Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Dynamic function tracing support.
4 *
5 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
6 *
7 * Thanks goes to Ingo Molnar, for suggesting the idea.
8 * Mathieu Desnoyers, for suggesting postponing the modifications.
9 * Arjan van de Ven, for keeping me straight, and explaining to me
10 * the dangers of modifying code on the run.
11 */
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15#include <linux/spinlock.h>
16#include <linux/hardirq.h>
17#include <linux/uaccess.h>
18#include <linux/ftrace.h>
19#include <linux/percpu.h>
20#include <linux/sched.h>
21#include <linux/slab.h>
22#include <linux/init.h>
23#include <linux/list.h>
24#include <linux/module.h>
25#include <linux/memory.h>
26#include <linux/vmalloc.h>
27
28#include <trace/syscall.h>
29
30#include <asm/set_memory.h>
31#include <asm/kprobes.h>
32#include <asm/ftrace.h>
33#include <asm/nops.h>
34#include <asm/text-patching.h>
35
36#ifdef CONFIG_DYNAMIC_FTRACE
37
38static int ftrace_poke_late = 0;
39
40int ftrace_arch_code_modify_prepare(void)
41 __acquires(&text_mutex)
42{
43 /*
44 * Need to grab text_mutex to prevent a race from module loading
45 * and live kernel patching from changing the text permissions while
46 * ftrace has it set to "read/write".
47 */
48 mutex_lock(&text_mutex);
49 ftrace_poke_late = 1;
50 return 0;
51}
52
53int ftrace_arch_code_modify_post_process(void)
54 __releases(&text_mutex)
55{
56 /*
57 * ftrace_make_{call,nop}() may be called during
58 * module load, and we need to finish the text_poke_queue()
59 * that they do, here.
60 */
61 text_poke_finish();
62 ftrace_poke_late = 0;
63 mutex_unlock(&text_mutex);
64 return 0;
65}
66
67static const char *ftrace_nop_replace(void)
68{
69 return x86_nops[5];
70}
71
72static const char *ftrace_call_replace(unsigned long ip, unsigned long addr)
73{
74 return text_gen_insn(CALL_INSN_OPCODE, (void *)ip, (void *)addr);
75}
76
77static int ftrace_verify_code(unsigned long ip, const char *old_code)
78{
79 char cur_code[MCOUNT_INSN_SIZE];
80
81 /*
82 * Note:
83 * We are paranoid about modifying text, as if a bug was to happen, it
84 * could cause us to read or write to someplace that could cause harm.
85 * Carefully read and modify the code with probe_kernel_*(), and make
86 * sure what we read is what we expected it to be before modifying it.
87 */
88 /* read the text we want to modify */
89 if (copy_from_kernel_nofault(cur_code, (void *)ip, MCOUNT_INSN_SIZE)) {
90 WARN_ON(1);
91 return -EFAULT;
92 }
93
94 /* Make sure it is what we expect it to be */
95 if (memcmp(cur_code, old_code, MCOUNT_INSN_SIZE) != 0) {
96 WARN_ON(1);
97 return -EINVAL;
98 }
99
100 return 0;
101}
102
103/*
104 * Marked __ref because it calls text_poke_early() which is .init.text. That is
105 * ok because that call will happen early, during boot, when .init sections are
106 * still present.
107 */
108static int __ref
109ftrace_modify_code_direct(unsigned long ip, const char *old_code,
110 const char *new_code)
111{
112 int ret = ftrace_verify_code(ip, old_code);
113 if (ret)
114 return ret;
115
116 /* replace the text with the new text */
117 if (ftrace_poke_late)
118 text_poke_queue((void *)ip, new_code, MCOUNT_INSN_SIZE, NULL);
119 else
120 text_poke_early((void *)ip, new_code, MCOUNT_INSN_SIZE);
121 return 0;
122}
123
124int ftrace_make_nop(struct module *mod, struct dyn_ftrace *rec, unsigned long addr)
125{
126 unsigned long ip = rec->ip;
127 const char *new, *old;
128
129 old = ftrace_call_replace(ip, addr);
130 new = ftrace_nop_replace();
131
132 /*
133 * On boot up, and when modules are loaded, the MCOUNT_ADDR
134 * is converted to a nop, and will never become MCOUNT_ADDR
135 * again. This code is either running before SMP (on boot up)
136 * or before the code will ever be executed (module load).
137 * We do not want to use the breakpoint version in this case,
138 * just modify the code directly.
139 */
140 if (addr == MCOUNT_ADDR)
141 return ftrace_modify_code_direct(ip, old, new);
142
143 /*
144 * x86 overrides ftrace_replace_code -- this function will never be used
145 * in this case.
146 */
147 WARN_ONCE(1, "invalid use of ftrace_make_nop");
148 return -EINVAL;
149}
150
151int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
152{
153 unsigned long ip = rec->ip;
154 const char *new, *old;
155
156 old = ftrace_nop_replace();
157 new = ftrace_call_replace(ip, addr);
158
159 /* Should only be called when module is loaded */
160 return ftrace_modify_code_direct(rec->ip, old, new);
161}
162
163/*
164 * Should never be called:
165 * As it is only called by __ftrace_replace_code() which is called by
166 * ftrace_replace_code() that x86 overrides, and by ftrace_update_code()
167 * which is called to turn mcount into nops or nops into function calls
168 * but not to convert a function from not using regs to one that uses
169 * regs, which ftrace_modify_call() is for.
