ip, (void *)rec->ip);
2668 /* Ftrace is shutting down, return anything */
2669 return (unsigned long)FTRACE_ADDR;
2670 }
2671 return ops->trampoline;
2672 }
2673
2674 if (rec->flags & FTRACE_FL_REGS_EN)
2675 return (unsigned long)FTRACE_REGS_ADDR;
2676 else
2677 return (unsigned long)FTRACE_ADDR;
2678 }
2679
2680 static int
__ftrace_replace_code(struct dyn_ftrace * rec,bool enable)2681 __ftrace_replace_code(struct dyn_ftrace *rec, bool enable)
2682 {
2683 unsigned long ftrace_old_addr;
2684 unsigned long ftrace_addr;
2685 int ret;
2686
2687 ftrace_addr = ftrace_get_addr_new(rec);
2688
2689 /* This needs to be done before we call ftrace_update_record */
2690 ftrace_old_addr = ftrace_get_addr_curr(rec);
2691
2692 ret = ftrace_update_record(rec, enable);
2693
2694 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2695
2696 switch (ret) {
2697 case FTRACE_UPDATE_IGNORE:
2698 return 0;
2699
2700 case FTRACE_UPDATE_MAKE_CALL:
2701 ftrace_bug_type = FTRACE_BUG_CALL;
2702 return ftrace_make_call(rec, ftrace_addr);
2703
2704 case FTRACE_UPDATE_MAKE_NOP:
2705 ftrace_bug_type = FTRACE_BUG_NOP;
2706 return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2707
2708 case FTRACE_UPDATE_MODIFY_CALL:
2709 ftrace_bug_type = FTRACE_BUG_UPDATE;
2710 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2711 }
2712
2713 return -1; /* unknown ftrace bug */
2714 }
2715
ftrace_replace_code(int mod_flags)2716 void __weak ftrace_replace_code(int mod_flags)
2717 {
2718 struct dyn_ftrace *rec;
2719 struct ftrace_page *pg;
2720 bool enable = mod_flags & FTRACE_MODIFY_ENABLE_FL;
2721 int schedulable = mod_flags & FTRACE_MODIFY_MAY_SLEEP_FL;
2722 int failed;
2723
2724 if (unlikely(ftrace_disabled))
2725 return;
2726
2727 do_for_each_ftrace_rec(pg, rec) {
2728
2729 if (skip_record(rec))
2730 continue;
2731
2732 failed = __ftrace_replace_code(rec, enable);
2733 if (failed) {
2734 ftrace_bug(failed, rec);
2735 /* Stop processing */
2736 return;
2737 }
2738 if (schedulable)
2739 cond_resched();
2740 } while_for_each_ftrace_rec();
2741 }
2742
2743 struct ftrace_rec_iter {
2744 struct ftrace_page *pg;
2745 int index;
2746 };
2747
2748 /**
2749 * ftrace_rec_iter_start - start up iterating over traced functions
2750 *
2751 * Returns: an iterator handle that is used to iterate over all
2752 * the records that represent address locations where functions
2753 * are traced.
2754 *
2755 * May return NULL if no records are available.
2756 */
ftrace_rec_iter_start(void)2757 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2758 {
2759 /*
2760 * We only use a single iterator.
2761 * Protected by the ftrace_lock mutex.
2762 */
2763 static struct ftrace_rec_iter ftrace_rec_iter;
2764 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2765
2766 iter->pg = ftrace_pages_start;
2767 iter->index = 0;
2768
2769 /* Could have empty pages */
2770 while (iter->pg && !iter->pg->index)
2771 iter->pg = iter->pg->next;
2772
2773 if (!iter->pg)
2774 return NULL;
2775
2776 return iter;
2777 }
2778
2779 /**
2780 * ftrace_rec_iter_next - get the next record to process.
2781 * @iter: The handle to the iterator.
2782 *
2783 * Returns: the next iterator after the given iterator @iter.
2784 */
ftrace_rec_iter_next(struct ftrace_rec_iter * iter)2785 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2786 {
2787 iter->index++;
2788
2789 if (iter->index >= iter->pg->index) {
2790 iter->pg = iter->pg->next;
2791 iter->index = 0;
2792
2793 /* Could have empty pages */
2794 while (iter->pg && !iter->pg->index)
2795 iter->pg = iter->pg->next;
2796 }
2797
2798 if (!iter->pg)
2799 return NULL;
2800
2801 return iter;
2802 }
2803
2804 /**
2805 * ftrace_rec_iter_record - get the record at the iterator location
2806 * @iter: The current iterator location
2807 *
2808 * Returns: the record that the current @iter is at.
2809 */
ftrace_rec_iter_record(struct ftrace_rec_iter * iter)2810 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2811 {
2812 return &iter->pg->records[iter->index];
2813 }
2814
2815 static int
ftrace_nop_initialize(struct module * mod,struct dyn_ftrace * rec)2816 ftrace_nop_initialize(struct module *mod, struct dyn_ftrace *rec)
2817 {
2818 int ret;
2819
2820 if (unlikely(ftrace_disabled))
2821 return 0;
2822
2823 ret = ftrace_init_nop(mod, rec);
2824 if (ret) {
2825 ftrace_bug_type = FTRACE_BUG_INIT;
2826 ftrace_bug(ret, rec);
2827 return 0;
2828 }
2829 return 1;
2830 }
2831
2832 /*
2833 * archs can override this function if they must do something
2834 * before the modifying code is performed.
2835 */
ftrace_arch_code_modify_prepare(void)2836 void __weak ftrace_arch_code_modify_prepare(void)
2837 {
2838 }
2839
2840 /*
2841 * archs can override this function if they must do something
2842 * after the modifying code is performed.
2843 */
ftrace_arch_code_modify_post_process(void)2844 void __weak ftrace_arch_code_modify_post_process(void)
2845 {
2846 }
2847
update_ftrace_func(ftrace_func_t func)2848 static int update_ftrace_func(ftrace_func_t func)
2849 {
2850 static ftrace_func_t save_func;
2851
2852 /* Avoid updating if it hasn't changed */
2853 if (func == save_func)
2854 return 0;
2855
2856 save_func = func;
2857
2858 return ftrace_update_ftrace_func(func);
2859 }
2860
ftrace_modify_all_code(int command)2861 void ftrace_modify_all_code(int command)
2862 {
2863 int update = command & FTRACE_UPDATE_TRACE_FUNC;
2864 int mod_flags = 0;
2865 int err = 0;
2866
2867 if (command & FTRACE_MAY_SLEEP)
2868 mod_flags = FTRACE_MODIFY_MAY_SLEEP_FL;
2869
2870 /*
2871 * If the ftrace_caller calls a ftrace_ops func directly,
2872 * we need to make sure that it only traces functions it
2873 * expects to trace. When doing the switch of functions,
2874 * we need to update to the ftrace_ops_list_func first
2875 * before the transition between old and new calls are set,
2876 * as the ftrace_ops_list_func will check the ops hashes
2877 * to make sure the ops are having the right functions
2878 * traced.
2879 */
2880 if (update) {
2881 err = update_ftrace_func(ftrace_ops_list_func);
2882 if (FTRACE_WARN_ON(err))
2883 return;
2884 }
2885
2886 if (command & FTRACE_UPDATE_CALLS)
2887 ftrace_replace_code(mod_flags | FTRACE_MODIFY_ENABLE_FL);
2888 else if (command & FTRACE_DISABLE_CALLS)
2889 ftrace_replace_code(mod_flags);
2890
2891 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2892 function_trace_op = set_function_trace_op;
2893 smp_wmb();
2894 /* If irqs are disabled, we are in stop machine */
2895 if (!irqs_disabled())
2896 smp_call_function(ftrace_sync_ipi, NULL, 1);
2897 err = update_ftrace_func(ftrace_trace_function);
2898 if (FTRACE_WARN_ON(err))
2899 return;
2900 }
2901
2902 if (command & FTRACE_START_FUNC_RET)
2903 err = ftrace_enable_ftrace_graph_caller();
2904 else if (command & FTRACE_STOP_FUNC_RET)
2905 err = ftrace_disable_ftrace_graph_caller();
2906 FTRACE_WARN_ON(err);
2907 }
2908
__ftrace_modify_code(void * data)2909 static int __ftrace_modify_code(void *data)
2910 {
2911 int *command = data;
2912
2913 ftrace_modify_all_code(*command);
2914
2915 return 0;
2916 }
2917
2918 /**
2919 * ftrace_run_stop_machine - go back to the stop machine method
2920 * @command: The command to tell ftrace what to do
2921 *
2922 * If an arch needs to fall back to the stop machine method, the
2923 * it can call this function.
2924 */
ftrace_run_stop_machine(int command)2925 void ftrace_run_stop_machine(int command)
2926 {
2927 stop_machine(__ftrace_modify_code, &command, NULL);
2928 }
2929
2930 /**
2931 * arch_ftrace_update_code - modify the code to trace or not trace
2932 * @command: The command that needs to be done
2933 *
2934 * Archs can override this function if it does not need to
2935 * run stop_machine() to modify code.
2936 */
arch_ftrace_update_code(int command)2937 void __weak arch_ftrace_update_code(int command)
2938 {
2939 ftrace_run_stop_machine(command);
2940 }
2941
ftrace_run_update_code(int command)2942 static void ftrace_run_update_code(int command)
2943 {
2944 ftrace_arch_code_modify_prepare();
2945
2946 /*
2947 * By default we use stop_machine() to modify the code.
2948 * But archs can do what ever they want as long as it
2949 * is safe. The stop_machine() is the safest, but also
2950 * produces the most overhead.
2951 */
2952 arch_ftrace_update_code(command);
2953
2954 ftrace_arch_code_modify_post_process();
2955 }
2956
ftrace_run_modify_code(struct ftrace_ops * ops,int command,struct ftrace_ops_hash * old_hash)2957 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2958 struct ftrace_ops_hash *old_hash)
2959 {
2960 ops->flags |= FTRACE_OPS_FL_MODIFYING;
2961 ops->old_hash.filter_hash = old_hash->filter_hash;
2962 ops->old_hash.notrace_hash = old_hash->notrace_hash;
2963 ftrace_run_update_code(command);
2964 ops->old_hash.filter_hash = NULL;
2965 ops->old_hash.notrace_hash = NULL;
2966 ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2967 }
2968
2969 static ftrace_func_t saved_ftrace_func;
2970 static int ftrace_start_up;
2971
arch_ftrace_trampoline_free(struct ftrace_ops * ops)2972 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2973 {
2974 }
2975
2976 /* List of trace_ops that have allocated trampolines */
2977 static LIST_HEAD(ftrace_ops_trampoline_list);
2978
ftrace_add_trampoline_to_kallsyms(struct ftrace_ops * ops)2979 static void ftrace_add_trampoline_to_kallsyms(struct ftrace_ops *ops)
2980 {
2981 lockdep_assert_held(&ftrace_lock);
2982 list_add_rcu(&ops->list, &ftrace_ops_trampoline_list);
2983 }
2984
ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops * ops)2985 static void ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops *ops)
2986 {
2987 lockdep_assert_held(&ftrace_lock);
2988 list_del_rcu(&ops->list);
2989 synchronize_rcu();
2990 }
2991
2992 /*
2993 * "__builtin__ftrace" is used as a module name in /proc/kallsyms for symbols
2994 * for pages allocated for ftrace purposes, even though "__builtin__ftrace" is
2995 * not a module.
2996 */
2997 #define FTRACE_TRAMPOLINE_MOD "__builtin__ftrace"
2998 #define FTRACE_TRAMPOLINE_SYM "ftrace_trampoline"
2999
ftrace_trampoline_free(struct ftrace_ops * ops)3000 static void ftrace_trampoline_free(struct ftrace_ops *ops)
3001 {
3002 if (ops && (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP) &&
3003 ops->trampoline) {
3004 /*
3005 * Record the text poke event before the ksymbol unregister
3006 * event.
3007 */
3008 perf_event_text_poke((void *)ops->trampoline,
3009 (void *)ops->trampoline,
3010 ops->trampoline_size, NULL, 0);
3011 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
3012 ops->trampoline, ops->trampoline_size,
3013 true, FTRACE_TRAMPOLINE_SYM);
3014 /* Remove from kallsyms after the perf events */
3015 ftrace_remove_trampoline_from_kallsyms(ops);
3016 }
3017
3018 arch_ftrace_trampoline_free(ops);
3019 }
3020
ftrace_startup_enable(int command)3021 static void ftrace_startup_enable(int command)
3022 {
3023 if (saved_ftrace_func != ftrace_trace_function) {
3024 saved_ftrace_func = ftrace_trace_function;
3025 command |= FTRACE_UPDATE_TRACE_FUNC;
3026 }
3027
3028 if (!command || !ftrace_enabled)
3029 return;
3030
3031 ftrace_run_update_code(command);
3032 }
3033
ftrace_startup_all(int command)3034 static void ftrace_startup_all(int command)
3035 {
3036 update_all_ops = true;
3037 ftrace_startup_enable(command);
3038 update_all_ops = false;
3039 }
3040
ftrace_startup(struct ftrace_ops * ops,int command)3041 int ftrace_startup(struct ftrace_ops *ops, int command)
3042 {
3043 int ret;
3044
3045 if (unlikely(ftrace_disabled))
3046 return -ENODEV;
3047
3048 ret = __register_ftrace_function(ops);
3049 if (ret)
3050 return ret;
3051
3052 ftrace_start_up++;
3053
3054 /*
3055 * Note that ftrace probes uses this to start up
3056 * and modify functions it will probe. But we still
3057 * set the ADDING flag for modification, as probes
3058 * do not have trampolines. If they add them in the
3059 * future, then the probes will need to distinguish
3060 * between adding and updating probes.
3061 */
3062 ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
3063
3064 ret = ftrace_hash_ipmodify_enable(ops);
3065 if (ret < 0) {
3066 /* Rollback registration process */
3067 __unregister_ftrace_function(ops);
3068 ftrace_start_up--;
3069 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
3070 if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
3071 ftrace_trampoline_free(ops);
3072 return ret;
3073 }
3074
3075 if (ftrace_hash_rec_enable(ops))
3076 command |= FTRACE_UPDATE_CALLS;
3077
3078 ftrace_startup_enable(command);
3079
3080 /*
3081 * If ftrace is in an undefined state, we just remove ops from list
3082 * to prevent the NULL pointer, instead of totally rolling it back and
3083 * free trampoline, because those actions could cause further damage.
3084 */
3085 if (unlikely(ftrace_disabled)) {
3086 __unregister_ftrace_function(ops);
3087 return -ENODEV;
3088 }
3089
3090 ops->flags &= ~FTRACE_OPS_FL_ADDING;
3091
3092 return 0;
3093 }
3094
ftrace_shutdown(struct ftrace_ops * ops,int command)3095 int ftrace_shutdown(struct ftrace_ops *ops, int command)
3096 {
3097 int ret;
3098
3099 if (unlikely(ftrace_disabled))
3100 return -ENODEV;
3101
3102 ret = __unregister_ftrace_function(ops);
3103 if (ret)
3104 return ret;
3105
3106 ftrace_start_up--;
3107 /*
3108 * Just warn in case of unbalance, no need to kill ftrace, it's not
3109 * critical but the ftrace_call callers may be never nopped again after
3110 * further ftrace uses.
3111 */
3112 WARN_ON_ONCE(ftrace_start_up < 0);
3113
3114 /* Disabling ipmodify never fails */
3115 ftrace_hash_ipmodify_disable(ops);
3116
3117 if (ftrace_hash_rec_disable(ops))
3118 command |= FTRACE_UPDATE_CALLS;
3119
3120 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
3121
3122 if (saved_ftrace_func != ftrace_trace_function) {
3123 saved_ftrace_func = ftrace_trace_function;
3124 command |= FTRACE_UPDATE_TRACE_FUNC;
3125 }
3126
3127 if (!command || !ftrace_enabled)
3128 goto out;
3129
3130 /*
3131 * If the ops uses a trampoline, then it needs to be
3132 * tested first on update.
3133 */
3134 ops->flags |= FTRACE_OPS_FL_REMOVING;
3135 removed_ops = ops;
3136
3137 /* The trampoline logic checks the old hashes */
3138 ops->old_hash.filter_hash = ops->func_hash->filter_hash;
3139 ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
3140
3141 ftrace_run_update_code(command);
3142
3143 /*
3144 * If there's no more ops registered with ftrace, run a
3145 * sanity check to make sure all rec flags are cleared.
3146 */
3147 if (rcu_dereference_protected(ftrace_ops_list,
3148 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
3149 struct ftrace_page *pg;
3150 struct dyn_ftrace *rec;
3151
3152 do_for_each_ftrace_rec(pg, rec) {
3153 if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_NOCLEAR_FLAGS))
3154 pr_warn(" %pS flags:%lx\n",
3155 (void *)rec->ip, rec->flags);
3156 } while_for_each_ftrace_rec();
3157 }
3158
3159 ops->old_hash.filter_hash = NULL;
3160 ops->old_hash.notrace_hash = NULL;
3161
3162 removed_ops = NULL;
3163 ops->flags &= ~FTRACE_OPS_FL_REMOVING;
3164
3165 out:
3166 /*
3167 * Dynamic ops may be freed, we must make sure that all
3168 * callers are done before leaving this function.
3169 */
3170 if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
3171 /*
3172 * We need to do a hard force of sched synchronization.
3173 * This is because we use preempt_disable() to do RCU, but
3174 * the function tracers can be called where RCU is not watching
3175 * (like before user_exit()). We can not rely on the RCU
3176 * infrastructure to do the synchronization, thus we must do it
3177 * ourselves.
3178 */
3179 synchronize_rcu_tasks_rude();
3180
3181 /*
3182 * When the kernel is preemptive, tasks can be preempted
3183 * while on a ftrace trampoline. Just scheduling a task on
3184 * a CPU is not good enough to flush them. Calling
3185 * synchronize_rcu_tasks() will wait for those tasks to
3186 * execute and either schedule voluntarily or enter user space.
3187 */
3188 synchronize_rcu_tasks();
3189
3190 ftrace_trampoline_free(ops);
3191 }
3192
3193 return 0;
3194 }
3195
3196 /* Simply make a copy of @src and return it */
copy_hash(struct ftrace_hash * src)3197 static struct ftrace_hash *copy_hash(struct ftrace_hash *src)
3198 {
3199 if (ftrace_hash_empty(src))
3200 return EMPTY_HASH;
3201
3202 return alloc_and_copy_ftrace_hash(src->size_bits, src);
3203 }
3204
3205 /*
3206 * Append @new_hash entries to @hash:
3207 *
3208 * If @hash is the EMPTY_HASH then it traces all functions and nothing
3209 * needs to be done.
3210 *
3211 * If @new_hash is the EMPTY_HASH, then make *hash the EMPTY_HASH so
3212 * that it traces everything.
3213 *
3214 * Otherwise, go through all of @new_hash and add anything that @hash
3215 * doesn't already have, to @hash.
3216 *
3217 * The filter_hash updates uses just the append_hash() function
3218 * and the notrace_hash does not.
3219 */
append_hash(struct ftrace_hash ** hash,struct ftrace_hash * new_hash,int size_bits)3220 static int append_hash(struct ftrace_hash **hash, struct ftrace_hash *new_hash,
3221 int size_bits)
3222 {
3223 struct ftrace_func_entry *entry;
3224 int size;
3225 int i;
3226
3227 if (*hash) {
3228 /* An empty hash does everything */
3229 if (ftrace_hash_empty(*hash))
3230 return 0;
3231 } else {
3232 *hash = alloc_ftrace_hash(size_bits);
3233 if (!*hash)
3234 return -ENOMEM;
3235 }
3236
3237 /* If new_hash has everything make hash have everything */
3238 if (ftrace_hash_empty(new_hash)) {
3239 free_ftrace_hash(*hash);
3240 *hash = EMPTY_HASH;
3241 return 0;
3242 }
3243
3244 size = 1 << new_hash->size_bits;
3245 for (i = 0; i < size; i++) {
3246 hlist_for_each_entry(entry, &new_hash->buckets[i], hlist) {
3247 /* Only add if not already in hash */
3248 if (!__ftrace_lookup_ip(*hash, entry->ip) &&
3249 add_hash_entry(*hash, entry->ip) == NULL)
3250 return -ENOMEM;
3251 }
3252 }
3253 return 0;
3254 }
3255
3256 /*
3257 * Remove functions from @hash that are in @notrace_hash
3258 */
remove_hash(struct ftrace_hash * hash,struct ftrace_hash * notrace_hash)3259 static void remove_hash(struct ftrace_hash *hash, struct ftrace_hash *notrace_hash)
3260 {
3261 struct ftrace_func_entry *entry;
3262 struct hlist_node *tmp;
3263 int size;
3264 int i;
3265
3266 /* If the notrace hash is empty, there's nothing to do */
3267 if (ftrace_hash_empty(notrace_hash))
3268 return;
3269
3270 size = 1 << hash->size_bits;
3271 for (i = 0; i < size; i++) {
3272 hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
3273 if (!__ftrace_lookup_ip(notrace_hash, entry->ip))
3274 continue;
3275 remove_hash_entry(hash, entry);
3276 kfree(entry);
3277 }
3278 }
3279 }
3280
3281 /*
3282 * Add to @hash only those that are in both @new_hash1 and @new_hash2
3283 *
3284 * The notrace_hash updates uses just the intersect_hash() function
3285 * and the filter_hash does not.
3286 */
intersect_hash(struct ftrace_hash ** hash,struct ftrace_hash * new_hash1,struct ftrace_hash * new_hash2)3287 static int intersect_hash(struct ftrace_hash **hash, struct ftrace_hash *new_hash1,
3288 struct ftrace_hash *new_hash2)
3289 {
3290 struct ftrace_func_entry *entry;
3291 int size;
3292 int i;
3293
3294 /*
3295 * If new_hash1 or new_hash2 is the EMPTY_HASH then make the hash
3296 * empty as well as empty for notrace means none are notraced.
3297 */
3298 if (ftrace_hash_empty(new_hash1) || ftrace_hash_empty(new_hash2)) {
3299 free_ftrace_hash(*hash);
3300 *hash = EMPTY_HASH;
3301 return 0;
3302 }
3303
3304 size = 1 << new_hash1->size_bits;
3305 for (i = 0; i < size; i++) {
3306 hlist_for_each_entry(entry, &new_hash1->buckets[i], hlist) {
3307 /* Only add if in both @new_hash1 and @new_hash2 */
3308 if (__ftrace_lookup_ip(new_hash2, entry->ip) &&
3309 add_hash_entry(*hash, entry->ip) == NULL)
3310 return -ENOMEM;
3311 }
3312 }
3313 /* If nothing intersects, make it the empty set */
3314 if (ftrace_hash_empty(*hash)) {
3315 free_ftrace_hash(*hash);
3316 *hash = EMPTY_HASH;
3317 }
3318 return 0;
3319 }
3320
ops_equal(struct ftrace_hash * A,struct ftrace_hash * B)3321 static bool ops_equal(struct ftrace_hash *A, struct ftrace_hash *B)
3322 {
3323 struct ftrace_func_entry *entry;
3324 int size;
3325 int i;
3326
3327 if (ftrace_hash_empty(A))
3328 return ftrace_hash_empty(B);
3329
3330 if (ftrace_hash_empty(B))
3331 return ftrace_hash_empty(A);
3332
3333 if (A->count != B->count)
3334 return false;
3335
3336 size = 1 << A->size_bits;
3337 for (i = 0; i < size; i++) {
3338 hlist_for_each_entry(entry, &A->buckets[i], hlist) {
3339 if (!__ftrace_lookup_ip(B, entry->ip))
3340 return false;
3341 }
3342 }
3343
3344 return true;
3345 }
3346
3347 static void ftrace_ops_update_code(struct ftrace_ops *ops,
3348 struct ftrace_ops_hash *old_hash);
3349
__ftrace_hash_move_and_update_ops(struct ftrace_ops * ops,struct ftrace_hash ** orig_hash,struct ftrace_hash * hash,int enable)3350 static int __ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
3351 struct ftrace_hash **orig_hash,
3352 struct ftrace_hash *hash,
3353 int enable)
3354 {
3355 struct ftrace_ops_hash old_hash_ops;
3356 struct ftrace_hash *old_hash;
3357 int ret;
3358
3359 old_hash = *orig_hash;
3360 old_hash_ops.filter_hash = ops->func_hash->filter_hash;
3361 old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
3362 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3363 if (!ret) {
3364 ftrace_ops_update_code(ops, &old_hash_ops);
3365 free_ftrace_hash_rcu(old_hash);
3366 }
3367 return ret;
3368 }
3369
ftrace_update_ops(struct ftrace_ops * ops,struct ftrace_hash * filter_hash,struct ftrace_hash * notrace_hash)3370 static int ftrace_update_ops(struct ftrace_ops *ops, struct ftrace_hash *filter_hash,
3371 struct ftrace_hash *notrace_hash)
3372 {
3373 int ret;
3374
3375 if (!ops_equal(filter_hash, ops->func_hash->filter_hash)) {
3376 ret = __ftrace_hash_move_and_update_ops(ops, &ops->func_hash->filter_hash,
3377 filter_hash, 1);
3378 if (ret < 0)
3379 return ret;
3380 }
3381
3382 if (!ops_equal(notrace_hash, ops->func_hash->notrace_hash)) {
3383 ret = __ftrace_hash_move_and_update_ops(ops, &ops->func_hash->notrace_hash,
3384 notrace_hash, 0);
3385 if (ret < 0)
3386 return ret;
3387 }
3388
3389 return 0;
3390 }
3391
add_first_hash(struct ftrace_hash ** filter_hash,struct ftrace_hash ** notrace_hash,struct ftrace_ops_hash * func_hash)3392 static int add_first_hash(struct ftrace_hash **filter_hash, struct ftrace_hash **notrace_hash,
3393 struct ftrace_ops_hash *func_hash)
3394 {
3395 /* If the filter hash is not empty, simply remove the nohash from it */
3396 if (!ftrace_hash_empty(func_hash->filter_hash)) {
3397 *filter_hash = copy_hash(func_hash->filter_hash);
3398 if (!*filter_hash)
3399 return -ENOMEM;
3400 remove_hash(*filter_hash, func_hash->notrace_hash);
3401 *notrace_hash = EMPTY_HASH;
3402
3403 } else {
3404 *notrace_hash = copy_hash(func_hash->notrace_hash);
3405 if (!*notrace_hash)
3406 return -ENOMEM;
3407 *filter_hash = EMPTY_HASH;
3408 }
3409 return 0;
3410 }
3411
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)3412 static int add_next_hash(struct ftrace_hash **filter_hash, struct ftrace_hash **notrace_hash,
3413 struct ftrace_ops_hash *ops_hash, struct ftrace_ops_hash *subops_hash)
3414 {
3415 int size_bits;
3416 int ret;
3417
3418 /* If the subops trace all functions so must the main ops */
3419 if (ftrace_hash_empty(ops_hash->filter_hash) ||
3420 ftrace_hash_empty(subops_hash->filter_hash)) {
3421 *filter_hash = EMPTY_HASH;
3422 } else {
3423 /*
3424 * The main ops filter hash is not empty, so its
3425 * notrace_hash had better be, as the notrace hash
3426 * is only used for empty main filter hashes.
