• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
32 #include <asm/text-patching.h>
33 #include <asm/cacheflush.h>
34 #include <asm/desc.h>
35 #include <asm/pgtable.h>
36 #include <asm/uaccess.h>
37 #include <asm/alternative.h>
38 #include <asm/insn.h>
39 #include <asm/debugreg.h>
40 #include <asm/nospec-branch.h>
41 
42 #include "common.h"
43 
__recover_optprobed_insn(kprobe_opcode_t * buf,unsigned long addr)44 unsigned long __recover_optprobed_insn(kprobe_opcode_t *buf, unsigned long addr)
45 {
46 	struct optimized_kprobe *op;
47 	struct kprobe *kp;
48 	long offs;
49 	int i;
50 
51 	for (i = 0; i < RELATIVEJUMP_SIZE; i++) {
52 		kp = get_kprobe((void *)addr - i);
53 		/* This function only handles jump-optimized kprobe */
54 		if (kp && kprobe_optimized(kp)) {
55 			op = container_of(kp, struct optimized_kprobe, kp);
56 			/* If op->list is not empty, op is under optimizing */
57 			if (list_empty(&op->list))
58 				goto found;
59 		}
60 	}
61 
62 	return addr;
63 found:
64 	/*
65 	 * If the kprobe can be optimized, original bytes which can be
66 	 * overwritten by jump destination address. In this case, original
67 	 * bytes must be recovered from op->optinsn.copied_insn buffer.
68 	 */
69 	memcpy(buf, (void *)addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
70 	if (addr == (unsigned long)kp->addr) {
71 		buf[0] = kp->opcode;
72 		memcpy(buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
73 	} else {
74 		offs = addr - (unsigned long)kp->addr - 1;
75 		memcpy(buf, op->optinsn.copied_insn + offs, RELATIVE_ADDR_SIZE - offs);
76 	}
77 
78 	return (unsigned long)buf;
79 }
80 
81 /* Insert a move instruction which sets a pointer to eax/rdi (1st arg). */
synthesize_set_arg1(kprobe_opcode_t * addr,unsigned long val)82 static void synthesize_set_arg1(kprobe_opcode_t *addr, unsigned long val)
83 {
84 #ifdef CONFIG_X86_64
85 	*addr++ = 0x48;
86 	*addr++ = 0xbf;
87 #else
88 	*addr++ = 0xb8;
89 #endif
90 	*(unsigned long *)addr = val;
91 }
92 
93 asm (
94 			".global optprobe_template_entry\n"
95 			"optprobe_template_entry:\n"
96 #ifdef CONFIG_X86_64
97 			/* We don't bother saving the ss register */
98 			"	pushq %rsp\n"
99 			"	pushfq\n"
100 			SAVE_REGS_STRING
101 			"	movq %rsp, %rsi\n"
102 			".global optprobe_template_val\n"
103 			"optprobe_template_val:\n"
104 			ASM_NOP5
105 			ASM_NOP5
106 			".global optprobe_template_call\n"
107 			"optprobe_template_call:\n"
108 			ASM_NOP5
109 			/* Move flags to rsp */
110 			"	movq 144(%rsp), %rdx\n"
111 			"	movq %rdx, 152(%rsp)\n"
112 			RESTORE_REGS_STRING
113 			/* Skip flags entry */
114 			"	addq $8, %rsp\n"
115 			"	popfq\n"
116 #else /* CONFIG_X86_32 */
117 			"	pushf\n"
118 			SAVE_REGS_STRING
119 			"	movl %esp, %edx\n"
120 			".global optprobe_template_val\n"
121 			"optprobe_template_val:\n"
122 			ASM_NOP5
123 			".global optprobe_template_call\n"
124 			"optprobe_template_call:\n"
125 			ASM_NOP5
126 			RESTORE_REGS_STRING
127 			"	addl $4, %esp\n"	/* skip cs */
128 			"	popf\n"
129 #endif
130 			".global optprobe_template_end\n"
131 			"optprobe_template_end:\n");
132 
133 #define TMPL_MOVE_IDX \
134 	((long)&optprobe_template_val - (long)&optprobe_template_entry)
135 #define TMPL_CALL_IDX \
136 	((long)&optprobe_template_call - (long)&optprobe_template_entry)
137 #define TMPL_END_IDX \
138 	((long)&optprobe_template_end - (long)&optprobe_template_entry)
139 
140 #define INT3_SIZE sizeof(kprobe_opcode_t)
141 
142 /* Optimized kprobe call back function: called from optinsn */
143 static void
optimized_callback(struct optimized_kprobe * op,struct pt_regs * regs)144 optimized_callback(struct optimized_kprobe *op, struct pt_regs *regs)
145 {
146 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
147 	unsigned long flags;
148 
