xref: /openbmc/linux/arch/x86/kernel/kprobes/opt.c (revision 6aa7de05)
1 /*
2  *  Kernel Probes Jump Optimization (Optprobes)
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License as published by
6  * the Free Software Foundation; either version 2 of the License, or
7  * (at your option) any later version.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write to the Free Software
16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17  *
18  * Copyright (C) IBM Corporation, 2002, 2004
19  * Copyright (C) Hitachi Ltd., 2012
20  */
21 #include <linux/kprobes.h>
22 #include <linux/ptrace.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/hardirq.h>
26 #include <linux/preempt.h>
27 #include <linux/extable.h>
28 #include <linux/kdebug.h>
29 #include <linux/kallsyms.h>
30 #include <linux/ftrace.h>
31 #include <linux/frame.h>
32 
33 #include <asm/text-patching.h>
34 #include <asm/cacheflush.h>
35 #include <asm/desc.h>
36 #include <asm/pgtable.h>
37 #include <linux/uaccess.h>
38 #include <asm/alternative.h>
39 #include <asm/insn.h>
40 #include <asm/debugreg.h>
41 #include <asm/set_memory.h>
42 #include <asm/sections.h>
43 
44 #include "common.h"
45 
46 unsigned long __recover_optprobed_insn(kprobe_opcode_t *buf, unsigned long addr)
47 {
48 	struct optimized_kprobe *op;
49 	struct kprobe *kp;
50 	long offs;
51 	int i;
52 
53 	for (i = 0; i < RELATIVEJUMP_SIZE; i++) {
54 		kp = get_kprobe((void *)addr - i);
55 		/* This function only handles jump-optimized kprobe */
56 		if (kp && kprobe_optimized(kp)) {
57 			op = container_of(kp, struct optimized_kprobe, kp);
58 			/* If op->list is not empty, op is under optimizing */
59 			if (list_empty(&op->list))
60 				goto found;
61 		}
62 	}
63 
64 	return addr;
65 found:
66 	/*
67 	 * If the kprobe can be optimized, original bytes which can be
68 	 * overwritten by jump destination address. In this case, original
69 	 * bytes must be recovered from op->optinsn.copied_insn buffer.
70 	 */
71 	if (probe_kernel_read(buf, (void *)addr,
72 		MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
73 		return 0UL;
74 
75 	if (addr == (unsigned long)kp->addr) {
76 		buf[0] = kp->opcode;
77 		memcpy(buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
78 	} else {
79 		offs = addr - (unsigned long)kp->addr - 1;
80 		memcpy(buf, op->optinsn.copied_insn + offs, RELATIVE_ADDR_SIZE - offs);
81 	}
82 
83 	return (unsigned long)buf;
84 }
85 
86 /* Insert a move instruction which sets a pointer to eax/rdi (1st arg). */
87 static void synthesize_set_arg1(kprobe_opcode_t *addr, unsigned long val)
88 {
89 #ifdef CONFIG_X86_64
90 	*addr++ = 0x48;
91 	*addr++ = 0xbf;
92 #else
93 	*addr++ = 0xb8;
94 #endif
95 	*(unsigned long *)addr = val;
96 }
97 
98 asm (
99 			"optprobe_template_func:\n"
100 			".global optprobe_template_entry\n"
101 			"optprobe_template_entry:\n"
102 #ifdef CONFIG_X86_64
103 			/* We don't bother saving the ss register */
104 			"	pushq %rsp\n"
105 			"	pushfq\n"
106 			SAVE_REGS_STRING
107 			"	movq %rsp, %rsi\n"
108 			".global optprobe_template_val\n"
109 			"optprobe_template_val:\n"
110 			ASM_NOP5
111 			ASM_NOP5
112 			".global optprobe_template_call\n"
113 			"optprobe_template_call:\n"
114 			ASM_NOP5
115 			/* Move flags to rsp */
116 			"	movq 144(%rsp), %rdx\n"
117 			"	movq %rdx, 152(%rsp)\n"
118 			RESTORE_REGS_STRING
119 			/* Skip flags entry */
120 			"	addq $8, %rsp\n"
121 			"	popfq\n"
122 #else /* CONFIG_X86_32 */
123 			"	pushf\n"
124 			SAVE_REGS_STRING
125 			"	movl %esp, %edx\n"
126 			".global optprobe_template_val\n"
127 			"optprobe_template_val:\n"
128 			ASM_NOP5
129 			".global optprobe_template_call\n"
130 			"optprobe_template_call:\n"
131 			ASM_NOP5
132 			RESTORE_REGS_STRING
133 			"	addl $4, %esp\n"	/* skip cs */
134 			"	popf\n"
135 #endif
136 			".global optprobe_template_end\n"