3427 */
3428 WARN_ON_ONCE(!ftrace_hash_empty(ops_hash->notrace_hash));
3429
3430 size_bits = max(ops_hash->filter_hash->size_bits,
3431 subops_hash->filter_hash->size_bits);
3432
3433 /* Copy the subops hash */
3434 *filter_hash = alloc_and_copy_ftrace_hash(size_bits, subops_hash->filter_hash);
3435 if (!*filter_hash)
3436 return -ENOMEM;
3437 /* Remove any notrace functions from the copy */
3438 remove_hash(*filter_hash, subops_hash->notrace_hash);
3439
3440 ret = append_hash(filter_hash, ops_hash->filter_hash,
3441 size_bits);
3442 if (ret < 0) {
3443 free_ftrace_hash(*filter_hash);
3444 *filter_hash = EMPTY_HASH;
3445 return ret;
3446 }
3447 }
3448
3449 /*
3450 * Only process notrace hashes if the main filter hash is empty
3451 * (tracing all functions), otherwise the filter hash will just
3452 * remove the notrace hash functions, and the notrace hash is
3453 * not needed.
3454 */
3455 if (ftrace_hash_empty(*filter_hash)) {
3456 /*
3457 * Intersect the notrace functions. That is, if two
3458 * subops are not tracing a set of functions, the
3459 * main ops will only not trace the functions that are
3460 * in both subops, but has to trace the functions that
3461 * are only notrace in one of the subops, for the other
3462 * subops to be able to trace them.
3463 */
3464 size_bits = max(ops_hash->notrace_hash->size_bits,
3465 subops_hash->notrace_hash->size_bits);
3466 *notrace_hash = alloc_ftrace_hash(size_bits);
3467 if (!*notrace_hash)
3468 return -ENOMEM;
3469
3470 ret = intersect_hash(notrace_hash, ops_hash->notrace_hash,
3471 subops_hash->notrace_hash);
3472 if (ret < 0) {
3473 free_ftrace_hash(*notrace_hash);
3474 *notrace_hash = EMPTY_HASH;
3475 return ret;
3476 }
3477 }
3478 return 0;
3479 }
3480
3481 /**
3482 * ftrace_startup_subops - enable tracing for subops of an ops
3483 * @ops: Manager ops (used to pick all the functions of its subops)
3484 * @subops: A new ops to add to @ops
3485 * @command: Extra commands to use to enable tracing
3486 *
3487 * The @ops is a manager @ops that has the filter that includes all the functions
3488 * that its list of subops are tracing. Adding a new @subops will add the
3489 * functions of @subops to @ops.
3490 */
ftrace_startup_subops(struct ftrace_ops * ops,struct ftrace_ops * subops,int command)3491 int ftrace_startup_subops(struct ftrace_ops *ops, struct ftrace_ops *subops, int command)
3492 {
3493 struct ftrace_hash *filter_hash = EMPTY_HASH;
3494 struct ftrace_hash *notrace_hash = EMPTY_HASH;
3495 struct ftrace_hash *save_filter_hash;
3496 struct ftrace_hash *save_notrace_hash;
3497 int ret;
3498
3499 if (unlikely(ftrace_disabled))
3500 return -ENODEV;
3501
3502 ftrace_ops_init(ops);
3503 ftrace_ops_init(subops);
3504
3505 if (WARN_ON_ONCE(subops->flags & FTRACE_OPS_FL_ENABLED))
3506 return -EBUSY;
3507
3508 /* Make everything canonical (Just in case!) */
3509 if (!ops->func_hash->filter_hash)
3510 ops->func_hash->filter_hash = EMPTY_HASH;
3511 if (!ops->func_hash->notrace_hash)
3512 ops->func_hash->notrace_hash = EMPTY_HASH;
3513 if (!subops->func_hash->filter_hash)
3514 subops->func_hash->filter_hash = EMPTY_HASH;
3515 if (!subops->func_hash->notrace_hash)
3516 subops->func_hash->notrace_hash = EMPTY_HASH;
3517
3518 /* For the first subops to ops just enable it normally */
3519 if (list_empty(&ops->subop_list)) {
3520
3521 /* The ops was empty, should have empty hashes */
3522 WARN_ON_ONCE(!ftrace_hash_empty(ops->func_hash->filter_hash));
3523 WARN_ON_ONCE(!ftrace_hash_empty(ops->func_hash->notrace_hash));
3524
3525 ret = add_first_hash(&filter_hash, ¬race_hash, subops->func_hash);
3526 if (ret < 0)
3527 return ret;
3528
3529 save_filter_hash = ops->func_hash->filter_hash;
3530 save_notrace_hash = ops->func_hash->notrace_hash;
3531
3532 ops->func_hash->filter_hash = filter_hash;
3533 ops->func_hash->notrace_hash = notrace_hash;
3534 list_add(&subops->list, &ops->subop_list);
3535 ret = ftrace_startup(ops, command);
3536 if (ret < 0) {
3537 list_del(&subops->list);
3538 ops->func_hash->filter_hash = save_filter_hash;
3539 ops->func_hash->notrace_hash = save_notrace_hash;
3540 free_ftrace_hash(filter_hash);
3541 free_ftrace_hash(notrace_hash);
3542 } else {
3543 free_ftrace_hash(save_filter_hash);
3544 free_ftrace_hash(save_notrace_hash);
3545 subops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP;
3546 subops->managed = ops;
3547 }
3548 return ret;
3549 }
3550
3551 /*
3552 * Here there's already something attached. Here are the rules:
3553 * If the new subops and main ops filter hashes are not empty:
3554 * o Make a copy of the subops filter hash
3555 * o Remove all functions in the nohash from it.
3556 * o Add in the main hash filter functions
3557 * o Remove any of these functions from the main notrace hash
3558 */
3559
3560 ret = add_next_hash(&filter_hash, ¬race_hash, ops->func_hash, subops->func_hash);
3561 if (ret < 0)
3562 return ret;
3563
3564 list_add(&subops->list, &ops->subop_list);
3565
3566 ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3567 free_ftrace_hash(filter_hash);
3568 free_ftrace_hash(notrace_hash);
3569 if (ret < 0) {
3570 list_del(&subops->list);
3571 } else {
3572 subops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP;
3573 subops->managed = ops;
3574 }
3575 return ret;
3576 }
3577
rebuild_hashes(struct ftrace_hash ** filter_hash,struct ftrace_hash ** notrace_hash,struct ftrace_ops * ops)3578 static int rebuild_hashes(struct ftrace_hash **filter_hash, struct ftrace_hash **notrace_hash,
3579 struct ftrace_ops *ops)
3580 {
3581 struct ftrace_ops_hash temp_hash;
3582 struct ftrace_ops *subops;
3583 bool first = true;
3584 int ret;
3585
3586 temp_hash.filter_hash = EMPTY_HASH;
3587 temp_hash.notrace_hash = EMPTY_HASH;
3588
3589 list_for_each_entry(subops, &ops->subop_list, list) {
3590 *filter_hash = EMPTY_HASH;
3591 *notrace_hash = EMPTY_HASH;
3592
3593 if (first) {
3594 ret = add_first_hash(filter_hash, notrace_hash, subops->func_hash);
3595 if (ret < 0)
3596 return ret;
3597 first = false;
3598 } else {
3599 ret = add_next_hash(filter_hash, notrace_hash,
3600 &temp_hash, subops->func_hash);
3601 if (ret < 0) {
3602 free_ftrace_hash(temp_hash.filter_hash);
3603 free_ftrace_hash(temp_hash.notrace_hash);
3604 return ret;
3605 }
3606 }
3607
3608 free_ftrace_hash(temp_hash.filter_hash);
3609 free_ftrace_hash(temp_hash.notrace_hash);
3610
3611 temp_hash.filter_hash = *filter_hash;
3612 temp_hash.notrace_hash = *notrace_hash;
3613 }
3614 return 0;
3615 }
3616
3617 /**
3618 * ftrace_shutdown_subops - Remove a subops from a manager ops
3619 * @ops: A manager ops to remove @subops from
3620 * @subops: The subops to remove from @ops
3621 * @command: Any extra command flags to add to modifying the text
3622 *
3623 * Removes the functions being traced by the @subops from @ops. Note, it
3624 * will not affect functions that are being traced by other subops that
3625 * still exist in @ops.
3626 *
3627 * If the last subops is removed from @ops, then @ops is shutdown normally.
3628 */
ftrace_shutdown_subops(struct ftrace_ops * ops,struct ftrace_ops * subops,int command)3629 int ftrace_shutdown_subops(struct ftrace_ops *ops, struct ftrace_ops *subops, int command)
3630 {
3631 struct ftrace_hash *filter_hash = EMPTY_HASH;
3632 struct ftrace_hash *notrace_hash = EMPTY_HASH;
3633 int ret;
3634
3635 if (unlikely(ftrace_disabled))
3636 return -ENODEV;
3637
3638 if (WARN_ON_ONCE(!(subops->flags & FTRACE_OPS_FL_ENABLED)))
3639 return -EINVAL;
3640
3641 list_del(&subops->list);
3642
3643 if (list_empty(&ops->subop_list)) {
3644 /* Last one, just disable the current ops */
3645
3646 ret = ftrace_shutdown(ops, command);
3647 if (ret < 0) {
3648 list_add(&subops->list, &ops->subop_list);
3649 return ret;
3650 }
3651
3652 subops->flags &= ~FTRACE_OPS_FL_ENABLED;
3653
3654 free_ftrace_hash(ops->func_hash->filter_hash);
3655 free_ftrace_hash(ops->func_hash->notrace_hash);
3656 ops->func_hash->filter_hash = EMPTY_HASH;
3657 ops->func_hash->notrace_hash = EMPTY_HASH;
3658 subops->flags &= ~(FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP);
3659 subops->managed = NULL;
3660
3661 return 0;
3662 }
3663
3664 /* Rebuild the hashes without subops */
3665 ret = rebuild_hashes(&filter_hash, ¬race_hash, ops);
3666 if (ret < 0)
3667 return ret;
3668
3669 ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3670 if (ret < 0) {
3671 list_add(&subops->list, &ops->subop_list);
3672 } else {
3673 subops->flags &= ~(FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP);
3674 subops->managed = NULL;
3675 }
3676 free_ftrace_hash(filter_hash);
3677 free_ftrace_hash(notrace_hash);
3678 return ret;
3679 }
3680
ftrace_hash_move_and_update_subops(struct ftrace_ops * subops,struct ftrace_hash ** orig_subhash,struct ftrace_hash * hash)3681 static int ftrace_hash_move_and_update_subops(struct ftrace_ops *subops,
3682 struct ftrace_hash **orig_subhash,
3683 struct ftrace_hash *hash)
3684 {
3685 struct ftrace_ops *ops = subops->managed;
3686 struct ftrace_hash *notrace_hash;
3687 struct ftrace_hash *filter_hash;
3688 struct ftrace_hash *save_hash;
3689 struct ftrace_hash *new_hash;
3690 int ret;
3691
3692 /* Manager ops can not be subops (yet) */
3693 if (WARN_ON_ONCE(!ops || ops->flags & FTRACE_OPS_FL_SUBOP))
3694 return -EINVAL;
3695
3696 /* Move the new hash over to the subops hash */
3697 save_hash = *orig_subhash;
3698 *orig_subhash = __ftrace_hash_move(hash);
3699 if (!*orig_subhash) {
3700 *orig_subhash = save_hash;
3701 return -ENOMEM;
3702 }
3703
3704 ret = rebuild_hashes(&filter_hash, ¬race_hash, ops);
3705 if (!ret) {
3706 ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3707 free_ftrace_hash(filter_hash);
3708 free_ftrace_hash(notrace_hash);
3709 }
3710
3711 if (ret) {
3712 /* Put back the original hash */
3713 new_hash = *orig_subhash;
3714 *orig_subhash = save_hash;
3715 free_ftrace_hash_rcu(new_hash);
3716 } else {
3717 free_ftrace_hash_rcu(save_hash);
3718 }
3719 return ret;
3720 }
3721
3722
3723 u64 ftrace_update_time;
3724 u64 ftrace_total_mod_time;
3725 unsigned long ftrace_update_tot_cnt;
3726 unsigned long ftrace_number_of_pages;
3727 unsigned long ftrace_number_of_groups;
3728
ops_traces_mod(struct ftrace_ops * ops)3729 static inline int ops_traces_mod(struct ftrace_ops *ops)
3730 {
3731 /*
3732 * Filter_hash being empty will default to trace module.
3733 * But notrace hash requires a test of individual module functions.
3734 */
3735 return ftrace_hash_empty(ops->func_hash->filter_hash) &&
3736 ftrace_hash_empty(ops->func_hash->notrace_hash);
3737 }
3738
ftrace_update_code(struct module * mod,struct ftrace_page * new_pgs)3739 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
3740 {
3741 bool init_nop = ftrace_need_init_nop();
3742 struct ftrace_page *pg;
3743 struct dyn_ftrace *p;
3744 u64 start, stop, update_time;
3745 unsigned long update_cnt = 0;
3746 unsigned long rec_flags = 0;
3747 int i;
3748
3749 start = ftrace_now(raw_smp_processor_id());
3750
3751 /*
3752 * When a module is loaded, this function is called to convert
3753 * the calls to mcount in its text to nops, and also to create
3754 * an entry in the ftrace data. Now, if ftrace is activated
3755 * after this call, but before the module sets its text to
3756 * read-only, the modification of enabling ftrace can fail if
3757 * the read-only is done while ftrace is converting the calls.
3758 * To prevent this, the module's records are set as disabled
3759 * and will be enabled after the call to set the module's text
3760 * to read-only.
3761 */
3762 if (mod)
3763 rec_flags |= FTRACE_FL_DISABLED;
3764
3765 for (pg = new_pgs; pg; pg = pg->next) {
3766
3767 for (i = 0; i < pg->index; i++) {
3768
3769 /* If something went wrong, bail without enabling anything */
3770 if (unlikely(ftrace_disabled))
3771 return -1;
3772
3773 p = &pg->records[i];
3774 p->flags = rec_flags;
3775
3776 /*
3777 * Do the initial record conversion from mcount jump
3778 * to the NOP instructions.
3779 */
3780 if (init_nop && !ftrace_nop_initialize(mod, p))
3781 break;
3782
3783 update_cnt++;
3784 }
3785 }
3786
3787 stop = ftrace_now(raw_smp_processor_id());
3788 update_time = stop - start;
3789 if (mod)
3790 ftrace_total_mod_time += update_time;
3791 else
3792 ftrace_update_time = update_time;
3793 ftrace_update_tot_cnt += update_cnt;
3794
3795 return 0;
3796 }
3797
ftrace_allocate_records(struct ftrace_page * pg,int count)3798 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
3799 {
3800 int order;
3801 int pages;
3802 int cnt;
3803
3804 if (WARN_ON(!count))
3805 return -EINVAL;
3806
3807 /* We want to fill as much as possible, with no empty pages */
3808 pages = DIV_ROUND_UP(count, ENTRIES_PER_PAGE);
3809 order = fls(pages) - 1;
3810
3811 again:
3812 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
3813
3814 if (!pg->records) {
3815 /* if we can't allocate this size, try something smaller */
3816 if (!order)
3817 return -ENOMEM;
3818 order--;
3819 goto again;
3820 }
3821
3822 ftrace_number_of_pages += 1 << order;
3823 ftrace_number_of_groups++;
3824
3825 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
3826 pg->order = order;
3827
3828 if (cnt > count)
3829 cnt = count;
3830
3831 return cnt;
3832 }
3833
ftrace_free_pages(struct ftrace_page * pages)3834 static void ftrace_free_pages(struct ftrace_page *pages)
3835 {
3836 struct ftrace_page *pg = pages;
3837
3838 while (pg) {
3839 if (pg->records) {
3840 free_pages((unsigned long)pg->records, pg->order);
3841 ftrace_number_of_pages -= 1 << pg->order;
3842 }
3843 pages = pg->next;
3844 kfree(pg);
3845 pg = pages;
3846 ftrace_number_of_groups--;
3847 }
3848 }
3849
3850 static struct ftrace_page *
ftrace_allocate_pages(unsigned long num_to_init)3851 ftrace_allocate_pages(unsigned long num_to_init)
3852 {
3853 struct ftrace_page *start_pg;
3854 struct ftrace_page *pg;
3855 int cnt;
3856
3857 if (!num_to_init)
3858 return NULL;
3859
3860 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3861 if (!pg)
3862 return NULL;
3863
3864 /*
3865 * Try to allocate as much as possible in one continues
3866 * location that fills in all of the space. We want to
3867 * waste as little space as possible.
3868 */
3869 for (;;) {
3870 cnt = ftrace_allocate_records(pg, num_to_init);
3871 if (cnt < 0)
3872 goto free_pages;
3873
3874 num_to_init -= cnt;
3875 if (!num_to_init)
3876 break;
3877
3878 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3879 if (!pg->next)
3880 goto free_pages;
3881
3882 pg = pg->next;
3883 }
3884
3885 return start_pg;
3886
3887 free_pages:
3888 ftrace_free_pages(start_pg);
3889 pr_info("ftrace: FAILED to allocate memory for functions\n");
3890 return NULL;
3891 }
3892
3893 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3894
3895 struct ftrace_iterator {
3896 loff_t pos;
3897 loff_t func_pos;
3898 loff_t mod_pos;
3899 struct ftrace_page *pg;
3900 struct dyn_ftrace *func;
3901 struct ftrace_func_probe *probe;
3902 struct ftrace_func_entry *probe_entry;
3903 struct trace_parser parser;
3904 struct ftrace_hash *hash;
3905 struct ftrace_ops *ops;
3906 struct trace_array *tr;
3907 struct list_head *mod_list;
3908 int pidx;
3909 int idx;
3910 unsigned flags;
3911 };
3912
3913 static void *
t_probe_next(struct seq_file * m,loff_t * pos)3914 t_probe_next(struct seq_file *m, loff_t *pos)
3915 {
3916 struct ftrace_iterator *iter = m->private;
3917 struct trace_array *tr = iter->ops->private;
3918 struct list_head *func_probes;
3919 struct ftrace_hash *hash;
3920 struct list_head *next;
3921 struct hlist_node *hnd = NULL;
3922 struct hlist_head *hhd;
3923 int size;
3924
3925 (*pos)++;
3926 iter->pos = *pos;
3927
3928 if (!tr)
3929 return NULL;
3930
3931 func_probes = &tr->func_probes;
3932 if (list_empty(func_probes))
3933 return NULL;
3934
3935 if (!iter->probe) {
3936 next = func_probes->next;
3937 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3938 }
3939
3940 if (iter->probe_entry)
3941 hnd = &iter->probe_entry->hlist;
3942
3943 hash = iter->probe->ops.func_hash->filter_hash;
3944
3945 /*
3946 * A probe being registered may temporarily have an empty hash
3947 * and it's at the end of the func_probes list.
3948 */
3949 if (!hash || hash == EMPTY_HASH)
3950 return NULL;
3951
3952 size = 1 << hash->size_bits;
3953
3954 retry:
3955 if (iter->pidx >= size) {
3956 if (iter->probe->list.next == func_probes)
3957 return NULL;
3958 next = iter->probe->list.next;
3959 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3960 hash = iter->probe->ops.func_hash->filter_hash;
3961 size = 1 << hash->size_bits;
3962 iter->pidx = 0;
3963 }
3964
3965 hhd = &hash->buckets[iter->pidx];
3966
3967 if (hlist_empty(hhd)) {
3968 iter->pidx++;
3969 hnd = NULL;
3970 goto retry;
3971 }
3972
3973 if (!hnd)
3974 hnd = hhd->first;
3975 else {
3976 hnd = hnd->next;
3977 if (!hnd) {
3978 iter->pidx++;
3979 goto retry;
3980 }
3981 }
3982
3983 if (WARN_ON_ONCE(!hnd))
3984 return NULL;
3985
3986 iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
3987
3988 return iter;
3989 }
3990
t_probe_start(struct seq_file * m,loff_t * pos)3991 static void *t_probe_start(struct seq_file *m, loff_t *pos)
3992 {
3993 struct ftrace_iterator *iter = m->private;
3994 void *p = NULL;
3995 loff_t l;
3996
3997 if (!(iter->flags & FTRACE_ITER_DO_PROBES))
3998 return NULL;
3999
4000 if (iter->mod_pos > *pos)
4001 return NULL;
4002
4003 iter->probe = NULL;
4004 iter->probe_entry = NULL;
4005 iter->pidx = 0;
4006 for (l = 0; l <= (*pos - iter->mod_pos); ) {
4007 p = t_probe_next(m, &l);
4008 if (!p)
4009 break;
4010 }
4011 if (!p)
4012 return NULL;
4013
4014 /* Only set this if we have an item */
4015 iter->flags |= FTRACE_ITER_PROBE;
4016
4017 return iter;
4018 }
4019
4020 static int
t_probe_show(struct seq_file * m,struct ftrace_iterator * iter)4021 t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
4022 {
4023 struct ftrace_func_entry *probe_entry;
4024 struct ftrace_probe_ops *probe_ops;
4025 struct ftrace_func_probe *probe;
4026
4027 probe = iter->probe;
4028 probe_entry = iter->probe_entry;
4029
4030 if (WARN_ON_ONCE(!probe || !probe_entry))
4031 return -EIO;
4032
4033 probe_ops = probe->probe_ops;
4034
4035 if (probe_ops->print)
4036 return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
4037
4038 seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
4039 (void *)probe_ops->func);
4040
4041 return 0;
4042 }
4043
4044 static void *
t_mod_next(struct seq_file * m,loff_t * pos)4045 t_mod_next(struct seq_file *m, loff_t *pos)
4046 {
4047 struct ftrace_iterator *iter = m->private;
4048 struct trace_array *tr = iter->tr;
4049
4050 (*pos)++;
4051 iter->pos = *pos;
4052
4053 iter->mod_list = iter->mod_list->next;
4054
4055 if (iter->mod_list == &tr->mod_trace ||
4056 iter->mod_list == &tr->mod_notrace) {
4057 iter->flags &= ~FTRACE_ITER_MOD;
4058 return NULL;
4059 }
4060
4061 iter->mod_pos = *pos;
4062
4063 return iter;
4064 }
4065
t_mod_start(struct seq_file * m,loff_t * pos)4066 static void *t_mod_start(struct seq_file *m, loff_t *pos)
4067 {
4068 struct ftrace_iterator *iter = m->private;
4069 void *p = NULL;
4070 loff_t l;
4071
4072 if (iter->func_pos > *pos)
4073 return NULL;
4074
4075 iter->mod_pos = iter->func_pos;
4076
4077 /* probes are only available if tr is set */
4078 if (!iter->tr)
4079 return NULL;
4080
4081 for (l = 0; l <= (*pos - iter->func_pos); ) {
4082 p = t_mod_next(m, &l);
4083 if (!p)
4084 break;
4085 }
4086 if (!p) {
4087 iter->flags &= ~FTRACE_ITER_MOD;
4088 return t_probe_start(m, pos);
4089 }
4090
4091 /* Only set this if we have an item */
4092 iter->flags |= FTRACE_ITER_MOD;
4093
4094 return iter;
4095 }
4096
4097 static int
t_mod_show(struct seq_file * m,struct ftrace_iterator * iter)4098 t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
4099 {
4100 struct ftrace_mod_load *ftrace_mod;
4101 struct trace_array *tr = iter->tr;
4102
4103 if (WARN_ON_ONCE(!iter->mod_list) ||
4104 iter->mod_list == &tr->mod_trace ||
4105 iter->mod_list == &tr->mod_notrace)
4106 return -EIO;
4107
4108 ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
4109
4110 if (ftrace_mod->func)
4111 seq_printf(m, "%s", ftrace_mod->func);
4112 else
4113 seq_putc(m, '*');
4114
4115 seq_printf(m, ":mod:%s\n", ftrace_mod->module);
4116
4117 return 0;
4118 }
4119
4120 static void *
t_func_next(struct seq_file * m,loff_t * pos)4121 t_func_next(struct seq_file *m, loff_t *pos)
4122 {
4123 struct ftrace_iterator *iter = m->private;
4124 struct dyn_ftrace *rec = NULL;
4125
4126 (*pos)++;
4127
4128 retry:
4129 if (iter->idx >= iter->pg->index) {
4130 if (iter->pg->next) {
4131 iter->pg = iter->pg->next;
4132 iter->idx = 0;
4133 goto retry;
4134 }
4135 } else {
4136 rec = &iter->pg->records[iter->idx++];
4137 if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
4138 !ftrace_lookup_ip(iter->hash, rec->ip)) ||
4139
4140 ((iter->flags & FTRACE_ITER_ENABLED) &&
4141 !(rec->flags & FTRACE_FL_ENABLED)) ||
4142
4143 ((iter->flags & FTRACE_ITER_TOUCHED) &&
4144 !(rec->flags & FTRACE_FL_TOUCHED))) {
4145
4146 rec = NULL;
4147 goto retry;
4148 }
4149 }
4150
4151 if (!rec)
4152 return NULL;
4153
4154 iter->pos = iter->func_pos = *pos;
4155 iter->func = rec;
4156
4157 return iter;
4158 }
4159
4160 static void *
t_next(struct seq_file * m,void * v,loff_t * pos)4161 t_next(struct seq_file *m, void *v, loff_t *pos)
4162 {
4163 struct ftrace_iterator *iter = m->private;
4164 loff_t l = *pos; /* t_probe_start() must use original pos */
4165 void *ret;
4166
4167 if (unlikely(ftrace_disabled))
4168 return NULL;
4169
4170 if (iter->flags & FTRACE_ITER_PROBE)
4171 return t_probe_next(m, pos);
4172
4173 if (iter->flags & FTRACE_ITER_MOD)
4174 return t_mod_next(m, pos);
4175
4176 if (iter->flags & FTRACE_ITER_PRINTALL) {
4177 /* next must increment pos, and t_probe_start does not */
4178 (*pos)++;
4179 return t_mod_start(m, &l);
4180 }
4181
4182 ret = t_func_next(m, pos);
4183
4184 if (!ret)
4185 return t_mod_start(m, &l);
4186
4187 return ret;
4188 }
4189
reset_iter_read(struct ftrace_iterator * iter)4190 static void reset_iter_read(struct ftrace_iterator *iter)
4191 {
4192 iter->pos = 0;
4193 iter->func_pos = 0;
4194 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
4195 }
4196
t_start(struct seq_file * m,loff_t * pos)4197 static void *t_start(struct seq_file *m, loff_t *pos)
4198 {
4199 struct ftrace_iterator *iter = m->private;
4200 void *p = NULL;
4201 loff_t l;
4202
4203 mutex_lock(&ftrace_lock);
4204
4205 if (unlikely(ftrace_disabled))
4206 return NULL;
4207
4208 /*
4209 * If an lseek was done, then reset and start from beginning.
