| 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | #ifndef _ASM_X86_TEXT_PATCHING_H |
| 3 | #define _ASM_X86_TEXT_PATCHING_H |
| 4 | |
| 5 | #include <linux/types.h> |
| 6 | #include <linux/stddef.h> |
| 7 | #include <asm/ptrace.h> |
| 8 | |
| 9 | /* |
| 10 | * Currently, the max observed size in the kernel code is |
| 11 | * JUMP_LABEL_NOP_SIZE/RELATIVEJUMP_SIZE, which are 5. |
| 12 | * Raise it if needed. |
| 13 | */ |
| 14 | #define TEXT_POKE_MAX_OPCODE_SIZE 5 |
| 15 | |
| 16 | extern void text_poke_early(void *addr, const void *opcode, size_t len); |
| 17 | |
| 18 | extern void text_poke_apply_relocation(u8 *buf, const u8 * const instr, size_t instrlen, u8 *repl, size_t repl_len); |
| 19 | |
| 20 | /* |
| 21 | * Clear and restore the kernel write-protection flag on the local CPU. |
| 22 | * Allows the kernel to edit read-only pages. |
| 23 | * Side-effect: any interrupt handler running between save and restore will have |
| 24 | * the ability to write to read-only pages. |
| 25 | * |
| 26 | * Warning: |
| 27 | * Code patching in the UP case is safe if NMIs and MCE handlers are stopped and |
| 28 | * no thread can be preempted in the instructions being modified (no iret to an |
| 29 | * invalid instruction possible) or if the instructions are changed from a |
| 30 | * consistent state to another consistent state atomically. |
| 31 | * On the local CPU you need to be protected against NMI or MCE handlers seeing |
| 32 | * an inconsistent instruction while you patch. |
| 33 | */ |
| 34 | extern void *text_poke(void *addr, const void *opcode, size_t len); |
| 35 | extern void smp_text_poke_sync_each_cpu(void); |
| 36 | extern void *text_poke_kgdb(void *addr, const void *opcode, size_t len); |
| 37 | extern void *text_poke_copy(void *addr, const void *opcode, size_t len); |
| 38 | #define text_poke_copy text_poke_copy |
| 39 | extern void *text_poke_copy_locked(void *addr, const void *opcode, size_t len, bool core_ok); |
| 40 | extern void *text_poke_set(void *addr, int c, size_t len); |
| 41 | extern int smp_text_poke_int3_handler(struct pt_regs *regs); |
| 42 | extern void smp_text_poke_single(void *addr, const void *opcode, size_t len, const void *emulate); |
| 43 | |
| 44 | extern void smp_text_poke_batch_add(void *addr, const void *opcode, size_t len, const void *emulate); |
| 45 | extern void smp_text_poke_batch_finish(void); |
| 46 | |
| 47 | #define INT3_INSN_SIZE 1 |
| 48 | #define INT3_INSN_OPCODE 0xCC |
| 49 | |
| 50 | #define RET_INSN_SIZE 1 |
| 51 | #define RET_INSN_OPCODE 0xC3 |
| 52 | |
| 53 | #define CALL_INSN_SIZE 5 |
| 54 | #define CALL_INSN_OPCODE 0xE8 |
| 55 | |
| 56 | #define JMP32_INSN_SIZE 5 |
| 57 | #define JMP32_INSN_OPCODE 0xE9 |
| 58 | |
| 59 | #define JMP8_INSN_SIZE 2 |
| 60 | #define JMP8_INSN_OPCODE 0xEB |
| 61 | |
| 62 | #define DISP32_SIZE 4 |
| 63 | |
| 64 | static __always_inline int text_opcode_size(u8 opcode) |
| 65 | { |
| 66 | int size = 0; |
| 67 | |
| 68 | #define __CASE(insn) \ |
| 69 | case insn##_INSN_OPCODE: size = insn##_INSN_SIZE; break |
| 70 | |
| 71 | switch(opcode) { |
| 72 | __CASE(INT3); |
| 73 | __CASE(RET); |
| 74 | __CASE(CALL); |
| 75 | __CASE(JMP32); |
| 76 | __CASE(JMP8); |
| 77 | } |
| 78 | |
| 79 | #undef __CASE |
| 80 | |
| 81 | return size; |
| 82 | } |
| 83 | |
| 84 | union text_poke_insn { |
| 85 | u8 text[TEXT_POKE_MAX_OPCODE_SIZE]; |
| 86 | struct { |
| 87 | u8 opcode; |
| 88 | s32 disp; |
| 89 | } __attribute__((packed)); |
| 90 | }; |
| 91 | |
| 92 | static __always_inline |
| 93 | void __text_gen_insn(void *buf, u8 opcode, const void *addr, const void *dest, int size) |
| 94 | { |
| 95 | union text_poke_insn *insn = buf; |
| 96 | |
| 97 | BUG_ON(size < text_opcode_size(opcode)); |
| 98 | |
| 99 | /* |
| 100 | * Hide the addresses to avoid the compiler folding in constants when |
| 101 | * referencing code, these can mess up annotations like |
