]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - arch/x86/kernel/kprobes/opt.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Kernel Probes Jump Optimization (Optprobes)
5 * Copyright (C) IBM Corporation, 2002, 2004
6 * Copyright (C) Hitachi Ltd., 2012
8 #include <linux/kprobes.h>
9 #include <linux/ptrace.h>
10 #include <linux/string.h>
11 #include <linux/slab.h>
12 #include <linux/hardirq.h>
13 #include <linux/preempt.h>
14 #include <linux/extable.h>
15 #include <linux/kdebug.h>
16 #include <linux/kallsyms.h>
17 #include <linux/ftrace.h>
18 #include <linux/frame.h>
20 #include <asm/text-patching.h>
21 #include <asm/cacheflush.h>
23 #include <asm/pgtable.h>
24 #include <linux/uaccess.h>
25 #include <asm/alternative.h>
27 #include <asm/debugreg.h>
28 #include <asm/set_memory.h>
29 #include <asm/sections.h>
30 #include <asm/nospec-branch.h>
34 unsigned long __recover_optprobed_insn(kprobe_opcode_t
*buf
, unsigned long addr
)
36 struct optimized_kprobe
*op
;
41 for (i
= 0; i
< RELATIVEJUMP_SIZE
; i
++) {
42 kp
= get_kprobe((void *)addr
- i
);
43 /* This function only handles jump-optimized kprobe */
44 if (kp
&& kprobe_optimized(kp
)) {
45 op
= container_of(kp
, struct optimized_kprobe
, kp
);
46 /* If op->list is not empty, op is under optimizing */
47 if (list_empty(&op
->list
))
55 * If the kprobe can be optimized, original bytes which can be
56 * overwritten by jump destination address. In this case, original
57 * bytes must be recovered from op->optinsn.copied_insn buffer.
59 if (probe_kernel_read(buf
, (void *)addr
,
60 MAX_INSN_SIZE
* sizeof(kprobe_opcode_t
)))
63 if (addr
== (unsigned long)kp
->addr
) {
65 memcpy(buf
+ 1, op
->optinsn
.copied_insn
, RELATIVE_ADDR_SIZE
);
67 offs
= addr
- (unsigned long)kp
->addr
- 1;
68 memcpy(buf
, op
->optinsn
.copied_insn
+ offs
, RELATIVE_ADDR_SIZE
- offs
);
71 return (unsigned long)buf
;
74 /* Insert a move instruction which sets a pointer to eax/rdi (1st arg). */
75 static void synthesize_set_arg1(kprobe_opcode_t
*addr
, unsigned long val
)
83 *(unsigned long *)addr
= val
;
87 ".pushsection .rodata\n"
88 "optprobe_template_func:\n"
89 ".global optprobe_template_entry\n"
90 "optprobe_template_entry:\n"
92 /* We don't bother saving the ss register */
97 ".global optprobe_template_val\n"
98 "optprobe_template_val:\n"
101 ".global optprobe_template_call\n"
102 "optprobe_template_call:\n"
104 /* Move flags to rsp */
105 " movq 18*8(%rsp), %rdx\n"
106 " movq %rdx, 19*8(%rsp)\n"
108 /* Skip flags entry */
111 #else /* CONFIG_X86_32 */
116 ".global optprobe_template_val\n"
117 "optprobe_template_val:\n"
119 ".global optprobe_template_call\n"
120 "optprobe_template_call:\n"
122 /* Move flags into esp */
123 " movl 14*4(%esp), %edx\n"
124 " movl %edx, 15*4(%esp)\n"
126 /* Skip flags entry */
130 ".global optprobe_template_end\n"
131 "optprobe_template_end:\n"
134 void optprobe_template_func(void);
135 STACK_FRAME_NON_STANDARD(optprobe_template_func
);
137 #define TMPL_MOVE_IDX \
138 ((long)optprobe_template_val - (long)optprobe_template_entry)
139 #define TMPL_CALL_IDX \
140 ((long)optprobe_template_call - (long)optprobe_template_entry)
141 #define TMPL_END_IDX \
142 ((long)optprobe_template_end - (long)optprobe_template_entry)