4210 */
4211 if (*pos < iter->pos)
4212 reset_iter_read(iter);
4213
4214 /*
4215 * For set_ftrace_filter reading, if we have the filter
4216 * off, we can short cut and just print out that all
4217 * functions are enabled.
4218 */
4219 if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
4220 ftrace_hash_empty(iter->hash)) {
4221 iter->func_pos = 1; /* Account for the message */
4222 if (*pos > 0)
4223 return t_mod_start(m, pos);
4224 iter->flags |= FTRACE_ITER_PRINTALL;
4225 /* reset in case of seek/pread */
4226 iter->flags &= ~FTRACE_ITER_PROBE;
4227 return iter;
4228 }
4229
4230 if (iter->flags & FTRACE_ITER_MOD)
4231 return t_mod_start(m, pos);
4232
4233 /*
4234 * Unfortunately, we need to restart at ftrace_pages_start
4235 * every time we let go of the ftrace_mutex. This is because
4236 * those pointers can change without the lock.
4237 */
4238 iter->pg = ftrace_pages_start;
4239 iter->idx = 0;
4240 for (l = 0; l <= *pos; ) {
4241 p = t_func_next(m, &l);
4242 if (!p)
4243 break;
4244 }
4245
4246 if (!p)
4247 return t_mod_start(m, pos);
4248
4249 return iter;
4250 }
4251
t_stop(struct seq_file * m,void * p)4252 static void t_stop(struct seq_file *m, void *p)
4253 {
4254 mutex_unlock(&ftrace_lock);
4255 }
4256
4257 void * __weak
arch_ftrace_trampoline_func(struct ftrace_ops * ops,struct dyn_ftrace * rec)4258 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
4259 {
4260 return NULL;
4261 }
4262
add_trampoline_func(struct seq_file * m,struct ftrace_ops * ops,struct dyn_ftrace * rec)4263 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
4264 struct dyn_ftrace *rec)
4265 {
4266 void *ptr;
4267
4268 ptr = arch_ftrace_trampoline_func(ops, rec);
4269 if (ptr)
4270 seq_printf(m, " ->%pS", ptr);
4271 }
4272
4273 #ifdef FTRACE_MCOUNT_MAX_OFFSET
4274 /*
4275 * Weak functions can still have an mcount/fentry that is saved in
4276 * the __mcount_loc section. These can be detected by having a
4277 * symbol offset of greater than FTRACE_MCOUNT_MAX_OFFSET, as the
4278 * symbol found by kallsyms is not the function that the mcount/fentry
4279 * is part of. The offset is much greater in these cases.
4280 *
4281 * Test the record to make sure that the ip points to a valid kallsyms
4282 * and if not, mark it disabled.
4283 */
test_for_valid_rec(struct dyn_ftrace * rec)4284 static int test_for_valid_rec(struct dyn_ftrace *rec)
4285 {
4286 char str[KSYM_SYMBOL_LEN];
4287 unsigned long offset;
4288 const char *ret;
4289
4290 ret = kallsyms_lookup(rec->ip, NULL, &offset, NULL, str);
4291
4292 /* Weak functions can cause invalid addresses */
4293 if (!ret || offset > FTRACE_MCOUNT_MAX_OFFSET) {
4294 rec->flags |= FTRACE_FL_DISABLED;
4295 return 0;
4296 }
4297 return 1;
4298 }
4299
4300 static struct workqueue_struct *ftrace_check_wq __initdata;
4301 static struct work_struct ftrace_check_work __initdata;
4302
4303 /*
4304 * Scan all the mcount/fentry entries to make sure they are valid.
4305 */
ftrace_check_work_func(struct work_struct * work)4306 static __init void ftrace_check_work_func(struct work_struct *work)
4307 {
4308 struct ftrace_page *pg;
4309 struct dyn_ftrace *rec;
4310
4311 mutex_lock(&ftrace_lock);
4312 do_for_each_ftrace_rec(pg, rec) {
4313 test_for_valid_rec(rec);
4314 } while_for_each_ftrace_rec();
4315 mutex_unlock(&ftrace_lock);
4316 }
4317
ftrace_check_for_weak_functions(void)4318 static int __init ftrace_check_for_weak_functions(void)
4319 {
4320 INIT_WORK(&ftrace_check_work, ftrace_check_work_func);
4321
4322 ftrace_check_wq = alloc_workqueue("ftrace_check_wq", WQ_UNBOUND, 0);
4323
4324 queue_work(ftrace_check_wq, &ftrace_check_work);
4325 return 0;
4326 }
4327
ftrace_check_sync(void)4328 static int __init ftrace_check_sync(void)
4329 {
4330 /* Make sure the ftrace_check updates are finished */
4331 if (ftrace_check_wq)
4332 destroy_workqueue(ftrace_check_wq);
4333 return 0;
4334 }
4335
4336 late_initcall_sync(ftrace_check_sync);
4337 subsys_initcall(ftrace_check_for_weak_functions);
4338
print_rec(struct seq_file * m,unsigned long ip)4339 static int print_rec(struct seq_file *m, unsigned long ip)
4340 {
4341 unsigned long offset;
4342 char str[KSYM_SYMBOL_LEN];
4343 char *modname;
4344 const char *ret;
4345
4346 ret = kallsyms_lookup(ip, NULL, &offset, &modname, str);
4347 /* Weak functions can cause invalid addresses */
4348 if (!ret || offset > FTRACE_MCOUNT_MAX_OFFSET) {
4349 snprintf(str, KSYM_SYMBOL_LEN, "%s_%ld",
4350 FTRACE_INVALID_FUNCTION, offset);
4351 ret = NULL;
4352 }
4353
4354 seq_puts(m, str);
4355 if (modname)
4356 seq_printf(m, " [%s]", modname);
4357 return ret == NULL ? -1 : 0;
4358 }
4359 #else
test_for_valid_rec(struct dyn_ftrace * rec)4360 static inline int test_for_valid_rec(struct dyn_ftrace *rec)
4361 {
4362 return 1;
4363 }
4364
print_rec(struct seq_file * m,unsigned long ip)4365 static inline int print_rec(struct seq_file *m, unsigned long ip)
4366 {
4367 seq_printf(m, "%ps", (void *)ip);
4368 return 0;
4369 }
4370 #endif
4371
print_subops(struct seq_file * m,struct ftrace_ops * ops,struct dyn_ftrace * rec)4372 static void print_subops(struct seq_file *m, struct ftrace_ops *ops, struct dyn_ftrace *rec)
4373 {
4374 struct ftrace_ops *subops;
4375 bool first = true;
4376
4377 list_for_each_entry(subops, &ops->subop_list, list) {
4378 if (!((subops->flags & FTRACE_OPS_FL_ENABLED) &&
4379 hash_contains_ip(rec->ip, subops->func_hash)))
4380 continue;
4381 if (first) {
4382 seq_printf(m, "\tsubops:");
4383 first = false;
4384 }
4385 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4386 if (subops->flags & FTRACE_OPS_FL_GRAPH) {
4387 struct fgraph_ops *gops;
4388
4389 gops = container_of(subops, struct fgraph_ops, ops);
4390 seq_printf(m, " {ent:%pS ret:%pS}",
4391 (void *)gops->entryfunc,
4392 (void *)gops->retfunc);
4393 continue;
4394 }
4395 #endif
4396 if (subops->trampoline) {
4397 seq_printf(m, " {%pS (%pS)}",
4398 (void *)subops->trampoline,
4399 (void *)subops->func);
4400 add_trampoline_func(m, subops, rec);
4401 } else {
4402 seq_printf(m, " {%pS}",
4403 (void *)subops->func);
4404 }
4405 }
4406 }
4407
t_show(struct seq_file * m,void * v)4408 static int t_show(struct seq_file *m, void *v)
4409 {
4410 struct ftrace_iterator *iter = m->private;
4411 struct dyn_ftrace *rec;
4412
4413 if (iter->flags & FTRACE_ITER_PROBE)
4414 return t_probe_show(m, iter);
4415
4416 if (iter->flags & FTRACE_ITER_MOD)
4417 return t_mod_show(m, iter);
4418
4419 if (iter->flags & FTRACE_ITER_PRINTALL) {
4420 if (iter->flags & FTRACE_ITER_NOTRACE)
4421 seq_puts(m, "#### no functions disabled ####\n");
4422 else
4423 seq_puts(m, "#### all functions enabled ####\n");
4424 return 0;
4425 }
4426
4427 rec = iter->func;
4428
4429 if (!rec)
4430 return 0;
4431
4432 if (iter->flags & FTRACE_ITER_ADDRS)
4433 seq_printf(m, "%lx ", rec->ip);
4434
4435 if (print_rec(m, rec->ip)) {
4436 /* This should only happen when a rec is disabled */
4437 WARN_ON_ONCE(!(rec->flags & FTRACE_FL_DISABLED));
4438 seq_putc(m, '\n');
4439 return 0;
4440 }
4441
4442 if (iter->flags & (FTRACE_ITER_ENABLED | FTRACE_ITER_TOUCHED)) {
4443 struct ftrace_ops *ops;
4444
4445 seq_printf(m, " (%ld)%s%s%s%s%s",
4446 ftrace_rec_count(rec),
4447 rec->flags & FTRACE_FL_REGS ? " R" : " ",
4448 rec->flags & FTRACE_FL_IPMODIFY ? " I" : " ",
4449 rec->flags & FTRACE_FL_DIRECT ? " D" : " ",
4450 rec->flags & FTRACE_FL_CALL_OPS ? " O" : " ",
4451 rec->flags & FTRACE_FL_MODIFIED ? " M " : " ");
4452 if (rec->flags & FTRACE_FL_TRAMP_EN) {
4453 ops = ftrace_find_tramp_ops_any(rec);
4454 if (ops) {
4455 do {
4456 seq_printf(m, "\ttramp: %pS (%pS)",
4457 (void *)ops->trampoline,
4458 (void *)ops->func);
4459 add_trampoline_func(m, ops, rec);
4460 print_subops(m, ops, rec);
4461 ops = ftrace_find_tramp_ops_next(rec, ops);
4462 } while (ops);
4463 } else
4464 seq_puts(m, "\ttramp: ERROR!");
4465 } else {
4466 add_trampoline_func(m, NULL, rec);
4467 }
4468 if (rec->flags & FTRACE_FL_CALL_OPS_EN) {
4469 ops = ftrace_find_unique_ops(rec);
4470 if (ops) {
4471 seq_printf(m, "\tops: %pS (%pS)",
4472 ops, ops->func);
4473 print_subops(m, ops, rec);
4474 } else {
4475 seq_puts(m, "\tops: ERROR!");
4476 }
4477 }
4478 if (rec->flags & FTRACE_FL_DIRECT) {
4479 unsigned long direct;
4480
4481 direct = ftrace_find_rec_direct(rec->ip);
4482 if (direct)
4483 seq_printf(m, "\n\tdirect-->%pS", (void *)direct);
4484 }
4485 }
4486
4487 seq_putc(m, '\n');
4488
4489 return 0;
4490 }
4491
4492 static const struct seq_operations show_ftrace_seq_ops = {
4493 .start = t_start,
4494 .next = t_next,
4495 .stop = t_stop,
4496 .show = t_show,
4497 };
4498
4499 static int
ftrace_avail_open(struct inode * inode,struct file * file)4500 ftrace_avail_open(struct inode *inode, struct file *file)
4501 {
4502 struct ftrace_iterator *iter;
4503 int ret;
4504
4505 ret = security_locked_down(LOCKDOWN_TRACEFS);
4506 if (ret)
4507 return ret;
4508
4509 if (unlikely(ftrace_disabled))
4510 return -ENODEV;
4511
4512 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4513 if (!iter)
4514 return -ENOMEM;
4515
4516 iter->pg = ftrace_pages_start;
4517 iter->ops = &global_ops;
4518
4519 return 0;
4520 }
4521
4522 static int
ftrace_enabled_open(struct inode * inode,struct file * file)4523 ftrace_enabled_open(struct inode *inode, struct file *file)
4524 {
4525 struct ftrace_iterator *iter;
4526
4527 /*
4528 * This shows us what functions are currently being
4529 * traced and by what. Not sure if we want lockdown
4530 * to hide such critical information for an admin.
4531 * Although, perhaps it can show information we don't
4532 * want people to see, but if something is tracing
4533 * something, we probably want to know about it.
4534 */
4535
4536 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4537 if (!iter)
4538 return -ENOMEM;
4539
4540 iter->pg = ftrace_pages_start;
4541 iter->flags = FTRACE_ITER_ENABLED;
4542 iter->ops = &global_ops;
4543
4544 return 0;
4545 }
4546
4547 static int
ftrace_touched_open(struct inode * inode,struct file * file)4548 ftrace_touched_open(struct inode *inode, struct file *file)
4549 {
4550 struct ftrace_iterator *iter;
4551
4552 /*
4553 * This shows us what functions have ever been enabled
4554 * (traced, direct, patched, etc). Not sure if we want lockdown
4555 * to hide such critical information for an admin.
4556 * Although, perhaps it can show information we don't
4557 * want people to see, but if something had traced
4558 * something, we probably want to know about it.
4559 */
4560
4561 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4562 if (!iter)
4563 return -ENOMEM;
4564
4565 iter->pg = ftrace_pages_start;
4566 iter->flags = FTRACE_ITER_TOUCHED;
4567 iter->ops = &global_ops;
4568
4569 return 0;
4570 }
4571
4572 static int
ftrace_avail_addrs_open(struct inode * inode,struct file * file)4573 ftrace_avail_addrs_open(struct inode *inode, struct file *file)
4574 {
4575 struct ftrace_iterator *iter;
4576 int ret;
4577
4578 ret = security_locked_down(LOCKDOWN_TRACEFS);
4579 if (ret)
4580 return ret;
4581
4582 if (unlikely(ftrace_disabled))
4583 return -ENODEV;
4584
4585 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4586 if (!iter)
4587 return -ENOMEM;
4588
4589 iter->pg = ftrace_pages_start;
4590 iter->flags = FTRACE_ITER_ADDRS;
4591 iter->ops = &global_ops;
4592
4593 return 0;
4594 }
4595
4596 /**
4597 * ftrace_regex_open - initialize function tracer filter files
4598 * @ops: The ftrace_ops that hold the hash filters
4599 * @flag: The type of filter to process
4600 * @inode: The inode, usually passed in to your open routine
4601 * @file: The file, usually passed in to your open routine
4602 *
4603 * ftrace_regex_open() initializes the filter files for the
4604 * @ops. Depending on @flag it may process the filter hash or
4605 * the notrace hash of @ops. With this called from the open
4606 * routine, you can use ftrace_filter_write() for the write
4607 * routine if @flag has FTRACE_ITER_FILTER set, or
4608 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
4609 * tracing_lseek() should be used as the lseek routine, and
4610 * release must call ftrace_regex_release().
4611 *
4612 * Returns: 0 on success or a negative errno value on failure
4613 */
4614 int
ftrace_regex_open(struct ftrace_ops * ops,int flag,struct inode * inode,struct file * file)4615 ftrace_regex_open(struct ftrace_ops *ops, int flag,
4616 struct inode *inode, struct file *file)
4617 {
4618 struct ftrace_iterator *iter;
4619 struct ftrace_hash *hash;
4620 struct list_head *mod_head;
4621 struct trace_array *tr = ops->private;
4622 int ret = -ENOMEM;
4623
4624 ftrace_ops_init(ops);
4625
4626 if (unlikely(ftrace_disabled))
4627 return -ENODEV;
4628
4629 if (tracing_check_open_get_tr(tr))
4630 return -ENODEV;
4631
4632 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4633 if (!iter)
4634 goto out;
4635
4636 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX))
4637 goto out;
4638
4639 iter->ops = ops;
4640 iter->flags = flag;
4641 iter->tr = tr;
4642
4643 mutex_lock(&ops->func_hash->regex_lock);
4644
4645 if (flag & FTRACE_ITER_NOTRACE) {
4646 hash = ops->func_hash->notrace_hash;
4647 mod_head = tr ? &tr->mod_notrace : NULL;
4648 } else {
4649 hash = ops->func_hash->filter_hash;
4650 mod_head = tr ? &tr->mod_trace : NULL;
4651 }
4652
4653 iter->mod_list = mod_head;
4654
4655 if (file->f_mode & FMODE_WRITE) {
4656 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
4657
4658 if (file->f_flags & O_TRUNC) {
4659 iter->hash = alloc_ftrace_hash(size_bits);
4660 clear_ftrace_mod_list(mod_head);
4661 } else {
4662 iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
4663 }
4664
4665 if (!iter->hash) {
4666 trace_parser_put(&iter->parser);
4667 goto out_unlock;
4668 }
4669 } else
4670 iter->hash = hash;
4671
4672 ret = 0;
4673
4674 if (file->f_mode & FMODE_READ) {
4675 iter->pg = ftrace_pages_start;
4676
4677 ret = seq_open(file, &show_ftrace_seq_ops);
4678 if (!ret) {
4679 struct seq_file *m = file->private_data;
4680 m->private = iter;
4681 } else {
4682 /* Failed */
4683 free_ftrace_hash(iter->hash);
4684 trace_parser_put(&iter->parser);
4685 }
4686 } else
4687 file->private_data = iter;
4688
4689 out_unlock:
4690 mutex_unlock(&ops->func_hash->regex_lock);
4691
4692 out:
4693 if (ret) {
4694 kfree(iter);
4695 if (tr)
4696 trace_array_put(tr);
4697 }
4698
4699 return ret;
4700 }
4701
4702 static int
ftrace_filter_open(struct inode * inode,struct file * file)4703 ftrace_filter_open(struct inode *inode, struct file *file)
4704 {
4705 struct ftrace_ops *ops = inode->i_private;
4706
4707 /* Checks for tracefs lockdown */
4708 return ftrace_regex_open(ops,
4709 FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
4710 inode, file);
4711 }
4712
4713 static int
ftrace_notrace_open(struct inode * inode,struct file * file)4714 ftrace_notrace_open(struct inode *inode, struct file *file)
4715 {
4716 struct ftrace_ops *ops = inode->i_private;
4717
4718 /* Checks for tracefs lockdown */
4719 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
4720 inode, file);
4721 }
4722
4723 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
4724 struct ftrace_glob {
4725 char *search;
4726 unsigned len;
4727 int type;
4728 };
4729
4730 /*
4731 * If symbols in an architecture don't correspond exactly to the user-visible
4732 * name of what they represent, it is possible to define this function to
4733 * perform the necessary adjustments.
4734 */
arch_ftrace_match_adjust(char * str,const char * search)4735 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
4736 {
4737 return str;
4738 }
4739
ftrace_match(char * str,struct ftrace_glob * g)4740 static int ftrace_match(char *str, struct ftrace_glob *g)
4741 {
4742 int matched = 0;
4743 int slen;
4744
4745 str = arch_ftrace_match_adjust(str, g->search);
4746
4747 switch (g->type) {
4748 case MATCH_FULL:
4749 if (strcmp(str, g->search) == 0)
4750 matched = 1;
4751 break;
4752 case MATCH_FRONT_ONLY:
4753 if (strncmp(str, g->search, g->len) == 0)
4754 matched = 1;
4755 break;
4756 case MATCH_MIDDLE_ONLY:
4757 if (strstr(str, g->search))
4758 matched = 1;
4759 break;
4760 case MATCH_END_ONLY:
4761 slen = strlen(str);
4762 if (slen >= g->len &&
4763 memcmp(str + slen - g->len, g->search, g->len) == 0)
4764 matched = 1;
4765 break;
4766 case MATCH_GLOB:
4767 if (glob_match(g->search, str))
4768 matched = 1;
4769 break;
4770 }
4771
4772 return matched;
4773 }
4774
4775 static int
enter_record(struct ftrace_hash * hash,struct dyn_ftrace * rec,int clear_filter)4776 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
4777 {
4778 struct ftrace_func_entry *entry;
4779 int ret = 0;
4780
4781 entry = ftrace_lookup_ip(hash, rec->ip);
4782 if (clear_filter) {
4783 /* Do nothing if it doesn't exist */
4784 if (!entry)
4785 return 0;
4786
4787 free_hash_entry(hash, entry);
4788 } else {
4789 /* Do nothing if it exists */
4790 if (entry)
4791 return 0;
4792 if (add_hash_entry(hash, rec->ip) == NULL)
4793 ret = -ENOMEM;
4794 }
4795 return ret;
4796 }
4797
4798 static int
add_rec_by_index(struct ftrace_hash * hash,struct ftrace_glob * func_g,int clear_filter)4799 add_rec_by_index(struct ftrace_hash *hash, struct ftrace_glob *func_g,
4800 int clear_filter)
4801 {
4802 long index;
4803 struct ftrace_page *pg;
4804 struct dyn_ftrace *rec;
4805
4806 /* The index starts at 1 */
4807 if (kstrtoul(func_g->search, 0, &index) || --index < 0)
4808 return 0;
4809
4810 do_for_each_ftrace_rec(pg, rec) {
4811 if (pg->index <= index) {
4812 index -= pg->index;
4813 /* this is a double loop, break goes to the next page */
4814 break;
4815 }
4816 rec = &pg->records[index];
4817 enter_record(hash, rec, clear_filter);
4818 return 1;
4819 } while_for_each_ftrace_rec();
4820 return 0;
4821 }
4822
4823 #ifdef FTRACE_MCOUNT_MAX_OFFSET
lookup_ip(unsigned long ip,char ** modname,char * str)4824 static int lookup_ip(unsigned long ip, char **modname, char *str)
4825 {
4826 unsigned long offset;
4827
4828 kallsyms_lookup(ip, NULL, &offset, modname, str);
4829 if (offset > FTRACE_MCOUNT_MAX_OFFSET)
4830 return -1;
4831 return 0;
4832 }
4833 #else
lookup_ip(unsigned long ip,char ** modname,char * str)4834 static int lookup_ip(unsigned long ip, char **modname, char *str)
4835 {
4836 kallsyms_lookup(ip, NULL, NULL, modname, str);
4837 return 0;
4838 }
4839 #endif
4840
4841 static int
ftrace_match_record(struct dyn_ftrace * rec,struct ftrace_glob * func_g,struct ftrace_glob * mod_g,int exclude_mod)4842 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
4843 struct ftrace_glob *mod_g, int exclude_mod)
4844 {
4845 char str[KSYM_SYMBOL_LEN];
4846 char *modname;
4847
4848 if (lookup_ip(rec->ip, &modname, str)) {
4849 /* This should only happen when a rec is disabled */
4850 WARN_ON_ONCE(system_state == SYSTEM_RUNNING &&
4851 !(rec->flags & FTRACE_FL_DISABLED));
4852 return 0;
4853 }
4854
4855 if (mod_g) {
4856 int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
4857
4858 /* blank module name to match all modules */
4859 if (!mod_g->len) {
4860 /* blank module globbing: modname xor exclude_mod */
4861 if (!exclude_mod != !modname)
4862 goto func_match;
4863 return 0;
4864 }
4865
4866 /*
4867 * exclude_mod is set to trace everything but the given
4868 * module. If it is set and the module matches, then
4869 * return 0. If it is not set, and the module doesn't match
4870 * also return 0. Otherwise, check the function to see if
4871 * that matches.
4872 */
4873 if (!mod_matches == !exclude_mod)
4874 return 0;
4875 func_match:
4876 /* blank search means to match all funcs in the mod */
4877 if (!func_g->len)
4878 return 1;
4879 }
4880
4881 return ftrace_match(str, func_g);
4882 }
4883
4884 static int
match_records(struct ftrace_hash * hash,char * func,int len,char * mod)4885 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
4886 {
4887 struct ftrace_page *pg;
4888 struct dyn_ftrace *rec;
4889 struct ftrace_glob func_g = { .type = MATCH_FULL };
4890 struct ftrace_glob mod_g = { .type = MATCH_FULL };
4891 struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
4892 int exclude_mod = 0;
4893 int found = 0;
4894 int ret;
4895 int clear_filter = 0;
4896
4897 if (func) {
4898 func_g.type = filter_parse_regex(func, len, &func_g.search,
4899 &clear_filter);
4900 func_g.len = strlen(func_g.search);
4901 }
4902
4903 if (mod) {
4904 mod_g.type = filter_parse_regex(mod, strlen(mod),
4905 &mod_g.search, &exclude_mod);
4906 mod_g.len = strlen(mod_g.search);
4907 }
4908
4909 guard(mutex)(&ftrace_lock);
4910
4911 if (unlikely(ftrace_disabled))
4912 return 0;
4913
4914 if (func_g.type == MATCH_INDEX)
4915 return add_rec_by_index(hash, &func_g, clear_filter);
4916
4917 do_for_each_ftrace_rec(pg, rec) {
4918
4919 if (rec->flags & FTRACE_FL_DISABLED)
4920 continue;
4921
4922 if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
4923 ret = enter_record(hash, rec, clear_filter);
4924 if (ret < 0)
4925 return ret;
4926 found = 1;
4927 }
4928 cond_resched();
4929 } while_for_each_ftrace_rec();
4930
4931 return found;
4932 }
4933
4934 static int
ftrace_match_records(struct ftrace_hash * hash,char * buff,int len)4935 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
4936 {
4937 return match_records(hash, buff, len, NULL);
4938 }
4939
ftrace_ops_update_code(struct ftrace_ops * ops,struct ftrace_ops_hash * old_hash)4940 static void ftrace_ops_update_code(struct ftrace_ops *ops,
4941 struct ftrace_ops_hash *old_hash)
4942 {
4943 struct ftrace_ops *op;
4944
4945 if (!ftrace_enabled)
4946 return;
4947
4948 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
4949 ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
4950 return;
4951 }
4952
4953 /*
4954 * If this is the shared global_ops filter, then we need to
4955 * check if there is another ops that shares it, is enabled.
4956 * If so, we still need to run the modify code.
4957 */
4958 if (ops->func_hash != &global_ops.local_hash)
4959 return;
4960
4961 do_for_each_ftrace_op(op, ftrace_ops_list) {
4962 if (op->func_hash == &global_ops.local_hash &&
4963 op->flags & FTRACE_OPS_FL_ENABLED) {
4964 ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
4965 /* Only need to do this once */
4966 return;
4967 }
4968 } while_for_each_ftrace_op(op);
4969 }
4970
ftrace_hash_move_and_update_ops(struct ftrace_ops * ops,struct ftrace_hash ** orig_hash,struct ftrace_hash * hash,int enable)4971 static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
4972 struct ftrace_hash **orig_hash,
4973 struct ftrace_hash *hash,
4974 int enable)
4975 {
4976 if (ops->flags & FTRACE_OPS_FL_SUBOP)
4977 return ftrace_hash_move_and_update_subops(ops, orig_hash, hash);
4978
4979 /*
4980 * If this ops is not enabled, it could be sharing its filters
4981 * with a subop. If that's the case, update the subop instead of
4982 * this ops. Shared filters are only allowed to have one ops set
4983 * at a time, and if we update the ops that is not enabled,
4984 * it will not affect subops that share it.
