xref: /linux/arch/x86/kernel/step.c (revision 7aacf86b75bc5523d20fd9127104384fce51ce9c)
1 /*
2  * x86 single-step support code, common to 32-bit and 64-bit.
3  */
4 #include <linux/sched.h>
5 #include <linux/sched/task_stack.h>
6 #include <linux/mm.h>
7 #include <linux/ptrace.h>
8 #include <asm/desc.h>
9 #include <asm/mmu_context.h>
10 
11 unsigned long convert_ip_to_linear(struct task_struct *child, struct pt_regs *regs)
12 {
13 	unsigned long addr, seg;
14 
15 	addr = regs->ip;
16 	seg = regs->cs & 0xffff;
17 	if (v8086_mode(regs)) {
18 		addr = (addr & 0xffff) + (seg << 4);
19 		return addr;
20 	}
21 
22 #ifdef CONFIG_MODIFY_LDT_SYSCALL
23 	/*
24 	 * We'll assume that the code segments in the GDT
25 	 * are all zero-based. That is largely true: the
26 	 * TLS segments are used for data, and the PNPBIOS
27 	 * and APM bios ones we just ignore here.
28 	 */
29 	if ((seg & SEGMENT_TI_MASK) == SEGMENT_LDT) {
30 		struct desc_struct *desc;
31 		unsigned long base;
32 
33 		seg >>= 3;
34 
35 		mutex_lock(&child->mm->context.lock);
36 		if (unlikely(!child->mm->context.ldt ||
37 			     seg >= child->mm->context.ldt->nr_entries))
38 			addr = -1L; /* bogus selector, access would fault */
39 		else {
40 			desc = &child->mm->context.ldt->entries[seg];
41 			base = get_desc_base(desc);
42 
43 			/* 16-bit code segment? */
44 			if (!desc->d)
45 				addr &= 0xffff;
46 			addr += base;
47 		}
48 		mutex_unlock(&child->mm->context.lock);
49 	}
50 #endif
51 
52 	return addr;
53 }
54 
55 static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
56 {
57 	int i, copied;
58 	unsigned char opcode[15];
59 	unsigned long addr = convert_ip_to_linear(child, regs);
60 
61 	copied = access_process_vm(child, addr, opcode, sizeof(opcode),
62 			FOLL_FORCE);
63 	for (i = 0; i < copied; i++) {
64 		switch (opcode[i]) {
65 		/* popf and iret */
66 		case 0x9d: case 0xcf:
67 			return 1;
68 
69 			/* CHECKME: 64 65 */
70 
71 		/* opcode and address size prefixes */
72 		case 0x66: case 0x67:
73 			continue;
74 		/* irrelevant prefixes (segment overrides and repeats) */
75 		case 0x26: case 0x2e:
76 		case 0x36: case 0x3e:
77 		case 0x64: case 0x65:
78 		case 0xf0: case 0xf2: case 0xf3:
79 			continue;
80 
81 #ifdef CONFIG_X86_64
82 		case 0x40 ... 0x4f:
83 			if (!user_64bit_mode(regs))
84 				/* 32-bit mode: register increment */
85 				return 0;
86 			/* 64-bit mode: REX prefix */
87 			continue;
88 #endif
89 
90 			/* CHECKME: f2, f3 */
91 
92 		/*
93 		 * pushf: NOTE! We should probably not let
94 		 * the user see the TF bit being set. But
95 		 * it's more pain than it's worth to avoid
96 		 * it, and a debugger could emulate this
97 		 * all in user space if it _really_ cares.
98 		 */
99 		case 0x9c:
100 		default:
101 			return 0;
102 		}
103 	}
104 	return 0;
105 }
106 
107 /*
108  * Enable single-stepping.  Return nonzero if user mode is not using TF itself.
109  */
110 static int enable_single_step(struct task_struct *child)
111 {
112 	struct pt_regs *regs = task_pt_regs(child);
113 	unsigned long oflags;
114 
115 	/*
116 	 * If we stepped into a sysenter/syscall insn, it trapped in
117 	 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
118 	 * If user-mode had set TF itself, then it's still clear from
119 	 * do_debug() and we need to set it again to restore the user
120 	 * state so we don't wrongly set TIF_FORCED_TF below.
121 	 * If enable_single_step() was used last and that is what
122 	 * set TIF_SINGLESTEP, then both TF and TIF_FORCED_TF are
123 	 * already set and our bookkeeping is fine.
