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