170 */
171int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
172 unsigned long addr)
173{
174 WARN_ON(1);
175 return -EINVAL;
176}
177
178int ftrace_update_ftrace_func(ftrace_func_t func)
179{
180 unsigned long ip;
181 const char *new;
182
183 ip = (unsigned long)(&ftrace_call);
184 new = ftrace_call_replace(ip, (unsigned long)func);
185 text_poke_bp((void *)ip, new, MCOUNT_INSN_SIZE, NULL);
186
187 ip = (unsigned long)(&ftrace_regs_call);
188 new = ftrace_call_replace(ip, (unsigned long)func);
189 text_poke_bp((void *)ip, new, MCOUNT_INSN_SIZE, NULL);
190
191 return 0;
192}
193
194void ftrace_replace_code(int enable)
195{
196 struct ftrace_rec_iter *iter;
197 struct dyn_ftrace *rec;
198 const char *new, *old;
199 int ret;
200
201 for_ftrace_rec_iter(iter) {
202 rec = ftrace_rec_iter_record(iter);
203
204 switch (ftrace_test_record(rec, enable)) {
205 case FTRACE_UPDATE_IGNORE:
206 default:
207 continue;
208
209 case FTRACE_UPDATE_MAKE_CALL:
210 old = ftrace_nop_replace();
211 break;
212
213 case FTRACE_UPDATE_MODIFY_CALL:
214 case FTRACE_UPDATE_MAKE_NOP:
215 old = ftrace_call_replace(rec->ip, ftrace_get_addr_curr(rec));
216 break;
217 }
218
219 ret = ftrace_verify_code(rec->ip, old);
220 if (ret) {
221 ftrace_bug(ret, rec);
222 return;
223 }
224 }
225
226 for_ftrace_rec_iter(iter) {
227 rec = ftrace_rec_iter_record(iter);
228
229 switch (ftrace_test_record(rec, enable)) {
230 case FTRACE_UPDATE_IGNORE:
231 default:
232 continue;
233
234 case FTRACE_UPDATE_MAKE_CALL:
235 case FTRACE_UPDATE_MODIFY_CALL:
236 new = ftrace_call_replace(rec->ip, ftrace_get_addr_new(rec));
237 break;
238
239 case FTRACE_UPDATE_MAKE_NOP:
240 new = ftrace_nop_replace();
241 break;
242 }
243
244 text_poke_queue((void *)rec->ip, new, MCOUNT_INSN_SIZE, NULL);
245 ftrace_update_record(rec, enable);
246 }
247 text_poke_finish();
248}
249
250void arch_ftrace_update_code(int command)
251{
252 ftrace_modify_all_code(command);
253}
254
255int __init ftrace_dyn_arch_init(void)
256{
257 return 0;
258}
259
260/* Currently only x86_64 supports dynamic trampolines */
261#ifdef CONFIG_X86_64
262
263#ifdef CONFIG_MODULES
264#include <linux/moduleloader.h>
265/* Module allocation simplifies allocating memory for code */
266static inline void *alloc_tramp(unsigned long size)
267{
268 return module_alloc(size);
269}
270static inline void tramp_free(void *tramp)
271{
272 module_memfree(tramp);
273}
274#else
275/* Trampolines can only be created if modules are supported */
276static inline void *alloc_tramp(unsigned long size)
277{
278 return NULL;
279}
280static inline void tramp_free(void *tramp) { }
281#endif
282
283/* Defined as markers to the end of the ftrace default trampolines */
284extern void ftrace_regs_caller_end(void);
285extern void ftrace_regs_caller_ret(void);
286extern void ftrace_caller_end(void);
287extern void ftrace_caller_op_ptr(void);
288extern void ftrace_regs_caller_op_ptr(void);
289extern void ftrace_regs_caller_jmp(void);
290
291/* movq function_trace_op(%rip), %rdx */
292/* 0x48 0x8b 0x15 <offset-to-ftrace_trace_op (4 bytes)> */
293#define OP_REF_SIZE 7
294
295/*
296 * The ftrace_ops is passed to the function callback. Since the
297 * trampoline only services a single ftrace_ops, we can pass in
298 * that ops directly.
299 *
300 * The ftrace_op_code_union is used to create a pointer to the
301 * ftrace_ops that will be passed to the callback function.
302 */
303union ftrace_op_code_union {
304 char code[OP_REF_SIZE];
305 struct {
306 char op[3];
307 int offset;
308 } __attribute__((packed));
309};
310
311#define RET_SIZE 1
312
313static unsigned long
314create_trampoline(struct ftrace_ops *ops, unsigned int *tramp_size)
315{
316 unsigned long start_offset;
317 unsigned long end_offset;
318 unsigned long op_offset;
319 unsigned long call_offset;
320 unsigned long jmp_offset;
321 unsigned long offset;
322 unsigned long npages;
323 unsigned long size;
324 unsigned long retq;
325 unsigned long *ptr;
326 void *trampoline;
327 void *ip;
328 /* 48 8b 15 <offset> is movq <offset>(%rip), %rdx */
329 unsigned const char op_ref[] = { 0x48, 0x8b, 0x15 };
330 union ftrace_op_code_union op_ptr;
331 int ret;
332
333 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
334 start_offset = (unsigned long)ftrace_regs_caller;
335 end_offset = (unsigned long)ftrace_regs_caller_end;
336 op_offset = (unsigned long)ftrace_regs_caller_op_ptr;
337 call_offset = (unsigned long)ftrace_regs_call;
338 jmp_offset = (unsigned long)ftrace_regs_caller_jmp;
339 } else {
340 start_offset = (unsigned long)ftrace_caller;
341 end_offset = (unsigned long)ftrace_caller_end;
342 op_offset = (unsigned long)ftrace_caller_op_ptr;
343 call_offset = (unsigned long)ftrace_call;
344 jmp_offset = 0;
345 }
346
347 size = end_offset - start_offset;
348
349 /*
350 * Allocate enough size to store the ftrace_caller code,
351 * the iret , as well as the address of the ftrace_ops this
352 * trampoline is used for.
