Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 2 | /* |
| 3 | * x86 single-step support code, common to 32-bit and 64-bit. |
| 4 | */ |
| 5 | #include <linux/sched.h> |
Ingo Molnar | 68db0cf | 2017-02-08 18:51:37 +0100 | [diff] [blame] | 6 | #include <linux/sched/task_stack.h> |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 7 | #include <linux/mm.h> |
| 8 | #include <linux/ptrace.h> |
Akinobu Mita | 254e0a6 | 2009-07-19 00:08:54 +0900 | [diff] [blame] | 9 | #include <asm/desc.h> |
Andy Lutomirski | 37868fe | 2015-07-30 14:31:32 -0700 | [diff] [blame] | 10 | #include <asm/mmu_context.h> |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 11 | |
Harvey Harrison | 37cd9cf | 2008-01-30 13:33:12 +0100 | [diff] [blame] | 12 | unsigned long convert_ip_to_linear(struct task_struct *child, struct pt_regs *regs) |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 13 | { |
| 14 | unsigned long addr, seg; |
| 15 | |
H. Peter Anvin | 65ea5b0 | 2008-01-30 13:30:56 +0100 | [diff] [blame] | 16 | addr = regs->ip; |
Andy Lutomirski | 9950481 | 2017-07-28 06:00:32 -0700 | [diff] [blame] | 17 | seg = regs->cs; |
H. Peter Anvin | 65ea5b0 | 2008-01-30 13:30:56 +0100 | [diff] [blame] | 18 | if (v8086_mode(regs)) { |
Roland McGrath | 7122ec8 | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 19 | addr = (addr & 0xffff) + (seg << 4); |
| 20 | return addr; |
| 21 | } |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 22 | |
Andy Lutomirski | a5b9e5a | 2015-07-30 14:31:34 -0700 | [diff] [blame] | 23 | #ifdef CONFIG_MODIFY_LDT_SYSCALL |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 24 | /* |
| 25 | * We'll assume that the code segments in the GDT |
| 26 | * are all zero-based. That is largely true: the |
| 27 | * TLS segments are used for data, and the PNPBIOS |
| 28 | * and APM bios ones we just ignore here. |
| 29 | */ |
Roland McGrath | 3f80c1a | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 30 | if ((seg & SEGMENT_TI_MASK) == SEGMENT_LDT) { |
Akinobu Mita | 254e0a6 | 2009-07-19 00:08:54 +0900 | [diff] [blame] | 31 | struct desc_struct *desc; |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 32 | unsigned long base; |
| 33 | |
Juergen Gross | 136d9d8 | 2015-08-06 10:04:38 +0200 | [diff] [blame] | 34 | seg >>= 3; |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 35 | |
| 36 | mutex_lock(&child->mm->context.lock); |
Andy Lutomirski | 37868fe | 2015-07-30 14:31:32 -0700 | [diff] [blame] | 37 | if (unlikely(!child->mm->context.ldt || |
Borislav Petkov | bbf79d2 | 2017-06-06 19:31:16 +0200 | [diff] [blame] | 38 | seg >= child->mm->context.ldt->nr_entries)) |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 39 | addr = -1L; /* bogus selector, access would fault */ |
| 40 | else { |
Andy Lutomirski | 37868fe | 2015-07-30 14:31:32 -0700 | [diff] [blame] | 41 | desc = &child->mm->context.ldt->entries[seg]; |
Akinobu Mita | 254e0a6 | 2009-07-19 00:08:54 +0900 | [diff] [blame] | 42 | base = get_desc_base(desc); |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 43 | |
| 44 | /* 16-bit code segment? */ |
Akinobu Mita | 254e0a6 | 2009-07-19 00:08:54 +0900 | [diff] [blame] | 45 | if (!desc->d) |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 46 | addr &= 0xffff; |
| 47 | addr += base; |
| 48 | } |
| 49 | mutex_unlock(&child->mm->context.lock); |
| 50 | } |