144 #define INT3_SIZE sizeof(kprobe_opcode_t)
146 /* Optimized kprobe call back function: called from optinsn */
148 optimized_callback(struct optimized_kprobe
*op
, struct pt_regs
*regs
)
150 /* This is possible if op is under delayed unoptimizing */
151 if (kprobe_disabled(&op
->kp
))
155 if (kprobe_running()) {
156 kprobes_inc_nmissed_count(&op
->kp
);
158 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
159 /* Save skipped registers */
160 regs
->cs
= __KERNEL_CS
;
162 regs
->cs
|= get_kernel_rpl();
165 regs
->ip
= (unsigned long)op
->kp
.addr
+ INT3_SIZE
;
166 regs
->orig_ax
= ~0UL;
168 __this_cpu_write(current_kprobe
, &op
->kp
);
169 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
170 opt_pre_handler(&op
->kp
, regs
);
171 __this_cpu_write(current_kprobe
, NULL
);
175 NOKPROBE_SYMBOL(optimized_callback
);
177 static int copy_optimized_instructions(u8
*dest
, u8
*src
, u8
*real
)
182 while (len
< RELATIVEJUMP_SIZE
) {
183 ret
= __copy_instruction(dest
+ len
, src
+ len
, real
+ len
, &insn
);
184 if (!ret
|| !can_boost(&insn
, src
+ len
))
188 /* Check whether the address range is reserved */
189 if (ftrace_text_reserved(src
, src
+ len
- 1) ||
190 alternatives_text_reserved(src
, src
+ len
- 1) ||
191 jump_label_text_reserved(src
, src
+ len
- 1))
197 /* Check whether insn is indirect jump */
198 static int __insn_is_indirect_jump(struct insn
*insn
)
200 return ((insn
->opcode
.bytes
[0] == 0xff &&
201 (X86_MODRM_REG(insn
->modrm
.value
) & 6) == 4) || /* Jump */
202 insn
->opcode
.bytes
[0] == 0xea); /* Segment based jump */
205 /* Check whether insn jumps into specified address range */
206 static int insn_jump_into_range(struct insn
*insn
, unsigned long start
, int len
)
208 unsigned long target
= 0;
210 switch (insn
->opcode
.bytes
[0]) {
211 case 0xe0: /* loopne */
212 case 0xe1: /* loope */
213 case 0xe2: /* loop */
214 case 0xe3: /* jcxz */
215 case 0xe9: /* near relative jump */
216 case 0xeb: /* short relative jump */
219 if ((insn
->opcode
.bytes
[1] & 0xf0) == 0x80) /* jcc near */
223 if ((insn
->opcode
.bytes
[0] & 0xf0) == 0x70) /* jcc short */
227 target
= (unsigned long)insn
->next_byte
+ insn
->immediate
.value
;
229 return (start
<= target
&& target
<= start
+ len
);
232 static int insn_is_indirect_jump(struct insn
*insn
)
234 int ret
= __insn_is_indirect_jump(insn
);
236 #ifdef CONFIG_RETPOLINE
238 * Jump to x86_indirect_thunk_* is treated as an indirect jump.
239 * Note that even with CONFIG_RETPOLINE=y, the kernel compiled with
240 * older gcc may use indirect jump. So we add this check instead of
241 * replace indirect-jump check.
244 ret
= insn_jump_into_range(insn
,
245 (unsigned long)__indirect_thunk_start
,
246 (unsigned long)__indirect_thunk_end
-
247 (unsigned long)__indirect_thunk_start
);
252 /* Decode whole function to ensure any instructions don't jump into target */
253 static int can_optimize(unsigned long paddr
)
255 unsigned long addr
, size
= 0, offset
= 0;
257 kprobe_opcode_t buf
[MAX_INSN_SIZE
];
259 /* Lookup symbol including addr */
260 if (!kallsyms_lookup_size_offset(paddr
, &size
, &offset
))
264 * Do not optimize in the entry code due to the unstable
265 * stack handling and registers setup.
