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 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
 144 {
 145         struct task_struct *tsk;
 146         unsigned long caller;
 147         int vsyscall_nr;
 148         long ret;
 149
 150         /*
 151          * No point in checking CS -- the only way to get here is a user mode
 152          * trap to a high address, which means that we're in 64-bit user code.
 153          */
 154
 155         WARN_ON_ONCE(address != regs->ip);
 156
 157         if (vsyscall_mode == NONE) {
 158                 warn_bad_vsyscall(KERN_INFO, regs,
 159                                   "vsyscall attempted with vsyscall=none");
 160                 return false;
 161         }
 162
 163         vsyscall_nr = addr_to_vsyscall_nr(address);
 164
 165         trace_emulate_vsyscall(vsyscall_nr);
 166
 167         if (vsyscall_nr < 0) {
 168                 warn_bad_vsyscall(KERN_WARNING, regs,
 169                                   "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
 170                 goto sigsegv;
 171         }
 172
 173         if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
 174                 warn_bad_vsyscall(KERN_WARNING, regs,
 175                                   "vsyscall with bad stack (exploit attempt?)");
 176                 goto sigsegv;
 177         }
 178
 179         tsk = current;
 180         if (seccomp_mode(&tsk->seccomp))
 181                 do_exit(SIGKILL);
 182
 183         switch (vsyscall_nr) {
 184         case 0:
 185                 ret = sys_gettimeofday(
 186                         (struct timeval __user *)regs->di,
 187                         (struct timezone __user *)regs->si);
 188                 break;
 189
 190         case 1:
 191                 ret = sys_time((time_t __user *)regs->di);
 192                 break;
 193
 194         case 2:
 195                 ret = sys_getcpu((unsigned __user *)regs->di,
 196                                  (unsigned __user *)regs->si,
 197                                  0);
 198                 break;
 199         }
 200
 201         if (ret == -EFAULT) {
 202                 /*
 203                  * Bad news -- userspace fed a bad pointer to a vsyscall.
 204                  *
 205                  * With a real vsyscall, that would have caused SIGSEGV.
 206                  * To make writing reliable exploits using the emulated
 207                  * vsyscalls harder, generate SIGSEGV here as well.
 208                  */
 209                 warn_bad_vsyscall(KERN_INFO, regs,
 210                                   "vsyscall fault (exploit attempt?)");
 211                 goto sigsegv;
 212         }
 213
 214         regs->ax = ret;
 215
 216         /* Emulate a ret instruction. */
 217         regs->ip = caller;
 218         regs->sp += 8;
 219
 220         return true;
 221
 222 sigsegv:
 223         force_sig(SIGSEGV, current);
 224         return true;
 225 }
 226
 227 /*
 228  * Assume __initcall executes before all user space. Hopefully kmod
 229  * doesn't violate that. We'll find out if it does.
 230  */
 231 static void __cpuinit vsyscall_set_cpu(int cpu)
 232 {
 233         unsigned long d;
 234         unsigned long node = 0;
 235 #ifdef CONFIG_NUMA
 236         node = cpu_to_node(cpu);
 237 #endif
 238         if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
 239                 write_rdtscp_aux((node << 12) | cpu);
 240
 241         /*
 242          * Store cpu number in limit so that it can be loaded quickly
 243          * in user space in vgetcpu. (12 bits for the CPU and 8 bits for the node)
 244          */
 245         d = 0x0f40000000000ULL;
 246         d |= cpu;
 247         d |= (node & 0xf) << 12;
 248         d |= (node >> 4) << 48;
 249
 250         write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
 251 }
 252
 253 static void __cpuinit cpu_vsyscall_init(void *arg)
 254 {
 255         /* preemption should be already off */
 256         vsyscall_set_cpu(raw_smp_processor_id());
 257 }
 258
 259 static int __cpuinit
 260 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
 261 {
 262         long cpu = (long)arg;
 263
 264         if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
 265                 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
 266
 267         return NOTIFY_DONE;
 268 }
 269
 270 void __init map_vsyscall(void)
 271 {
 272         extern char __vsyscall_page;
 273         unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
 274         extern char __vvar_page;
 275         unsigned long physaddr_vvar_page = __pa_symbol(&__vvar_page);
 276
 277         __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_vsyscall,
 278                      vsyscall_mode == NATIVE
 279                      ? PAGE_KERNEL_VSYSCALL
 280                      : PAGE_KERNEL_VVAR);
 281         BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_FIRST_PAGE) !=
 282                      (unsigned long)VSYSCALL_START);
 283
 284         __set_fixmap(VVAR_PAGE, physaddr_vvar_page, PAGE_KERNEL_VVAR);
 285         BUILD_BUG_ON((unsigned long)__fix_to_virt(VVAR_PAGE) !=
 286                      (unsigned long)VVAR_ADDRESS);
 287 }
 288
 289 static int __init vsyscall_init(void)
 290 {
 291         BUG_ON(VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE));
 292
 293         on_each_cpu(cpu_vsyscall_init, NULL, 1);
 294         /* notifier priority > KVM */
 295         hotcpu_notifier(cpu_vsyscall_notifier, 30);
 296
 297         return 0;
 298 }
 299 __initcall(vsyscall_init);