| 102 | * ANNOTATE_NOENDBR. |
| 103 | */ |
| 104 | OPTIMIZER_HIDE_VAR(insn); |
| 105 | OPTIMIZER_HIDE_VAR(addr); |
| 106 | OPTIMIZER_HIDE_VAR(dest); |
| 107 | |
| 108 | insn->opcode = opcode; |
| 109 | |
| 110 | if (size > 1) { |
| 111 | insn->disp = (long)dest - (long)(addr + size); |
| 112 | if (size == 2) { |
| 113 | /* |
| 114 | * Ensure that for JMP8 the displacement |
| 115 | * actually fits the signed byte. |
| 116 | */ |
| 117 | BUG_ON((insn->disp >> 31) != (insn->disp >> 7)); |
| 118 | } |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | static __always_inline |
| 123 | void *text_gen_insn(u8 opcode, const void *addr, const void *dest) |
| 124 | { |
| 125 | static union text_poke_insn insn; /* per instance */ |
| 126 | __text_gen_insn(buf: &insn, opcode, addr, dest, size: text_opcode_size(opcode)); |
| 127 | return &insn.text; |
| 128 | } |
| 129 | |
| 130 | extern int after_bootmem; |
| 131 | extern __ro_after_init struct mm_struct *text_poke_mm; |
| 132 | extern __ro_after_init unsigned long text_poke_mm_addr; |
| 133 | |
| 134 | #ifndef CONFIG_UML_X86 |
| 135 | static __always_inline |
| 136 | void int3_emulate_jmp(struct pt_regs *regs, unsigned long ip) |
| 137 | { |
| 138 | regs->ip = ip; |
| 139 | } |
| 140 | |
| 141 | static __always_inline |
| 142 | void int3_emulate_push(struct pt_regs *regs, unsigned long val) |
| 143 | { |
| 144 | /* |
| 145 | * The INT3 handler in entry_64.S adds a gap between the |
| 146 | * stack where the break point happened, and the saving of |
| 147 | * pt_regs. We can extend the original stack because of |
| 148 | * this gap. See the idtentry macro's X86_TRAP_BP logic. |
| 149 | * |
| 150 | * Similarly, entry_32.S will have a gap on the stack for |
| 151 | * (any) hardware exception and pt_regs; see the |
| 152 | * FIXUP_FRAME macro. |
| 153 | */ |
| 154 | regs->sp -= sizeof(unsigned long); |
| 155 | *(unsigned long *)regs->sp = val; |
| 156 | } |
| 157 | |
| 158 | static __always_inline |
| 159 | unsigned long int3_emulate_pop(struct pt_regs *regs) |
| 160 | { |
| 161 | unsigned long val = *(unsigned long *)regs->sp; |
| 162 | regs->sp += sizeof(unsigned long); |
| 163 | return val; |
| 164 | } |
| 165 | |
| 166 | static __always_inline |
| 167 | void int3_emulate_call(struct pt_regs *regs, unsigned long func) |
| 168 | { |
| 169 | int3_emulate_push(regs, val: regs->ip - INT3_INSN_SIZE + CALL_INSN_SIZE); |
| 170 | int3_emulate_jmp(regs, ip: func); |
| 171 | } |
| 172 | |
| 173 | static __always_inline |
| 174 | void int3_emulate_ret(struct pt_regs *regs) |
| 175 | { |
| 176 | unsigned long ip = int3_emulate_pop(regs); |
| 177 | int3_emulate_jmp(regs, ip); |
| 178 | } |
| 179 | |
| 180 | static __always_inline |
| 181 | bool __emulate_cc(unsigned long flags, u8 cc) |
| 182 | { |
| 183 | static const unsigned long cc_mask[6] = { |
| 184 | [0] = X86_EFLAGS_OF, |
| 185 | [1] = X86_EFLAGS_CF, |
| 186 | [2] = X86_EFLAGS_ZF, |
| 187 | [3] = X86_EFLAGS_CF | X86_EFLAGS_ZF, |
| 188 | [4] = X86_EFLAGS_SF, |
| 189 | [5] = X86_EFLAGS_PF, |
| 190 | }; |
| 191 | |
| 192 | bool invert = cc & 1; |
| 193 | bool match; |
| 194 | |
| 195 | if (cc < 0xc) { |
| 196 | match = flags & cc_mask[cc >> 1]; |
| 197 | } else { |
| 198 | match = ((flags & X86_EFLAGS_SF) >> X86_EFLAGS_SF_BIT) ^ |
| 199 | ((flags & X86_EFLAGS_OF) >> X86_EFLAGS_OF_BIT); |
| 200 | if (cc >= 0xe) |
| 201 | match = match || (flags & X86_EFLAGS_ZF); |
| 202 | } |
| 203 | |
| 204 | return (match && !invert) || (!match && invert); |
| 205 | } |
| 206 | |
| 207 | static __always_inline |
| 208 | void int3_emulate_jcc(struct pt_regs *regs, u8 cc, unsigned long ip, unsigned long disp) |
| 209 | { |
| 210 | if (__emulate_cc(flags: regs->flags, cc)) |
| 211 | ip += disp; |
| 212 | |
| 213 | int3_emulate_jmp(regs, ip); |
| 214 | } |
| 215 | |
| 216 | #endif /* !CONFIG_UML_X86 */ |
| 217 | |
| 218 | #endif /* _ASM_X86_TEXT_PATCHING_H */ |
| 219 | |