4985 */
4986 if (!(ops->flags & FTRACE_OPS_FL_ENABLED)) {
4987 struct ftrace_ops *op;
4988
4989 /* Check if any other manager subops maps to this hash */
4990 do_for_each_ftrace_op(op, ftrace_ops_list) {
4991 struct ftrace_ops *subops;
4992
4993 list_for_each_entry(subops, &op->subop_list, list) {
4994 if ((subops->flags & FTRACE_OPS_FL_ENABLED) &&
4995 subops->func_hash == ops->func_hash) {
4996 return ftrace_hash_move_and_update_subops(subops, orig_hash, hash);
4997 }
4998 }
4999 } while_for_each_ftrace_op(op);
5000 }
5001
5002 return __ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
5003 }
5004
cache_mod(struct trace_array * tr,const char * func,char * module,int enable)5005 static int cache_mod(struct trace_array *tr,
5006 const char *func, char *module, int enable)
5007 {
5008 struct ftrace_mod_load *ftrace_mod, *n;
5009 struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
5010
5011 guard(mutex)(&ftrace_lock);
5012
5013 /* We do not cache inverse filters */
5014 if (func[0] == '!') {
5015 int ret = -EINVAL;
5016
5017 func++;
5018
5019 /* Look to remove this hash */
5020 list_for_each_entry_safe(ftrace_mod, n, head, list) {
5021 if (strcmp(ftrace_mod->module, module) != 0)
5022 continue;
5023
5024 /* no func matches all */
5025 if (strcmp(func, "*") == 0 ||
5026 (ftrace_mod->func &&
5027 strcmp(ftrace_mod->func, func) == 0)) {
5028 ret = 0;
5029 free_ftrace_mod(ftrace_mod);
5030 continue;
5031 }
5032 }
5033 return ret;
5034 }
5035
5036 /* We only care about modules that have not been loaded yet */
5037 if (module_exists(module))
5038 return -EINVAL;
5039
5040 /* Save this string off, and execute it when the module is loaded */
5041 return ftrace_add_mod(tr, func, module, enable);
5042 }
5043
5044 #ifdef CONFIG_MODULES
process_mod_list(struct list_head * head,struct ftrace_ops * ops,char * mod,bool enable)5045 static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
5046 char *mod, bool enable)
5047 {
5048 struct ftrace_mod_load *ftrace_mod, *n;
5049 struct ftrace_hash **orig_hash, *new_hash;
5050 LIST_HEAD(process_mods);
5051 char *func;
5052
5053 mutex_lock(&ops->func_hash->regex_lock);
5054
5055 if (enable)
5056 orig_hash = &ops->func_hash->filter_hash;
5057 else
5058 orig_hash = &ops->func_hash->notrace_hash;
5059
5060 new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
5061 *orig_hash);
5062 if (!new_hash)
5063 goto out; /* warn? */
5064
5065 mutex_lock(&ftrace_lock);
5066
5067 list_for_each_entry_safe(ftrace_mod, n, head, list) {
5068
5069 if (strcmp(ftrace_mod->module, mod) != 0)
5070 continue;
5071
5072 if (ftrace_mod->func)
5073 func = kstrdup(ftrace_mod->func, GFP_KERNEL);
5074 else
5075 func = kstrdup("*", GFP_KERNEL);
5076
5077 if (!func) /* warn? */
5078 continue;
5079
5080 list_move(&ftrace_mod->list, &process_mods);
5081
5082 /* Use the newly allocated func, as it may be "*" */
5083 kfree(ftrace_mod->func);
5084 ftrace_mod->func = func;
5085 }
5086
5087 mutex_unlock(&ftrace_lock);
5088
5089 list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
5090
5091 func = ftrace_mod->func;
5092
5093 /* Grabs ftrace_lock, which is why we have this extra step */
5094 match_records(new_hash, func, strlen(func), mod);
5095 free_ftrace_mod(ftrace_mod);
5096 }
5097
5098 if (enable && list_empty(head))
5099 new_hash->flags &= ~FTRACE_HASH_FL_MOD;
5100
5101 mutex_lock(&ftrace_lock);
5102
5103 ftrace_hash_move_and_update_ops(ops, orig_hash,
5104 new_hash, enable);
5105 mutex_unlock(&ftrace_lock);
5106
5107 out:
5108 mutex_unlock(&ops->func_hash->regex_lock);
5109
5110 free_ftrace_hash(new_hash);
5111 }
5112
process_cached_mods(const char * mod_name)5113 static void process_cached_mods(const char *mod_name)
5114 {
5115 struct trace_array *tr;
5116 char *mod;
5117
5118 mod = kstrdup(mod_name, GFP_KERNEL);
5119 if (!mod)
5120 return;
5121
5122 mutex_lock(&trace_types_lock);
5123 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5124 if (!list_empty(&tr->mod_trace))
5125 process_mod_list(&tr->mod_trace, tr->ops, mod, true);
5126 if (!list_empty(&tr->mod_notrace))
5127 process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
5128 }
5129 mutex_unlock(&trace_types_lock);
5130
5131 kfree(mod);
5132 }
5133 #endif
5134
5135 /*
5136 * We register the module command as a template to show others how
5137 * to register the a command as well.
5138 */
5139
5140 static int
ftrace_mod_callback(struct trace_array * tr,struct ftrace_hash * hash,char * func_orig,char * cmd,char * module,int enable)5141 ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
5142 char *func_orig, char *cmd, char *module, int enable)
5143 {
5144 char *func;
5145 int ret;
5146
5147 if (!tr)
5148 return -ENODEV;
5149
5150 /* match_records() modifies func, and we need the original */
5151 func = kstrdup(func_orig, GFP_KERNEL);
5152 if (!func)
5153 return -ENOMEM;
5154
5155 /*
5156 * cmd == 'mod' because we only registered this func
5157 * for the 'mod' ftrace_func_command.
5158 * But if you register one func with multiple commands,
5159 * you can tell which command was used by the cmd
5160 * parameter.
5161 */
5162 ret = match_records(hash, func, strlen(func), module);
5163 kfree(func);
5164
5165 if (!ret)
5166 return cache_mod(tr, func_orig, module, enable);
5167 if (ret < 0)
5168 return ret;
5169 return 0;
5170 }
5171
5172 static struct ftrace_func_command ftrace_mod_cmd = {
5173 .name = "mod",
5174 .func = ftrace_mod_callback,
5175 };
5176
ftrace_mod_cmd_init(void)5177 static int __init ftrace_mod_cmd_init(void)
5178 {
5179 return register_ftrace_command(&ftrace_mod_cmd);
5180 }
5181 core_initcall(ftrace_mod_cmd_init);
5182
function_trace_probe_call(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)5183 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
5184 struct ftrace_ops *op, struct ftrace_regs *fregs)
5185 {
5186 struct ftrace_probe_ops *probe_ops;
5187 struct ftrace_func_probe *probe;
5188
5189 probe = container_of(op, struct ftrace_func_probe, ops);
5190 probe_ops = probe->probe_ops;
5191
5192 /*
5193 * Disable preemption for these calls to prevent a RCU grace
5194 * period. This syncs the hash iteration and freeing of items
5195 * on the hash. rcu_read_lock is too dangerous here.
5196 */
5197 preempt_disable_notrace();
5198 probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
5199 preempt_enable_notrace();
5200 }
5201
5202 struct ftrace_func_map {
5203 struct ftrace_func_entry entry;
5204 void *data;
5205 };
5206
5207 /*
5208 * Note, ftrace_func_mapper is freed by free_ftrace_hash(&mapper->hash).
5209 * The hash field must be the first field.
5210 */
5211 struct ftrace_func_mapper {
5212 struct ftrace_hash hash; /* Must be first! */
5213 };
5214
5215 /**
5216 * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
5217 *
5218 * Returns: a ftrace_func_mapper descriptor that can be used to map ips to data.
5219 */
allocate_ftrace_func_mapper(void)5220 struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
5221 {
5222 struct ftrace_hash *hash;
5223
5224 /*
5225 * The mapper is simply a ftrace_hash, but since the entries
5226 * in the hash are not ftrace_func_entry type, we define it
5227 * as a separate structure.
5228 */
5229 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5230 return (struct ftrace_func_mapper *)hash;
5231 }
5232
5233 /**
5234 * ftrace_func_mapper_find_ip - Find some data mapped to an ip
5235 * @mapper: The mapper that has the ip maps
5236 * @ip: the instruction pointer to find the data for
5237 *
5238 * Returns: the data mapped to @ip if found otherwise NULL. The return
5239 * is actually the address of the mapper data pointer. The address is
5240 * returned for use cases where the data is no bigger than a long, and
5241 * the user can use the data pointer as its data instead of having to
5242 * allocate more memory for the reference.
5243 */
ftrace_func_mapper_find_ip(struct ftrace_func_mapper * mapper,unsigned long ip)5244 void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
5245 unsigned long ip)
5246 {
5247 struct ftrace_func_entry *entry;
5248 struct ftrace_func_map *map;
5249
5250 entry = ftrace_lookup_ip(&mapper->hash, ip);
5251 if (!entry)
5252 return NULL;
5253
5254 map = (struct ftrace_func_map *)entry;
5255 return &map->data;
5256 }
5257
5258 /**
5259 * ftrace_func_mapper_add_ip - Map some data to an ip
5260 * @mapper: The mapper that has the ip maps
5261 * @ip: The instruction pointer address to map @data to
5262 * @data: The data to map to @ip
5263 *
5264 * Returns: 0 on success otherwise an error.
5265 */
ftrace_func_mapper_add_ip(struct ftrace_func_mapper * mapper,unsigned long ip,void * data)5266 int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
5267 unsigned long ip, void *data)
5268 {
5269 struct ftrace_func_entry *entry;
5270 struct ftrace_func_map *map;
5271
5272 entry = ftrace_lookup_ip(&mapper->hash, ip);
5273 if (entry)
5274 return -EBUSY;
5275
5276 map = kmalloc(sizeof(*map), GFP_KERNEL);
5277 if (!map)
5278 return -ENOMEM;
5279
5280 map->entry.ip = ip;
5281 map->data = data;
5282
5283 __add_hash_entry(&mapper->hash, &map->entry);
5284
5285 return 0;
5286 }
5287
5288 /**
5289 * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
5290 * @mapper: The mapper that has the ip maps
5291 * @ip: The instruction pointer address to remove the data from
5292 *
5293 * Returns: the data if it is found, otherwise NULL.
5294 * Note, if the data pointer is used as the data itself, (see
5295 * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
5296 * if the data pointer was set to zero.
5297 */
ftrace_func_mapper_remove_ip(struct ftrace_func_mapper * mapper,unsigned long ip)5298 void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
5299 unsigned long ip)
5300 {
5301 struct ftrace_func_entry *entry;
5302 struct ftrace_func_map *map;
5303 void *data;
5304
5305 entry = ftrace_lookup_ip(&mapper->hash, ip);
5306 if (!entry)
5307 return NULL;
5308
5309 map = (struct ftrace_func_map *)entry;
5310 data = map->data;
5311
5312 remove_hash_entry(&mapper->hash, entry);
5313 kfree(entry);
5314
5315 return data;
5316 }
5317
5318 /**
5319 * free_ftrace_func_mapper - free a mapping of ips and data
5320 * @mapper: The mapper that has the ip maps
5321 * @free_func: A function to be called on each data item.
5322 *
5323 * This is used to free the function mapper. The @free_func is optional
5324 * and can be used if the data needs to be freed as well.
5325 */
free_ftrace_func_mapper(struct ftrace_func_mapper * mapper,ftrace_mapper_func free_func)5326 void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
5327 ftrace_mapper_func free_func)
5328 {
5329 struct ftrace_func_entry *entry;
5330 struct ftrace_func_map *map;
5331 struct hlist_head *hhd;
5332 int size, i;
5333
5334 if (!mapper)
5335 return;
5336
5337 if (free_func && mapper->hash.count) {
5338 size = 1 << mapper->hash.size_bits;
5339 for (i = 0; i < size; i++) {
5340 hhd = &mapper->hash.buckets[i];
5341 hlist_for_each_entry(entry, hhd, hlist) {
5342 map = (struct ftrace_func_map *)entry;
5343 free_func(map);
5344 }
5345 }
5346 }
5347 /* This also frees the mapper itself */
5348 free_ftrace_hash(&mapper->hash);
5349 }
5350
release_probe(struct ftrace_func_probe * probe)5351 static void release_probe(struct ftrace_func_probe *probe)
5352 {
5353 struct ftrace_probe_ops *probe_ops;
5354
5355 guard(mutex)(&ftrace_lock);
5356
5357 WARN_ON(probe->ref <= 0);
5358
5359 /* Subtract the ref that was used to protect this instance */
5360 probe->ref--;
5361
5362 if (!probe->ref) {
5363 probe_ops = probe->probe_ops;
5364 /*
5365 * Sending zero as ip tells probe_ops to free
5366 * the probe->data itself
5367 */
5368 if (probe_ops->free)
5369 probe_ops->free(probe_ops, probe->tr, 0, probe->data);
5370 list_del(&probe->list);
5371 kfree(probe);
5372 }
5373 }
5374
acquire_probe_locked(struct ftrace_func_probe * probe)5375 static void acquire_probe_locked(struct ftrace_func_probe *probe)
5376 {
5377 /*
5378 * Add one ref to keep it from being freed when releasing the
5379 * ftrace_lock mutex.
5380 */
5381 probe->ref++;
5382 }
5383
5384 int
register_ftrace_function_probe(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops,void * data)5385 register_ftrace_function_probe(char *glob, struct trace_array *tr,
5386 struct ftrace_probe_ops *probe_ops,
5387 void *data)
5388 {
5389 struct ftrace_func_probe *probe = NULL, *iter;
5390 struct ftrace_func_entry *entry;
5391 struct ftrace_hash **orig_hash;
5392 struct ftrace_hash *old_hash;
5393 struct ftrace_hash *hash;
5394 int count = 0;
5395 int size;
5396 int ret;
5397 int i;
5398
5399 if (WARN_ON(!tr))
5400 return -EINVAL;
5401
5402 /* We do not support '!' for function probes */
5403 if (WARN_ON(glob[0] == '!'))
5404 return -EINVAL;
5405
5406
5407 mutex_lock(&ftrace_lock);
5408 /* Check if the probe_ops is already registered */
5409 list_for_each_entry(iter, &tr->func_probes, list) {
5410 if (iter->probe_ops == probe_ops) {
5411 probe = iter;
5412 break;
5413 }
5414 }
5415 if (!probe) {
5416 probe = kzalloc(sizeof(*probe), GFP_KERNEL);
5417 if (!probe) {
5418 mutex_unlock(&ftrace_lock);
5419 return -ENOMEM;
5420 }
5421 probe->probe_ops = probe_ops;
5422 probe->ops.func = function_trace_probe_call;
5423 probe->tr = tr;
5424 ftrace_ops_init(&probe->ops);
5425 list_add(&probe->list, &tr->func_probes);
5426 }
5427
5428 acquire_probe_locked(probe);
5429
5430 mutex_unlock(&ftrace_lock);
5431
5432 /*
5433 * Note, there's a small window here that the func_hash->filter_hash
5434 * may be NULL or empty. Need to be careful when reading the loop.
5435 */
5436 mutex_lock(&probe->ops.func_hash->regex_lock);
5437
5438 orig_hash = &probe->ops.func_hash->filter_hash;
5439 old_hash = *orig_hash;
5440 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
5441
5442 if (!hash) {
5443 ret = -ENOMEM;
5444 goto out;
5445 }
5446
5447 ret = ftrace_match_records(hash, glob, strlen(glob));
5448
5449 /* Nothing found? */
5450 if (!ret)
5451 ret = -EINVAL;
5452
5453 if (ret < 0)
5454 goto out;
5455
5456 size = 1 << hash->size_bits;
5457 for (i = 0; i < size; i++) {
5458 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5459 if (ftrace_lookup_ip(old_hash, entry->ip))
5460 continue;
5461 /*
5462 * The caller might want to do something special
5463 * for each function we find. We call the callback
5464 * to give the caller an opportunity to do so.
5465 */
5466 if (probe_ops->init) {
5467 ret = probe_ops->init(probe_ops, tr,
5468 entry->ip, data,
5469 &probe->data);
5470 if (ret < 0) {
5471 if (probe_ops->free && count)
5472 probe_ops->free(probe_ops, tr,
5473 0, probe->data);
5474 probe->data = NULL;
5475 goto out;
5476 }
5477 }
5478 count++;
5479 }
5480 }
5481
5482 mutex_lock(&ftrace_lock);
5483
5484 if (!count) {
5485 /* Nothing was added? */
5486 ret = -EINVAL;
5487 goto out_unlock;
5488 }
5489
5490 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
5491 hash, 1);
5492 if (ret < 0)
5493 goto err_unlock;
5494
5495 /* One ref for each new function traced */
5496 probe->ref += count;
5497
5498 if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
5499 ret = ftrace_startup(&probe->ops, 0);
5500
5501 out_unlock:
5502 mutex_unlock(&ftrace_lock);
5503
5504 if (!ret)
5505 ret = count;
5506 out:
5507 mutex_unlock(&probe->ops.func_hash->regex_lock);
5508 free_ftrace_hash(hash);
5509
5510 release_probe(probe);
5511
5512 return ret;
5513
5514 err_unlock:
5515 if (!probe_ops->free || !count)
5516 goto out_unlock;
5517
5518 /* Failed to do the move, need to call the free functions */
5519 for (i = 0; i < size; i++) {
5520 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5521 if (ftrace_lookup_ip(old_hash, entry->ip))
5522 continue;
5523 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
5524 }
5525 }
5526 goto out_unlock;
5527 }
5528
5529 int
unregister_ftrace_function_probe_func(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops)5530 unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
5531 struct ftrace_probe_ops *probe_ops)
5532 {
5533 struct ftrace_func_probe *probe = NULL, *iter;
5534 struct ftrace_ops_hash old_hash_ops;
5535 struct ftrace_func_entry *entry;
5536 struct ftrace_glob func_g;
5537 struct ftrace_hash **orig_hash;
5538 struct ftrace_hash *old_hash;
5539 struct ftrace_hash *hash = NULL;
5540 struct hlist_node *tmp;
5541 struct hlist_head hhd;
5542 char str[KSYM_SYMBOL_LEN];
5543 int count = 0;
5544 int i, ret = -ENODEV;
5545 int size;
5546
5547 if (!glob || !strlen(glob) || !strcmp(glob, "*"))
5548 func_g.search = NULL;
5549 else {
5550 int not;
5551
5552 func_g.type = filter_parse_regex(glob, strlen(glob),
5553 &func_g.search, ¬);
5554 func_g.len = strlen(func_g.search);
5555
5556 /* we do not support '!' for function probes */
5557 if (WARN_ON(not))
5558 return -EINVAL;
5559 }
5560
5561 mutex_lock(&ftrace_lock);
5562 /* Check if the probe_ops is already registered */
5563 list_for_each_entry(iter, &tr->func_probes, list) {
5564 if (iter->probe_ops == probe_ops) {
5565 probe = iter;
5566 break;
5567 }
5568 }
5569 if (!probe)
5570 goto err_unlock_ftrace;
5571
5572 ret = -EINVAL;
5573 if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
5574 goto err_unlock_ftrace;
5575
5576 acquire_probe_locked(probe);
5577
5578 mutex_unlock(&ftrace_lock);
5579
5580 mutex_lock(&probe->ops.func_hash->regex_lock);
5581
5582 orig_hash = &probe->ops.func_hash->filter_hash;
5583 old_hash = *orig_hash;
5584
5585 if (ftrace_hash_empty(old_hash))
5586 goto out_unlock;
5587
5588 old_hash_ops.filter_hash = old_hash;
5589 /* Probes only have filters */
5590 old_hash_ops.notrace_hash = NULL;
5591
5592 ret = -ENOMEM;
5593 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
5594 if (!hash)
5595 goto out_unlock;
5596
5597 INIT_HLIST_HEAD(&hhd);
5598
5599 size = 1 << hash->size_bits;
5600 for (i = 0; i < size; i++) {
5601 hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
5602
5603 if (func_g.search) {
5604 kallsyms_lookup(entry->ip, NULL, NULL,
5605 NULL, str);
5606 if (!ftrace_match(str, &func_g))
5607 continue;
5608 }
5609 count++;
5610 remove_hash_entry(hash, entry);
5611 hlist_add_head(&entry->hlist, &hhd);
5612 }
5613 }
5614
5615 /* Nothing found? */
5616 if (!count) {
5617 ret = -EINVAL;
5618 goto out_unlock;
5619 }
5620
5621 mutex_lock(&ftrace_lock);
5622
5623 WARN_ON(probe->ref < count);
5624
5625 probe->ref -= count;
5626
5627 if (ftrace_hash_empty(hash))
5628 ftrace_shutdown(&probe->ops, 0);
5629
5630 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
5631 hash, 1);
5632
5633 /* still need to update the function call sites */
5634 if (ftrace_enabled && !ftrace_hash_empty(hash))
5635 ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
5636 &old_hash_ops);
5637 synchronize_rcu();
5638
5639 hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
5640 hlist_del(&entry->hlist);
5641 if (probe_ops->free)
5642 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
5643 kfree(entry);
5644 }
5645 mutex_unlock(&ftrace_lock);
5646
5647 out_unlock:
5648 mutex_unlock(&probe->ops.func_hash->regex_lock);
5649 free_ftrace_hash(hash);
5650
5651 release_probe(probe);
5652
5653 return ret;
5654
5655 err_unlock_ftrace:
5656 mutex_unlock(&ftrace_lock);
5657 return ret;
5658 }
5659
clear_ftrace_function_probes(struct trace_array * tr)5660 void clear_ftrace_function_probes(struct trace_array *tr)
5661 {
5662 struct ftrace_func_probe *probe, *n;
5663
5664 list_for_each_entry_safe(probe, n, &tr->func_probes, list)
5665 unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
5666 }
5667
5668 static LIST_HEAD(ftrace_commands);
5669 static DEFINE_MUTEX(ftrace_cmd_mutex);
5670
5671 /*
5672 * Currently we only register ftrace commands from __init, so mark this
5673 * __init too.
5674 */
register_ftrace_command(struct ftrace_func_command * cmd)5675 __init int register_ftrace_command(struct ftrace_func_command *cmd)
5676 {
5677 struct ftrace_func_command *p;
5678
5679 guard(mutex)(&ftrace_cmd_mutex);
5680 list_for_each_entry(p, &ftrace_commands, list) {
5681 if (strcmp(cmd->name, p->name) == 0)
5682 return -EBUSY;
5683 }
5684 list_add(&cmd->list, &ftrace_commands);
5685
5686 return 0;
5687 }
5688
5689 /*
5690 * Currently we only unregister ftrace commands from __init, so mark
5691 * this __init too.
5692 */
unregister_ftrace_command(struct ftrace_func_command * cmd)5693 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
5694 {
5695 struct ftrace_func_command *p, *n;
5696
5697 guard(mutex)(&ftrace_cmd_mutex);
5698
5699 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
5700 if (strcmp(cmd->name, p->name) == 0) {
5701 list_del_init(&p->list);
5702 return 0;
5703 }
5704 }
5705
5706 return -ENODEV;
5707 }
5708
ftrace_process_regex(struct ftrace_iterator * iter,char * buff,int len,int enable)5709 static int ftrace_process_regex(struct ftrace_iterator *iter,
5710 char *buff, int len, int enable)
5711 {
5712 struct ftrace_hash *hash = iter->hash;
5713 struct trace_array *tr = iter->ops->private;
5714 char *func, *command, *next = buff;
5715 struct ftrace_func_command *p;
5716 int ret;
5717
5718 func = strsep(&next, ":");
5719
5720 if (!next) {
5721 ret = ftrace_match_records(hash, func, len);
5722 if (!ret)
5723 ret = -EINVAL;
5724 if (ret < 0)
5725 return ret;
5726 return 0;
5727 }
5728
5729 /* command found */
5730
5731 command = strsep(&next, ":");
5732
5733 guard(mutex)(&ftrace_cmd_mutex);
5734
5735 list_for_each_entry(p, &ftrace_commands, list) {
5736 if (strcmp(p->name, command) == 0)
5737 return p->func(tr, hash, func, command, next, enable);
5738 }
5739
5740 return -EINVAL;
5741 }
5742
5743 static ssize_t
ftrace_regex_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos,int enable)5744 ftrace_regex_write(struct file *file, const char __user *ubuf,
5745 size_t cnt, loff_t *ppos, int enable)
5746 {
5747 struct ftrace_iterator *iter;
5748 struct trace_parser *parser;
5749 ssize_t ret, read;
5750
5751 if (!cnt)
5752 return 0;
5753
5754 if (file->f_mode & FMODE_READ) {
5755 struct seq_file *m = file->private_data;
5756 iter = m->private;
5757 } else
5758 iter = file->private_data;
5759
5760 if (unlikely(ftrace_disabled))
5761 return -ENODEV;
5762
5763 /* iter->hash is a local copy, so we don't need regex_lock */
5764
5765 parser = &iter->parser;
5766 read = trace_get_user(parser, ubuf, cnt, ppos);
5767
5768 if (read >= 0 && trace_parser_loaded(parser) &&
5769 !trace_parser_cont(parser)) {
5770 ret = ftrace_process_regex(iter, parser->buffer,
5771 parser->idx, enable);
5772 trace_parser_clear(parser);
5773 if (ret < 0)
5774 return ret;
5775 }
5776
5777 return read;
5778 }
5779
5780 ssize_t
ftrace_filter_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5781 ftrace_filter_write(struct file *file, const char __user *ubuf,
5782 size_t cnt, loff_t *ppos)
5783 {
5784 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
5785 }
5786
5787 ssize_t
ftrace_notrace_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5788 ftrace_notrace_write(struct file *file, const char __user *ubuf,
5789 size_t cnt, loff_t *ppos)
5790 {
5791 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
5792 }
5793
5794 static int
__ftrace_match_addr(struct ftrace_hash * hash,unsigned long ip,int remove)5795 __ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
5796 {
5797 struct ftrace_func_entry *entry;
5798
5799 ip = ftrace_location(ip);
5800 if (!ip)
5801 return -EINVAL;
5802
5803 if (remove) {
5804 entry = ftrace_lookup_ip(hash, ip);
5805 if (!entry)
5806 return -ENOENT;
5807 free_hash_entry(hash, entry);
5808 return 0;
5809 } else if (__ftrace_lookup_ip(hash, ip) != NULL) {
5810 /* Already exists */
5811 return 0;
5812 }
5813
5814 entry = add_hash_entry(hash, ip);
5815 return entry ? 0 : -ENOMEM;
5816 }
5817
5818 static int
ftrace_match_addr(struct ftrace_hash * hash,unsigned long * ips,unsigned int cnt,int remove)5819 ftrace_match_addr(struct ftrace_hash *hash, unsigned long *ips,
5820 unsigned int cnt, int remove)
5821 {
5822 unsigned int i;
5823 int err;
5824
5825 for (i = 0; i < cnt; i++) {
5826 err = __ftrace_match_addr(hash, ips[i], remove);
5827 if (err) {
5828 /*
5829 * This expects the @hash is a temporary hash and if this
5830 * fails the caller must free the @hash.
