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/clocksource.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/vgtod.h>
  51 #include <asm/traps.h>
  52
  53 #define CREATE_TRACE_POINTS
  54 #include "vsyscall_trace.h"
  55
  56 DEFINE_VVAR(int, vgetcpu_mode);
  57 DEFINE_VVAR(struct vsyscall_gtod_data, vsyscall_gtod_data);
  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         vsyscall_gtod_data.sys_tz = sys_tz;
  83 }
  84
  85 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
  86                         struct clocksource *clock, u32 mult)
  87 {
  88         struct timespec monotonic;
  89
  90         write_seqcount_begin(&vsyscall_gtod_data.seq);
  91
  92         /* copy vsyscall data */
  93         vsyscall_gtod_data.clock.vclock_mode    = clock->archdata.vclock_mode;
  94         vsyscall_gtod_data.clock.cycle_last     = clock->cycle_last;
  95         vsyscall_gtod_data.clock.mask           = clock->mask;
  96         vsyscall_gtod_data.clock.mult           = mult;
  97         vsyscall_gtod_data.clock.shift          = clock->shift;
  98
  99         vsyscall_gtod_data.wall_time_sec        = wall_time->tv_sec;
 100         vsyscall_gtod_data.wall_time_nsec       = wall_time->tv_nsec;
 101
 102         monotonic = timespec_add(*wall_time, *wtm);
 103         vsyscall_gtod_data.monotonic_time_sec   = monotonic.tv_sec;
 104         vsyscall_gtod_data.monotonic_time_nsec  = monotonic.tv_nsec;
 105
 106         vsyscall_gtod_data.wall_time_coarse     = __current_kernel_time();
 107         vsyscall_gtod_data.monotonic_time_coarse =
 108                 timespec_add(vsyscall_gtod_data.wall_time_coarse, *wtm);
 109
 110         write_seqcount_end(&vsyscall_gtod_data.seq);
 111 }
 112
 113 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
 114                               const char *message)
 115 {
 116         if (!show_unhandled_signals)
 117                 return;
 118
 119         pr_notice_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
 120                               level, current->comm, task_pid_nr(current),
 121                               message, regs->ip, regs->cs,
 122                               regs->sp, regs->ax, regs->si, regs->di);
 123 }
 124
 125 static int addr_to_vsyscall_nr(unsigned long addr)
 126 {
 127         int nr;
 128
 129         if ((addr & ~0xC00UL) != VSYSCALL_START)
 130                 return -EINVAL;
 131
 132         nr = (addr & 0xC00UL) >> 10;
 133         if (nr >= 3)
 134                 return -EINVAL;
 135
 136         return nr;
 137 }
 138
 139 #ifdef CONFIG_SECCOMP
 140 static int vsyscall_seccomp(struct task_struct *tsk, int syscall_nr)
 141 {
 142         if (!seccomp_mode(&tsk->seccomp))
 143                 return 0;
 144         task_pt_regs(tsk)->orig_ax = syscall_nr;
 145         task_pt_regs(tsk)->ax = syscall_nr;
 146         return __secure_computing(syscall_nr);
 147 }
 148 #else
 149 #define vsyscall_seccomp(_tsk, _nr) 0
 150 #endif
 151
 152 static bool write_ok_or_segv(unsigned long ptr, size_t size)
 153 {
 154         /*
 155          * XXX: if access_ok, get_user, and put_user handled
 156          * sig_on_uaccess_error, this could go away.
 157          */
 158
 159         if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
 160                 siginfo_t info;
 161                 struct thread_struct *thread = &current->thread;
 162
 163                 thread->error_code      = 6;  /* user fault, no page, write */
 164                 thread->cr2             = ptr;
 165                 thread->trap_nr         = X86_TRAP_PF;
 166
 167                 memset(&info, 0, sizeof(info));
 168                 info.si_signo           = SIGSEGV;
 169                 info.si_errno           = 0;
 170                 info.si_code            = SEGV_MAPERR;
 171                 info.si_addr            = (void __user *)ptr;
 172
 173                 force_sig_info(SIGSEGV, &info, current);
 174                 return false;
 175         } else {
 176                 return true;
 177         }
 178 }
 179
 180 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
 181 {
 182         struct task_struct *tsk;
 183         unsigned long caller;
 184         int vsyscall_nr;
 185         int prev_sig_on_uaccess_error;
 186         long ret;
 187         int skip;
 188
 189         /*
 190          * No point in checking CS -- the only way to get here is a user mode
 191          * trap to a high address, which means that we're in 64-bit user code.
 192          */
 193
 194         WARN_ON_ONCE(address != regs->ip);
 195
 196         if (vsyscall_mode == NONE) {
 197                 warn_bad_vsyscall(KERN_INFO, regs,
 198                                   "vsyscall attempted with vsyscall=none");
 199                 return false;
 200         }
 201
 202         vsyscall_nr = addr_to_vsyscall_nr(address);
 203
 204         trace_emulate_vsyscall(vsyscall_nr);
 205
 206         if (vsyscall_nr < 0) {
 207                 warn_bad_vsyscall(KERN_WARNING, regs,
 208                                   "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
 209                 goto sigsegv;
 210         }
 211
 212         if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
 213                 warn_bad_vsyscall(KERN_WARNING, regs,
 214                                   "vsyscall with bad stack (exploit attempt?)");
 215                 goto sigsegv;
 216         }
 217
 218         tsk = current;
 219         /*
 220          * With a real vsyscall, page faults cause SIGSEGV.  We want to
 221          * preserve that behavior to make writing exploits harder.
