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,bool update_target)1972 static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1973 struct ftrace_hash *old_hash,
1974 struct ftrace_hash *new_hash,
1975 bool update_target)
1976 {
1977 struct ftrace_page *pg;
1978 struct dyn_ftrace *rec, *end = NULL;
1979 int in_old, in_new;
1980 bool is_ipmodify, is_direct;
1981
1982 /* Only update if the ops has been registered */
1983 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1984 return 0;
1985
1986 is_ipmodify = ops->flags & FTRACE_OPS_FL_IPMODIFY;
1987 is_direct = ops->flags & FTRACE_OPS_FL_DIRECT;
1988
1989 /* neither IPMODIFY nor DIRECT, skip */
1990 if (!is_ipmodify && !is_direct)
1991 return 0;
1992
1993 if (WARN_ON_ONCE(is_ipmodify && is_direct))
1994 return 0;
1995
1996 /*
1997 * Since the IPMODIFY and DIRECT are very address sensitive
1998 * actions, we do not allow ftrace_ops to set all functions to new
1999 * hash.
2000 */
2001 if (!new_hash || !old_hash)
2002 return -EINVAL;
2003
2004 /* Update rec->flags */
2005 do_for_each_ftrace_rec(pg, rec) {
2006
2007 if (rec->flags & FTRACE_FL_DISABLED)
2008 continue;
2009
2010 /*
2011 * Unless we are updating the target of a direct function,
2012 * we only need to update differences of filter_hash
2013 */
2014 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
2015 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
2016 if (!update_target && (in_old == in_new))
2017 continue;
2018
2019 if (in_new) {
2020 if (rec->flags & FTRACE_FL_IPMODIFY) {
2021 int ret;
2022
2023 /* Cannot have two ipmodify on same rec */
2024 if (is_ipmodify)
2025 goto rollback;
2026
2027 /*
2028 * If this is called by __modify_ftrace_direct()
2029 * then it is only changing where the direct
2030 * pointer is jumping to, and the record already
2031 * points to a direct trampoline. If it isn't,
2032 * then it is a bug to update ipmodify on a direct
2033 * caller.
2034 */
2035 FTRACE_WARN_ON(!update_target &&
2036 (rec->flags & FTRACE_FL_DIRECT));
2037
2038 /*
2039 * Another ops with IPMODIFY is already
2040 * attached. We are now attaching a direct
2041 * ops. Run SHARE_IPMODIFY_SELF, to check
2042 * whether sharing is supported.
2043 */
2044 if (!ops->ops_func)
2045 return -EBUSY;
2046 ret = ops->ops_func(ops, FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_SELF);
2047 if (ret)
2048 return ret;
2049 } else if (is_ipmodify) {
2050 rec->flags |= FTRACE_FL_IPMODIFY;
2051 }
2052 } else if (is_ipmodify) {
2053 rec->flags &= ~FTRACE_FL_IPMODIFY;
2054 }
2055 } while_for_each_ftrace_rec();
2056
2057 return 0;
2058
2059 rollback:
2060 end = rec;
2061
2062 /* Roll back what we did above */
2063 do_for_each_ftrace_rec(pg, rec) {
2064
2065 if (rec->flags & FTRACE_FL_DISABLED)
2066 continue;
2067
2068 if (rec == end)
2069 return -EBUSY;
2070
2071 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
2072 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
2073 if (in_old == in_new)
2074 continue;
2075
2076 if (in_new)
2077 rec->flags &= ~FTRACE_FL_IPMODIFY;
2078 else
2079 rec->flags |= FTRACE_FL_IPMODIFY;
2080 } while_for_each_ftrace_rec();
2081
2082 return -EBUSY;
2083 }
2084
ftrace_hash_ipmodify_enable(struct ftrace_ops * ops)2085 static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
2086 {
2087 struct ftrace_hash *hash = ops->func_hash->filter_hash;
2088
2089 if (ftrace_hash_empty(hash))
2090 hash = NULL;
2091
2092 return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash, false);
2093 }
2094
2095 /* Disabling always succeeds */
ftrace_hash_ipmodify_disable(struct ftrace_ops * ops)2096 static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
2097 {
2098 struct ftrace_hash *hash = ops->func_hash->filter_hash;
2099
2100 if (ftrace_hash_empty(hash))
2101 hash = NULL;
2102
2103 __ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH, false);
2104 }
2105
ftrace_hash_ipmodify_update(struct ftrace_ops * ops,struct ftrace_hash * new_hash)2106 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
2107 struct ftrace_hash *new_hash)
2108 {
2109 struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
2110
2111 if (ftrace_hash_empty(old_hash))
2112 old_hash = NULL;
2113
2114 if (ftrace_hash_empty(new_hash))
2115 new_hash = NULL;
2116
2117 return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash, false);
2118 }
2119
print_ip_ins(const char * fmt,const unsigned char * p)2120 static void print_ip_ins(const char *fmt, const unsigned char *p)
2121 {
2122 char ins[MCOUNT_INSN_SIZE];
2123
2124 if (copy_from_kernel_nofault(ins, p, MCOUNT_INSN_SIZE)) {
2125 printk(KERN_CONT "%s[FAULT] %px\n", fmt, p);
2126 return;
2127 }
2128
2129 printk(KERN_CONT "%s", fmt);
2130 pr_cont("%*phC", MCOUNT_INSN_SIZE, ins);
2131 }
2132
2133 enum ftrace_bug_type ftrace_bug_type;
2134 const void *ftrace_expected;
2135
print_bug_type(void)2136 static void print_bug_type(void)
2137 {
2138 switch (ftrace_bug_type) {
2139 case FTRACE_BUG_UNKNOWN:
2140 break;
2141 case FTRACE_BUG_INIT:
2142 pr_info("Initializing ftrace call sites\n");
2143 break;
2144 case FTRACE_BUG_NOP:
2145 pr_info("Setting ftrace call site to NOP\n");
2146 break;
2147 case FTRACE_BUG_CALL:
2148 pr_info("Setting ftrace call site to call ftrace function\n");
2149 break;
2150 case FTRACE_BUG_UPDATE:
2151 pr_info("Updating ftrace call site to call a different ftrace function\n");
2152 break;
2153 }
2154 }
2155
2156 /**
2157 * ftrace_bug - report and shutdown function tracer
2158 * @failed: The failed type (EFAULT, EINVAL, EPERM)
2159 * @rec: The record that failed
2160 *
2161 * The arch code that enables or disables the function tracing
2162 * can call ftrace_bug() when it has detected a problem in
2163 * modifying the code. @failed should be one of either:
2164 * EFAULT - if the problem happens on reading the @ip address
2165 * EINVAL - if what is read at @ip is not what was expected
2166 * EPERM - if the problem happens on writing to the @ip address
2167 */
ftrace_bug(int failed,struct dyn_ftrace * rec)2168 void ftrace_bug(int failed, struct dyn_ftrace *rec)
2169 {
2170 unsigned long ip = rec ? rec->ip : 0;
2171
2172 pr_info("------------[ ftrace bug ]------------\n");
2173
2174 switch (failed) {
2175 case -EFAULT:
2176 pr_info("ftrace faulted on modifying ");
2177 print_ip_sym(KERN_INFO, ip);
2178 break;
2179 case -EINVAL:
2180 pr_info("ftrace failed to modify ");
2181 print_ip_sym(KERN_INFO, ip);
2182 print_ip_ins(" actual: ", (unsigned char *)ip);
2183 pr_cont("\n");
2184 if (ftrace_expected) {
2185 print_ip_ins(" expected: ", ftrace_expected);
2186 pr_cont("\n");
2187 }
2188 break;
2189 case -EPERM:
2190 pr_info("ftrace faulted on writing ");
2191 print_ip_sym(KERN_INFO, ip);
2192 break;
2193 default:
2194 pr_info("ftrace faulted on unknown error ");
2195 print_ip_sym(KERN_INFO, ip);
2196 }
2197 print_bug_type();
2198 if (rec) {
2199 struct ftrace_ops *ops = NULL;
2200
2201 pr_info("ftrace record flags: %lx\n", rec->flags);
2202 pr_cont(" (%ld)%s%s", ftrace_rec_count(rec),
2203 rec->flags & FTRACE_FL_REGS ? " R" : " ",
2204 rec->flags & FTRACE_FL_CALL_OPS ? " O" : " ");
2205 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2206 ops = ftrace_find_tramp_ops_any(rec);
2207 if (ops) {
2208 do {
2209 pr_cont("\ttramp: %pS (%pS)",
2210 (void *)ops->trampoline,
2211 (void *)ops->func);
2212 ops = ftrace_find_tramp_ops_next(rec, ops);
2213 } while (ops);
2214 } else
2215 pr_cont("\ttramp: ERROR!");
2216
2217 }
2218 ip = ftrace_get_addr_curr(rec);
2219 pr_cont("\n expected tramp: %lx\n", ip);
2220 }
2221
2222 FTRACE_WARN_ON_ONCE(1);
2223 }
2224
ftrace_check_record(struct dyn_ftrace * rec,bool enable,bool update)2225 static int ftrace_check_record(struct dyn_ftrace *rec, bool enable, bool update)
2226 {
2227 unsigned long flag = 0UL;
2228
2229 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2230
2231 if (skip_record(rec))
2232 return FTRACE_UPDATE_IGNORE;
2233
2234 /*
2235 * If we are updating calls:
2236 *
2237 * If the record has a ref count, then we need to enable it
2238 * because someone is using it.
2239 *
2240 * Otherwise we make sure its disabled.
2241 *
2242 * If we are disabling calls, then disable all records that
2243 * are enabled.
2244 */
2245 if (enable && ftrace_rec_count(rec))
2246 flag = FTRACE_FL_ENABLED;
2247
2248 /*
2249 * If enabling and the REGS flag does not match the REGS_EN, or
2250 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2251 * this record. Set flags to fail the compare against ENABLED.
2252 * Same for direct calls.
2253 */
2254 if (flag) {
2255 if (!(rec->flags & FTRACE_FL_REGS) !=
2256 !(rec->flags & FTRACE_FL_REGS_EN))
2257 flag |= FTRACE_FL_REGS;
2258
2259 if (!(rec->flags & FTRACE_FL_TRAMP) !=
2260 !(rec->flags & FTRACE_FL_TRAMP_EN))
2261 flag |= FTRACE_FL_TRAMP;
2262
2263 /*
2264 * Direct calls are special, as count matters.
2265 * We must test the record for direct, if the
2266 * DIRECT and DIRECT_EN do not match, but only
2267 * if the count is 1. That's because, if the
2268 * count is something other than one, we do not
2269 * want the direct enabled (it will be done via the
2270 * direct helper). But if DIRECT_EN is set, and
2271 * the count is not one, we need to clear it.
2272 *
2273 */
2274 if (ftrace_rec_count(rec) == 1) {
2275 if (!(rec->flags & FTRACE_FL_DIRECT) !=
2276 !(rec->flags & FTRACE_FL_DIRECT_EN))
2277 flag |= FTRACE_FL_DIRECT;
2278 } else if (rec->flags & FTRACE_FL_DIRECT_EN) {
2279 flag |= FTRACE_FL_DIRECT;
2280 }
2281
2282 /*
2283 * Ops calls are special, as count matters.
2284 * As with direct calls, they must only be enabled when count
2285 * is one, otherwise they'll be handled via the list ops.
2286 */
2287 if (ftrace_rec_count(rec) == 1) {
2288 if (!(rec->flags & FTRACE_FL_CALL_OPS) !=
2289 !(rec->flags & FTRACE_FL_CALL_OPS_EN))
2290 flag |= FTRACE_FL_CALL_OPS;
2291 } else if (rec->flags & FTRACE_FL_CALL_OPS_EN) {
2292 flag |= FTRACE_FL_CALL_OPS;
2293 }
2294 }
2295
2296 /* If the state of this record hasn't changed, then do nothing */
2297 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2298 return FTRACE_UPDATE_IGNORE;
2299
2300 if (flag) {
2301 /* Save off if rec is being enabled (for return value) */
2302 flag ^= rec->flags & FTRACE_FL_ENABLED;
2303
2304 if (update) {
2305 rec->flags |= FTRACE_FL_ENABLED | FTRACE_FL_TOUCHED;
2306 if (flag & FTRACE_FL_REGS) {
2307 if (rec->flags & FTRACE_FL_REGS)
2308 rec->flags |= FTRACE_FL_REGS_EN;
2309 else
2310 rec->flags &= ~FTRACE_FL_REGS_EN;
2311 }
2312 if (flag & FTRACE_FL_TRAMP) {
2313 if (rec->flags & FTRACE_FL_TRAMP)
2314 rec->flags |= FTRACE_FL_TRAMP_EN;
2315 else
2316 rec->flags &= ~FTRACE_FL_TRAMP_EN;
2317 }
2318
2319 /* Keep track of anything that modifies the function */
2320 if (rec->flags & (FTRACE_FL_DIRECT | FTRACE_FL_IPMODIFY))
2321 rec->flags |= FTRACE_FL_MODIFIED;
2322
2323 if (flag & FTRACE_FL_DIRECT) {
2324 /*
2325 * If there's only one user (direct_ops helper)
2326 * then we can call the direct function
2327 * directly (no ftrace trampoline).
2328 */
2329 if (ftrace_rec_count(rec) == 1) {
2330 if (rec->flags & FTRACE_FL_DIRECT)
2331 rec->flags |= FTRACE_FL_DIRECT_EN;
2332 else
2333 rec->flags &= ~FTRACE_FL_DIRECT_EN;
2334 } else {
2335 /*
2336 * Can only call directly if there's
2337 * only one callback to the function.
2338 */
2339 rec->flags &= ~FTRACE_FL_DIRECT_EN;
2340 }
2341 }
2342
2343 if (flag & FTRACE_FL_CALL_OPS) {
2344 if (ftrace_rec_count(rec) == 1) {
2345 if (rec->flags & FTRACE_FL_CALL_OPS)
2346 rec->flags |= FTRACE_FL_CALL_OPS_EN;
2347 else
2348 rec->flags &= ~FTRACE_FL_CALL_OPS_EN;
2349 } else {
2350 /*
2351 * Can only call directly if there's
2352 * only one set of associated ops.
2353 */
2354 rec->flags &= ~FTRACE_FL_CALL_OPS_EN;
2355 }
2356 }
2357 }
2358
2359 /*
2360 * If this record is being updated from a nop, then
2361 * return UPDATE_MAKE_CALL.
2362 * Otherwise,
2363 * return UPDATE_MODIFY_CALL to tell the caller to convert
2364 * from the save regs, to a non-save regs function or
2365 * vice versa, or from a trampoline call.
2366 */
2367 if (flag & FTRACE_FL_ENABLED) {
2368 ftrace_bug_type = FTRACE_BUG_CALL;
2369 return FTRACE_UPDATE_MAKE_CALL;
2370 }
2371
2372 ftrace_bug_type = FTRACE_BUG_UPDATE;
2373 return FTRACE_UPDATE_MODIFY_CALL;
2374 }
2375
2376 if (update) {
2377 /* If there's no more users, clear all flags */
2378 if (!ftrace_rec_count(rec))
2379 rec->flags &= FTRACE_NOCLEAR_FLAGS;
2380 else
2381 /*
2382 * Just disable the record, but keep the ops TRAMP
2383 * and REGS states. The _EN flags must be disabled though.
2384 */
2385 rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2386 FTRACE_FL_REGS_EN | FTRACE_FL_DIRECT_EN |
2387 FTRACE_FL_CALL_OPS_EN);
2388 }
2389
2390 ftrace_bug_type = FTRACE_BUG_NOP;
2391 return FTRACE_UPDATE_MAKE_NOP;
2392 }
2393
2394 /**
2395 * ftrace_update_record - set a record that now is tracing or not
2396 * @rec: the record to update
2397 * @enable: set to true if the record is tracing, false to force disable
2398 *
2399 * The records that represent all functions that can be traced need
2400 * to be updated when tracing has been enabled.
2401 */
ftrace_update_record(struct dyn_ftrace * rec,bool enable)2402 int ftrace_update_record(struct dyn_ftrace *rec, bool enable)
2403 {
2404 return ftrace_check_record(rec, enable, true);
2405 }
2406
2407 /**
2408 * ftrace_test_record - check if the record has been enabled or not
2409 * @rec: the record to test
2410 * @enable: set to true to check if enabled, false if it is disabled
2411 *
2412 * The arch code may need to test if a record is already set to
2413 * tracing to determine how to modify the function code that it
2414 * represents.
2415 */
ftrace_test_record(struct dyn_ftrace * rec,bool enable)2416 int ftrace_test_record(struct dyn_ftrace *rec, bool enable)
2417 {
2418 return ftrace_check_record(rec, enable, false);
2419 }
2420
2421 static struct ftrace_ops *
ftrace_find_tramp_ops_any(struct dyn_ftrace * rec)2422 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2423 {
2424 struct ftrace_ops *op;
2425 unsigned long ip = rec->ip;
2426
2427 do_for_each_ftrace_op(op, ftrace_ops_list) {
2428
2429 if (!op->trampoline)
2430 continue;
2431
2432 if (hash_contains_ip(ip, op->func_hash))
2433 return op;
2434 } while_for_each_ftrace_op(op);
2435
2436 return NULL;
2437 }
2438
2439 static struct ftrace_ops *
ftrace_find_tramp_ops_any_other(struct dyn_ftrace * rec,struct ftrace_ops * op_exclude)2440 ftrace_find_tramp_ops_any_other(struct dyn_ftrace *rec, struct ftrace_ops *op_exclude)
2441 {
2442 struct ftrace_ops *op;
2443 unsigned long ip = rec->ip;
2444
2445 do_for_each_ftrace_op(op, ftrace_ops_list) {
2446
2447 if (op == op_exclude || !op->trampoline)
2448 continue;
2449
2450 if (hash_contains_ip(ip, op->func_hash))
2451 return op;
2452 } while_for_each_ftrace_op(op);
2453
2454 return NULL;
2455 }
2456
2457 static struct ftrace_ops *
ftrace_find_tramp_ops_next(struct dyn_ftrace * rec,struct ftrace_ops * op)2458 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2459 struct ftrace_ops *op)
2460 {
2461 unsigned long ip = rec->ip;
2462
2463 while_for_each_ftrace_op(op) {
2464
2465 if (!op->trampoline)
2466 continue;
2467
2468 if (hash_contains_ip(ip, op->func_hash))
2469 return op;
2470 }
2471
2472 return NULL;
2473 }
2474
2475 static struct ftrace_ops *
ftrace_find_tramp_ops_curr(struct dyn_ftrace * rec)2476 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2477 {
2478 struct ftrace_ops *op;
2479 unsigned long ip = rec->ip;
2480
2481 /*
2482 * Need to check removed ops first.
2483 * If they are being removed, and this rec has a tramp,
2484 * and this rec is in the ops list, then it would be the
2485 * one with the tramp.
2486 */
2487 if (removed_ops) {
2488 if (hash_contains_ip(ip, &removed_ops->old_hash))
2489 return removed_ops;
2490 }
2491
2492 /*
2493 * Need to find the current trampoline for a rec.
2494 * Now, a trampoline is only attached to a rec if there
2495 * was a single 'ops' attached to it. But this can be called
2496 * when we are adding another op to the rec or removing the
2497 * current one. Thus, if the op is being added, we can
2498 * ignore it because it hasn't attached itself to the rec
2499 * yet.
2500 *
2501 * If an ops is being modified (hooking to different functions)
2502 * then we don't care about the new functions that are being
2503 * added, just the old ones (that are probably being removed).
2504 *
2505 * If we are adding an ops to a function that already is using
2506 * a trampoline, it needs to be removed (trampolines are only
2507 * for single ops connected), then an ops that is not being
2508 * modified also needs to be checked.
2509 */
2510 do_for_each_ftrace_op(op, ftrace_ops_list) {
2511
2512 if (!op->trampoline)
2513 continue;
2514
2515 /*
2516 * If the ops is being added, it hasn't gotten to
2517 * the point to be removed from this tree yet.
2518 */
2519 if (op->flags & FTRACE_OPS_FL_ADDING)
2520 continue;
2521
2522
2523 /*
2524 * If the ops is being modified and is in the old
2525 * hash, then it is probably being removed from this
2526 * function.
2527 */
2528 if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2529 hash_contains_ip(ip, &op->old_hash))
2530 return op;
2531 /*
2532 * If the ops is not being added or modified, and it's
2533 * in its normal filter hash, then this must be the one
2534 * we want!
2535 */
2536 if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2537 hash_contains_ip(ip, op->func_hash))
2538 return op;
2539
2540 } while_for_each_ftrace_op(op);
2541
2542 return NULL;
2543 }
2544
2545 static struct ftrace_ops *
ftrace_find_tramp_ops_new(struct dyn_ftrace * rec)2546 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2547 {
2548 struct ftrace_ops *op;
2549 unsigned long ip = rec->ip;
2550
2551 do_for_each_ftrace_op(op, ftrace_ops_list) {
2552 /* pass rec in as regs to have non-NULL val */
2553 if (hash_contains_ip(ip, op->func_hash))
2554 return op;
2555 } while_for_each_ftrace_op(op);
2556
2557 return NULL;
2558 }
2559
2560 struct ftrace_ops *
ftrace_find_unique_ops(struct dyn_ftrace * rec)2561 ftrace_find_unique_ops(struct dyn_ftrace *rec)
2562 {
2563 struct ftrace_ops *op, *found = NULL;
2564 unsigned long ip = rec->ip;
2565
2566 do_for_each_ftrace_op(op, ftrace_ops_list) {
2567
2568 if (hash_contains_ip(ip, op->func_hash)) {
2569 if (found)
2570 return NULL;
2571 found = op;
2572 }
2573
2574 } while_for_each_ftrace_op(op);
2575
2576 return found;
2577 }
2578
2579 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
2580 /* Protected by rcu_tasks for reading, and direct_mutex for writing */
2581 static struct ftrace_hash __rcu *direct_functions = EMPTY_HASH;
2582 static DEFINE_MUTEX(direct_mutex);
2583
2584 /*
2585 * Search the direct_functions hash to see if the given instruction pointer
2586 * has a direct caller attached to it.
2587 */
ftrace_find_rec_direct(unsigned long ip)2588 unsigned long ftrace_find_rec_direct(unsigned long ip)
2589 {
2590 struct ftrace_func_entry *entry;
2591
2592 entry = __ftrace_lookup_ip(direct_functions, ip);
2593 if (!entry)
2594 return 0;
2595
2596 return entry->direct;
2597 }
2598
call_direct_funcs(unsigned long ip,unsigned long pip,struct ftrace_ops * ops,struct ftrace_regs * fregs)2599 static void call_direct_funcs(unsigned long ip, unsigned long pip,
2600 struct ftrace_ops *ops, struct ftrace_regs *fregs)
2601 {
2602 unsigned long addr = READ_ONCE(ops->direct_call);
2603
2604 if (!addr)
2605 return;
2606
2607 arch_ftrace_set_direct_caller(fregs, addr);
2608 }
2609 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
2610
2611 /**
2612 * ftrace_get_addr_new - Get the call address to set to
2613 * @rec: The ftrace record descriptor
2614 *
2615 * If the record has the FTRACE_FL_REGS set, that means that it
2616 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2617 * is not set, then it wants to convert to the normal callback.
2618 *
2619 * Returns: the address of the trampoline to set to
2620 */
ftrace_get_addr_new(struct dyn_ftrace * rec)2621 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2622 {
2623 struct ftrace_ops *ops;
2624 unsigned long addr;
2625
2626 if ((rec->flags & FTRACE_FL_DIRECT) &&
2627 (ftrace_rec_count(rec) == 1)) {
2628 addr = ftrace_find_rec_direct(rec->ip);
2629 if (addr)
2630 return addr;
2631 WARN_ON_ONCE(1);
2632 }
2633
2634 /* Trampolines take precedence over regs */
2635 if (rec->flags & FTRACE_FL_TRAMP) {
2636 ops = ftrace_find_tramp_ops_new(rec);
2637 if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2638 pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2639 (void *)rec->ip, (void *)rec->ip, rec->flags);
2640 /* Ftrace is shutting down, return anything */
2641 return (unsigned long)FTRACE_ADDR;
2642 }
2643 return ops->trampoline;
2644 }
2645
2646 if (rec->flags & FTRACE_FL_REGS)
2647 return (unsigned long)FTRACE_REGS_ADDR;
2648 else
2649 return (unsigned long)FTRACE_ADDR;
2650 }
2651
2652 /**
2653 * ftrace_get_addr_curr - Get the call address that is already there
2654 * @rec: The ftrace record descriptor
2655 *
2656 * The FTRACE_FL_REGS_EN is set when the record already points to
2657 * a function that saves all the regs. Basically the '_EN' version
2658 * represents the current state of the function.
2659 *
2660 * Returns: the address of the trampoline that is currently being called
2661 */
ftrace_get_addr_curr(struct dyn_ftrace * rec)2662 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2663 {
2664 struct ftrace_ops *ops;
2665 unsigned long addr;
2666
2667 /* Direct calls take precedence over trampolines */
2668 if (rec->flags & FTRACE_FL_DIRECT_EN) {
2669 addr = ftrace_find_rec_direct(rec->ip);
2670 if (addr)
2671 return addr;
2672 WARN_ON_ONCE(1);
2673 }
2674
2675 /* Trampolines take precedence over regs */
2676 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2677 ops = ftrace_find_tramp_ops_curr(rec);
2678 if (FTRACE_WARN_ON(!ops)) {
2679 pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2680 (void *)rec->ip, (void *)rec->ip);
2681 /* Ftrace is shutting down, return anything */
2682 return (unsigned long)FTRACE_ADDR;
2683 }
2684 return ops->trampoline;
2685 }
2686
2687 if (rec->flags & FTRACE_FL_REGS_EN)
2688 return (unsigned long)FTRACE_REGS_ADDR;
2689 else
2690 return (unsigned long)FTRACE_ADDR;
2691 }
2692
2693 static int
__ftrace_replace_code(struct dyn_ftrace * rec,bool enable)2694 __ftrace_replace_code(struct dyn_ftrace *rec, bool enable)
2695 {
2696 unsigned long ftrace_old_addr;
2697 unsigned long ftrace_addr;
2698 int ret;
2699
2700 ftrace_addr = ftrace_get_addr_new(rec);
2701
2702 /* This needs to be done before we call ftrace_update_record */
2703 ftrace_old_addr = ftrace_get_addr_curr(rec);
2704
2705 ret = ftrace_update_record(rec, enable);
2706
2707 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2708
2709 switch (ret) {
2710 case FTRACE_UPDATE_IGNORE:
2711 return 0;
2712
2713 case FTRACE_UPDATE_MAKE_CALL:
2714 ftrace_bug_type = FTRACE_BUG_CALL;
2715 return ftrace_make_call(rec, ftrace_addr);
2716
2717 case FTRACE_UPDATE_MAKE_NOP:
2718 ftrace_bug_type = FTRACE_BUG_NOP;
2719 return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2720
2721 case FTRACE_UPDATE_MODIFY_CALL:
2722 ftrace_bug_type = FTRACE_BUG_UPDATE;
2723 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2724 }
2725
2726 return -1; /* unknown ftrace bug */
2727 }
2728
ftrace_replace_code(int mod_flags)2729 void __weak ftrace_replace_code(int mod_flags)
2730 {
2731 struct dyn_ftrace *rec;
2732 struct ftrace_page *pg;
2733 bool enable = mod_flags & FTRACE_MODIFY_ENABLE_FL;
2734 int schedulable = mod_flags & FTRACE_MODIFY_MAY_SLEEP_FL;
2735 int failed;
2736
2737 if (unlikely(ftrace_disabled))
2738 return;
2739
2740 do_for_each_ftrace_rec(pg, rec) {
2741
2742 if (skip_record(rec))
2743 continue;
2744
2745 failed = __ftrace_replace_code(rec, enable);
2746 if (failed) {
2747 ftrace_bug(failed, rec);
2748 /* Stop processing */
2749 return;
2750 }
2751 if (schedulable)
2752 cond_resched();
2753 } while_for_each_ftrace_rec();
2754 }
2755
2756 struct ftrace_rec_iter {
2757 struct ftrace_page *pg;
2758 int index;
2759 };
2760
2761 /**
2762 * ftrace_rec_iter_start - start up iterating over traced functions
2763 *
2764 * Returns: an iterator handle that is used to iterate over all
2765 * the records that represent address locations where functions
2766 * are traced.
2767 *
2768 * May return NULL if no records are available.
2769 */
ftrace_rec_iter_start(void)2770 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2771 {
2772 /*
2773 * We only use a single iterator.
2774 * Protected by the ftrace_lock mutex.
2775 */
2776 static struct ftrace_rec_iter ftrace_rec_iter;
2777 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2778
2779 iter->pg = ftrace_pages_start;
2780 iter->index = 0;
2781
2782 /* Could have empty pages */
2783 while (iter->pg && !iter->pg->index)
2784 iter->pg = iter->pg->next;
2785
2786 if (!iter->pg)
2787 return NULL;
2788
2789 return iter;
2790 }
2791
2792 /**
2793 * ftrace_rec_iter_next - get the next record to process.
2794 * @iter: The handle to the iterator.
2795 *
2796 * Returns: the next iterator after the given iterator @iter.
2797 */
ftrace_rec_iter_next(struct ftrace_rec_iter * iter)2798 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2799 {
2800 iter->index++;
2801
2802 if (iter->index >= iter->pg->index) {
2803 iter->pg = iter->pg->next;
2804 iter->index = 0;
2805
2806 /* Could have empty pages */
2807 while (iter->pg && !iter->pg->index)
2808 iter->pg = iter->pg->next;
2809 }
2810
2811 if (!iter->pg)
2812 return NULL;
2813
2814 return iter;
2815 }
2816
2817 /**
2818 * ftrace_rec_iter_record - get the record at the iterator location
2819 * @iter: The current iterator location
2820 *
2821 * Returns: the record that the current @iter is at.