5831 */
5832 return err;
5833 }
5834 }
5835 return 0;
5836 }
5837
5838 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)5839 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
5840 unsigned long *ips, unsigned int cnt,
5841 int remove, int reset, int enable, char *mod)
5842 {
5843 struct ftrace_hash **orig_hash;
5844 struct ftrace_hash *hash;
5845 int ret;
5846
5847 if (unlikely(ftrace_disabled))
5848 return -ENODEV;
5849
5850 mutex_lock(&ops->func_hash->regex_lock);
5851
5852 if (enable)
5853 orig_hash = &ops->func_hash->filter_hash;
5854 else
5855 orig_hash = &ops->func_hash->notrace_hash;
5856
5857 if (reset)
5858 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5859 else
5860 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
5861
5862 if (!hash) {
5863 ret = -ENOMEM;
5864 goto out_regex_unlock;
5865 }
5866
5867 if (buf && !match_records(hash, buf, len, mod)) {
5868 /* If this was for a module and nothing was enabled, flag it */
5869 if (mod)
5870 (*orig_hash)->flags |= FTRACE_HASH_FL_MOD;
5871
5872 /*
5873 * Even if it is a mod, return error to let caller know
5874 * nothing was added
5875 */
5876 ret = -EINVAL;
5877 goto out_regex_unlock;
5878 }
5879 if (ips) {
5880 ret = ftrace_match_addr(hash, ips, cnt, remove);
5881 if (ret < 0)
5882 goto out_regex_unlock;
5883 }
5884
5885 mutex_lock(&ftrace_lock);
5886 ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
5887 mutex_unlock(&ftrace_lock);
5888
5889 out_regex_unlock:
5890 mutex_unlock(&ops->func_hash->regex_lock);
5891
5892 free_ftrace_hash(hash);
5893 return ret;
5894 }
5895
5896 static int
ftrace_set_addr(struct ftrace_ops * ops,unsigned long * ips,unsigned int cnt,int remove,int reset,int enable)5897 ftrace_set_addr(struct ftrace_ops *ops, unsigned long *ips, unsigned int cnt,
5898 int remove, int reset, int enable)
5899 {
5900 return ftrace_set_hash(ops, NULL, 0, ips, cnt, remove, reset, enable, NULL);
5901 }
5902
5903 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
5904
5905 static int register_ftrace_function_nolock(struct ftrace_ops *ops);
5906
5907 /*
5908 * If there are multiple ftrace_ops, use SAVE_REGS by default, so that direct
5909 * call will be jumped from ftrace_regs_caller. Only if the architecture does
5910 * not support ftrace_regs_caller but direct_call, use SAVE_ARGS so that it
5911 * jumps from ftrace_caller for multiple ftrace_ops.
5912 */
5913 #ifndef CONFIG_HAVE_DYNAMIC_FTRACE_WITH_REGS
5914 #define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_ARGS)
5915 #else
5916 #define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_REGS)
5917 #endif
5918
check_direct_multi(struct ftrace_ops * ops)5919 static int check_direct_multi(struct ftrace_ops *ops)
5920 {
5921 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED))
5922 return -EINVAL;
5923 if ((ops->flags & MULTI_FLAGS) != MULTI_FLAGS)
5924 return -EINVAL;
5925 return 0;
5926 }
5927
remove_direct_functions_hash(struct ftrace_hash * hash,unsigned long addr)5928 static void remove_direct_functions_hash(struct ftrace_hash *hash, unsigned long addr)
5929 {
5930 struct ftrace_func_entry *entry, *del;
5931 int size, i;
5932
5933 size = 1 << hash->size_bits;
5934 for (i = 0; i < size; i++) {
5935 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5936 del = __ftrace_lookup_ip(direct_functions, entry->ip);
5937 if (del && del->direct == addr) {
5938 remove_hash_entry(direct_functions, del);
5939 kfree(del);
5940 }
5941 }
5942 }
5943 }
5944
register_ftrace_direct_cb(struct rcu_head * rhp)5945 static void register_ftrace_direct_cb(struct rcu_head *rhp)
5946 {
5947 struct ftrace_hash *fhp = container_of(rhp, struct ftrace_hash, rcu);
5948
5949 free_ftrace_hash(fhp);
5950 }
5951
5952 /**
5953 * register_ftrace_direct - Call a custom trampoline directly
5954 * for multiple functions registered in @ops
5955 * @ops: The address of the struct ftrace_ops object
5956 * @addr: The address of the trampoline to call at @ops functions
5957 *
5958 * This is used to connect a direct calls to @addr from the nop locations
5959 * of the functions registered in @ops (with by ftrace_set_filter_ip
5960 * function).
5961 *
5962 * The location that it calls (@addr) must be able to handle a direct call,
5963 * and save the parameters of the function being traced, and restore them
5964 * (or inject new ones if needed), before returning.
5965 *
5966 * Returns:
5967 * 0 on success
5968 * -EINVAL - The @ops object was already registered with this call or
5969 * when there are no functions in @ops object.
5970 * -EBUSY - Another direct function is already attached (there can be only one)
5971 * -ENODEV - @ip does not point to a ftrace nop location (or not supported)
5972 * -ENOMEM - There was an allocation failure.
5973 */
register_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)5974 int register_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
5975 {
5976 struct ftrace_hash *hash, *new_hash = NULL, *free_hash = NULL;
5977 struct ftrace_func_entry *entry, *new;
5978 int err = -EBUSY, size, i;
5979
5980 if (ops->func || ops->trampoline)
5981 return -EINVAL;
5982 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED))
5983 return -EINVAL;
5984 if (ops->flags & FTRACE_OPS_FL_ENABLED)
5985 return -EINVAL;
5986
5987 hash = ops->func_hash->filter_hash;
5988 if (ftrace_hash_empty(hash))
5989 return -EINVAL;
5990
5991 mutex_lock(&direct_mutex);
5992
5993 /* Make sure requested entries are not already registered.. */
5994 size = 1 << hash->size_bits;
5995 for (i = 0; i < size; i++) {
5996 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5997 if (ftrace_find_rec_direct(entry->ip))
5998 goto out_unlock;
5999 }
6000 }
6001
6002 err = -ENOMEM;
6003
6004 /* Make a copy hash to place the new and the old entries in */
6005 size = hash->count + direct_functions->count;
6006 size = fls(size);
6007 if (size > FTRACE_HASH_MAX_BITS)
6008 size = FTRACE_HASH_MAX_BITS;
6009 new_hash = alloc_ftrace_hash(size);
6010 if (!new_hash)
6011 goto out_unlock;
6012
6013 /* Now copy over the existing direct entries */
6014 size = 1 << direct_functions->size_bits;
6015 for (i = 0; i < size; i++) {
6016 hlist_for_each_entry(entry, &direct_functions->buckets[i], hlist) {
6017 new = add_hash_entry(new_hash, entry->ip);
6018 if (!new)
6019 goto out_unlock;
6020 new->direct = entry->direct;
6021 }
6022 }
6023
6024 /* ... and add the new entries */
6025 size = 1 << hash->size_bits;
6026 for (i = 0; i < size; i++) {
6027 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
6028 new = add_hash_entry(new_hash, entry->ip);
6029 if (!new)
6030 goto out_unlock;
6031 /* Update both the copy and the hash entry */
6032 new->direct = addr;
6033 entry->direct = addr;
6034 }
6035 }
6036
6037 free_hash = direct_functions;
6038 rcu_assign_pointer(direct_functions, new_hash);
6039 new_hash = NULL;
6040
6041 ops->func = call_direct_funcs;
6042 ops->flags = MULTI_FLAGS;
6043 ops->trampoline = FTRACE_REGS_ADDR;
6044 ops->direct_call = addr;
6045
6046 err = register_ftrace_function_nolock(ops);
6047
6048 out_unlock:
6049 mutex_unlock(&direct_mutex);
6050
6051 if (free_hash && free_hash != EMPTY_HASH)
6052 call_rcu_tasks(&free_hash->rcu, register_ftrace_direct_cb);
6053
6054 if (new_hash)
6055 free_ftrace_hash(new_hash);
6056
6057 return err;
6058 }
6059 EXPORT_SYMBOL_GPL(register_ftrace_direct);
6060
6061 /**
6062 * unregister_ftrace_direct - Remove calls to custom trampoline
6063 * previously registered by register_ftrace_direct for @ops object.
6064 * @ops: The address of the struct ftrace_ops object
6065 * @addr: The address of the direct function that is called by the @ops functions
6066 * @free_filters: Set to true to remove all filters for the ftrace_ops, false otherwise
6067 *
6068 * This is used to remove a direct calls to @addr from the nop locations
6069 * of the functions registered in @ops (with by ftrace_set_filter_ip
6070 * function).
6071 *
6072 * Returns:
6073 * 0 on success
6074 * -EINVAL - The @ops object was not properly registered.
6075 */
unregister_ftrace_direct(struct ftrace_ops * ops,unsigned long addr,bool free_filters)6076 int unregister_ftrace_direct(struct ftrace_ops *ops, unsigned long addr,
6077 bool free_filters)
6078 {
6079 struct ftrace_hash *hash = ops->func_hash->filter_hash;
6080 int err;
6081
6082 if (check_direct_multi(ops))
6083 return -EINVAL;
6084 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6085 return -EINVAL;
6086
6087 mutex_lock(&direct_mutex);
6088 err = unregister_ftrace_function(ops);
6089 remove_direct_functions_hash(hash, addr);
6090 mutex_unlock(&direct_mutex);
6091
6092 /* cleanup for possible another register call */
6093 ops->func = NULL;
6094 ops->trampoline = 0;
6095
6096 if (free_filters)
6097 ftrace_free_filter(ops);
6098 return err;
6099 }
6100 EXPORT_SYMBOL_GPL(unregister_ftrace_direct);
6101
6102 static int
__modify_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)6103 __modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
6104 {
6105 struct ftrace_hash *hash;
6106 struct ftrace_func_entry *entry, *iter;
6107 static struct ftrace_ops tmp_ops = {
6108 .func = ftrace_stub,
6109 .flags = FTRACE_OPS_FL_STUB,
6110 };
6111 int i, size;
6112 int err;
6113
6114 lockdep_assert_held_once(&direct_mutex);
6115
6116 /* Enable the tmp_ops to have the same functions as the direct ops */
6117 ftrace_ops_init(&tmp_ops);
6118 tmp_ops.func_hash = ops->func_hash;
6119 tmp_ops.direct_call = addr;
6120
6121 err = register_ftrace_function_nolock(&tmp_ops);
6122 if (err)
6123 return err;
6124
6125 /*
6126 * Now the ftrace_ops_list_func() is called to do the direct callers.
6127 * We can safely change the direct functions attached to each entry.
6128 */
6129 mutex_lock(&ftrace_lock);
6130
6131 hash = ops->func_hash->filter_hash;
6132 size = 1 << hash->size_bits;
6133 for (i = 0; i < size; i++) {
6134 hlist_for_each_entry(iter, &hash->buckets[i], hlist) {
6135 entry = __ftrace_lookup_ip(direct_functions, iter->ip);
6136 if (!entry)
6137 continue;
6138 entry->direct = addr;
6139 }
6140 }
6141 /* Prevent store tearing if a trampoline concurrently accesses the value */
6142 WRITE_ONCE(ops->direct_call, addr);
6143
6144 mutex_unlock(&ftrace_lock);
6145
6146 /* Removing the tmp_ops will add the updated direct callers to the functions */
6147 unregister_ftrace_function(&tmp_ops);
6148
6149 return err;
6150 }
6151
6152 /**
6153 * modify_ftrace_direct_nolock - Modify an existing direct 'multi' call
6154 * to call something else
6155 * @ops: The address of the struct ftrace_ops object
6156 * @addr: The address of the new trampoline to call at @ops functions
6157 *
6158 * This is used to unregister currently registered direct caller and
6159 * register new one @addr on functions registered in @ops object.
6160 *
6161 * Note there's window between ftrace_shutdown and ftrace_startup calls
6162 * where there will be no callbacks called.
6163 *
6164 * Caller should already have direct_mutex locked, so we don't lock
6165 * direct_mutex here.
6166 *
6167 * Returns: zero on success. Non zero on error, which includes:
6168 * -EINVAL - The @ops object was not properly registered.
6169 */
modify_ftrace_direct_nolock(struct ftrace_ops * ops,unsigned long addr)6170 int modify_ftrace_direct_nolock(struct ftrace_ops *ops, unsigned long addr)
6171 {
6172 if (check_direct_multi(ops))
6173 return -EINVAL;
6174 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6175 return -EINVAL;
6176
6177 return __modify_ftrace_direct(ops, addr);
6178 }
6179 EXPORT_SYMBOL_GPL(modify_ftrace_direct_nolock);
6180
6181 /**
6182 * modify_ftrace_direct - Modify an existing direct 'multi' call
6183 * to call something else
6184 * @ops: The address of the struct ftrace_ops object
6185 * @addr: The address of the new trampoline to call at @ops functions
6186 *
6187 * This is used to unregister currently registered direct caller and
6188 * register new one @addr on functions registered in @ops object.
6189 *
6190 * Note there's window between ftrace_shutdown and ftrace_startup calls
6191 * where there will be no callbacks called.
6192 *
6193 * Returns: zero on success. Non zero on error, which includes:
6194 * -EINVAL - The @ops object was not properly registered.
6195 */
modify_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)6196 int modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
6197 {
6198 int err;
6199
6200 if (check_direct_multi(ops))
6201 return -EINVAL;
6202 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6203 return -EINVAL;
6204
6205 mutex_lock(&direct_mutex);
6206 err = __modify_ftrace_direct(ops, addr);
6207 mutex_unlock(&direct_mutex);
6208 return err;
6209 }
6210 EXPORT_SYMBOL_GPL(modify_ftrace_direct);
6211 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
6212
6213 /**
6214 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
6215 * @ops: the ops to set the filter with
6216 * @ip: the address to add to or remove from the filter.
6217 * @remove: non zero to remove the ip from the filter
6218 * @reset: non zero to reset all filters before applying this filter.
6219 *
6220 * Filters denote which functions should be enabled when tracing is enabled
6221 * If @ip is NULL, it fails to update filter.
6222 *
6223 * This can allocate memory which must be freed before @ops can be freed,
6224 * either by removing each filtered addr or by using
6225 * ftrace_free_filter(@ops).
6226 */
ftrace_set_filter_ip(struct ftrace_ops * ops,unsigned long ip,int remove,int reset)6227 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
6228 int remove, int reset)
6229 {
6230 ftrace_ops_init(ops);
6231 return ftrace_set_addr(ops, &ip, 1, remove, reset, 1);
6232 }
6233 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
6234
6235 /**
6236 * ftrace_set_filter_ips - set functions to filter on in ftrace by addresses
6237 * @ops: the ops to set the filter with
6238 * @ips: the array of addresses to add to or remove from the filter.
6239 * @cnt: the number of addresses in @ips
6240 * @remove: non zero to remove ips from the filter
6241 * @reset: non zero to reset all filters before applying this filter.
6242 *
6243 * Filters denote which functions should be enabled when tracing is enabled
6244 * If @ips array or any ip specified within is NULL , it fails to update filter.
6245 *
6246 * This can allocate memory which must be freed before @ops can be freed,
6247 * either by removing each filtered addr or by using
6248 * ftrace_free_filter(@ops).
6249 */
ftrace_set_filter_ips(struct ftrace_ops * ops,unsigned long * ips,unsigned int cnt,int remove,int reset)6250 int ftrace_set_filter_ips(struct ftrace_ops *ops, unsigned long *ips,
6251 unsigned int cnt, int remove, int reset)
6252 {
6253 ftrace_ops_init(ops);
6254 return ftrace_set_addr(ops, ips, cnt, remove, reset, 1);
6255 }
6256 EXPORT_SYMBOL_GPL(ftrace_set_filter_ips);
6257
6258 /**
6259 * ftrace_ops_set_global_filter - setup ops to use global filters
6260 * @ops: the ops which will use the global filters
6261 *
6262 * ftrace users who need global function trace filtering should call this.
6263 * It can set the global filter only if ops were not initialized before.
6264 */
ftrace_ops_set_global_filter(struct ftrace_ops * ops)6265 void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
6266 {
6267 if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
6268 return;
6269
6270 ftrace_ops_init(ops);
6271 ops->func_hash = &global_ops.local_hash;
6272 }
6273 EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
6274
6275 static int
ftrace_set_regex(struct ftrace_ops * ops,unsigned char * buf,int len,int reset,int enable)6276 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
6277 int reset, int enable)
6278 {
6279 char *mod = NULL, *func, *command, *next = buf;
6280 char *tmp __free(kfree) = NULL;
6281 struct trace_array *tr = ops->private;
6282 int ret;
6283
6284 func = strsep(&next, ":");
6285
6286 /* This can also handle :mod: parsing */
6287 if (next) {
6288 if (!tr)
6289 return -EINVAL;
6290
6291 command = strsep(&next, ":");
6292 if (strcmp(command, "mod") != 0)
6293 return -EINVAL;
6294
6295 mod = next;
6296 len = command - func;
6297 /* Save the original func as ftrace_set_hash() can modify it */
6298 tmp = kstrdup(func, GFP_KERNEL);
6299 }
6300
6301 ret = ftrace_set_hash(ops, func, len, NULL, 0, 0, reset, enable, mod);
6302
6303 if (tr && mod && ret < 0) {
6304 /* Did tmp fail to allocate? */
6305 if (!tmp)
6306 return -ENOMEM;
6307 ret = cache_mod(tr, tmp, mod, enable);
6308 }
6309
6310 return ret;
6311 }
6312
6313 /**
6314 * ftrace_set_filter - set a function to filter on in ftrace
6315 * @ops: the ops to set the filter with
6316 * @buf: the string that holds the function filter text.
6317 * @len: the length of the string.
6318 * @reset: non-zero to reset all filters before applying this filter.
6319 *
6320 * Filters denote which functions should be enabled when tracing is enabled.
6321 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
6322 *
6323 * This can allocate memory which must be freed before @ops can be freed,
6324 * either by removing each filtered addr or by using
6325 * ftrace_free_filter(@ops).
6326 */
ftrace_set_filter(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)6327 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
6328 int len, int reset)
6329 {
6330 ftrace_ops_init(ops);
6331 return ftrace_set_regex(ops, buf, len, reset, 1);
6332 }
6333 EXPORT_SYMBOL_GPL(ftrace_set_filter);
6334
6335 /**
6336 * ftrace_set_notrace - set a function to not trace in ftrace
6337 * @ops: the ops to set the notrace filter with
6338 * @buf: the string that holds the function notrace text.
6339 * @len: the length of the string.
6340 * @reset: non-zero to reset all filters before applying this filter.
6341 *
6342 * Notrace Filters denote which functions should not be enabled when tracing
6343 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
6344 * for tracing.
6345 *
6346 * This can allocate memory which must be freed before @ops can be freed,
6347 * either by removing each filtered addr or by using
6348 * ftrace_free_filter(@ops).
6349 */
ftrace_set_notrace(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)6350 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
6351 int len, int reset)
6352 {
6353 ftrace_ops_init(ops);
6354 return ftrace_set_regex(ops, buf, len, reset, 0);
6355 }
6356 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
6357 /**
6358 * ftrace_set_global_filter - set a function to filter on with global tracers
6359 * @buf: the string that holds the function filter text.
6360 * @len: the length of the string.
6361 * @reset: non-zero to reset all filters before applying this filter.
6362 *
6363 * Filters denote which functions should be enabled when tracing is enabled.
6364 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
6365 */
ftrace_set_global_filter(unsigned char * buf,int len,int reset)6366 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
6367 {
6368 ftrace_set_regex(&global_ops, buf, len, reset, 1);
6369 }
6370 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
6371
6372 /**
6373 * ftrace_set_global_notrace - set a function to not trace with global tracers
6374 * @buf: the string that holds the function notrace text.
6375 * @len: the length of the string.
6376 * @reset: non-zero to reset all filters before applying this filter.
6377 *
6378 * Notrace Filters denote which functions should not be enabled when tracing
6379 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
6380 * for tracing.
6381 */
ftrace_set_global_notrace(unsigned char * buf,int len,int reset)6382 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
6383 {
6384 ftrace_set_regex(&global_ops, buf, len, reset, 0);
6385 }
6386 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
6387
6388 /*
6389 * command line interface to allow users to set filters on boot up.
6390 */
6391 #define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
6392 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
6393 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
6394
6395 /* Used by function selftest to not test if filter is set */
6396 bool ftrace_filter_param __initdata;
6397
set_ftrace_notrace(char * str)6398 static int __init set_ftrace_notrace(char *str)
6399 {
6400 ftrace_filter_param = true;
6401 strscpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
6402 return 1;
6403 }
6404 __setup("ftrace_notrace=", set_ftrace_notrace);
6405
set_ftrace_filter(char * str)6406 static int __init set_ftrace_filter(char *str)
6407 {
6408 ftrace_filter_param = true;
6409 strscpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
6410 return 1;
6411 }
6412 __setup("ftrace_filter=", set_ftrace_filter);
6413
6414 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6415 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
6416 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
6417 static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
6418
set_graph_function(char * str)6419 static int __init set_graph_function(char *str)
6420 {
6421 strscpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
6422 return 1;
6423 }
6424 __setup("ftrace_graph_filter=", set_graph_function);
6425
set_graph_notrace_function(char * str)6426 static int __init set_graph_notrace_function(char *str)
6427 {
6428 strscpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
6429 return 1;
6430 }
6431 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
6432
set_graph_max_depth_function(char * str)6433 static int __init set_graph_max_depth_function(char *str)
6434 {
6435 if (!str || kstrtouint(str, 0, &fgraph_max_depth))
6436 return 0;
6437 return 1;
6438 }
6439 __setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
6440
set_ftrace_early_graph(char * buf,int enable)6441 static void __init set_ftrace_early_graph(char *buf, int enable)
6442 {
6443 int ret;
6444 char *func;
6445 struct ftrace_hash *hash;
6446
6447 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
6448 if (MEM_FAIL(!hash, "Failed to allocate hash\n"))
6449 return;
6450
6451 while (buf) {
6452 func = strsep(&buf, ",");
6453 /* we allow only one expression at a time */
6454 ret = ftrace_graph_set_hash(hash, func);
6455 if (ret)
6456 printk(KERN_DEBUG "ftrace: function %s not "
6457 "traceable\n", func);
6458 }
6459
6460 if (enable)
6461 ftrace_graph_hash = hash;
6462 else
6463 ftrace_graph_notrace_hash = hash;
6464 }
6465 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6466
6467 void __init
ftrace_set_early_filter(struct ftrace_ops * ops,char * buf,int enable)6468 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
6469 {
6470 char *func;
6471
6472 ftrace_ops_init(ops);
6473
6474 /* The trace_array is needed for caching module function filters */
6475 if (!ops->private) {
6476 struct trace_array *tr = trace_get_global_array();
6477
6478 ops->private = tr;
6479 ftrace_init_trace_array(tr);
6480 }
6481
6482 while (buf) {
6483 func = strsep(&buf, ",");
6484 ftrace_set_regex(ops, func, strlen(func), 0, enable);
6485 }
6486 }
6487
set_ftrace_early_filters(void)6488 static void __init set_ftrace_early_filters(void)
6489 {
6490 if (ftrace_filter_buf[0])
6491 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
6492 if (ftrace_notrace_buf[0])
6493 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
6494 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6495 if (ftrace_graph_buf[0])
6496 set_ftrace_early_graph(ftrace_graph_buf, 1);
6497 if (ftrace_graph_notrace_buf[0])
6498 set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
6499 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6500 }
6501
ftrace_regex_release(struct inode * inode,struct file * file)6502 int ftrace_regex_release(struct inode *inode, struct file *file)
6503 {
6504 struct seq_file *m = (struct seq_file *)file->private_data;
6505 struct ftrace_iterator *iter;
6506 struct ftrace_hash **orig_hash;
6507 struct trace_parser *parser;
6508 int filter_hash;
6509
6510 if (file->f_mode & FMODE_READ) {
6511 iter = m->private;
6512 seq_release(inode, file);
6513 } else
6514 iter = file->private_data;
6515
6516 parser = &iter->parser;
6517 if (trace_parser_loaded(parser)) {
6518 int enable = !(iter->flags & FTRACE_ITER_NOTRACE);
6519
6520 ftrace_process_regex(iter, parser->buffer,
6521 parser->idx, enable);
6522 }
6523
6524 trace_parser_put(parser);
6525
6526 mutex_lock(&iter->ops->func_hash->regex_lock);
6527
6528 if (file->f_mode & FMODE_WRITE) {
6529 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
6530
6531 if (filter_hash) {
6532 orig_hash = &iter->ops->func_hash->filter_hash;
6533 if (iter->tr) {
6534 if (list_empty(&iter->tr->mod_trace))
6535 iter->hash->flags &= ~FTRACE_HASH_FL_MOD;
6536 else
6537 iter->hash->flags |= FTRACE_HASH_FL_MOD;
6538 }
6539 } else
6540 orig_hash = &iter->ops->func_hash->notrace_hash;
6541
6542 mutex_lock(&ftrace_lock);
6543 ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
6544 iter->hash, filter_hash);
6545 mutex_unlock(&ftrace_lock);
6546 } else {
6547 /* For read only, the hash is the ops hash */
6548 iter->hash = NULL;
6549 }
6550
6551 mutex_unlock(&iter->ops->func_hash->regex_lock);
6552 free_ftrace_hash(iter->hash);
6553 if (iter->tr)
6554 trace_array_put(iter->tr);
6555 kfree(iter);
6556
6557 return 0;
6558 }
6559
6560 static const struct file_operations ftrace_avail_fops = {
6561 .open = ftrace_avail_open,
6562 .read = seq_read,
6563 .llseek = seq_lseek,
6564 .release = seq_release_private,
6565 };
6566
6567 static const struct file_operations ftrace_enabled_fops = {
6568 .open = ftrace_enabled_open,
6569 .read = seq_read,
6570 .llseek = seq_lseek,
6571 .release = seq_release_private,
6572 };
6573
6574 static const struct file_operations ftrace_touched_fops = {
6575 .open = ftrace_touched_open,
6576 .read = seq_read,
6577 .llseek = seq_lseek,
6578 .release = seq_release_private,
6579 };
6580
6581 static const struct file_operations ftrace_avail_addrs_fops = {
6582 .open = ftrace_avail_addrs_open,
6583 .read = seq_read,
6584 .llseek = seq_lseek,
6585 .release = seq_release_private,
6586 };
6587
6588 static const struct file_operations ftrace_filter_fops = {
6589 .open = ftrace_filter_open,
6590 .read = seq_read,
6591 .write = ftrace_filter_write,
6592 .llseek = tracing_lseek,
6593 .release = ftrace_regex_release,
6594 };
6595
6596 static const struct file_operations ftrace_notrace_fops = {
6597 .open = ftrace_notrace_open,
6598 .read = seq_read,
6599 .write = ftrace_notrace_write,
6600 .llseek = tracing_lseek,
6601 .release = ftrace_regex_release,
6602 };
6603
6604 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6605
6606 static DEFINE_MUTEX(graph_lock);
6607
6608 struct ftrace_hash __rcu *ftrace_graph_hash = EMPTY_HASH;
6609 struct ftrace_hash __rcu *ftrace_graph_notrace_hash = EMPTY_HASH;
6610
6611 enum graph_filter_type {
6612 GRAPH_FILTER_NOTRACE = 0,
6613 GRAPH_FILTER_FUNCTION,
6614 };
6615
6616 #define FTRACE_GRAPH_EMPTY ((void *)1)
6617
6618 struct ftrace_graph_data {
6619 struct ftrace_hash *hash;
6620 struct ftrace_func_entry *entry;
6621 int idx; /* for hash table iteration */
6622 enum graph_filter_type type;
6623 struct ftrace_hash *new_hash;
6624 const struct seq_operations *seq_ops;
6625 struct trace_parser parser;
6626 };
6627
6628 static void *
__g_next(struct seq_file * m,loff_t * pos)6629 __g_next(struct seq_file *m, loff_t *pos)
6630 {
6631 struct ftrace_graph_data *fgd = m->private;
6632 struct ftrace_func_entry *entry = fgd->entry;
6633 struct hlist_head *head;
6634 int i, idx = fgd->idx;
6635
6636 if (*pos >= fgd->hash->count)
6637 return NULL;
6638
6639 if (entry) {
6640 hlist_for_each_entry_continue(entry, hlist) {
6641 fgd->entry = entry;
6642 return entry;
6643 }
6644
6645 idx++;
6646 }
6647
6648 for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
6649 head = &fgd->hash->buckets[i];
6650 hlist_for_each_entry(entry, head, hlist) {
6651 fgd->entry = entry;
6652 fgd->idx = i;
6653 return entry;
6654 }
6655 }
6656 return NULL;
6657 }
6658
6659 static void *
g_next(struct seq_file * m,void * v,loff_t * pos)6660 g_next(struct seq_file *m, void *v, loff_t *pos)
6661 {
6662 (*pos)++;
6663 return __g_next(m, pos);
6664 }
6665
g_start(struct seq_file * m,loff_t * pos)6666 static void *g_start(struct seq_file *m, loff_t *pos)
6667 {
6668 struct ftrace_graph_data *fgd = m->private;
6669
6670 mutex_lock(&graph_lock);
6671
6672 if (fgd->type == GRAPH_FILTER_FUNCTION)
6673 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
6674 lockdep_is_held(&graph_lock));
6675 else
6676 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6677 lockdep_is_held(&graph_lock));
6678
6679 /* Nothing, tell g_show to print all functions are enabled */
6680 if (ftrace_hash_empty(fgd->hash) && !*pos)
6681 return FTRACE_GRAPH_EMPTY;
6682
6683 fgd->idx = 0;
6684 fgd->entry = NULL;
6685 return __g_next(m, pos);
6686 }
6687
g_stop(struct seq_file * m,void * p)6688 static void g_stop(struct seq_file *m, void *p)
6689 {
6690 mutex_unlock(&graph_lock);
6691 }
6692
g_show(struct seq_file * m,void * v)6693 static int g_show(struct seq_file *m, void *v)
6694 {
6695 struct ftrace_func_entry *entry = v;
6696
6697 if (!entry)
6698 return 0;
6699
6700 if (entry == FTRACE_GRAPH_EMPTY) {
6701 struct ftrace_graph_data *fgd = m->private;
6702
6703 if (fgd->type == GRAPH_FILTER_FUNCTION)
6704 seq_puts(m, "#### all functions enabled ####\n");
6705 else
6706 seq_puts(m, "#### no functions disabled ####\n");
6707 return 0;
6708 }
6709
6710 seq_printf(m, "%ps\n", (void *)entry->ip);
6711
6712 return 0;
6713 }
6714
6715 static const struct seq_operations ftrace_graph_seq_ops = {
6716 .start = g_start,
6717 .next = g_next,
6718 .stop = g_stop,
6719 .show = g_show,
6720 };
6721
6722 static int
__ftrace_graph_open(struct inode * inode,struct file * file,struct ftrace_graph_data * fgd)6723 __ftrace_graph_open(struct inode *inode, struct file *file,
6724 struct ftrace_graph_data *fgd)
6725 {
6726 int ret;
6727 struct ftrace_hash *new_hash = NULL;
6728
6729 ret = security_locked_down(LOCKDOWN_TRACEFS);
6730 if (ret)
6731 return ret;
6732
6733 if (file->f_mode & FMODE_WRITE) {
6734 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
6735
6736 if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
6737 return -ENOMEM;
6738
6739 if (file->f_flags & O_TRUNC)
6740 new_hash = alloc_ftrace_hash(size_bits);
6741 else
6742 new_hash = alloc_and_copy_ftrace_hash(size_bits,
6743 fgd->hash);
6744 if (!new_hash) {
6745 ret = -ENOMEM;
6746 goto out;
6747 }
6748 }
6749
6750 if (file->f_mode & FMODE_READ) {
6751 ret = seq_open(file, &ftrace_graph_seq_ops);
6752 if (!ret) {
6753 struct seq_file *m = file->private_data;
6754 m->private = fgd;
6755 } else {
6756 /* Failed */
6757 free_ftrace_hash(new_hash);
6758 new_hash = NULL;
6759 }
6760 } else
6761 file->private_data = fgd;
6762
6763 out:
6764 if (ret < 0 && file->f_mode & FMODE_WRITE)
6765 trace_parser_put(&fgd->parser);
6766
6767 fgd->new_hash = new_hash;
6768
6769 /*
6770 * All uses of fgd->hash must be taken with the graph_lock
6771 * held. The graph_lock is going to be released, so force
6772 * fgd->hash to be reinitialized when it is taken again.