353 */
354 trampoline = alloc_tramp(size + RET_SIZE + sizeof(void *));
355 if (!trampoline)
356 return 0;
357
358 *tramp_size = size + RET_SIZE + sizeof(void *);
359 npages = DIV_ROUND_UP(*tramp_size, PAGE_SIZE);
360
361 /* Copy ftrace_caller onto the trampoline memory */
362 ret = copy_from_kernel_nofault(trampoline, (void *)start_offset, size);
363 if (WARN_ON(ret < 0))
364 goto fail;
365
366 ip = trampoline + size;
367
368 /* The trampoline ends with ret(q) */
369 retq = (unsigned long)ftrace_stub;
370 ret = copy_from_kernel_nofault(ip, (void *)retq, RET_SIZE);
371 if (WARN_ON(ret < 0))
372 goto fail;
373
374 /* No need to test direct calls on created trampolines */
375 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
376 /* NOP the jnz 1f; but make sure it's a 2 byte jnz */
377 ip = trampoline + (jmp_offset - start_offset);
378 if (WARN_ON(*(char *)ip != 0x75))
379 goto fail;
380 ret = copy_from_kernel_nofault(ip, x86_nops[2], 2);
381 if (ret < 0)
382 goto fail;
383 }
384
385 /*
386 * The address of the ftrace_ops that is used for this trampoline
387 * is stored at the end of the trampoline. This will be used to
388 * load the third parameter for the callback. Basically, that
389 * location at the end of the trampoline takes the place of
390 * the global function_trace_op variable.
391 */
392
393 ptr = (unsigned long *)(trampoline + size + RET_SIZE);
394 *ptr = (unsigned long)ops;
395
396 op_offset -= start_offset;
397 memcpy(&op_ptr, trampoline + op_offset, OP_REF_SIZE);
398
399 /* Are we pointing to the reference? */
400 if (WARN_ON(memcmp(op_ptr.op, op_ref, 3) != 0))
401 goto fail;
402
403 /* Load the contents of ptr into the callback parameter */
404 offset = (unsigned long)ptr;
405 offset -= (unsigned long)trampoline + op_offset + OP_REF_SIZE;
406
407 op_ptr.offset = offset;
408
409 /* put in the new offset to the ftrace_ops */
410 memcpy(trampoline + op_offset, &op_ptr, OP_REF_SIZE);
411
412 /* put in the call to the function */
413 mutex_lock(&text_mutex);
414 call_offset -= start_offset;
415 memcpy(trampoline + call_offset,
416 text_gen_insn(CALL_INSN_OPCODE,
417 trampoline + call_offset,
418 ftrace_ops_get_func(ops)), CALL_INSN_SIZE);
419 mutex_unlock(&text_mutex);
420
421 /* ALLOC_TRAMP flags lets us know we created it */
422 ops->flags |= FTRACE_OPS_FL_ALLOC_TRAMP;
423
424 set_vm_flush_reset_perms(trampoline);
425
426 if (likely(system_state != SYSTEM_BOOTING))
427 set_memory_ro((unsigned long)trampoline, npages);
428 set_memory_x((unsigned long)trampoline, npages);
429 return (unsigned long)trampoline;
430fail:
431 tramp_free(trampoline);
432 return 0;
433}
434
435void set_ftrace_ops_ro(void)
436{
437 struct ftrace_ops *ops;
438 unsigned long start_offset;
439 unsigned long end_offset;
440 unsigned long npages;
441 unsigned long size;
442
443 do_for_each_ftrace_op(ops, ftrace_ops_list) {
444 if (!(ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP))
445 continue;
446
447 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
448 start_offset = (unsigned long)ftrace_regs_caller;
449 end_offset = (unsigned long)ftrace_regs_caller_end;
450 } else {
451 start_offset = (unsigned long)ftrace_caller;
452 end_offset = (unsigned long)ftrace_caller_end;
453 }
454 size = end_offset - start_offset;
455 size = size + RET_SIZE + sizeof(void *);
456 npages = DIV_ROUND_UP(size, PAGE_SIZE);
457 set_memory_ro((unsigned long)ops->trampoline, npages);
458 } while_for_each_ftrace_op(ops);
459}
460
461static unsigned long calc_trampoline_call_offset(bool save_regs)
462{
463 unsigned long start_offset;
464 unsigned long call_offset;
465
466 if (save_regs) {
467 start_offset = (unsigned long)ftrace_regs_caller;
468 call_offset = (unsigned long)ftrace_regs_call;
469 } else {
470 start_offset = (unsigned long)ftrace_caller;
471 call_offset = (unsigned long)ftrace_call;
472 }
473
474 return call_offset - start_offset;
475}
476
477void arch_ftrace_update_trampoline(struct ftrace_ops *ops)
478{
479 ftrace_func_t func;
480 unsigned long offset;
481 unsigned long ip;
482 unsigned int size;
483 const char *new;
484
485 if (!ops->trampoline) {
486 ops->trampoline = create_trampoline(ops, &size);
487 if (!ops->trampoline)
488 return;
489 ops->trampoline_size = size;
490 return;
491 }
492
493 /*
494 * The ftrace_ops caller may set up its own trampoline.
495 * In such a case, this code must not modify it.