 222          */
 223         prev_sig_on_uaccess_error = current_thread_info()->sig_on_uaccess_error;
 224         current_thread_info()->sig_on_uaccess_error = 1;
 225
 226         /*
 227          * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
 228          * 64-bit, so we don't need to special-case it here.  For all the
 229          * vsyscalls, NULL means "don't write anything" not "write it at
 230          * address 0".
 231          */
 232         ret = -EFAULT;
 233         skip = 0;
 234         switch (vsyscall_nr) {
 235         case 0:
 236                 skip = vsyscall_seccomp(tsk, __NR_gettimeofday);
 237                 if (skip)
 238                         break;
 239
 240                 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
 241                     !write_ok_or_segv(regs->si, sizeof(struct timezone)))
 242                         break;
 243
 244                 ret = sys_gettimeofday(
 245                         (struct timeval __user *)regs->di,
 246                         (struct timezone __user *)regs->si);
 247                 break;
 248
 249         case 1:
 250                 skip = vsyscall_seccomp(tsk, __NR_time);
 251                 if (skip)
 252                         break;
 253
 254                 if (!write_ok_or_segv(regs->di, sizeof(time_t)))
 255                         break;
 256
 257                 ret = sys_time((time_t __user *)regs->di);
 258                 break;
 259
 260         case 2:
 261                 skip = vsyscall_seccomp(tsk, __NR_getcpu);
 262                 if (skip)
 263                         break;
 264
 265                 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
 266                     !write_ok_or_segv(regs->si, sizeof(unsigned)))
 267                         break;
 268
 269                 ret = sys_getcpu((unsigned __user *)regs->di,
 270                                  (unsigned __user *)regs->si,
 271                                  NULL);
 272                 break;
 273         }
 274
 275         current_thread_info()->sig_on_uaccess_error = prev_sig_on_uaccess_error;
 276
 277         if (skip) {
 278                 if ((long)regs->ax <= 0L) /* seccomp errno emulation */
 279                         goto do_ret;
 280                 goto done; /* seccomp trace/trap */
 281         }
 282
 283         if (ret == -EFAULT) {
 284                 /* Bad news -- userspace fed a bad pointer to a vsyscall. */
 285                 warn_bad_vsyscall(KERN_INFO, regs,
 286                                   "vsyscall fault (exploit attempt?)");
 287
 288                 /*
 289                  * If we failed to generate a signal for any reason,
 290                  * generate one here.  (This should be impossible.)
 291                  */
 292                 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
 293                                  !sigismember(&tsk->pending.signal, SIGSEGV)))
 294                         goto sigsegv;
 295
 296                 return true;  /* Don't emulate the ret. */
 297         }
 298
 299         regs->ax = ret;
 300
 301 do_ret:
 302         /* Emulate a ret instruction. */
 303         regs->ip = caller;
 304         regs->sp += 8;
 305 done:
 306         return true;
 307
 308 sigsegv:
 309         force_sig(SIGSEGV, current);
 310         return true;
 311 }
 312
 313 /*
 314  * Assume __initcall executes before all user space. Hopefully kmod
 315  * doesn't violate that. We'll find out if it does.
 316  */
 317 static void __cpuinit vsyscall_set_cpu(int cpu)
 318 {
 319         unsigned long d;
 320         unsigned long node = 0;
 321 #ifdef CONFIG_NUMA
 322         node = cpu_to_node(cpu);
 323 #endif
 324         if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
 325                 write_rdtscp_aux((node << 12) | cpu);
 326
 327         /*
 328          * Store cpu number in limit so that it can be loaded quickly
 329          * in user space in vgetcpu. (12 bits for the CPU and 8 bits for the node)
 330          */
 331         d = 0x0f40000000000ULL;
 332         d |= cpu;
 333         d |= (node & 0xf) << 12;
 334         d |= (node >> 4) << 48;
 335
 336         write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
 337 }
 338
 339 static void __cpuinit cpu_vsyscall_init(void *arg)
 340 {
 341         /* preemption should be already off */
 342         vsyscall_set_cpu(raw_smp_processor_id());
 343 }
 344
 345 static int __cpuinit
 346 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
 347 {
 348         long cpu = (long)arg;
 349
 350         if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
 351                 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
 352
 353         return NOTIFY_DONE;
 354 }
 355
 356 void __init map_vsyscall(void)
 357 {
 358         extern char __vsyscall_page;
 359         unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
 360         extern char __vvar_page;
 361         unsigned long physaddr_vvar_page = __pa_symbol(&__vvar_page);
 362
 363         __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_vsyscall,
 364                      vsyscall_mode == NATIVE
 365                      ? PAGE_KERNEL_VSYSCALL
 366                      : PAGE_KERNEL_VVAR);
 367         BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_FIRST_PAGE) !=
 368                      (unsigned long)VSYSCALL_START);
 369
 370         __set_fixmap(VVAR_PAGE, physaddr_vvar_page, PAGE_KERNEL_VVAR);
 371         BUILD_BUG_ON((unsigned long)__fix_to_virt(VVAR_PAGE) !=
 372                      (unsigned long)VVAR_ADDRESS);
 373 }
 374
 375 static int __init vsyscall_init(void)
 376 {
 377         BUG_ON(VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE));
 378
 379         on_each_cpu(cpu_vsyscall_init, NULL, 1);
 380         /* notifier priority > KVM */
 381         hotcpu_notifier(cpu_vsyscall_notifier, 30);
 382
 383         return 0;
 384 }
 385 __initcall(vsyscall_init);