6773 */
6774 fgd->hash = NULL;
6775
6776 return ret;
6777 }
6778
6779 static int
ftrace_graph_open(struct inode * inode,struct file * file)6780 ftrace_graph_open(struct inode *inode, struct file *file)
6781 {
6782 struct ftrace_graph_data *fgd;
6783 int ret;
6784
6785 if (unlikely(ftrace_disabled))
6786 return -ENODEV;
6787
6788 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
6789 if (fgd == NULL)
6790 return -ENOMEM;
6791
6792 mutex_lock(&graph_lock);
6793
6794 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
6795 lockdep_is_held(&graph_lock));
6796 fgd->type = GRAPH_FILTER_FUNCTION;
6797 fgd->seq_ops = &ftrace_graph_seq_ops;
6798
6799 ret = __ftrace_graph_open(inode, file, fgd);
6800 if (ret < 0)
6801 kfree(fgd);
6802
6803 mutex_unlock(&graph_lock);
6804 return ret;
6805 }
6806
6807 static int
ftrace_graph_notrace_open(struct inode * inode,struct file * file)6808 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
6809 {
6810 struct ftrace_graph_data *fgd;
6811 int ret;
6812
6813 if (unlikely(ftrace_disabled))
6814 return -ENODEV;
6815
6816 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
6817 if (fgd == NULL)
6818 return -ENOMEM;
6819
6820 mutex_lock(&graph_lock);
6821
6822 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6823 lockdep_is_held(&graph_lock));
6824 fgd->type = GRAPH_FILTER_NOTRACE;
6825 fgd->seq_ops = &ftrace_graph_seq_ops;
6826
6827 ret = __ftrace_graph_open(inode, file, fgd);
6828 if (ret < 0)
6829 kfree(fgd);
6830
6831 mutex_unlock(&graph_lock);
6832 return ret;
6833 }
6834
6835 static int
ftrace_graph_release(struct inode * inode,struct file * file)6836 ftrace_graph_release(struct inode *inode, struct file *file)
6837 {
6838 struct ftrace_graph_data *fgd;
6839 struct ftrace_hash *old_hash, *new_hash;
6840 struct trace_parser *parser;
6841 int ret = 0;
6842
6843 if (file->f_mode & FMODE_READ) {
6844 struct seq_file *m = file->private_data;
6845
6846 fgd = m->private;
6847 seq_release(inode, file);
6848 } else {
6849 fgd = file->private_data;
6850 }
6851
6852
6853 if (file->f_mode & FMODE_WRITE) {
6854
6855 parser = &fgd->parser;
6856
6857 if (trace_parser_loaded((parser))) {
6858 ret = ftrace_graph_set_hash(fgd->new_hash,
6859 parser->buffer);
6860 }
6861
6862 trace_parser_put(parser);
6863
6864 new_hash = __ftrace_hash_move(fgd->new_hash);
6865 if (!new_hash) {
6866 ret = -ENOMEM;
6867 goto out;
6868 }
6869
6870 mutex_lock(&graph_lock);
6871
6872 if (fgd->type == GRAPH_FILTER_FUNCTION) {
6873 old_hash = rcu_dereference_protected(ftrace_graph_hash,
6874 lockdep_is_held(&graph_lock));
6875 rcu_assign_pointer(ftrace_graph_hash, new_hash);
6876 } else {
6877 old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6878 lockdep_is_held(&graph_lock));
6879 rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
6880 }
6881
6882 mutex_unlock(&graph_lock);
6883
6884 /*
6885 * We need to do a hard force of sched synchronization.
6886 * This is because we use preempt_disable() to do RCU, but
6887 * the function tracers can be called where RCU is not watching
6888 * (like before user_exit()). We can not rely on the RCU
6889 * infrastructure to do the synchronization, thus we must do it
6890 * ourselves.
6891 */
6892 if (old_hash != EMPTY_HASH)
6893 synchronize_rcu_tasks_rude();
6894
6895 free_ftrace_hash(old_hash);
6896 }
6897
6898 out:
6899 free_ftrace_hash(fgd->new_hash);
6900 kfree(fgd);
6901
6902 return ret;
6903 }
6904
6905 static int
ftrace_graph_set_hash(struct ftrace_hash * hash,char * buffer)6906 ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
6907 {
6908 struct ftrace_glob func_g;
6909 struct dyn_ftrace *rec;
6910 struct ftrace_page *pg;
6911 struct ftrace_func_entry *entry;
6912 int fail = 1;
6913 int not;
6914
6915 /* decode regex */
6916 func_g.type = filter_parse_regex(buffer, strlen(buffer),
6917 &func_g.search, ¬);
6918
6919 func_g.len = strlen(func_g.search);
6920
6921 guard(mutex)(&ftrace_lock);
6922
6923 if (unlikely(ftrace_disabled))
6924 return -ENODEV;
6925
6926 do_for_each_ftrace_rec(pg, rec) {
6927
6928 if (rec->flags & FTRACE_FL_DISABLED)
6929 continue;
6930
6931 if (ftrace_match_record(rec, &func_g, NULL, 0)) {
6932 entry = ftrace_lookup_ip(hash, rec->ip);
6933
6934 if (!not) {
6935 fail = 0;
6936
6937 if (entry)
6938 continue;
6939 if (add_hash_entry(hash, rec->ip) == NULL)
6940 return 0;
6941 } else {
6942 if (entry) {
6943 free_hash_entry(hash, entry);
6944 fail = 0;
6945 }
6946 }
6947 }
6948 cond_resched();
6949 } while_for_each_ftrace_rec();
6950
6951 return fail ? -EINVAL : 0;
6952 }
6953
6954 static ssize_t
ftrace_graph_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)6955 ftrace_graph_write(struct file *file, const char __user *ubuf,
6956 size_t cnt, loff_t *ppos)
6957 {
6958 ssize_t read, ret = 0;
6959 struct ftrace_graph_data *fgd = file->private_data;
6960 struct trace_parser *parser;
6961
6962 if (!cnt)
6963 return 0;
6964
6965 /* Read mode uses seq functions */
6966 if (file->f_mode & FMODE_READ) {
6967 struct seq_file *m = file->private_data;
6968 fgd = m->private;
6969 }
6970
6971 parser = &fgd->parser;
6972
6973 read = trace_get_user(parser, ubuf, cnt, ppos);
6974
6975 if (read >= 0 && trace_parser_loaded(parser) &&
6976 !trace_parser_cont(parser)) {
6977
6978 ret = ftrace_graph_set_hash(fgd->new_hash,
6979 parser->buffer);
6980 trace_parser_clear(parser);
6981 }
6982
6983 if (!ret)
6984 ret = read;
6985
6986 return ret;
6987 }
6988
6989 static const struct file_operations ftrace_graph_fops = {
6990 .open = ftrace_graph_open,
6991 .read = seq_read,
6992 .write = ftrace_graph_write,
6993 .llseek = tracing_lseek,
6994 .release = ftrace_graph_release,
6995 };
6996
6997 static const struct file_operations ftrace_graph_notrace_fops = {
6998 .open = ftrace_graph_notrace_open,
6999 .read = seq_read,
7000 .write = ftrace_graph_write,
7001 .llseek = tracing_lseek,
7002 .release = ftrace_graph_release,
7003 };
7004 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
7005
ftrace_create_filter_files(struct ftrace_ops * ops,struct dentry * parent)7006 void ftrace_create_filter_files(struct ftrace_ops *ops,
7007 struct dentry *parent)
7008 {
7009
7010 trace_create_file("set_ftrace_filter", TRACE_MODE_WRITE, parent,
7011 ops, &ftrace_filter_fops);
7012
7013 trace_create_file("set_ftrace_notrace", TRACE_MODE_WRITE, parent,
7014 ops, &ftrace_notrace_fops);
7015 }
7016
7017 /*
7018 * The name "destroy_filter_files" is really a misnomer. Although
7019 * in the future, it may actually delete the files, but this is
7020 * really intended to make sure the ops passed in are disabled
7021 * and that when this function returns, the caller is free to
7022 * free the ops.
7023 *
7024 * The "destroy" name is only to match the "create" name that this
7025 * should be paired with.
7026 */
ftrace_destroy_filter_files(struct ftrace_ops * ops)7027 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
7028 {
7029 mutex_lock(&ftrace_lock);
7030 if (ops->flags & FTRACE_OPS_FL_ENABLED)
7031 ftrace_shutdown(ops, 0);
7032 ops->flags |= FTRACE_OPS_FL_DELETED;
7033 ftrace_free_filter(ops);
7034 mutex_unlock(&ftrace_lock);
7035 }
7036
ftrace_init_dyn_tracefs(struct dentry * d_tracer)7037 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
7038 {
7039
7040 trace_create_file("available_filter_functions", TRACE_MODE_READ,
7041 d_tracer, NULL, &ftrace_avail_fops);
7042
7043 trace_create_file("available_filter_functions_addrs", TRACE_MODE_READ,
7044 d_tracer, NULL, &ftrace_avail_addrs_fops);
7045
7046 trace_create_file("enabled_functions", TRACE_MODE_READ,
7047 d_tracer, NULL, &ftrace_enabled_fops);
7048
7049 trace_create_file("touched_functions", TRACE_MODE_READ,
7050 d_tracer, NULL, &ftrace_touched_fops);
7051
7052 ftrace_create_filter_files(&global_ops, d_tracer);
7053
7054 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
7055 trace_create_file("set_graph_function", TRACE_MODE_WRITE, d_tracer,
7056 NULL,
7057 &ftrace_graph_fops);
7058 trace_create_file("set_graph_notrace", TRACE_MODE_WRITE, d_tracer,
7059 NULL,
7060 &ftrace_graph_notrace_fops);
7061 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
7062
7063 return 0;
7064 }
7065
ftrace_cmp_ips(const void * a,const void * b)7066 static int ftrace_cmp_ips(const void *a, const void *b)
7067 {
7068 const unsigned long *ipa = a;
7069 const unsigned long *ipb = b;
7070
7071 if (*ipa > *ipb)
7072 return 1;
7073 if (*ipa < *ipb)
7074 return -1;
7075 return 0;
7076 }
7077
7078 #ifdef CONFIG_FTRACE_SORT_STARTUP_TEST
test_is_sorted(unsigned long * start,unsigned long count)7079 static void test_is_sorted(unsigned long *start, unsigned long count)
7080 {
7081 int i;
7082
7083 for (i = 1; i < count; i++) {
7084 if (WARN(start[i - 1] > start[i],
7085 "[%d] %pS at %lx is not sorted with %pS at %lx\n", i,
7086 (void *)start[i - 1], start[i - 1],
7087 (void *)start[i], start[i]))
7088 break;
7089 }
7090 if (i == count)
7091 pr_info("ftrace section at %px sorted properly\n", start);
7092 }
7093 #else
test_is_sorted(unsigned long * start,unsigned long count)7094 static void test_is_sorted(unsigned long *start, unsigned long count)
7095 {
7096 }
7097 #endif
7098
ftrace_process_locs(struct module * mod,unsigned long * start,unsigned long * end)7099 static int ftrace_process_locs(struct module *mod,
7100 unsigned long *start,
7101 unsigned long *end)
7102 {
7103 struct ftrace_page *pg_unuse = NULL;
7104 struct ftrace_page *start_pg;
7105 struct ftrace_page *pg;
7106 struct dyn_ftrace *rec;
7107 unsigned long skipped = 0;
7108 unsigned long count;
7109 unsigned long *p;
7110 unsigned long addr;
7111 unsigned long flags = 0; /* Shut up gcc */
7112 unsigned long pages;
7113 int ret = -ENOMEM;
7114
7115 count = end - start;
7116
7117 if (!count)
7118 return 0;
7119
7120 pages = DIV_ROUND_UP(count, ENTRIES_PER_PAGE);
7121
7122 /*
7123 * Sorting mcount in vmlinux at build time depend on
7124 * CONFIG_BUILDTIME_MCOUNT_SORT, while mcount loc in
7125 * modules can not be sorted at build time.
7126 */
7127 if (!IS_ENABLED(CONFIG_BUILDTIME_MCOUNT_SORT) || mod) {
7128 sort(start, count, sizeof(*start),
7129 ftrace_cmp_ips, NULL);
7130 } else {
7131 test_is_sorted(start, count);
7132 }
7133
7134 start_pg = ftrace_allocate_pages(count);
7135 if (!start_pg)
7136 return -ENOMEM;
7137
7138 mutex_lock(&ftrace_lock);
7139
7140 /*
7141 * Core and each module needs their own pages, as
7142 * modules will free them when they are removed.
7143 * Force a new page to be allocated for modules.
7144 */
7145 if (!mod) {
7146 WARN_ON(ftrace_pages || ftrace_pages_start);
7147 /* First initialization */
7148 ftrace_pages = ftrace_pages_start = start_pg;
7149 } else {
7150 if (!ftrace_pages)
7151 goto out;
7152
7153 if (WARN_ON(ftrace_pages->next)) {
7154 /* Hmm, we have free pages? */
7155 while (ftrace_pages->next)
7156 ftrace_pages = ftrace_pages->next;
7157 }
7158
7159 ftrace_pages->next = start_pg;
7160 }
7161
7162 p = start;
7163 pg = start_pg;
7164 while (p < end) {
7165 unsigned long end_offset;
7166
7167 addr = *p++;
7168
7169 /*
7170 * Some architecture linkers will pad between
7171 * the different mcount_loc sections of different
7172 * object files to satisfy alignments.
7173 * Skip any NULL pointers.
7174 */
7175 if (!addr) {
7176 skipped++;
7177 continue;
7178 }
7179
7180 /*
7181 * If this is core kernel, make sure the address is in core
7182 * or inittext, as weak functions get zeroed and KASLR can
7183 * move them to something other than zero. It just will not
7184 * move it to an area where kernel text is.
7185 */
7186 if (!mod && !(is_kernel_text(addr) || is_kernel_inittext(addr))) {
7187 skipped++;
7188 continue;
7189 }
7190
7191 addr = ftrace_call_adjust(addr);
7192
7193 end_offset = (pg->index+1) * sizeof(pg->records[0]);
7194 if (end_offset > PAGE_SIZE << pg->order) {
7195 /* We should have allocated enough */
7196 if (WARN_ON(!pg->next))
7197 break;
7198 pg = pg->next;
7199 }
7200
7201 rec = &pg->records[pg->index++];
7202 rec->ip = addr;
7203 }
7204
7205 if (pg->next) {
7206 pg_unuse = pg->next;
7207 pg->next = NULL;
7208 }
7209
7210 /* Assign the last page to ftrace_pages */
7211 ftrace_pages = pg;
7212
7213 /*
7214 * We only need to disable interrupts on start up
7215 * because we are modifying code that an interrupt
7216 * may execute, and the modification is not atomic.
7217 * But for modules, nothing runs the code we modify
7218 * until we are finished with it, and there's no
7219 * reason to cause large interrupt latencies while we do it.
7220 */
7221 if (!mod)
7222 local_irq_save(flags);
7223 ftrace_update_code(mod, start_pg);
7224 if (!mod)
7225 local_irq_restore(flags);
7226 ret = 0;
7227 out:
7228 mutex_unlock(&ftrace_lock);
7229
7230 /* We should have used all pages unless we skipped some */
7231 if (pg_unuse) {
7232 unsigned long pg_remaining, remaining = 0;
7233 unsigned long skip;
7234
7235 /* Count the number of entries unused and compare it to skipped. */
7236 pg_remaining = (ENTRIES_PER_PAGE << pg->order) - pg->index;
7237
7238 if (!WARN(skipped < pg_remaining, "Extra allocated pages for ftrace")) {
7239
7240 skip = skipped - pg_remaining;
7241
7242 for (pg = pg_unuse; pg; pg = pg->next)
7243 remaining += 1 << pg->order;
7244
7245 pages -= remaining;
7246
7247 skip = DIV_ROUND_UP(skip, ENTRIES_PER_PAGE);
7248
7249 /*
7250 * Check to see if the number of pages remaining would
7251 * just fit the number of entries skipped.
7252 */
7253 WARN(skip != remaining, "Extra allocated pages for ftrace: %lu with %lu skipped",
7254 remaining, skipped);
7255 }
7256 /* Need to synchronize with ftrace_location_range() */
7257 synchronize_rcu();
7258 ftrace_free_pages(pg_unuse);
7259 }
7260
7261 if (!mod) {
7262 count -= skipped;
7263 pr_info("ftrace: allocating %ld entries in %ld pages\n",
7264 count, pages);
7265 }
7266
7267 return ret;
7268 }
7269
7270 struct ftrace_mod_func {
7271 struct list_head list;
7272 char *name;
7273 unsigned long ip;
7274 unsigned int size;
7275 };
7276
7277 struct ftrace_mod_map {
7278 struct rcu_head rcu;
7279 struct list_head list;
7280 struct module *mod;
7281 unsigned long start_addr;
7282 unsigned long end_addr;
7283 struct list_head funcs;
7284 unsigned int num_funcs;
7285 };
7286
ftrace_get_trampoline_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7287 static int ftrace_get_trampoline_kallsym(unsigned int symnum,
7288 unsigned long *value, char *type,
7289 char *name, char *module_name,
7290 int *exported)
7291 {
7292 struct ftrace_ops *op;
7293
7294 list_for_each_entry_rcu(op, &ftrace_ops_trampoline_list, list) {
7295 if (!op->trampoline || symnum--)
7296 continue;
7297 *value = op->trampoline;
7298 *type = 't';
7299 strscpy(name, FTRACE_TRAMPOLINE_SYM, KSYM_NAME_LEN);
7300 strscpy(module_name, FTRACE_TRAMPOLINE_MOD, MODULE_NAME_LEN);
7301 *exported = 0;
7302 return 0;
7303 }
7304
7305 return -ERANGE;
7306 }
7307
7308 #if defined(CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS) || defined(CONFIG_MODULES)
7309 /*
7310 * Check if the current ops references the given ip.
7311 *
7312 * If the ops traces all functions, then it was already accounted for.
7313 * If the ops does not trace the current record function, skip it.
7314 * If the ops ignores the function via notrace filter, skip it.
7315 */
7316 static bool
ops_references_ip(struct ftrace_ops * ops,unsigned long ip)7317 ops_references_ip(struct ftrace_ops *ops, unsigned long ip)
7318 {
7319 /* If ops isn't enabled, ignore it */
7320 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
7321 return false;
7322
7323 /* If ops traces all then it includes this function */
7324 if (ops_traces_mod(ops))
7325 return true;
7326
7327 /* The function must be in the filter */
7328 if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
7329 !__ftrace_lookup_ip(ops->func_hash->filter_hash, ip))
7330 return false;
7331
7332 /* If in notrace hash, we ignore it too */
7333 if (ftrace_lookup_ip(ops->func_hash->notrace_hash, ip))
7334 return false;
7335
7336 return true;
7337 }
7338 #endif
7339
7340 #ifdef CONFIG_MODULES
7341
7342 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
7343
7344 static LIST_HEAD(ftrace_mod_maps);
7345
referenced_filters(struct dyn_ftrace * rec)7346 static int referenced_filters(struct dyn_ftrace *rec)
7347 {
7348 struct ftrace_ops *ops;
7349 int cnt = 0;
7350
7351 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
7352 if (ops_references_ip(ops, rec->ip)) {
7353 if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_DIRECT))
7354 continue;
7355 if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_IPMODIFY))
7356 continue;
7357 cnt++;
7358 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
7359 rec->flags |= FTRACE_FL_REGS;
7360 if (cnt == 1 && ops->trampoline)
7361 rec->flags |= FTRACE_FL_TRAMP;
7362 else
7363 rec->flags &= ~FTRACE_FL_TRAMP;
7364 }
7365 }
7366
7367 return cnt;
7368 }
7369
7370 static void
clear_mod_from_hash(struct ftrace_page * pg,struct ftrace_hash * hash)7371 clear_mod_from_hash(struct ftrace_page *pg, struct ftrace_hash *hash)
7372 {
7373 struct ftrace_func_entry *entry;
7374 struct dyn_ftrace *rec;
7375 int i;
7376
7377 if (ftrace_hash_empty(hash))
7378 return;
7379
7380 for (i = 0; i < pg->index; i++) {
7381 rec = &pg->records[i];
7382 entry = __ftrace_lookup_ip(hash, rec->ip);
7383 /*
7384 * Do not allow this rec to match again.
7385 * Yeah, it may waste some memory, but will be removed
7386 * if/when the hash is modified again.