149 	/* This is possible if op is under delayed unoptimizing */
150 	if (kprobe_disabled(&op->kp))
151 		return;
152 
153 	local_irq_save(flags);
154 	if (kprobe_running()) {
155 		kprobes_inc_nmissed_count(&op->kp);
156 	} else {
157 		/* Save skipped registers */
158 #ifdef CONFIG_X86_64
159 		regs->cs = __KERNEL_CS;
160 #else
161 		regs->cs = __KERNEL_CS | get_kernel_rpl();
162 		regs->gs = 0;
163 #endif
164 		regs->ip = (unsigned long)op->kp.addr + INT3_SIZE;
165 		regs->orig_ax = ~0UL;
166 
167 		__this_cpu_write(current_kprobe, &op->kp);
168 		kcb->kprobe_status = KPROBE_HIT_ACTIVE;
169 		opt_pre_handler(&op->kp, regs);
170 		__this_cpu_write(current_kprobe, NULL);
171 	}
172 	local_irq_restore(flags);
173 }
174 NOKPROBE_SYMBOL(optimized_callback);
175 
copy_optimized_instructions(u8 * dest,u8 * src)176 static int copy_optimized_instructions(u8 *dest, u8 *src)
177 {
178 	int len = 0, ret;
179 
180 	while (len < RELATIVEJUMP_SIZE) {
181 		ret = __copy_instruction(dest + len, src + len);
182 		if (!ret || !can_boost(dest + len, src + len))
183 			return -EINVAL;
184 		len += ret;
185 	}
186 	/* Check whether the address range is reserved */
187 	if (ftrace_text_reserved(src, src + len - 1) ||
188 	    alternatives_text_reserved(src, src + len - 1) ||
189 	    jump_label_text_reserved(src, src + len - 1))
190 		return -EBUSY;
191 
192 	return len;
193 }
194 
195 /* Check whether insn is indirect jump */
__insn_is_indirect_jump(struct insn * insn)196 static int __insn_is_indirect_jump(struct insn *insn)
197 {
198 	return ((insn->opcode.bytes[0] == 0xff &&
199 		(X86_MODRM_REG(insn->modrm.value) & 6) == 4) || /* Jump */
200 		insn->opcode.bytes[0] == 0xea);	/* Segment based jump */
201 }
202 
203 /* Check whether insn jumps into specified address range */
insn_jump_into_range(struct insn * insn,unsigned long start,int len)204 static int insn_jump_into_range(struct insn *insn, unsigned long start, int len)
205 {
206 	unsigned long target = 0;
207 
208 	switch (insn->opcode.bytes[0]) {
209 	case 0xe0:	/* loopne */
210 	case 0xe1:	/* loope */
211 	case 0xe2:	/* loop */
212 	case 0xe3:	/* jcxz */
213 	case 0xe9:	/* near relative jump */
214 	case 0xeb:	/* short relative jump */
215 		break;
216 	case 0x0f:
217 		if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */
218 			break;
219 		return 0;
220 	default:
221 		if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
222 			break;
223 		return 0;
224 	}
225 	target = (unsigned long)insn->next_byte + insn->immediate.value;
226 
227 	return (start <= target && target <= start + len);
228 }
229 
insn_is_indirect_jump(struct insn * insn)230 static int insn_is_indirect_jump(struct insn *insn)
231 {
232 	int ret = __insn_is_indirect_jump(insn);
233 
234 #ifdef CONFIG_RETPOLINE
235 	/*
236 	 * Jump to x86_indirect_thunk_* is treated as an indirect jump.
237 	 * Note that even with CONFIG_RETPOLINE=y, the kernel compiled with
238 	 * older gcc may use indirect jump. So we add this check instead of
239 	 * replace indirect-jump check.
240 	 */
241 	if (!ret)
242 		ret = insn_jump_into_range(insn,
243 				(unsigned long)__indirect_thunk_start,
244 				(unsigned long)__indirect_thunk_end -
245 				(unsigned long)__indirect_thunk_start);
246 #endif
247 	return ret;
248 }
249 
250 /* Decode whole function to ensure any instructions don't jump into target */
can_optimize(unsigned long paddr)251 static int can_optimize(unsigned long paddr)
252 {
253 	unsigned long addr, size = 0, offset = 0;
254 	struct insn insn;
255 	kprobe_opcode_t buf[MAX_INSN_SIZE];
256 
257 	/* Lookup symbol including addr */
258 	if (!kallsyms_lookup_size_offset(paddr, &size, &offset))
259 		return 0;
260 
261 	/*
262 	 * Do not optimize in the entry code due to the unstable
263 	 * stack handling.
