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 #include <linux/time.h>
  22 #include <linux/init.h>
  23 #include <linux/kernel.h>
  24 #include <linux/timer.h>
  25 #include <linux/seqlock.h>
  26 #include <linux/jiffies.h>
  27 #include <linux/sysctl.h>
  28 #include <linux/topology.h>
  29 #include <linux/clocksource.h>
  30 #include <linux/getcpu.h>
  31 #include <linux/cpu.h>
  32 #include <linux/smp.h>
  33 #include <linux/notifier.h>
  34 #include <linux/syscalls.h>
  35 #include <linux/ratelimit.h>
  36
  37 #include <asm/vsyscall.h>
  38 #include <asm/pgtable.h>
  39 #include <asm/compat.h>
  40 #include <asm/page.h>
  41 #include <asm/unistd.h>
  42 #include <asm/fixmap.h>
  43 #include <asm/errno.h>
  44 #include <asm/io.h>
  45 #include <asm/segment.h>
  46 #include <asm/desc.h>
  47 #include <asm/topology.h>
  48 #include <asm/vgtod.h>
  49 #include <asm/traps.h>
  50
  51 #define CREATE_TRACE_POINTS
  52 #include "vsyscall_trace.h"
  53
  54 DEFINE_VVAR(int, vgetcpu_mode);
  55 DEFINE_VVAR(struct vsyscall_gtod_data, vsyscall_gtod_data) =
  56 {
  57         .lock = __SEQLOCK_UNLOCKED(__vsyscall_gtod_data.lock),
  58 };
  59
  60 static enum { EMULATE, NATIVE, NONE } vsyscall_mode = EMULATE;
  61
  62 static int __init vsyscall_setup(char *str)
  63 {
  64         if (str) {
  65                 if (!strcmp("emulate", str))
  66                         vsyscall_mode = EMULATE;
  67                 else if (!strcmp("native", str))
  68                         vsyscall_mode = NATIVE;
  69                 else if (!strcmp("none", str))
  70                         vsyscall_mode = NONE;
  71                 else
  72                         return -EINVAL;
  73
  74                 return 0;
  75         }
  76
  77         return -EINVAL;
  78 }
  79 early_param("vsyscall", vsyscall_setup);
  80
  81 void update_vsyscall_tz(void)
  82 {
  83         unsigned long flags;
  84
  85         write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
  86         /* sys_tz has changed */
  87         vsyscall_gtod_data.sys_tz = sys_tz;
  88         write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
  89 }
  90
  91 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
  92                         struct clocksource *clock, u32 mult)
  93 {
  94         unsigned long flags;
  95
  96         write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
  97
  98         /* copy vsyscall data */
  99         vsyscall_gtod_data.clock.vclock_mode    = clock->archdata.vclock_mode;
 100         vsyscall_gtod_data.clock.cycle_last     = clock->cycle_last;
 101         vsyscall_gtod_data.clock.mask           = clock->mask;
 102         vsyscall_gtod_data.clock.mult           = mult;
 103         vsyscall_gtod_data.clock.shift          = clock->shift;
 104         vsyscall_gtod_data.wall_time_sec        = wall_time->tv_sec;
 105         vsyscall_gtod_data.wall_time_nsec       = wall_time->tv_nsec;
 106         vsyscall_gtod_data.wall_to_monotonic    = *wtm;
 107         vsyscall_gtod_data.wall_time_coarse     = __current_kernel_time();
 108
 109         write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
 110 }
 111
 112 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
 113                               const char *message)
 114 {
 115         static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
 116         struct task_struct *tsk;
 117
 118         if (!show_unhandled_signals || !__ratelimit(&rs))
 119                 return;
 120
 121         tsk = current;
 122
 123         printk("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
 124                level, tsk->comm, task_pid_nr(tsk),
 125                message, regs->ip, regs->cs,
 126                regs->sp, regs->ax, regs->si, regs->di);
 127 }
 128
 129 static int addr_to_vsyscall_nr(unsigned long addr)
 130 {
 131         int nr;
 132
 133         if ((addr & ~0xC00UL) != VSYSCALL_START)
 134                 return -EINVAL;
 135
 136         nr = (addr & 0xC00UL) >> 10;
 137         if (nr >= 3)
 138                 return -EINVAL;
 139
 140         return nr;
 141 }
 142
 143 static bool write_ok_or_segv(unsigned long ptr, size_t size)
 144 {
 145         /*
 146          * XXX: if access_ok, get_user, and put_user handled
 147          * sig_on_uaccess_error, this could go away.