7387 */
7388 if (entry)
7389 entry->ip = 0;
7390 }
7391 }
7392
7393 /* Clear any records from hashes */
clear_mod_from_hashes(struct ftrace_page * pg)7394 static void clear_mod_from_hashes(struct ftrace_page *pg)
7395 {
7396 struct trace_array *tr;
7397
7398 mutex_lock(&trace_types_lock);
7399 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7400 if (!tr->ops || !tr->ops->func_hash)
7401 continue;
7402 mutex_lock(&tr->ops->func_hash->regex_lock);
7403 clear_mod_from_hash(pg, tr->ops->func_hash->filter_hash);
7404 clear_mod_from_hash(pg, tr->ops->func_hash->notrace_hash);
7405 mutex_unlock(&tr->ops->func_hash->regex_lock);
7406 }
7407 mutex_unlock(&trace_types_lock);
7408 }
7409
ftrace_free_mod_map(struct rcu_head * rcu)7410 static void ftrace_free_mod_map(struct rcu_head *rcu)
7411 {
7412 struct ftrace_mod_map *mod_map = container_of(rcu, struct ftrace_mod_map, rcu);
7413 struct ftrace_mod_func *mod_func;
7414 struct ftrace_mod_func *n;
7415
7416 /* All the contents of mod_map are now not visible to readers */
7417 list_for_each_entry_safe(mod_func, n, &mod_map->funcs, list) {
7418 kfree(mod_func->name);
7419 list_del(&mod_func->list);
7420 kfree(mod_func);
7421 }
7422
7423 kfree(mod_map);
7424 }
7425
ftrace_release_mod(struct module * mod)7426 void ftrace_release_mod(struct module *mod)
7427 {
7428 struct ftrace_mod_map *mod_map;
7429 struct ftrace_mod_map *n;
7430 struct dyn_ftrace *rec;
7431 struct ftrace_page **last_pg;
7432 struct ftrace_page *tmp_page = NULL;
7433 struct ftrace_page *pg;
7434
7435 mutex_lock(&ftrace_lock);
7436
7437 /*
7438 * To avoid the UAF problem after the module is unloaded, the
7439 * 'mod_map' resource needs to be released unconditionally.
7440 */
7441 list_for_each_entry_safe(mod_map, n, &ftrace_mod_maps, list) {
7442 if (mod_map->mod == mod) {
7443 list_del_rcu(&mod_map->list);
7444 call_rcu(&mod_map->rcu, ftrace_free_mod_map);
7445 break;
7446 }
7447 }
7448
7449 if (ftrace_disabled)
7450 goto out_unlock;
7451
7452 /*
7453 * Each module has its own ftrace_pages, remove
7454 * them from the list.
7455 */
7456 last_pg = &ftrace_pages_start;
7457 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
7458 rec = &pg->records[0];
7459 if (within_module(rec->ip, mod)) {
7460 /*
7461 * As core pages are first, the first
7462 * page should never be a module page.
7463 */
7464 if (WARN_ON(pg == ftrace_pages_start))
7465 goto out_unlock;
7466
7467 /* Check if we are deleting the last page */
7468 if (pg == ftrace_pages)
7469 ftrace_pages = next_to_ftrace_page(last_pg);
7470
7471 ftrace_update_tot_cnt -= pg->index;
7472 *last_pg = pg->next;
7473
7474 pg->next = tmp_page;
7475 tmp_page = pg;
7476 } else
7477 last_pg = &pg->next;
7478 }
7479 out_unlock:
7480 mutex_unlock(&ftrace_lock);
7481
7482 /* Need to synchronize with ftrace_location_range() */
7483 if (tmp_page)
7484 synchronize_rcu();
7485 for (pg = tmp_page; pg; pg = tmp_page) {
7486
7487 /* Needs to be called outside of ftrace_lock */
7488 clear_mod_from_hashes(pg);
7489
7490 if (pg->records) {
7491 free_pages((unsigned long)pg->records, pg->order);
7492 ftrace_number_of_pages -= 1 << pg->order;
7493 }
7494 tmp_page = pg->next;
7495 kfree(pg);
7496 ftrace_number_of_groups--;
7497 }
7498 }
7499
ftrace_module_enable(struct module * mod)7500 void ftrace_module_enable(struct module *mod)
7501 {
7502 struct dyn_ftrace *rec;
7503 struct ftrace_page *pg;
7504
7505 mutex_lock(&ftrace_lock);
7506
7507 if (ftrace_disabled)
7508 goto out_unlock;
7509
7510 /*
7511 * If the tracing is enabled, go ahead and enable the record.
7512 *
7513 * The reason not to enable the record immediately is the
7514 * inherent check of ftrace_make_nop/ftrace_make_call for
7515 * correct previous instructions. Making first the NOP
7516 * conversion puts the module to the correct state, thus
7517 * passing the ftrace_make_call check.
7518 *
7519 * We also delay this to after the module code already set the
7520 * text to read-only, as we now need to set it back to read-write
7521 * so that we can modify the text.
7522 */
7523 if (ftrace_start_up)
7524 ftrace_arch_code_modify_prepare();
7525
7526 do_for_each_ftrace_rec(pg, rec) {
7527 int cnt;
7528 /*
7529 * do_for_each_ftrace_rec() is a double loop.
7530 * module text shares the pg. If a record is
7531 * not part of this module, then skip this pg,
7532 * which the "break" will do.
7533 */
7534 if (!within_module(rec->ip, mod))
7535 break;
7536
7537 /* Weak functions should still be ignored */
7538 if (!test_for_valid_rec(rec)) {
7539 /* Clear all other flags. Should not be enabled anyway */
7540 rec->flags = FTRACE_FL_DISABLED;
7541 continue;
7542 }
7543
7544 cnt = 0;
7545
7546 /*
7547 * When adding a module, we need to check if tracers are
7548 * currently enabled and if they are, and can trace this record,
7549 * we need to enable the module functions as well as update the
7550 * reference counts for those function records.
7551 */
7552 if (ftrace_start_up)
7553 cnt += referenced_filters(rec);
7554
7555 rec->flags &= ~FTRACE_FL_DISABLED;
7556 rec->flags += cnt;
7557
7558 if (ftrace_start_up && cnt) {
7559 int failed = __ftrace_replace_code(rec, 1);
7560 if (failed) {
7561 ftrace_bug(failed, rec);
7562 goto out_loop;
7563 }
7564 }
7565
7566 } while_for_each_ftrace_rec();
7567
7568 out_loop:
7569 if (ftrace_start_up)
7570 ftrace_arch_code_modify_post_process();
7571
7572 out_unlock:
7573 mutex_unlock(&ftrace_lock);
7574
7575 process_cached_mods(mod->name);
7576 }
7577
ftrace_module_init(struct module * mod)7578 void ftrace_module_init(struct module *mod)
7579 {
7580 int ret;
7581
7582 if (ftrace_disabled || !mod->num_ftrace_callsites)
7583 return;
7584
7585 ret = ftrace_process_locs(mod, mod->ftrace_callsites,
7586 mod->ftrace_callsites + mod->num_ftrace_callsites);
7587 if (ret)
7588 pr_warn("ftrace: failed to allocate entries for module '%s' functions\n",
7589 mod->name);
7590 }
7591
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)7592 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
7593 struct dyn_ftrace *rec)
7594 {
7595 struct ftrace_mod_func *mod_func;
7596 unsigned long symsize;
7597 unsigned long offset;
7598 char str[KSYM_SYMBOL_LEN];
7599 char *modname;
7600 const char *ret;
7601
7602 ret = kallsyms_lookup(rec->ip, &symsize, &offset, &modname, str);
7603 if (!ret)
7604 return;
7605
7606 mod_func = kmalloc(sizeof(*mod_func), GFP_KERNEL);
7607 if (!mod_func)
7608 return;
7609
7610 mod_func->name = kstrdup(str, GFP_KERNEL);
7611 if (!mod_func->name) {
7612 kfree(mod_func);
7613 return;
7614 }
7615
7616 mod_func->ip = rec->ip - offset;
7617 mod_func->size = symsize;
7618
7619 mod_map->num_funcs++;
7620
7621 list_add_rcu(&mod_func->list, &mod_map->funcs);
7622 }
7623
7624 static struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)7625 allocate_ftrace_mod_map(struct module *mod,
7626 unsigned long start, unsigned long end)
7627 {
7628 struct ftrace_mod_map *mod_map;
7629
7630 if (ftrace_disabled)
7631 return NULL;
7632
7633 mod_map = kmalloc(sizeof(*mod_map), GFP_KERNEL);
7634 if (!mod_map)
7635 return NULL;
7636
7637 mod_map->mod = mod;
7638 mod_map->start_addr = start;
7639 mod_map->end_addr = end;
7640 mod_map->num_funcs = 0;
7641
7642 INIT_LIST_HEAD_RCU(&mod_map->funcs);
7643
7644 list_add_rcu(&mod_map->list, &ftrace_mod_maps);
7645
7646 return mod_map;
7647 }
7648
7649 static int
ftrace_func_address_lookup(struct ftrace_mod_map * mod_map,unsigned long addr,unsigned long * size,unsigned long * off,char * sym)7650 ftrace_func_address_lookup(struct ftrace_mod_map *mod_map,
7651 unsigned long addr, unsigned long *size,
7652 unsigned long *off, char *sym)
7653 {
7654 struct ftrace_mod_func *found_func = NULL;
7655 struct ftrace_mod_func *mod_func;
7656
7657 list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
7658 if (addr >= mod_func->ip &&
7659 addr < mod_func->ip + mod_func->size) {
7660 found_func = mod_func;
7661 break;
7662 }
7663 }
7664
7665 if (found_func) {
7666 if (size)
7667 *size = found_func->size;
7668 if (off)
7669 *off = addr - found_func->ip;
7670 return strscpy(sym, found_func->name, KSYM_NAME_LEN);
7671 }
7672
7673 return 0;
7674 }
7675
7676 int
ftrace_mod_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)7677 ftrace_mod_address_lookup(unsigned long addr, unsigned long *size,
7678 unsigned long *off, char **modname, char *sym)
7679 {
7680 struct ftrace_mod_map *mod_map;
7681 int ret = 0;
7682
7683 /* mod_map is freed via call_rcu() */
7684 preempt_disable();
7685 list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
7686 ret = ftrace_func_address_lookup(mod_map, addr, size, off, sym);
7687 if (ret) {
7688 if (modname)
7689 *modname = mod_map->mod->name;
7690 break;
7691 }
7692 }
7693 preempt_enable();
7694
7695 return ret;
7696 }
7697
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7698 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
7699 char *type, char *name,
7700 char *module_name, int *exported)
7701 {
7702 struct ftrace_mod_map *mod_map;
7703 struct ftrace_mod_func *mod_func;
7704 int ret;
7705
7706 preempt_disable();
7707 list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
7708
7709 if (symnum >= mod_map->num_funcs) {
7710 symnum -= mod_map->num_funcs;
7711 continue;
7712 }
7713
7714 list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
7715 if (symnum > 1) {
7716 symnum--;
7717 continue;
7718 }
7719
7720 *value = mod_func->ip;
7721 *type = 'T';
7722 strscpy(name, mod_func->name, KSYM_NAME_LEN);
7723 strscpy(module_name, mod_map->mod->name, MODULE_NAME_LEN);
7724 *exported = 1;
7725 preempt_enable();
7726 return 0;
7727 }
7728 WARN_ON(1);
7729 break;
7730 }
7731 ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
7732 module_name, exported);
7733 preempt_enable();
7734 return ret;
7735 }
7736
7737 #else
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)7738 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
7739 struct dyn_ftrace *rec) { }
7740 static inline struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)7741 allocate_ftrace_mod_map(struct module *mod,
7742 unsigned long start, unsigned long end)
7743 {
7744 return NULL;
7745 }
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7746 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
7747 char *type, char *name, char *module_name,
7748 int *exported)
7749 {
7750 int ret;
7751
7752 preempt_disable();
7753 ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
7754 module_name, exported);
7755 preempt_enable();
7756 return ret;
7757 }
7758 #endif /* CONFIG_MODULES */
7759
7760 struct ftrace_init_func {
7761 struct list_head list;
7762 unsigned long ip;
7763 };
7764
7765 /* Clear any init ips from hashes */
7766 static void
clear_func_from_hash(struct ftrace_init_func * func,struct ftrace_hash * hash)7767 clear_func_from_hash(struct ftrace_init_func *func, struct ftrace_hash *hash)
7768 {
7769 struct ftrace_func_entry *entry;
7770
7771 entry = ftrace_lookup_ip(hash, func->ip);
7772 /*
7773 * Do not allow this rec to match again.
7774 * Yeah, it may waste some memory, but will be removed
7775 * if/when the hash is modified again.
7776 */
7777 if (entry)
7778 entry->ip = 0;
7779 }
7780
7781 static void
clear_func_from_hashes(struct ftrace_init_func * func)7782 clear_func_from_hashes(struct ftrace_init_func *func)
7783 {
7784 struct trace_array *tr;
7785
7786 mutex_lock(&trace_types_lock);
7787 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7788 if (!tr->ops || !tr->ops->func_hash)
7789 continue;
7790 mutex_lock(&tr->ops->func_hash->regex_lock);
7791 clear_func_from_hash(func, tr->ops->func_hash->filter_hash);
7792 clear_func_from_hash(func, tr->ops->func_hash->notrace_hash);
7793 mutex_unlock(&tr->ops->func_hash->regex_lock);
7794 }
7795 mutex_unlock(&trace_types_lock);
7796 }
7797
add_to_clear_hash_list(struct list_head * clear_list,struct dyn_ftrace * rec)7798 static void add_to_clear_hash_list(struct list_head *clear_list,
7799 struct dyn_ftrace *rec)
7800 {
7801 struct ftrace_init_func *func;
7802
7803 func = kmalloc(sizeof(*func), GFP_KERNEL);
7804 if (!func) {
7805 MEM_FAIL(1, "alloc failure, ftrace filter could be stale\n");
7806 return;
7807 }
7808
7809 func->ip = rec->ip;
7810 list_add(&func->list, clear_list);
7811 }
7812
ftrace_free_mem(struct module * mod,void * start_ptr,void * end_ptr)7813 void ftrace_free_mem(struct module *mod, void *start_ptr, void *end_ptr)
7814 {
7815 unsigned long start = (unsigned long)(start_ptr);
7816 unsigned long end = (unsigned long)(end_ptr);
7817 struct ftrace_page **last_pg = &ftrace_pages_start;
7818 struct ftrace_page *tmp_page = NULL;
7819 struct ftrace_page *pg;
7820 struct dyn_ftrace *rec;
7821 struct dyn_ftrace key;
7822 struct ftrace_mod_map *mod_map = NULL;
7823 struct ftrace_init_func *func, *func_next;
7824 LIST_HEAD(clear_hash);
7825
7826 key.ip = start;
7827 key.flags = end; /* overload flags, as it is unsigned long */
7828
7829 mutex_lock(&ftrace_lock);
7830
7831 /*
7832 * If we are freeing module init memory, then check if
7833 * any tracer is active. If so, we need to save a mapping of
7834 * the module functions being freed with the address.
7835 */
7836 if (mod && ftrace_ops_list != &ftrace_list_end)
7837 mod_map = allocate_ftrace_mod_map(mod, start, end);
7838
7839 for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
7840 if (end < pg->records[0].ip ||
7841 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
7842 continue;
7843 again:
7844 rec = bsearch(&key, pg->records, pg->index,
7845 sizeof(struct dyn_ftrace),
7846 ftrace_cmp_recs);
7847 if (!rec)
7848 continue;
7849
7850 /* rec will be cleared from hashes after ftrace_lock unlock */
7851 add_to_clear_hash_list(&clear_hash, rec);
7852
7853 if (mod_map)
7854 save_ftrace_mod_rec(mod_map, rec);
7855
7856 pg->index--;
7857 ftrace_update_tot_cnt--;
7858 if (!pg->index) {
7859 *last_pg = pg->next;
7860 pg->next = tmp_page;
7861 tmp_page = pg;
7862 pg = container_of(last_pg, struct ftrace_page, next);
7863 if (!(*last_pg))
7864 ftrace_pages = pg;
7865 continue;
7866 }
7867 memmove(rec, rec + 1,
7868 (pg->index - (rec - pg->records)) * sizeof(*rec));
7869 /* More than one function may be in this block */
7870 goto again;
7871 }
7872 mutex_unlock(&ftrace_lock);
7873
7874 list_for_each_entry_safe(func, func_next, &clear_hash, list) {
7875 clear_func_from_hashes(func);
7876 kfree(func);
7877 }
7878 /* Need to synchronize with ftrace_location_range() */
7879 if (tmp_page) {
7880 synchronize_rcu();
7881 ftrace_free_pages(tmp_page);
7882 }
7883 }
7884
ftrace_free_init_mem(void)7885 void __init ftrace_free_init_mem(void)
7886 {
7887 void *start = (void *)(&__init_begin);
7888 void *end = (void *)(&__init_end);
7889
7890 ftrace_boot_snapshot();
7891
7892 ftrace_free_mem(NULL, start, end);
7893 }
7894
ftrace_dyn_arch_init(void)7895 int __init __weak ftrace_dyn_arch_init(void)
7896 {
7897 return 0;
7898 }
7899
ftrace_init(void)7900 void __init ftrace_init(void)
7901 {
7902 extern unsigned long __start_mcount_loc[];
7903 extern unsigned long __stop_mcount_loc[];
7904 unsigned long count, flags;
7905 int ret;
7906
7907 local_irq_save(flags);
7908 ret = ftrace_dyn_arch_init();
7909 local_irq_restore(flags);
7910 if (ret)
7911 goto failed;
7912
7913 count = __stop_mcount_loc - __start_mcount_loc;
7914 if (!count) {
7915 pr_info("ftrace: No functions to be traced?\n");
7916 goto failed;
7917 }
7918
7919 ret = ftrace_process_locs(NULL,
7920 __start_mcount_loc,
7921 __stop_mcount_loc);
7922 if (ret) {
7923 pr_warn("ftrace: failed to allocate entries for functions\n");
7924 goto failed;
7925 }
7926
7927 pr_info("ftrace: allocated %ld pages with %ld groups\n",
7928 ftrace_number_of_pages, ftrace_number_of_groups);
7929
7930 last_ftrace_enabled = ftrace_enabled = 1;
7931
7932 set_ftrace_early_filters();
7933
7934 return;
7935 failed:
7936 ftrace_disabled = 1;
7937 }
7938
7939 /* Do nothing if arch does not support this */
arch_ftrace_update_trampoline(struct ftrace_ops * ops)7940 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
7941 {
7942 }
7943
ftrace_update_trampoline(struct ftrace_ops * ops)7944 static void ftrace_update_trampoline(struct ftrace_ops *ops)
7945 {
7946 unsigned long trampoline = ops->trampoline;
7947
7948 arch_ftrace_update_trampoline(ops);
7949 if (ops->trampoline && ops->trampoline != trampoline &&
7950 (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP)) {
7951 /* Add to kallsyms before the perf events */
7952 ftrace_add_trampoline_to_kallsyms(ops);
7953 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
7954 ops->trampoline, ops->trampoline_size, false,
7955 FTRACE_TRAMPOLINE_SYM);
7956 /*
7957 * Record the perf text poke event after the ksymbol register
7958 * event.
7959 */
7960 perf_event_text_poke((void *)ops->trampoline, NULL, 0,
7961 (void *)ops->trampoline,
7962 ops->trampoline_size);
7963 }
7964 }
7965
ftrace_init_trace_array(struct trace_array * tr)7966 void ftrace_init_trace_array(struct trace_array *tr)
7967 {
7968 if (tr->flags & TRACE_ARRAY_FL_MOD_INIT)
7969 return;
7970
7971 INIT_LIST_HEAD(&tr->func_probes);
7972 INIT_LIST_HEAD(&tr->mod_trace);
7973 INIT_LIST_HEAD(&tr->mod_notrace);
7974
7975 tr->flags |= TRACE_ARRAY_FL_MOD_INIT;
7976 }
7977 #else
7978
7979 struct ftrace_ops global_ops = {
7980 .func = ftrace_stub,
7981 .flags = FTRACE_OPS_FL_INITIALIZED |
7982 FTRACE_OPS_FL_PID,
7983 };
7984
ftrace_nodyn_init(void)7985 static int __init ftrace_nodyn_init(void)
7986 {
7987 ftrace_enabled = 1;
7988 return 0;
7989 }
7990 core_initcall(ftrace_nodyn_init);
7991
ftrace_init_dyn_tracefs(struct dentry * d_tracer)7992 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
ftrace_startup_all(int command)7993 static inline void ftrace_startup_all(int command) { }
7994
ftrace_update_trampoline(struct ftrace_ops * ops)7995 static void ftrace_update_trampoline(struct ftrace_ops *ops)
7996 {
7997 }
7998
7999 #endif /* CONFIG_DYNAMIC_FTRACE */
8000
ftrace_init_global_array_ops(struct trace_array * tr)8001 __init void ftrace_init_global_array_ops(struct trace_array *tr)
8002 {
8003 tr->ops = &global_ops;
8004 if (!global_ops.private)
8005 global_ops.private = tr;
8006 ftrace_init_trace_array(tr);
8007 init_array_fgraph_ops(tr, tr->ops);
8008 }
8009
ftrace_init_array_ops(struct trace_array * tr,ftrace_func_t func)8010 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
8011 {
8012 /* If we filter on pids, update to use the pid function */
8013 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
8014 if (WARN_ON(tr->ops->func != ftrace_stub))
8015 printk("ftrace ops had %pS for function\n",
8016 tr->ops->func);
8017 }
8018 tr->ops->func = func;
8019 tr->ops->private = tr;
8020 }
8021
ftrace_reset_array_ops(struct trace_array * tr)8022 void ftrace_reset_array_ops(struct trace_array *tr)
8023 {
8024 tr->ops->func = ftrace_stub;
8025 }
8026
8027 static nokprobe_inline void
__ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * ignored,struct ftrace_regs * fregs)8028 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
8029 struct ftrace_ops *ignored, struct ftrace_regs *fregs)
8030 {
8031 struct pt_regs *regs = ftrace_get_regs(fregs);
8032 struct ftrace_ops *op;
8033 int bit;
8034
8035 /*
8036 * The ftrace_test_and_set_recursion() will disable preemption,
8037 * which is required since some of the ops may be dynamically
8038 * allocated, they must be freed after a synchronize_rcu().
8039 */
8040 bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
8041 if (bit < 0)
8042 return;
8043
8044 do_for_each_ftrace_op(op, ftrace_ops_list) {
8045 /* Stub functions don't need to be called nor tested */
8046 if (op->flags & FTRACE_OPS_FL_STUB)
8047 continue;
8048 /*
8049 * Check the following for each ops before calling their func:
8050 * if RCU flag is set, then rcu_is_watching() must be true
8051 * Otherwise test if the ip matches the ops filter
8052 *
8053 * If any of the above fails then the op->func() is not executed.
8054 */
8055 if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
8056 ftrace_ops_test(op, ip, regs)) {
8057 if (FTRACE_WARN_ON(!op->func)) {
8058 pr_warn("op=%p %pS\n", op, op);
8059 goto out;
8060 }
8061 op->func(ip, parent_ip, op, fregs);
8062 }
8063 } while_for_each_ftrace_op(op);
8064 out:
8065 trace_clear_recursion(bit);
8066 }
8067
8068 /*
8069 * Some archs only support passing ip and parent_ip. Even though
8070 * the list function ignores the op parameter, we do not want any
8071 * C side effects, where a function is called without the caller
8072 * sending a third parameter.
8073 * Archs are to support both the regs and ftrace_ops at the same time.
8074 * If they support ftrace_ops, it is assumed they support regs.
8075 * If call backs want to use regs, they must either check for regs
8076 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
8077 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
8078 * An architecture can pass partial regs with ftrace_ops and still
8079 * set the ARCH_SUPPORTS_FTRACE_OPS.
8080 *
8081 * In vmlinux.lds.h, ftrace_ops_list_func() is defined to be
8082 * arch_ftrace_ops_list_func.
8083 */
8084 #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)8085 void arch_ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
8086 struct ftrace_ops *op, struct ftrace_regs *fregs)
8087 {
8088 kmsan_unpoison_memory(fregs, ftrace_regs_size());
8089 __ftrace_ops_list_func(ip, parent_ip, NULL, fregs);
8090 }
8091 #else
arch_ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip)8092 void arch_ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
8093 {
8094 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
8095 }
8096 #endif
8097 NOKPROBE_SYMBOL(arch_ftrace_ops_list_func);
8098
8099 /*
8100 * If there's only one function registered but it does not support
8101 * recursion, needs RCU protection, then this function will be called
8102 * by the mcount trampoline.
8103 */
ftrace_ops_assist_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)8104 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
8105 struct ftrace_ops *op, struct ftrace_regs *fregs)
8106 {
8107 int bit;
8108
8109 bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
8110 if (bit < 0)
8111 return;
8112
8113 if (!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching())
8114 op->func(ip, parent_ip, op, fregs);
8115
8116 trace_clear_recursion(bit);
8117 }
8118 NOKPROBE_SYMBOL(ftrace_ops_assist_func);
8119
8120 /**
8121 * ftrace_ops_get_func - get the function a trampoline should call
8122 * @ops: the ops to get the function for
8123 *
8124 * Normally the mcount trampoline will call the ops->func, but there
8125 * are times that it should not. For example, if the ops does not
8126 * have its own recursion protection, then it should call the
8127 * ftrace_ops_assist_func() instead.
8128 *
8129 * Returns: the function that the trampoline should call for @ops.
8130 */
ftrace_ops_get_func(struct ftrace_ops * ops)8131 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
8132 {
8133 /*
8134 * If the function does not handle recursion or needs to be RCU safe,
8135 * then we need to call the assist handler.