2822 */
ftrace_rec_iter_record(struct ftrace_rec_iter * iter)2823 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2824 {
2825 return &iter->pg->records[iter->index];
2826 }
2827
2828 static int
ftrace_nop_initialize(struct module * mod,struct dyn_ftrace * rec)2829 ftrace_nop_initialize(struct module *mod, struct dyn_ftrace *rec)
2830 {
2831 int ret;
2832
2833 if (unlikely(ftrace_disabled))
2834 return 0;
2835
2836 ret = ftrace_init_nop(mod, rec);
2837 if (ret) {
2838 ftrace_bug_type = FTRACE_BUG_INIT;
2839 ftrace_bug(ret, rec);
2840 return 0;
2841 }
2842 return 1;
2843 }
2844
2845 /*
2846 * archs can override this function if they must do something
2847 * before the modifying code is performed.
2848 */
ftrace_arch_code_modify_prepare(void)2849 void __weak ftrace_arch_code_modify_prepare(void)
2850 {
2851 }
2852
2853 /*
2854 * archs can override this function if they must do something
2855 * after the modifying code is performed.
2856 */
ftrace_arch_code_modify_post_process(void)2857 void __weak ftrace_arch_code_modify_post_process(void)
2858 {
2859 }
2860
update_ftrace_func(ftrace_func_t func)2861 static int update_ftrace_func(ftrace_func_t func)
2862 {
2863 static ftrace_func_t save_func;
2864
2865 /* Avoid updating if it hasn't changed */
2866 if (func == save_func)
2867 return 0;
2868
2869 save_func = func;
2870
2871 return ftrace_update_ftrace_func(func);
2872 }
2873
ftrace_modify_all_code(int command)2874 void ftrace_modify_all_code(int command)
2875 {
2876 int update = command & FTRACE_UPDATE_TRACE_FUNC;
2877 int mod_flags = 0;
2878 int err = 0;
2879
2880 if (command & FTRACE_MAY_SLEEP)
2881 mod_flags = FTRACE_MODIFY_MAY_SLEEP_FL;
2882
2883 /*
2884 * If the ftrace_caller calls a ftrace_ops func directly,
2885 * we need to make sure that it only traces functions it
2886 * expects to trace. When doing the switch of functions,
2887 * we need to update to the ftrace_ops_list_func first
2888 * before the transition between old and new calls are set,
2889 * as the ftrace_ops_list_func will check the ops hashes
2890 * to make sure the ops are having the right functions
2891 * traced.
2892 */
2893 if (update) {
2894 err = update_ftrace_func(ftrace_ops_list_func);
2895 if (FTRACE_WARN_ON(err))
2896 return;
2897 }
2898
2899 if (command & FTRACE_UPDATE_CALLS)
2900 ftrace_replace_code(mod_flags | FTRACE_MODIFY_ENABLE_FL);
2901 else if (command & FTRACE_DISABLE_CALLS)
2902 ftrace_replace_code(mod_flags);
2903
2904 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2905 function_trace_op = set_function_trace_op;
2906 smp_wmb();
2907 /* If irqs are disabled, we are in stop machine */
2908 if (!irqs_disabled())
2909 smp_call_function(ftrace_sync_ipi, NULL, 1);
2910 err = update_ftrace_func(ftrace_trace_function);
2911 if (FTRACE_WARN_ON(err))
2912 return;
2913 }
2914
2915 if (command & FTRACE_START_FUNC_RET)
2916 err = ftrace_enable_ftrace_graph_caller();
2917 else if (command & FTRACE_STOP_FUNC_RET)
2918 err = ftrace_disable_ftrace_graph_caller();
2919 FTRACE_WARN_ON(err);
2920 }
2921
__ftrace_modify_code(void * data)2922 static int __ftrace_modify_code(void *data)
2923 {
2924 int *command = data;
2925
2926 ftrace_modify_all_code(*command);
2927
2928 return 0;
2929 }
2930
2931 /**
2932 * ftrace_run_stop_machine - go back to the stop machine method
2933 * @command: The command to tell ftrace what to do
2934 *
2935 * If an arch needs to fall back to the stop machine method, the
2936 * it can call this function.
2937 */
ftrace_run_stop_machine(int command)2938 void ftrace_run_stop_machine(int command)
2939 {
2940 stop_machine(__ftrace_modify_code, &command, NULL);
2941 }
2942
2943 /**
2944 * arch_ftrace_update_code - modify the code to trace or not trace
2945 * @command: The command that needs to be done
2946 *
2947 * Archs can override this function if it does not need to
2948 * run stop_machine() to modify code.
2949 */
arch_ftrace_update_code(int command)2950 void __weak arch_ftrace_update_code(int command)
2951 {
2952 ftrace_run_stop_machine(command);
2953 }
2954
ftrace_run_update_code(int command)2955 static void ftrace_run_update_code(int command)
2956 {
2957 ftrace_arch_code_modify_prepare();
2958
2959 /*
2960 * By default we use stop_machine() to modify the code.
2961 * But archs can do what ever they want as long as it
2962 * is safe. The stop_machine() is the safest, but also
2963 * produces the most overhead.
2964 */
2965 arch_ftrace_update_code(command);
2966
2967 ftrace_arch_code_modify_post_process();
2968 }
2969
ftrace_run_modify_code(struct ftrace_ops * ops,int command,struct ftrace_ops_hash * old_hash)2970 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2971 struct ftrace_ops_hash *old_hash)
2972 {
2973 ops->flags |= FTRACE_OPS_FL_MODIFYING;
2974 ops->old_hash.filter_hash = old_hash->filter_hash;
2975 ops->old_hash.notrace_hash = old_hash->notrace_hash;
2976 ftrace_run_update_code(command);
2977 ops->old_hash.filter_hash = NULL;
2978 ops->old_hash.notrace_hash = NULL;
2979 ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2980 }
2981
2982 static ftrace_func_t saved_ftrace_func;
2983 static int ftrace_start_up;
2984
arch_ftrace_trampoline_free(struct ftrace_ops * ops)2985 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2986 {
2987 }
2988
2989 /* List of trace_ops that have allocated trampolines */
2990 static LIST_HEAD(ftrace_ops_trampoline_list);
2991
ftrace_add_trampoline_to_kallsyms(struct ftrace_ops * ops)2992 static void ftrace_add_trampoline_to_kallsyms(struct ftrace_ops *ops)
2993 {
2994 lockdep_assert_held(&ftrace_lock);
2995 list_add_rcu(&ops->list, &ftrace_ops_trampoline_list);
2996 }
2997
ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops * ops)2998 static void ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops *ops)
2999 {
3000 lockdep_assert_held(&ftrace_lock);
3001 list_del_rcu(&ops->list);
3002 synchronize_rcu();
3003 }
3004
3005 /*
3006 * "__builtin__ftrace" is used as a module name in /proc/kallsyms for symbols
3007 * for pages allocated for ftrace purposes, even though "__builtin__ftrace" is
3008 * not a module.
3009 */
3010 #define FTRACE_TRAMPOLINE_MOD "__builtin__ftrace"
3011 #define FTRACE_TRAMPOLINE_SYM "ftrace_trampoline"
3012
ftrace_trampoline_free(struct ftrace_ops * ops)3013 static void ftrace_trampoline_free(struct ftrace_ops *ops)
3014 {
3015 if (ops && (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP) &&
3016 ops->trampoline) {
3017 /*
3018 * Record the text poke event before the ksymbol unregister
3019 * event.
3020 */
3021 perf_event_text_poke((void *)ops->trampoline,
3022 (void *)ops->trampoline,
3023 ops->trampoline_size, NULL, 0);
3024 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
3025 ops->trampoline, ops->trampoline_size,
3026 true, FTRACE_TRAMPOLINE_SYM);
3027 /* Remove from kallsyms after the perf events */
3028 ftrace_remove_trampoline_from_kallsyms(ops);
3029 }
3030
3031 arch_ftrace_trampoline_free(ops);
3032 }
3033
ftrace_startup_enable(int command)3034 static void ftrace_startup_enable(int command)
3035 {
3036 if (saved_ftrace_func != ftrace_trace_function) {
3037 saved_ftrace_func = ftrace_trace_function;
3038 command |= FTRACE_UPDATE_TRACE_FUNC;
3039 }
3040
3041 if (!command || !ftrace_enabled)
3042 return;
3043
3044 ftrace_run_update_code(command);
3045 }
3046
ftrace_startup_all(int command)3047 static void ftrace_startup_all(int command)
3048 {
3049 update_all_ops = true;
3050 ftrace_startup_enable(command);
3051 update_all_ops = false;
3052 }
3053
ftrace_startup(struct ftrace_ops * ops,int command)3054 int ftrace_startup(struct ftrace_ops *ops, int command)
3055 {
3056 int ret;
3057
3058 if (unlikely(ftrace_disabled))
3059 return -ENODEV;
3060
3061 ret = __register_ftrace_function(ops);
3062 if (ret)
3063 return ret;
3064
3065 ftrace_start_up++;
3066
3067 /*
3068 * Note that ftrace probes uses this to start up
3069 * and modify functions it will probe. But we still
3070 * set the ADDING flag for modification, as probes
3071 * do not have trampolines. If they add them in the
3072 * future, then the probes will need to distinguish
3073 * between adding and updating probes.
3074 */
3075 ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
3076
3077 ret = ftrace_hash_ipmodify_enable(ops);
3078 if (ret < 0) {
3079 /* Rollback registration process */
3080 __unregister_ftrace_function(ops);
3081 ftrace_start_up--;
3082 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
3083 if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
3084 ftrace_trampoline_free(ops);
3085 return ret;
3086 }
3087
3088 if (ftrace_hash_rec_enable(ops))
3089 command |= FTRACE_UPDATE_CALLS;
3090
3091 ftrace_startup_enable(command);
3092
3093 /*
3094 * If ftrace is in an undefined state, we just remove ops from list
3095 * to prevent the NULL pointer, instead of totally rolling it back and
3096 * free trampoline, because those actions could cause further damage.
3097 */
3098 if (unlikely(ftrace_disabled)) {
3099 __unregister_ftrace_function(ops);
3100 return -ENODEV;
3101 }
3102
3103 ops->flags &= ~FTRACE_OPS_FL_ADDING;
3104
3105 return 0;
3106 }
3107
ftrace_shutdown(struct ftrace_ops * ops,int command)3108 int ftrace_shutdown(struct ftrace_ops *ops, int command)
3109 {
3110 int ret;
3111
3112 if (unlikely(ftrace_disabled))
3113 return -ENODEV;
3114
3115 ret = __unregister_ftrace_function(ops);
3116 if (ret)
3117 return ret;
3118
3119 ftrace_start_up--;
3120 /*
3121 * Just warn in case of unbalance, no need to kill ftrace, it's not
3122 * critical but the ftrace_call callers may be never nopped again after
3123 * further ftrace uses.
3124 */
3125 WARN_ON_ONCE(ftrace_start_up < 0);
3126
3127 /* Disabling ipmodify never fails */
3128 ftrace_hash_ipmodify_disable(ops);
3129
3130 if (ftrace_hash_rec_disable(ops))
3131 command |= FTRACE_UPDATE_CALLS;
3132
3133 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
3134
3135 if (saved_ftrace_func != ftrace_trace_function) {
3136 saved_ftrace_func = ftrace_trace_function;
3137 command |= FTRACE_UPDATE_TRACE_FUNC;
3138 }
3139
3140 if (!command || !ftrace_enabled)
3141 goto out;
3142
3143 /*
3144 * If the ops uses a trampoline, then it needs to be
3145 * tested first on update.
3146 */
3147 ops->flags |= FTRACE_OPS_FL_REMOVING;
3148 removed_ops = ops;
3149
3150 /* The trampoline logic checks the old hashes */
3151 ops->old_hash.filter_hash = ops->func_hash->filter_hash;
3152 ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
3153
3154 ftrace_run_update_code(command);
3155
3156 /*
3157 * If there's no more ops registered with ftrace, run a
3158 * sanity check to make sure all rec flags are cleared.
3159 */
3160 if (rcu_dereference_protected(ftrace_ops_list,
3161 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
3162 struct ftrace_page *pg;
3163 struct dyn_ftrace *rec;
3164
3165 do_for_each_ftrace_rec(pg, rec) {
3166 if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_NOCLEAR_FLAGS))
3167 pr_warn(" %pS flags:%lx\n",
3168 (void *)rec->ip, rec->flags);
3169 } while_for_each_ftrace_rec();
3170 }
3171
3172 ops->old_hash.filter_hash = NULL;
3173 ops->old_hash.notrace_hash = NULL;
3174
3175 removed_ops = NULL;
3176 ops->flags &= ~FTRACE_OPS_FL_REMOVING;
3177
3178 out:
3179 /*
3180 * Dynamic ops may be freed, we must make sure that all
3181 * callers are done before leaving this function.
3182 */
3183 if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
3184 /*
3185 * We need to do a hard force of sched synchronization.
3186 * This is because we use preempt_disable() to do RCU, but
3187 * the function tracers can be called where RCU is not watching
3188 * (like before user_exit()). We can not rely on the RCU
3189 * infrastructure to do the synchronization, thus we must do it
3190 * ourselves.
3191 */
3192 synchronize_rcu_tasks_rude();
3193
3194 /*
3195 * When the kernel is preemptive, tasks can be preempted
3196 * while on a ftrace trampoline. Just scheduling a task on
3197 * a CPU is not good enough to flush them. Calling
3198 * synchronize_rcu_tasks() will wait for those tasks to
3199 * execute and either schedule voluntarily or enter user space.
3200 */
3201 synchronize_rcu_tasks();
3202
3203 ftrace_trampoline_free(ops);
3204 }
3205
3206 return 0;
3207 }
3208
3209 /* Simply make a copy of @src and return it */
copy_hash(struct ftrace_hash * src)3210 static struct ftrace_hash *copy_hash(struct ftrace_hash *src)
3211 {
3212 if (ftrace_hash_empty(src))
3213 return EMPTY_HASH;
3214
3215 return alloc_and_copy_ftrace_hash(src->size_bits, src);
3216 }
3217
3218 /*
3219 * Append @new_hash entries to @hash:
3220 *
3221 * If @hash is the EMPTY_HASH then it traces all functions and nothing
3222 * needs to be done.
3223 *
3224 * If @new_hash is the EMPTY_HASH, then make *hash the EMPTY_HASH so
3225 * that it traces everything.
3226 *
3227 * Otherwise, go through all of @new_hash and add anything that @hash
3228 * doesn't already have, to @hash.
3229 *
3230 * The filter_hash updates uses just the append_hash() function
3231 * and the notrace_hash does not.
3232 */
append_hash(struct ftrace_hash ** hash,struct ftrace_hash * new_hash,int size_bits)3233 static int append_hash(struct ftrace_hash **hash, struct ftrace_hash *new_hash,
3234 int size_bits)
3235 {
3236 struct ftrace_func_entry *entry;
3237 int size;
3238 int i;
3239
3240 if (*hash) {
3241 /* An empty hash does everything */
3242 if (ftrace_hash_empty(*hash))
3243 return 0;
3244 } else {
3245 *hash = alloc_ftrace_hash(size_bits);
3246 if (!*hash)
3247 return -ENOMEM;
3248 }
3249
3250 /* If new_hash has everything make hash have everything */
3251 if (ftrace_hash_empty(new_hash)) {
3252 free_ftrace_hash(*hash);
3253 *hash = EMPTY_HASH;
3254 return 0;
3255 }
3256
3257 size = 1 << new_hash->size_bits;
3258 for (i = 0; i < size; i++) {
3259 hlist_for_each_entry(entry, &new_hash->buckets[i], hlist) {
3260 /* Only add if not already in hash */
3261 if (!__ftrace_lookup_ip(*hash, entry->ip) &&
3262 add_hash_entry(*hash, entry->ip) == NULL)
3263 return -ENOMEM;
3264 }
3265 }
3266 return 0;
3267 }
3268
3269 /*
3270 * Remove functions from @hash that are in @notrace_hash
3271 */
remove_hash(struct ftrace_hash * hash,struct ftrace_hash * notrace_hash)3272 static void remove_hash(struct ftrace_hash *hash, struct ftrace_hash *notrace_hash)
3273 {
3274 struct ftrace_func_entry *entry;
3275 struct hlist_node *tmp;
3276 int size;
3277 int i;
3278
3279 /* If the notrace hash is empty, there's nothing to do */
3280 if (ftrace_hash_empty(notrace_hash))
3281 return;
3282
3283 size = 1 << hash->size_bits;
3284 for (i = 0; i < size; i++) {
3285 hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
3286 if (!__ftrace_lookup_ip(notrace_hash, entry->ip))
3287 continue;
3288 remove_hash_entry(hash, entry);
3289 kfree(entry);
3290 }
3291 }
3292 }
3293
3294 /*
3295 * Add to @hash only those that are in both @new_hash1 and @new_hash2
3296 *
3297 * The notrace_hash updates uses just the intersect_hash() function
3298 * and the filter_hash does not.
3299 */
intersect_hash(struct ftrace_hash ** hash,struct ftrace_hash * new_hash1,struct ftrace_hash * new_hash2)3300 static int intersect_hash(struct ftrace_hash **hash, struct ftrace_hash *new_hash1,
3301 struct ftrace_hash *new_hash2)
3302 {
3303 struct ftrace_func_entry *entry;
3304 int size;
3305 int i;
3306
3307 /*
3308 * If new_hash1 or new_hash2 is the EMPTY_HASH then make the hash
3309 * empty as well as empty for notrace means none are notraced.
3310 */
3311 if (ftrace_hash_empty(new_hash1) || ftrace_hash_empty(new_hash2)) {
3312 free_ftrace_hash(*hash);
3313 *hash = EMPTY_HASH;
3314 return 0;
3315 }
3316
3317 size = 1 << new_hash1->size_bits;
3318 for (i = 0; i < size; i++) {
3319 hlist_for_each_entry(entry, &new_hash1->buckets[i], hlist) {
3320 /* Only add if in both @new_hash1 and @new_hash2 */
3321 if (__ftrace_lookup_ip(new_hash2, entry->ip) &&
3322 add_hash_entry(*hash, entry->ip) == NULL)
3323 return -ENOMEM;
3324 }
3325 }
3326 /* If nothing intersects, make it the empty set */
3327 if (ftrace_hash_empty(*hash)) {
3328 free_ftrace_hash(*hash);
3329 *hash = EMPTY_HASH;
3330 }
3331 return 0;
3332 }
3333
ops_equal(struct ftrace_hash * A,struct ftrace_hash * B)3334 static bool ops_equal(struct ftrace_hash *A, struct ftrace_hash *B)
3335 {
3336 struct ftrace_func_entry *entry;
3337 int size;
3338 int i;
3339
3340 if (ftrace_hash_empty(A))
3341 return ftrace_hash_empty(B);
3342
3343 if (ftrace_hash_empty(B))
3344 return ftrace_hash_empty(A);
3345
3346 if (A->count != B->count)
3347 return false;
3348
3349 size = 1 << A->size_bits;
3350 for (i = 0; i < size; i++) {
3351 hlist_for_each_entry(entry, &A->buckets[i], hlist) {
3352 if (!__ftrace_lookup_ip(B, entry->ip))
3353 return false;
3354 }
3355 }
3356
3357 return true;
3358 }
3359
3360 static void ftrace_ops_update_code(struct ftrace_ops *ops,
3361 struct ftrace_ops_hash *old_hash);
3362
__ftrace_hash_move_and_update_ops(struct ftrace_ops * ops,struct ftrace_hash ** orig_hash,struct ftrace_hash * hash,int enable)3363 static int __ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
3364 struct ftrace_hash **orig_hash,
3365 struct ftrace_hash *hash,
3366 int enable)
3367 {
3368 struct ftrace_ops_hash old_hash_ops;
3369 struct ftrace_hash *old_hash;
3370 int ret;
3371
3372 old_hash = *orig_hash;
3373 old_hash_ops.filter_hash = ops->func_hash->filter_hash;
3374 old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
3375 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3376 if (!ret) {
3377 ftrace_ops_update_code(ops, &old_hash_ops);
3378 free_ftrace_hash_rcu(old_hash);
3379 }
3380 return ret;
3381 }
3382
ftrace_update_ops(struct ftrace_ops * ops,struct ftrace_hash * filter_hash,struct ftrace_hash * notrace_hash)3383 static int ftrace_update_ops(struct ftrace_ops *ops, struct ftrace_hash *filter_hash,
3384 struct ftrace_hash *notrace_hash)
3385 {
3386 int ret;
3387
3388 if (!ops_equal(filter_hash, ops->func_hash->filter_hash)) {
3389 ret = __ftrace_hash_move_and_update_ops(ops, &ops->func_hash->filter_hash,
3390 filter_hash, 1);
3391 if (ret < 0)
3392 return ret;
3393 }
3394
3395 if (!ops_equal(notrace_hash, ops->func_hash->notrace_hash)) {
3396 ret = __ftrace_hash_move_and_update_ops(ops, &ops->func_hash->notrace_hash,
3397 notrace_hash, 0);
3398 if (ret < 0)
3399 return ret;
3400 }
3401
3402 return 0;
3403 }
3404
add_first_hash(struct ftrace_hash ** filter_hash,struct ftrace_hash ** notrace_hash,struct ftrace_ops_hash * func_hash)3405 static int add_first_hash(struct ftrace_hash **filter_hash, struct ftrace_hash **notrace_hash,
3406 struct ftrace_ops_hash *func_hash)
3407 {
3408 /* If the filter hash is not empty, simply remove the nohash from it */
3409 if (!ftrace_hash_empty(func_hash->filter_hash)) {
3410 *filter_hash = copy_hash(func_hash->filter_hash);
3411 if (!*filter_hash)
3412 return -ENOMEM;
3413 remove_hash(*filter_hash, func_hash->notrace_hash);
3414 *notrace_hash = EMPTY_HASH;
3415
3416 } else {
3417 *notrace_hash = copy_hash(func_hash->notrace_hash);
3418 if (!*notrace_hash)
3419 return -ENOMEM;
3420 *filter_hash = EMPTY_HASH;
3421 }
3422 return 0;
3423 }
3424
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)3425 static int add_next_hash(struct ftrace_hash **filter_hash, struct ftrace_hash **notrace_hash,
3426 struct ftrace_ops_hash *ops_hash, struct ftrace_ops_hash *subops_hash)
3427 {
3428 int size_bits;
3429 int ret;
3430
3431 /* If the subops trace all functions so must the main ops */
3432 if (ftrace_hash_empty(ops_hash->filter_hash) ||
3433 ftrace_hash_empty(subops_hash->filter_hash)) {
3434 *filter_hash = EMPTY_HASH;
3435 } else {
3436 /*
3437 * The main ops filter hash is not empty, so its
3438 * notrace_hash had better be, as the notrace hash
3439 * is only used for empty main filter hashes.
3440 */
3441 WARN_ON_ONCE(!ftrace_hash_empty(ops_hash->notrace_hash));
3442
3443 size_bits = max(ops_hash->filter_hash->size_bits,
3444 subops_hash->filter_hash->size_bits);
3445
3446 /* Copy the subops hash */
3447 *filter_hash = alloc_and_copy_ftrace_hash(size_bits, subops_hash->filter_hash);
3448 if (!*filter_hash)
3449 return -ENOMEM;
3450 /* Remove any notrace functions from the copy */
3451 remove_hash(*filter_hash, subops_hash->notrace_hash);
3452
3453 ret = append_hash(filter_hash, ops_hash->filter_hash,
3454 size_bits);
3455 if (ret < 0) {
3456 free_ftrace_hash(*filter_hash);
3457 *filter_hash = EMPTY_HASH;
3458 return ret;
3459 }
3460 }
3461
3462 /*
3463 * Only process notrace hashes if the main filter hash is empty
3464 * (tracing all functions), otherwise the filter hash will just
3465 * remove the notrace hash functions, and the notrace hash is
3466 * not needed.
3467 */
3468 if (ftrace_hash_empty(*filter_hash)) {
3469 /*
3470 * Intersect the notrace functions. That is, if two
3471 * subops are not tracing a set of functions, the
3472 * main ops will only not trace the functions that are
3473 * in both subops, but has to trace the functions that
3474 * are only notrace in one of the subops, for the other
3475 * subops to be able to trace them.
3476 */
3477 size_bits = max(ops_hash->notrace_hash->size_bits,
3478 subops_hash->notrace_hash->size_bits);
3479 *notrace_hash = alloc_ftrace_hash(size_bits);
3480 if (!*notrace_hash)
3481 return -ENOMEM;
3482
3483 ret = intersect_hash(notrace_hash, ops_hash->notrace_hash,
3484 subops_hash->notrace_hash);
3485 if (ret < 0) {
3486 free_ftrace_hash(*notrace_hash);
3487 *notrace_hash = EMPTY_HASH;
3488 return ret;
3489 }
3490 }
3491 return 0;
3492 }
3493
3494 /**
3495 * ftrace_startup_subops - enable tracing for subops of an ops
3496 * @ops: Manager ops (used to pick all the functions of its subops)
3497 * @subops: A new ops to add to @ops
3498 * @command: Extra commands to use to enable tracing
3499 *
3500 * The @ops is a manager @ops that has the filter that includes all the functions
3501 * that its list of subops are tracing. Adding a new @subops will add the
3502 * functions of @subops to @ops.
3503 */
ftrace_startup_subops(struct ftrace_ops * ops,struct ftrace_ops * subops,int command)3504 int ftrace_startup_subops(struct ftrace_ops *ops, struct ftrace_ops *subops, int command)
3505 {
3506 struct ftrace_hash *filter_hash = EMPTY_HASH;
3507 struct ftrace_hash *notrace_hash = EMPTY_HASH;
3508 struct ftrace_hash *save_filter_hash;
3509 struct ftrace_hash *save_notrace_hash;
3510 int ret;
3511
3512 if (unlikely(ftrace_disabled))
3513 return -ENODEV;
3514
3515 ftrace_ops_init(ops);
3516 ftrace_ops_init(subops);
3517
3518 if (WARN_ON_ONCE(subops->flags & FTRACE_OPS_FL_ENABLED))
3519 return -EBUSY;
3520
3521 /* Make everything canonical (Just in case!) */
3522 if (!ops->func_hash->filter_hash)
3523 ops->func_hash->filter_hash = EMPTY_HASH;
3524 if (!ops->func_hash->notrace_hash)
3525 ops->func_hash->notrace_hash = EMPTY_HASH;
3526 if (!subops->func_hash->filter_hash)
3527 subops->func_hash->filter_hash = EMPTY_HASH;
3528 if (!subops->func_hash->notrace_hash)
3529 subops->func_hash->notrace_hash = EMPTY_HASH;
3530
3531 /* For the first subops to ops just enable it normally */
3532 if (list_empty(&ops->subop_list)) {
3533
3534 /* The ops was empty, should have empty hashes */
3535 WARN_ON_ONCE(!ftrace_hash_empty(ops->func_hash->filter_hash));
3536 WARN_ON_ONCE(!ftrace_hash_empty(ops->func_hash->notrace_hash));
3537
3538 ret = add_first_hash(&filter_hash, ¬race_hash, subops->func_hash);
3539 if (ret < 0)
3540 return ret;
3541
3542 save_filter_hash = ops->func_hash->filter_hash;
3543 save_notrace_hash = ops->func_hash->notrace_hash;
3544
3545 ops->func_hash->filter_hash = filter_hash;
3546 ops->func_hash->notrace_hash = notrace_hash;
3547 list_add(&subops->list, &ops->subop_list);
3548 ret = ftrace_startup(ops, command);
3549 if (ret < 0) {
3550 list_del(&subops->list);
3551 ops->func_hash->filter_hash = save_filter_hash;
3552 ops->func_hash->notrace_hash = save_notrace_hash;
3553 free_ftrace_hash(filter_hash);
3554 free_ftrace_hash(notrace_hash);
3555 } else {
3556 free_ftrace_hash(save_filter_hash);
3557 free_ftrace_hash(save_notrace_hash);
3558 subops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP;
3559 subops->managed = ops;
3560 }
3561 return ret;
3562 }
3563
3564 /*
3565 * Here there's already something attached. Here are the rules:
3566 * If the new subops and main ops filter hashes are not empty:
3567 * o Make a copy of the subops filter hash
3568 * o Remove all functions in the nohash from it.
3569 * o Add in the main hash filter functions
3570 * o Remove any of these functions from the main notrace hash
3571 */
3572
3573 ret = add_next_hash(&filter_hash, ¬race_hash, ops->func_hash, subops->func_hash);
3574 if (ret < 0)
3575 return ret;
3576
3577 list_add(&subops->list, &ops->subop_list);
3578
3579 ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3580 free_ftrace_hash(filter_hash);
3581 free_ftrace_hash(notrace_hash);
3582 if (ret < 0) {
3583 list_del(&subops->list);
3584 } else {
3585 subops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP;
3586 subops->managed = ops;
3587 }
3588 return ret;
3589 }
3590
rebuild_hashes(struct ftrace_hash ** filter_hash,struct ftrace_hash ** notrace_hash,struct ftrace_ops * ops)3591 static int rebuild_hashes(struct ftrace_hash **filter_hash, struct ftrace_hash **notrace_hash,
3592 struct ftrace_ops *ops)
3593 {
3594 struct ftrace_ops_hash temp_hash;
3595 struct ftrace_ops *subops;
3596 bool first = true;
3597 int ret;
3598
3599 temp_hash.filter_hash = EMPTY_HASH;
3600 temp_hash.notrace_hash = EMPTY_HASH;
3601
3602 list_for_each_entry(subops, &ops->subop_list, list) {
3603 *filter_hash = EMPTY_HASH;
3604 *notrace_hash = EMPTY_HASH;
3605
3606 if (first) {
3607 ret = add_first_hash(filter_hash, notrace_hash, subops->func_hash);
3608 if (ret < 0)
3609 return ret;
3610 first = false;
3611 } else {
3612 ret = add_next_hash(filter_hash, notrace_hash,
3613 &temp_hash, subops->func_hash);
3614 if (ret < 0) {
3615 free_ftrace_hash(temp_hash.filter_hash);
3616 free_ftrace_hash(temp_hash.notrace_hash);
3617 return ret;
3618 }
3619 }
3620
3621 free_ftrace_hash(temp_hash.filter_hash);
3622 free_ftrace_hash(temp_hash.notrace_hash);
3623
3624 temp_hash.filter_hash = *filter_hash;
3625 temp_hash.notrace_hash = *notrace_hash;
3626 }
3627 return 0;
3628 }
3629
3630 /**
3631 * ftrace_shutdown_subops - Remove a subops from a manager ops
3632 * @ops: A manager ops to remove @subops from
3633 * @subops: The subops to remove from @ops
3634 * @command: Any extra command flags to add to modifying the text
3635 *
3636 * Removes the functions being traced by the @subops from @ops. Note, it
3637 * will not affect functions that are being traced by other subops that
3638 * still exist in @ops.
