1/*
2 * Copyright (c) 2012-2014 Andy Lutomirski <luto@amacapital.net>
3 *
4 * Based on the original implementation which is:
5 *  Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
6 *  Copyright 2003 Andi Kleen, SuSE Labs.
7 *
8 *  Parts of the original code have been moved to arch/x86/vdso/vma.c
9 *
10 * This file implements vsyscall emulation.  vsyscalls are a legacy ABI:
11 * Userspace can request certain kernel services by calling fixed
12 * addresses.  This concept is problematic:
13 *
14 * - It interferes with ASLR.
15 * - It's awkward to write code that lives in kernel addresses but is
16 *   callable by userspace at fixed addresses.
17 * - The whole concept is impossible for 32-bit compat userspace.
18 * - UML cannot easily virtualize a vsyscall.
19 *
20 * As of mid-2014, I believe that there is no new userspace code that
21 * will use a vsyscall if the vDSO is present.  I hope that there will
22 * soon be no new userspace code that will ever use a vsyscall.
23 *
24 * The code in this file emulates vsyscalls when notified of a page
25 * fault to a vsyscall address.
26 */
27
28#include <linux/kernel.h>
29#include <linux/timer.h>
30#include <linux/syscalls.h>
31#include <linux/ratelimit.h>
32
33#include <asm/vsyscall.h>
34#include <asm/unistd.h>
35#include <asm/fixmap.h>
36#include <asm/traps.h>
37
38#define CREATE_TRACE_POINTS
39#include "vsyscall_trace.h"
40
41static enum { EMULATE, NATIVE, NONE } vsyscall_mode = EMULATE;
42
43static int __init vsyscall_setup(char *str)
44{
45	if (str) {
46		if (!strcmp("emulate", str))
47			vsyscall_mode = EMULATE;
48		else if (!strcmp("native", str))
49			vsyscall_mode = NATIVE;
50		else if (!strcmp("none", str))
51			vsyscall_mode = NONE;
52		else
53			return -EINVAL;
54
55		return 0;
56	}
57
58	return -EINVAL;
59}
60early_param("vsyscall", vsyscall_setup);
61
62static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
63			      const char *message)
64{
65	if (!show_unhandled_signals)
66		return;
67
68	printk_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
69			   level, current->comm, task_pid_nr(current),
70			   message, regs->ip, regs->cs,
71			   regs->sp, regs->ax, regs->si, regs->di);
72}
73
74static int addr_to_vsyscall_nr(unsigned long addr)
75{
76	int nr;
77
78	if ((addr & ~0xC00UL) != VSYSCALL_ADDR)
79		return -EINVAL;
80
81	nr = (addr & 0xC00UL) >> 10;
82	if (nr >= 3)
83		return -EINVAL;
84
85	return nr;
86}
87
88static bool write_ok_or_segv(unsigned long ptr, size_t size)
89{
90	/*
91	 * XXX: if access_ok, get_user, and put_user handled
92	 * sig_on_uaccess_error, this could go away.
93	 */
94
95	if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
96		siginfo_t info;
97		struct thread_struct *thread = &current->thread;
98
99		thread->error_code	= 6;  /* user fault, no page, write */
100		thread->cr2		= ptr;
101		thread->trap_nr		= X86_TRAP_PF;
102
103		memset(&info, 0, sizeof(info));
104		info.si_signo		= SIGSEGV;
105		info.si_errno		= 0;
106		info.si_code		= SEGV_MAPERR;
107		info.si_addr		= (void __user *)ptr;
108
109		force_sig_info(SIGSEGV, &info, current);
110		return false;
111	} else {
112		return true;
113	}
114}
115
116bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
117{
118	struct task_struct *tsk;
119	unsigned long caller;
120	int vsyscall_nr, syscall_nr, tmp;
121	int prev_sig_on_uaccess_error;
122	long ret;
123
124	/*
125	 * No point in checking CS -- the only way to get here is a user mode
126	 * trap to a high address, which means that we're in 64-bit user code.
127	 */
128
129	WARN_ON_ONCE(address != regs->ip);
130
131	if (vsyscall_mode == NONE) {
132		warn_bad_vsyscall(KERN_INFO, regs,
133				  "vsyscall attempted with vsyscall=none");
134		return false;
135	}
136
137	vsyscall_nr = addr_to_vsyscall_nr(address);
138
139	trace_emulate_vsyscall(vsyscall_nr);
140
141	if (vsyscall_nr < 0) {
142		warn_bad_vsyscall(KERN_WARNING, regs,
143				  "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
144		goto sigsegv;
145	}
146
147	if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
148		warn_bad_vsyscall(KERN_WARNING, regs,
149				  "vsyscall with bad stack (exploit attempt?)");
150		goto sigsegv;
151	}
152
153	tsk = current;
154
155	/*
156	 * Check for access_ok violations and find the syscall nr.
157	 *
158	 * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
159	 * 64-bit, so we don't need to special-case it here.  For all the
160	 * vsyscalls, NULL means "don't write anything" not "write it at
161	 * address 0".