Andy Lutomirski | a5b9e5a | 2015-07-30 14:31:34 -0700 | [diff] [blame] | 51 | #endif |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 52 | |
| 53 | return addr; |
| 54 | } |
| 55 | |
| 56 | static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs) |
| 57 | { |
| 58 | int i, copied; |
| 59 | unsigned char opcode[15]; |
Harvey Harrison | 37cd9cf | 2008-01-30 13:33:12 +0100 | [diff] [blame] | 60 | unsigned long addr = convert_ip_to_linear(child, regs); |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 61 | |
Lorenzo Stoakes | f307ab6 | 2016-10-13 01:20:20 +0100 | [diff] [blame] | 62 | copied = access_process_vm(child, addr, opcode, sizeof(opcode), |
| 63 | FOLL_FORCE); |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 64 | for (i = 0; i < copied; i++) { |
| 65 | switch (opcode[i]) { |
| 66 | /* popf and iret */ |
| 67 | case 0x9d: case 0xcf: |
| 68 | return 1; |
| 69 | |
| 70 | /* CHECKME: 64 65 */ |
| 71 | |
| 72 | /* opcode and address size prefixes */ |
| 73 | case 0x66: case 0x67: |
| 74 | continue; |
| 75 | /* irrelevant prefixes (segment overrides and repeats) */ |
| 76 | case 0x26: case 0x2e: |
| 77 | case 0x36: case 0x3e: |
| 78 | case 0x64: case 0x65: |
Roland McGrath | 5f76cb1 | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 79 | case 0xf0: case 0xf2: case 0xf3: |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 80 | continue; |
| 81 | |
Roland McGrath | 7122ec8 | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 82 | #ifdef CONFIG_X86_64 |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 83 | case 0x40 ... 0x4f: |
Andy Lutomirski | 318f5a2 | 2011-08-03 09:31:53 -0400 | [diff] [blame] | 84 | if (!user_64bit_mode(regs)) |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 85 | /* 32-bit mode: register increment */ |
| 86 | return 0; |
| 87 | /* 64-bit mode: REX prefix */ |
| 88 | continue; |
Roland McGrath | 7122ec8 | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 89 | #endif |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 90 | |
| 91 | /* CHECKME: f2, f3 */ |
| 92 | |
| 93 | /* |
| 94 | * pushf: NOTE! We should probably not let |
| 95 | * the user see the TF bit being set. But |
| 96 | * it's more pain than it's worth to avoid |
| 97 | * it, and a debugger could emulate this |
| 98 | * all in user space if it _really_ cares. |
| 99 | */ |
| 100 | case 0x9c: |
| 101 | default: |
| 102 | return 0; |
| 103 | } |
| 104 | } |
| 105 | return 0; |
| 106 | } |
| 107 | |
Roland McGrath | 10faa81 | 2008-01-30 13:30:54 +0100 | [diff] [blame] | 108 | /* |
| 109 | * Enable single-stepping. Return nonzero if user mode is not using TF itself. |
| 110 | */ |
| 111 | static int enable_single_step(struct task_struct *child) |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 112 | { |
| 113 | struct pt_regs *regs = task_pt_regs(child); |
Roland McGrath | 6718d0d | 2008-07-09 01:07:02 -0700 | [diff] [blame] | 114 | unsigned long oflags; |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 115 | |
| 116 | /* |
Roland McGrath | 380fdd7 | 2008-07-09 02:39:29 -0700 | [diff] [blame] | 117 | * If we stepped into a sysenter/syscall insn, it trapped in |
| 118 | * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP. |
| 119 | * If user-mode had set TF itself, then it's still clear from |
| 120 | * do_debug() and we need to set it again to restore the user |