3639 *
3640 * If the last subops is removed from @ops, then @ops is shutdown normally.
3641 */
ftrace_shutdown_subops(struct ftrace_ops * ops,struct ftrace_ops * subops,int command)3642 int ftrace_shutdown_subops(struct ftrace_ops *ops, struct ftrace_ops *subops, int command)
3643 {
3644 struct ftrace_hash *filter_hash = EMPTY_HASH;
3645 struct ftrace_hash *notrace_hash = EMPTY_HASH;
3646 int ret;
3647
3648 if (unlikely(ftrace_disabled))
3649 return -ENODEV;
3650
3651 if (WARN_ON_ONCE(!(subops->flags & FTRACE_OPS_FL_ENABLED)))
3652 return -EINVAL;
3653
3654 list_del(&subops->list);
3655
3656 if (list_empty(&ops->subop_list)) {
3657 /* Last one, just disable the current ops */
3658
3659 ret = ftrace_shutdown(ops, command);
3660 if (ret < 0) {
3661 list_add(&subops->list, &ops->subop_list);
3662 return ret;
3663 }
3664
3665 subops->flags &= ~FTRACE_OPS_FL_ENABLED;
3666
3667 free_ftrace_hash(ops->func_hash->filter_hash);
3668 free_ftrace_hash(ops->func_hash->notrace_hash);
3669 ops->func_hash->filter_hash = EMPTY_HASH;
3670 ops->func_hash->notrace_hash = EMPTY_HASH;
3671 subops->flags &= ~(FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP);
3672 subops->managed = NULL;
3673
3674 return 0;
3675 }
3676
3677 /* Rebuild the hashes without subops */
3678 ret = rebuild_hashes(&filter_hash, ¬race_hash, ops);
3679 if (ret < 0)
3680 return ret;
3681
3682 ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3683 if (ret < 0) {
3684 list_add(&subops->list, &ops->subop_list);
3685 } else {
3686 subops->flags &= ~(FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_SUBOP);
3687 subops->managed = NULL;
3688 }
3689 free_ftrace_hash(filter_hash);
3690 free_ftrace_hash(notrace_hash);
3691 return ret;
3692 }
3693
ftrace_hash_move_and_update_subops(struct ftrace_ops * subops,struct ftrace_hash ** orig_subhash,struct ftrace_hash * hash)3694 static int ftrace_hash_move_and_update_subops(struct ftrace_ops *subops,
3695 struct ftrace_hash **orig_subhash,
3696 struct ftrace_hash *hash)
3697 {
3698 struct ftrace_ops *ops = subops->managed;
3699 struct ftrace_hash *notrace_hash;
3700 struct ftrace_hash *filter_hash;
3701 struct ftrace_hash *save_hash;
3702 struct ftrace_hash *new_hash;
3703 int ret;
3704
3705 /* Manager ops can not be subops (yet) */
3706 if (WARN_ON_ONCE(!ops || ops->flags & FTRACE_OPS_FL_SUBOP))
3707 return -EINVAL;
3708
3709 /* Move the new hash over to the subops hash */
3710 save_hash = *orig_subhash;
3711 *orig_subhash = __ftrace_hash_move(hash);
3712 if (!*orig_subhash) {
3713 *orig_subhash = save_hash;
3714 return -ENOMEM;
3715 }
3716
3717 ret = rebuild_hashes(&filter_hash, ¬race_hash, ops);
3718 if (!ret) {
3719 ret = ftrace_update_ops(ops, filter_hash, notrace_hash);
3720 free_ftrace_hash(filter_hash);
3721 free_ftrace_hash(notrace_hash);
3722 }
3723
3724 if (ret) {
3725 /* Put back the original hash */
3726 new_hash = *orig_subhash;
3727 *orig_subhash = save_hash;
3728 free_ftrace_hash_rcu(new_hash);
3729 } else {
3730 free_ftrace_hash_rcu(save_hash);
3731 }
3732 return ret;
3733 }
3734
3735
3736 u64 ftrace_update_time;
3737 u64 ftrace_total_mod_time;
3738 unsigned long ftrace_update_tot_cnt;
3739 unsigned long ftrace_number_of_pages;
3740 unsigned long ftrace_number_of_groups;
3741
ops_traces_mod(struct ftrace_ops * ops)3742 static inline int ops_traces_mod(struct ftrace_ops *ops)
3743 {
3744 /*
3745 * Filter_hash being empty will default to trace module.
3746 * But notrace hash requires a test of individual module functions.
3747 */
3748 return ftrace_hash_empty(ops->func_hash->filter_hash) &&
3749 ftrace_hash_empty(ops->func_hash->notrace_hash);
3750 }
3751
ftrace_update_code(struct module * mod,struct ftrace_page * new_pgs)3752 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
3753 {
3754 bool init_nop = ftrace_need_init_nop();
3755 struct ftrace_page *pg;
3756 struct dyn_ftrace *p;
3757 u64 start, stop, update_time;
3758 unsigned long update_cnt = 0;
3759 unsigned long rec_flags = 0;
3760 int i;
3761
3762 start = ftrace_now(raw_smp_processor_id());
3763
3764 /*
3765 * When a module is loaded, this function is called to convert
3766 * the calls to mcount in its text to nops, and also to create
3767 * an entry in the ftrace data. Now, if ftrace is activated
3768 * after this call, but before the module sets its text to
3769 * read-only, the modification of enabling ftrace can fail if
3770 * the read-only is done while ftrace is converting the calls.
3771 * To prevent this, the module's records are set as disabled
3772 * and will be enabled after the call to set the module's text
3773 * to read-only.
3774 */
3775 if (mod)
3776 rec_flags |= FTRACE_FL_DISABLED;
3777
3778 for (pg = new_pgs; pg; pg = pg->next) {
3779
3780 for (i = 0; i < pg->index; i++) {
3781
3782 /* If something went wrong, bail without enabling anything */
3783 if (unlikely(ftrace_disabled))
3784 return -1;
3785
3786 p = &pg->records[i];
3787 p->flags = rec_flags;
3788
3789 /*
3790 * Do the initial record conversion from mcount jump
3791 * to the NOP instructions.
3792 */
3793 if (init_nop && !ftrace_nop_initialize(mod, p))
3794 break;
3795
3796 update_cnt++;
3797 }
3798 }
3799
3800 stop = ftrace_now(raw_smp_processor_id());
3801 update_time = stop - start;
3802 if (mod)
3803 ftrace_total_mod_time += update_time;
3804 else
3805 ftrace_update_time = update_time;
3806 ftrace_update_tot_cnt += update_cnt;
3807
3808 return 0;
3809 }
3810
ftrace_allocate_records(struct ftrace_page * pg,int count)3811 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
3812 {
3813 int order;
3814 int pages;
3815 int cnt;
3816
3817 if (WARN_ON(!count))
3818 return -EINVAL;
3819
3820 /* We want to fill as much as possible, with no empty pages */
3821 pages = DIV_ROUND_UP(count, ENTRIES_PER_PAGE);
3822 order = fls(pages) - 1;
3823
3824 again:
3825 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
3826
3827 if (!pg->records) {
3828 /* if we can't allocate this size, try something smaller */
3829 if (!order)
3830 return -ENOMEM;
3831 order--;
3832 goto again;
3833 }
3834
3835 ftrace_number_of_pages += 1 << order;
3836 ftrace_number_of_groups++;
3837
3838 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
3839 pg->order = order;
3840
3841 if (cnt > count)
3842 cnt = count;
3843
3844 return cnt;
3845 }
3846
ftrace_free_pages(struct ftrace_page * pages)3847 static void ftrace_free_pages(struct ftrace_page *pages)
3848 {
3849 struct ftrace_page *pg = pages;
3850
3851 while (pg) {
3852 if (pg->records) {
3853 free_pages((unsigned long)pg->records, pg->order);
3854 ftrace_number_of_pages -= 1 << pg->order;
3855 }
3856 pages = pg->next;
3857 kfree(pg);
3858 pg = pages;
3859 ftrace_number_of_groups--;
3860 }
3861 }
3862
3863 static struct ftrace_page *
ftrace_allocate_pages(unsigned long num_to_init)3864 ftrace_allocate_pages(unsigned long num_to_init)
3865 {
3866 struct ftrace_page *start_pg;
3867 struct ftrace_page *pg;
3868 int cnt;
3869
3870 if (!num_to_init)
3871 return NULL;
3872
3873 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3874 if (!pg)
3875 return NULL;
3876
3877 /*
3878 * Try to allocate as much as possible in one continues
3879 * location that fills in all of the space. We want to
3880 * waste as little space as possible.
3881 */
3882 for (;;) {
3883 cnt = ftrace_allocate_records(pg, num_to_init);
3884 if (cnt < 0)
3885 goto free_pages;
3886
3887 num_to_init -= cnt;
3888 if (!num_to_init)
3889 break;
3890
3891 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3892 if (!pg->next)
3893 goto free_pages;
3894
3895 pg = pg->next;
3896 }
3897
3898 return start_pg;
3899
3900 free_pages:
3901 ftrace_free_pages(start_pg);
3902 pr_info("ftrace: FAILED to allocate memory for functions\n");
3903 return NULL;
3904 }
3905
3906 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3907
3908 struct ftrace_iterator {
3909 loff_t pos;
3910 loff_t func_pos;
3911 loff_t mod_pos;
3912 struct ftrace_page *pg;
3913 struct dyn_ftrace *func;
3914 struct ftrace_func_probe *probe;
3915 struct ftrace_func_entry *probe_entry;
3916 struct trace_parser parser;
3917 struct ftrace_hash *hash;
3918 struct ftrace_ops *ops;
3919 struct trace_array *tr;
3920 struct list_head *mod_list;
3921 int pidx;
3922 int idx;
3923 unsigned flags;
3924 };
3925
3926 static void *
t_probe_next(struct seq_file * m,loff_t * pos)3927 t_probe_next(struct seq_file *m, loff_t *pos)
3928 {
3929 struct ftrace_iterator *iter = m->private;
3930 struct trace_array *tr = iter->ops->private;
3931 struct list_head *func_probes;
3932 struct ftrace_hash *hash;
3933 struct list_head *next;
3934 struct hlist_node *hnd = NULL;
3935 struct hlist_head *hhd;
3936 int size;
3937
3938 (*pos)++;
3939 iter->pos = *pos;
3940
3941 if (!tr)
3942 return NULL;
3943
3944 func_probes = &tr->func_probes;
3945 if (list_empty(func_probes))
3946 return NULL;
3947
3948 if (!iter->probe) {
3949 next = func_probes->next;
3950 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3951 }
3952
3953 if (iter->probe_entry)
3954 hnd = &iter->probe_entry->hlist;
3955
3956 hash = iter->probe->ops.func_hash->filter_hash;
3957
3958 /*
3959 * A probe being registered may temporarily have an empty hash
3960 * and it's at the end of the func_probes list.
3961 */
3962 if (!hash || hash == EMPTY_HASH)
3963 return NULL;
3964
3965 size = 1 << hash->size_bits;
3966
3967 retry:
3968 if (iter->pidx >= size) {
3969 if (iter->probe->list.next == func_probes)
3970 return NULL;
3971 next = iter->probe->list.next;
3972 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3973 hash = iter->probe->ops.func_hash->filter_hash;
3974 size = 1 << hash->size_bits;
3975 iter->pidx = 0;
3976 }
3977
3978 hhd = &hash->buckets[iter->pidx];
3979
3980 if (hlist_empty(hhd)) {
3981 iter->pidx++;
3982 hnd = NULL;
3983 goto retry;
3984 }
3985
3986 if (!hnd)
3987 hnd = hhd->first;
3988 else {
3989 hnd = hnd->next;
3990 if (!hnd) {
3991 iter->pidx++;
3992 goto retry;
3993 }
3994 }
3995
3996 if (WARN_ON_ONCE(!hnd))
3997 return NULL;
3998
3999 iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
4000
4001 return iter;
4002 }
4003
t_probe_start(struct seq_file * m,loff_t * pos)4004 static void *t_probe_start(struct seq_file *m, loff_t *pos)
4005 {
4006 struct ftrace_iterator *iter = m->private;
4007 void *p = NULL;
4008 loff_t l;
4009
4010 if (!(iter->flags & FTRACE_ITER_DO_PROBES))
4011 return NULL;
4012
4013 if (iter->mod_pos > *pos)
4014 return NULL;
4015
4016 iter->probe = NULL;
4017 iter->probe_entry = NULL;
4018 iter->pidx = 0;
4019 for (l = 0; l <= (*pos - iter->mod_pos); ) {
4020 p = t_probe_next(m, &l);
4021 if (!p)
4022 break;
4023 }
4024 if (!p)
4025 return NULL;
4026
4027 /* Only set this if we have an item */
4028 iter->flags |= FTRACE_ITER_PROBE;
4029
4030 return iter;
4031 }
4032
4033 static int
t_probe_show(struct seq_file * m,struct ftrace_iterator * iter)4034 t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
4035 {
4036 struct ftrace_func_entry *probe_entry;
4037 struct ftrace_probe_ops *probe_ops;
4038 struct ftrace_func_probe *probe;
4039
4040 probe = iter->probe;
4041 probe_entry = iter->probe_entry;
4042
4043 if (WARN_ON_ONCE(!probe || !probe_entry))
4044 return -EIO;
4045
4046 probe_ops = probe->probe_ops;
4047
4048 if (probe_ops->print)
4049 return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
4050
4051 seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
4052 (void *)probe_ops->func);
4053
4054 return 0;
4055 }
4056
4057 static void *
t_mod_next(struct seq_file * m,loff_t * pos)4058 t_mod_next(struct seq_file *m, loff_t *pos)
4059 {
4060 struct ftrace_iterator *iter = m->private;
4061 struct trace_array *tr = iter->tr;
4062
4063 (*pos)++;
4064 iter->pos = *pos;
4065
4066 iter->mod_list = iter->mod_list->next;
4067
4068 if (iter->mod_list == &tr->mod_trace ||
4069 iter->mod_list == &tr->mod_notrace) {
4070 iter->flags &= ~FTRACE_ITER_MOD;
4071 return NULL;
4072 }
4073
4074 iter->mod_pos = *pos;
4075
4076 return iter;
4077 }
4078
t_mod_start(struct seq_file * m,loff_t * pos)4079 static void *t_mod_start(struct seq_file *m, loff_t *pos)
4080 {
4081 struct ftrace_iterator *iter = m->private;
4082 void *p = NULL;
4083 loff_t l;
4084
4085 if (iter->func_pos > *pos)
4086 return NULL;
4087
4088 iter->mod_pos = iter->func_pos;
4089
4090 /* probes are only available if tr is set */
4091 if (!iter->tr)
4092 return NULL;
4093
4094 for (l = 0; l <= (*pos - iter->func_pos); ) {
4095 p = t_mod_next(m, &l);
4096 if (!p)
4097 break;
4098 }
4099 if (!p) {
4100 iter->flags &= ~FTRACE_ITER_MOD;
4101 return t_probe_start(m, pos);
4102 }
4103
4104 /* Only set this if we have an item */
4105 iter->flags |= FTRACE_ITER_MOD;
4106
4107 return iter;
4108 }
4109
4110 static int
t_mod_show(struct seq_file * m,struct ftrace_iterator * iter)4111 t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
4112 {
4113 struct ftrace_mod_load *ftrace_mod;
4114 struct trace_array *tr = iter->tr;
4115
4116 if (WARN_ON_ONCE(!iter->mod_list) ||
4117 iter->mod_list == &tr->mod_trace ||
4118 iter->mod_list == &tr->mod_notrace)
4119 return -EIO;
4120
4121 ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
4122
4123 if (ftrace_mod->func)
4124 seq_printf(m, "%s", ftrace_mod->func);
4125 else
4126 seq_putc(m, '*');
4127
4128 seq_printf(m, ":mod:%s\n", ftrace_mod->module);
4129
4130 return 0;
4131 }
4132
4133 static void *
t_func_next(struct seq_file * m,loff_t * pos)4134 t_func_next(struct seq_file *m, loff_t *pos)
4135 {
4136 struct ftrace_iterator *iter = m->private;
4137 struct dyn_ftrace *rec = NULL;
4138
4139 (*pos)++;
4140
4141 retry:
4142 if (iter->idx >= iter->pg->index) {
4143 if (iter->pg->next) {
4144 iter->pg = iter->pg->next;
4145 iter->idx = 0;
4146 goto retry;
4147 }
4148 } else {
4149 rec = &iter->pg->records[iter->idx++];
4150 if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
4151 !ftrace_lookup_ip(iter->hash, rec->ip)) ||
4152
4153 ((iter->flags & FTRACE_ITER_ENABLED) &&
4154 !(rec->flags & FTRACE_FL_ENABLED)) ||
4155
4156 ((iter->flags & FTRACE_ITER_TOUCHED) &&
4157 !(rec->flags & FTRACE_FL_TOUCHED))) {
4158
4159 rec = NULL;
4160 goto retry;
4161 }
4162 }
4163
4164 if (!rec)
4165 return NULL;
4166
4167 iter->pos = iter->func_pos = *pos;
4168 iter->func = rec;
4169
4170 return iter;
4171 }
4172
4173 static void *
t_next(struct seq_file * m,void * v,loff_t * pos)4174 t_next(struct seq_file *m, void *v, loff_t *pos)
4175 {
4176 struct ftrace_iterator *iter = m->private;
4177 loff_t l = *pos; /* t_probe_start() must use original pos */
4178 void *ret;
4179
4180 if (unlikely(ftrace_disabled))
4181 return NULL;
4182
4183 if (iter->flags & FTRACE_ITER_PROBE)
4184 return t_probe_next(m, pos);
4185
4186 if (iter->flags & FTRACE_ITER_MOD)
4187 return t_mod_next(m, pos);
4188
4189 if (iter->flags & FTRACE_ITER_PRINTALL) {
4190 /* next must increment pos, and t_probe_start does not */
4191 (*pos)++;
4192 return t_mod_start(m, &l);
4193 }
4194
4195 ret = t_func_next(m, pos);
4196
4197 if (!ret)
4198 return t_mod_start(m, &l);
4199
4200 return ret;
4201 }
4202
reset_iter_read(struct ftrace_iterator * iter)4203 static void reset_iter_read(struct ftrace_iterator *iter)
4204 {
4205 iter->pos = 0;
4206 iter->func_pos = 0;
4207 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
4208 }
4209
t_start(struct seq_file * m,loff_t * pos)4210 static void *t_start(struct seq_file *m, loff_t *pos)
4211 {
4212 struct ftrace_iterator *iter = m->private;
4213 void *p = NULL;
4214 loff_t l;
4215
4216 mutex_lock(&ftrace_lock);
4217
4218 if (unlikely(ftrace_disabled))
4219 return NULL;
4220
4221 /*
4222 * If an lseek was done, then reset and start from beginning.
4223 */
4224 if (*pos < iter->pos)
4225 reset_iter_read(iter);
4226
4227 /*
4228 * For set_ftrace_filter reading, if we have the filter
4229 * off, we can short cut and just print out that all
4230 * functions are enabled.
4231 */
4232 if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
4233 ftrace_hash_empty(iter->hash)) {
4234 iter->func_pos = 1; /* Account for the message */
4235 if (*pos > 0)
4236 return t_mod_start(m, pos);
4237 iter->flags |= FTRACE_ITER_PRINTALL;
4238 /* reset in case of seek/pread */
4239 iter->flags &= ~FTRACE_ITER_PROBE;
4240 return iter;
4241 }
4242
4243 if (iter->flags & FTRACE_ITER_MOD)
4244 return t_mod_start(m, pos);
4245
4246 /*
4247 * Unfortunately, we need to restart at ftrace_pages_start
4248 * every time we let go of the ftrace_mutex. This is because
4249 * those pointers can change without the lock.
4250 */
4251 iter->pg = ftrace_pages_start;
4252 iter->idx = 0;
4253 for (l = 0; l <= *pos; ) {
4254 p = t_func_next(m, &l);
4255 if (!p)
4256 break;
4257 }
4258
4259 if (!p)
4260 return t_mod_start(m, pos);
4261
4262 return iter;
4263 }
4264
t_stop(struct seq_file * m,void * p)4265 static void t_stop(struct seq_file *m, void *p)
4266 {
4267 mutex_unlock(&ftrace_lock);
4268 }
4269
4270 void * __weak
arch_ftrace_trampoline_func(struct ftrace_ops * ops,struct dyn_ftrace * rec)4271 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
4272 {
4273 return NULL;
4274 }
4275
add_trampoline_func(struct seq_file * m,struct ftrace_ops * ops,struct dyn_ftrace * rec)4276 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
4277 struct dyn_ftrace *rec)
4278 {
4279 void *ptr;
4280
4281 ptr = arch_ftrace_trampoline_func(ops, rec);
4282 if (ptr)
4283 seq_printf(m, " ->%pS", ptr);
4284 }
4285
4286 #ifdef FTRACE_MCOUNT_MAX_OFFSET
4287 /*
4288 * Weak functions can still have an mcount/fentry that is saved in
4289 * the __mcount_loc section. These can be detected by having a
4290 * symbol offset of greater than FTRACE_MCOUNT_MAX_OFFSET, as the
4291 * symbol found by kallsyms is not the function that the mcount/fentry
4292 * is part of. The offset is much greater in these cases.
4293 *
4294 * Test the record to make sure that the ip points to a valid kallsyms
4295 * and if not, mark it disabled.
4296 */
test_for_valid_rec(struct dyn_ftrace * rec)4297 static int test_for_valid_rec(struct dyn_ftrace *rec)
4298 {
4299 char str[KSYM_SYMBOL_LEN];
4300 unsigned long offset;
4301 const char *ret;
4302
4303 ret = kallsyms_lookup(rec->ip, NULL, &offset, NULL, str);
4304
4305 /* Weak functions can cause invalid addresses */
4306 if (!ret || offset > FTRACE_MCOUNT_MAX_OFFSET) {
4307 rec->flags |= FTRACE_FL_DISABLED;
4308 return 0;
4309 }
4310 return 1;
4311 }
4312
4313 static struct workqueue_struct *ftrace_check_wq __initdata;
4314 static struct work_struct ftrace_check_work __initdata;
4315
4316 /*
4317 * Scan all the mcount/fentry entries to make sure they are valid.
4318 */
ftrace_check_work_func(struct work_struct * work)4319 static __init void ftrace_check_work_func(struct work_struct *work)
4320 {
4321 struct ftrace_page *pg;
4322 struct dyn_ftrace *rec;
4323
4324 mutex_lock(&ftrace_lock);
4325 do_for_each_ftrace_rec(pg, rec) {
4326 test_for_valid_rec(rec);
4327 } while_for_each_ftrace_rec();
4328 mutex_unlock(&ftrace_lock);
4329 }
4330
ftrace_check_for_weak_functions(void)4331 static int __init ftrace_check_for_weak_functions(void)
4332 {
4333 INIT_WORK(&ftrace_check_work, ftrace_check_work_func);
4334
4335 ftrace_check_wq = alloc_workqueue("ftrace_check_wq", WQ_UNBOUND, 0);
4336
4337 queue_work(ftrace_check_wq, &ftrace_check_work);
4338 return 0;
4339 }
4340
ftrace_check_sync(void)4341 static int __init ftrace_check_sync(void)
4342 {
4343 /* Make sure the ftrace_check updates are finished */
4344 if (ftrace_check_wq)
4345 destroy_workqueue(ftrace_check_wq);
4346 return 0;
4347 }
4348
4349 late_initcall_sync(ftrace_check_sync);
4350 subsys_initcall(ftrace_check_for_weak_functions);
4351
print_rec(struct seq_file * m,unsigned long ip)4352 static int print_rec(struct seq_file *m, unsigned long ip)
4353 {
4354 unsigned long offset;
4355 char str[KSYM_SYMBOL_LEN];
4356 char *modname;
4357 const char *ret;
4358
4359 ret = kallsyms_lookup(ip, NULL, &offset, &modname, str);
4360 /* Weak functions can cause invalid addresses */
4361 if (!ret || offset > FTRACE_MCOUNT_MAX_OFFSET) {
4362 snprintf(str, KSYM_SYMBOL_LEN, "%s_%ld",
4363 FTRACE_INVALID_FUNCTION, offset);
4364 ret = NULL;
4365 }
4366
4367 seq_puts(m, str);
4368 if (modname)
4369 seq_printf(m, " [%s]", modname);
4370 return ret == NULL ? -1 : 0;
4371 }
4372 #else
test_for_valid_rec(struct dyn_ftrace * rec)4373 static inline int test_for_valid_rec(struct dyn_ftrace *rec)
4374 {
4375 return 1;
4376 }
4377
print_rec(struct seq_file * m,unsigned long ip)4378 static inline int print_rec(struct seq_file *m, unsigned long ip)
4379 {
4380 seq_printf(m, "%ps", (void *)ip);
4381 return 0;
4382 }
4383 #endif
4384
print_subops(struct seq_file * m,struct ftrace_ops * ops,struct dyn_ftrace * rec)4385 static void print_subops(struct seq_file *m, struct ftrace_ops *ops, struct dyn_ftrace *rec)
4386 {
4387 struct ftrace_ops *subops;
4388 bool first = true;
4389
4390 list_for_each_entry(subops, &ops->subop_list, list) {
4391 if (!((subops->flags & FTRACE_OPS_FL_ENABLED) &&
4392 hash_contains_ip(rec->ip, subops->func_hash)))
4393 continue;
4394 if (first) {
4395 seq_printf(m, "\tsubops:");
4396 first = false;
4397 }
4398 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4399 if (subops->flags & FTRACE_OPS_FL_GRAPH) {
4400 struct fgraph_ops *gops;
4401
4402 gops = container_of(subops, struct fgraph_ops, ops);
4403 seq_printf(m, " {ent:%pS ret:%pS}",
4404 (void *)gops->entryfunc,
4405 (void *)gops->retfunc);
4406 continue;
4407 }
4408 #endif
4409 if (subops->trampoline) {
4410 seq_printf(m, " {%pS (%pS)}",
4411 (void *)subops->trampoline,
4412 (void *)subops->func);
4413 add_trampoline_func(m, subops, rec);
4414 } else {
4415 seq_printf(m, " {%pS}",
4416 (void *)subops->func);
4417 }
4418 }
4419 }
4420
t_show(struct seq_file * m,void * v)4421 static int t_show(struct seq_file *m, void *v)
4422 {
4423 struct ftrace_iterator *iter = m->private;
4424 struct dyn_ftrace *rec;
4425
4426 if (iter->flags & FTRACE_ITER_PROBE)
4427 return t_probe_show(m, iter);
4428
4429 if (iter->flags & FTRACE_ITER_MOD)
4430 return t_mod_show(m, iter);
4431
4432 if (iter->flags & FTRACE_ITER_PRINTALL) {
4433 if (iter->flags & FTRACE_ITER_NOTRACE)
4434 seq_puts(m, "#### no functions disabled ####\n");
4435 else
4436 seq_puts(m, "#### all functions enabled ####\n");
4437 return 0;
4438 }
4439
4440 rec = iter->func;
4441
4442 if (!rec)
4443 return 0;
4444
4445 if (iter->flags & FTRACE_ITER_ADDRS)
4446 seq_printf(m, "%lx ", rec->ip);
4447
4448 if (print_rec(m, rec->ip)) {
4449 /* This should only happen when a rec is disabled */
4450 WARN_ON_ONCE(!(rec->flags & FTRACE_FL_DISABLED));
4451 seq_putc(m, '\n');
4452 return 0;
4453 }
4454
4455 if (iter->flags & (FTRACE_ITER_ENABLED | FTRACE_ITER_TOUCHED)) {
4456 struct ftrace_ops *ops;
4457
4458 seq_printf(m, " (%ld)%s%s%s%s%s",
4459 ftrace_rec_count(rec),
4460 rec->flags & FTRACE_FL_REGS ? " R" : " ",
4461 rec->flags & FTRACE_FL_IPMODIFY ? " I" : " ",
4462 rec->flags & FTRACE_FL_DIRECT ? " D" : " ",
4463 rec->flags & FTRACE_FL_CALL_OPS ? " O" : " ",
4464 rec->flags & FTRACE_FL_MODIFIED ? " M " : " ");
4465 if (rec->flags & FTRACE_FL_TRAMP_EN) {
4466 ops = ftrace_find_tramp_ops_any(rec);
4467 if (ops) {
4468 do {
4469 seq_printf(m, "\ttramp: %pS (%pS)",
4470 (void *)ops->trampoline,
4471 (void *)ops->func);
4472 add_trampoline_func(m, ops, rec);
4473 print_subops(m, ops, rec);
4474 ops = ftrace_find_tramp_ops_next(rec, ops);
4475 } while (ops);
4476 } else
4477 seq_puts(m, "\ttramp: ERROR!");
4478 } else {
4479 add_trampoline_func(m, NULL, rec);
4480 }
4481 if (rec->flags & FTRACE_FL_CALL_OPS_EN) {
4482 ops = ftrace_find_unique_ops(rec);
4483 if (ops) {
4484 seq_printf(m, "\tops: %pS (%pS)",
4485 ops, ops->func);
4486 print_subops(m, ops, rec);
4487 } else {
4488 seq_puts(m, "\tops: ERROR!");
4489 }
4490 }
4491 if (rec->flags & FTRACE_FL_DIRECT) {
4492 unsigned long direct;
4493
4494 direct = ftrace_find_rec_direct(rec->ip);
4495 if (direct)
4496 seq_printf(m, "\n\tdirect-->%pS", (void *)direct);
4497 }
4498 }
4499
4500 seq_putc(m, '\n');
4501
4502 return 0;
4503 }
4504
4505 static const struct seq_operations show_ftrace_seq_ops = {
4506 .start = t_start,
4507 .next = t_next,
4508 .stop = t_stop,
4509 .show = t_show,
4510 };
4511
4512 static int
ftrace_avail_open(struct inode * inode,struct file * file)4513 ftrace_avail_open(struct inode *inode, struct file *file)
4514 {
4515 struct ftrace_iterator *iter;
4516 int ret;
4517
4518 ret = security_locked_down(LOCKDOWN_TRACEFS);
4519 if (ret)
4520 return ret;
4521
4522 if (unlikely(ftrace_disabled))
4523 return -ENODEV;
4524
4525 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4526 if (!iter)
4527 return -ENOMEM;
4528
4529 iter->pg = ftrace_pages_start;
4530 iter->ops = &global_ops;
4531
4532 return 0;
4533 }
4534
4535 static int
ftrace_enabled_open(struct inode * inode,struct file * file)4536 ftrace_enabled_open(struct inode *inode, struct file *file)
4537 {
4538 struct ftrace_iterator *iter;
4539
4540 /*
4541 * This shows us what functions are currently being
4542 * traced and by what. Not sure if we want lockdown
4543 * to hide such critical information for an admin.