162	 */
163	switch (vsyscall_nr) {
164	case 0:
165		if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
166		    !write_ok_or_segv(regs->si, sizeof(struct timezone))) {
167			ret = -EFAULT;
168			goto check_fault;
169		}
170
171		syscall_nr = __NR_gettimeofday;
172		break;
173
174	case 1:
175		if (!write_ok_or_segv(regs->di, sizeof(time_t))) {
176			ret = -EFAULT;
177			goto check_fault;
178		}
179
180		syscall_nr = __NR_time;
181		break;
182
183	case 2:
184		if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
185		    !write_ok_or_segv(regs->si, sizeof(unsigned))) {
186			ret = -EFAULT;
187			goto check_fault;
188		}
189
190		syscall_nr = __NR_getcpu;
191		break;
192	}
193
194	/*
195	 * Handle seccomp.  regs->ip must be the original value.
196	 * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt.
197	 *
198	 * We could optimize the seccomp disabled case, but performance
199	 * here doesn't matter.
200	 */
201	regs->orig_ax = syscall_nr;
202	regs->ax = -ENOSYS;
203	tmp = secure_computing();
204	if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) {
205		warn_bad_vsyscall(KERN_DEBUG, regs,
206				  "seccomp tried to change syscall nr or ip");
207		do_exit(SIGSYS);
208	}
209	regs->orig_ax = -1;
210	if (tmp)
211		goto do_ret;  /* skip requested */
212
213	/*
214	 * With a real vsyscall, page faults cause SIGSEGV.  We want to
215	 * preserve that behavior to make writing exploits harder.
216	 */
217	prev_sig_on_uaccess_error = current_thread_info()->sig_on_uaccess_error;
218	current_thread_info()->sig_on_uaccess_error = 1;
219
220	ret = -EFAULT;
221	switch (vsyscall_nr) {
222	case 0:
223		ret = sys_gettimeofday(
224			(struct timeval __user *)regs->di,
225			(struct timezone __user *)regs->si);
226		break;
227
228	case 1:
229		ret = sys_time((time_t __user *)regs->di);
230		break;
231
232	case 2:
233		ret = sys_getcpu((unsigned __user *)regs->di,
234				 (unsigned __user *)regs->si,
235				 NULL);
236		break;
237	}
238
239	current_thread_info()->sig_on_uaccess_error = prev_sig_on_uaccess_error;
240
241check_fault:
242	if (ret == -EFAULT) {
243		/* Bad news -- userspace fed a bad pointer to a vsyscall. */
244		warn_bad_vsyscall(KERN_INFO, regs,
245				  "vsyscall fault (exploit attempt?)");
246
247		/*
248		 * If we failed to generate a signal for any reason,
249		 * generate one here.  (This should be impossible.)
250		 */
251		if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
252				 !sigismember(&tsk->pending.signal, SIGSEGV)))
253			goto sigsegv;
254
255		return true;  /* Don't emulate the ret. */
256	}
257
258	regs->ax = ret;
259
260do_ret:
261	/* Emulate a ret instruction. */
262	regs->ip = caller;
263	regs->sp += 8;
264	return true;
265
266sigsegv:
267	force_sig(SIGSEGV, current);
268	return true;
269}
270
271/*
272 * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
273 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
274 * not need special handling anymore:
275 */
276static const char *gate_vma_name(struct vm_area_struct *vma)
277{
278	return "[vsyscall]";
279}
280static struct vm_operations_struct gate_vma_ops = {
281	.name = gate_vma_name,
282};
283static struct vm_area_struct gate_vma = {
284	.vm_start	= VSYSCALL_ADDR,
285	.vm_end		= VSYSCALL_ADDR + PAGE_SIZE,
286	.vm_page_prot	= PAGE_READONLY_EXEC,
287	.vm_flags	= VM_READ | VM_EXEC,
288	.vm_ops		= &gate_vma_ops,
289};
290
291struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
292{
293#ifdef CONFIG_IA32_EMULATION
294	if (!mm || mm->context.ia32_compat)
295		return NULL;
296#endif
297	if (vsyscall_mode == NONE)
298		return NULL;
299	return &gate_vma;
300}
301
302int in_gate_area(struct mm_struct *mm, unsigned long addr)
303{
304	struct vm_area_struct *vma = get_gate_vma(mm);
305
306	if (!vma)
307		return 0;
308
309	return (addr >= vma->vm_start) && (addr < vma->vm_end);
310}
311
312/*
313 * Use this when you have no reliable mm, typically from interrupt
314 * context. It is less reliable than using a task's mm and may give
315 * false positives.
316 */
317int in_gate_area_no_mm(unsigned long addr)
318{
319	return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR;
320}
321
322void __init map_vsyscall(void)
323{
324	extern char __vsyscall_page;
325	unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
326
327	if (vsyscall_mode != NONE)
328		__set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall,
329			     vsyscall_mode == NATIVE
330			     ? PAGE_KERNEL_VSYSCALL
331			     : PAGE_KERNEL_VVAR);
332
333	BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) !=
334		     (unsigned long)VSYSCALL_ADDR);
335}
336