| 121 | * state so we don't wrongly set TIF_FORCED_TF below. |
| 122 | * If enable_single_step() was used last and that is what |
| 123 | * set TIF_SINGLESTEP, then both TF and TIF_FORCED_TF are |
| 124 | * already set and our bookkeeping is fine. |
| 125 | */ |
| 126 | if (unlikely(test_tsk_thread_flag(child, TIF_SINGLESTEP))) |
| 127 | regs->flags |= X86_EFLAGS_TF; |
| 128 | |
| 129 | /* |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 130 | * Always set TIF_SINGLESTEP - this guarantees that |
| 131 | * we single-step system calls etc.. This will also |
| 132 | * cause us to set TF when returning to user mode. |
| 133 | */ |
| 134 | set_tsk_thread_flag(child, TIF_SINGLESTEP); |
| 135 | |
Roland McGrath | 6718d0d | 2008-07-09 01:07:02 -0700 | [diff] [blame] | 136 | oflags = regs->flags; |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 137 | |
| 138 | /* Set TF on the kernel stack.. */ |
H. Peter Anvin | 65ea5b0 | 2008-01-30 13:30:56 +0100 | [diff] [blame] | 139 | regs->flags |= X86_EFLAGS_TF; |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 140 | |
| 141 | /* |
| 142 | * ..but if TF is changed by the instruction we will trace, |
| 143 | * don't mark it as being "us" that set it, so that we |
| 144 | * won't clear it by hand later. |
Roland McGrath | 6718d0d | 2008-07-09 01:07:02 -0700 | [diff] [blame] | 145 | * |
| 146 | * Note that if we don't actually execute the popf because |
| 147 | * of a signal arriving right now or suchlike, we will lose |
| 148 | * track of the fact that it really was "us" that set it. |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 149 | */ |
Roland McGrath | 6718d0d | 2008-07-09 01:07:02 -0700 | [diff] [blame] | 150 | if (is_setting_trap_flag(child, regs)) { |
| 151 | clear_tsk_thread_flag(child, TIF_FORCED_TF); |
Roland McGrath | 10faa81 | 2008-01-30 13:30:54 +0100 | [diff] [blame] | 152 | return 0; |
Roland McGrath | 6718d0d | 2008-07-09 01:07:02 -0700 | [diff] [blame] | 153 | } |
| 154 | |
| 155 | /* |
| 156 | * If TF was already set, check whether it was us who set it. |
| 157 | * If not, we should never attempt a block step. |
| 158 | */ |
| 159 | if (oflags & X86_EFLAGS_TF) |
| 160 | return test_tsk_thread_flag(child, TIF_FORCED_TF); |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 161 | |
Roland McGrath | e1f2877 | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 162 | set_tsk_thread_flag(child, TIF_FORCED_TF); |
Roland McGrath | 10faa81 | 2008-01-30 13:30:54 +0100 | [diff] [blame] | 163 | |
| 164 | return 1; |
| 165 | } |
| 166 | |
Oleg Nesterov | 9bd1190 | 2012-09-03 15:24:17 +0200 | [diff] [blame] | 167 | void set_task_blockstep(struct task_struct *task, bool on) |
Oleg Nesterov | 848e8f5 | 2012-08-03 17:31:46 +0200 | [diff] [blame] | 168 | { |
| 169 | unsigned long debugctl; |
| 170 | |
Oleg Nesterov | 95cf00f | 2012-08-11 18:06:42 +0200 | [diff] [blame] | 171 | /* |
| 172 | * Ensure irq/preemption can't change debugctl in between. |
| 173 | * Note also that both TIF_BLOCKSTEP and debugctl should |
| 174 | * be changed atomically wrt preemption. |
Oleg Nesterov | 9899d11 | 2013-01-21 20:48:00 +0100 | [diff] [blame] | 175 | * |
| 176 | * NOTE: this means that set/clear TIF_BLOCKSTEP is only safe if |
| 177 | * task is current or it can't be running, otherwise we can race |