4544 * Although, perhaps it can show information we don't
4545 * want people to see, but if something is tracing
4546 * something, we probably want to know about it.
4547 */
4548
4549 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4550 if (!iter)
4551 return -ENOMEM;
4552
4553 iter->pg = ftrace_pages_start;
4554 iter->flags = FTRACE_ITER_ENABLED;
4555 iter->ops = &global_ops;
4556
4557 return 0;
4558 }
4559
4560 static int
ftrace_touched_open(struct inode * inode,struct file * file)4561 ftrace_touched_open(struct inode *inode, struct file *file)
4562 {
4563 struct ftrace_iterator *iter;
4564
4565 /*
4566 * This shows us what functions have ever been enabled
4567 * (traced, direct, patched, etc). Not sure if we want lockdown
4568 * to hide such critical information for an admin.
4569 * Although, perhaps it can show information we don't
4570 * want people to see, but if something had traced
4571 * something, we probably want to know about it.
4572 */
4573
4574 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4575 if (!iter)
4576 return -ENOMEM;
4577
4578 iter->pg = ftrace_pages_start;
4579 iter->flags = FTRACE_ITER_TOUCHED;
4580 iter->ops = &global_ops;
4581
4582 return 0;
4583 }
4584
4585 static int
ftrace_avail_addrs_open(struct inode * inode,struct file * file)4586 ftrace_avail_addrs_open(struct inode *inode, struct file *file)
4587 {
4588 struct ftrace_iterator *iter;
4589 int ret;
4590
4591 ret = security_locked_down(LOCKDOWN_TRACEFS);
4592 if (ret)
4593 return ret;
4594
4595 if (unlikely(ftrace_disabled))
4596 return -ENODEV;
4597
4598 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
4599 if (!iter)
4600 return -ENOMEM;
4601
4602 iter->pg = ftrace_pages_start;
4603 iter->flags = FTRACE_ITER_ADDRS;
4604 iter->ops = &global_ops;
4605
4606 return 0;
4607 }
4608
4609 /**
4610 * ftrace_regex_open - initialize function tracer filter files
4611 * @ops: The ftrace_ops that hold the hash filters
4612 * @flag: The type of filter to process
4613 * @inode: The inode, usually passed in to your open routine
4614 * @file: The file, usually passed in to your open routine
4615 *
4616 * ftrace_regex_open() initializes the filter files for the
4617 * @ops. Depending on @flag it may process the filter hash or
4618 * the notrace hash of @ops. With this called from the open
4619 * routine, you can use ftrace_filter_write() for the write
4620 * routine if @flag has FTRACE_ITER_FILTER set, or
4621 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
4622 * tracing_lseek() should be used as the lseek routine, and
4623 * release must call ftrace_regex_release().
4624 *
4625 * Returns: 0 on success or a negative errno value on failure
4626 */
4627 int
ftrace_regex_open(struct ftrace_ops * ops,int flag,struct inode * inode,struct file * file)4628 ftrace_regex_open(struct ftrace_ops *ops, int flag,
4629 struct inode *inode, struct file *file)
4630 {
4631 struct ftrace_iterator *iter;
4632 struct ftrace_hash *hash;
4633 struct list_head *mod_head;
4634 struct trace_array *tr = ops->private;
4635 int ret = -ENOMEM;
4636
4637 ftrace_ops_init(ops);
4638
4639 if (unlikely(ftrace_disabled))
4640 return -ENODEV;
4641
4642 if (tracing_check_open_get_tr(tr))
4643 return -ENODEV;
4644
4645 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4646 if (!iter)
4647 goto out;
4648
4649 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX))
4650 goto out;
4651
4652 iter->ops = ops;
4653 iter->flags = flag;
4654 iter->tr = tr;
4655
4656 mutex_lock(&ops->func_hash->regex_lock);
4657
4658 if (flag & FTRACE_ITER_NOTRACE) {
4659 hash = ops->func_hash->notrace_hash;
4660 mod_head = tr ? &tr->mod_notrace : NULL;
4661 } else {
4662 hash = ops->func_hash->filter_hash;
4663 mod_head = tr ? &tr->mod_trace : NULL;
4664 }
4665
4666 iter->mod_list = mod_head;
4667
4668 if (file->f_mode & FMODE_WRITE) {
4669 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
4670
4671 if (file->f_flags & O_TRUNC) {
4672 iter->hash = alloc_ftrace_hash(size_bits);
4673 clear_ftrace_mod_list(mod_head);
4674 } else {
4675 iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
4676 }
4677 } else {
4678 if (hash)
4679 iter->hash = alloc_and_copy_ftrace_hash(hash->size_bits, hash);
4680 else
4681 iter->hash = EMPTY_HASH;
4682 }
4683
4684 if (!iter->hash) {
4685 trace_parser_put(&iter->parser);
4686 goto out_unlock;
4687 }
4688
4689 ret = 0;
4690
4691 if (file->f_mode & FMODE_READ) {
4692 iter->pg = ftrace_pages_start;
4693
4694 ret = seq_open(file, &show_ftrace_seq_ops);
4695 if (!ret) {
4696 struct seq_file *m = file->private_data;
4697 m->private = iter;
4698 } else {
4699 /* Failed */
4700 free_ftrace_hash(iter->hash);
4701 trace_parser_put(&iter->parser);
4702 }
4703 } else
4704 file->private_data = iter;
4705
4706 out_unlock:
4707 mutex_unlock(&ops->func_hash->regex_lock);
4708
4709 out:
4710 if (ret) {
4711 kfree(iter);
4712 if (tr)
4713 trace_array_put(tr);
4714 }
4715
4716 return ret;
4717 }
4718
4719 static int
ftrace_filter_open(struct inode * inode,struct file * file)4720 ftrace_filter_open(struct inode *inode, struct file *file)
4721 {
4722 struct ftrace_ops *ops = inode->i_private;
4723
4724 /* Checks for tracefs lockdown */
4725 return ftrace_regex_open(ops,
4726 FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
4727 inode, file);
4728 }
4729
4730 static int
ftrace_notrace_open(struct inode * inode,struct file * file)4731 ftrace_notrace_open(struct inode *inode, struct file *file)
4732 {
4733 struct ftrace_ops *ops = inode->i_private;
4734
4735 /* Checks for tracefs lockdown */
4736 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
4737 inode, file);
4738 }
4739
4740 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
4741 struct ftrace_glob {
4742 char *search;
4743 unsigned len;
4744 int type;
4745 };
4746
4747 /*
4748 * If symbols in an architecture don't correspond exactly to the user-visible
4749 * name of what they represent, it is possible to define this function to
4750 * perform the necessary adjustments.
4751 */
arch_ftrace_match_adjust(char * str,const char * search)4752 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
4753 {
4754 return str;
4755 }
4756
ftrace_match(char * str,struct ftrace_glob * g)4757 static int ftrace_match(char *str, struct ftrace_glob *g)
4758 {
4759 int matched = 0;
4760 int slen;
4761
4762 str = arch_ftrace_match_adjust(str, g->search);
4763
4764 switch (g->type) {
4765 case MATCH_FULL:
4766 if (strcmp(str, g->search) == 0)
4767 matched = 1;
4768 break;
4769 case MATCH_FRONT_ONLY:
4770 if (strncmp(str, g->search, g->len) == 0)
4771 matched = 1;
4772 break;
4773 case MATCH_MIDDLE_ONLY:
4774 if (strstr(str, g->search))
4775 matched = 1;
4776 break;
4777 case MATCH_END_ONLY:
4778 slen = strlen(str);
4779 if (slen >= g->len &&
4780 memcmp(str + slen - g->len, g->search, g->len) == 0)
4781 matched = 1;
4782 break;
4783 case MATCH_GLOB:
4784 if (glob_match(g->search, str))
4785 matched = 1;
4786 break;
4787 }
4788
4789 return matched;
4790 }
4791
4792 static int
enter_record(struct ftrace_hash * hash,struct dyn_ftrace * rec,int clear_filter)4793 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
4794 {
4795 struct ftrace_func_entry *entry;
4796 int ret = 0;
4797
4798 entry = ftrace_lookup_ip(hash, rec->ip);
4799 if (clear_filter) {
4800 /* Do nothing if it doesn't exist */
4801 if (!entry)
4802 return 0;
4803
4804 free_hash_entry(hash, entry);
4805 } else {
4806 /* Do nothing if it exists */
4807 if (entry)
4808 return 0;
4809 if (add_hash_entry(hash, rec->ip) == NULL)
4810 ret = -ENOMEM;
4811 }
4812 return ret;
4813 }
4814
4815 static int
add_rec_by_index(struct ftrace_hash * hash,struct ftrace_glob * func_g,int clear_filter)4816 add_rec_by_index(struct ftrace_hash *hash, struct ftrace_glob *func_g,
4817 int clear_filter)
4818 {
4819 long index;
4820 struct ftrace_page *pg;
4821 struct dyn_ftrace *rec;
4822
4823 /* The index starts at 1 */
4824 if (kstrtoul(func_g->search, 0, &index) || --index < 0)
4825 return 0;
4826
4827 do_for_each_ftrace_rec(pg, rec) {
4828 if (pg->index <= index) {
4829 index -= pg->index;
4830 /* this is a double loop, break goes to the next page */
4831 break;
4832 }
4833 rec = &pg->records[index];
4834 enter_record(hash, rec, clear_filter);
4835 return 1;
4836 } while_for_each_ftrace_rec();
4837 return 0;
4838 }
4839
4840 #ifdef FTRACE_MCOUNT_MAX_OFFSET
lookup_ip(unsigned long ip,char ** modname,char * str)4841 static int lookup_ip(unsigned long ip, char **modname, char *str)
4842 {
4843 unsigned long offset;
4844
4845 kallsyms_lookup(ip, NULL, &offset, modname, str);
4846 if (offset > FTRACE_MCOUNT_MAX_OFFSET)
4847 return -1;
4848 return 0;
4849 }
4850 #else
lookup_ip(unsigned long ip,char ** modname,char * str)4851 static int lookup_ip(unsigned long ip, char **modname, char *str)
4852 {
4853 kallsyms_lookup(ip, NULL, NULL, modname, str);
4854 return 0;
4855 }
4856 #endif
4857
4858 static int
ftrace_match_record(struct dyn_ftrace * rec,struct ftrace_glob * func_g,struct ftrace_glob * mod_g,int exclude_mod)4859 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
4860 struct ftrace_glob *mod_g, int exclude_mod)
4861 {
4862 char str[KSYM_SYMBOL_LEN];
4863 char *modname;
4864
4865 if (lookup_ip(rec->ip, &modname, str)) {
4866 /* This should only happen when a rec is disabled */
4867 WARN_ON_ONCE(system_state == SYSTEM_RUNNING &&
4868 !(rec->flags & FTRACE_FL_DISABLED));
4869 return 0;
4870 }
4871
4872 if (mod_g) {
4873 int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
4874
4875 /* blank module name to match all modules */
4876 if (!mod_g->len) {
4877 /* blank module globbing: modname xor exclude_mod */
4878 if (!exclude_mod != !modname)
4879 goto func_match;
4880 return 0;
4881 }
4882
4883 /*
4884 * exclude_mod is set to trace everything but the given
4885 * module. If it is set and the module matches, then
4886 * return 0. If it is not set, and the module doesn't match
4887 * also return 0. Otherwise, check the function to see if
4888 * that matches.
4889 */
4890 if (!mod_matches == !exclude_mod)
4891 return 0;
4892 func_match:
4893 /* blank search means to match all funcs in the mod */
4894 if (!func_g->len)
4895 return 1;
4896 }
4897
4898 return ftrace_match(str, func_g);
4899 }
4900
4901 static int
match_records(struct ftrace_hash * hash,char * func,int len,char * mod)4902 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
4903 {
4904 struct ftrace_page *pg;
4905 struct dyn_ftrace *rec;
4906 struct ftrace_glob func_g = { .type = MATCH_FULL };
4907 struct ftrace_glob mod_g = { .type = MATCH_FULL };
4908 struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
4909 int exclude_mod = 0;
4910 int found = 0;
4911 int ret;
4912 int clear_filter = 0;
4913
4914 if (func) {
4915 func_g.type = filter_parse_regex(func, len, &func_g.search,
4916 &clear_filter);
4917 func_g.len = strlen(func_g.search);
4918 }
4919
4920 if (mod) {
4921 mod_g.type = filter_parse_regex(mod, strlen(mod),
4922 &mod_g.search, &exclude_mod);
4923 mod_g.len = strlen(mod_g.search);
4924 }
4925
4926 guard(mutex)(&ftrace_lock);
4927
4928 if (unlikely(ftrace_disabled))
4929 return 0;
4930
4931 if (func_g.type == MATCH_INDEX)
4932 return add_rec_by_index(hash, &func_g, clear_filter);
4933
4934 do_for_each_ftrace_rec(pg, rec) {
4935
4936 if (rec->flags & FTRACE_FL_DISABLED)
4937 continue;
4938
4939 if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
4940 ret = enter_record(hash, rec, clear_filter);
4941 if (ret < 0)
4942 return ret;
4943 found = 1;
4944 }
4945 cond_resched();
4946 } while_for_each_ftrace_rec();
4947
4948 return found;
4949 }
4950
4951 static int
ftrace_match_records(struct ftrace_hash * hash,char * buff,int len)4952 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
4953 {
4954 return match_records(hash, buff, len, NULL);
4955 }
4956
ftrace_ops_update_code(struct ftrace_ops * ops,struct ftrace_ops_hash * old_hash)4957 static void ftrace_ops_update_code(struct ftrace_ops *ops,
4958 struct ftrace_ops_hash *old_hash)
4959 {
4960 struct ftrace_ops *op;
4961
4962 if (!ftrace_enabled)
4963 return;
4964
4965 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
4966 ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
4967 return;
4968 }
4969
4970 /*
4971 * If this is the shared global_ops filter, then we need to
4972 * check if there is another ops that shares it, is enabled.
4973 * If so, we still need to run the modify code.
4974 */
4975 if (ops->func_hash != &global_ops.local_hash)
4976 return;
4977
4978 do_for_each_ftrace_op(op, ftrace_ops_list) {
4979 if (op->func_hash == &global_ops.local_hash &&
4980 op->flags & FTRACE_OPS_FL_ENABLED) {
4981 ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
4982 /* Only need to do this once */
4983 return;
4984 }
4985 } while_for_each_ftrace_op(op);
4986 }
4987
ftrace_hash_move_and_update_ops(struct ftrace_ops * ops,struct ftrace_hash ** orig_hash,struct ftrace_hash * hash,int enable)4988 static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
4989 struct ftrace_hash **orig_hash,
4990 struct ftrace_hash *hash,
4991 int enable)
4992 {
4993 if (ops->flags & FTRACE_OPS_FL_SUBOP)
4994 return ftrace_hash_move_and_update_subops(ops, orig_hash, hash);
4995
4996 /*
4997 * If this ops is not enabled, it could be sharing its filters
4998 * with a subop. If that's the case, update the subop instead of
4999 * this ops. Shared filters are only allowed to have one ops set
5000 * at a time, and if we update the ops that is not enabled,
5001 * it will not affect subops that share it.
5002 */
5003 if (!(ops->flags & FTRACE_OPS_FL_ENABLED)) {
5004 struct ftrace_ops *op;
5005
5006 /* Check if any other manager subops maps to this hash */
5007 do_for_each_ftrace_op(op, ftrace_ops_list) {
5008 struct ftrace_ops *subops;
5009
5010 list_for_each_entry(subops, &op->subop_list, list) {
5011 if ((subops->flags & FTRACE_OPS_FL_ENABLED) &&
5012 subops->func_hash == ops->func_hash) {
5013 return ftrace_hash_move_and_update_subops(subops, orig_hash, hash);
5014 }
5015 }
5016 } while_for_each_ftrace_op(op);
5017 }
5018
5019 return __ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
5020 }
5021
cache_mod(struct trace_array * tr,const char * func,char * module,int enable)5022 static int cache_mod(struct trace_array *tr,
5023 const char *func, char *module, int enable)
5024 {
5025 struct ftrace_mod_load *ftrace_mod, *n;
5026 struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
5027
5028 guard(mutex)(&ftrace_lock);
5029
5030 /* We do not cache inverse filters */
5031 if (func[0] == '!') {
5032 int ret = -EINVAL;
5033
5034 func++;
5035
5036 /* Look to remove this hash */
5037 list_for_each_entry_safe(ftrace_mod, n, head, list) {
5038 if (strcmp(ftrace_mod->module, module) != 0)
5039 continue;
5040
5041 /* no func matches all */
5042 if (strcmp(func, "*") == 0 ||
5043 (ftrace_mod->func &&
5044 strcmp(ftrace_mod->func, func) == 0)) {
5045 ret = 0;
5046 free_ftrace_mod(ftrace_mod);
5047 continue;
5048 }
5049 }
5050 return ret;
5051 }
5052
5053 /* We only care about modules that have not been loaded yet */
5054 if (module_exists(module))
5055 return -EINVAL;
5056
5057 /* Save this string off, and execute it when the module is loaded */
5058 return ftrace_add_mod(tr, func, module, enable);
5059 }
5060
5061 #ifdef CONFIG_MODULES
process_mod_list(struct list_head * head,struct ftrace_ops * ops,char * mod,bool enable)5062 static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
5063 char *mod, bool enable)
5064 {
5065 struct ftrace_mod_load *ftrace_mod, *n;
5066 struct ftrace_hash **orig_hash, *new_hash;
5067 LIST_HEAD(process_mods);
5068 char *func;
5069
5070 mutex_lock(&ops->func_hash->regex_lock);
5071
5072 if (enable)
5073 orig_hash = &ops->func_hash->filter_hash;
5074 else
5075 orig_hash = &ops->func_hash->notrace_hash;
5076
5077 new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
5078 *orig_hash);
5079 if (!new_hash)
5080 goto out; /* warn? */
5081
5082 mutex_lock(&ftrace_lock);
5083
5084 list_for_each_entry_safe(ftrace_mod, n, head, list) {
5085
5086 if (strcmp(ftrace_mod->module, mod) != 0)
5087 continue;
5088
5089 if (ftrace_mod->func)
5090 func = kstrdup(ftrace_mod->func, GFP_KERNEL);
5091 else
5092 func = kstrdup("*", GFP_KERNEL);
5093
5094 if (!func) /* warn? */
5095 continue;
5096
5097 list_move(&ftrace_mod->list, &process_mods);
5098
5099 /* Use the newly allocated func, as it may be "*" */
5100 kfree(ftrace_mod->func);
5101 ftrace_mod->func = func;
5102 }
5103
5104 mutex_unlock(&ftrace_lock);
5105
5106 list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
5107
5108 func = ftrace_mod->func;
5109
5110 /* Grabs ftrace_lock, which is why we have this extra step */
5111 match_records(new_hash, func, strlen(func), mod);
5112 free_ftrace_mod(ftrace_mod);
5113 }
5114
5115 if (enable && list_empty(head))
5116 new_hash->flags &= ~FTRACE_HASH_FL_MOD;
5117
5118 mutex_lock(&ftrace_lock);
5119
5120 ftrace_hash_move_and_update_ops(ops, orig_hash,
5121 new_hash, enable);
5122 mutex_unlock(&ftrace_lock);
5123
5124 out:
5125 mutex_unlock(&ops->func_hash->regex_lock);
5126
5127 free_ftrace_hash(new_hash);
5128 }
5129
process_cached_mods(const char * mod_name)5130 static void process_cached_mods(const char *mod_name)
5131 {
5132 struct trace_array *tr;
5133 char *mod;
5134
5135 mod = kstrdup(mod_name, GFP_KERNEL);
5136 if (!mod)
5137 return;
5138
5139 mutex_lock(&trace_types_lock);
5140 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5141 if (!list_empty(&tr->mod_trace))
5142 process_mod_list(&tr->mod_trace, tr->ops, mod, true);
5143 if (!list_empty(&tr->mod_notrace))
5144 process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
5145 }
5146 mutex_unlock(&trace_types_lock);
5147
5148 kfree(mod);
5149 }
5150 #endif
5151
5152 /*
5153 * We register the module command as a template to show others how
5154 * to register the a command as well.
5155 */
5156
5157 static int
ftrace_mod_callback(struct trace_array * tr,struct ftrace_hash * hash,char * func_orig,char * cmd,char * module,int enable)5158 ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
5159 char *func_orig, char *cmd, char *module, int enable)
5160 {
5161 char *func;
5162 int ret;
5163
5164 if (!tr)
5165 return -ENODEV;
5166
5167 /* match_records() modifies func, and we need the original */
5168 func = kstrdup(func_orig, GFP_KERNEL);
5169 if (!func)
5170 return -ENOMEM;
5171
5172 /*
5173 * cmd == 'mod' because we only registered this func
5174 * for the 'mod' ftrace_func_command.
5175 * But if you register one func with multiple commands,
5176 * you can tell which command was used by the cmd
5177 * parameter.
5178 */
5179 ret = match_records(hash, func, strlen(func), module);
5180 kfree(func);
5181
5182 if (!ret)
5183 return cache_mod(tr, func_orig, module, enable);
5184 if (ret < 0)
5185 return ret;
5186 return 0;
5187 }
5188
5189 static struct ftrace_func_command ftrace_mod_cmd = {
5190 .name = "mod",
5191 .func = ftrace_mod_callback,
5192 };
5193
ftrace_mod_cmd_init(void)5194 static int __init ftrace_mod_cmd_init(void)
5195 {
5196 return register_ftrace_command(&ftrace_mod_cmd);
5197 }
5198 core_initcall(ftrace_mod_cmd_init);
5199
function_trace_probe_call(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)5200 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
5201 struct ftrace_ops *op, struct ftrace_regs *fregs)
5202 {
5203 struct ftrace_probe_ops *probe_ops;
5204 struct ftrace_func_probe *probe;
5205
5206 probe = container_of(op, struct ftrace_func_probe, ops);
5207 probe_ops = probe->probe_ops;
5208
5209 /*
5210 * Disable preemption for these calls to prevent a RCU grace
5211 * period. This syncs the hash iteration and freeing of items
5212 * on the hash. rcu_read_lock is too dangerous here.
5213 */
5214 preempt_disable_notrace();
5215 probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
5216 preempt_enable_notrace();
5217 }
5218
5219 struct ftrace_func_map {
5220 struct ftrace_func_entry entry;
5221 void *data;
5222 };
5223
5224 /*
5225 * Note, ftrace_func_mapper is freed by free_ftrace_hash(&mapper->hash).
5226 * The hash field must be the first field.
5227 */
5228 struct ftrace_func_mapper {
5229 struct ftrace_hash hash; /* Must be first! */
5230 };
5231
5232 /**
5233 * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
5234 *
5235 * Returns: a ftrace_func_mapper descriptor that can be used to map ips to data.
5236 */
allocate_ftrace_func_mapper(void)5237 struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
5238 {
5239 struct ftrace_hash *hash;
5240
5241 /*
5242 * The mapper is simply a ftrace_hash, but since the entries
5243 * in the hash are not ftrace_func_entry type, we define it
5244 * as a separate structure.
5245 */
5246 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5247 return (struct ftrace_func_mapper *)hash;
5248 }
5249
5250 /**
5251 * ftrace_func_mapper_find_ip - Find some data mapped to an ip
5252 * @mapper: The mapper that has the ip maps
5253 * @ip: the instruction pointer to find the data for
5254 *
5255 * Returns: the data mapped to @ip if found otherwise NULL. The return
5256 * is actually the address of the mapper data pointer. The address is
5257 * returned for use cases where the data is no bigger than a long, and
5258 * the user can use the data pointer as its data instead of having to
5259 * allocate more memory for the reference.
5260 */
ftrace_func_mapper_find_ip(struct ftrace_func_mapper * mapper,unsigned long ip)5261 void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
5262 unsigned long ip)
5263 {
5264 struct ftrace_func_entry *entry;
5265 struct ftrace_func_map *map;
5266
5267 entry = ftrace_lookup_ip(&mapper->hash, ip);
5268 if (!entry)
5269 return NULL;
5270
5271 map = (struct ftrace_func_map *)entry;
5272 return &map->data;
5273 }
5274
5275 /**
5276 * ftrace_func_mapper_add_ip - Map some data to an ip
5277 * @mapper: The mapper that has the ip maps
5278 * @ip: The instruction pointer address to map @data to
5279 * @data: The data to map to @ip
5280 *
5281 * Returns: 0 on success otherwise an error.
5282 */
ftrace_func_mapper_add_ip(struct ftrace_func_mapper * mapper,unsigned long ip,void * data)5283 int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
5284 unsigned long ip, void *data)
5285 {
5286 struct ftrace_func_entry *entry;
5287 struct ftrace_func_map *map;
5288
5289 entry = ftrace_lookup_ip(&mapper->hash, ip);
5290 if (entry)
5291 return -EBUSY;
5292
5293 map = kmalloc(sizeof(*map), GFP_KERNEL);
5294 if (!map)
5295 return -ENOMEM;
5296
5297 map->entry.ip = ip;
5298 map->data = data;
5299
5300 __add_hash_entry(&mapper->hash, &map->entry);
5301
5302 return 0;
5303 }
5304
5305 /**
5306 * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
5307 * @mapper: The mapper that has the ip maps
5308 * @ip: The instruction pointer address to remove the data from
5309 *
5310 * Returns: the data if it is found, otherwise NULL.
5311 * Note, if the data pointer is used as the data itself, (see
5312 * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
5313 * if the data pointer was set to zero.
5314 */
ftrace_func_mapper_remove_ip(struct ftrace_func_mapper * mapper,unsigned long ip)5315 void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
5316 unsigned long ip)
5317 {
5318 struct ftrace_func_entry *entry;
5319 struct ftrace_func_map *map;
5320 void *data;
5321
5322 entry = ftrace_lookup_ip(&mapper->hash, ip);
5323 if (!entry)
5324 return NULL;
5325
5326 map = (struct ftrace_func_map *)entry;
5327 data = map->data;
5328
5329 remove_hash_entry(&mapper->hash, entry);
5330 kfree(entry);
5331
5332 return data;
5333 }
5334
5335 /**
5336 * free_ftrace_func_mapper - free a mapping of ips and data
5337 * @mapper: The mapper that has the ip maps
5338 * @free_func: A function to be called on each data item.
5339 *
5340 * This is used to free the function mapper. The @free_func is optional
5341 * and can be used if the data needs to be freed as well.
5342 */
free_ftrace_func_mapper(struct ftrace_func_mapper * mapper,ftrace_mapper_func free_func)5343 void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
5344 ftrace_mapper_func free_func)
5345 {
5346 struct ftrace_func_entry *entry;
5347 struct ftrace_func_map *map;
5348 struct hlist_head *hhd;
5349 int size, i;
5350
5351 if (!mapper)
5352 return;
5353
5354 if (free_func && mapper->hash.count) {
5355 size = 1 << mapper->hash.size_bits;
5356 for (i = 0; i < size; i++) {
5357 hhd = &mapper->hash.buckets[i];
5358 hlist_for_each_entry(entry, hhd, hlist) {
5359 map = (struct ftrace_func_map *)entry;
5360 free_func(map);
5361 }
5362 }
5363 }
5364 /* This also frees the mapper itself */
5365 free_ftrace_hash(&mapper->hash);
5366 }
5367
release_probe(struct ftrace_func_probe * probe)5368 static void release_probe(struct ftrace_func_probe *probe)
5369 {
5370 struct ftrace_probe_ops *probe_ops;
5371
5372 guard(mutex)(&ftrace_lock);
5373
5374 WARN_ON(probe->ref <= 0);
5375
5376 /* Subtract the ref that was used to protect this instance */
5377 probe->ref--;
5378
5379 if (!probe->ref) {
5380 probe_ops = probe->probe_ops;
5381 /*
5382 * Sending zero as ip tells probe_ops to free
5383 * the probe->data itself
5384 */
5385 if (probe_ops->free)
5386 probe_ops->free(probe_ops, probe->tr, 0, probe->data);
5387 list_del(&probe->list);
5388 kfree(probe);
5389 }
5390 }
5391
acquire_probe_locked(struct ftrace_func_probe * probe)5392 static void acquire_probe_locked(struct ftrace_func_probe *probe)
5393 {
5394 /*
5395 * Add one ref to keep it from being freed when releasing the
5396 * ftrace_lock mutex.
5397 */
5398 probe->ref++;
5399 }
5400
5401 int
register_ftrace_function_probe(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops,void * data)5402 register_ftrace_function_probe(char *glob, struct trace_array *tr,
5403 struct ftrace_probe_ops *probe_ops,
5404 void *data)
5405 {
5406 struct ftrace_func_probe *probe = NULL, *iter;
5407 struct ftrace_func_entry *entry;
5408 struct ftrace_hash **orig_hash;
5409 struct ftrace_hash *old_hash;
5410 struct ftrace_hash *hash;
5411 int count = 0;
5412 int size;
5413 int ret;
5414 int i;
5415
5416 if (WARN_ON(!tr))
5417 return -EINVAL;
5418
5419 /* We do not support '!' for function probes */
5420 if (WARN_ON(glob[0] == '!'))
5421 return -EINVAL;
5422
5423
5424 mutex_lock(&ftrace_lock);
5425 /* Check if the probe_ops is already registered */
5426 list_for_each_entry(iter, &tr->func_probes, list) {
5427 if (iter->probe_ops == probe_ops) {
5428 probe = iter;
5429 break;
5430 }
5431 }
5432 if (!probe) {
5433 probe = kzalloc(sizeof(*probe), GFP_KERNEL);
5434 if (!probe) {
5435 mutex_unlock(&ftrace_lock);
5436 return -ENOMEM;
5437 }
5438 probe->probe_ops = probe_ops;
5439 probe->ops.func = function_trace_probe_call;
5440 probe->tr = tr;
5441 ftrace_ops_init(&probe->ops);
5442 list_add(&probe->list, &tr->func_probes);
5443 }
5444
5445 acquire_probe_locked(probe);
5446
5447 mutex_unlock(&ftrace_lock);
5448
5449 /*
5450 * Note, there's a small window here that the func_hash->filter_hash
5451 * may be NULL or empty. Need to be careful when reading the loop.