137 			"optprobe_template_end:\n"
138 			".type optprobe_template_func, @function\n"
139 			".size optprobe_template_func, .-optprobe_template_func\n");
140 
141 void optprobe_template_func(void);
142 STACK_FRAME_NON_STANDARD(optprobe_template_func);
143 
144 #define TMPL_MOVE_IDX \
145 	((long)&optprobe_template_val - (long)&optprobe_template_entry)
146 #define TMPL_CALL_IDX \
147 	((long)&optprobe_template_call - (long)&optprobe_template_entry)
148 #define TMPL_END_IDX \
149 	((long)&optprobe_template_end - (long)&optprobe_template_entry)
150 
151 #define INT3_SIZE sizeof(kprobe_opcode_t)
152 
153 /* Optimized kprobe call back function: called from optinsn */
154 static void
155 optimized_callback(struct optimized_kprobe *op, struct pt_regs *regs)
156 {
157 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
158 	unsigned long flags;
159 
160 	/* This is possible if op is under delayed unoptimizing */
161 	if (kprobe_disabled(&op->kp))
162 		return;
163 
164 	local_irq_save(flags);
165 	if (kprobe_running()) {
166 		kprobes_inc_nmissed_count(&op->kp);
167 	} else {
168 		/* Save skipped registers */
169 #ifdef CONFIG_X86_64
170 		regs->cs = __KERNEL_CS;
171 #else
172 		regs->cs = __KERNEL_CS | get_kernel_rpl();
173 		regs->gs = 0;
174 #endif
175 		regs->ip = (unsigned long)op->kp.addr + INT3_SIZE;
176 		regs->orig_ax = ~0UL;
177 
178 		__this_cpu_write(current_kprobe, &op->kp);
179 		kcb->kprobe_status = KPROBE_HIT_ACTIVE;
180 		opt_pre_handler(&op->kp, regs);
181 		__this_cpu_write(current_kprobe, NULL);
182 	}
183 	local_irq_restore(flags);
184 }
185 NOKPROBE_SYMBOL(optimized_callback);
186 
187 static int copy_optimized_instructions(u8 *dest, u8 *src)
188 {
189 	struct insn insn;
190 	int len = 0, ret;
191 
192 	while (len < RELATIVEJUMP_SIZE) {
193 		ret = __copy_instruction(dest + len, src + len, &insn);
194 		if (!ret || !can_boost(&insn, src + len))
195 			return -EINVAL;
196 		len += ret;
197 	}
198 	/* Check whether the address range is reserved */
199 	if (ftrace_text_reserved(src, src + len - 1) ||
200 	    alternatives_text_reserved(src, src + len - 1) ||
201 	    jump_label_text_reserved(src, src + len - 1))
202 		return -EBUSY;
203 
204 	return len;
205 }
206 
207 /* Check whether insn is indirect jump */
208 static int insn_is_indirect_jump(struct insn *insn)
209 {
210 	return ((insn->opcode.bytes[0] == 0xff &&
211 		(X86_MODRM_REG(insn->modrm.value) & 6) == 4) || /* Jump */
212 		insn->opcode.bytes[0] == 0xea);	/* Segment based jump */
213 }
214 
215 /* Check whether insn jumps into specified address range */
216 static int insn_jump_into_range(struct insn *insn, unsigned long start, int len)
217 {
218 	unsigned long target = 0;
219 
220 	switch (insn->opcode.bytes[0]) {
221 	case 0xe0:	/* loopne */
222 	case 0xe1:	/* loope */
223 	case 0xe2:	/* loop */
224 	case 0xe3:	/* jcxz */
225 	case 0xe9:	/* near relative jump */
226 	case 0xeb:	/* short relative jump */
227 		break;
228 	case 0x0f:
229 		if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */
230 			break;
231 		return 0;
232 	default:
233 		if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
234 			break;
235 		return 0;
236 	}
237 	target = (unsigned long)insn->next_byte + insn->immediate.value;
238 
239 	return (start <= target && target <= start + len);
240 }
241 
242 /* Decode whole function to ensure any instructions don't jump into target */
243 static int can_optimize(unsigned long paddr)
244 {
245 	unsigned long addr, size = 0, offset = 0;
246 	struct insn insn;
247 	kprobe_opcode_t buf[MAX_INSN_SIZE];
248 
249 	/* Lookup symbol including addr */
250 	if (!kallsyms_lookup_size_offset(paddr, &size, &offset))
251 		return 0;
252 
253 	/*
254 	 * Do not optimize in the entry code due to the unstable
255 	 * stack handling and registers setup.