267 if (((paddr
>= (unsigned long)__entry_text_start
) &&
268 (paddr
< (unsigned long)__entry_text_end
)) ||
269 ((paddr
>= (unsigned long)__irqentry_text_start
) &&
270 (paddr
< (unsigned long)__irqentry_text_end
)))
273 /* Check there is enough space for a relative jump. */
274 if (size
- offset
< RELATIVEJUMP_SIZE
)
277 /* Decode instructions */
278 addr
= paddr
- offset
;
279 while (addr
< paddr
- offset
+ size
) { /* Decode until function end */
280 unsigned long recovered_insn
;
281 if (search_exception_tables(addr
))
283 * Since some fixup code will jumps into this function,
284 * we can't optimize kprobe in this function.
287 recovered_insn
= recover_probed_instruction(buf
, addr
);
290 kernel_insn_init(&insn
, (void *)recovered_insn
, MAX_INSN_SIZE
);
291 insn_get_length(&insn
);
292 /* Another subsystem puts a breakpoint */
293 if (insn
.opcode
.bytes
[0] == BREAKPOINT_INSTRUCTION
)
295 /* Recover address */
296 insn
.kaddr
= (void *)addr
;
297 insn
.next_byte
= (void *)(addr
+ insn
.length
);
298 /* Check any instructions don't jump into target */
299 if (insn_is_indirect_jump(&insn
) ||
300 insn_jump_into_range(&insn
, paddr
+ INT3_SIZE
,
309 /* Check optimized_kprobe can actually be optimized. */
310 int arch_check_optimized_kprobe(struct optimized_kprobe
*op
)
315 for (i
= 1; i
< op
->optinsn
.size
; i
++) {
316 p
= get_kprobe(op
->kp
.addr
+ i
);
317 if (p
&& !kprobe_disabled(p
))
324 /* Check the addr is within the optimized instructions. */
325 int arch_within_optimized_kprobe(struct optimized_kprobe
*op
,
328 return ((unsigned long)op
->kp
.addr
<= addr
&&
329 (unsigned long)op
->kp
.addr
+ op
->optinsn
.size
> addr
);
332 /* Free optimized instruction slot */
334 void __arch_remove_optimized_kprobe(struct optimized_kprobe
*op
, int dirty
)
336 if (op
->optinsn
.insn
) {
337 free_optinsn_slot(op
->optinsn
.insn
, dirty
);
338 op
->optinsn
.insn
= NULL
;
339 op
->optinsn
.size
= 0;
343 void arch_remove_optimized_kprobe(struct optimized_kprobe
*op
)
345 __arch_remove_optimized_kprobe(op
, 1);
349 * Copy replacing target instructions
350 * Target instructions MUST be relocatable (checked inside)
351 * This is called when new aggr(opt)probe is allocated or reused.
353 int arch_prepare_optimized_kprobe(struct optimized_kprobe
*op
,
354 struct kprobe
*__unused
)
356 u8
*buf
= NULL
, *slot
;
360 if (!can_optimize((unsigned long)op
->kp
.addr
))
363 buf
= kzalloc(MAX_OPTINSN_SIZE
, GFP_KERNEL
);
367 op
->optinsn
.insn
= slot
= get_optinsn_slot();
374 * Verify if the address gap is in 2GB range, because this uses
377 rel
= (long)slot
- (long)op
->kp
.addr
+ RELATIVEJUMP_SIZE
;
378 if (abs(rel
) > 0x7fffffff) {
383 /* Copy arch-dep-instance from template */
384 memcpy(buf
, optprobe_template_entry
, TMPL_END_IDX
);
386 /* Copy instructions into the out-of-line buffer */
387 ret
= copy_optimized_instructions(buf
+ TMPL_END_IDX
, op
->kp
.addr
,
388 slot
+ TMPL_END_IDX
);
391 op
->optinsn
.size
= ret
;
392 len
= TMPL_END_IDX
+ op
->optinsn
.size
;
394 /* Set probe information */
395 synthesize_set_arg1(buf
+ TMPL_MOVE_IDX
, (unsigned long)op
);
397 /* Set probe function call */
398 synthesize_relcall(buf
+ TMPL_CALL_IDX
,
399 slot
+ TMPL_CALL_IDX
, optimized_callback
);
401 /* Set returning jmp instruction at the tail of out-of-line buffer */
402 synthesize_reljump(buf
+ len
, slot
+ len
,
403 (u8
*)op
->kp
.addr
+ op
->optinsn
.size
);
404 len
+= RELATIVEJUMP_SIZE
;
406 /* We have to use text_poke() for instruction buffer because it is RO */
407 text_poke(slot
, buf
, len
);
414 __arch_remove_optimized_kprobe(op
, 0);
419 * Replace breakpoints (int3) with relative jumps.