264 	 */
265 	if ((paddr >= (unsigned long)__entry_text_start) &&
266 	    (paddr <  (unsigned long)__entry_text_end))
267 		return 0;
268 
269 	/* Check there is enough space for a relative jump. */
270 	if (size - offset < RELATIVEJUMP_SIZE)
271 		return 0;
272 
273 	/* Decode instructions */
274 	addr = paddr - offset;
275 	while (addr < paddr - offset + size) { /* Decode until function end */
276 		unsigned long recovered_insn;
277 		if (search_exception_tables(addr))
278 			/*
279 			 * Since some fixup code will jumps into this function,
280 			 * we can't optimize kprobe in this function.
281 			 */
282 			return 0;
283 		recovered_insn = recover_probed_instruction(buf, addr);
284 		if (!recovered_insn)
285 			return 0;
286 		kernel_insn_init(&insn, (void *)recovered_insn, MAX_INSN_SIZE);
287 		insn_get_length(&insn);
288 		/* Another subsystem puts a breakpoint */
289 		if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
290 			return 0;
291 		/* Recover address */
292 		insn.kaddr = (void *)addr;
293 		insn.next_byte = (void *)(addr + insn.length);
294 		/* Check any instructions don't jump into target */
295 		if (insn_is_indirect_jump(&insn) ||
296 		    insn_jump_into_range(&insn, paddr + INT3_SIZE,
297 					 RELATIVE_ADDR_SIZE))
298 			return 0;
299 		addr += insn.length;
300 	}
301 
302 	return 1;
303 }
304 
305 /* Check optimized_kprobe can actually be optimized. */
arch_check_optimized_kprobe(struct optimized_kprobe * op)306 int arch_check_optimized_kprobe(struct optimized_kprobe *op)
307 {
308 	int i;
309 	struct kprobe *p;
310 
311 	for (i = 1; i < op->optinsn.size; i++) {
312 		p = get_kprobe(op->kp.addr + i);
313 		if (p && !kprobe_disabled(p))
314 			return -EEXIST;
315 	}
316 
317 	return 0;
318 }
319 
320 /* Check the addr is within the optimized instructions. */
arch_within_optimized_kprobe(struct optimized_kprobe * op,unsigned long addr)321 int arch_within_optimized_kprobe(struct optimized_kprobe *op,
322 				 unsigned long addr)
323 {
324 	return ((unsigned long)op->kp.addr <= addr &&
325 		(unsigned long)op->kp.addr + op->optinsn.size > addr);
326 }
327 
328 /* Free optimized instruction slot */
329 static
__arch_remove_optimized_kprobe(struct optimized_kprobe * op,int dirty)330 void __arch_remove_optimized_kprobe(struct optimized_kprobe *op, int dirty)
331 {
332 	if (op->optinsn.insn) {
333 		free_optinsn_slot(op->optinsn.insn, dirty);
334 		op->optinsn.insn = NULL;
335 		op->optinsn.size = 0;
336 	}
337 }
338 
arch_remove_optimized_kprobe(struct optimized_kprobe * op)339 void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
340 {
341 	__arch_remove_optimized_kprobe(op, 1);
342 }
343 
344 /*
345  * Copy replacing target instructions
346  * Target instructions MUST be relocatable (checked inside)
347  * This is called when new aggr(opt)probe is allocated or reused.
348  */
arch_prepare_optimized_kprobe(struct optimized_kprobe * op,struct kprobe * __unused)349 int arch_prepare_optimized_kprobe(struct optimized_kprobe *op,
350 				  struct kprobe *__unused)
351 {
352 	u8 *buf;
353 	int ret;
354 	long rel;
355 
356 	if (!can_optimize((unsigned long)op->kp.addr))
357 		return -EILSEQ;
358 
359 	op->optinsn.insn = get_optinsn_slot();
360 	if (!op->optinsn.insn)
361 		return -ENOMEM;
362 
363 	/*
364 	 * Verify if the address gap is in 2GB range, because this uses
365 	 * a relative jump.