256 	 */
257 	if (((paddr >= (unsigned long)__entry_text_start) &&
258 	     (paddr <  (unsigned long)__entry_text_end)) ||
259 	    ((paddr >= (unsigned long)__irqentry_text_start) &&
260 	     (paddr <  (unsigned long)__irqentry_text_end)))
261 		return 0;
262 
263 	/* Check there is enough space for a relative jump. */
264 	if (size - offset < RELATIVEJUMP_SIZE)
265 		return 0;
266 
267 	/* Decode instructions */
268 	addr = paddr - offset;
269 	while (addr < paddr - offset + size) { /* Decode until function end */
270 		unsigned long recovered_insn;
271 		if (search_exception_tables(addr))
272 			/*
273 			 * Since some fixup code will jumps into this function,
274 			 * we can't optimize kprobe in this function.
275 			 */
276 			return 0;
277 		recovered_insn = recover_probed_instruction(buf, addr);
278 		if (!recovered_insn)
279 			return 0;
280 		kernel_insn_init(&insn, (void *)recovered_insn, MAX_INSN_SIZE);
281 		insn_get_length(&insn);
282 		/* Another subsystem puts a breakpoint */
283 		if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
284 			return 0;
285 		/* Recover address */
286 		insn.kaddr = (void *)addr;
287 		insn.next_byte = (void *)(addr + insn.length);
288 		/* Check any instructions don't jump into target */
289 		if (insn_is_indirect_jump(&insn) ||
290 		    insn_jump_into_range(&insn, paddr + INT3_SIZE,
291 					 RELATIVE_ADDR_SIZE))
292 			return 0;
293 		addr += insn.length;
294 	}
295 
296 	return 1;
297 }
298 
299 /* Check optimized_kprobe can actually be optimized. */
300 int arch_check_optimized_kprobe(struct optimized_kprobe *op)
301 {
302 	int i;
303 	struct kprobe *p;
304 
305 	for (i = 1; i < op->optinsn.size; i++) {
306 		p = get_kprobe(op->kp.addr + i);
307 		if (p && !kprobe_disabled(p))
308 			return -EEXIST;
309 	}
310 
311 	return 0;
312 }
313 
314 /* Check the addr is within the optimized instructions. */
315 int arch_within_optimized_kprobe(struct optimized_kprobe *op,
316 				 unsigned long addr)
317 {
318 	return ((unsigned long)op->kp.addr <= addr &&
319 		(unsigned long)op->kp.addr + op->optinsn.size > addr);
320 }
321 
322 /* Free optimized instruction slot */
323 static
324 void __arch_remove_optimized_kprobe(struct optimized_kprobe *op, int dirty)
325 {
326 	if (op->optinsn.insn) {
327 		free_optinsn_slot(op->optinsn.insn, dirty);
328 		op->optinsn.insn = NULL;
329 		op->optinsn.size = 0;
330 	}
331 }
332 
333 void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
334 {
335 	__arch_remove_optimized_kprobe(op, 1);
336 }
337 
338 /*
339  * Copy replacing target instructions
340  * Target instructions MUST be relocatable (checked inside)
341  * This is called when new aggr(opt)probe is allocated or reused.
342  */
343 int arch_prepare_optimized_kprobe(struct optimized_kprobe *op,
344 				  struct kprobe *__unused)
345 {
346 	u8 *buf;
347 	int ret;
348 	long rel;
349 
350 	if (!can_optimize((unsigned long)op->kp.addr))
351 		return -EILSEQ;
352 
353 	op->optinsn.insn = get_optinsn_slot();
354 	if (!op->optinsn.insn)
355 		return -ENOMEM;
356 
357 	/*
358 	 * Verify if the address gap is in 2GB range, because this uses
359 	 * a relative jump.