| 178 | * with __switch_to_xtra(). We rely on ptrace_freeze_traced() but |
| 179 | * PTRACE_KILL is not safe. |
Oleg Nesterov | 95cf00f | 2012-08-11 18:06:42 +0200 | [diff] [blame] | 180 | */ |
| 181 | local_irq_disable(); |
Oleg Nesterov | 848e8f5 | 2012-08-03 17:31:46 +0200 | [diff] [blame] | 182 | debugctl = get_debugctlmsr(); |
| 183 | if (on) { |
| 184 | debugctl |= DEBUGCTLMSR_BTF; |
| 185 | set_tsk_thread_flag(task, TIF_BLOCKSTEP); |
| 186 | } else { |
| 187 | debugctl &= ~DEBUGCTLMSR_BTF; |
| 188 | clear_tsk_thread_flag(task, TIF_BLOCKSTEP); |
| 189 | } |
Oleg Nesterov | 95cf00f | 2012-08-11 18:06:42 +0200 | [diff] [blame] | 190 | if (task == current) |
| 191 | update_debugctlmsr(debugctl); |
| 192 | local_irq_enable(); |
Oleg Nesterov | 848e8f5 | 2012-08-03 17:31:46 +0200 | [diff] [blame] | 193 | } |
| 194 | |
Roland McGrath | 10faa81 | 2008-01-30 13:30:54 +0100 | [diff] [blame] | 195 | /* |
Roland McGrath | 10faa81 | 2008-01-30 13:30:54 +0100 | [diff] [blame] | 196 | * Enable single or block step. |
| 197 | */ |
| 198 | static void enable_step(struct task_struct *child, bool block) |
| 199 | { |
| 200 | /* |
| 201 | * Make sure block stepping (BTF) is not enabled unless it should be. |
| 202 | * Note that we don't try to worry about any is_setting_trap_flag() |
| 203 | * instructions after the first when using block stepping. |
Lucas De Marchi | 0d2eb44 | 2011-03-17 16:24:16 -0300 | [diff] [blame] | 204 | * So no one should try to use debugger block stepping in a program |
Roland McGrath | 10faa81 | 2008-01-30 13:30:54 +0100 | [diff] [blame] | 205 | * that uses user-mode single stepping itself. |
| 206 | */ |
Oleg Nesterov | 848e8f5 | 2012-08-03 17:31:46 +0200 | [diff] [blame] | 207 | if (enable_single_step(child) && block) |
| 208 | set_task_blockstep(child, true); |
| 209 | else if (test_tsk_thread_flag(child, TIF_BLOCKSTEP)) |
| 210 | set_task_blockstep(child, false); |
Roland McGrath | 10faa81 | 2008-01-30 13:30:54 +0100 | [diff] [blame] | 211 | } |
| 212 | |
| 213 | void user_enable_single_step(struct task_struct *child) |
| 214 | { |
| 215 | enable_step(child, 0); |
| 216 | } |
| 217 | |
| 218 | void user_enable_block_step(struct task_struct *child) |
| 219 | { |
| 220 | enable_step(child, 1); |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 221 | } |
| 222 | |
| 223 | void user_disable_single_step(struct task_struct *child) |
| 224 | { |
Roland McGrath | 10faa81 | 2008-01-30 13:30:54 +0100 | [diff] [blame] | 225 | /* |
| 226 | * Make sure block stepping (BTF) is disabled. |
| 227 | */ |
Oleg Nesterov | 848e8f5 | 2012-08-03 17:31:46 +0200 | [diff] [blame] | 228 | if (test_tsk_thread_flag(child, TIF_BLOCKSTEP)) |
| 229 | set_task_blockstep(child, false); |
Roland McGrath | 10faa81 | 2008-01-30 13:30:54 +0100 | [diff] [blame] | 230 | |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 231 | /* Always clear TIF_SINGLESTEP... */ |
| 232 | clear_tsk_thread_flag(child, TIF_SINGLESTEP); |
| 233 | |
| 234 | /* But touch TF only if it was set by us.. */ |
Roland McGrath | e1f2877 | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 235 | if (test_and_clear_tsk_thread_flag(child, TIF_FORCED_TF)) |
H. Peter Anvin | 65ea5b0 | 2008-01-30 13:30:56 +0100 | [diff] [blame] | 236 | task_pt_regs(child)->flags &= ~X86_EFLAGS_TF; |
Roland McGrath | fa1e03e | 2008-01-30 13:30:50 +0100 | [diff] [blame] | 237 | } |