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, XONLY, NONE } vsyscall_mode __ro_after_init =
46 #ifdef CONFIG_LEGACY_VSYSCALL_NONE
48 #elif defined(CONFIG_LEGACY_VSYSCALL_XONLY)
54 static int __init vsyscall_setup(char *str)
57 if (!strcmp("emulate", str))
58 vsyscall_mode = EMULATE;
59 else if (!strcmp("xonly", str))
60 vsyscall_mode = XONLY;
61 else if (!strcmp("none", str))
71 early_param("vsyscall", vsyscall_setup);
73 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
76 if (!show_unhandled_signals)
79 printk_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
80 level, current->comm, task_pid_nr(current),
81 message, regs->ip, regs->cs,
82 regs->sp, regs->ax, regs->si, regs->di);
85 static int addr_to_vsyscall_nr(unsigned long addr)
89 if ((addr & ~0xC00UL) != VSYSCALL_ADDR)
92 nr = (addr & 0xC00UL) >> 10;
99 static bool write_ok_or_segv(unsigned long ptr, size_t size)
102 * XXX: if access_ok, get_user, and put_user handled
103 * sig_on_uaccess_err, this could go away.
106 if (!access_ok((void __user *)ptr, size)) {
107 struct thread_struct *thread = ¤t->thread;
109 thread->error_code = X86_PF_USER | X86_PF_WRITE;
111 thread->trap_nr = X86_TRAP_PF;
113 force_sig_fault(SIGSEGV, SEGV_MAPERR, (void __user *)ptr);
120 bool emulate_vsyscall(unsigned long error_code,
121 struct pt_regs *regs, unsigned long address)
123 struct task_struct *tsk;
124 unsigned long caller;
125 int vsyscall_nr, syscall_nr, tmp;
126 int prev_sig_on_uaccess_err;
128 unsigned long orig_dx;
130 /* Write faults or kernel-privilege faults never get fixed up. */
131 if ((error_code & (X86_PF_WRITE | X86_PF_USER)) != X86_PF_USER)
134 if (!(error_code & X86_PF_INSTR)) {
135 /* Failed vsyscall read */
136 if (vsyscall_mode == EMULATE)
140 * User code tried and failed to read the vsyscall page.
142 warn_bad_vsyscall(KERN_INFO, regs, "vsyscall read attempt denied -- look up the vsyscall kernel parameter if you need a workaround");
147 * No point in checking CS -- the only way to get here is a user mode
148 * trap to a high address, which means that we're in 64-bit user code.
151 WARN_ON_ONCE(address != regs->ip);
153 if (vsyscall_mode == NONE) {
154 warn_bad_vsyscall(KERN_INFO, regs,
155 "vsyscall attempted with vsyscall=none");
159 vsyscall_nr = addr_to_vsyscall_nr(address);
161 trace_emulate_vsyscall(vsyscall_nr);
163 if (vsyscall_nr < 0) {
164 warn_bad_vsyscall(KERN_WARNING, regs,
165 "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
169 if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
170 warn_bad_vsyscall(KERN_WARNING, regs,
171 "vsyscall with bad stack (exploit attempt?)");
178 * Check for access_ok violations and find the syscall nr.
180 * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
181 * 64-bit, so we don't need to special-case it here. For all the
182 * vsyscalls, NULL means "don't write anything" not "write it at
185 switch (vsyscall_nr) {
187 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
188 !write_ok_or_segv(regs->si, sizeof(struct timezone))) {
193 syscall_nr = __NR_gettimeofday;
197 if (!write_ok_or_segv(regs->di, sizeof(time_t))) {
202 syscall_nr = __NR_time;
206 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
207 !write_ok_or_segv(regs->si, sizeof(unsigned))) {
212 syscall_nr = __NR_getcpu;
217 * Handle seccomp. regs->ip must be the original value.
218 * See seccomp_send_sigsys and Documentation/userspace-api/seccomp_filter.rst.
220 * We could optimize the seccomp disabled case, but performance
221 * here doesn't matter.
223 regs->orig_ax = syscall_nr;
225 tmp = secure_computing(NULL);
226 if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) {
227 warn_bad_vsyscall(KERN_DEBUG, regs,
228 "seccomp tried to change syscall nr or ip");
233 goto do_ret; /* skip requested */
236 * With a real vsyscall, page faults cause SIGSEGV. We want to
237 * preserve that behavior to make writing exploits harder.