496 */
497 if (!(ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP))
498 return;
499
500 offset = calc_trampoline_call_offset(ops->flags & FTRACE_OPS_FL_SAVE_REGS);
501 ip = ops->trampoline + offset;
502 func = ftrace_ops_get_func(ops);
503
504 mutex_lock(&text_mutex);
505 /* Do a safe modify in case the trampoline is executing */
506 new = ftrace_call_replace(ip, (unsigned long)func);
507 text_poke_bp((void *)ip, new, MCOUNT_INSN_SIZE, NULL);
508 mutex_unlock(&text_mutex);
509}
510
511/* Return the address of the function the trampoline calls */
512static void *addr_from_call(void *ptr)
513{
514 union text_poke_insn call;
515 int ret;
516
517 ret = copy_from_kernel_nofault(&call, ptr, CALL_INSN_SIZE);
518 if (WARN_ON_ONCE(ret < 0))
519 return NULL;
520
521 /* Make sure this is a call */
522 if (WARN_ON_ONCE(call.opcode != CALL_INSN_OPCODE)) {
523 pr_warn("Expected E8, got %x\n", call.opcode);
524 return NULL;
525 }
526
527 return ptr + CALL_INSN_SIZE + call.disp;
528}
529
530void prepare_ftrace_return(unsigned long self_addr, unsigned long *parent,
531 unsigned long frame_pointer);
532
533/*
534 * If the ops->trampoline was not allocated, then it probably
535 * has a static trampoline func, or is the ftrace caller itself.
536 */
537static void *static_tramp_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
538{
539 unsigned long offset;
540 bool save_regs = rec->flags & FTRACE_FL_REGS_EN;
541 void *ptr;
542
543 if (ops && ops->trampoline) {
544#ifdef CONFIG_FUNCTION_GRAPH_TRACER
545 /*
546 * We only know about function graph tracer setting as static
547 * trampoline.
548 */
549 if (ops->trampoline == FTRACE_GRAPH_ADDR)
550 return (void *)prepare_ftrace_return;
551#endif
552 return NULL;
553 }
554
555 offset = calc_trampoline_call_offset(save_regs);
556
557 if (save_regs)
558 ptr = (void *)FTRACE_REGS_ADDR + offset;
559 else
560 ptr = (void *)FTRACE_ADDR + offset;
561
562 return addr_from_call(ptr);
563}
564
565void *arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
566{
567 unsigned long offset;
568
569 /* If we didn't allocate this trampoline, consider it static */
570 if (!ops || !(ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP))
571 return static_tramp_func(ops, rec);
572
573 offset = calc_trampoline_call_offset(ops->flags & FTRACE_OPS_FL_SAVE_REGS);
574 return addr_from_call((void *)ops->trampoline + offset);
575}
576
577void arch_ftrace_trampoline_free(struct ftrace_ops *ops)
578{
579 if (!ops || !(ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP))
580 return;
581
582 tramp_free((void *)ops->trampoline);
583 ops->trampoline = 0;
584}
585
586#endif /* CONFIG_X86_64 */
587#endif /* CONFIG_DYNAMIC_FTRACE */
588
589#ifdef CONFIG_FUNCTION_GRAPH_TRACER
590
591#ifdef CONFIG_DYNAMIC_FTRACE
592extern void ftrace_graph_call(void);
593
594static const char *ftrace_jmp_replace(unsigned long ip, unsigned long addr)
595{
596 return text_gen_insn(JMP32_INSN_OPCODE, (void *)ip, (void *)addr);
597}
598
599static int ftrace_mod_jmp(unsigned long ip, void *func)
600{
601 const char *new;
602
603 new = ftrace_jmp_replace(ip, (unsigned long)func);
604 text_poke_bp((void *)ip, new, MCOUNT_INSN_SIZE, NULL);
605 return 0;
606}
607
608int ftrace_enable_ftrace_graph_caller(void)
609{
610 unsigned long ip = (unsigned long)(&ftrace_graph_call);
611
612 return ftrace_mod_jmp(ip, &ftrace_graph_caller);
613}
614
615int ftrace_disable_ftrace_graph_caller(void)
616{
617 unsigned long ip = (unsigned long)(&ftrace_graph_call);
618
619 return ftrace_mod_jmp(ip, &ftrace_stub);
620}
621
622#endif /* !CONFIG_DYNAMIC_FTRACE */
623
624/*
625 * Hook the return address and push it in the stack of return addrs
626 * in current thread info.
627 */
628void prepare_ftrace_return(unsigned long self_addr, unsigned long *parent,
629 unsigned long frame_pointer)
630{
631 unsigned long return_hooker = (unsigned long)&return_to_handler;
632 unsigned long old;
633 int faulted;
634
635 /*
636 * When resuming from suspend-to-ram, this function can be indirectly
637 * called from early CPU startup code while the CPU is in real mode,
638 * which would fail miserably. Make sure the stack pointer is a
639 * virtual address.
640 *
641 * This check isn't as accurate as virt_addr_valid(), but it should be
642 * good enough for this purpose, and it's fast.
643 */
644 if (unlikely((long)__builtin_frame_address(0) >= 0))
645 return;
646
647 if (unlikely(ftrace_graph_is_dead()))
648 return;
649
650 if (unlikely(atomic_read(¤t->tracing_graph_pause)))
651 return;
652
653 /*
654 * Protect against fault, even if it shouldn't
655 * happen. This tool is too much intrusive to
656 * ignore such a protection.
657 */
658 asm volatile(
659 "1: " _ASM_MOV " (%[parent]), %[old]\n"
660 "2: " _ASM_MOV " %[return_hooker], (%[parent])\n"
661 " movl $0, %[faulted]\n"
662 "3:\n"
663
664 ".section .fixup, \"ax\"\n"
665 "4: movl $1, %[faulted]\n"
666 " jmp 3b\n"
667 ".previous\n"
668
669 _ASM_EXTABLE(1b, 4b)
670 _ASM_EXTABLE(2b, 4b)
671
672 : [old] "=&r" (old), [faulted] "=r" (faulted)
673 : [parent] "r" (parent), [return_hooker] "r" (return_hooker)
674 : "memory"
675 );
676
677 if (unlikely(faulted)) {
678 ftrace_graph_stop();
679 WARN_ON(1);
680 return;
681 }
682
683 if (function_graph_enter(old, self_addr, frame_pointer, parent))
684 *parent = old;
685}
686#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Dynamic function tracing support.