360 	 */
361 	rel = (long)op->optinsn.insn - (long)op->kp.addr + RELATIVEJUMP_SIZE;
362 	if (abs(rel) > 0x7fffffff) {
363 		__arch_remove_optimized_kprobe(op, 0);
364 		return -ERANGE;
365 	}
366 
367 	buf = (u8 *)op->optinsn.insn;
368 	set_memory_rw((unsigned long)buf & PAGE_MASK, 1);
369 
370 	/* Copy instructions into the out-of-line buffer */
371 	ret = copy_optimized_instructions(buf + TMPL_END_IDX, op->kp.addr);
372 	if (ret < 0) {
373 		__arch_remove_optimized_kprobe(op, 0);
374 		return ret;
375 	}
376 	op->optinsn.size = ret;
377 
378 	/* Copy arch-dep-instance from template */
379 	memcpy(buf, &optprobe_template_entry, TMPL_END_IDX);
380 
381 	/* Set probe information */
382 	synthesize_set_arg1(buf + TMPL_MOVE_IDX, (unsigned long)op);
383 
384 	/* Set probe function call */
385 	synthesize_relcall(buf + TMPL_CALL_IDX, optimized_callback);
386 
387 	/* Set returning jmp instruction at the tail of out-of-line buffer */
388 	synthesize_reljump(buf + TMPL_END_IDX + op->optinsn.size,
389 			   (u8 *)op->kp.addr + op->optinsn.size);
390 
391 	set_memory_ro((unsigned long)buf & PAGE_MASK, 1);
392 
393 	flush_icache_range((unsigned long) buf,
394 			   (unsigned long) buf + TMPL_END_IDX +
395 			   op->optinsn.size + RELATIVEJUMP_SIZE);
396 	return 0;
397 }
398 
399 /*
400  * Replace breakpoints (int3) with relative jumps.
401  * Caller must call with locking kprobe_mutex and text_mutex.
402  */
403 void arch_optimize_kprobes(struct list_head *oplist)
404 {
405 	struct optimized_kprobe *op, *tmp;
406 	u8 insn_buf[RELATIVEJUMP_SIZE];
407 
408 	list_for_each_entry_safe(op, tmp, oplist, list) {
409 		s32 rel = (s32)((long)op->optinsn.insn -
410 			((long)op->kp.addr + RELATIVEJUMP_SIZE));
411 
412 		WARN_ON(kprobe_disabled(&op->kp));
413 
414 		/* Backup instructions which will be replaced by jump address */
415 		memcpy(op->optinsn.copied_insn, op->kp.addr + INT3_SIZE,
416 		       RELATIVE_ADDR_SIZE);
417 
418 		insn_buf[0] = RELATIVEJUMP_OPCODE;
419 		*(s32 *)(&insn_buf[1]) = rel;
420 
421 		text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
422 			     op->optinsn.insn);
423 
424 		list_del_init(&op->list);
425 	}
426 }
427 
428 /* Replace a relative jump with a breakpoint (int3).  */
429 void arch_unoptimize_kprobe(struct optimized_kprobe *op)
430 {
431 	u8 insn_buf[RELATIVEJUMP_SIZE];
432 
433 	/* Set int3 to first byte for kprobes */
434 	insn_buf[0] = BREAKPOINT_INSTRUCTION;
435 	memcpy(insn_buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
436 	text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
437 		     op->optinsn.insn);
438 }
439 
440 /*
441  * Recover original instructions and breakpoints from relative jumps.
442  * Caller must call with locking kprobe_mutex.
443  */
444 extern void arch_unoptimize_kprobes(struct list_head *oplist,
445 				    struct list_head *done_list)
446 {
447 	struct optimized_kprobe *op, *tmp;
448 
449 	list_for_each_entry_safe(op, tmp, oplist, list) {
450 		arch_unoptimize_kprobe(op);
451 		list_move(&op->list, done_list);
452 	}
453 }
454 
455 int setup_detour_execution(struct kprobe *p, struct pt_regs *regs, int reenter)
456 {
457 	struct optimized_kprobe *op;
458 
459 	if (p->flags & KPROBE_FLAG_OPTIMIZED) {
460 		/* This kprobe is really able to run optimized path. */
461 		op = container_of(p, struct optimized_kprobe, kp);
462 		/* Detour through copied instructions */
463 		regs->ip = (unsigned long)op->optinsn.insn + TMPL_END_IDX;
464 		if (!reenter)
465 			reset_current_kprobe();
466 		preempt_enable_no_resched();
467 		return 1;
468 	}
469 	return 0;
470 }
471 NOKPROBE_SYMBOL(setup_detour_execution);
472