2 * Copyright (c) 2012-2014 Andy Lutomirski <luto@amacapital.net>
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.
8 * Parts of the original code have been moved to arch/x86/vdso/vma.c
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:
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.
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.
24 * The code in this file emulates vsyscalls when notified of a page
25 * fault to a vsyscall address.
28 #include <linux/kernel.h>
29 #include <linux/timer.h>
30 #include <linux/sched/signal.h>
31 #include <linux/syscalls.h>
32 #include <linux/ratelimit.h>
34 #include <asm/vsyscall.h>
35 #include <asm/unistd.h>
36 #include <asm/fixmap.h>
37 #include <asm/traps.h>
39 #define CREATE_TRACE_POINTS
40 #include "vsyscall_trace.h"
42 static enum { EMULATE, NATIVE, NONE } vsyscall_mode =
43 #if defined(CONFIG_LEGACY_VSYSCALL_NATIVE)
45 #elif defined(CONFIG_LEGACY_VSYSCALL_NONE)
51 static int __init vsyscall_setup(char *str)
54 if (!strcmp("emulate", str))
55 vsyscall_mode = EMULATE;
56 else if (!strcmp("native", str))
57 vsyscall_mode = NATIVE;
58 else if (!strcmp("none", str))
68 early_param("vsyscall", vsyscall_setup);
70 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
73 if (!show_unhandled_signals)
76 printk_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
77 level, current->comm, task_pid_nr(current),
78 message, regs->ip, regs->cs,
79 regs->sp, regs->ax, regs->si, regs->di);
82 static int addr_to_vsyscall_nr(unsigned long addr)
86 if ((addr & ~0xC00UL) != VSYSCALL_ADDR)
89 nr = (addr & 0xC00UL) >> 10;
96 static bool write_ok_or_segv(unsigned long ptr, size_t size)
99 * XXX: if access_ok, get_user, and put_user handled
100 * sig_on_uaccess_err, this could go away.
103 if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
105 struct thread_struct *thread = ¤t->thread;
107 thread->error_code = 6; /* user fault, no page, write */
109 thread->trap_nr = X86_TRAP_PF;
111 memset(&info, 0, sizeof(info));
112 info.si_signo = SIGSEGV;
114 info.si_code = SEGV_MAPERR;
115 info.si_addr = (void __user *)ptr;
117 force_sig_info(SIGSEGV, &info, current);
124 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
126 struct task_struct *tsk;
127 unsigned long caller;
128 int vsyscall_nr, syscall_nr, tmp;
129 int prev_sig_on_uaccess_err;
133 * No point in checking CS -- the only way to get here is a user mode
134 * trap to a high address, which means that we're in 64-bit user code.
137 WARN_ON_ONCE(address != regs->ip);
139 if (vsyscall_mode == NONE) {
140 warn_bad_vsyscall(KERN_INFO, regs,
141 "vsyscall attempted with vsyscall=none");
145 vsyscall_nr = addr_to_vsyscall_nr(address);
147 trace_emulate_vsyscall(vsyscall_nr);
149 if (vsyscall_nr < 0) {
150 warn_bad_vsyscall(KERN_WARNING, regs,
151 "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
155 if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
156 warn_bad_vsyscall(KERN_WARNING, regs,
157 "vsyscall with bad stack (exploit attempt?)");
164 * Check for access_ok violations and find the syscall nr.
166 * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
167 * 64-bit, so we don't need to special-case it here. For all the
168 * vsyscalls, NULL means "don't write anything" not "write it at
171 switch (vsyscall_nr) {
173 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
174 !write_ok_or_segv(regs->si, sizeof(struct timezone))) {
179 syscall_nr = __NR_gettimeofday;
183 if (!write_ok_or_segv(regs->di, sizeof(time_t))) {
188 syscall_nr = __NR_time;
192 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
193 !write_ok_or_segv(regs->si, sizeof(unsigned))) {
198 syscall_nr = __NR_getcpu;
203 * Handle seccomp. regs->ip must be the original value.
204 * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt.
206 * We could optimize the seccomp disabled case, but performance
207 * here doesn't matter.
209 regs->orig_ax = syscall_nr;
211 tmp = secure_computing(NULL);
212 if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) {
213 warn_bad_vsyscall(KERN_DEBUG, regs,
214 "seccomp tried to change syscall nr or ip");
219 goto do_ret; /* skip requested */
222 * With a real vsyscall, page faults cause SIGSEGV. We want to
223 * preserve that behavior to make writing exploits harder.
225 prev_sig_on_uaccess_err = current->thread.sig_on_uaccess_err;
226 current->thread.sig_on_uaccess_err = 1;
229 switch (vsyscall_nr) {
231 ret = sys_gettimeofday(
232 (struct timeval __user *)regs->di,
233 (struct timezone __user *)regs->si);
237 ret = sys_time((time_t __user *)regs->di);
241 ret = sys_getcpu((unsigned __user *)regs->di,
242 (unsigned __user *)regs->si,
247 current->thread.sig_on_uaccess_err = prev_sig_on_uaccess_err;
250 if (ret == -EFAULT) {
251 /* Bad news -- userspace fed a bad pointer to a vsyscall. */
252 warn_bad_vsyscall(KERN_INFO, regs,
253 "vsyscall fault (exploit attempt?)");
256 * If we failed to generate a signal for any reason,
257 * generate one here. (This should be impossible.)
259 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
260 !sigismember(&tsk->pending.signal, SIGSEGV)))
263 return true; /* Don't emulate the ret. */
269 /* Emulate a ret instruction. */
275 force_sig(SIGSEGV, current);
280 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
281 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
282 * not need special handling anymore:
284 static const char *gate_vma_name(struct vm_area_struct *vma)
288 static const struct vm_operations_struct gate_vma_ops = {
289 .name = gate_vma_name,
291 static struct vm_area_struct gate_vma = {
292 .vm_start = VSYSCALL_ADDR,
293 .vm_end = VSYSCALL_ADDR + PAGE_SIZE,
294 .vm_page_prot = PAGE_READONLY_EXEC,
295 .vm_flags = VM_READ | VM_EXEC,
296 .vm_ops = &gate_vma_ops,
299 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
302 if (!mm || mm->context.ia32_compat)
305 if (vsyscall_mode == NONE)
310 int in_gate_area(struct mm_struct *mm, unsigned long addr)
312 struct vm_area_struct *vma = get_gate_vma(mm);
317 return (addr >= vma->vm_start) && (addr < vma->vm_end);
321 * Use this when you have no reliable mm, typically from interrupt
322 * context. It is less reliable than using a task's mm and may give
325 int in_gate_area_no_mm(unsigned long addr)
327 return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR;
330 void __init map_vsyscall(void)
332 extern char __vsyscall_page;
333 unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
335 if (vsyscall_mode != NONE)
336 __set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall,
337 vsyscall_mode == NATIVE
338 ? PAGE_KERNEL_VSYSCALL
341 BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) !=
342 (unsigned long)VSYSCALL_ADDR);