arch/x86/kernel/vsyscall_64.c

   1 /*
   2  *  Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
   3  *  Copyright 2003 Andi Kleen, SuSE Labs.
   4  *
   5  *  [ NOTE: this mechanism is now deprecated in favor of the vDSO. ]
   6  *
   7  *  Thanks to hpa@transmeta.com for some useful hint.
   8  *  Special thanks to Ingo Molnar for his early experience with
   9  *  a different vsyscall implementation for Linux/IA32 and for the name.
  10  *
  11  *  vsyscall 1 is located at -10Mbyte, vsyscall 2 is located
  12  *  at virtual address -10Mbyte+1024bytes etc... There are at max 4
  13  *  vsyscalls. One vsyscall can reserve more than 1 slot to avoid
  14  *  jumping out of line if necessary. We cannot add more with this
  15  *  mechanism because older kernels won't return -ENOSYS.
  16  *
  17  *  Note: the concept clashes with user mode linux.  UML users should
  18  *  use the vDSO.
  19  */
  20
  21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  22
  23 #include <linux/time.h>
  24 #include <linux/init.h>
  25 #include <linux/kernel.h>
  26 #include <linux/timer.h>
  27 #include <linux/seqlock.h>
  28 #include <linux/jiffies.h>
  29 #include <linux/sysctl.h>
  30 #include <linux/topology.h>
  31 #include <linux/timekeeper_internal.h>
  32 #include <linux/getcpu.h>
  33 #include <linux/cpu.h>
  34 #include <linux/smp.h>
  35 #include <linux/notifier.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/ratelimit.h>
  38
  39 #include <asm/vsyscall.h>
  40 #include <asm/pgtable.h>
  41 #include <asm/compat.h>
  42 #include <asm/page.h>
  43 #include <asm/unistd.h>
  44 #include <asm/fixmap.h>
  45 #include <asm/errno.h>
  46 #include <asm/io.h>
  47 #include <asm/segment.h>
  48 #include <asm/desc.h>
  49 #include <asm/topology.h>
  50 #include <asm/traps.h>
  51
  52 #define CREATE_TRACE_POINTS
  53 #include "vsyscall_trace.h"
  54
  55 static enum { EMULATE, NATIVE, NONE } vsyscall_mode = EMULATE;
  56
  57 static int __init vsyscall_setup(char *str)
  58 {
  59         if (str) {
  60                 if (!strcmp("emulate", str))
  61                         vsyscall_mode = EMULATE;
  62                 else if (!strcmp("native", str))
  63                         vsyscall_mode = NATIVE;
  64                 else if (!strcmp("none", str))
  65                         vsyscall_mode = NONE;
  66                 else
  67                         return -EINVAL;
  68
  69                 return 0;
  70         }
  71
  72         return -EINVAL;
  73 }
  74 early_param("vsyscall", vsyscall_setup);
  75
  76 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
  77                               const char *message)
  78 {
  79         if (!show_unhandled_signals)
  80                 return;
  81
  82         printk_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
  83                            level, current->comm, task_pid_nr(current),
  84                            message, regs->ip, regs->cs,
  85                            regs->sp, regs->ax, regs->si, regs->di);
  86 }
  87
  88 static int addr_to_vsyscall_nr(unsigned long addr)
  89 {
  90         int nr;
  91
  92         if ((addr & ~0xC00UL) != VSYSCALL_ADDR)
  93                 return -EINVAL;
  94
  95         nr = (addr & 0xC00UL) >> 10;
  96         if (nr >= 3)
  97                 return -EINVAL;
  98
  99         return nr;
 100 }
 101
 102 static bool write_ok_or_segv(unsigned long ptr, size_t size)
 103 {
 104         /*
 105          * XXX: if access_ok, get_user, and put_user handled
 106          * sig_on_uaccess_error, this could go away.
 107          */
 108
 109         if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
 110                 siginfo_t info;
 111                 struct thread_struct *thread = &current->thread;
 112
 113                 thread->error_code      = 6;  /* user fault, no page, write */
 114                 thread->cr2             = ptr;
 115                 thread->trap_nr         = X86_TRAP_PF;
 116
 117                 memset(&info, 0, sizeof(info));
 118                 info.si_signo           = SIGSEGV;
 119                 info.si_errno           = 0;
 120                 info.si_code            = SEGV_MAPERR;
 121                 info.si_addr            = (void __user *)ptr;
 122
 123                 force_sig_info(SIGSEGV, &info, current);
 124                 return false;
 125         } else {
 126                 return true;
 127         }
 128 }
 129
 130 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
 131 {
 132         struct task_struct *tsk;
 133         unsigned long caller;
 134         int vsyscall_nr, syscall_nr, tmp;
 135         int prev_sig_on_uaccess_error;
 136         long ret;
 137
 138         /*
 139          * No point in checking CS -- the only way to get here is a user mode
 140          * trap to a high address, which means that we're in 64-bit user code.
 141          */
 142
 143         WARN_ON_ONCE(address != regs->ip);
 144
 145         if (vsyscall_mode == NONE) {
 146                 warn_bad_vsyscall(KERN_INFO, regs,
 147                                   "vsyscall attempted with vsyscall=none");
 148                 return false;
 149         }
 150
 151         vsyscall_nr = addr_to_vsyscall_nr(address);
 152
 153         trace_emulate_vsyscall(vsyscall_nr);
 154
 155         if (vsyscall_nr < 0) {
 156                 warn_bad_vsyscall(KERN_WARNING, regs,
 157                                   "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
 158                 goto sigsegv;
 159         }
 160
 161         if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
 162                 warn_bad_vsyscall(KERN_WARNING, regs,
 163                                   "vsyscall with bad stack (exploit attempt?)");
 164                 goto sigsegv;
 165         }
 166
 167         tsk = current;
 168
 169         /*
 170          * Check for access_ok violations and find the syscall nr.