366 	 */
367 	rel = (long)op->optinsn.insn - (long)op->kp.addr + RELATIVEJUMP_SIZE;
368 	if (abs(rel) > 0x7fffffff) {
369 		__arch_remove_optimized_kprobe(op, 0);
370 		return -ERANGE;
371 	}
372 
373 	buf = (u8 *)op->optinsn.insn;
374 	set_memory_rw((unsigned long)buf & PAGE_MASK, 1);
375 
376 	/* Copy instructions into the out-of-line buffer */
377 	ret = copy_optimized_instructions(buf + TMPL_END_IDX, op->kp.addr);
378 	if (ret < 0) {
379 		__arch_remove_optimized_kprobe(op, 0);
380 		return ret;
381 	}
382 	op->optinsn.size = ret;
383 
384 	/* Copy arch-dep-instance from template */
385 	memcpy(buf, &optprobe_template_entry, TMPL_END_IDX);
386 
387 	/* Set probe information */
388 	synthesize_set_arg1(buf + TMPL_MOVE_IDX, (unsigned long)op);
389 
390 	/* Set probe function call */
391 	synthesize_relcall(buf + TMPL_CALL_IDX, optimized_callback);
392 
393 	/* Set returning jmp instruction at the tail of out-of-line buffer */
394 	synthesize_reljump(buf + TMPL_END_IDX + op->optinsn.size,
395 			   (u8 *)op->kp.addr + op->optinsn.size);
396 
397 	set_memory_ro((unsigned long)buf & PAGE_MASK, 1);
398 
399 	flush_icache_range((unsigned long) buf,
400 			   (unsigned long) buf + TMPL_END_IDX +
401 			   op->optinsn.size + RELATIVEJUMP_SIZE);
402 	return 0;
403 }
404 
405 /*
406  * Replace breakpoints (int3) with relative jumps.
407  * Caller must call with locking kprobe_mutex and text_mutex.
408  */
arch_optimize_kprobes(struct list_head * oplist)409 void arch_optimize_kprobes(struct list_head *oplist)
410 {
411 	struct optimized_kprobe *op, *tmp;
412 	u8 insn_buf[RELATIVEJUMP_SIZE];
413 
414 	list_for_each_entry_safe(op, tmp, oplist, list) {
415 		s32 rel = (s32)((long)op->optinsn.insn -
416 			((long)op->kp.addr + RELATIVEJUMP_SIZE));
417 
418 		WARN_ON(kprobe_disabled(&op->kp));
419 
420 		/* Backup instructions which will be replaced by jump address */
421 		memcpy(op->optinsn.copied_insn, op->kp.addr + INT3_SIZE,
422 		       RELATIVE_ADDR_SIZE);
423 
424 		insn_buf[0] = RELATIVEJUMP_OPCODE;
425 		*(s32 *)(&insn_buf[1]) = rel;
426 
427 		text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
428 			     op->optinsn.insn);
429 
430 		list_del_init(&op->list);
431 	}
432 }
433 
434 /* Replace a relative jump with a breakpoint (int3).  */
arch_unoptimize_kprobe(struct optimized_kprobe * op)435 void arch_unoptimize_kprobe(struct optimized_kprobe *op)
436 {
437 	u8 insn_buf[RELATIVEJUMP_SIZE];
438 
439 	/* Set int3 to first byte for kprobes */
440 	insn_buf[0] = BREAKPOINT_INSTRUCTION;
441 	memcpy(insn_buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
442 	text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
443 		     op->optinsn.insn);
444 }
445 
446 /*
447  * Recover original instructions and breakpoints from relative jumps.
448  * Caller must call with locking kprobe_mutex.
449  */
arch_unoptimize_kprobes(struct list_head * oplist,struct list_head * done_list)450 extern void arch_unoptimize_kprobes(struct list_head *oplist,
451 				    struct list_head *done_list)
452 {
453 	struct optimized_kprobe *op, *tmp;
454 
455 	list_for_each_entry_safe(op, tmp, oplist, list) {
456 		arch_unoptimize_kprobe(op);
457 		list_move(&op->list, done_list);
458 	}
459 }
460 
setup_detour_execution(struct kprobe * p,struct pt_regs * regs,int reenter)461 int setup_detour_execution(struct kprobe *p, struct pt_regs *regs, int reenter)
462 {
463 	struct optimized_kprobe *op;
464 
465 	if (p->flags & KPROBE_FLAG_OPTIMIZED) {
466 		/* This kprobe is really able to run optimized path. */
467 		op = container_of(p, struct optimized_kprobe, kp);
468 		/* Detour through copied instructions */
469 		regs->ip = (unsigned long)op->optinsn.insn + TMPL_END_IDX;
470 		if (!reenter)
471 			reset_current_kprobe();
472 		preempt_enable_no_resched();
473 		return 1;
474 	}
475 	return 0;
476 }
477 NOKPROBE_SYMBOL(setup_detour_execution);
478