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 static enum { EMULATE, NATIVE, NONE } vsyscall_mode = EMULATE;
  58
  59 static int __init vsyscall_setup(char *str)
  60 {
  61         if (str) {
  62                 if (!strcmp("emulate", str))
  63                         vsyscall_mode = EMULATE;
  64                 else if (!strcmp("native", str))
  65                         vsyscall_mode = NATIVE;
  66                 else if (!strcmp("none", str))
  67                         vsyscall_mode = NONE;
  68                 else
  69                         return -EINVAL;
  70
  71                 return 0;
  72         }
  73
  74         return -EINVAL;
  75 }
  76 early_param("vsyscall", vsyscall_setup);
  77
  78 void update_vsyscall_tz(void)
  79 {
  80         vsyscall_gtod_data.sys_tz = sys_tz;
  81 }
  82
  83 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
  84                         struct clocksource *clock, u32 mult)
  85 {
  86         struct timespec monotonic;
  87
  88         write_seqcount_begin(&vsyscall_gtod_data.seq);
  89
  90         /* copy vsyscall data */
  91         vsyscall_gtod_data.clock.vclock_mode    = clock->archdata.vclock_mode;
  92         vsyscall_gtod_data.clock.cycle_last     = clock->cycle_last;
  93         vsyscall_gtod_data.clock.mask           = clock->mask;
  94         vsyscall_gtod_data.clock.mult           = mult;
  95         vsyscall_gtod_data.clock.shift          = clock->shift;
  96
  97         vsyscall_gtod_data.wall_time_sec        = wall_time->tv_sec;
  98         vsyscall_gtod_data.wall_time_nsec       = wall_time->tv_nsec;
  99
 100         monotonic = timespec_add(*wall_time, *wtm);
 101         vsyscall_gtod_data.monotonic_time_sec   = monotonic.tv_sec;
 102         vsyscall_gtod_data.monotonic_time_nsec  = monotonic.tv_nsec;
 103
 104         vsyscall_gtod_data.wall_time_coarse     = __current_kernel_time();
 105         vsyscall_gtod_data.monotonic_time_coarse =
 106                 timespec_add(vsyscall_gtod_data.wall_time_coarse, *wtm);
 107
 108         write_seqcount_end(&vsyscall_gtod_data.seq);
 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 #ifdef CONFIG_SECCOMP
 143 static int vsyscall_seccomp(struct task_struct *tsk, int syscall_nr)
 144 {
 145         if (!seccomp_mode(&tsk->seccomp))
 146                 return 0;
 147         task_pt_regs(tsk)->orig_ax = syscall_nr;
 148         task_pt_regs(tsk)->ax = syscall_nr;
 149         return __secure_computing(syscall_nr);
 150 }
 151 #else
 152 #define vsyscall_seccomp(_tsk, _nr) 0
 153 #endif
 154
 155 static bool write_ok_or_segv(unsigned long ptr, size_t size)
 156 {
 157         /*
 158          * XXX: if access_ok, get_user, and put_user handled
 159          * sig_on_uaccess_error, this could go away.
 160          */
 161
 162         if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
 163                 siginfo_t info;
 164                 struct thread_struct *thread = &current->thread;
 165
 166                 thread->error_code      = 6;  /* user fault, no page, write */
 167                 thread->cr2             = ptr;
 168                 thread->trap_nr         = X86_TRAP_PF;
 169
 170                 memset(&info, 0, sizeof(info));
 171                 info.si_signo           = SIGSEGV;
 172                 info.si_errno           = 0;
 173                 info.si_code            = SEGV_MAPERR;
 174                 info.si_addr            = (void __user *)ptr;
 175
 176                 force_sig_info(SIGSEGV, &info, current);
 177                 return false;
 178         } else {
 179                 return true;
 180         }
 181 }
 182
 183 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
 184 {
 185         struct task_struct *tsk;
 186         unsigned long caller;
 187         int vsyscall_nr;
 188         int prev_sig_on_uaccess_error;
 189         long ret;
 190         int skip;
 191
 192         /*
 193          * No point in checking CS -- the only way to get here is a user mode
 194          * trap to a high address, which means that we're in 64-bit user code.
 195          */
 196
 197         WARN_ON_ONCE(address != regs->ip);
 198
 199         if (vsyscall_mode == NONE) {
 200                 warn_bad_vsyscall(KERN_INFO, regs,
 201                                   "vsyscall attempted with vsyscall=none");
 202                 return false;
 203         }
 204
 205         vsyscall_nr = addr_to_vsyscall_nr(address);
 206
 207         trace_emulate_vsyscall(vsyscall_nr);
 208
 209         if (vsyscall_nr < 0) {
 210                 warn_bad_vsyscall(KERN_WARNING, regs,
 211                                   "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
 212                 goto sigsegv;
 213         }
 214
 215         if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
 216                 warn_bad_vsyscall(KERN_WARNING, regs,
 217                                   "vsyscall with bad stack (exploit attempt?)");
 218                 goto sigsegv;
 219         }
 220
 221         tsk = current;
 222         /*
 223          * With a real vsyscall, page faults cause SIGSEGV.  We want to
 224          * preserve that behavior to make writing exploits harder.
