1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2012-2014 Andy Lutomirski <luto@amacapital.net>
5 * Based on the original implementation which is:
6 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
7 * Copyright 2003 Andi Kleen, SuSE Labs.
9 * Parts of the original code have been moved to arch/x86/vdso/vma.c
11 * This file implements vsyscall emulation. vsyscalls are a legacy ABI:
12 * Userspace can request certain kernel services by calling fixed
13 * addresses. This concept is problematic:
15 * - It interferes with ASLR.
16 * - It's awkward to write code that lives in kernel addresses but is
17 * callable by userspace at fixed addresses.
18 * - The whole concept is impossible for 32-bit compat userspace.
19 * - UML cannot easily virtualize a vsyscall.
21 * As of mid-2014, I believe that there is no new userspace code that
22 * will use a vsyscall if the vDSO is present. I hope that there will
23 * soon be no new userspace code that will ever use a vsyscall.
25 * The code in this file emulates vsyscalls when notified of a page
26 * fault to a vsyscall address.
29 #include <linux/kernel.h>
30 #include <linux/timer.h>
31 #include <linux/sched/signal.h>
32 #include <linux/mm_types.h>
33 #include <linux/syscalls.h>
34 #include <linux/ratelimit.h>
36 #include <asm/vsyscall.h>
37 #include <asm/unistd.h>
38 #include <asm/fixmap.h>
39 #include <asm/traps.h>
40 #include <asm/paravirt.h>
42 #define CREATE_TRACE_POINTS
43 #include "vsyscall_trace.h"
45 static enum { EMULATE
, NONE
} vsyscall_mode
=
46 #ifdef CONFIG_LEGACY_VSYSCALL_NONE
52 static int __init
vsyscall_setup(char *str
)
55 if (!strcmp("emulate", str
))
56 vsyscall_mode
= EMULATE
;
57 else if (!strcmp("none", str
))
67 early_param("vsyscall", vsyscall_setup
);
69 static void warn_bad_vsyscall(const char *level
, struct pt_regs
*regs
,
72 if (!show_unhandled_signals
)
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
);
81 static int addr_to_vsyscall_nr(unsigned long addr
)
85 if ((addr
& ~0xC00UL
) != VSYSCALL_ADDR
)
88 nr
= (addr
& 0xC00UL
) >> 10;
95 static bool write_ok_or_segv(unsigned long ptr
, size_t size
)
98 * XXX: if access_ok, get_user, and put_user handled
99 * sig_on_uaccess_err, this could go away.
102 if (!access_ok(VERIFY_WRITE
, (void __user
*)ptr
, size
)) {
104 struct thread_struct
*thread
= ¤t
->thread
;
106 thread
->error_code
= 6; /* user fault, no page, write */
108 thread
->trap_nr
= X86_TRAP_PF
;
110 memset(&info
, 0, sizeof(info
));
111 info
.si_signo
= SIGSEGV
;
113 info
.si_code
= SEGV_MAPERR
;
114 info
.si_addr
= (void __user
*)ptr
;
116 force_sig_info(SIGSEGV
, &info
, current
);
123 bool emulate_vsyscall(struct pt_regs
*regs
, unsigned long address
)
125 struct task_struct
*tsk
;
126 unsigned long caller
;
127 int vsyscall_nr
, syscall_nr
, tmp
;
128 int prev_sig_on_uaccess_err
;
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.
136 WARN_ON_ONCE(address
!= regs
->ip
);
138 if (vsyscall_mode
== NONE
) {
139 warn_bad_vsyscall(KERN_INFO
, regs
,
140 "vsyscall attempted with vsyscall=none");
144 vsyscall_nr
= addr_to_vsyscall_nr(address
);
146 trace_emulate_vsyscall(vsyscall_nr
);
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");
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?)");
163 * Check for access_ok violations and find the syscall nr.
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
170 switch (vsyscall_nr
) {
172 if (!write_ok_or_segv(regs
->di
, sizeof(struct timeval
)) ||
173 !write_ok_or_segv(regs
->si
, sizeof(struct timezone
))) {
178 syscall_nr
= __NR_gettimeofday
;
182 if (!write_ok_or_segv(regs
->di
, sizeof(time_t))) {
187 syscall_nr
= __NR_time
;
191 if (!write_ok_or_segv(regs
->di
, sizeof(unsigned)) ||
192 !write_ok_or_segv(regs
->si
, sizeof(unsigned))) {
197 syscall_nr
= __NR_getcpu
;
202 * Handle seccomp. regs->ip must be the original value.