 148          */
 149
 150         if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
 151                 siginfo_t info;
 152                 struct thread_struct *thread = &current->thread;
 153
 154                 thread->error_code      = 6;  /* user fault, no page, write */
 155                 thread->cr2             = ptr;
 156                 thread->trap_no         = 14;
 157
 158                 memset(&info, 0, sizeof(info));
 159                 info.si_signo           = SIGSEGV;
 160                 info.si_errno           = 0;
 161                 info.si_code            = SEGV_MAPERR;
 162                 info.si_addr            = (void __user *)ptr;
 163
 164                 force_sig_info(SIGSEGV, &info, current);
 165                 return false;
 166         } else {
 167                 return true;
 168         }
 169 }
 170
 171 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
 172 {
 173         struct task_struct *tsk;
 174         unsigned long caller;
 175         int vsyscall_nr;
 176         int prev_sig_on_uaccess_error;
 177         long ret;
 178
 179         /*
 180          * No point in checking CS -- the only way to get here is a user mode
 181          * trap to a high address, which means that we're in 64-bit user code.
 182          */
 183
 184         WARN_ON_ONCE(address != regs->ip);
 185
 186         if (vsyscall_mode == NONE) {
 187                 warn_bad_vsyscall(KERN_INFO, regs,
 188                                   "vsyscall attempted with vsyscall=none");
 189                 return false;
 190         }
 191
 192         vsyscall_nr = addr_to_vsyscall_nr(address);
 193
 194         trace_emulate_vsyscall(vsyscall_nr);
 195
 196         if (vsyscall_nr < 0) {
 197                 warn_bad_vsyscall(KERN_WARNING, regs,
 198                                   "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
 199                 goto sigsegv;
 200         }
 201
 202         if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
 203                 warn_bad_vsyscall(KERN_WARNING, regs,
 204                                   "vsyscall with bad stack (exploit attempt?)");
 205                 goto sigsegv;
 206         }
 207
 208         tsk = current;
 209         if (seccomp_mode(&tsk->seccomp))
 210                 do_exit(SIGKILL);
 211
 212         /*
 213          * With a real vsyscall, page faults cause SIGSEGV.  We want to
 214          * preserve that behavior to make writing exploits harder.
 215          */
 216         prev_sig_on_uaccess_error = current_thread_info()->sig_on_uaccess_error;
 217         current_thread_info()->sig_on_uaccess_error = 1;
 218
 219         /*
 220          * 0 is a valid user pointer (in the access_ok sense) on 32-bit and
 221          * 64-bit, so we don't need to special-case it here.  For all the
 222          * vsyscalls, 0 means "don't write anything" not "write it at
 223          * address 0".