5452 */
5453 mutex_lock(&probe->ops.func_hash->regex_lock);
5454
5455 orig_hash = &probe->ops.func_hash->filter_hash;
5456 old_hash = *orig_hash;
5457 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
5458
5459 if (!hash) {
5460 ret = -ENOMEM;
5461 goto out;
5462 }
5463
5464 ret = ftrace_match_records(hash, glob, strlen(glob));
5465
5466 /* Nothing found? */
5467 if (!ret)
5468 ret = -EINVAL;
5469
5470 if (ret < 0)
5471 goto out;
5472
5473 size = 1 << hash->size_bits;
5474 for (i = 0; i < size; i++) {
5475 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5476 if (ftrace_lookup_ip(old_hash, entry->ip))
5477 continue;
5478 /*
5479 * The caller might want to do something special
5480 * for each function we find. We call the callback
5481 * to give the caller an opportunity to do so.
5482 */
5483 if (probe_ops->init) {
5484 ret = probe_ops->init(probe_ops, tr,
5485 entry->ip, data,
5486 &probe->data);
5487 if (ret < 0) {
5488 if (probe_ops->free && count)
5489 probe_ops->free(probe_ops, tr,
5490 0, probe->data);
5491 probe->data = NULL;
5492 goto out;
5493 }
5494 }
5495 count++;
5496 }
5497 }
5498
5499 mutex_lock(&ftrace_lock);
5500
5501 if (!count) {
5502 /* Nothing was added? */
5503 ret = -EINVAL;
5504 goto out_unlock;
5505 }
5506
5507 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
5508 hash, 1);
5509 if (ret < 0)
5510 goto err_unlock;
5511
5512 /* One ref for each new function traced */
5513 probe->ref += count;
5514
5515 if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
5516 ret = ftrace_startup(&probe->ops, 0);
5517
5518 out_unlock:
5519 mutex_unlock(&ftrace_lock);
5520
5521 if (!ret)
5522 ret = count;
5523 out:
5524 mutex_unlock(&probe->ops.func_hash->regex_lock);
5525 free_ftrace_hash(hash);
5526
5527 release_probe(probe);
5528
5529 return ret;
5530
5531 err_unlock:
5532 if (!probe_ops->free || !count)
5533 goto out_unlock;
5534
5535 /* Failed to do the move, need to call the free functions */
5536 for (i = 0; i < size; i++) {
5537 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5538 if (ftrace_lookup_ip(old_hash, entry->ip))
5539 continue;
5540 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
5541 }
5542 }
5543 goto out_unlock;
5544 }
5545
5546 int
unregister_ftrace_function_probe_func(char * glob,struct trace_array * tr,struct ftrace_probe_ops * probe_ops)5547 unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
5548 struct ftrace_probe_ops *probe_ops)
5549 {
5550 struct ftrace_func_probe *probe = NULL, *iter;
5551 struct ftrace_ops_hash old_hash_ops;
5552 struct ftrace_func_entry *entry;
5553 struct ftrace_glob func_g;
5554 struct ftrace_hash **orig_hash;
5555 struct ftrace_hash *old_hash;
5556 struct ftrace_hash *hash = NULL;
5557 struct hlist_node *tmp;
5558 struct hlist_head hhd;
5559 char str[KSYM_SYMBOL_LEN];
5560 int count = 0;
5561 int i, ret = -ENODEV;
5562 int size;
5563
5564 if (!glob || !strlen(glob) || !strcmp(glob, "*"))
5565 func_g.search = NULL;
5566 else {
5567 int not;
5568
5569 func_g.type = filter_parse_regex(glob, strlen(glob),
5570 &func_g.search, ¬);
5571 func_g.len = strlen(func_g.search);
5572
5573 /* we do not support '!' for function probes */
5574 if (WARN_ON(not))
5575 return -EINVAL;
5576 }
5577
5578 mutex_lock(&ftrace_lock);
5579 /* Check if the probe_ops is already registered */
5580 list_for_each_entry(iter, &tr->func_probes, list) {
5581 if (iter->probe_ops == probe_ops) {
5582 probe = iter;
5583 break;
5584 }
5585 }
5586 if (!probe)
5587 goto err_unlock_ftrace;
5588
5589 ret = -EINVAL;
5590 if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
5591 goto err_unlock_ftrace;
5592
5593 acquire_probe_locked(probe);
5594
5595 mutex_unlock(&ftrace_lock);
5596
5597 mutex_lock(&probe->ops.func_hash->regex_lock);
5598
5599 orig_hash = &probe->ops.func_hash->filter_hash;
5600 old_hash = *orig_hash;
5601
5602 if (ftrace_hash_empty(old_hash))
5603 goto out_unlock;
5604
5605 old_hash_ops.filter_hash = old_hash;
5606 /* Probes only have filters */
5607 old_hash_ops.notrace_hash = NULL;
5608
5609 ret = -ENOMEM;
5610 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
5611 if (!hash)
5612 goto out_unlock;
5613
5614 INIT_HLIST_HEAD(&hhd);
5615
5616 size = 1 << hash->size_bits;
5617 for (i = 0; i < size; i++) {
5618 hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
5619
5620 if (func_g.search) {
5621 kallsyms_lookup(entry->ip, NULL, NULL,
5622 NULL, str);
5623 if (!ftrace_match(str, &func_g))
5624 continue;
5625 }
5626 count++;
5627 remove_hash_entry(hash, entry);
5628 hlist_add_head(&entry->hlist, &hhd);
5629 }
5630 }
5631
5632 /* Nothing found? */
5633 if (!count) {
5634 ret = -EINVAL;
5635 goto out_unlock;
5636 }
5637
5638 mutex_lock(&ftrace_lock);
5639
5640 WARN_ON(probe->ref < count);
5641
5642 probe->ref -= count;
5643
5644 if (ftrace_hash_empty(hash))
5645 ftrace_shutdown(&probe->ops, 0);
5646
5647 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
5648 hash, 1);
5649
5650 /* still need to update the function call sites */
5651 if (ftrace_enabled && !ftrace_hash_empty(hash))
5652 ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
5653 &old_hash_ops);
5654 synchronize_rcu();
5655
5656 hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
5657 hlist_del(&entry->hlist);
5658 if (probe_ops->free)
5659 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
5660 kfree(entry);
5661 }
5662 mutex_unlock(&ftrace_lock);
5663
5664 out_unlock:
5665 mutex_unlock(&probe->ops.func_hash->regex_lock);
5666 free_ftrace_hash(hash);
5667
5668 release_probe(probe);
5669
5670 return ret;
5671
5672 err_unlock_ftrace:
5673 mutex_unlock(&ftrace_lock);
5674 return ret;
5675 }
5676
clear_ftrace_function_probes(struct trace_array * tr)5677 void clear_ftrace_function_probes(struct trace_array *tr)
5678 {
5679 struct ftrace_func_probe *probe, *n;
5680
5681 list_for_each_entry_safe(probe, n, &tr->func_probes, list)
5682 unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
5683 }
5684
5685 static LIST_HEAD(ftrace_commands);
5686 static DEFINE_MUTEX(ftrace_cmd_mutex);
5687
5688 /*
5689 * Currently we only register ftrace commands from __init, so mark this
5690 * __init too.
5691 */
register_ftrace_command(struct ftrace_func_command * cmd)5692 __init int register_ftrace_command(struct ftrace_func_command *cmd)
5693 {
5694 struct ftrace_func_command *p;
5695
5696 guard(mutex)(&ftrace_cmd_mutex);
5697 list_for_each_entry(p, &ftrace_commands, list) {
5698 if (strcmp(cmd->name, p->name) == 0)
5699 return -EBUSY;
5700 }
5701 list_add(&cmd->list, &ftrace_commands);
5702
5703 return 0;
5704 }
5705
5706 /*
5707 * Currently we only unregister ftrace commands from __init, so mark
5708 * this __init too.
5709 */
unregister_ftrace_command(struct ftrace_func_command * cmd)5710 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
5711 {
5712 struct ftrace_func_command *p, *n;
5713
5714 guard(mutex)(&ftrace_cmd_mutex);
5715
5716 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
5717 if (strcmp(cmd->name, p->name) == 0) {
5718 list_del_init(&p->list);
5719 return 0;
5720 }
5721 }
5722
5723 return -ENODEV;
5724 }
5725
ftrace_process_regex(struct ftrace_iterator * iter,char * buff,int len,int enable)5726 static int ftrace_process_regex(struct ftrace_iterator *iter,
5727 char *buff, int len, int enable)
5728 {
5729 struct ftrace_hash *hash = iter->hash;
5730 struct trace_array *tr = iter->ops->private;
5731 char *func, *command, *next = buff;
5732 struct ftrace_func_command *p;
5733 int ret;
5734
5735 func = strsep(&next, ":");
5736
5737 if (!next) {
5738 ret = ftrace_match_records(hash, func, len);
5739 if (!ret)
5740 ret = -EINVAL;
5741 if (ret < 0)
5742 return ret;
5743 return 0;
5744 }
5745
5746 /* command found */
5747
5748 command = strsep(&next, ":");
5749
5750 guard(mutex)(&ftrace_cmd_mutex);
5751
5752 list_for_each_entry(p, &ftrace_commands, list) {
5753 if (strcmp(p->name, command) == 0)
5754 return p->func(tr, hash, func, command, next, enable);
5755 }
5756
5757 return -EINVAL;
5758 }
5759
5760 static ssize_t
ftrace_regex_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos,int enable)5761 ftrace_regex_write(struct file *file, const char __user *ubuf,
5762 size_t cnt, loff_t *ppos, int enable)
5763 {
5764 struct ftrace_iterator *iter;
5765 struct trace_parser *parser;
5766 ssize_t ret, read;
5767
5768 if (!cnt)
5769 return 0;
5770
5771 if (file->f_mode & FMODE_READ) {
5772 struct seq_file *m = file->private_data;
5773 iter = m->private;
5774 } else
5775 iter = file->private_data;
5776
5777 if (unlikely(ftrace_disabled))
5778 return -ENODEV;
5779
5780 /* iter->hash is a local copy, so we don't need regex_lock */
5781
5782 parser = &iter->parser;
5783 read = trace_get_user(parser, ubuf, cnt, ppos);
5784
5785 if (read >= 0 && trace_parser_loaded(parser) &&
5786 !trace_parser_cont(parser)) {
5787 ret = ftrace_process_regex(iter, parser->buffer,
5788 parser->idx, enable);
5789 trace_parser_clear(parser);
5790 if (ret < 0)
5791 return ret;
5792 }
5793
5794 return read;
5795 }
5796
5797 ssize_t
ftrace_filter_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5798 ftrace_filter_write(struct file *file, const char __user *ubuf,
5799 size_t cnt, loff_t *ppos)
5800 {
5801 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
5802 }
5803
5804 ssize_t
ftrace_notrace_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)5805 ftrace_notrace_write(struct file *file, const char __user *ubuf,
5806 size_t cnt, loff_t *ppos)
5807 {
5808 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
5809 }
5810
5811 static int
__ftrace_match_addr(struct ftrace_hash * hash,unsigned long ip,int remove)5812 __ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
5813 {
5814 struct ftrace_func_entry *entry;
5815
5816 ip = ftrace_location(ip);
5817 if (!ip)
5818 return -EINVAL;
5819
5820 if (remove) {
5821 entry = ftrace_lookup_ip(hash, ip);
5822 if (!entry)
5823 return -ENOENT;
5824 free_hash_entry(hash, entry);
5825 return 0;
5826 } else if (__ftrace_lookup_ip(hash, ip) != NULL) {
5827 /* Already exists */
5828 return 0;
5829 }
5830
5831 entry = add_hash_entry(hash, ip);
5832 return entry ? 0 : -ENOMEM;
5833 }
5834
5835 static int
ftrace_match_addr(struct ftrace_hash * hash,unsigned long * ips,unsigned int cnt,int remove)5836 ftrace_match_addr(struct ftrace_hash *hash, unsigned long *ips,
5837 unsigned int cnt, int remove)
5838 {
5839 unsigned int i;
5840 int err;
5841
5842 for (i = 0; i < cnt; i++) {
5843 err = __ftrace_match_addr(hash, ips[i], remove);
5844 if (err) {
5845 /*
5846 * This expects the @hash is a temporary hash and if this
5847 * fails the caller must free the @hash.
5848 */
5849 return err;
5850 }
5851 }
5852 return 0;
5853 }
5854
5855 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)5856 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
5857 unsigned long *ips, unsigned int cnt,
5858 int remove, int reset, int enable, char *mod)
5859 {
5860 struct ftrace_hash **orig_hash;
5861 struct ftrace_hash *hash;
5862 int ret;
5863
5864 if (unlikely(ftrace_disabled))
5865 return -ENODEV;
5866
5867 mutex_lock(&ops->func_hash->regex_lock);
5868
5869 if (enable)
5870 orig_hash = &ops->func_hash->filter_hash;
5871 else
5872 orig_hash = &ops->func_hash->notrace_hash;
5873
5874 if (reset)
5875 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5876 else
5877 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
5878
5879 if (!hash) {
5880 ret = -ENOMEM;
5881 goto out_regex_unlock;
5882 }
5883
5884 if (buf && !match_records(hash, buf, len, mod)) {
5885 /* If this was for a module and nothing was enabled, flag it */
5886 if (mod)
5887 (*orig_hash)->flags |= FTRACE_HASH_FL_MOD;
5888
5889 /*
5890 * Even if it is a mod, return error to let caller know
5891 * nothing was added
5892 */
5893 ret = -EINVAL;
5894 goto out_regex_unlock;
5895 }
5896 if (ips) {
5897 ret = ftrace_match_addr(hash, ips, cnt, remove);
5898 if (ret < 0)
5899 goto out_regex_unlock;
5900 }
5901
5902 mutex_lock(&ftrace_lock);
5903 ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
5904 mutex_unlock(&ftrace_lock);
5905
5906 out_regex_unlock:
5907 mutex_unlock(&ops->func_hash->regex_lock);
5908
5909 free_ftrace_hash(hash);
5910 return ret;
5911 }
5912
5913 static int
ftrace_set_addr(struct ftrace_ops * ops,unsigned long * ips,unsigned int cnt,int remove,int reset,int enable)5914 ftrace_set_addr(struct ftrace_ops *ops, unsigned long *ips, unsigned int cnt,
5915 int remove, int reset, int enable)
5916 {
5917 return ftrace_set_hash(ops, NULL, 0, ips, cnt, remove, reset, enable, NULL);
5918 }
5919
5920 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
5921
5922 static int register_ftrace_function_nolock(struct ftrace_ops *ops);
5923
5924 /*
5925 * If there are multiple ftrace_ops, use SAVE_REGS by default, so that direct
5926 * call will be jumped from ftrace_regs_caller. Only if the architecture does
5927 * not support ftrace_regs_caller but direct_call, use SAVE_ARGS so that it
5928 * jumps from ftrace_caller for multiple ftrace_ops.
5929 */
5930 #ifndef CONFIG_HAVE_DYNAMIC_FTRACE_WITH_REGS
5931 #define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_ARGS)
5932 #else
5933 #define MULTI_FLAGS (FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_REGS)
5934 #endif
5935
check_direct_multi(struct ftrace_ops * ops)5936 static int check_direct_multi(struct ftrace_ops *ops)
5937 {
5938 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED))
5939 return -EINVAL;
5940 if ((ops->flags & MULTI_FLAGS) != MULTI_FLAGS)
5941 return -EINVAL;
5942 return 0;
5943 }
5944
remove_direct_functions_hash(struct ftrace_hash * hash,unsigned long addr)5945 static void remove_direct_functions_hash(struct ftrace_hash *hash, unsigned long addr)
5946 {
5947 struct ftrace_func_entry *entry, *del;
5948 int size, i;
5949
5950 size = 1 << hash->size_bits;
5951 for (i = 0; i < size; i++) {
5952 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
5953 del = __ftrace_lookup_ip(direct_functions, entry->ip);
5954 if (del && del->direct == addr) {
5955 remove_hash_entry(direct_functions, del);
5956 kfree(del);
5957 }
5958 }
5959 }
5960 }
5961
register_ftrace_direct_cb(struct rcu_head * rhp)5962 static void register_ftrace_direct_cb(struct rcu_head *rhp)
5963 {
5964 struct ftrace_hash *fhp = container_of(rhp, struct ftrace_hash, rcu);
5965
5966 free_ftrace_hash(fhp);
5967 }
5968
reset_direct(struct ftrace_ops * ops,unsigned long addr)5969 static void reset_direct(struct ftrace_ops *ops, unsigned long addr)
5970 {
5971 struct ftrace_hash *hash = ops->func_hash->filter_hash;
5972
5973 remove_direct_functions_hash(hash, addr);
5974
5975 /* cleanup for possible another register call */
5976 ops->func = NULL;
5977 ops->trampoline = 0;
5978 }
5979
5980 /**
5981 * register_ftrace_direct - Call a custom trampoline directly
5982 * for multiple functions registered in @ops
5983 * @ops: The address of the struct ftrace_ops object
5984 * @addr: The address of the trampoline to call at @ops functions
5985 *
5986 * This is used to connect a direct calls to @addr from the nop locations
5987 * of the functions registered in @ops (with by ftrace_set_filter_ip
5988 * function).
5989 *
5990 * The location that it calls (@addr) must be able to handle a direct call,
5991 * and save the parameters of the function being traced, and restore them
5992 * (or inject new ones if needed), before returning.
5993 *
5994 * Returns:
5995 * 0 on success
5996 * -EINVAL - The @ops object was already registered with this call or
5997 * when there are no functions in @ops object.
5998 * -EBUSY - Another direct function is already attached (there can be only one)
5999 * -ENODEV - @ip does not point to a ftrace nop location (or not supported)
6000 * -ENOMEM - There was an allocation failure.
6001 */
register_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)6002 int register_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
6003 {
6004 struct ftrace_hash *hash, *new_hash = NULL, *free_hash = NULL;
6005 struct ftrace_func_entry *entry, *new;
6006 int err = -EBUSY, size, i;
6007
6008 if (ops->func || ops->trampoline)
6009 return -EINVAL;
6010 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED))
6011 return -EINVAL;
6012 if (ops->flags & FTRACE_OPS_FL_ENABLED)
6013 return -EINVAL;
6014
6015 hash = ops->func_hash->filter_hash;
6016 if (ftrace_hash_empty(hash))
6017 return -EINVAL;
6018
6019 mutex_lock(&direct_mutex);
6020
6021 /* Make sure requested entries are not already registered.. */
6022 size = 1 << hash->size_bits;
6023 for (i = 0; i < size; i++) {
6024 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
6025 if (ftrace_find_rec_direct(entry->ip))
6026 goto out_unlock;
6027 }
6028 }
6029
6030 err = -ENOMEM;
6031
6032 /* Make a copy hash to place the new and the old entries in */
6033 size = hash->count + direct_functions->count;
6034 size = fls(size);
6035 if (size > FTRACE_HASH_MAX_BITS)
6036 size = FTRACE_HASH_MAX_BITS;
6037 new_hash = alloc_ftrace_hash(size);
6038 if (!new_hash)
6039 goto out_unlock;
6040
6041 /* Now copy over the existing direct entries */
6042 size = 1 << direct_functions->size_bits;
6043 for (i = 0; i < size; i++) {
6044 hlist_for_each_entry(entry, &direct_functions->buckets[i], hlist) {
6045 new = add_hash_entry(new_hash, entry->ip);
6046 if (!new)
6047 goto out_unlock;
6048 new->direct = entry->direct;
6049 }
6050 }
6051
6052 /* ... and add the new entries */
6053 size = 1 << hash->size_bits;
6054 for (i = 0; i < size; i++) {
6055 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
6056 new = add_hash_entry(new_hash, entry->ip);
6057 if (!new)
6058 goto out_unlock;
6059 /* Update both the copy and the hash entry */
6060 new->direct = addr;
6061 entry->direct = addr;
6062 }
6063 }
6064
6065 free_hash = direct_functions;
6066 rcu_assign_pointer(direct_functions, new_hash);
6067 new_hash = NULL;
6068
6069 ops->func = call_direct_funcs;
6070 ops->flags = MULTI_FLAGS;
6071 ops->trampoline = FTRACE_REGS_ADDR;
6072 ops->direct_call = addr;
6073
6074 err = register_ftrace_function_nolock(ops);
6075 if (err)
6076 reset_direct(ops, addr);
6077
6078 out_unlock:
6079 mutex_unlock(&direct_mutex);
6080
6081 if (free_hash && free_hash != EMPTY_HASH)
6082 call_rcu_tasks(&free_hash->rcu, register_ftrace_direct_cb);
6083
6084 if (new_hash)
6085 free_ftrace_hash(new_hash);
6086
6087 return err;
6088 }
6089 EXPORT_SYMBOL_GPL(register_ftrace_direct);
6090
6091 /**
6092 * unregister_ftrace_direct - Remove calls to custom trampoline
6093 * previously registered by register_ftrace_direct for @ops object.
6094 * @ops: The address of the struct ftrace_ops object
6095 * @addr: The address of the direct function that is called by the @ops functions
6096 * @free_filters: Set to true to remove all filters for the ftrace_ops, false otherwise
6097 *
6098 * This is used to remove a direct calls to @addr from the nop locations
6099 * of the functions registered in @ops (with by ftrace_set_filter_ip
6100 * function).
6101 *
6102 * Returns:
6103 * 0 on success
6104 * -EINVAL - The @ops object was not properly registered.
6105 */
unregister_ftrace_direct(struct ftrace_ops * ops,unsigned long addr,bool free_filters)6106 int unregister_ftrace_direct(struct ftrace_ops *ops, unsigned long addr,
6107 bool free_filters)
6108 {
6109 int err;
6110
6111 if (check_direct_multi(ops))
6112 return -EINVAL;
6113 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6114 return -EINVAL;
6115
6116 mutex_lock(&direct_mutex);
6117 err = unregister_ftrace_function(ops);
6118 reset_direct(ops, addr);
6119 mutex_unlock(&direct_mutex);
6120
6121 if (free_filters)
6122 ftrace_free_filter(ops);
6123 return err;
6124 }
6125 EXPORT_SYMBOL_GPL(unregister_ftrace_direct);
6126
6127 static int
__modify_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)6128 __modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
6129 {
6130 struct ftrace_hash *hash = ops->func_hash->filter_hash;
6131 struct ftrace_func_entry *entry, *iter;
6132 static struct ftrace_ops tmp_ops = {
6133 .func = ftrace_stub,
6134 .flags = FTRACE_OPS_FL_STUB,
6135 };
6136 int i, size;
6137 int err;
6138
6139 lockdep_assert_held_once(&direct_mutex);
6140
6141 /* Enable the tmp_ops to have the same functions as the direct ops */
6142 ftrace_ops_init(&tmp_ops);
6143 tmp_ops.func_hash = ops->func_hash;
6144 tmp_ops.direct_call = addr;
6145
6146 err = register_ftrace_function_nolock(&tmp_ops);
6147 if (err)
6148 return err;
6149
6150 /*
6151 * Call __ftrace_hash_update_ipmodify() here, so that we can call
6152 * ops->ops_func for the ops. This is needed because the above
6153 * register_ftrace_function_nolock() worked on tmp_ops.
6154 */
6155 err = __ftrace_hash_update_ipmodify(ops, hash, hash, true);
6156 if (err)
6157 goto out;
6158
6159 /*
6160 * Now the ftrace_ops_list_func() is called to do the direct callers.
6161 * We can safely change the direct functions attached to each entry.
6162 */
6163 mutex_lock(&ftrace_lock);
6164
6165 size = 1 << hash->size_bits;
6166 for (i = 0; i < size; i++) {
6167 hlist_for_each_entry(iter, &hash->buckets[i], hlist) {
6168 entry = __ftrace_lookup_ip(direct_functions, iter->ip);
6169 if (!entry)
6170 continue;
6171 entry->direct = addr;
6172 }
6173 }
6174 /* Prevent store tearing if a trampoline concurrently accesses the value */
6175 WRITE_ONCE(ops->direct_call, addr);
6176
6177 mutex_unlock(&ftrace_lock);
6178
6179 out:
6180 /* Removing the tmp_ops will add the updated direct callers to the functions */
6181 unregister_ftrace_function(&tmp_ops);
6182
6183 return err;
6184 }
6185
6186 /**
6187 * modify_ftrace_direct_nolock - Modify an existing direct 'multi' call
6188 * to call something else
6189 * @ops: The address of the struct ftrace_ops object
6190 * @addr: The address of the new trampoline to call at @ops functions
6191 *
6192 * This is used to unregister currently registered direct caller and
6193 * register new one @addr on functions registered in @ops object.
6194 *
6195 * Note there's window between ftrace_shutdown and ftrace_startup calls
6196 * where there will be no callbacks called.
6197 *
6198 * Caller should already have direct_mutex locked, so we don't lock
6199 * direct_mutex here.
6200 *
6201 * Returns: zero on success. Non zero on error, which includes:
6202 * -EINVAL - The @ops object was not properly registered.
6203 */
modify_ftrace_direct_nolock(struct ftrace_ops * ops,unsigned long addr)6204 int modify_ftrace_direct_nolock(struct ftrace_ops *ops, unsigned long addr)
6205 {
6206 if (check_direct_multi(ops))
6207 return -EINVAL;
6208 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6209 return -EINVAL;
6210
6211 return __modify_ftrace_direct(ops, addr);
6212 }
6213 EXPORT_SYMBOL_GPL(modify_ftrace_direct_nolock);
6214
6215 /**
6216 * modify_ftrace_direct - Modify an existing direct 'multi' call
6217 * to call something else
6218 * @ops: The address of the struct ftrace_ops object
6219 * @addr: The address of the new trampoline to call at @ops functions
6220 *
6221 * This is used to unregister currently registered direct caller and
6222 * register new one @addr on functions registered in @ops object.
6223 *
6224 * Note there's window between ftrace_shutdown and ftrace_startup calls
6225 * where there will be no callbacks called.
6226 *
6227 * Returns: zero on success. Non zero on error, which includes:
6228 * -EINVAL - The @ops object was not properly registered.
6229 */
modify_ftrace_direct(struct ftrace_ops * ops,unsigned long addr)6230 int modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr)
6231 {
6232 int err;
6233
6234 if (check_direct_multi(ops))
6235 return -EINVAL;
6236 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
6237 return -EINVAL;
6238
6239 mutex_lock(&direct_mutex);
6240 err = __modify_ftrace_direct(ops, addr);
6241 mutex_unlock(&direct_mutex);
6242 return err;
6243 }
6244 EXPORT_SYMBOL_GPL(modify_ftrace_direct);
6245 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
6246
6247 /**
6248 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
6249 * @ops: the ops to set the filter with
6250 * @ip: the address to add to or remove from the filter.
6251 * @remove: non zero to remove the ip from the filter
6252 * @reset: non zero to reset all filters before applying this filter.
6253 *
6254 * Filters denote which functions should be enabled when tracing is enabled
6255 * If @ip is NULL, it fails to update filter.
6256 *
6257 * This can allocate memory which must be freed before @ops can be freed,
6258 * either by removing each filtered addr or by using
6259 * ftrace_free_filter(@ops).
6260 */
ftrace_set_filter_ip(struct ftrace_ops * ops,unsigned long ip,int remove,int reset)6261 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
6262 int remove, int reset)
6263 {
6264 ftrace_ops_init(ops);
6265 return ftrace_set_addr(ops, &ip, 1, remove, reset, 1);
6266 }
6267 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
6268
6269 /**
6270 * ftrace_set_filter_ips - set functions to filter on in ftrace by addresses
6271 * @ops: the ops to set the filter with
6272 * @ips: the array of addresses to add to or remove from the filter.
6273 * @cnt: the number of addresses in @ips
6274 * @remove: non zero to remove ips from the filter
6275 * @reset: non zero to reset all filters before applying this filter.
6276 *
6277 * Filters denote which functions should be enabled when tracing is enabled
6278 * If @ips array or any ip specified within is NULL , it fails to update filter.
6279 *
6280 * This can allocate memory which must be freed before @ops can be freed,
6281 * either by removing each filtered addr or by using
6282 * ftrace_free_filter(@ops).
6283 */
ftrace_set_filter_ips(struct ftrace_ops * ops,unsigned long * ips,unsigned int cnt,int remove,int reset)6284 int ftrace_set_filter_ips(struct ftrace_ops *ops, unsigned long *ips,
6285 unsigned int cnt, int remove, int reset)
6286 {
6287 ftrace_ops_init(ops);
6288 return ftrace_set_addr(ops, ips, cnt, remove, reset, 1);
6289 }
6290 EXPORT_SYMBOL_GPL(ftrace_set_filter_ips);
6291
6292 /**
6293 * ftrace_ops_set_global_filter - setup ops to use global filters
6294 * @ops: the ops which will use the global filters
6295 *
6296 * ftrace users who need global function trace filtering should call this.
6297 * It can set the global filter only if ops were not initialized before.
6298 */
ftrace_ops_set_global_filter(struct ftrace_ops * ops)6299 void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
6300 {
6301 if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
6302 return;
6303
6304 ftrace_ops_init(ops);
6305 ops->func_hash = &global_ops.local_hash;
6306 }
6307 EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
6308
6309 static int
ftrace_set_regex(struct ftrace_ops * ops,unsigned char * buf,int len,int reset,int enable)6310 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
6311 int reset, int enable)
6312 {
6313 char *mod = NULL, *func, *command, *next = buf;
6314 char *tmp __free(kfree) = NULL;
6315 struct trace_array *tr = ops->private;
6316 int ret;
6317
6318 func = strsep(&next, ":");
6319
6320 /* This can also handle :mod: parsing */
6321 if (next) {
6322 if (!tr)
6323 return -EINVAL;
6324
6325 command = strsep(&next, ":");
6326 if (strcmp(command, "mod") != 0)
6327 return -EINVAL;
6328
6329 mod = next;
6330 len = command - func;
6331 /* Save the original func as ftrace_set_hash() can modify it */
6332 tmp = kstrdup(func, GFP_KERNEL);
6333 }
6334
6335 ret = ftrace_set_hash(ops, func, len, NULL, 0, 0, reset, enable, mod);
6336
6337 if (tr && mod && ret < 0) {
6338 /* Did tmp fail to allocate? */
6339 if (!tmp)
6340 return -ENOMEM;
6341 ret = cache_mod(tr, tmp, mod, enable);
6342 }
6343
6344 return ret;
6345 }
6346
6347 /**
6348 * ftrace_set_filter - set a function to filter on in ftrace
6349 * @ops: the ops to set the filter with
6350 * @buf: the string that holds the function filter text.