4 *
5 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
6 *
7 * Thanks goes to Ingo Molnar, for suggesting the idea.
8 * Mathieu Desnoyers, for suggesting postponing the modifications.
9 * Arjan van de Ven, for keeping me straight, and explaining to me
10 * the dangers of modifying code on the run.
11 */
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15#include <linux/spinlock.h>
16#include <linux/hardirq.h>
17#include <linux/uaccess.h>
18#include <linux/ftrace.h>
19#include <linux/percpu.h>
20#include <linux/sched.h>
21#include <linux/slab.h>
22#include <linux/init.h>
23#include <linux/list.h>
24#include <linux/module.h>
25#include <linux/memory.h>
26#include <linux/vmalloc.h>
27#include <linux/set_memory.h>
28
29#include <trace/syscall.h>
30
31#include <asm/kprobes.h>
32#include <asm/ftrace.h>
33#include <asm/nops.h>
34#include <asm/text-patching.h>
35
36#ifdef CONFIG_DYNAMIC_FTRACE
37
38static int ftrace_poke_late = 0;
39
40void ftrace_arch_code_modify_prepare(void)
41 __acquires(&text_mutex)
42{
43 /*
44 * Need to grab text_mutex to prevent a race from module loading
45 * and live kernel patching from changing the text permissions while
46 * ftrace has it set to "read/write".
47 */
48 mutex_lock(&text_mutex);
49 ftrace_poke_late = 1;
50}
51
52void ftrace_arch_code_modify_post_process(void)
53 __releases(&text_mutex)
54{
55 /*
56 * ftrace_make_{call,nop}() may be called during
57 * module load, and we need to finish the text_poke_queue()
58 * that they do, here.
59 */
60 text_poke_finish();
61 ftrace_poke_late = 0;
62 mutex_unlock(&text_mutex);
63}
64
65static const char *ftrace_nop_replace(void)
66{
67 return x86_nops[5];
68}
69
70static const char *ftrace_call_replace(unsigned long ip, unsigned long addr)
71{
72 /*
73 * No need to translate into a callthunk. The trampoline does
74 * the depth accounting itself.
75 */
76 return text_gen_insn(CALL_INSN_OPCODE, (void *)ip, (void *)addr);
77}
78
79static int ftrace_verify_code(unsigned long ip, const char *old_code)
80{
81 char cur_code[MCOUNT_INSN_SIZE];
82
83 /*
84 * Note:
85 * We are paranoid about modifying text, as if a bug was to happen, it
86 * could cause us to read or write to someplace that could cause harm.
87 * Carefully read and modify the code with probe_kernel_*(), and make
88 * sure what we read is what we expected it to be before modifying it.
89 */
90 /* read the text we want to modify */
91 if (copy_from_kernel_nofault(cur_code, (void *)ip, MCOUNT_INSN_SIZE)) {
92 WARN_ON(1);
93 return -EFAULT;
94 }
95
96 /* Make sure it is what we expect it to be */
97 if (memcmp(cur_code, old_code, MCOUNT_INSN_SIZE) != 0) {
98 ftrace_expected = old_code;
99 WARN_ON(1);
100 return -EINVAL;
101 }
102
103 return 0;
104}
105
106/*
107 * Marked __ref because it calls text_poke_early() which is .init.text. That is
108 * ok because that call will happen early, during boot, when .init sections are
109 * still present.
110 */
111static int __ref
112ftrace_modify_code_direct(unsigned long ip, const char *old_code,
113 const char *new_code)
114{
115 int ret = ftrace_verify_code(ip, old_code);
116 if (ret)
117 return ret;
118
119 /* replace the text with the new text */
120 if (ftrace_poke_late)
121 text_poke_queue((void *)ip, new_code, MCOUNT_INSN_SIZE, NULL);
122 else
123 text_poke_early((void *)ip, new_code, MCOUNT_INSN_SIZE);
124 return 0;
125}
126
127int ftrace_make_nop(struct module *mod, struct dyn_ftrace *rec, unsigned long addr)
128{
129 unsigned long ip = rec->ip;
130 const char *new, *old;
131
132 old = ftrace_call_replace(ip, addr);
133 new = ftrace_nop_replace();
134
135 /*
136 * On boot up, and when modules are loaded, the MCOUNT_ADDR
137 * is converted to a nop, and will never become MCOUNT_ADDR
138 * again. This code is either running before SMP (on boot up)
139 * or before the code will ever be executed (module load).
140 * We do not want to use the breakpoint version in this case,
141 * just modify the code directly.
142 */
143 if (addr == MCOUNT_ADDR)
144 return ftrace_modify_code_direct(ip, old, new);
145
146 /*
147 * x86 overrides ftrace_replace_code -- this function will never be used
148 * in this case.
149 */
150 WARN_ONCE(1, "invalid use of ftrace_make_nop");
151 return -EINVAL;
152}
153
154int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
155{
156 unsigned long ip = rec->ip;
157 const char *new, *old;
158
159 old = ftrace_nop_replace();
160 new = ftrace_call_replace(ip, addr);
161
162 /* Should only be called when module is loaded */
163 return ftrace_modify_code_direct(rec->ip, old, new);
164}
165
166/*
167 * Should never be called:
168 * As it is only called by __ftrace_replace_code() which is called by
169 * ftrace_replace_code() that x86 overrides, and by ftrace_update_code()
170 * which is called to turn mcount into nops or nops into function calls
171 * but not to convert a function from not using regs to one that uses
172 * regs, which ftrace_modify_call() is for.
