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 
41 static enum { EMULATE, NATIVE, NONE } vsyscall_mode =
42 #if defined(CONFIG_LEGACY_VSYSCALL_NATIVE)
43 	NATIVE;
44 #elif defined(CONFIG_LEGACY_VSYSCALL_NONE)
45 	NONE;
46 #else
47 	EMULATE;
48 #endif
49 
50 static int __init vsyscall_setup(char *str)
51 {
52 	if (str) {
53 		if (!strcmp("emulate", str))
54 			vsyscall_mode = EMULATE;
55 		else if (!strcmp("native", str))
56 			vsyscall_mode = NATIVE;
57 		else if (!strcmp("none", str))
58 			vsyscall_mode = NONE;
59 		else
60 			return -EINVAL;
61 
62 		return 0;
63 	}
64 
65 	return -EINVAL;
66 }
67 early_param("vsyscall", vsyscall_setup);
68 
69 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
70 			      const char *message)
71 {
72 	if (!show_unhandled_signals)
73 		return;
74 
75 	printk_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
76 			   level, current->comm, task_pid_nr(current),
77 			   message, regs->ip, regs->cs,
78 			   regs->sp, regs->ax, regs->si, regs->di);
79 }
80 
81 static int addr_to_vsyscall_nr(unsigned long addr)
82 {
83 	int nr;
84 
85 	if ((addr & ~0xC00UL) != VSYSCALL_ADDR)
86 		return -EINVAL;
87 
88 	nr = (addr & 0xC00UL) >> 10;
89 	if (nr >= 3)
90 		return -EINVAL;
91 
92 	return nr;
93 }
94 
95 static bool write_ok_or_segv(unsigned long ptr, size_t size)
96 {
97 	/*
98 	 * XXX: if access_ok, get_user, and put_user handled
99 	 * sig_on_uaccess_error, this could go away.
100 	 */
101 
102 	if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
103 		siginfo_t info;
104 		struct thread_struct *thread = &current->thread;
105 
106 		thread->error_code	= 6;  /* user fault, no page, write */
107 		thread->cr2		= ptr;
108 		thread->trap_nr		= X86_TRAP_PF;
109 
110 		memset(&info, 0, sizeof(info));
111 		info.si_signo		= SIGSEGV;
112 		info.si_errno		= 0;
113 		info.si_code		= SEGV_MAPERR;
114 		info.si_addr		= (void __user *)ptr;
115 
116 		force_sig_info(SIGSEGV, &info, current);
117 		return false;
118 	} else {
119 		return true;
120 	}
121 }
122 
123 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
124 {
125 	struct task_struct *tsk;
126 	unsigned long caller;
127 	int vsyscall_nr, syscall_nr, tmp;
128 	int prev_sig_on_uaccess_error;
129 	long ret;
130 
131 	/*
132 	 * No point in checking CS -- the only way to get here is a user mode
133 	 * trap to a high address, which means that we're in 64-bit user code.
134 	 */
135 
136 	WARN_ON_ONCE(address != regs->ip);
137 
138 	if (vsyscall_mode == NONE) {
139 		warn_bad_vsyscall(KERN_INFO, regs,
140 				  "vsyscall attempted with vsyscall=none");
141 		return false;
142 	}
143 
144 	vsyscall_nr = addr_to_vsyscall_nr(address);
145 
146 	trace_emulate_vsyscall(vsyscall_nr);
147 
148 	if (vsyscall_nr < 0) {
149 		warn_bad_vsyscall(KERN_WARNING, regs,
150 				  "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
151 		goto sigsegv;
152 	}
153 
154 	if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
155 		warn_bad_vsyscall(KERN_WARNING, regs,
156 				  "vsyscall with bad stack (exploit attempt?)");
157 		goto sigsegv;
158 	}
159 
160 	tsk = current;
161 
162 	/*
163 	 * Check for access_ok violations and find the syscall nr.
164 	 *
165 	 * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
166 	 * 64-bit, so we don't need to special-case it here.  For all the
167 	 * vsyscalls, NULL means "don't write anything" not "write it at
168 	 * address 0".
169 	 */
170 	switch (vsyscall_nr) {
171 	case 0:
172 		if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
173 		    !write_ok_or_segv(regs->si, sizeof(struct timezone))) {
174 			ret = -EFAULT;
175 			goto check_fault;
176 		}
177 
178 		syscall_nr = __NR_gettimeofday;
179 		break;
180 
181 	case 1:
182 		if (!write_ok_or_segv(regs->di, sizeof(time_t))) {
183 			ret = -EFAULT;
184 			goto check_fault;
185 		}
186 
187 		syscall_nr = __NR_time;
188 		break;
189 
190 	case 2:
191 		if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
192 		    !write_ok_or_segv(regs->si, sizeof(unsigned))) {
193 			ret = -EFAULT;
194 			goto check_fault;
195 		}
196 
197 		syscall_nr = __NR_getcpu;
198 		break;
199 	}
200 
201 	/*
202 	 * Handle seccomp.  regs->ip must be the original value.
