arch/x86/entry/vsyscall/vsyscall_64.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2012-2014 Andy Lutomirski <luto@amacapital.net>
   4  *
   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.
   8  *
   9  *  Parts of the original code have been moved to arch/x86/vdso/vma.c
  10  *
  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:
  14  *
  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.
  20  *
  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.
  24  *
  25  * The code in this file emulates vsyscalls when notified of a page
  26  * fault to a vsyscall address.
  27  */
  28
  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>
  35
  36 #include <asm/vsyscall.h>
  37 #include <asm/unistd.h>
  38 #include <asm/fixmap.h>
  39 #include <asm/traps.h>
  40
  41 #define CREATE_TRACE_POINTS
  42 #include "vsyscall_trace.h"
  43
  44 static enum { EMULATE, NATIVE, NONE } vsyscall_mode =
  45 #if defined(CONFIG_LEGACY_VSYSCALL_NATIVE)
  46         NATIVE;
  47 #elif defined(CONFIG_LEGACY_VSYSCALL_NONE)
  48         NONE;
  49 #else
  50         EMULATE;
  51 #endif
  52
  53 static int __init vsyscall_setup(char *str)
  54 {
  55         if (str) {
  56                 if (!strcmp("emulate", str))
  57                         vsyscall_mode = EMULATE;
  58                 else if (!strcmp("native", str))
  59                         vsyscall_mode = NATIVE;
  60                 else if (!strcmp("none", str))
  61                         vsyscall_mode = NONE;
  62                 else
  63                         return -EINVAL;
  64
  65                 return 0;
  66         }
  67
  68         return -EINVAL;
  69 }
  70 early_param("vsyscall", vsyscall_setup);
  71
  72 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
  73                               const char *message)
  74 {
  75         if (!show_unhandled_signals)
  76                 return;
  77
  78         printk_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
  79                            level, current->comm, task_pid_nr(current),
  80                            message, regs->ip, regs->cs,
  81                            regs->sp, regs->ax, regs->si, regs->di);
  82 }
  83
  84 static int addr_to_vsyscall_nr(unsigned long addr)
  85 {
  86         int nr;
  87
  88         if ((addr & ~0xC00UL) != VSYSCALL_ADDR)
  89                 return -EINVAL;
  90
  91         nr = (addr & 0xC00UL) >> 10;
  92         if (nr >= 3)
  93                 return -EINVAL;
  94
  95         return nr;
  96 }
  97
  98 static bool write_ok_or_segv(unsigned long ptr, size_t size)
  99 {
 100         /*
 101          * XXX: if access_ok, get_user, and put_user handled
 102          * sig_on_uaccess_err, this could go away.
 103          */
 104
 105         if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
 106                 siginfo_t info;
 107                 struct thread_struct *thread = &current->thread;
 108
 109                 thread->error_code      = 6;  /* user fault, no page, write */
 110                 thread->cr2             = ptr;
 111                 thread->trap_nr         = X86_TRAP_PF;
 112
 113                 memset(&info, 0, sizeof(info));
 114                 info.si_signo           = SIGSEGV;
 115                 info.si_errno           = 0;
 116                 info.si_code            = SEGV_MAPERR;
 117                 info.si_addr            = (void __user *)ptr;
 118
 119                 force_sig_info(SIGSEGV, &info, current);
 120                 return false;
 121         } else {
 122                 return true;
 123         }
 124 }
 125
 126 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
 127 {
 128         struct task_struct *tsk;
 129         unsigned long caller;
 130         int vsyscall_nr, syscall_nr, tmp;
 131         int prev_sig_on_uaccess_err;
 132         long ret;
 133
 134         /*
 135          * No point in checking CS -- the only way to get here is a user mode
 136          * trap to a high address, which means that we're in 64-bit user code.
 137          */
 138
 139         WARN_ON_ONCE(address != regs->ip);
 140
 141         if (vsyscall_mode == NONE) {
 142                 warn_bad_vsyscall(KERN_INFO, regs,
 143                                   "vsyscall attempted with vsyscall=none");
 144                 return false;
 145         }
 146
 147         vsyscall_nr = addr_to_vsyscall_nr(address);
 148
 149         trace_emulate_vsyscall(vsyscall_nr);
 150
 151         if (vsyscall_nr < 0) {
 152                 warn_bad_vsyscall(KERN_WARNING, regs,
 153                                   "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
 154                 goto sigsegv;
 155         }
 156
 157         if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
 158                 warn_bad_vsyscall(KERN_WARNING, regs,
 159                                   "vsyscall with bad stack (exploit attempt?)");
 160                 goto sigsegv;
 161         }
 162
 163         tsk = current;
 164
 165         /*
 166          * Check for access_ok violations and find the syscall nr.
 167          *
 168          * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
 169          * 64-bit, so we don't need to special-case it here.  For all the
 170          * vsyscalls, NULL means "don't write anything" not "write it at
 171          * address 0".