8136 */
8137 if (ops->flags & (FTRACE_OPS_FL_RECURSION |
8138 FTRACE_OPS_FL_RCU))
8139 return ftrace_ops_assist_func;
8140
8141 return ops->func;
8142 }
8143
8144 static void
ftrace_filter_pid_sched_switch_probe(void * data,bool preempt,struct task_struct * prev,struct task_struct * next,unsigned int prev_state)8145 ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
8146 struct task_struct *prev,
8147 struct task_struct *next,
8148 unsigned int prev_state)
8149 {
8150 struct trace_array *tr = data;
8151 struct trace_pid_list *pid_list;
8152 struct trace_pid_list *no_pid_list;
8153
8154 pid_list = rcu_dereference_sched(tr->function_pids);
8155 no_pid_list = rcu_dereference_sched(tr->function_no_pids);
8156
8157 if (trace_ignore_this_task(pid_list, no_pid_list, next))
8158 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8159 FTRACE_PID_IGNORE);
8160 else
8161 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8162 next->pid);
8163 }
8164
8165 static void
ftrace_pid_follow_sched_process_fork(void * data,struct task_struct * self,struct task_struct * task)8166 ftrace_pid_follow_sched_process_fork(void *data,
8167 struct task_struct *self,
8168 struct task_struct *task)
8169 {
8170 struct trace_pid_list *pid_list;
8171 struct trace_array *tr = data;
8172
8173 pid_list = rcu_dereference_sched(tr->function_pids);
8174 trace_filter_add_remove_task(pid_list, self, task);
8175
8176 pid_list = rcu_dereference_sched(tr->function_no_pids);
8177 trace_filter_add_remove_task(pid_list, self, task);
8178 }
8179
8180 static void
ftrace_pid_follow_sched_process_exit(void * data,struct task_struct * task)8181 ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
8182 {
8183 struct trace_pid_list *pid_list;
8184 struct trace_array *tr = data;
8185
8186 pid_list = rcu_dereference_sched(tr->function_pids);
8187 trace_filter_add_remove_task(pid_list, NULL, task);
8188
8189 pid_list = rcu_dereference_sched(tr->function_no_pids);
8190 trace_filter_add_remove_task(pid_list, NULL, task);
8191 }
8192
ftrace_pid_follow_fork(struct trace_array * tr,bool enable)8193 void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
8194 {
8195 if (enable) {
8196 register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
8197 tr);
8198 register_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
8199 tr);
8200 } else {
8201 unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
8202 tr);
8203 unregister_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
8204 tr);
8205 }
8206 }
8207
clear_ftrace_pids(struct trace_array * tr,int type)8208 static void clear_ftrace_pids(struct trace_array *tr, int type)
8209 {
8210 struct trace_pid_list *pid_list;
8211 struct trace_pid_list *no_pid_list;
8212 int cpu;
8213
8214 pid_list = rcu_dereference_protected(tr->function_pids,
8215 lockdep_is_held(&ftrace_lock));
8216 no_pid_list = rcu_dereference_protected(tr->function_no_pids,
8217 lockdep_is_held(&ftrace_lock));
8218
8219 /* Make sure there's something to do */
8220 if (!pid_type_enabled(type, pid_list, no_pid_list))
8221 return;
8222
8223 /* See if the pids still need to be checked after this */
8224 if (!still_need_pid_events(type, pid_list, no_pid_list)) {
8225 unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
8226 for_each_possible_cpu(cpu)
8227 per_cpu_ptr(tr->array_buffer.data, cpu)->ftrace_ignore_pid = FTRACE_PID_TRACE;
8228 }
8229
8230 if (type & TRACE_PIDS)
8231 rcu_assign_pointer(tr->function_pids, NULL);
8232
8233 if (type & TRACE_NO_PIDS)
8234 rcu_assign_pointer(tr->function_no_pids, NULL);
8235
8236 /* Wait till all users are no longer using pid filtering */
8237 synchronize_rcu();
8238
8239 if ((type & TRACE_PIDS) && pid_list)
8240 trace_pid_list_free(pid_list);
8241
8242 if ((type & TRACE_NO_PIDS) && no_pid_list)
8243 trace_pid_list_free(no_pid_list);
8244 }
8245
ftrace_clear_pids(struct trace_array * tr)8246 void ftrace_clear_pids(struct trace_array *tr)
8247 {
8248 mutex_lock(&ftrace_lock);
8249
8250 clear_ftrace_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
8251
8252 mutex_unlock(&ftrace_lock);
8253 }
8254
ftrace_pid_reset(struct trace_array * tr,int type)8255 static void ftrace_pid_reset(struct trace_array *tr, int type)
8256 {
8257 mutex_lock(&ftrace_lock);
8258 clear_ftrace_pids(tr, type);
8259
8260 ftrace_update_pid_func();
8261 ftrace_startup_all(0);
8262
8263 mutex_unlock(&ftrace_lock);
8264 }
8265
8266 /* Greater than any max PID */
8267 #define FTRACE_NO_PIDS (void *)(PID_MAX_LIMIT + 1)
8268
fpid_start(struct seq_file * m,loff_t * pos)8269 static void *fpid_start(struct seq_file *m, loff_t *pos)
8270 __acquires(RCU)
8271 {
8272 struct trace_pid_list *pid_list;
8273 struct trace_array *tr = m->private;
8274
8275 mutex_lock(&ftrace_lock);
8276 rcu_read_lock_sched();
8277
8278 pid_list = rcu_dereference_sched(tr->function_pids);
8279
8280 if (!pid_list)
8281 return !(*pos) ? FTRACE_NO_PIDS : NULL;
8282
8283 return trace_pid_start(pid_list, pos);
8284 }
8285
fpid_next(struct seq_file * m,void * v,loff_t * pos)8286 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
8287 {
8288 struct trace_array *tr = m->private;
8289 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
8290
8291 if (v == FTRACE_NO_PIDS) {
8292 (*pos)++;
8293 return NULL;
8294 }
8295 return trace_pid_next(pid_list, v, pos);
8296 }
8297
fpid_stop(struct seq_file * m,void * p)8298 static void fpid_stop(struct seq_file *m, void *p)
8299 __releases(RCU)
8300 {
8301 rcu_read_unlock_sched();
8302 mutex_unlock(&ftrace_lock);
8303 }
8304
fpid_show(struct seq_file * m,void * v)8305 static int fpid_show(struct seq_file *m, void *v)
8306 {
8307 if (v == FTRACE_NO_PIDS) {
8308 seq_puts(m, "no pid\n");
8309 return 0;
8310 }
8311
8312 return trace_pid_show(m, v);
8313 }
8314
8315 static const struct seq_operations ftrace_pid_sops = {
8316 .start = fpid_start,
8317 .next = fpid_next,
8318 .stop = fpid_stop,
8319 .show = fpid_show,
8320 };
8321
fnpid_start(struct seq_file * m,loff_t * pos)8322 static void *fnpid_start(struct seq_file *m, loff_t *pos)
8323 __acquires(RCU)
8324 {
8325 struct trace_pid_list *pid_list;
8326 struct trace_array *tr = m->private;
8327
8328 mutex_lock(&ftrace_lock);
8329 rcu_read_lock_sched();
8330
8331 pid_list = rcu_dereference_sched(tr->function_no_pids);
8332
8333 if (!pid_list)
8334 return !(*pos) ? FTRACE_NO_PIDS : NULL;
8335
8336 return trace_pid_start(pid_list, pos);
8337 }
8338
fnpid_next(struct seq_file * m,void * v,loff_t * pos)8339 static void *fnpid_next(struct seq_file *m, void *v, loff_t *pos)
8340 {
8341 struct trace_array *tr = m->private;
8342 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_no_pids);
8343
8344 if (v == FTRACE_NO_PIDS) {
8345 (*pos)++;
8346 return NULL;
8347 }
8348 return trace_pid_next(pid_list, v, pos);
8349 }
8350
8351 static const struct seq_operations ftrace_no_pid_sops = {
8352 .start = fnpid_start,
8353 .next = fnpid_next,
8354 .stop = fpid_stop,
8355 .show = fpid_show,
8356 };
8357
pid_open(struct inode * inode,struct file * file,int type)8358 static int pid_open(struct inode *inode, struct file *file, int type)
8359 {
8360 const struct seq_operations *seq_ops;
8361 struct trace_array *tr = inode->i_private;
8362 struct seq_file *m;
8363 int ret = 0;
8364
8365 ret = tracing_check_open_get_tr(tr);
8366 if (ret)
8367 return ret;
8368
8369 if ((file->f_mode & FMODE_WRITE) &&
8370 (file->f_flags & O_TRUNC))
8371 ftrace_pid_reset(tr, type);
8372
8373 switch (type) {
8374 case TRACE_PIDS:
8375 seq_ops = &ftrace_pid_sops;
8376 break;
8377 case TRACE_NO_PIDS:
8378 seq_ops = &ftrace_no_pid_sops;
8379 break;
8380 default:
8381 trace_array_put(tr);
8382 WARN_ON_ONCE(1);
8383 return -EINVAL;
8384 }
8385
8386 ret = seq_open(file, seq_ops);
8387 if (ret < 0) {
8388 trace_array_put(tr);
8389 } else {
8390 m = file->private_data;
8391 /* copy tr over to seq ops */
8392 m->private = tr;
8393 }
8394
8395 return ret;
8396 }
8397
8398 static int
ftrace_pid_open(struct inode * inode,struct file * file)8399 ftrace_pid_open(struct inode *inode, struct file *file)
8400 {
8401 return pid_open(inode, file, TRACE_PIDS);
8402 }
8403
8404 static int
ftrace_no_pid_open(struct inode * inode,struct file * file)8405 ftrace_no_pid_open(struct inode *inode, struct file *file)
8406 {
8407 return pid_open(inode, file, TRACE_NO_PIDS);
8408 }
8409
ignore_task_cpu(void * data)8410 static void ignore_task_cpu(void *data)
8411 {
8412 struct trace_array *tr = data;
8413 struct trace_pid_list *pid_list;
8414 struct trace_pid_list *no_pid_list;
8415
8416 /*
8417 * This function is called by on_each_cpu() while the
8418 * event_mutex is held.
8419 */
8420 pid_list = rcu_dereference_protected(tr->function_pids,
8421 mutex_is_locked(&ftrace_lock));
8422 no_pid_list = rcu_dereference_protected(tr->function_no_pids,
8423 mutex_is_locked(&ftrace_lock));
8424
8425 if (trace_ignore_this_task(pid_list, no_pid_list, current))
8426 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8427 FTRACE_PID_IGNORE);
8428 else
8429 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8430 current->pid);
8431 }
8432
8433 static ssize_t
pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos,int type)8434 pid_write(struct file *filp, const char __user *ubuf,
8435 size_t cnt, loff_t *ppos, int type)
8436 {
8437 struct seq_file *m = filp->private_data;
8438 struct trace_array *tr = m->private;
8439 struct trace_pid_list *filtered_pids;
8440 struct trace_pid_list *other_pids;
8441 struct trace_pid_list *pid_list;
8442 ssize_t ret;
8443
8444 if (!cnt)
8445 return 0;
8446
8447 guard(mutex)(&ftrace_lock);
8448
8449 switch (type) {
8450 case TRACE_PIDS:
8451 filtered_pids = rcu_dereference_protected(tr->function_pids,
8452 lockdep_is_held(&ftrace_lock));
8453 other_pids = rcu_dereference_protected(tr->function_no_pids,
8454 lockdep_is_held(&ftrace_lock));
8455 break;
8456 case TRACE_NO_PIDS:
8457 filtered_pids = rcu_dereference_protected(tr->function_no_pids,
8458 lockdep_is_held(&ftrace_lock));
8459 other_pids = rcu_dereference_protected(tr->function_pids,
8460 lockdep_is_held(&ftrace_lock));
8461 break;
8462 default:
8463 WARN_ON_ONCE(1);
8464 return -EINVAL;
8465 }
8466
8467 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
8468 if (ret < 0)
8469 return ret;
8470
8471 switch (type) {
8472 case TRACE_PIDS:
8473 rcu_assign_pointer(tr->function_pids, pid_list);
8474 break;
8475 case TRACE_NO_PIDS:
8476 rcu_assign_pointer(tr->function_no_pids, pid_list);
8477 break;
8478 }
8479
8480
8481 if (filtered_pids) {
8482 synchronize_rcu();
8483 trace_pid_list_free(filtered_pids);
8484 } else if (pid_list && !other_pids) {
8485 /* Register a probe to set whether to ignore the tracing of a task */
8486 register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
8487 }
8488
8489 /*
8490 * Ignoring of pids is done at task switch. But we have to
8491 * check for those tasks that are currently running.
8492 * Always do this in case a pid was appended or removed.
8493 */
8494 on_each_cpu(ignore_task_cpu, tr, 1);
8495
8496 ftrace_update_pid_func();
8497 ftrace_startup_all(0);
8498
8499 *ppos += ret;
8500
8501 return ret;
8502 }
8503
8504 static ssize_t
ftrace_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)8505 ftrace_pid_write(struct file *filp, const char __user *ubuf,
8506 size_t cnt, loff_t *ppos)
8507 {
8508 return pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
8509 }
8510
8511 static ssize_t
ftrace_no_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)8512 ftrace_no_pid_write(struct file *filp, const char __user *ubuf,
8513 size_t cnt, loff_t *ppos)
8514 {
8515 return pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
8516 }
8517
8518 static int
ftrace_pid_release(struct inode * inode,struct file * file)8519 ftrace_pid_release(struct inode *inode, struct file *file)
8520 {
8521 struct trace_array *tr = inode->i_private;
8522
8523 trace_array_put(tr);
8524
8525 return seq_release(inode, file);
8526 }
8527
8528 static const struct file_operations ftrace_pid_fops = {
8529 .open = ftrace_pid_open,
8530 .write = ftrace_pid_write,
8531 .read = seq_read,
8532 .llseek = tracing_lseek,
8533 .release = ftrace_pid_release,
8534 };
8535
8536 static const struct file_operations ftrace_no_pid_fops = {
8537 .open = ftrace_no_pid_open,
8538 .write = ftrace_no_pid_write,
8539 .read = seq_read,
8540 .llseek = tracing_lseek,
8541 .release = ftrace_pid_release,
8542 };
8543
ftrace_init_tracefs(struct trace_array * tr,struct dentry * d_tracer)8544 void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
8545 {
8546 trace_create_file("set_ftrace_pid", TRACE_MODE_WRITE, d_tracer,
8547 tr, &ftrace_pid_fops);
8548 trace_create_file("set_ftrace_notrace_pid", TRACE_MODE_WRITE,
8549 d_tracer, tr, &ftrace_no_pid_fops);
8550 }
8551
ftrace_init_tracefs_toplevel(struct trace_array * tr,struct dentry * d_tracer)8552 void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
8553 struct dentry *d_tracer)
8554 {
8555 /* Only the top level directory has the dyn_tracefs and profile */
8556 WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
8557
8558 ftrace_init_dyn_tracefs(d_tracer);
8559 ftrace_profile_tracefs(d_tracer);
8560 }
8561
8562 /**
8563 * ftrace_kill - kill ftrace
8564 *
8565 * This function should be used by panic code. It stops ftrace
8566 * but in a not so nice way. If you need to simply kill ftrace
8567 * from a non-atomic section, use ftrace_kill.
8568 */
ftrace_kill(void)8569 void ftrace_kill(void)
8570 {
8571 ftrace_disabled = 1;
8572 ftrace_enabled = 0;
8573 ftrace_trace_function = ftrace_stub;
8574 kprobe_ftrace_kill();
8575 }
8576
8577 /**
8578 * ftrace_is_dead - Test if ftrace is dead or not.
8579 *
8580 * Returns: 1 if ftrace is "dead", zero otherwise.
8581 */
ftrace_is_dead(void)8582 int ftrace_is_dead(void)
8583 {
8584 return ftrace_disabled;
8585 }
8586
8587 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
8588 /*
8589 * When registering ftrace_ops with IPMODIFY, it is necessary to make sure
8590 * it doesn't conflict with any direct ftrace_ops. If there is existing
8591 * direct ftrace_ops on a kernel function being patched, call
8592 * FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_PEER on it to enable sharing.
8593 *
8594 * @ops: ftrace_ops being registered.
8595 *
8596 * Returns:
8597 * 0 on success;
8598 * Negative on failure.
8599 */
prepare_direct_functions_for_ipmodify(struct ftrace_ops * ops)8600 static int prepare_direct_functions_for_ipmodify(struct ftrace_ops *ops)
8601 {
8602 struct ftrace_func_entry *entry;
8603 struct ftrace_hash *hash;
8604 struct ftrace_ops *op;
8605 int size, i, ret;
8606
8607 lockdep_assert_held_once(&direct_mutex);
8608
8609 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
8610 return 0;
8611
8612 hash = ops->func_hash->filter_hash;
8613 size = 1 << hash->size_bits;
8614 for (i = 0; i < size; i++) {
8615 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
8616 unsigned long ip = entry->ip;
8617 bool found_op = false;
8618
8619 mutex_lock(&ftrace_lock);
8620 do_for_each_ftrace_op(op, ftrace_ops_list) {
8621 if (!(op->flags & FTRACE_OPS_FL_DIRECT))
8622 continue;
8623 if (ops_references_ip(op, ip)) {
8624 found_op = true;
8625 break;
8626 }
8627 } while_for_each_ftrace_op(op);
8628 mutex_unlock(&ftrace_lock);
8629
8630 if (found_op) {
8631 if (!op->ops_func)
8632 return -EBUSY;
8633
8634 ret = op->ops_func(op, FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_PEER);
8635 if (ret)
8636 return ret;
8637 }
8638 }
8639 }
8640
8641 return 0;
8642 }
8643
8644 /*
8645 * Similar to prepare_direct_functions_for_ipmodify, clean up after ops
8646 * with IPMODIFY is unregistered. The cleanup is optional for most DIRECT
8647 * ops.
8648 */
cleanup_direct_functions_after_ipmodify(struct ftrace_ops * ops)8649 static void cleanup_direct_functions_after_ipmodify(struct ftrace_ops *ops)
8650 {
8651 struct ftrace_func_entry *entry;
8652 struct ftrace_hash *hash;
8653 struct ftrace_ops *op;
8654 int size, i;
8655
8656 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
8657 return;
8658
8659 mutex_lock(&direct_mutex);
8660
8661 hash = ops->func_hash->filter_hash;
8662 size = 1 << hash->size_bits;
8663 for (i = 0; i < size; i++) {
8664 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
8665 unsigned long ip = entry->ip;
8666 bool found_op = false;
8667
8668 mutex_lock(&ftrace_lock);
8669 do_for_each_ftrace_op(op, ftrace_ops_list) {
8670 if (!(op->flags & FTRACE_OPS_FL_DIRECT))
8671 continue;
8672 if (ops_references_ip(op, ip)) {
8673 found_op = true;
8674 break;
8675 }
8676 } while_for_each_ftrace_op(op);
8677 mutex_unlock(&ftrace_lock);
8678
8679 /* The cleanup is optional, ignore any errors */
8680 if (found_op && op->ops_func)
8681 op->ops_func(op, FTRACE_OPS_CMD_DISABLE_SHARE_IPMODIFY_PEER);
8682 }
8683 }
8684 mutex_unlock(&direct_mutex);
8685 }
8686
8687 #define lock_direct_mutex() mutex_lock(&direct_mutex)
8688 #define unlock_direct_mutex() mutex_unlock(&direct_mutex)
8689
8690 #else /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
8691
prepare_direct_functions_for_ipmodify(struct ftrace_ops * ops)8692 static int prepare_direct_functions_for_ipmodify(struct ftrace_ops *ops)
8693 {
8694 return 0;
8695 }
8696
cleanup_direct_functions_after_ipmodify(struct ftrace_ops * ops)8697 static void cleanup_direct_functions_after_ipmodify(struct ftrace_ops *ops)
8698 {
8699 }
8700
8701 #define lock_direct_mutex() do { } while (0)
8702 #define unlock_direct_mutex() do { } while (0)
8703
8704 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
8705
8706 /*
8707 * Similar to register_ftrace_function, except we don't lock direct_mutex.
8708 */
register_ftrace_function_nolock(struct ftrace_ops * ops)8709 static int register_ftrace_function_nolock(struct ftrace_ops *ops)
8710 {
8711 int ret;
8712
8713 ftrace_ops_init(ops);
8714
8715 mutex_lock(&ftrace_lock);
8716
8717 ret = ftrace_startup(ops, 0);
8718
8719 mutex_unlock(&ftrace_lock);
8720
8721 return ret;
8722 }
8723
8724 /**
8725 * register_ftrace_function - register a function for profiling
8726 * @ops: ops structure that holds the function for profiling.
8727 *
8728 * Register a function to be called by all functions in the
8729 * kernel.
8730 *
8731 * Note: @ops->func and all the functions it calls must be labeled
8732 * with "notrace", otherwise it will go into a
8733 * recursive loop.
8734 */
register_ftrace_function(struct ftrace_ops * ops)8735 int register_ftrace_function(struct ftrace_ops *ops)
8736 {
8737 int ret;
8738
8739 lock_direct_mutex();
8740 ret = prepare_direct_functions_for_ipmodify(ops);
8741 if (ret < 0)
8742 goto out_unlock;
8743
8744 ret = register_ftrace_function_nolock(ops);
8745
8746 out_unlock:
8747 unlock_direct_mutex();
8748 return ret;
8749 }
8750 EXPORT_SYMBOL_GPL(register_ftrace_function);
8751
8752 /**
8753 * unregister_ftrace_function - unregister a function for profiling.
8754 * @ops: ops structure that holds the function to unregister
8755 *
8756 * Unregister a function that was added to be called by ftrace profiling.
8757 */
unregister_ftrace_function(struct ftrace_ops * ops)8758 int unregister_ftrace_function(struct ftrace_ops *ops)
8759 {
8760 int ret;
8761
8762 mutex_lock(&ftrace_lock);
8763 ret = ftrace_shutdown(ops, 0);
8764 mutex_unlock(&ftrace_lock);
8765
8766 cleanup_direct_functions_after_ipmodify(ops);
8767 return ret;
8768 }
8769 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
8770
symbols_cmp(const void * a,const void * b)8771 static int symbols_cmp(const void *a, const void *b)
8772 {
8773 const char **str_a = (const char **) a;
8774 const char **str_b = (const char **) b;
8775
8776 return strcmp(*str_a, *str_b);
8777 }
8778
8779 struct kallsyms_data {
8780 unsigned long *addrs;
8781 const char **syms;
8782 size_t cnt;
8783 size_t found;
8784 };
8785
8786 /* This function gets called for all kernel and module symbols
8787 * and returns 1 in case we resolved all the requested symbols,
8788 * 0 otherwise.
8789 */
kallsyms_callback(void * data,const char * name,unsigned long addr)8790 static int kallsyms_callback(void *data, const char *name, unsigned long addr)
8791 {
8792 struct kallsyms_data *args = data;
8793 const char **sym;
8794 int idx;
8795
8796 sym = bsearch(&name, args->syms, args->cnt, sizeof(*args->syms), symbols_cmp);
8797 if (!sym)
8798 return 0;
8799
8800 idx = sym - args->syms;
8801 if (args->addrs[idx])
8802 return 0;
8803
8804 if (!ftrace_location(addr))
8805 return 0;
8806
8807 args->addrs[idx] = addr;
8808 args->found++;
8809 return args->found == args->cnt ? 1 : 0;
8810 }
8811
8812 /**
8813 * ftrace_lookup_symbols - Lookup addresses for array of symbols
8814 *
8815 * @sorted_syms: array of symbols pointers symbols to resolve,
8816 * must be alphabetically sorted
8817 * @cnt: number of symbols/addresses in @syms/@addrs arrays
8818 * @addrs: array for storing resulting addresses
8819 *
8820 * This function looks up addresses for array of symbols provided in
8821 * @syms array (must be alphabetically sorted) and stores them in
8822 * @addrs array, which needs to be big enough to store at least @cnt
8823 * addresses.
8824 *
8825 * Returns: 0 if all provided symbols are found, -ESRCH otherwise.
8826 */
ftrace_lookup_symbols(const char ** sorted_syms,size_t cnt,unsigned long * addrs)8827 int ftrace_lookup_symbols(const char **sorted_syms, size_t cnt, unsigned long *addrs)
8828 {
8829 struct kallsyms_data args;
8830 int found_all;
8831
8832 memset(addrs, 0, sizeof(*addrs) * cnt);
8833 args.addrs = addrs;
8834 args.syms = sorted_syms;
8835 args.cnt = cnt;
8836 args.found = 0;
8837
8838 found_all = kallsyms_on_each_symbol(kallsyms_callback, &args);
8839 if (found_all)
8840 return 0;
8841 found_all = module_kallsyms_on_each_symbol(NULL, kallsyms_callback, &args);
8842 return found_all ? 0 : -ESRCH;
8843 }
8844
8845 #ifdef CONFIG_SYSCTL
8846
8847 #ifdef CONFIG_DYNAMIC_FTRACE
ftrace_startup_sysctl(void)8848 static void ftrace_startup_sysctl(void)
8849 {
8850 int command;
8851
8852 if (unlikely(ftrace_disabled))
8853 return;
8854
8855 /* Force update next time */
8856 saved_ftrace_func = NULL;
8857 /* ftrace_start_up is true if we want ftrace running */
8858 if (ftrace_start_up) {
8859 command = FTRACE_UPDATE_CALLS;
8860 if (ftrace_graph_active)
8861 command |= FTRACE_START_FUNC_RET;
8862 ftrace_startup_enable(command);
8863 }
8864 }
8865
ftrace_shutdown_sysctl(void)8866 static void ftrace_shutdown_sysctl(void)
8867 {
8868 int command;
8869
8870 if (unlikely(ftrace_disabled))
8871 return;
8872
8873 /* ftrace_start_up is true if ftrace is running */
8874 if (ftrace_start_up) {
8875 command = FTRACE_DISABLE_CALLS;
8876 if (ftrace_graph_active)
8877 command |= FTRACE_STOP_FUNC_RET;
8878 ftrace_run_update_code(command);
8879 }
8880 }
8881 #else
8882 # define ftrace_startup_sysctl() do { } while (0)
8883 # define ftrace_shutdown_sysctl() do { } while (0)
8884 #endif /* CONFIG_DYNAMIC_FTRACE */
8885
is_permanent_ops_registered(void)8886 static bool is_permanent_ops_registered(void)
8887 {
8888 struct ftrace_ops *op;
8889
8890 do_for_each_ftrace_op(op, ftrace_ops_list) {
8891 if (op->flags & FTRACE_OPS_FL_PERMANENT)
8892 return true;
8893 } while_for_each_ftrace_op(op);
8894
8895 return false;
8896 }
8897
8898 static int
ftrace_enable_sysctl(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)8899 ftrace_enable_sysctl(const struct ctl_table *table, int write,
8900 void *buffer, size_t *lenp, loff_t *ppos)
8901 {
8902 int ret;
8903
8904 guard(mutex)(&ftrace_lock);
8905
8906 if (unlikely(ftrace_disabled))
8907 return -ENODEV;
8908
8909 ret = proc_dointvec(table, write, buffer, lenp, ppos);
8910
8911 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
8912 return ret;
8913
8914 if (ftrace_enabled) {
8915
8916 /* we are starting ftrace again */
8917 if (rcu_dereference_protected(ftrace_ops_list,
8918 lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
8919 update_ftrace_function();
8920
8921 ftrace_startup_sysctl();
8922
8923 } else {
8924 if (is_permanent_ops_registered()) {
8925 ftrace_enabled = true;
8926 return -EBUSY;
8927 }
8928
8929 /* stopping ftrace calls (just send to ftrace_stub) */
8930 ftrace_trace_function = ftrace_stub;
8931
8932 ftrace_shutdown_sysctl();
8933 }
8934
8935 last_ftrace_enabled = !!ftrace_enabled;
8936 return 0;
8937 }
8938
8939 static const struct ctl_table ftrace_sysctls[] = {
8940 {
8941 .procname = "ftrace_enabled",
8942 .data = &ftrace_enabled,
8943 .maxlen = sizeof(int),
8944 .mode = 0644,
8945 .proc_handler = ftrace_enable_sysctl,
8946 },
8947 };
8948
ftrace_sysctl_init(void)8949 static int __init ftrace_sysctl_init(void)
8950 {
8951 register_sysctl_init("kernel", ftrace_sysctls);
8952 return 0;
8953 }
8954 late_initcall(ftrace_sysctl_init);
8955 #endif
8956