203 	 * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt.
204 	 *
205 	 * We could optimize the seccomp disabled case, but performance
206 	 * here doesn't matter.
207 	 */
208 	regs->orig_ax = syscall_nr;
209 	regs->ax = -ENOSYS;
210 	tmp = secure_computing();
211 	if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) {
212 		warn_bad_vsyscall(KERN_DEBUG, regs,
213 				  "seccomp tried to change syscall nr or ip");
214 		do_exit(SIGSYS);
215 	}
216 	regs->orig_ax = -1;
217 	if (tmp)
218 		goto do_ret;  /* skip requested */
219 
220 	/*
221 	 * With a real vsyscall, page faults cause SIGSEGV.  We want to
222 	 * preserve that behavior to make writing exploits harder.
223 	 */
224 	prev_sig_on_uaccess_error = current_thread_info()->sig_on_uaccess_error;
225 	current_thread_info()->sig_on_uaccess_error = 1;
226 
227 	ret = -EFAULT;
228 	switch (vsyscall_nr) {
229 	case 0:
230 		ret = sys_gettimeofday(
231 			(struct timeval __user *)regs->di,
232 			(struct timezone __user *)regs->si);
233 		break;
234 
235 	case 1:
236 		ret = sys_time((time_t __user *)regs->di);
237 		break;
238 
239 	case 2:
240 		ret = sys_getcpu((unsigned __user *)regs->di,
241 				 (unsigned __user *)regs->si,
242 				 NULL);
243 		break;
244 	}
245 
246 	current_thread_info()->sig_on_uaccess_error = prev_sig_on_uaccess_error;
247 
248 check_fault:
249 	if (ret == -EFAULT) {
250 		/* Bad news -- userspace fed a bad pointer to a vsyscall. */
251 		warn_bad_vsyscall(KERN_INFO, regs,
252 				  "vsyscall fault (exploit attempt?)");
253 
254 		/*
255 		 * If we failed to generate a signal for any reason,
256 		 * generate one here.  (This should be impossible.)
257 		 */
258 		if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
259 				 !sigismember(&tsk->pending.signal, SIGSEGV)))
260 			goto sigsegv;
261 
262 		return true;  /* Don't emulate the ret. */
263 	}
264 
265 	regs->ax = ret;
266 
267 do_ret:
268 	/* Emulate a ret instruction. */
269 	regs->ip = caller;
270 	regs->sp += 8;
271 	return true;
272 
273 sigsegv:
274 	force_sig(SIGSEGV, current);
275 	return true;
276 }
277 
278 /*
279  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
280  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
281  * not need special handling anymore:
282  */
283 static const char *gate_vma_name(struct vm_area_struct *vma)
284 {
285 	return "[vsyscall]";
286 }
287 static const struct vm_operations_struct gate_vma_ops = {
288 	.name = gate_vma_name,
289 };
290 static struct vm_area_struct gate_vma = {
291 	.vm_start	= VSYSCALL_ADDR,
292 	.vm_end		= VSYSCALL_ADDR + PAGE_SIZE,
293 	.vm_page_prot	= PAGE_READONLY_EXEC,
294 	.vm_flags	= VM_READ | VM_EXEC,
295 	.vm_ops		= &gate_vma_ops,
296 };
297 
298 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
299 {
300 #ifdef CONFIG_COMPAT
301 	if (!mm || mm->context.ia32_compat)
302 		return NULL;
303 #endif
304 	if (vsyscall_mode == NONE)
305 		return NULL;
306 	return &gate_vma;
307 }
308 
309 int in_gate_area(struct mm_struct *mm, unsigned long addr)
310 {
311 	struct vm_area_struct *vma = get_gate_vma(mm);
312 
313 	if (!vma)
314 		return 0;
315 
316 	return (addr >= vma->vm_start) && (addr < vma->vm_end);
317 }
318 
319 /*
320  * Use this when you have no reliable mm, typically from interrupt
321  * context. It is less reliable than using a task's mm and may give
322  * false positives.
323  */
324 int in_gate_area_no_mm(unsigned long addr)
325 {
326 	return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR;
327 }
328 
329 void __init map_vsyscall(void)
330 {
331 	extern char __vsyscall_page;
332 	unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
333 
334 	if (vsyscall_mode != NONE)
335 		__set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall,
336 			     vsyscall_mode == NATIVE
337 			     ? PAGE_KERNEL_VSYSCALL
338 			     : PAGE_KERNEL_VVAR);
339 
340 	BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) !=
341 		     (unsigned long)VSYSCALL_ADDR);
342 }
343