239 prev_sig_on_uaccess_err = current->thread.sig_on_uaccess_err;
240 current->thread.sig_on_uaccess_err = 1;
243 switch (vsyscall_nr) {
245 /* this decodes regs->di and regs->si on its own */
246 ret = __x64_sys_gettimeofday(regs);
250 /* this decodes regs->di on its own */
251 ret = __x64_sys_time(regs);
255 /* while we could clobber regs->dx, we didn't in the past... */
258 /* this decodes regs->di, regs->si and regs->dx on its own */
259 ret = __x64_sys_getcpu(regs);
264 current->thread.sig_on_uaccess_err = prev_sig_on_uaccess_err;
267 if (ret == -EFAULT) {
268 /* Bad news -- userspace fed a bad pointer to a vsyscall. */
269 warn_bad_vsyscall(KERN_INFO, regs,
270 "vsyscall fault (exploit attempt?)");
273 * If we failed to generate a signal for any reason,
274 * generate one here. (This should be impossible.)
276 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
277 !sigismember(&tsk->pending.signal, SIGSEGV)))
280 return true; /* Don't emulate the ret. */
286 /* Emulate a ret instruction. */
297 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
298 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
299 * not need special handling anymore:
301 static const char *gate_vma_name(struct vm_area_struct *vma)
305 static const struct vm_operations_struct gate_vma_ops = {
306 .name = gate_vma_name,
308 static struct vm_area_struct gate_vma __ro_after_init = {
309 .vm_start = VSYSCALL_ADDR,
310 .vm_end = VSYSCALL_ADDR + PAGE_SIZE,
311 .vm_page_prot = PAGE_READONLY_EXEC,
312 .vm_flags = VM_READ | VM_EXEC,
313 .vm_ops = &gate_vma_ops,
316 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
319 if (!mm || mm->context.ia32_compat)
322 if (vsyscall_mode == NONE)
327 int in_gate_area(struct mm_struct *mm, unsigned long addr)
329 struct vm_area_struct *vma = get_gate_vma(mm);
334 return (addr >= vma->vm_start) && (addr < vma->vm_end);
338 * Use this when you have no reliable mm, typically from interrupt
339 * context. It is less reliable than using a task's mm and may give
342 int in_gate_area_no_mm(unsigned long addr)
344 return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR;
348 * The VSYSCALL page is the only user-accessible page in the kernel address
349 * range. Normally, the kernel page tables can have _PAGE_USER clear, but
350 * the tables covering VSYSCALL_ADDR need _PAGE_USER set if vsyscalls
353 * Some day we may create a "minimal" vsyscall mode in which we emulate
354 * vsyscalls but leave the page not present. If so, we skip calling
357 void __init set_vsyscall_pgtable_user_bits(pgd_t *root)
364 pgd = pgd_offset_pgd(root, VSYSCALL_ADDR);
365 set_pgd(pgd, __pgd(pgd_val(*pgd) | _PAGE_USER));
366 p4d = p4d_offset(pgd, VSYSCALL_ADDR);
367 #if CONFIG_PGTABLE_LEVELS >= 5
368 set_p4d(p4d, __p4d(p4d_val(*p4d) | _PAGE_USER));
370 pud = pud_offset(p4d, VSYSCALL_ADDR);
371 set_pud(pud, __pud(pud_val(*pud) | _PAGE_USER));
372 pmd = pmd_offset(pud, VSYSCALL_ADDR);
373 set_pmd(pmd, __pmd(pmd_val(*pmd) | _PAGE_USER));
376 void __init map_vsyscall(void)
378 extern char __vsyscall_page;
379 unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
382 * For full emulation, the page needs to exist for real. In
383 * execute-only mode, there is no PTE at all backing the vsyscall
386 if (vsyscall_mode == EMULATE) {
387 __set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall,
389 set_vsyscall_pgtable_user_bits(swapper_pg_dir);
392 if (vsyscall_mode == XONLY)
393 gate_vma.vm_flags = VM_EXEC;
395 BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) !=
396 (unsigned long)VSYSCALL_ADDR);