 224          */
 225         ret = -EFAULT;
 226         switch (vsyscall_nr) {
 227         case 0:
 228                 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
 229                     !write_ok_or_segv(regs->si, sizeof(struct timezone)))
 230                         break;
 231
 232                 ret = sys_gettimeofday(
 233                         (struct timeval __user *)regs->di,
 234                         (struct timezone __user *)regs->si);
 235                 break;
 236
 237         case 1:
 238                 if (!write_ok_or_segv(regs->di, sizeof(time_t)))
 239                         break;
 240
 241                 ret = sys_time((time_t __user *)regs->di);
 242                 break;
 243
 244         case 2:
 245                 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
 246                     !write_ok_or_segv(regs->si, sizeof(unsigned)))
 247                         break;
 248
 249                 ret = sys_getcpu((unsigned __user *)regs->di,
 250                                  (unsigned __user *)regs->si,
 251                                  0);
 252                 break;
 253         }
 254
 255         current_thread_info()->sig_on_uaccess_error = prev_sig_on_uaccess_error;
 256
 257         if (ret == -EFAULT) {
 258                 /* Bad news -- userspace fed a bad pointer to a vsyscall. */
 259                 warn_bad_vsyscall(KERN_INFO, regs,
 260                                   "vsyscall fault (exploit attempt?)");
 261
 262                 /*
 263                  * If we failed to generate a signal for any reason,
 264                  * generate one here.  (This should be impossible.)
 265                  */
 266                 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
 267                                  !sigismember(&tsk->pending.signal, SIGSEGV)))
 268                         goto sigsegv;
 269
 270                 return true;  /* Don't emulate the ret. */
 271         }
 272
 273         regs->ax = ret;
 274
 275         /* Emulate a ret instruction. */
 276         regs->ip = caller;
 277         regs->sp += 8;
 278
 279         return true;
 280
 281 sigsegv:
 282         force_sig(SIGSEGV, current);
 283         return true;
 284 }
 285
 286 /*
 287  * Assume __initcall executes before all user space. Hopefully kmod
 288  * doesn't violate that. We'll find out if it does.
 289  */
 290 static void __cpuinit vsyscall_set_cpu(int cpu)
 291 {
 292         unsigned long d;
 293         unsigned long node = 0;
 294 #ifdef CONFIG_NUMA
 295         node = cpu_to_node(cpu);
 296 #endif
 297         if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
 298                 write_rdtscp_aux((node << 12) | cpu);
 299
 300         /*
 301          * Store cpu number in limit so that it can be loaded quickly
 302          * in user space in vgetcpu. (12 bits for the CPU and 8 bits for the node)
 303          */
 304         d = 0x0f40000000000ULL;
 305         d |= cpu;
 306         d |= (node & 0xf) << 12;
 307         d |= (node >> 4) << 48;
 308
 309         write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
 310 }
 311
 312 static void __cpuinit cpu_vsyscall_init(void *arg)
 313 {
 314         /* preemption should be already off */
 315         vsyscall_set_cpu(raw_smp_processor_id());
 316 }
 317
 318 static int __cpuinit
 319 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
 320 {
 321         long cpu = (long)arg;
 322
 323         if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
 324                 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
 325
 326         return NOTIFY_DONE;
 327 }
 328
 329 void __init map_vsyscall(void)
 330 {
 331         extern char __vsyscall_page;
 332         unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
 333         extern char __vvar_page;
 334         unsigned long physaddr_vvar_page = __pa_symbol(&__vvar_page);
 335
 336         __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_vsyscall,
 337                      vsyscall_mode == NATIVE
 338                      ? PAGE_KERNEL_VSYSCALL
 339                      : PAGE_KERNEL_VVAR);
 340         BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_FIRST_PAGE) !=
 341                      (unsigned long)VSYSCALL_START);
 342
 343         __set_fixmap(VVAR_PAGE, physaddr_vvar_page, PAGE_KERNEL_VVAR);
 344         BUILD_BUG_ON((unsigned long)__fix_to_virt(VVAR_PAGE) !=
 345                      (unsigned long)VVAR_ADDRESS);
 346 }
 347
 348 static int __init vsyscall_init(void)
 349 {
 350         BUG_ON(VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE));
 351
 352         on_each_cpu(cpu_vsyscall_init, NULL, 1);
 353         /* notifier priority > KVM */
 354         hotcpu_notifier(cpu_vsyscall_notifier, 30);
 355
 356         return 0;
 357 }
 358 __initcall(vsyscall_init);