420 * Caller must call with locking kprobe_mutex and text_mutex.
422 void arch_optimize_kprobes(struct list_head
*oplist
)
424 struct optimized_kprobe
*op
, *tmp
;
425 u8 insn_buff
[RELATIVEJUMP_SIZE
];
427 list_for_each_entry_safe(op
, tmp
, oplist
, list
) {
428 s32 rel
= (s32
)((long)op
->optinsn
.insn
-
429 ((long)op
->kp
.addr
+ RELATIVEJUMP_SIZE
));
431 WARN_ON(kprobe_disabled(&op
->kp
));
433 /* Backup instructions which will be replaced by jump address */
434 memcpy(op
->optinsn
.copied_insn
, op
->kp
.addr
+ INT3_SIZE
,
437 insn_buff
[0] = RELATIVEJUMP_OPCODE
;
438 *(s32
*)(&insn_buff
[1]) = rel
;
440 text_poke_bp(op
->kp
.addr
, insn_buff
, RELATIVEJUMP_SIZE
, NULL
);
442 list_del_init(&op
->list
);
446 /* Replace a relative jump with a breakpoint (int3). */
447 void arch_unoptimize_kprobe(struct optimized_kprobe
*op
)
449 u8 insn_buff
[RELATIVEJUMP_SIZE
];
450 u8 emulate_buff
[RELATIVEJUMP_SIZE
];
452 /* Set int3 to first byte for kprobes */
453 insn_buff
[0] = BREAKPOINT_INSTRUCTION
;
454 memcpy(insn_buff
+ 1, op
->optinsn
.copied_insn
, RELATIVE_ADDR_SIZE
);
456 emulate_buff
[0] = RELATIVEJUMP_OPCODE
;
457 *(s32
*)(&emulate_buff
[1]) = (s32
)((long)op
->optinsn
.insn
-
458 ((long)op
->kp
.addr
+ RELATIVEJUMP_SIZE
));
460 text_poke_bp(op
->kp
.addr
, insn_buff
, RELATIVEJUMP_SIZE
,
465 * Recover original instructions and breakpoints from relative jumps.
466 * Caller must call with locking kprobe_mutex.
468 extern void arch_unoptimize_kprobes(struct list_head
*oplist
,
469 struct list_head
*done_list
)
471 struct optimized_kprobe
*op
, *tmp
;
473 list_for_each_entry_safe(op
, tmp
, oplist
, list
) {
474 arch_unoptimize_kprobe(op
);
475 list_move(&op
->list
, done_list
);
479 int setup_detour_execution(struct kprobe
*p
, struct pt_regs
*regs
, int reenter
)
481 struct optimized_kprobe
*op
;
483 if (p
->flags
& KPROBE_FLAG_OPTIMIZED
) {
484 /* This kprobe is really able to run optimized path. */
485 op
= container_of(p
, struct optimized_kprobe
, kp
);
486 /* Detour through copied instructions */
487 regs
->ip
= (unsigned long)op
->optinsn
.insn
+ TMPL_END_IDX
;
489 reset_current_kprobe();
494 NOKPROBE_SYMBOL(setup_detour_execution
);