6351 * @len: the length of the string.
6352 * @reset: non-zero to reset all filters before applying this filter.
6353 *
6354 * Filters denote which functions should be enabled when tracing is enabled.
6355 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
6356 *
6357 * This can allocate memory which must be freed before @ops can be freed,
6358 * either by removing each filtered addr or by using
6359 * ftrace_free_filter(@ops).
6360 */
ftrace_set_filter(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)6361 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
6362 int len, int reset)
6363 {
6364 ftrace_ops_init(ops);
6365 return ftrace_set_regex(ops, buf, len, reset, 1);
6366 }
6367 EXPORT_SYMBOL_GPL(ftrace_set_filter);
6368
6369 /**
6370 * ftrace_set_notrace - set a function to not trace in ftrace
6371 * @ops: the ops to set the notrace filter with
6372 * @buf: the string that holds the function notrace text.
6373 * @len: the length of the string.
6374 * @reset: non-zero to reset all filters before applying this filter.
6375 *
6376 * Notrace Filters denote which functions should not be enabled when tracing
6377 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
6378 * for tracing.
6379 *
6380 * This can allocate memory which must be freed before @ops can be freed,
6381 * either by removing each filtered addr or by using
6382 * ftrace_free_filter(@ops).
6383 */
ftrace_set_notrace(struct ftrace_ops * ops,unsigned char * buf,int len,int reset)6384 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
6385 int len, int reset)
6386 {
6387 ftrace_ops_init(ops);
6388 return ftrace_set_regex(ops, buf, len, reset, 0);
6389 }
6390 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
6391 /**
6392 * ftrace_set_global_filter - set a function to filter on with global tracers
6393 * @buf: the string that holds the function filter text.
6394 * @len: the length of the string.
6395 * @reset: non-zero to reset all filters before applying this filter.
6396 *
6397 * Filters denote which functions should be enabled when tracing is enabled.
6398 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
6399 */
ftrace_set_global_filter(unsigned char * buf,int len,int reset)6400 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
6401 {
6402 ftrace_set_regex(&global_ops, buf, len, reset, 1);
6403 }
6404 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
6405
6406 /**
6407 * ftrace_set_global_notrace - set a function to not trace with global tracers
6408 * @buf: the string that holds the function notrace text.
6409 * @len: the length of the string.
6410 * @reset: non-zero to reset all filters before applying this filter.
6411 *
6412 * Notrace Filters denote which functions should not be enabled when tracing
6413 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
6414 * for tracing.
6415 */
ftrace_set_global_notrace(unsigned char * buf,int len,int reset)6416 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
6417 {
6418 ftrace_set_regex(&global_ops, buf, len, reset, 0);
6419 }
6420 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
6421
6422 /*
6423 * command line interface to allow users to set filters on boot up.
6424 */
6425 #define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
6426 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
6427 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
6428
6429 /* Used by function selftest to not test if filter is set */
6430 bool ftrace_filter_param __initdata;
6431
set_ftrace_notrace(char * str)6432 static int __init set_ftrace_notrace(char *str)
6433 {
6434 ftrace_filter_param = true;
6435 strscpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
6436 return 1;
6437 }
6438 __setup("ftrace_notrace=", set_ftrace_notrace);
6439
set_ftrace_filter(char * str)6440 static int __init set_ftrace_filter(char *str)
6441 {
6442 ftrace_filter_param = true;
6443 strscpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
6444 return 1;
6445 }
6446 __setup("ftrace_filter=", set_ftrace_filter);
6447
6448 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6449 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
6450 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
6451 static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
6452
set_graph_function(char * str)6453 static int __init set_graph_function(char *str)
6454 {
6455 strscpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
6456 return 1;
6457 }
6458 __setup("ftrace_graph_filter=", set_graph_function);
6459
set_graph_notrace_function(char * str)6460 static int __init set_graph_notrace_function(char *str)
6461 {
6462 strscpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
6463 return 1;
6464 }
6465 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
6466
set_graph_max_depth_function(char * str)6467 static int __init set_graph_max_depth_function(char *str)
6468 {
6469 if (!str || kstrtouint(str, 0, &fgraph_max_depth))
6470 return 0;
6471 return 1;
6472 }
6473 __setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
6474
set_ftrace_early_graph(char * buf,int enable)6475 static void __init set_ftrace_early_graph(char *buf, int enable)
6476 {
6477 int ret;
6478 char *func;
6479 struct ftrace_hash *hash;
6480
6481 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
6482 if (MEM_FAIL(!hash, "Failed to allocate hash\n"))
6483 return;
6484
6485 while (buf) {
6486 func = strsep(&buf, ",");
6487 /* we allow only one expression at a time */
6488 ret = ftrace_graph_set_hash(hash, func);
6489 if (ret)
6490 printk(KERN_DEBUG "ftrace: function %s not "
6491 "traceable\n", func);
6492 }
6493
6494 if (enable)
6495 ftrace_graph_hash = hash;
6496 else
6497 ftrace_graph_notrace_hash = hash;
6498 }
6499 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6500
6501 void __init
ftrace_set_early_filter(struct ftrace_ops * ops,char * buf,int enable)6502 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
6503 {
6504 char *func;
6505
6506 ftrace_ops_init(ops);
6507
6508 /* The trace_array is needed for caching module function filters */
6509 if (!ops->private) {
6510 struct trace_array *tr = trace_get_global_array();
6511
6512 ops->private = tr;
6513 ftrace_init_trace_array(tr);
6514 }
6515
6516 while (buf) {
6517 func = strsep(&buf, ",");
6518 ftrace_set_regex(ops, func, strlen(func), 0, enable);
6519 }
6520 }
6521
set_ftrace_early_filters(void)6522 static void __init set_ftrace_early_filters(void)
6523 {
6524 if (ftrace_filter_buf[0])
6525 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
6526 if (ftrace_notrace_buf[0])
6527 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
6528 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6529 if (ftrace_graph_buf[0])
6530 set_ftrace_early_graph(ftrace_graph_buf, 1);
6531 if (ftrace_graph_notrace_buf[0])
6532 set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
6533 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6534 }
6535
ftrace_regex_release(struct inode * inode,struct file * file)6536 int ftrace_regex_release(struct inode *inode, struct file *file)
6537 {
6538 struct seq_file *m = (struct seq_file *)file->private_data;
6539 struct ftrace_iterator *iter;
6540 struct ftrace_hash **orig_hash;
6541 struct trace_parser *parser;
6542 int filter_hash;
6543
6544 if (file->f_mode & FMODE_READ) {
6545 iter = m->private;
6546 seq_release(inode, file);
6547 } else
6548 iter = file->private_data;
6549
6550 parser = &iter->parser;
6551 if (trace_parser_loaded(parser)) {
6552 int enable = !(iter->flags & FTRACE_ITER_NOTRACE);
6553
6554 ftrace_process_regex(iter, parser->buffer,
6555 parser->idx, enable);
6556 }
6557
6558 trace_parser_put(parser);
6559
6560 mutex_lock(&iter->ops->func_hash->regex_lock);
6561
6562 if (file->f_mode & FMODE_WRITE) {
6563 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
6564
6565 if (filter_hash) {
6566 orig_hash = &iter->ops->func_hash->filter_hash;
6567 if (iter->tr) {
6568 if (list_empty(&iter->tr->mod_trace))
6569 iter->hash->flags &= ~FTRACE_HASH_FL_MOD;
6570 else
6571 iter->hash->flags |= FTRACE_HASH_FL_MOD;
6572 }
6573 } else
6574 orig_hash = &iter->ops->func_hash->notrace_hash;
6575
6576 mutex_lock(&ftrace_lock);
6577 ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
6578 iter->hash, filter_hash);
6579 mutex_unlock(&ftrace_lock);
6580 }
6581
6582 mutex_unlock(&iter->ops->func_hash->regex_lock);
6583 free_ftrace_hash(iter->hash);
6584 if (iter->tr)
6585 trace_array_put(iter->tr);
6586 kfree(iter);
6587
6588 return 0;
6589 }
6590
6591 static const struct file_operations ftrace_avail_fops = {
6592 .open = ftrace_avail_open,
6593 .read = seq_read,
6594 .llseek = seq_lseek,
6595 .release = seq_release_private,
6596 };
6597
6598 static const struct file_operations ftrace_enabled_fops = {
6599 .open = ftrace_enabled_open,
6600 .read = seq_read,
6601 .llseek = seq_lseek,
6602 .release = seq_release_private,
6603 };
6604
6605 static const struct file_operations ftrace_touched_fops = {
6606 .open = ftrace_touched_open,
6607 .read = seq_read,
6608 .llseek = seq_lseek,
6609 .release = seq_release_private,
6610 };
6611
6612 static const struct file_operations ftrace_avail_addrs_fops = {
6613 .open = ftrace_avail_addrs_open,
6614 .read = seq_read,
6615 .llseek = seq_lseek,
6616 .release = seq_release_private,
6617 };
6618
6619 static const struct file_operations ftrace_filter_fops = {
6620 .open = ftrace_filter_open,
6621 .read = seq_read,
6622 .write = ftrace_filter_write,
6623 .llseek = tracing_lseek,
6624 .release = ftrace_regex_release,
6625 };
6626
6627 static const struct file_operations ftrace_notrace_fops = {
6628 .open = ftrace_notrace_open,
6629 .read = seq_read,
6630 .write = ftrace_notrace_write,
6631 .llseek = tracing_lseek,
6632 .release = ftrace_regex_release,
6633 };
6634
6635 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6636
6637 static DEFINE_MUTEX(graph_lock);
6638
6639 struct ftrace_hash __rcu *ftrace_graph_hash = EMPTY_HASH;
6640 struct ftrace_hash __rcu *ftrace_graph_notrace_hash = EMPTY_HASH;
6641
6642 enum graph_filter_type {
6643 GRAPH_FILTER_NOTRACE = 0,
6644 GRAPH_FILTER_FUNCTION,
6645 };
6646
6647 #define FTRACE_GRAPH_EMPTY ((void *)1)
6648
6649 struct ftrace_graph_data {
6650 struct ftrace_hash *hash;
6651 struct ftrace_func_entry *entry;
6652 int idx; /* for hash table iteration */
6653 enum graph_filter_type type;
6654 struct ftrace_hash *new_hash;
6655 const struct seq_operations *seq_ops;
6656 struct trace_parser parser;
6657 };
6658
6659 static void *
__g_next(struct seq_file * m,loff_t * pos)6660 __g_next(struct seq_file *m, loff_t *pos)
6661 {
6662 struct ftrace_graph_data *fgd = m->private;
6663 struct ftrace_func_entry *entry = fgd->entry;
6664 struct hlist_head *head;
6665 int i, idx = fgd->idx;
6666
6667 if (*pos >= fgd->hash->count)
6668 return NULL;
6669
6670 if (entry) {
6671 hlist_for_each_entry_continue(entry, hlist) {
6672 fgd->entry = entry;
6673 return entry;
6674 }
6675
6676 idx++;
6677 }
6678
6679 for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
6680 head = &fgd->hash->buckets[i];
6681 hlist_for_each_entry(entry, head, hlist) {
6682 fgd->entry = entry;
6683 fgd->idx = i;
6684 return entry;
6685 }
6686 }
6687 return NULL;
6688 }
6689
6690 static void *
g_next(struct seq_file * m,void * v,loff_t * pos)6691 g_next(struct seq_file *m, void *v, loff_t *pos)
6692 {
6693 (*pos)++;
6694 return __g_next(m, pos);
6695 }
6696
g_start(struct seq_file * m,loff_t * pos)6697 static void *g_start(struct seq_file *m, loff_t *pos)
6698 {
6699 struct ftrace_graph_data *fgd = m->private;
6700
6701 mutex_lock(&graph_lock);
6702
6703 if (fgd->type == GRAPH_FILTER_FUNCTION)
6704 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
6705 lockdep_is_held(&graph_lock));
6706 else
6707 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6708 lockdep_is_held(&graph_lock));
6709
6710 /* Nothing, tell g_show to print all functions are enabled */
6711 if (ftrace_hash_empty(fgd->hash) && !*pos)
6712 return FTRACE_GRAPH_EMPTY;
6713
6714 fgd->idx = 0;
6715 fgd->entry = NULL;
6716 return __g_next(m, pos);
6717 }
6718
g_stop(struct seq_file * m,void * p)6719 static void g_stop(struct seq_file *m, void *p)
6720 {
6721 mutex_unlock(&graph_lock);
6722 }
6723
g_show(struct seq_file * m,void * v)6724 static int g_show(struct seq_file *m, void *v)
6725 {
6726 struct ftrace_func_entry *entry = v;
6727
6728 if (!entry)
6729 return 0;
6730
6731 if (entry == FTRACE_GRAPH_EMPTY) {
6732 struct ftrace_graph_data *fgd = m->private;
6733
6734 if (fgd->type == GRAPH_FILTER_FUNCTION)
6735 seq_puts(m, "#### all functions enabled ####\n");
6736 else
6737 seq_puts(m, "#### no functions disabled ####\n");
6738 return 0;
6739 }
6740
6741 seq_printf(m, "%ps\n", (void *)entry->ip);
6742
6743 return 0;
6744 }
6745
6746 static const struct seq_operations ftrace_graph_seq_ops = {
6747 .start = g_start,
6748 .next = g_next,
6749 .stop = g_stop,
6750 .show = g_show,
6751 };
6752
6753 static int
__ftrace_graph_open(struct inode * inode,struct file * file,struct ftrace_graph_data * fgd)6754 __ftrace_graph_open(struct inode *inode, struct file *file,
6755 struct ftrace_graph_data *fgd)
6756 {
6757 int ret;
6758 struct ftrace_hash *new_hash = NULL;
6759
6760 ret = security_locked_down(LOCKDOWN_TRACEFS);
6761 if (ret)
6762 return ret;
6763
6764 if (file->f_mode & FMODE_WRITE) {
6765 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
6766
6767 if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
6768 return -ENOMEM;
6769
6770 if (file->f_flags & O_TRUNC)
6771 new_hash = alloc_ftrace_hash(size_bits);
6772 else
6773 new_hash = alloc_and_copy_ftrace_hash(size_bits,
6774 fgd->hash);
6775 if (!new_hash) {
6776 ret = -ENOMEM;
6777 goto out;
6778 }
6779 }
6780
6781 if (file->f_mode & FMODE_READ) {
6782 ret = seq_open(file, &ftrace_graph_seq_ops);
6783 if (!ret) {
6784 struct seq_file *m = file->private_data;
6785 m->private = fgd;
6786 } else {
6787 /* Failed */
6788 free_ftrace_hash(new_hash);
6789 new_hash = NULL;
6790 }
6791 } else
6792 file->private_data = fgd;
6793
6794 out:
6795 if (ret < 0 && file->f_mode & FMODE_WRITE)
6796 trace_parser_put(&fgd->parser);
6797
6798 fgd->new_hash = new_hash;
6799
6800 /*
6801 * All uses of fgd->hash must be taken with the graph_lock
6802 * held. The graph_lock is going to be released, so force
6803 * fgd->hash to be reinitialized when it is taken again.
6804 */
6805 fgd->hash = NULL;
6806
6807 return ret;
6808 }
6809
6810 static int
ftrace_graph_open(struct inode * inode,struct file * file)6811 ftrace_graph_open(struct inode *inode, struct file *file)
6812 {
6813 struct ftrace_graph_data *fgd;
6814 int ret;
6815
6816 if (unlikely(ftrace_disabled))
6817 return -ENODEV;
6818
6819 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
6820 if (fgd == NULL)
6821 return -ENOMEM;
6822
6823 mutex_lock(&graph_lock);
6824
6825 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
6826 lockdep_is_held(&graph_lock));
6827 fgd->type = GRAPH_FILTER_FUNCTION;
6828 fgd->seq_ops = &ftrace_graph_seq_ops;
6829
6830 ret = __ftrace_graph_open(inode, file, fgd);
6831 if (ret < 0)
6832 kfree(fgd);
6833
6834 mutex_unlock(&graph_lock);
6835 return ret;
6836 }
6837
6838 static int
ftrace_graph_notrace_open(struct inode * inode,struct file * file)6839 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
6840 {
6841 struct ftrace_graph_data *fgd;
6842 int ret;
6843
6844 if (unlikely(ftrace_disabled))
6845 return -ENODEV;
6846
6847 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
6848 if (fgd == NULL)
6849 return -ENOMEM;
6850
6851 mutex_lock(&graph_lock);
6852
6853 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6854 lockdep_is_held(&graph_lock));
6855 fgd->type = GRAPH_FILTER_NOTRACE;
6856 fgd->seq_ops = &ftrace_graph_seq_ops;
6857
6858 ret = __ftrace_graph_open(inode, file, fgd);
6859 if (ret < 0)
6860 kfree(fgd);
6861
6862 mutex_unlock(&graph_lock);
6863 return ret;
6864 }
6865
6866 static int
ftrace_graph_release(struct inode * inode,struct file * file)6867 ftrace_graph_release(struct inode *inode, struct file *file)
6868 {
6869 struct ftrace_graph_data *fgd;
6870 struct ftrace_hash *old_hash, *new_hash;
6871 struct trace_parser *parser;
6872 int ret = 0;
6873
6874 if (file->f_mode & FMODE_READ) {
6875 struct seq_file *m = file->private_data;
6876
6877 fgd = m->private;
6878 seq_release(inode, file);
6879 } else {
6880 fgd = file->private_data;
6881 }
6882
6883
6884 if (file->f_mode & FMODE_WRITE) {
6885
6886 parser = &fgd->parser;
6887
6888 if (trace_parser_loaded((parser))) {
6889 ret = ftrace_graph_set_hash(fgd->new_hash,
6890 parser->buffer);
6891 }
6892
6893 trace_parser_put(parser);
6894
6895 new_hash = __ftrace_hash_move(fgd->new_hash);
6896 if (!new_hash) {
6897 ret = -ENOMEM;
6898 goto out;
6899 }
6900
6901 mutex_lock(&graph_lock);
6902
6903 if (fgd->type == GRAPH_FILTER_FUNCTION) {
6904 old_hash = rcu_dereference_protected(ftrace_graph_hash,
6905 lockdep_is_held(&graph_lock));
6906 rcu_assign_pointer(ftrace_graph_hash, new_hash);
6907 } else {
6908 old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
6909 lockdep_is_held(&graph_lock));
6910 rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
6911 }
6912
6913 mutex_unlock(&graph_lock);
6914
6915 /*
6916 * We need to do a hard force of sched synchronization.
6917 * This is because we use preempt_disable() to do RCU, but
6918 * the function tracers can be called where RCU is not watching
6919 * (like before user_exit()). We can not rely on the RCU
6920 * infrastructure to do the synchronization, thus we must do it
6921 * ourselves.
6922 */
6923 if (old_hash != EMPTY_HASH)
6924 synchronize_rcu_tasks_rude();
6925
6926 free_ftrace_hash(old_hash);
6927 }
6928
6929 out:
6930 free_ftrace_hash(fgd->new_hash);
6931 kfree(fgd);
6932
6933 return ret;
6934 }
6935
6936 static int
ftrace_graph_set_hash(struct ftrace_hash * hash,char * buffer)6937 ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
6938 {
6939 struct ftrace_glob func_g;
6940 struct dyn_ftrace *rec;
6941 struct ftrace_page *pg;
6942 struct ftrace_func_entry *entry;
6943 int fail = 1;
6944 int not;
6945
6946 /* decode regex */
6947 func_g.type = filter_parse_regex(buffer, strlen(buffer),
6948 &func_g.search, ¬);
6949
6950 func_g.len = strlen(func_g.search);
6951
6952 guard(mutex)(&ftrace_lock);
6953
6954 if (unlikely(ftrace_disabled))
6955 return -ENODEV;
6956
6957 do_for_each_ftrace_rec(pg, rec) {
6958
6959 if (rec->flags & FTRACE_FL_DISABLED)
6960 continue;
6961
6962 if (ftrace_match_record(rec, &func_g, NULL, 0)) {
6963 entry = ftrace_lookup_ip(hash, rec->ip);
6964
6965 if (!not) {
6966 fail = 0;
6967
6968 if (entry)
6969 continue;
6970 if (add_hash_entry(hash, rec->ip) == NULL)
6971 return 0;
6972 } else {
6973 if (entry) {
6974 free_hash_entry(hash, entry);
6975 fail = 0;
6976 }
6977 }
6978 }
6979 cond_resched();
6980 } while_for_each_ftrace_rec();
6981
6982 return fail ? -EINVAL : 0;
6983 }
6984
6985 static ssize_t
ftrace_graph_write(struct file * file,const char __user * ubuf,size_t cnt,loff_t * ppos)6986 ftrace_graph_write(struct file *file, const char __user *ubuf,
6987 size_t cnt, loff_t *ppos)
6988 {
6989 ssize_t read, ret = 0;
6990 struct ftrace_graph_data *fgd = file->private_data;
6991 struct trace_parser *parser;
6992
6993 if (!cnt)
6994 return 0;
6995
6996 /* Read mode uses seq functions */
6997 if (file->f_mode & FMODE_READ) {
6998 struct seq_file *m = file->private_data;
6999 fgd = m->private;
7000 }
7001
7002 parser = &fgd->parser;
7003
7004 read = trace_get_user(parser, ubuf, cnt, ppos);
7005
7006 if (read >= 0 && trace_parser_loaded(parser) &&
7007 !trace_parser_cont(parser)) {
7008
7009 ret = ftrace_graph_set_hash(fgd->new_hash,
7010 parser->buffer);
7011 trace_parser_clear(parser);
7012 }
7013
7014 if (!ret)
7015 ret = read;
7016
7017 return ret;
7018 }
7019
7020 static const struct file_operations ftrace_graph_fops = {
7021 .open = ftrace_graph_open,
7022 .read = seq_read,
7023 .write = ftrace_graph_write,
7024 .llseek = tracing_lseek,
7025 .release = ftrace_graph_release,
7026 };
7027
7028 static const struct file_operations ftrace_graph_notrace_fops = {
7029 .open = ftrace_graph_notrace_open,
7030 .read = seq_read,
7031 .write = ftrace_graph_write,
7032 .llseek = tracing_lseek,
7033 .release = ftrace_graph_release,
7034 };
7035 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
7036
ftrace_create_filter_files(struct ftrace_ops * ops,struct dentry * parent)7037 void ftrace_create_filter_files(struct ftrace_ops *ops,
7038 struct dentry *parent)
7039 {
7040
7041 trace_create_file("set_ftrace_filter", TRACE_MODE_WRITE, parent,
7042 ops, &ftrace_filter_fops);
7043
7044 trace_create_file("set_ftrace_notrace", TRACE_MODE_WRITE, parent,
7045 ops, &ftrace_notrace_fops);
7046 }
7047
7048 /*
7049 * The name "destroy_filter_files" is really a misnomer. Although
7050 * in the future, it may actually delete the files, but this is
7051 * really intended to make sure the ops passed in are disabled
7052 * and that when this function returns, the caller is free to
7053 * free the ops.
7054 *
7055 * The "destroy" name is only to match the "create" name that this
7056 * should be paired with.
7057 */
ftrace_destroy_filter_files(struct ftrace_ops * ops)7058 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
7059 {
7060 mutex_lock(&ftrace_lock);
7061 if (ops->flags & FTRACE_OPS_FL_ENABLED)
7062 ftrace_shutdown(ops, 0);
7063 ops->flags |= FTRACE_OPS_FL_DELETED;
7064 ftrace_free_filter(ops);
7065 mutex_unlock(&ftrace_lock);
7066 }
7067
ftrace_init_dyn_tracefs(struct dentry * d_tracer)7068 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
7069 {
7070
7071 trace_create_file("available_filter_functions", TRACE_MODE_READ,
7072 d_tracer, NULL, &ftrace_avail_fops);
7073
7074 trace_create_file("available_filter_functions_addrs", TRACE_MODE_READ,
7075 d_tracer, NULL, &ftrace_avail_addrs_fops);
7076
7077 trace_create_file("enabled_functions", TRACE_MODE_READ,
7078 d_tracer, NULL, &ftrace_enabled_fops);
7079
7080 trace_create_file("touched_functions", TRACE_MODE_READ,
7081 d_tracer, NULL, &ftrace_touched_fops);
7082
7083 ftrace_create_filter_files(&global_ops, d_tracer);
7084
7085 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
7086 trace_create_file("set_graph_function", TRACE_MODE_WRITE, d_tracer,
7087 NULL,
7088 &ftrace_graph_fops);
7089 trace_create_file("set_graph_notrace", TRACE_MODE_WRITE, d_tracer,
7090 NULL,
7091 &ftrace_graph_notrace_fops);
7092 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
7093
7094 return 0;
7095 }
7096
ftrace_cmp_ips(const void * a,const void * b)7097 static int ftrace_cmp_ips(const void *a, const void *b)
7098 {
7099 const unsigned long *ipa = a;
7100 const unsigned long *ipb = b;
7101
7102 if (*ipa > *ipb)
7103 return 1;
7104 if (*ipa < *ipb)
7105 return -1;
7106 return 0;
7107 }
7108
7109 #ifdef CONFIG_FTRACE_SORT_STARTUP_TEST
test_is_sorted(unsigned long * start,unsigned long count)7110 static void test_is_sorted(unsigned long *start, unsigned long count)
7111 {
7112 int i;
7113
7114 for (i = 1; i < count; i++) {
7115 if (WARN(start[i - 1] > start[i],
7116 "[%d] %pS at %lx is not sorted with %pS at %lx\n", i,
7117 (void *)start[i - 1], start[i - 1],
7118 (void *)start[i], start[i]))
7119 break;
7120 }
7121 if (i == count)
7122 pr_info("ftrace section at %px sorted properly\n", start);
7123 }
7124 #else
test_is_sorted(unsigned long * start,unsigned long count)7125 static void test_is_sorted(unsigned long *start, unsigned long count)
7126 {
7127 }
7128 #endif
7129
ftrace_process_locs(struct module * mod,unsigned long * start,unsigned long * end)7130 static int ftrace_process_locs(struct module *mod,
7131 unsigned long *start,
7132 unsigned long *end)
7133 {
7134 struct ftrace_page *pg_unuse = NULL;
7135 struct ftrace_page *start_pg;
7136 struct ftrace_page *pg;
7137 struct dyn_ftrace *rec;
7138 unsigned long skipped = 0;
7139 unsigned long count;
7140 unsigned long *p;
7141 unsigned long addr;
7142 unsigned long flags = 0; /* Shut up gcc */
7143 unsigned long pages;
7144 int ret = -ENOMEM;
7145
7146 count = end - start;
7147
7148 if (!count)
7149 return 0;
7150
7151 pages = DIV_ROUND_UP(count, ENTRIES_PER_PAGE);
7152
7153 /*
7154 * Sorting mcount in vmlinux at build time depend on
7155 * CONFIG_BUILDTIME_MCOUNT_SORT, while mcount loc in
7156 * modules can not be sorted at build time.
7157 */
7158 if (!IS_ENABLED(CONFIG_BUILDTIME_MCOUNT_SORT) || mod) {
7159 sort(start, count, sizeof(*start),
7160 ftrace_cmp_ips, NULL);
7161 } else {
7162 test_is_sorted(start, count);
7163 }
7164
7165 start_pg = ftrace_allocate_pages(count);
7166 if (!start_pg)
7167 return -ENOMEM;
7168
7169 mutex_lock(&ftrace_lock);
7170
7171 /*
7172 * Core and each module needs their own pages, as
7173 * modules will free them when they are removed.
7174 * Force a new page to be allocated for modules.
7175 */
7176 if (!mod) {
7177 WARN_ON(ftrace_pages || ftrace_pages_start);
7178 /* First initialization */
7179 ftrace_pages = ftrace_pages_start = start_pg;
7180 } else {
7181 if (!ftrace_pages)
7182 goto out;
7183
7184 if (WARN_ON(ftrace_pages->next)) {
7185 /* Hmm, we have free pages? */
7186 while (ftrace_pages->next)
7187 ftrace_pages = ftrace_pages->next;
7188 }
7189
7190 ftrace_pages->next = start_pg;
7191 }
7192
7193 p = start;
7194 pg = start_pg;
7195 while (p < end) {
7196 unsigned long end_offset;
7197
7198 addr = *p++;
7199
7200 /*
7201 * Some architecture linkers will pad between
7202 * the different mcount_loc sections of different
7203 * object files to satisfy alignments.
7204 * Skip any NULL pointers.
7205 */
7206 if (!addr) {
7207 skipped++;
7208 continue;
7209 }
7210
7211 /*
7212 * If this is core kernel, make sure the address is in core
7213 * or inittext, as weak functions get zeroed and KASLR can
7214 * move them to something other than zero. It just will not
7215 * move it to an area where kernel text is.
7216 */
7217 if (!mod && !(is_kernel_text(addr) || is_kernel_inittext(addr))) {
7218 skipped++;
7219 continue;
7220 }
7221
7222 addr = ftrace_call_adjust(addr);
7223
7224 end_offset = (pg->index+1) * sizeof(pg->records[0]);
7225 if (end_offset > PAGE_SIZE << pg->order) {
7226 /* We should have allocated enough */
7227 if (WARN_ON(!pg->next))
7228 break;
7229 pg = pg->next;
7230 }
7231
7232 rec = &pg->records[pg->index++];
7233 rec->ip = addr;
7234 }
7235
7236 if (pg->next) {
7237 pg_unuse = pg->next;
7238 pg->next = NULL;
7239 }
7240
7241 /* Assign the last page to ftrace_pages */
7242 ftrace_pages = pg;
7243
7244 /*
7245 * We only need to disable interrupts on start up
7246 * because we are modifying code that an interrupt
7247 * may execute, and the modification is not atomic.
