ftrace.c

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