173 */
174int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
175 unsigned long addr)
176{
177 WARN_ON(1);
178 return -EINVAL;
179}
180
181int ftrace_update_ftrace_func(ftrace_func_t func)
182{
183 unsigned long ip;
184 const char *new;
185
186 ip = (unsigned long)(&ftrace_call);
187 new = ftrace_call_replace(ip, (unsigned long)func);
188 text_poke_bp((void *)ip, new, MCOUNT_INSN_SIZE, NULL);
189
190 ip = (unsigned long)(&ftrace_regs_call);
191 new = ftrace_call_replace(ip, (unsigned long)func);
192 text_poke_bp((void *)ip, new, MCOUNT_INSN_SIZE, NULL);
193
194 return 0;
195}
196
197void ftrace_replace_code(int enable)
198{
199 struct ftrace_rec_iter *iter;
200 struct dyn_ftrace *rec;
201 const char *new, *old;
202 int ret;
203
204 for_ftrace_rec_iter(iter) {
205 rec = ftrace_rec_iter_record(iter);
206
207 switch (ftrace_test_record(rec, enable)) {
208 case FTRACE_UPDATE_IGNORE:
209 default:
210 continue;
211
212 case FTRACE_UPDATE_MAKE_CALL:
213 old = ftrace_nop_replace();
214 break;
215
216 case FTRACE_UPDATE_MODIFY_CALL:
217 case FTRACE_UPDATE_MAKE_NOP:
218 old = ftrace_call_replace(rec->ip, ftrace_get_addr_curr(rec));
219 break;
220 }
221
222 ret = ftrace_verify_code(rec->ip, old);
223 if (ret) {
224 ftrace_expected = old;
225 ftrace_bug(ret, rec);
226 ftrace_expected = NULL;
227 return;
228 }
229 }
230
231 for_ftrace_rec_iter(iter) {
232 rec = ftrace_rec_iter_record(iter);
233
234 switch (ftrace_test_record(rec, enable)) {
235 case FTRACE_UPDATE_IGNORE:
236 default:
237 continue;
238
239 case FTRACE_UPDATE_MAKE_CALL:
240 case FTRACE_UPDATE_MODIFY_CALL:
241 new = ftrace_call_replace(rec->ip, ftrace_get_addr_new(rec));
242 break;
243
244 case FTRACE_UPDATE_MAKE_NOP:
245 new = ftrace_nop_replace();
246 break;
247 }
248
249 text_poke_queue((void *)rec->ip, new, MCOUNT_INSN_SIZE, NULL);
250 ftrace_update_record(rec, enable);
251 }
252 text_poke_finish();
253}
254
255void arch_ftrace_update_code(int command)
256{
257 ftrace_modify_all_code(command);
258}
259
260/* Currently only x86_64 supports dynamic trampolines */
261#ifdef CONFIG_X86_64
262
263#ifdef CONFIG_MODULES
264#include <linux/moduleloader.h>
265/* Module allocation simplifies allocating memory for code */
266static inline void *alloc_tramp(unsigned long size)
267{
268 return module_alloc(size);
269}
270static inline void tramp_free(void *tramp)
271{
272 module_memfree(tramp);
273}
274#else
275/* Trampolines can only be created if modules are supported */
276static inline void *alloc_tramp(unsigned long size)
277{
278 return NULL;
279}
280static inline void tramp_free(void *tramp) { }
281#endif
282
283/* Defined as markers to the end of the ftrace default trampolines */
284extern void ftrace_regs_caller_end(void);
285extern void ftrace_caller_end(void);
286extern void ftrace_caller_op_ptr(void);
287extern void ftrace_regs_caller_op_ptr(void);
288extern void ftrace_regs_caller_jmp(void);
289
290/* movq function_trace_op(%rip), %rdx */
291/* 0x48 0x8b 0x15 <offset-to-ftrace_trace_op (4 bytes)> */
292#define OP_REF_SIZE 7
293
294/*
295 * The ftrace_ops is passed to the function callback. Since the
296 * trampoline only services a single ftrace_ops, we can pass in
297 * that ops directly.
298 *
299 * The ftrace_op_code_union is used to create a pointer to the
300 * ftrace_ops that will be passed to the callback function.
301 */
302union ftrace_op_code_union {
303 char code[OP_REF_SIZE];
304 struct {
305 char op[3];
306 int offset;
307 } __attribute__((packed));
308};
309
310#define RET_SIZE (IS_ENABLED(CONFIG_RETPOLINE) ? 5 : 1 + IS_ENABLED(CONFIG_SLS))
311
312static unsigned long
313create_trampoline(struct ftrace_ops *ops, unsigned int *tramp_size)
314{
315 unsigned long start_offset;
316 unsigned long end_offset;
317 unsigned long op_offset;
318 unsigned long call_offset;
319 unsigned long jmp_offset;
320 unsigned long offset;
321 unsigned long npages;
322 unsigned long size;
323 unsigned long *ptr;
324 void *trampoline;
325 void *ip, *dest;
326 /* 48 8b 15 <offset> is movq <offset>(%rip), %rdx */
327 unsigned const char op_ref[] = { 0x48, 0x8b, 0x15 };
328 unsigned const char retq[] = { RET_INSN_OPCODE, INT3_INSN_OPCODE };
329 union ftrace_op_code_union op_ptr;
330 int ret;
331
332 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
333 start_offset = (unsigned long)ftrace_regs_caller;
334 end_offset = (unsigned long)ftrace_regs_caller_end;
335 op_offset = (unsigned long)ftrace_regs_caller_op_ptr;
336 call_offset = (unsigned long)ftrace_regs_call;
337 jmp_offset = (unsigned long)ftrace_regs_caller_jmp;
338 } else {
339 start_offset = (unsigned long)ftrace_caller;
340 end_offset = (unsigned long)ftrace_caller_end;
341 op_offset = (unsigned long)ftrace_caller_op_ptr;
342 call_offset = (unsigned long)ftrace_call;
343 jmp_offset = 0;
344 }
345
346 size = end_offset - start_offset;
347
348 /*
349 * Allocate enough size to store the ftrace_caller code,
350 * the iret , as well as the address of the ftrace_ops this
351 * trampoline is used for.