 172          */
 173         switch (vsyscall_nr) {
 174         case 0:
 175                 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
 176                     !write_ok_or_segv(regs->si, sizeof(struct timezone))) {
 177                         ret = -EFAULT;
 178                         goto check_fault;
 179                 }
 180
 181                 syscall_nr = __NR_gettimeofday;
 182                 break;
 183
 184         case 1:
 185                 if (!write_ok_or_segv(regs->di, sizeof(time_t))) {
 186                         ret = -EFAULT;
 187                         goto check_fault;
 188                 }
 189
 190                 syscall_nr = __NR_time;
 191                 break;
 192
 193         case 2:
 194                 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
 195                     !write_ok_or_segv(regs->si, sizeof(unsigned))) {
 196                         ret = -EFAULT;
 197                         goto check_fault;
 198                 }
 199
 200                 syscall_nr = __NR_getcpu;
 201                 break;
 202         }
 203
 204         /*
 205          * Handle seccomp.  regs->ip must be the original value.
 206          * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt.
 207          *
 208          * We could optimize the seccomp disabled case, but performance
 209          * here doesn't matter.
 210          */
 211         regs->orig_ax = syscall_nr;
 212         regs->ax = -ENOSYS;
 213         tmp = secure_computing(NULL);
 214         if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) {
 215                 warn_bad_vsyscall(KERN_DEBUG, regs,
 216                                   "seccomp tried to change syscall nr or ip");
 217                 do_exit(SIGSYS);
 218         }
 219         regs->orig_ax = -1;
 220         if (tmp)
 221                 goto do_ret;  /* skip requested */
 222
 223         /*
 224          * With a real vsyscall, page faults cause SIGSEGV.  We want to
 225          * preserve that behavior to make writing exploits harder.
 226          */
 227         prev_sig_on_uaccess_err = current->thread.sig_on_uaccess_err;
 228         current->thread.sig_on_uaccess_err = 1;
 229
 230         ret = -EFAULT;
 231         switch (vsyscall_nr) {
 232         case 0:
 233                 ret = sys_gettimeofday(
 234                         (struct timeval __user *)regs->di,
 235                         (struct timezone __user *)regs->si);
 236                 break;
 237
 238         case 1:
 239                 ret = sys_time((time_t __user *)regs->di);
 240                 break;
 241
 242         case 2:
 243                 ret = sys_getcpu((unsigned __user *)regs->di,
 244                                  (unsigned __user *)regs->si,
 245                                  NULL);
 246                 break;
 247         }
 248
 249         current->thread.sig_on_uaccess_err = prev_sig_on_uaccess_err;
 250
 251 check_fault:
 252         if (ret == -EFAULT) {
 253                 /* Bad news -- userspace fed a bad pointer to a vsyscall. */
 254                 warn_bad_vsyscall(KERN_INFO, regs,
 255                                   "vsyscall fault (exploit attempt?)");
 256
 257                 /*
 258                  * If we failed to generate a signal for any reason,
 259                  * generate one here.  (This should be impossible.)
 260                  */
 261                 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
 262                                  !sigismember(&tsk->pending.signal, SIGSEGV)))
 263                         goto sigsegv;
 264
 265                 return true;  /* Don't emulate the ret. */
 266         }
 267
 268         regs->ax = ret;
 269
 270 do_ret:
 271         /* Emulate a ret instruction. */
 272         regs->ip = caller;
 273         regs->sp += 8;
 274         return true;
 275
 276 sigsegv:
 277         force_sig(SIGSEGV, current);
 278         return true;
 279 }
 280
 281 /*
 282  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 283  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 284  * not need special handling anymore:
 285  */
 286 static const char *gate_vma_name(struct vm_area_struct *vma)
 287 {
 288         return "[vsyscall]";
 289 }
 290 static const struct vm_operations_struct gate_vma_ops = {
 291         .name = gate_vma_name,
 292 };
 293 static struct vm_area_struct gate_vma = {
 294         .vm_start       = VSYSCALL_ADDR,
 295         .vm_end         = VSYSCALL_ADDR + PAGE_SIZE,
 296         .vm_page_prot   = PAGE_READONLY_EXEC,
 297         .vm_flags       = VM_READ | VM_EXEC,
 298         .vm_ops         = &gate_vma_ops,
 299 };
 300
 301 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
 302 {
 303 #ifdef CONFIG_COMPAT
 304         if (!mm || mm->context.ia32_compat)
 305                 return NULL;
 306 #endif
 307         if (vsyscall_mode == NONE)
 308                 return NULL;
 309         return &gate_vma;
 310 }
 311
 312 int in_gate_area(struct mm_struct *mm, unsigned long addr)
 313 {
 314         struct vm_area_struct *vma = get_gate_vma(mm);
 315
 316         if (!vma)
 317                 return 0;
 318
 319         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 320 }
 321
 322 /*
 323  * Use this when you have no reliable mm, typically from interrupt
 324  * context. It is less reliable than using a task's mm and may give
 325  * false positives.
 326  */
 327 int in_gate_area_no_mm(unsigned long addr)
 328 {
 329         return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR;
 330 }
 331
 332 void __init map_vsyscall(void)
 333 {
 334         extern char __vsyscall_page;
 335         unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
 336
 337         if (vsyscall_mode != NONE)
 338                 __set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall,
 339                              vsyscall_mode == NATIVE
 340                              ? PAGE_KERNEL_VSYSCALL
 341                              : PAGE_KERNEL_VVAR);
 342
 343         BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) !=
 344                      (unsigned long)VSYSCALL_ADDR);
 345 }