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/syscalls.h>
31 #include <linux/ratelimit.h>
33 #include <asm/vsyscall.h>
34 #include <asm/unistd.h>
35 #include <asm/fixmap.h>
36 #include <asm/traps.h>
38 #define CREATE_TRACE_POINTS
39 #include "vsyscall_trace.h"
41 static enum { EMULATE, NATIVE, NONE } vsyscall_mode =
42 #if defined(CONFIG_LEGACY_VSYSCALL_NATIVE)
44 #elif defined(CONFIG_LEGACY_VSYSCALL_NONE)
49 unsigned long vsyscall_pgprot = __PAGE_KERNEL_VSYSCALL;
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 bool vsyscall_enabled(void)
72 return vsyscall_mode != NONE;
75 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
78 if (!show_unhandled_signals)
81 printk_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
82 level, current->comm, task_pid_nr(current),
83 message, regs->ip, regs->cs,
84 regs->sp, regs->ax, regs->si, regs->di);
87 static int addr_to_vsyscall_nr(unsigned long addr)
91 if ((addr & ~0xC00UL) != VSYSCALL_ADDR)
94 nr = (addr & 0xC00UL) >> 10;
101 static bool write_ok_or_segv(unsigned long ptr, size_t size)
104 * XXX: if access_ok, get_user, and put_user handled
105 * sig_on_uaccess_err, this could go away.
108 if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
110 struct thread_struct *thread = ¤t->thread;
112 thread->error_code = 6; /* user fault, no page, write */
114 thread->trap_nr = X86_TRAP_PF;
116 memset(&info, 0, sizeof(info));
117 info.si_signo = SIGSEGV;
119 info.si_code = SEGV_MAPERR;
120 info.si_addr = (void __user *)ptr;
122 force_sig_info(SIGSEGV, &info, current);
129 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
131 struct task_struct *tsk;
132 unsigned long caller;
133 int vsyscall_nr, syscall_nr, tmp;
134 int prev_sig_on_uaccess_err;
138 * No point in checking CS -- the only way to get here is a user mode
139 * trap to a high address, which means that we're in 64-bit user code.
142 WARN_ON_ONCE(address != regs->ip);
144 if (vsyscall_mode == NONE) {
145 warn_bad_vsyscall(KERN_INFO, regs,
146 "vsyscall attempted with vsyscall=none");
150 vsyscall_nr = addr_to_vsyscall_nr(address);
152 trace_emulate_vsyscall(vsyscall_nr);
154 if (vsyscall_nr < 0) {
155 warn_bad_vsyscall(KERN_WARNING, regs,
156 "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
160 if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
161 warn_bad_vsyscall(KERN_WARNING, regs,
162 "vsyscall with bad stack (exploit attempt?)");
169 * Check for access_ok violations and find the syscall nr.
171 * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
172 * 64-bit, so we don't need to special-case it here. For all the
173 * vsyscalls, NULL means "don't write anything" not "write it at
176 switch (vsyscall_nr) {
178 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
179 !write_ok_or_segv(regs->si, sizeof(struct timezone))) {
184 syscall_nr = __NR_gettimeofday;
188 if (!write_ok_or_segv(regs->di, sizeof(time_t))) {
193 syscall_nr = __NR_time;
197 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
198 !write_ok_or_segv(regs->si, sizeof(unsigned))) {
203 syscall_nr = __NR_getcpu;
208 * Handle seccomp. regs->ip must be the original value.
209 * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt.
211 * We could optimize the seccomp disabled case, but performance
212 * here doesn't matter.
214 regs->orig_ax = syscall_nr;
216 tmp = secure_computing(NULL);
217 if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) {
218 warn_bad_vsyscall(KERN_DEBUG, regs,
219 "seccomp tried to change syscall nr or ip");
224 goto do_ret; /* skip requested */
227 * With a real vsyscall, page faults cause SIGSEGV. We want to
228 * preserve that behavior to make writing exploits harder.
230 prev_sig_on_uaccess_err = current->thread.sig_on_uaccess_err;
231 current->thread.sig_on_uaccess_err = 1;
234 switch (vsyscall_nr) {
236 ret = sys_gettimeofday(
237 (struct timeval __user *)regs->di,
238 (struct timezone __user *)regs->si);
242 ret = sys_time((time_t __user *)regs->di);
246 ret = sys_getcpu((unsigned __user *)regs->di,
247 (unsigned __user *)regs->si,
252 current->thread.sig_on_uaccess_err = prev_sig_on_uaccess_err;
255 if (ret == -EFAULT) {
256 /* Bad news -- userspace fed a bad pointer to a vsyscall. */
257 warn_bad_vsyscall(KERN_INFO, regs,
258 "vsyscall fault (exploit attempt?)");
261 * If we failed to generate a signal for any reason,
262 * generate one here. (This should be impossible.)
264 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
265 !sigismember(&tsk->pending.signal, SIGSEGV)))
268 return true; /* Don't emulate the ret. */
274 /* Emulate a ret instruction. */
280 force_sig(SIGSEGV, current);
285 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
286 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
287 * not need special handling anymore:
289 static const char *gate_vma_name(struct vm_area_struct *vma)
293 static const struct vm_operations_struct gate_vma_ops = {
294 .name = gate_vma_name,
296 static struct vm_area_struct gate_vma = {
297 .vm_start = VSYSCALL_ADDR,
298 .vm_end = VSYSCALL_ADDR + PAGE_SIZE,
299 .vm_page_prot = PAGE_READONLY_EXEC,
300 .vm_flags = VM_READ | VM_EXEC,
301 .vm_ops = &gate_vma_ops,
304 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
307 if (!mm || mm->context.ia32_compat)
310 if (vsyscall_mode == NONE)
315 int in_gate_area(struct mm_struct *mm, unsigned long addr)
317 struct vm_area_struct *vma = get_gate_vma(mm);
322 return (addr >= vma->vm_start) && (addr < vma->vm_end);
326 * Use this when you have no reliable mm, typically from interrupt
327 * context. It is less reliable than using a task's mm and may give
330 int in_gate_area_no_mm(unsigned long addr)
332 return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR;
335 void __init map_vsyscall(void)
337 extern char __vsyscall_page;
338 unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
340 if (vsyscall_mode != NATIVE)
341 vsyscall_pgprot = __PAGE_KERNEL_VVAR;
342 if (vsyscall_mode != NONE)
343 __set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall,
344 __pgprot(vsyscall_pgprot));
346 BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) !=
347 (unsigned long)VSYSCALL_ADDR);