7248 * But for modules, nothing runs the code we modify
7249 * until we are finished with it, and there's no
7250 * reason to cause large interrupt latencies while we do it.
7251 */
7252 if (!mod)
7253 local_irq_save(flags);
7254 ftrace_update_code(mod, start_pg);
7255 if (!mod)
7256 local_irq_restore(flags);
7257 ret = 0;
7258 out:
7259 mutex_unlock(&ftrace_lock);
7260
7261 /* We should have used all pages unless we skipped some */
7262 if (pg_unuse) {
7263 unsigned long pg_remaining, remaining = 0;
7264 unsigned long skip;
7265
7266 /* Count the number of entries unused and compare it to skipped. */
7267 pg_remaining = (ENTRIES_PER_PAGE << pg->order) - pg->index;
7268
7269 if (!WARN(skipped < pg_remaining, "Extra allocated pages for ftrace")) {
7270
7271 skip = skipped - pg_remaining;
7272
7273 for (pg = pg_unuse; pg; pg = pg->next)
7274 remaining += 1 << pg->order;
7275
7276 pages -= remaining;
7277
7278 skip = DIV_ROUND_UP(skip, ENTRIES_PER_PAGE);
7279
7280 /*
7281 * Check to see if the number of pages remaining would
7282 * just fit the number of entries skipped.
7283 */
7284 WARN(skip != remaining, "Extra allocated pages for ftrace: %lu with %lu skipped",
7285 remaining, skipped);
7286 }
7287 /* Need to synchronize with ftrace_location_range() */
7288 synchronize_rcu();
7289 ftrace_free_pages(pg_unuse);
7290 }
7291
7292 if (!mod) {
7293 count -= skipped;
7294 pr_info("ftrace: allocating %ld entries in %ld pages\n",
7295 count, pages);
7296 }
7297
7298 return ret;
7299 }
7300
7301 struct ftrace_mod_func {
7302 struct list_head list;
7303 char *name;
7304 unsigned long ip;
7305 unsigned int size;
7306 };
7307
7308 struct ftrace_mod_map {
7309 struct rcu_head rcu;
7310 struct list_head list;
7311 struct module *mod;
7312 unsigned long start_addr;
7313 unsigned long end_addr;
7314 struct list_head funcs;
7315 unsigned int num_funcs;
7316 };
7317
ftrace_get_trampoline_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7318 static int ftrace_get_trampoline_kallsym(unsigned int symnum,
7319 unsigned long *value, char *type,
7320 char *name, char *module_name,
7321 int *exported)
7322 {
7323 struct ftrace_ops *op;
7324
7325 list_for_each_entry_rcu(op, &ftrace_ops_trampoline_list, list) {
7326 if (!op->trampoline || symnum--)
7327 continue;
7328 *value = op->trampoline;
7329 *type = 't';
7330 strscpy(name, FTRACE_TRAMPOLINE_SYM, KSYM_NAME_LEN);
7331 strscpy(module_name, FTRACE_TRAMPOLINE_MOD, MODULE_NAME_LEN);
7332 *exported = 0;
7333 return 0;
7334 }
7335
7336 return -ERANGE;
7337 }
7338
7339 #if defined(CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS) || defined(CONFIG_MODULES)
7340 /*
7341 * Check if the current ops references the given ip.
7342 *
7343 * If the ops traces all functions, then it was already accounted for.
7344 * If the ops does not trace the current record function, skip it.
7345 * If the ops ignores the function via notrace filter, skip it.
7346 */
7347 static bool
ops_references_ip(struct ftrace_ops * ops,unsigned long ip)7348 ops_references_ip(struct ftrace_ops *ops, unsigned long ip)
7349 {
7350 /* If ops isn't enabled, ignore it */
7351 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
7352 return false;
7353
7354 /* If ops traces all then it includes this function */
7355 if (ops_traces_mod(ops))
7356 return true;
7357
7358 /* The function must be in the filter */
7359 if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
7360 !__ftrace_lookup_ip(ops->func_hash->filter_hash, ip))
7361 return false;
7362
7363 /* If in notrace hash, we ignore it too */
7364 if (ftrace_lookup_ip(ops->func_hash->notrace_hash, ip))
7365 return false;
7366
7367 return true;
7368 }
7369 #endif
7370
7371 #ifdef CONFIG_MODULES
7372
7373 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
7374
7375 static LIST_HEAD(ftrace_mod_maps);
7376
referenced_filters(struct dyn_ftrace * rec)7377 static int referenced_filters(struct dyn_ftrace *rec)
7378 {
7379 struct ftrace_ops *ops;
7380 int cnt = 0;
7381
7382 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
7383 if (ops_references_ip(ops, rec->ip)) {
7384 if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_DIRECT))
7385 continue;
7386 if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_IPMODIFY))
7387 continue;
7388 cnt++;
7389 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
7390 rec->flags |= FTRACE_FL_REGS;
7391 if (cnt == 1 && ops->trampoline)
7392 rec->flags |= FTRACE_FL_TRAMP;
7393 else
7394 rec->flags &= ~FTRACE_FL_TRAMP;
7395 }
7396 }
7397
7398 return cnt;
7399 }
7400
7401 static void
clear_mod_from_hash(struct ftrace_page * pg,struct ftrace_hash * hash)7402 clear_mod_from_hash(struct ftrace_page *pg, struct ftrace_hash *hash)
7403 {
7404 struct ftrace_func_entry *entry;
7405 struct dyn_ftrace *rec;
7406 int i;
7407
7408 if (ftrace_hash_empty(hash))
7409 return;
7410
7411 for (i = 0; i < pg->index; i++) {
7412 rec = &pg->records[i];
7413 entry = __ftrace_lookup_ip(hash, rec->ip);
7414 /*
7415 * Do not allow this rec to match again.
7416 * Yeah, it may waste some memory, but will be removed
7417 * if/when the hash is modified again.
7418 */
7419 if (entry)
7420 entry->ip = 0;
7421 }
7422 }
7423
7424 /* Clear any records from hashes */
clear_mod_from_hashes(struct ftrace_page * pg)7425 static void clear_mod_from_hashes(struct ftrace_page *pg)
7426 {
7427 struct trace_array *tr;
7428
7429 mutex_lock(&trace_types_lock);
7430 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7431 if (!tr->ops || !tr->ops->func_hash)
7432 continue;
7433 mutex_lock(&tr->ops->func_hash->regex_lock);
7434 clear_mod_from_hash(pg, tr->ops->func_hash->filter_hash);
7435 clear_mod_from_hash(pg, tr->ops->func_hash->notrace_hash);
7436 mutex_unlock(&tr->ops->func_hash->regex_lock);
7437 }
7438 mutex_unlock(&trace_types_lock);
7439 }
7440
ftrace_free_mod_map(struct rcu_head * rcu)7441 static void ftrace_free_mod_map(struct rcu_head *rcu)
7442 {
7443 struct ftrace_mod_map *mod_map = container_of(rcu, struct ftrace_mod_map, rcu);
7444 struct ftrace_mod_func *mod_func;
7445 struct ftrace_mod_func *n;
7446
7447 /* All the contents of mod_map are now not visible to readers */
7448 list_for_each_entry_safe(mod_func, n, &mod_map->funcs, list) {
7449 kfree(mod_func->name);
7450 list_del(&mod_func->list);
7451 kfree(mod_func);
7452 }
7453
7454 kfree(mod_map);
7455 }
7456
ftrace_release_mod(struct module * mod)7457 void ftrace_release_mod(struct module *mod)
7458 {
7459 struct ftrace_mod_map *mod_map;
7460 struct ftrace_mod_map *n;
7461 struct dyn_ftrace *rec;
7462 struct ftrace_page **last_pg;
7463 struct ftrace_page *tmp_page = NULL;
7464 struct ftrace_page *pg;
7465
7466 mutex_lock(&ftrace_lock);
7467
7468 /*
7469 * To avoid the UAF problem after the module is unloaded, the
7470 * 'mod_map' resource needs to be released unconditionally.
7471 */
7472 list_for_each_entry_safe(mod_map, n, &ftrace_mod_maps, list) {
7473 if (mod_map->mod == mod) {
7474 list_del_rcu(&mod_map->list);
7475 call_rcu(&mod_map->rcu, ftrace_free_mod_map);
7476 break;
7477 }
7478 }
7479
7480 if (ftrace_disabled)
7481 goto out_unlock;
7482
7483 /*
7484 * Each module has its own ftrace_pages, remove
7485 * them from the list.
7486 */
7487 last_pg = &ftrace_pages_start;
7488 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
7489 rec = &pg->records[0];
7490 if (within_module(rec->ip, mod)) {
7491 /*
7492 * As core pages are first, the first
7493 * page should never be a module page.
7494 */
7495 if (WARN_ON(pg == ftrace_pages_start))
7496 goto out_unlock;
7497
7498 /* Check if we are deleting the last page */
7499 if (pg == ftrace_pages)
7500 ftrace_pages = next_to_ftrace_page(last_pg);
7501
7502 ftrace_update_tot_cnt -= pg->index;
7503 *last_pg = pg->next;
7504
7505 pg->next = tmp_page;
7506 tmp_page = pg;
7507 } else
7508 last_pg = &pg->next;
7509 }
7510 out_unlock:
7511 mutex_unlock(&ftrace_lock);
7512
7513 /* Need to synchronize with ftrace_location_range() */
7514 if (tmp_page)
7515 synchronize_rcu();
7516 for (pg = tmp_page; pg; pg = tmp_page) {
7517
7518 /* Needs to be called outside of ftrace_lock */
7519 clear_mod_from_hashes(pg);
7520
7521 if (pg->records) {
7522 free_pages((unsigned long)pg->records, pg->order);
7523 ftrace_number_of_pages -= 1 << pg->order;
7524 }
7525 tmp_page = pg->next;
7526 kfree(pg);
7527 ftrace_number_of_groups--;
7528 }
7529 }
7530
ftrace_module_enable(struct module * mod)7531 void ftrace_module_enable(struct module *mod)
7532 {
7533 struct dyn_ftrace *rec;
7534 struct ftrace_page *pg;
7535
7536 mutex_lock(&ftrace_lock);
7537
7538 if (ftrace_disabled)
7539 goto out_unlock;
7540
7541 /*
7542 * If the tracing is enabled, go ahead and enable the record.
7543 *
7544 * The reason not to enable the record immediately is the
7545 * inherent check of ftrace_make_nop/ftrace_make_call for
7546 * correct previous instructions. Making first the NOP
7547 * conversion puts the module to the correct state, thus
7548 * passing the ftrace_make_call check.
7549 *
7550 * We also delay this to after the module code already set the
7551 * text to read-only, as we now need to set it back to read-write
7552 * so that we can modify the text.
7553 */
7554 if (ftrace_start_up)
7555 ftrace_arch_code_modify_prepare();
7556
7557 do_for_each_ftrace_rec(pg, rec) {
7558 int cnt;
7559 /*
7560 * do_for_each_ftrace_rec() is a double loop.
7561 * module text shares the pg. If a record is
7562 * not part of this module, then skip this pg,
7563 * which the "break" will do.
7564 */
7565 if (!within_module(rec->ip, mod))
7566 break;
7567
7568 cond_resched();
7569
7570 /* Weak functions should still be ignored */
7571 if (!test_for_valid_rec(rec)) {
7572 /* Clear all other flags. Should not be enabled anyway */
7573 rec->flags = FTRACE_FL_DISABLED;
7574 continue;
7575 }
7576
7577 cnt = 0;
7578
7579 /*
7580 * When adding a module, we need to check if tracers are
7581 * currently enabled and if they are, and can trace this record,
7582 * we need to enable the module functions as well as update the
7583 * reference counts for those function records.
7584 */
7585 if (ftrace_start_up)
7586 cnt += referenced_filters(rec);
7587
7588 rec->flags &= ~FTRACE_FL_DISABLED;
7589 rec->flags += cnt;
7590
7591 if (ftrace_start_up && cnt) {
7592 int failed = __ftrace_replace_code(rec, 1);
7593 if (failed) {
7594 ftrace_bug(failed, rec);
7595 goto out_loop;
7596 }
7597 }
7598
7599 } while_for_each_ftrace_rec();
7600
7601 out_loop:
7602 if (ftrace_start_up)
7603 ftrace_arch_code_modify_post_process();
7604
7605 out_unlock:
7606 mutex_unlock(&ftrace_lock);
7607
7608 process_cached_mods(mod->name);
7609 }
7610
ftrace_module_init(struct module * mod)7611 void ftrace_module_init(struct module *mod)
7612 {
7613 int ret;
7614
7615 if (ftrace_disabled || !mod->num_ftrace_callsites)
7616 return;
7617
7618 ret = ftrace_process_locs(mod, mod->ftrace_callsites,
7619 mod->ftrace_callsites + mod->num_ftrace_callsites);
7620 if (ret)
7621 pr_warn("ftrace: failed to allocate entries for module '%s' functions\n",
7622 mod->name);
7623 }
7624
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)7625 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
7626 struct dyn_ftrace *rec)
7627 {
7628 struct ftrace_mod_func *mod_func;
7629 unsigned long symsize;
7630 unsigned long offset;
7631 char str[KSYM_SYMBOL_LEN];
7632 char *modname;
7633 const char *ret;
7634
7635 ret = kallsyms_lookup(rec->ip, &symsize, &offset, &modname, str);
7636 if (!ret)
7637 return;
7638
7639 mod_func = kmalloc(sizeof(*mod_func), GFP_KERNEL);
7640 if (!mod_func)
7641 return;
7642
7643 mod_func->name = kstrdup(str, GFP_KERNEL);
7644 if (!mod_func->name) {
7645 kfree(mod_func);
7646 return;
7647 }
7648
7649 mod_func->ip = rec->ip - offset;
7650 mod_func->size = symsize;
7651
7652 mod_map->num_funcs++;
7653
7654 list_add_rcu(&mod_func->list, &mod_map->funcs);
7655 }
7656
7657 static struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)7658 allocate_ftrace_mod_map(struct module *mod,
7659 unsigned long start, unsigned long end)
7660 {
7661 struct ftrace_mod_map *mod_map;
7662
7663 if (ftrace_disabled)
7664 return NULL;
7665
7666 mod_map = kmalloc(sizeof(*mod_map), GFP_KERNEL);
7667 if (!mod_map)
7668 return NULL;
7669
7670 mod_map->mod = mod;
7671 mod_map->start_addr = start;
7672 mod_map->end_addr = end;
7673 mod_map->num_funcs = 0;
7674
7675 INIT_LIST_HEAD_RCU(&mod_map->funcs);
7676
7677 list_add_rcu(&mod_map->list, &ftrace_mod_maps);
7678
7679 return mod_map;
7680 }
7681
7682 static int
ftrace_func_address_lookup(struct ftrace_mod_map * mod_map,unsigned long addr,unsigned long * size,unsigned long * off,char * sym)7683 ftrace_func_address_lookup(struct ftrace_mod_map *mod_map,
7684 unsigned long addr, unsigned long *size,
7685 unsigned long *off, char *sym)
7686 {
7687 struct ftrace_mod_func *found_func = NULL;
7688 struct ftrace_mod_func *mod_func;
7689
7690 list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
7691 if (addr >= mod_func->ip &&
7692 addr < mod_func->ip + mod_func->size) {
7693 found_func = mod_func;
7694 break;
7695 }
7696 }
7697
7698 if (found_func) {
7699 if (size)
7700 *size = found_func->size;
7701 if (off)
7702 *off = addr - found_func->ip;
7703 return strscpy(sym, found_func->name, KSYM_NAME_LEN);
7704 }
7705
7706 return 0;
7707 }
7708
7709 int
ftrace_mod_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)7710 ftrace_mod_address_lookup(unsigned long addr, unsigned long *size,
7711 unsigned long *off, char **modname, char *sym)
7712 {
7713 struct ftrace_mod_map *mod_map;
7714 int ret = 0;
7715
7716 /* mod_map is freed via call_rcu() */
7717 preempt_disable();
7718 list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
7719 ret = ftrace_func_address_lookup(mod_map, addr, size, off, sym);
7720 if (ret) {
7721 if (modname)
7722 *modname = mod_map->mod->name;
7723 break;
7724 }
7725 }
7726 preempt_enable();
7727
7728 return ret;
7729 }
7730
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7731 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
7732 char *type, char *name,
7733 char *module_name, int *exported)
7734 {
7735 struct ftrace_mod_map *mod_map;
7736 struct ftrace_mod_func *mod_func;
7737 int ret;
7738
7739 preempt_disable();
7740 list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
7741
7742 if (symnum >= mod_map->num_funcs) {
7743 symnum -= mod_map->num_funcs;
7744 continue;
7745 }
7746
7747 list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
7748 if (symnum > 1) {
7749 symnum--;
7750 continue;
7751 }
7752
7753 *value = mod_func->ip;
7754 *type = 'T';
7755 strscpy(name, mod_func->name, KSYM_NAME_LEN);
7756 strscpy(module_name, mod_map->mod->name, MODULE_NAME_LEN);
7757 *exported = 1;
7758 preempt_enable();
7759 return 0;
7760 }
7761 WARN_ON(1);
7762 break;
7763 }
7764 ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
7765 module_name, exported);
7766 preempt_enable();
7767 return ret;
7768 }
7769
7770 #else
save_ftrace_mod_rec(struct ftrace_mod_map * mod_map,struct dyn_ftrace * rec)7771 static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
7772 struct dyn_ftrace *rec) { }
7773 static inline struct ftrace_mod_map *
allocate_ftrace_mod_map(struct module * mod,unsigned long start,unsigned long end)7774 allocate_ftrace_mod_map(struct module *mod,
7775 unsigned long start, unsigned long end)
7776 {
7777 return NULL;
7778 }
ftrace_mod_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * name,char * module_name,int * exported)7779 int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
7780 char *type, char *name, char *module_name,
7781 int *exported)
7782 {
7783 int ret;
7784
7785 preempt_disable();
7786 ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
7787 module_name, exported);
7788 preempt_enable();
7789 return ret;
7790 }
7791 #endif /* CONFIG_MODULES */
7792
7793 struct ftrace_init_func {
7794 struct list_head list;
7795 unsigned long ip;
7796 };
7797
7798 /* Clear any init ips from hashes */
7799 static void
clear_func_from_hash(struct ftrace_init_func * func,struct ftrace_hash * hash)7800 clear_func_from_hash(struct ftrace_init_func *func, struct ftrace_hash *hash)
7801 {
7802 struct ftrace_func_entry *entry;
7803
7804 entry = ftrace_lookup_ip(hash, func->ip);
7805 /*
7806 * Do not allow this rec to match again.
7807 * Yeah, it may waste some memory, but will be removed
7808 * if/when the hash is modified again.
7809 */
7810 if (entry)
7811 entry->ip = 0;
7812 }
7813
7814 static void
clear_func_from_hashes(struct ftrace_init_func * func)7815 clear_func_from_hashes(struct ftrace_init_func *func)
7816 {
7817 struct trace_array *tr;
7818
7819 mutex_lock(&trace_types_lock);
7820 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7821 if (!tr->ops || !tr->ops->func_hash)
7822 continue;
7823 mutex_lock(&tr->ops->func_hash->regex_lock);
7824 clear_func_from_hash(func, tr->ops->func_hash->filter_hash);
7825 clear_func_from_hash(func, tr->ops->func_hash->notrace_hash);
7826 mutex_unlock(&tr->ops->func_hash->regex_lock);
7827 }
7828 mutex_unlock(&trace_types_lock);
7829 }
7830
add_to_clear_hash_list(struct list_head * clear_list,struct dyn_ftrace * rec)7831 static void add_to_clear_hash_list(struct list_head *clear_list,
7832 struct dyn_ftrace *rec)
7833 {
7834 struct ftrace_init_func *func;
7835
7836 func = kmalloc(sizeof(*func), GFP_KERNEL);
7837 if (!func) {
7838 MEM_FAIL(1, "alloc failure, ftrace filter could be stale\n");
7839 return;
7840 }
7841
7842 func->ip = rec->ip;
7843 list_add(&func->list, clear_list);
7844 }
7845
ftrace_free_mem(struct module * mod,void * start_ptr,void * end_ptr)7846 void ftrace_free_mem(struct module *mod, void *start_ptr, void *end_ptr)
7847 {
7848 unsigned long start = (unsigned long)(start_ptr);
7849 unsigned long end = (unsigned long)(end_ptr);
7850 struct ftrace_page **last_pg = &ftrace_pages_start;
7851 struct ftrace_page *tmp_page = NULL;
7852 struct ftrace_page *pg;
7853 struct dyn_ftrace *rec;
7854 struct dyn_ftrace key;
7855 struct ftrace_mod_map *mod_map = NULL;
7856 struct ftrace_init_func *func, *func_next;
7857 LIST_HEAD(clear_hash);
7858
7859 key.ip = start;
7860 key.flags = end; /* overload flags, as it is unsigned long */
7861
7862 mutex_lock(&ftrace_lock);
7863
7864 /*
7865 * If we are freeing module init memory, then check if
7866 * any tracer is active. If so, we need to save a mapping of
7867 * the module functions being freed with the address.
7868 */
7869 if (mod && ftrace_ops_list != &ftrace_list_end)
7870 mod_map = allocate_ftrace_mod_map(mod, start, end);
7871
7872 for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
7873 if (end < pg->records[0].ip ||
7874 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
7875 continue;
7876 again:
7877 rec = bsearch(&key, pg->records, pg->index,
7878 sizeof(struct dyn_ftrace),
7879 ftrace_cmp_recs);
7880 if (!rec)
7881 continue;
7882
7883 /* rec will be cleared from hashes after ftrace_lock unlock */
7884 add_to_clear_hash_list(&clear_hash, rec);
7885
7886 if (mod_map)
7887 save_ftrace_mod_rec(mod_map, rec);
7888
7889 pg->index--;
7890 ftrace_update_tot_cnt--;
7891 if (!pg->index) {
7892 *last_pg = pg->next;
7893 pg->next = tmp_page;
7894 tmp_page = pg;
7895 pg = container_of(last_pg, struct ftrace_page, next);
7896 if (!(*last_pg))
7897 ftrace_pages = pg;
7898 continue;
7899 }
7900 memmove(rec, rec + 1,
7901 (pg->index - (rec - pg->records)) * sizeof(*rec));
7902 /* More than one function may be in this block */
7903 goto again;
7904 }
7905 mutex_unlock(&ftrace_lock);
7906
7907 list_for_each_entry_safe(func, func_next, &clear_hash, list) {
7908 clear_func_from_hashes(func);
7909 kfree(func);
7910 }
7911 /* Need to synchronize with ftrace_location_range() */
7912 if (tmp_page) {
7913 synchronize_rcu();
7914 ftrace_free_pages(tmp_page);
7915 }
7916 }
7917
ftrace_free_init_mem(void)7918 void __init ftrace_free_init_mem(void)
7919 {
7920 void *start = (void *)(&__init_begin);
7921 void *end = (void *)(&__init_end);
7922
7923 ftrace_boot_snapshot();
7924
7925 ftrace_free_mem(NULL, start, end);
7926 }
7927
ftrace_dyn_arch_init(void)7928 int __init __weak ftrace_dyn_arch_init(void)
7929 {
7930 return 0;
7931 }
7932
ftrace_init(void)7933 void __init ftrace_init(void)
7934 {
7935 extern unsigned long __start_mcount_loc[];
7936 extern unsigned long __stop_mcount_loc[];
7937 unsigned long count, flags;
7938 int ret;
7939
7940 local_irq_save(flags);
7941 ret = ftrace_dyn_arch_init();
7942 local_irq_restore(flags);
7943 if (ret)
7944 goto failed;
7945
7946 count = __stop_mcount_loc - __start_mcount_loc;
7947 if (!count) {
7948 pr_info("ftrace: No functions to be traced?\n");
7949 goto failed;
7950 }
7951
7952 ret = ftrace_process_locs(NULL,
7953 __start_mcount_loc,
7954 __stop_mcount_loc);
7955 if (ret) {
7956 pr_warn("ftrace: failed to allocate entries for functions\n");
7957 goto failed;
7958 }
7959
7960 pr_info("ftrace: allocated %ld pages with %ld groups\n",
7961 ftrace_number_of_pages, ftrace_number_of_groups);
7962
7963 last_ftrace_enabled = ftrace_enabled = 1;
7964
7965 set_ftrace_early_filters();
7966
7967 return;
7968 failed:
7969 ftrace_disabled = 1;
7970 }
7971
7972 /* Do nothing if arch does not support this */
arch_ftrace_update_trampoline(struct ftrace_ops * ops)7973 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
7974 {
7975 }
7976
ftrace_update_trampoline(struct ftrace_ops * ops)7977 static void ftrace_update_trampoline(struct ftrace_ops *ops)
7978 {
7979 unsigned long trampoline = ops->trampoline;
7980
7981 arch_ftrace_update_trampoline(ops);
7982 if (ops->trampoline && ops->trampoline != trampoline &&
7983 (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP)) {
7984 /* Add to kallsyms before the perf events */
7985 ftrace_add_trampoline_to_kallsyms(ops);
7986 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
7987 ops->trampoline, ops->trampoline_size, false,
7988 FTRACE_TRAMPOLINE_SYM);
7989 /*
7990 * Record the perf text poke event after the ksymbol register
7991 * event.
7992 */
7993 perf_event_text_poke((void *)ops->trampoline, NULL, 0,
7994 (void *)ops->trampoline,
7995 ops->trampoline_size);
7996 }
7997 }
7998
ftrace_init_trace_array(struct trace_array * tr)7999 void ftrace_init_trace_array(struct trace_array *tr)
8000 {
8001 if (tr->flags & TRACE_ARRAY_FL_MOD_INIT)
8002 return;
8003
8004 INIT_LIST_HEAD(&tr->func_probes);
8005 INIT_LIST_HEAD(&tr->mod_trace);
8006 INIT_LIST_HEAD(&tr->mod_notrace);
8007
8008 tr->flags |= TRACE_ARRAY_FL_MOD_INIT;
8009 }
8010 #else
8011
8012 struct ftrace_ops global_ops = {
8013 .func = ftrace_stub,
8014 .flags = FTRACE_OPS_FL_INITIALIZED |
8015 FTRACE_OPS_FL_PID,
8016 };
8017
ftrace_nodyn_init(void)8018 static int __init ftrace_nodyn_init(void)
8019 {
8020 ftrace_enabled = 1;
8021 return 0;
8022 }
8023 core_initcall(ftrace_nodyn_init);
8024
ftrace_init_dyn_tracefs(struct dentry * d_tracer)8025 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
ftrace_startup_all(int command)8026 static inline void ftrace_startup_all(int command) { }
8027
ftrace_update_trampoline(struct ftrace_ops * ops)8028 static void ftrace_update_trampoline(struct ftrace_ops *ops)
8029 {
8030 }
8031
8032 #endif /* CONFIG_DYNAMIC_FTRACE */
8033
ftrace_init_global_array_ops(struct trace_array * tr)8034 __init void ftrace_init_global_array_ops(struct trace_array *tr)
8035 {
8036 tr->ops = &global_ops;
8037 if (!global_ops.private)
8038 global_ops.private = tr;
8039 ftrace_init_trace_array(tr);
8040 init_array_fgraph_ops(tr, tr->ops);
8041 }
8042
ftrace_init_array_ops(struct trace_array * tr,ftrace_func_t func)8043 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
8044 {
8045 /* If we filter on pids, update to use the pid function */
8046 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
8047 if (WARN_ON(tr->ops->func != ftrace_stub))
8048 printk("ftrace ops had %pS for function\n",
8049 tr->ops->func);
8050 }
8051 tr->ops->func = func;
8052 tr->ops->private = tr;
8053 }
8054
ftrace_reset_array_ops(struct trace_array * tr)8055 void ftrace_reset_array_ops(struct trace_array *tr)
8056 {
8057 tr->ops->func = ftrace_stub;
8058 }
8059
8060 static nokprobe_inline void
__ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * ignored,struct ftrace_regs * fregs)8061 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
8062 struct ftrace_ops *ignored, struct ftrace_regs *fregs)
8063 {
8064 struct pt_regs *regs = ftrace_get_regs(fregs);
8065 struct ftrace_ops *op;
8066 int bit;
8067
8068 /*
8069 * The ftrace_test_and_set_recursion() will disable preemption,
8070 * which is required since some of the ops may be dynamically
8071 * allocated, they must be freed after a synchronize_rcu().
8072 */
8073 bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
8074 if (bit < 0)
8075 return;
8076
8077 do_for_each_ftrace_op(op, ftrace_ops_list) {
8078 /* Stub functions don't need to be called nor tested */
8079 if (op->flags & FTRACE_OPS_FL_STUB)
8080 continue;
8081 /*
8082 * Check the following for each ops before calling their func:
8083 * if RCU flag is set, then rcu_is_watching() must be true
8084 * Otherwise test if the ip matches the ops filter
8085 *
8086 * If any of the above fails then the op->func() is not executed.
8087 */
8088 if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
8089 ftrace_ops_test(op, ip, regs)) {
8090 if (FTRACE_WARN_ON(!op->func)) {
8091 pr_warn("op=%p %pS\n", op, op);
8092 goto out;
8093 }
8094 op->func(ip, parent_ip, op, fregs);
8095 }
8096 } while_for_each_ftrace_op(op);
8097 out:
8098 trace_clear_recursion(bit);
8099 }
8100
8101 /*
8102 * Some archs only support passing ip and parent_ip. Even though
8103 * the list function ignores the op parameter, we do not want any
8104 * C side effects, where a function is called without the caller
8105 * sending a third parameter.
8106 * Archs are to support both the regs and ftrace_ops at the same time.
8107 * If they support ftrace_ops, it is assumed they support regs.
8108 * If call backs want to use regs, they must either check for regs
8109 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
8110 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
8111 * An architecture can pass partial regs with ftrace_ops and still
8112 * set the ARCH_SUPPORTS_FTRACE_OPS.
8113 *
8114 * In vmlinux.lds.h, ftrace_ops_list_func() is defined to be
8115 * arch_ftrace_ops_list_func.