124 	 */
125 	if (unlikely(test_tsk_thread_flag(child, TIF_SINGLESTEP)))
126 		regs->flags |= X86_EFLAGS_TF;
127 
128 	/*
129 	 * Always set TIF_SINGLESTEP - this guarantees that
130 	 * we single-step system calls etc..  This will also
131 	 * cause us to set TF when returning to user mode.
132 	 */
133 	set_tsk_thread_flag(child, TIF_SINGLESTEP);
134 
135 	oflags = regs->flags;
136 
137 	/* Set TF on the kernel stack.. */
138 	regs->flags |= X86_EFLAGS_TF;
139 
140 	/*
141 	 * ..but if TF is changed by the instruction we will trace,
142 	 * don't mark it as being "us" that set it, so that we
143 	 * won't clear it by hand later.
144 	 *
145 	 * Note that if we don't actually execute the popf because
146 	 * of a signal arriving right now or suchlike, we will lose
147 	 * track of the fact that it really was "us" that set it.
148 	 */
149 	if (is_setting_trap_flag(child, regs)) {
150 		clear_tsk_thread_flag(child, TIF_FORCED_TF);
151 		return 0;
152 	}
153 
154 	/*
155 	 * If TF was already set, check whether it was us who set it.
156 	 * If not, we should never attempt a block step.
157 	 */
158 	if (oflags & X86_EFLAGS_TF)
159 		return test_tsk_thread_flag(child, TIF_FORCED_TF);
160 
161 	set_tsk_thread_flag(child, TIF_FORCED_TF);
162 
163 	return 1;
164 }
165 
166 void set_task_blockstep(struct task_struct *task, bool on)
167 {
168 	unsigned long debugctl;
169 
170 	/*
171 	 * Ensure irq/preemption can't change debugctl in between.
172 	 * Note also that both TIF_BLOCKSTEP and debugctl should
173 	 * be changed atomically wrt preemption.
174 	 *
175 	 * NOTE: this means that set/clear TIF_BLOCKSTEP is only safe if
176 	 * task is current or it can't be running, otherwise we can race
177 	 * with __switch_to_xtra(). We rely on ptrace_freeze_traced() but
178 	 * PTRACE_KILL is not safe.
179 	 */
180 	local_irq_disable();
181 	debugctl = get_debugctlmsr();
182 	if (on) {
183 		debugctl |= DEBUGCTLMSR_BTF;
184 		set_tsk_thread_flag(task, TIF_BLOCKSTEP);
185 	} else {
186 		debugctl &= ~DEBUGCTLMSR_BTF;
187 		clear_tsk_thread_flag(task, TIF_BLOCKSTEP);
188 	}
189 	if (task == current)
190 		update_debugctlmsr(debugctl);
191 	local_irq_enable();
192 }
193 
194 /*
195  * Enable single or block step.
196  */
197 static void enable_step(struct task_struct *child, bool block)
198 {
199 	/*
200 	 * Make sure block stepping (BTF) is not enabled unless it should be.
201 	 * Note that we don't try to worry about any is_setting_trap_flag()
202 	 * instructions after the first when using block stepping.
203 	 * So no one should try to use debugger block stepping in a program
204 	 * that uses user-mode single stepping itself.
205 	 */
206 	if (enable_single_step(child) && block)
207 		set_task_blockstep(child, true);
208 	else if (test_tsk_thread_flag(child, TIF_BLOCKSTEP))
209 		set_task_blockstep(child, false);
210 }
211 
212 void user_enable_single_step(struct task_struct *child)
213 {
214 	enable_step(child, 0);
215 }
216 
217 void user_enable_block_step(struct task_struct *child)
218 {
219 	enable_step(child, 1);
220 }
221 
222 void user_disable_single_step(struct task_struct *child)
223 {
224 	/*
225 	 * Make sure block stepping (BTF) is disabled.
226 	 */
227 	if (test_tsk_thread_flag(child, TIF_BLOCKSTEP))
228 		set_task_blockstep(child, false);
229 
230 	/* Always clear TIF_SINGLESTEP... */
231 	clear_tsk_thread_flag(child, TIF_SINGLESTEP);
232 
233 	/* But touch TF only if it was set by us.. */
234 	if (test_and_clear_tsk_thread_flag(child, TIF_FORCED_TF))
235 		task_pt_regs(child)->flags &= ~X86_EFLAGS_TF;
236 }
237