203 * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt.
205 * We could optimize the seccomp disabled case, but performance
206 * here doesn't matter.
208 regs
->orig_ax
= syscall_nr
;
210 tmp
= secure_computing(NULL
);
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");
218 goto do_ret
; /* skip requested */
221 * With a real vsyscall, page faults cause SIGSEGV. We want to
222 * preserve that behavior to make writing exploits harder.
224 prev_sig_on_uaccess_err
= current
->thread
.sig_on_uaccess_err
;
225 current
->thread
.sig_on_uaccess_err
= 1;
228 switch (vsyscall_nr
) {
230 ret
= sys_gettimeofday(
231 (struct timeval __user
*)regs
->di
,
232 (struct timezone __user
*)regs
->si
);
236 ret
= sys_time((time_t __user
*)regs
->di
);
240 ret
= sys_getcpu((unsigned __user
*)regs
->di
,
241 (unsigned __user
*)regs
->si
,
246 current
->thread
.sig_on_uaccess_err
= prev_sig_on_uaccess_err
;
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?)");
255 * If we failed to generate a signal for any reason,
256 * generate one here. (This should be impossible.)
258 if (WARN_ON_ONCE(!sigismember(&tsk
->pending
.signal
, SIGBUS
) &&
259 !sigismember(&tsk
->pending
.signal
, SIGSEGV
)))
262 return true; /* Don't emulate the ret. */
268 /* Emulate a ret instruction. */
274 force_sig(SIGSEGV
, current
);
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:
283 static const char *gate_vma_name(struct vm_area_struct
*vma
)
287 static const struct vm_operations_struct gate_vma_ops
= {
288 .name
= gate_vma_name
,
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
,
298 struct vm_area_struct
*get_gate_vma(struct mm_struct
*mm
)
301 if (!mm
|| mm
->context
.ia32_compat
)
304 if (vsyscall_mode
== NONE
)
309 int in_gate_area(struct mm_struct
*mm
, unsigned long addr
)
311 struct vm_area_struct
*vma
= get_gate_vma(mm
);
316 return (addr
>= vma
->vm_start
) && (addr
< vma
->vm_end
);
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
324 int in_gate_area_no_mm(unsigned long addr
)
326 return vsyscall_mode
!= NONE
&& (addr
& PAGE_MASK
) == VSYSCALL_ADDR
;
330 * The VSYSCALL page is the only user-accessible page in the kernel address
331 * range. Normally, the kernel page tables can have _PAGE_USER clear, but
332 * the tables covering VSYSCALL_ADDR need _PAGE_USER set if vsyscalls
335 * Some day we may create a "minimal" vsyscall mode in which we emulate
336 * vsyscalls but leave the page not present. If so, we skip calling
339 void __init
set_vsyscall_pgtable_user_bits(pgd_t
*root
)
346 pgd
= pgd_offset_pgd(root
, VSYSCALL_ADDR
);
347 set_pgd(pgd
, __pgd(pgd_val(*pgd
) | _PAGE_USER
));
348 p4d
= p4d_offset(pgd
, VSYSCALL_ADDR
);
349 #if CONFIG_PGTABLE_LEVELS >= 5
350 p4d
->p4d
|= _PAGE_USER
;
352 pud
= pud_offset(p4d
, VSYSCALL_ADDR
);
353 set_pud(pud
, __pud(pud_val(*pud
) | _PAGE_USER
));
354 pmd
= pmd_offset(pud
, VSYSCALL_ADDR
);
355 set_pmd(pmd
, __pmd(pmd_val(*pmd
) | _PAGE_USER
));
358 void __init
map_vsyscall(void)
360 extern char __vsyscall_page
;
361 unsigned long physaddr_vsyscall
= __pa_symbol(&__vsyscall_page
);
363 if (vsyscall_mode
!= NONE
) {
364 __set_fixmap(VSYSCALL_PAGE
, physaddr_vsyscall
,
366 set_vsyscall_pgtable_user_bits(swapper_pg_dir
);
369 BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE
) !=
370 (unsigned long)VSYSCALL_ADDR
);