352 */
353 trampoline = alloc_tramp(size + RET_SIZE + sizeof(void *));
354 if (!trampoline)
355 return 0;
356
357 *tramp_size = size + RET_SIZE + sizeof(void *);
358 npages = DIV_ROUND_UP(*tramp_size, PAGE_SIZE);
359
360 /* Copy ftrace_caller onto the trampoline memory */
361 ret = copy_from_kernel_nofault(trampoline, (void *)start_offset, size);
362 if (WARN_ON(ret < 0))
363 goto fail;
364
365 ip = trampoline + size;
366 if (cpu_feature_enabled(X86_FEATURE_RETHUNK))
367 __text_gen_insn(ip, JMP32_INSN_OPCODE, ip, x86_return_thunk, JMP32_INSN_SIZE);
368 else
369 memcpy(ip, retq, sizeof(retq));
370
371 /* No need to test direct calls on created trampolines */
372 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
373 /* NOP the jnz 1f; but make sure it's a 2 byte jnz */
374 ip = trampoline + (jmp_offset - start_offset);
375 if (WARN_ON(*(char *)ip != 0x75))
376 goto fail;
377 ret = copy_from_kernel_nofault(ip, x86_nops[2], 2);
378 if (ret < 0)
379 goto fail;
380 }
381
382 /*
383 * The address of the ftrace_ops that is used for this trampoline
384 * is stored at the end of the trampoline. This will be used to
385 * load the third parameter for the callback. Basically, that
386 * location at the end of the trampoline takes the place of
387 * the global function_trace_op variable.
388 */
389
390 ptr = (unsigned long *)(trampoline + size + RET_SIZE);
391 *ptr = (unsigned long)ops;
392
393 op_offset -= start_offset;
394 memcpy(&op_ptr, trampoline + op_offset, OP_REF_SIZE);
395
396 /* Are we pointing to the reference? */
397 if (WARN_ON(memcmp(op_ptr.op, op_ref, 3) != 0))
398 goto fail;
399
400 /* Load the contents of ptr into the callback parameter */
401 offset = (unsigned long)ptr;
402 offset -= (unsigned long)trampoline + op_offset + OP_REF_SIZE;
403
404 op_ptr.offset = offset;
405
406 /* put in the new offset to the ftrace_ops */
407 memcpy(trampoline + op_offset, &op_ptr, OP_REF_SIZE);
408
409 /* put in the call to the function */
410 mutex_lock(&text_mutex);
411 call_offset -= start_offset;
412 /*
413 * No need to translate into a callthunk. The trampoline does
414 * the depth accounting before the call already.
415 */
416 dest = ftrace_ops_get_func(ops);
417 memcpy(trampoline + call_offset,
418 text_gen_insn(CALL_INSN_OPCODE, trampoline + call_offset, dest),
419 CALL_INSN_SIZE);
420 mutex_unlock(&text_mutex);
421
422 /* ALLOC_TRAMP flags lets us know we created it */
423 ops->flags |= FTRACE_OPS_FL_ALLOC_TRAMP;
424
425 set_memory_rox((unsigned long)trampoline, npages);
426 return (unsigned long)trampoline;
427fail:
428 tramp_free(trampoline);
429 return 0;
430}
431
432void set_ftrace_ops_ro(void)
433{
434 struct ftrace_ops *ops;
435 unsigned long start_offset;
436 unsigned long end_offset;
437 unsigned long npages;
438 unsigned long size;
439
440 do_for_each_ftrace_op(ops, ftrace_ops_list) {
441 if (!(ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP))
442 continue;
443
444 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
445 start_offset = (unsigned long)ftrace_regs_caller;
446 end_offset = (unsigned long)ftrace_regs_caller_end;
447 } else {
448 start_offset = (unsigned long)ftrace_caller;
449 end_offset = (unsigned long)ftrace_caller_end;
450 }
451 size = end_offset - start_offset;
452 size = size + RET_SIZE + sizeof(void *);
453 npages = DIV_ROUND_UP(size, PAGE_SIZE);
454 set_memory_ro((unsigned long)ops->trampoline, npages);
455 } while_for_each_ftrace_op(ops);
456}
457
458static unsigned long calc_trampoline_call_offset(bool save_regs)
459{
460 unsigned long start_offset;
461 unsigned long call_offset;
462
463 if (save_regs) {
464 start_offset = (unsigned long)ftrace_regs_caller;
465 call_offset = (unsigned long)ftrace_regs_call;
466 } else {
467 start_offset = (unsigned long)ftrace_caller;
468 call_offset = (unsigned long)ftrace_call;
469 }
470
471 return call_offset - start_offset;
472}
473
474void arch_ftrace_update_trampoline(struct ftrace_ops *ops)
475{
476 ftrace_func_t func;
477 unsigned long offset;
478 unsigned long ip;
479 unsigned int size;
480 const char *new;
481
482 if (!ops->trampoline) {
483 ops->trampoline = create_trampoline(ops, &size);
484 if (!ops->trampoline)
485 return;
486 ops->trampoline_size = size;
487 return;
488 }
489
490 /*
491 * The ftrace_ops caller may set up its own trampoline.
492 * In such a case, this code must not modify it.