 171          *
 172          * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
 173          * 64-bit, so we don't need to special-case it here.  For all the
 174          * vsyscalls, NULL means "don't write anything" not "write it at
 175          * address 0".
 176          */
 177         switch (vsyscall_nr) {
 178         case 0:
 179                 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
 180                     !write_ok_or_segv(regs->si, sizeof(struct timezone))) {
 181                         ret = -EFAULT;
 182                         goto check_fault;
 183                 }
 184
 185                 syscall_nr = __NR_gettimeofday;
 186                 break;
 187
 188         case 1:
 189                 if (!write_ok_or_segv(regs->di, sizeof(time_t))) {
 190                         ret = -EFAULT;
 191                         goto check_fault;
 192                 }
 193
 194                 syscall_nr = __NR_time;
 195                 break;
 196
 197         case 2:
 198                 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
 199                     !write_ok_or_segv(regs->si, sizeof(unsigned))) {
 200                         ret = -EFAULT;
 201                         goto check_fault;
 202                 }
 203
 204                 syscall_nr = __NR_getcpu;
 205                 break;
 206         }
 207
 208         /*
 209          * Handle seccomp.  regs->ip must be the original value.
 210          * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt.
 211          *
 212          * We could optimize the seccomp disabled case, but performance
 213          * here doesn't matter.
 214          */
 215         regs->orig_ax = syscall_nr;
 216         regs->ax = -ENOSYS;
 217         tmp = secure_computing();
 218         if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) {
 219                 warn_bad_vsyscall(KERN_DEBUG, regs,
 220                                   "seccomp tried to change syscall nr or ip");
 221                 do_exit(SIGSYS);
 222         }
 223         if (tmp)
 224                 goto do_ret;  /* skip requested */
 225
 226         /*
 227          * With a real vsyscall, page faults cause SIGSEGV.  We want to
 228          * preserve that behavior to make writing exploits harder.
 229          */
 230         prev_sig_on_uaccess_error = current_thread_info()->sig_on_uaccess_error;
 231         current_thread_info()->sig_on_uaccess_error = 1;
 232
 233         ret = -EFAULT;
 234         switch (vsyscall_nr) {
 235         case 0:
 236                 ret = sys_gettimeofday(
 237                         (struct timeval __user *)regs->di,
 238                         (struct timezone __user *)regs->si);
 239                 break;
 240
 241         case 1:
 242                 ret = sys_time((time_t __user *)regs->di);
 243                 break;
 244
 245         case 2:
 246                 ret = sys_getcpu((unsigned __user *)regs->di,
 247                                  (unsigned __user *)regs->si,
 248                                  NULL);
 249                 break;
 250         }
 251
 252         current_thread_info()->sig_on_uaccess_error = prev_sig_on_uaccess_error;
 253
 254 check_fault:
 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?)");
 259
 260                 /*
 261                  * If we failed to generate a signal for any reason,
 262                  * generate one here.  (This should be impossible.)
 263                  */
 264                 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
 265                                  !sigismember(&tsk->pending.signal, SIGSEGV)))
 266                         goto sigsegv;
 267
 268                 return true;  /* Don't emulate the ret. */
 269         }
 270
 271         regs->ax = ret;
 272
 273 do_ret:
 274         /* Emulate a ret instruction. */
 275         regs->ip = caller;
 276         regs->sp += 8;
 277         return true;
 278
 279 sigsegv:
 280         force_sig(SIGSEGV, current);
 281         return true;
 282 }
 283
 284 /*
 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:
 288  */
 289 static const char *gate_vma_name(struct vm_area_struct *vma)
 290 {
 291         return "[vsyscall]";
 292 }
 293 static struct vm_operations_struct gate_vma_ops = {
 294         .name = gate_vma_name,
 295 };
 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,
 302 };
 303
 304 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
 305 {
 306 #ifdef CONFIG_IA32_EMULATION
 307         if (!mm || mm->context.ia32_compat)
 308                 return NULL;
 309 #endif
 310         if (vsyscall_mode == NONE)
 311                 return NULL;
 312         return &gate_vma;
 313 }
 314
 315 int in_gate_area(struct mm_struct *mm, unsigned long addr)
 316 {
 317         struct vm_area_struct *vma = get_gate_vma(mm);
 318
 319         if (!vma)
 320                 return 0;
 321
 322         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 323 }
 324
 325 /*
 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
 328  * false positives.
 329  */
 330 int in_gate_area_no_mm(unsigned long addr)
 331 {
 332         return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR;
 333 }
 334
 335 void __init map_vsyscall(void)
 336 {
 337         extern char __vsyscall_page;
 338         unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
 339
 340         if (vsyscall_mode != NONE)
 341                 __set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall,
 342                              vsyscall_mode == NATIVE
 343                              ? PAGE_KERNEL_VSYSCALL
 344                              : PAGE_KERNEL_VVAR);
 345
 346         BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) !=
 347                      (unsigned long)VSYSCALL_ADDR);
 348 }