4 * Based on the original implementation which is:
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 = EMULATE;
43 static int __init vsyscall_setup(char *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))
60 early_param("vsyscall", vsyscall_setup);
62 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
65 if (!show_unhandled_signals)
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);
74 static int addr_to_vsyscall_nr(unsigned long addr)
78 if ((addr & ~0xC00UL) != VSYSCALL_ADDR)
81 nr = (addr & 0xC00UL) >> 10;
88 static bool write_ok_or_segv(unsigned long ptr, size_t size)
91 * XXX: if access_ok, get_user, and put_user handled
92 * sig_on_uaccess_error, this could go away.
95 if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
97 struct thread_struct *thread = ¤t->thread;
99 thread->error_code = 6; /* user fault, no page, write */
101 thread->trap_nr = X86_TRAP_PF;
103 memset(&info, 0, sizeof(info));
104 info.si_signo = SIGSEGV;
106 info.si_code = SEGV_MAPERR;
107 info.si_addr = (void __user *)ptr;
109 force_sig_info(SIGSEGV, &info, current);
116 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
118 struct task_struct *tsk;
119 unsigned long caller;
120 int vsyscall_nr, syscall_nr, tmp;
121 int prev_sig_on_uaccess_error;
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.
129 WARN_ON_ONCE(address != regs->ip);
131 if (vsyscall_mode == NONE) {
132 warn_bad_vsyscall(KERN_INFO, regs,
133 "vsyscall attempted with vsyscall=none");
137 vsyscall_nr = addr_to_vsyscall_nr(address);
139 trace_emulate_vsyscall(vsyscall_nr);
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");
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?)");
156 * Check for access_ok violations and find the syscall nr.
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
163 switch (vsyscall_nr) {
165 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
166 !write_ok_or_segv(regs->si, sizeof(struct timezone))) {
171 syscall_nr = __NR_gettimeofday;
175 if (!write_ok_or_segv(regs->di, sizeof(time_t))) {
180 syscall_nr = __NR_time;
184 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
185 !write_ok_or_segv(regs->si, sizeof(unsigned))) {
190 syscall_nr = __NR_getcpu;
195 * Handle seccomp. regs->ip must be the original value.
196 * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt.
198 * We could optimize the seccomp disabled case, but performance
199 * here doesn't matter.
201 regs->orig_ax = syscall_nr;
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");
211 goto do_ret; /* skip requested */
214 * With a real vsyscall, page faults cause SIGSEGV. We want to
215 * preserve that behavior to make writing exploits harder.
217 prev_sig_on_uaccess_error = current_thread_info()->sig_on_uaccess_error;
218 current_thread_info()->sig_on_uaccess_error = 1;
221 switch (vsyscall_nr) {
223 ret = sys_gettimeofday(
224 (struct timeval __user *)regs->di,
225 (struct timezone __user *)regs->si);
229 ret = sys_time((time_t __user *)regs->di);
233 ret = sys_getcpu((unsigned __user *)regs->di,
234 (unsigned __user *)regs->si,
239 current_thread_info()->sig_on_uaccess_error = prev_sig_on_uaccess_error;
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?)");
248 * If we failed to generate a signal for any reason,
249 * generate one here. (This should be impossible.)
251 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
252 !sigismember(&tsk->pending.signal, SIGSEGV)))
255 return true; /* Don't emulate the ret. */
261 /* Emulate a ret instruction. */
267 force_sig(SIGSEGV, current);
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:
276 static const char *gate_vma_name(struct vm_area_struct *vma)
280 static struct vm_operations_struct gate_vma_ops = {
281 .name = gate_vma_name,
283 static 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,
291 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
293 #ifdef CONFIG_IA32_EMULATION
294 if (!mm || mm->context.ia32_compat)
297 if (vsyscall_mode == NONE)
302 int in_gate_area(struct mm_struct *mm, unsigned long addr)
304 struct vm_area_struct *vma = get_gate_vma(mm);
309 return (addr >= vma->vm_start) && (addr < vma->vm_end);
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
317 int in_gate_area_no_mm(unsigned long addr)
319 return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR;
322 void __init map_vsyscall(void)
324 extern char __vsyscall_page;
325 unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
327 if (vsyscall_mode != NONE)
328 __set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall,
329 vsyscall_mode == NATIVE
330 ? PAGE_KERNEL_VSYSCALL
333 BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) !=
334 (unsigned long)VSYSCALL_ADDR);