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