8116 */
8117 #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)8118 void arch_ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
8119 struct ftrace_ops *op, struct ftrace_regs *fregs)
8120 {
8121 kmsan_unpoison_memory(fregs, ftrace_regs_size());
8122 __ftrace_ops_list_func(ip, parent_ip, NULL, fregs);
8123 }
8124 #else
arch_ftrace_ops_list_func(unsigned long ip,unsigned long parent_ip)8125 void arch_ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
8126 {
8127 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
8128 }
8129 #endif
8130 NOKPROBE_SYMBOL(arch_ftrace_ops_list_func);
8131
8132 /*
8133 * If there's only one function registered but it does not support
8134 * recursion, needs RCU protection, then this function will be called
8135 * by the mcount trampoline.
8136 */
ftrace_ops_assist_func(unsigned long ip,unsigned long parent_ip,struct ftrace_ops * op,struct ftrace_regs * fregs)8137 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
8138 struct ftrace_ops *op, struct ftrace_regs *fregs)
8139 {
8140 int bit;
8141
8142 bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
8143 if (bit < 0)
8144 return;
8145
8146 if (!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching())
8147 op->func(ip, parent_ip, op, fregs);
8148
8149 trace_clear_recursion(bit);
8150 }
8151 NOKPROBE_SYMBOL(ftrace_ops_assist_func);
8152
8153 /**
8154 * ftrace_ops_get_func - get the function a trampoline should call
8155 * @ops: the ops to get the function for
8156 *
8157 * Normally the mcount trampoline will call the ops->func, but there
8158 * are times that it should not. For example, if the ops does not
8159 * have its own recursion protection, then it should call the
8160 * ftrace_ops_assist_func() instead.
8161 *
8162 * Returns: the function that the trampoline should call for @ops.
8163 */
ftrace_ops_get_func(struct ftrace_ops * ops)8164 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
8165 {
8166 /*
8167 * If the function does not handle recursion or needs to be RCU safe,
8168 * then we need to call the assist handler.
8169 */
8170 if (ops->flags & (FTRACE_OPS_FL_RECURSION |
8171 FTRACE_OPS_FL_RCU))
8172 return ftrace_ops_assist_func;
8173
8174 return ops->func;
8175 }
8176
8177 static void
ftrace_filter_pid_sched_switch_probe(void * data,bool preempt,struct task_struct * prev,struct task_struct * next,unsigned int prev_state)8178 ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
8179 struct task_struct *prev,
8180 struct task_struct *next,
8181 unsigned int prev_state)
8182 {
8183 struct trace_array *tr = data;
8184 struct trace_pid_list *pid_list;
8185 struct trace_pid_list *no_pid_list;
8186
8187 pid_list = rcu_dereference_sched(tr->function_pids);
8188 no_pid_list = rcu_dereference_sched(tr->function_no_pids);
8189
8190 if (trace_ignore_this_task(pid_list, no_pid_list, next))
8191 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8192 FTRACE_PID_IGNORE);
8193 else
8194 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8195 next->pid);
8196 }
8197
8198 static void
ftrace_pid_follow_sched_process_fork(void * data,struct task_struct * self,struct task_struct * task)8199 ftrace_pid_follow_sched_process_fork(void *data,
8200 struct task_struct *self,
8201 struct task_struct *task)
8202 {
8203 struct trace_pid_list *pid_list;
8204 struct trace_array *tr = data;
8205
8206 pid_list = rcu_dereference_sched(tr->function_pids);
8207 trace_filter_add_remove_task(pid_list, self, task);
8208
8209 pid_list = rcu_dereference_sched(tr->function_no_pids);
8210 trace_filter_add_remove_task(pid_list, self, task);
8211 }
8212
8213 static void
ftrace_pid_follow_sched_process_exit(void * data,struct task_struct * task)8214 ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
8215 {
8216 struct trace_pid_list *pid_list;
8217 struct trace_array *tr = data;
8218
8219 pid_list = rcu_dereference_sched(tr->function_pids);
8220 trace_filter_add_remove_task(pid_list, NULL, task);
8221
8222 pid_list = rcu_dereference_sched(tr->function_no_pids);
8223 trace_filter_add_remove_task(pid_list, NULL, task);
8224 }
8225
ftrace_pid_follow_fork(struct trace_array * tr,bool enable)8226 void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
8227 {
8228 if (enable) {
8229 register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
8230 tr);
8231 register_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
8232 tr);
8233 } else {
8234 unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
8235 tr);
8236 unregister_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
8237 tr);
8238 }
8239 }
8240
clear_ftrace_pids(struct trace_array * tr,int type)8241 static void clear_ftrace_pids(struct trace_array *tr, int type)
8242 {
8243 struct trace_pid_list *pid_list;
8244 struct trace_pid_list *no_pid_list;
8245 int cpu;
8246
8247 pid_list = rcu_dereference_protected(tr->function_pids,
8248 lockdep_is_held(&ftrace_lock));
8249 no_pid_list = rcu_dereference_protected(tr->function_no_pids,
8250 lockdep_is_held(&ftrace_lock));
8251
8252 /* Make sure there's something to do */
8253 if (!pid_type_enabled(type, pid_list, no_pid_list))
8254 return;
8255
8256 /* See if the pids still need to be checked after this */
8257 if (!still_need_pid_events(type, pid_list, no_pid_list)) {
8258 unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
8259 for_each_possible_cpu(cpu)
8260 per_cpu_ptr(tr->array_buffer.data, cpu)->ftrace_ignore_pid = FTRACE_PID_TRACE;
8261 }
8262
8263 if (type & TRACE_PIDS)
8264 rcu_assign_pointer(tr->function_pids, NULL);
8265
8266 if (type & TRACE_NO_PIDS)
8267 rcu_assign_pointer(tr->function_no_pids, NULL);
8268
8269 /* Wait till all users are no longer using pid filtering */
8270 synchronize_rcu();
8271
8272 if ((type & TRACE_PIDS) && pid_list)
8273 trace_pid_list_free(pid_list);
8274
8275 if ((type & TRACE_NO_PIDS) && no_pid_list)
8276 trace_pid_list_free(no_pid_list);
8277 }
8278
ftrace_clear_pids(struct trace_array * tr)8279 void ftrace_clear_pids(struct trace_array *tr)
8280 {
8281 mutex_lock(&ftrace_lock);
8282
8283 clear_ftrace_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
8284
8285 mutex_unlock(&ftrace_lock);
8286 }
8287
ftrace_pid_reset(struct trace_array * tr,int type)8288 static void ftrace_pid_reset(struct trace_array *tr, int type)
8289 {
8290 mutex_lock(&ftrace_lock);
8291 clear_ftrace_pids(tr, type);
8292
8293 ftrace_update_pid_func();
8294 ftrace_startup_all(0);
8295
8296 mutex_unlock(&ftrace_lock);
8297 }
8298
8299 /* Greater than any max PID */
8300 #define FTRACE_NO_PIDS (void *)(PID_MAX_LIMIT + 1)
8301
fpid_start(struct seq_file * m,loff_t * pos)8302 static void *fpid_start(struct seq_file *m, loff_t *pos)
8303 __acquires(RCU)
8304 {
8305 struct trace_pid_list *pid_list;
8306 struct trace_array *tr = m->private;
8307
8308 mutex_lock(&ftrace_lock);
8309 rcu_read_lock_sched();
8310
8311 pid_list = rcu_dereference_sched(tr->function_pids);
8312
8313 if (!pid_list)
8314 return !(*pos) ? FTRACE_NO_PIDS : NULL;
8315
8316 return trace_pid_start(pid_list, pos);
8317 }
8318
fpid_next(struct seq_file * m,void * v,loff_t * pos)8319 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
8320 {
8321 struct trace_array *tr = m->private;
8322 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
8323
8324 if (v == FTRACE_NO_PIDS) {
8325 (*pos)++;
8326 return NULL;
8327 }
8328 return trace_pid_next(pid_list, v, pos);
8329 }
8330
fpid_stop(struct seq_file * m,void * p)8331 static void fpid_stop(struct seq_file *m, void *p)
8332 __releases(RCU)
8333 {
8334 rcu_read_unlock_sched();
8335 mutex_unlock(&ftrace_lock);
8336 }
8337
fpid_show(struct seq_file * m,void * v)8338 static int fpid_show(struct seq_file *m, void *v)
8339 {
8340 if (v == FTRACE_NO_PIDS) {
8341 seq_puts(m, "no pid\n");
8342 return 0;
8343 }
8344
8345 return trace_pid_show(m, v);
8346 }
8347
8348 static const struct seq_operations ftrace_pid_sops = {
8349 .start = fpid_start,
8350 .next = fpid_next,
8351 .stop = fpid_stop,
8352 .show = fpid_show,
8353 };
8354
fnpid_start(struct seq_file * m,loff_t * pos)8355 static void *fnpid_start(struct seq_file *m, loff_t *pos)
8356 __acquires(RCU)
8357 {
8358 struct trace_pid_list *pid_list;
8359 struct trace_array *tr = m->private;
8360
8361 mutex_lock(&ftrace_lock);
8362 rcu_read_lock_sched();
8363
8364 pid_list = rcu_dereference_sched(tr->function_no_pids);
8365
8366 if (!pid_list)
8367 return !(*pos) ? FTRACE_NO_PIDS : NULL;
8368
8369 return trace_pid_start(pid_list, pos);
8370 }
8371
fnpid_next(struct seq_file * m,void * v,loff_t * pos)8372 static void *fnpid_next(struct seq_file *m, void *v, loff_t *pos)
8373 {
8374 struct trace_array *tr = m->private;
8375 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_no_pids);
8376
8377 if (v == FTRACE_NO_PIDS) {
8378 (*pos)++;
8379 return NULL;
8380 }
8381 return trace_pid_next(pid_list, v, pos);
8382 }
8383
8384 static const struct seq_operations ftrace_no_pid_sops = {
8385 .start = fnpid_start,
8386 .next = fnpid_next,
8387 .stop = fpid_stop,
8388 .show = fpid_show,
8389 };
8390
pid_open(struct inode * inode,struct file * file,int type)8391 static int pid_open(struct inode *inode, struct file *file, int type)
8392 {
8393 const struct seq_operations *seq_ops;
8394 struct trace_array *tr = inode->i_private;
8395 struct seq_file *m;
8396 int ret = 0;
8397
8398 ret = tracing_check_open_get_tr(tr);
8399 if (ret)
8400 return ret;
8401
8402 if ((file->f_mode & FMODE_WRITE) &&
8403 (file->f_flags & O_TRUNC))
8404 ftrace_pid_reset(tr, type);
8405
8406 switch (type) {
8407 case TRACE_PIDS:
8408 seq_ops = &ftrace_pid_sops;
8409 break;
8410 case TRACE_NO_PIDS:
8411 seq_ops = &ftrace_no_pid_sops;
8412 break;
8413 default:
8414 trace_array_put(tr);
8415 WARN_ON_ONCE(1);
8416 return -EINVAL;
8417 }
8418
8419 ret = seq_open(file, seq_ops);
8420 if (ret < 0) {
8421 trace_array_put(tr);
8422 } else {
8423 m = file->private_data;
8424 /* copy tr over to seq ops */
8425 m->private = tr;
8426 }
8427
8428 return ret;
8429 }
8430
8431 static int
ftrace_pid_open(struct inode * inode,struct file * file)8432 ftrace_pid_open(struct inode *inode, struct file *file)
8433 {
8434 return pid_open(inode, file, TRACE_PIDS);
8435 }
8436
8437 static int
ftrace_no_pid_open(struct inode * inode,struct file * file)8438 ftrace_no_pid_open(struct inode *inode, struct file *file)
8439 {
8440 return pid_open(inode, file, TRACE_NO_PIDS);
8441 }
8442
ignore_task_cpu(void * data)8443 static void ignore_task_cpu(void *data)
8444 {
8445 struct trace_array *tr = data;
8446 struct trace_pid_list *pid_list;
8447 struct trace_pid_list *no_pid_list;
8448
8449 /*
8450 * This function is called by on_each_cpu() while the
8451 * event_mutex is held.
8452 */
8453 pid_list = rcu_dereference_protected(tr->function_pids,
8454 mutex_is_locked(&ftrace_lock));
8455 no_pid_list = rcu_dereference_protected(tr->function_no_pids,
8456 mutex_is_locked(&ftrace_lock));
8457
8458 if (trace_ignore_this_task(pid_list, no_pid_list, current))
8459 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8460 FTRACE_PID_IGNORE);
8461 else
8462 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
8463 current->pid);
8464 }
8465
8466 static ssize_t
pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos,int type)8467 pid_write(struct file *filp, const char __user *ubuf,
8468 size_t cnt, loff_t *ppos, int type)
8469 {
8470 struct seq_file *m = filp->private_data;
8471 struct trace_array *tr = m->private;
8472 struct trace_pid_list *filtered_pids;
8473 struct trace_pid_list *other_pids;
8474 struct trace_pid_list *pid_list;
8475 ssize_t ret;
8476
8477 if (!cnt)
8478 return 0;
8479
8480 guard(mutex)(&ftrace_lock);
8481
8482 switch (type) {
8483 case TRACE_PIDS:
8484 filtered_pids = rcu_dereference_protected(tr->function_pids,
8485 lockdep_is_held(&ftrace_lock));
8486 other_pids = rcu_dereference_protected(tr->function_no_pids,
8487 lockdep_is_held(&ftrace_lock));
8488 break;
8489 case TRACE_NO_PIDS:
8490 filtered_pids = rcu_dereference_protected(tr->function_no_pids,
8491 lockdep_is_held(&ftrace_lock));
8492 other_pids = rcu_dereference_protected(tr->function_pids,
8493 lockdep_is_held(&ftrace_lock));
8494 break;
8495 default:
8496 WARN_ON_ONCE(1);
8497 return -EINVAL;
8498 }
8499
8500 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
8501 if (ret < 0)
8502 return ret;
8503
8504 switch (type) {
8505 case TRACE_PIDS:
8506 rcu_assign_pointer(tr->function_pids, pid_list);
8507 break;
8508 case TRACE_NO_PIDS:
8509 rcu_assign_pointer(tr->function_no_pids, pid_list);
8510 break;
8511 }
8512
8513
8514 if (filtered_pids) {
8515 synchronize_rcu();
8516 trace_pid_list_free(filtered_pids);
8517 } else if (pid_list && !other_pids) {
8518 /* Register a probe to set whether to ignore the tracing of a task */
8519 register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
8520 }
8521
8522 /*
8523 * Ignoring of pids is done at task switch. But we have to
8524 * check for those tasks that are currently running.
8525 * Always do this in case a pid was appended or removed.
8526 */
8527 on_each_cpu(ignore_task_cpu, tr, 1);
8528
8529 ftrace_update_pid_func();
8530 ftrace_startup_all(0);
8531
8532 *ppos += ret;
8533
8534 return ret;
8535 }
8536
8537 static ssize_t
ftrace_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)8538 ftrace_pid_write(struct file *filp, const char __user *ubuf,
8539 size_t cnt, loff_t *ppos)
8540 {
8541 return pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
8542 }
8543
8544 static ssize_t
ftrace_no_pid_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)8545 ftrace_no_pid_write(struct file *filp, const char __user *ubuf,
8546 size_t cnt, loff_t *ppos)
8547 {
8548 return pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
8549 }
8550
8551 static int
ftrace_pid_release(struct inode * inode,struct file * file)8552 ftrace_pid_release(struct inode *inode, struct file *file)
8553 {
8554 struct trace_array *tr = inode->i_private;
8555
8556 trace_array_put(tr);
8557
8558 return seq_release(inode, file);
8559 }
8560
8561 static const struct file_operations ftrace_pid_fops = {
8562 .open = ftrace_pid_open,
8563 .write = ftrace_pid_write,
8564 .read = seq_read,
8565 .llseek = tracing_lseek,
8566 .release = ftrace_pid_release,
8567 };
8568
8569 static const struct file_operations ftrace_no_pid_fops = {
8570 .open = ftrace_no_pid_open,
8571 .write = ftrace_no_pid_write,
8572 .read = seq_read,
8573 .llseek = tracing_lseek,
8574 .release = ftrace_pid_release,
8575 };
8576
ftrace_init_tracefs(struct trace_array * tr,struct dentry * d_tracer)8577 void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
8578 {
8579 trace_create_file("set_ftrace_pid", TRACE_MODE_WRITE, d_tracer,
8580 tr, &ftrace_pid_fops);
8581 trace_create_file("set_ftrace_notrace_pid", TRACE_MODE_WRITE,
8582 d_tracer, tr, &ftrace_no_pid_fops);
8583 }
8584
ftrace_init_tracefs_toplevel(struct trace_array * tr,struct dentry * d_tracer)8585 void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
8586 struct dentry *d_tracer)
8587 {
8588 /* Only the top level directory has the dyn_tracefs and profile */
8589 WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
8590
8591 ftrace_init_dyn_tracefs(d_tracer);
8592 ftrace_profile_tracefs(d_tracer);
8593 }
8594
8595 /**
8596 * ftrace_kill - kill ftrace
8597 *
8598 * This function should be used by panic code. It stops ftrace
8599 * but in a not so nice way. If you need to simply kill ftrace
8600 * from a non-atomic section, use ftrace_kill.
8601 */
ftrace_kill(void)8602 void ftrace_kill(void)
8603 {
8604 ftrace_disabled = 1;
8605 ftrace_enabled = 0;
8606 ftrace_trace_function = ftrace_stub;
8607 kprobe_ftrace_kill();
8608 }
8609
8610 /**
8611 * ftrace_is_dead - Test if ftrace is dead or not.
8612 *
8613 * Returns: 1 if ftrace is "dead", zero otherwise.
8614 */
ftrace_is_dead(void)8615 int ftrace_is_dead(void)
8616 {
8617 return ftrace_disabled;
8618 }
8619
8620 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
8621 /*
8622 * When registering ftrace_ops with IPMODIFY, it is necessary to make sure
8623 * it doesn't conflict with any direct ftrace_ops. If there is existing
8624 * direct ftrace_ops on a kernel function being patched, call
8625 * FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_PEER on it to enable sharing.
8626 *
8627 * @ops: ftrace_ops being registered.
8628 *
8629 * Returns:
8630 * 0 on success;
8631 * Negative on failure.
8632 */
prepare_direct_functions_for_ipmodify(struct ftrace_ops * ops)8633 static int prepare_direct_functions_for_ipmodify(struct ftrace_ops *ops)
8634 {
8635 struct ftrace_func_entry *entry;
8636 struct ftrace_hash *hash;
8637 struct ftrace_ops *op;
8638 int size, i, ret;
8639
8640 lockdep_assert_held_once(&direct_mutex);
8641
8642 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
8643 return 0;
8644
8645 hash = ops->func_hash->filter_hash;
8646 size = 1 << hash->size_bits;
8647 for (i = 0; i < size; i++) {
8648 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
8649 unsigned long ip = entry->ip;
8650 bool found_op = false;
8651
8652 mutex_lock(&ftrace_lock);
8653 do_for_each_ftrace_op(op, ftrace_ops_list) {
8654 if (!(op->flags & FTRACE_OPS_FL_DIRECT))
8655 continue;
8656 if (ops_references_ip(op, ip)) {
8657 found_op = true;
8658 break;
8659 }
8660 } while_for_each_ftrace_op(op);
8661 mutex_unlock(&ftrace_lock);
8662
8663 if (found_op) {
8664 if (!op->ops_func)
8665 return -EBUSY;
8666
8667 ret = op->ops_func(op, FTRACE_OPS_CMD_ENABLE_SHARE_IPMODIFY_PEER);
8668 if (ret)
8669 return ret;
8670 }
8671 }
8672 }
8673
8674 return 0;
8675 }
8676
8677 /*
8678 * Similar to prepare_direct_functions_for_ipmodify, clean up after ops
8679 * with IPMODIFY is unregistered. The cleanup is optional for most DIRECT
8680 * ops.
8681 */
cleanup_direct_functions_after_ipmodify(struct ftrace_ops * ops)8682 static void cleanup_direct_functions_after_ipmodify(struct ftrace_ops *ops)
8683 {
8684 struct ftrace_func_entry *entry;
8685 struct ftrace_hash *hash;
8686 struct ftrace_ops *op;
8687 int size, i;
8688
8689 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
8690 return;
8691
8692 mutex_lock(&direct_mutex);
8693
8694 hash = ops->func_hash->filter_hash;
8695 size = 1 << hash->size_bits;
8696 for (i = 0; i < size; i++) {
8697 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
8698 unsigned long ip = entry->ip;
8699 bool found_op = false;
8700
8701 mutex_lock(&ftrace_lock);
8702 do_for_each_ftrace_op(op, ftrace_ops_list) {
8703 if (!(op->flags & FTRACE_OPS_FL_DIRECT))
8704 continue;
8705 if (ops_references_ip(op, ip)) {
8706 found_op = true;
8707 break;
8708 }
8709 } while_for_each_ftrace_op(op);
8710 mutex_unlock(&ftrace_lock);
8711
8712 /* The cleanup is optional, ignore any errors */
8713 if (found_op && op->ops_func)
8714 op->ops_func(op, FTRACE_OPS_CMD_DISABLE_SHARE_IPMODIFY_PEER);
8715 }
8716 }
8717 mutex_unlock(&direct_mutex);
8718 }
8719
8720 #define lock_direct_mutex() mutex_lock(&direct_mutex)
8721 #define unlock_direct_mutex() mutex_unlock(&direct_mutex)
8722
8723 #else /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
8724
prepare_direct_functions_for_ipmodify(struct ftrace_ops * ops)8725 static int prepare_direct_functions_for_ipmodify(struct ftrace_ops *ops)
8726 {
8727 return 0;
8728 }
8729
cleanup_direct_functions_after_ipmodify(struct ftrace_ops * ops)8730 static void cleanup_direct_functions_after_ipmodify(struct ftrace_ops *ops)
8731 {
8732 }
8733
8734 #define lock_direct_mutex() do { } while (0)
8735 #define unlock_direct_mutex() do { } while (0)
8736
8737 #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
8738
8739 /*
8740 * Similar to register_ftrace_function, except we don't lock direct_mutex.
8741 */
register_ftrace_function_nolock(struct ftrace_ops * ops)8742 static int register_ftrace_function_nolock(struct ftrace_ops *ops)
8743 {
8744 int ret;
8745
8746 ftrace_ops_init(ops);
8747
8748 mutex_lock(&ftrace_lock);
8749
8750 ret = ftrace_startup(ops, 0);
8751
8752 mutex_unlock(&ftrace_lock);
8753
8754 return ret;
8755 }
8756
8757 /**
8758 * register_ftrace_function - register a function for profiling
8759 * @ops: ops structure that holds the function for profiling.
8760 *
8761 * Register a function to be called by all functions in the
8762 * kernel.
8763 *
8764 * Note: @ops->func and all the functions it calls must be labeled
8765 * with "notrace", otherwise it will go into a
8766 * recursive loop.
8767 */
register_ftrace_function(struct ftrace_ops * ops)8768 int register_ftrace_function(struct ftrace_ops *ops)
8769 {
8770 int ret;
8771
8772 lock_direct_mutex();
8773 ret = prepare_direct_functions_for_ipmodify(ops);
8774 if (ret < 0)
8775 goto out_unlock;
8776
8777 ret = register_ftrace_function_nolock(ops);
8778
8779 out_unlock:
8780 unlock_direct_mutex();
8781 return ret;
8782 }
8783 EXPORT_SYMBOL_GPL(register_ftrace_function);
8784
8785 /**
8786 * unregister_ftrace_function - unregister a function for profiling.
8787 * @ops: ops structure that holds the function to unregister
8788 *
8789 * Unregister a function that was added to be called by ftrace profiling.
8790 */
unregister_ftrace_function(struct ftrace_ops * ops)8791 int unregister_ftrace_function(struct ftrace_ops *ops)
8792 {
8793 int ret;
8794
8795 mutex_lock(&ftrace_lock);
8796 ret = ftrace_shutdown(ops, 0);
8797 mutex_unlock(&ftrace_lock);
8798
8799 cleanup_direct_functions_after_ipmodify(ops);
8800 return ret;
8801 }
8802 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
8803
symbols_cmp(const void * a,const void * b)8804 static int symbols_cmp(const void *a, const void *b)
8805 {
8806 const char **str_a = (const char **) a;
8807 const char **str_b = (const char **) b;
8808
8809 return strcmp(*str_a, *str_b);
8810 }
8811
8812 struct kallsyms_data {
8813 unsigned long *addrs;
8814 const char **syms;
8815 size_t cnt;
8816 size_t found;
8817 };
8818
8819 /* This function gets called for all kernel and module symbols
8820 * and returns 1 in case we resolved all the requested symbols,
8821 * 0 otherwise.
8822 */
kallsyms_callback(void * data,const char * name,unsigned long addr)8823 static int kallsyms_callback(void *data, const char *name, unsigned long addr)
8824 {
8825 struct kallsyms_data *args = data;
8826 const char **sym;
8827 int idx;
8828
8829 sym = bsearch(&name, args->syms, args->cnt, sizeof(*args->syms), symbols_cmp);
8830 if (!sym)
8831 return 0;
8832
8833 idx = sym - args->syms;
8834 if (args->addrs[idx])
8835 return 0;
8836
8837 if (!ftrace_location(addr))
8838 return 0;
8839
8840 args->addrs[idx] = addr;
8841 args->found++;
8842 return args->found == args->cnt ? 1 : 0;
8843 }
8844
8845 /**
8846 * ftrace_lookup_symbols - Lookup addresses for array of symbols
8847 *
8848 * @sorted_syms: array of symbols pointers symbols to resolve,
8849 * must be alphabetically sorted
8850 * @cnt: number of symbols/addresses in @syms/@addrs arrays
8851 * @addrs: array for storing resulting addresses
8852 *
8853 * This function looks up addresses for array of symbols provided in
8854 * @syms array (must be alphabetically sorted) and stores them in
8855 * @addrs array, which needs to be big enough to store at least @cnt
8856 * addresses.
8857 *
8858 * Returns: 0 if all provided symbols are found, -ESRCH otherwise.
8859 */
ftrace_lookup_symbols(const char ** sorted_syms,size_t cnt,unsigned long * addrs)8860 int ftrace_lookup_symbols(const char **sorted_syms, size_t cnt, unsigned long *addrs)
8861 {
8862 struct kallsyms_data args;
8863 int found_all;
8864
8865 memset(addrs, 0, sizeof(*addrs) * cnt);
8866 args.addrs = addrs;
8867 args.syms = sorted_syms;
8868 args.cnt = cnt;
8869 args.found = 0;
8870
8871 found_all = kallsyms_on_each_symbol(kallsyms_callback, &args);
8872 if (found_all)
8873 return 0;
8874 found_all = module_kallsyms_on_each_symbol(NULL, kallsyms_callback, &args);
8875 return found_all ? 0 : -ESRCH;
8876 }
8877
8878 #ifdef CONFIG_SYSCTL
8879
8880 #ifdef CONFIG_DYNAMIC_FTRACE
ftrace_startup_sysctl(void)8881 static void ftrace_startup_sysctl(void)
8882 {
8883 int command;
8884
8885 if (unlikely(ftrace_disabled))
8886 return;
8887
8888 /* Force update next time */
8889 saved_ftrace_func = NULL;
8890 /* ftrace_start_up is true if we want ftrace running */
8891 if (ftrace_start_up) {
8892 command = FTRACE_UPDATE_CALLS;
8893 if (ftrace_graph_active)
8894 command |= FTRACE_START_FUNC_RET;
8895 ftrace_startup_enable(command);
8896 }
8897 }
8898
ftrace_shutdown_sysctl(void)8899 static void ftrace_shutdown_sysctl(void)
8900 {
8901 int command;
8902
8903 if (unlikely(ftrace_disabled))
8904 return;
8905
8906 /* ftrace_start_up is true if ftrace is running */
8907 if (ftrace_start_up) {
8908 command = FTRACE_DISABLE_CALLS;
8909 if (ftrace_graph_active)
8910 command |= FTRACE_STOP_FUNC_RET;
8911 ftrace_run_update_code(command);
8912 }
8913 }
8914 #else
8915 # define ftrace_startup_sysctl() do { } while (0)
8916 # define ftrace_shutdown_sysctl() do { } while (0)
8917 #endif /* CONFIG_DYNAMIC_FTRACE */
8918
is_permanent_ops_registered(void)8919 static bool is_permanent_ops_registered(void)
8920 {
8921 struct ftrace_ops *op;
8922
8923 do_for_each_ftrace_op(op, ftrace_ops_list) {
8924 if (op->flags & FTRACE_OPS_FL_PERMANENT)
8925 return true;
8926 } while_for_each_ftrace_op(op);
8927
8928 return false;
8929 }
8930
8931 static int
ftrace_enable_sysctl(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)8932 ftrace_enable_sysctl(const struct ctl_table *table, int write,
8933 void *buffer, size_t *lenp, loff_t *ppos)
8934 {
8935 int ret;
8936
8937 guard(mutex)(&ftrace_lock);
8938
8939 if (unlikely(ftrace_disabled))
8940 return -ENODEV;
8941
8942 ret = proc_dointvec(table, write, buffer, lenp, ppos);
8943
8944 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
8945 return ret;
8946
8947 if (ftrace_enabled) {
8948
8949 /* we are starting ftrace again */
8950 if (rcu_dereference_protected(ftrace_ops_list,
8951 lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
8952 update_ftrace_function();
8953
8954 ftrace_startup_sysctl();
8955
8956 } else {
8957 if (is_permanent_ops_registered()) {
8958 ftrace_enabled = true;
8959 return -EBUSY;
8960 }
8961
8962 /* stopping ftrace calls (just send to ftrace_stub) */
8963 ftrace_trace_function = ftrace_stub;
8964
8965 ftrace_shutdown_sysctl();
8966 }
8967
8968 last_ftrace_enabled = !!ftrace_enabled;
8969 return 0;
8970 }
8971
8972 static const struct ctl_table ftrace_sysctls[] = {
8973 {
8974 .procname = "ftrace_enabled",
8975 .data = &ftrace_enabled,
8976 .maxlen = sizeof(int),
8977 .mode = 0644,
8978 .proc_handler = ftrace_enable_sysctl,
8979 },
8980 };
8981
ftrace_sysctl_init(void)8982 static int __init ftrace_sysctl_init(void)
8983 {
8984 register_sysctl_init("kernel", ftrace_sysctls);
8985 return 0;
8986 }
8987 late_initcall(ftrace_sysctl_init);
8988 #endif
8989