 225          */
 226         prev_sig_on_uaccess_error = current_thread_info()->sig_on_uaccess_error;
 227         current_thread_info()->sig_on_uaccess_error = 1;
 228
 229         /*
 230          * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
 231          * 64-bit, so we don't need to special-case it here.  For all the
 232          * vsyscalls, NULL means "don't write anything" not "write it at
 233          * address 0".
 234          */
 235         ret = -EFAULT;
 236         skip = 0;
 237         switch (vsyscall_nr) {
 238         case 0:
 239                 skip = vsyscall_seccomp(tsk, __NR_gettimeofday);
 240                 if (skip)
 241                         break;
 242
 243                 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
 244                     !write_ok_or_segv(regs->si, sizeof(struct timezone)))
 245                         break;
 246
 247                 ret = sys_gettimeofday(
 248                         (struct timeval __user *)regs->di,
 249                         (struct timezone __user *)regs->si);
 250                 break;
 251
 252         case 1:
 253                 skip = vsyscall_seccomp(tsk, __NR_time);
 254                 if (skip)
 255                         break;
 256
 257                 if (!write_ok_or_segv(regs->di, sizeof(time_t)))
 258                         break;
 259
 260                 ret = sys_time((time_t __user *)regs->di);
 261                 break;
 262
 263         case 2:
 264                 skip = vsyscall_seccomp(tsk, __NR_getcpu);
 265                 if (skip)
 266                         break;
 267
 268                 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
 269                     !write_ok_or_segv(regs->si, sizeof(unsigned)))
 270                         break;
 271
 272                 ret = sys_getcpu((unsigned __user *)regs->di,
 273                                  (unsigned __user *)regs->si,
 274                                  NULL);
 275                 break;
 276         }
 277
 278         current_thread_info()->sig_on_uaccess_error = prev_sig_on_uaccess_error;
 279
 280         if (skip) {
 281                 if ((long)regs->ax <= 0L) /* seccomp errno emulation */
 282                         goto do_ret;
 283                 goto done; /* seccomp trace/trap */
 284         }
 285
 286         if (ret == -EFAULT) {
 287                 /* Bad news -- userspace fed a bad pointer to a vsyscall. */
 288                 warn_bad_vsyscall(KERN_INFO, regs,
 289                                   "vsyscall fault (exploit attempt?)");
 290
 291                 /*
 292                  * If we failed to generate a signal for any reason,
 293                  * generate one here.  (This should be impossible.)
 294                  */
 295                 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
 296                                  !sigismember(&tsk->pending.signal, SIGSEGV)))
 297                         goto sigsegv;
 298
 299                 return true;  /* Don't emulate the ret. */
 300         }
 301
 302         regs->ax = ret;
 303
 304 do_ret:
 305         /* Emulate a ret instruction. */
 306         regs->ip = caller;
 307         regs->sp += 8;
 308 done:
 309         return true;
 310
 311 sigsegv:
 312         force_sig(SIGSEGV, current);
 313         return true;
 314 }
 315
 316 /*
 317  * Assume __initcall executes before all user space. Hopefully kmod
 318  * doesn't violate that. We'll find out if it does.
 319  */
 320 static void __cpuinit vsyscall_set_cpu(int cpu)
 321 {
 322         unsigned long d;
 323         unsigned long node = 0;
 324 #ifdef CONFIG_NUMA
 325         node = cpu_to_node(cpu);
 326 #endif
 327         if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
 328                 write_rdtscp_aux((node << 12) | cpu);
 329
 330         /*
 331          * Store cpu number in limit so that it can be loaded quickly
 332          * in user space in vgetcpu. (12 bits for the CPU and 8 bits for the node)
 333          */
 334         d = 0x0f40000000000ULL;
 335         d |= cpu;
 336         d |= (node & 0xf) << 12;
 337         d |= (node >> 4) << 48;
 338
 339         write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
 340 }
 341
 342 static void __cpuinit cpu_vsyscall_init(void *arg)
 343 {
 344         /* preemption should be already off */
 345         vsyscall_set_cpu(raw_smp_processor_id());
 346 }
 347
 348 static int __cpuinit
 349 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
 350 {
 351         long cpu = (long)arg;
 352
 353         if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
 354                 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
 355
 356         return NOTIFY_DONE;
 357 }
 358
 359 void __init map_vsyscall(void)
 360 {
 361         extern char __vsyscall_page;
 362         unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
 363         extern char __vvar_page;
 364         unsigned long physaddr_vvar_page = __pa_symbol(&__vvar_page);
 365
 366         __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_vsyscall,
 367                      vsyscall_mode == NATIVE
 368                      ? PAGE_KERNEL_VSYSCALL
 369                      : PAGE_KERNEL_VVAR);
 370         BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_FIRST_PAGE) !=
 371                      (unsigned long)VSYSCALL_START);
 372
 373         __set_fixmap(VVAR_PAGE, physaddr_vvar_page, PAGE_KERNEL_VVAR);
 374         BUILD_BUG_ON((unsigned long)__fix_to_virt(VVAR_PAGE) !=
 375                      (unsigned long)VVAR_ADDRESS);
 376 }
 377
 378 static int __init vsyscall_init(void)
 379 {
 380         BUG_ON(VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE));
 381
 382         on_each_cpu(cpu_vsyscall_init, NULL, 1);
 383         /* notifier priority > KVM */
 384         hotcpu_notifier(cpu_vsyscall_notifier, 30);
 385
 386         return 0;
 387 }
 388 __initcall(vsyscall_init);