493 */
494 if (!(ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP))
495 return;
496
497 offset = calc_trampoline_call_offset(ops->flags & FTRACE_OPS_FL_SAVE_REGS);
498 ip = ops->trampoline + offset;
499 func = ftrace_ops_get_func(ops);
500
501 mutex_lock(&text_mutex);
502 /* Do a safe modify in case the trampoline is executing */
503 new = ftrace_call_replace(ip, (unsigned long)func);
504 text_poke_bp((void *)ip, new, MCOUNT_INSN_SIZE, NULL);
505 mutex_unlock(&text_mutex);
506}
507
508/* Return the address of the function the trampoline calls */
509static void *addr_from_call(void *ptr)
510{
511 union text_poke_insn call;
512 int ret;
513
514 ret = copy_from_kernel_nofault(&call, ptr, CALL_INSN_SIZE);
515 if (WARN_ON_ONCE(ret < 0))
516 return NULL;
517
518 /* Make sure this is a call */
519 if (WARN_ON_ONCE(call.opcode != CALL_INSN_OPCODE)) {
520 pr_warn("Expected E8, got %x\n", call.opcode);
521 return NULL;
522 }
523
524 return ptr + CALL_INSN_SIZE + call.disp;
525}
526
527/*
528 * If the ops->trampoline was not allocated, then it probably
529 * has a static trampoline func, or is the ftrace caller itself.
530 */
531static void *static_tramp_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
532{
533 unsigned long offset;
534 bool save_regs = rec->flags & FTRACE_FL_REGS_EN;
535 void *ptr;
536
537 if (ops && ops->trampoline) {
538#if !defined(CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS) && \
539 defined(CONFIG_FUNCTION_GRAPH_TRACER)
540 /*
541 * We only know about function graph tracer setting as static
542 * trampoline.
543 */
544 if (ops->trampoline == FTRACE_GRAPH_ADDR)
545 return (void *)prepare_ftrace_return;
546#endif
547 return NULL;
548 }
549
550 offset = calc_trampoline_call_offset(save_regs);
551
552 if (save_regs)
553 ptr = (void *)FTRACE_REGS_ADDR + offset;
554 else
555 ptr = (void *)FTRACE_ADDR + offset;
556
557 return addr_from_call(ptr);
558}
559
560void *arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
561{
562 unsigned long offset;
563
564 /* If we didn't allocate this trampoline, consider it static */
565 if (!ops || !(ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP))
566 return static_tramp_func(ops, rec);
567
568 offset = calc_trampoline_call_offset(ops->flags & FTRACE_OPS_FL_SAVE_REGS);
569 return addr_from_call((void *)ops->trampoline + offset);
570}
571
572void arch_ftrace_trampoline_free(struct ftrace_ops *ops)
573{
574 if (!ops || !(ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP))
575 return;
576
577 tramp_free((void *)ops->trampoline);
578 ops->trampoline = 0;
579}
580
581#endif /* CONFIG_X86_64 */
582#endif /* CONFIG_DYNAMIC_FTRACE */
583
584#ifdef CONFIG_FUNCTION_GRAPH_TRACER
585
586#if defined(CONFIG_DYNAMIC_FTRACE) && !defined(CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS)
587extern void ftrace_graph_call(void);
588static const char *ftrace_jmp_replace(unsigned long ip, unsigned long addr)
589{
590 return text_gen_insn(JMP32_INSN_OPCODE, (void *)ip, (void *)addr);
591}
592
593static int ftrace_mod_jmp(unsigned long ip, void *func)
594{
595 const char *new;
596
597 new = ftrace_jmp_replace(ip, (unsigned long)func);
598 text_poke_bp((void *)ip, new, MCOUNT_INSN_SIZE, NULL);
599 return 0;
600}
601
602int ftrace_enable_ftrace_graph_caller(void)
603{
604 unsigned long ip = (unsigned long)(&ftrace_graph_call);
605
606 return ftrace_mod_jmp(ip, &ftrace_graph_caller);
607}
608
609int ftrace_disable_ftrace_graph_caller(void)
610{
611 unsigned long ip = (unsigned long)(&ftrace_graph_call);
612
613 return ftrace_mod_jmp(ip, &ftrace_stub);
614}
615#endif /* CONFIG_DYNAMIC_FTRACE && !CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS */
616
617/*
618 * Hook the return address and push it in the stack of return addrs
619 * in current thread info.
620 */
621void prepare_ftrace_return(unsigned long ip, unsigned long *parent,
622 unsigned long frame_pointer)
623{
624 unsigned long return_hooker = (unsigned long)&return_to_handler;
625 int bit;
626
627 /*
628 * When resuming from suspend-to-ram, this function can be indirectly
629 * called from early CPU startup code while the CPU is in real mode,
630 * which would fail miserably. Make sure the stack pointer is a
631 * virtual address.
632 *
633 * This check isn't as accurate as virt_addr_valid(), but it should be
634 * good enough for this purpose, and it's fast.
635 */
636 if (unlikely((long)__builtin_frame_address(0) >= 0))
637 return;
638
639 if (unlikely(ftrace_graph_is_dead()))
640 return;
641
642 if (unlikely(atomic_read(¤t->tracing_graph_pause)))
643 return;
644
645 bit = ftrace_test_recursion_trylock(ip, *parent);
646 if (bit < 0)
647 return;
648
649 if (!function_graph_enter(*parent, ip, frame_pointer, parent))
650 *parent = return_hooker;
651
652 ftrace_test_recursion_unlock(bit);
653}
654
655#ifdef CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS
656void ftrace_graph_func(unsigned long ip, unsigned long parent_ip,
657 struct ftrace_ops *op, struct ftrace_regs *fregs)
658{
659 struct pt_regs *regs = &fregs->regs;
660 unsigned long *stack = (unsigned long *)kernel_stack_pointer(regs);
661
662 prepare_ftrace_return(ip, (unsigned long *)stack, 0);
663}
664#endif
665
666#endif /* CONFIG_FUNCTION_GRAPH_TRACER */