linux/arch/x86/mm/extable.c
<<
>>
Prefs
   1#include <linux/extable.h>
   2#include <linux/uaccess.h>
   3#include <linux/sched/debug.h>
   4#include <xen/xen.h>
   5
   6#include <asm/fpu/internal.h>
   7#include <asm/traps.h>
   8#include <asm/kdebug.h>
   9
  10typedef bool (*ex_handler_t)(const struct exception_table_entry *,
  11                            struct pt_regs *, int, unsigned long,
  12                            unsigned long);
  13
  14static inline unsigned long
  15ex_fixup_addr(const struct exception_table_entry *x)
  16{
  17        return (unsigned long)&x->fixup + x->fixup;
  18}
  19static inline ex_handler_t
  20ex_fixup_handler(const struct exception_table_entry *x)
  21{
  22        return (ex_handler_t)((unsigned long)&x->handler + x->handler);
  23}
  24
  25__visible bool ex_handler_default(const struct exception_table_entry *fixup,
  26                                  struct pt_regs *regs, int trapnr,
  27                                  unsigned long error_code,
  28                                  unsigned long fault_addr)
  29{
  30        regs->ip = ex_fixup_addr(fixup);
  31        return true;
  32}
  33EXPORT_SYMBOL(ex_handler_default);
  34
  35__visible bool ex_handler_fault(const struct exception_table_entry *fixup,
  36                                struct pt_regs *regs, int trapnr,
  37                                unsigned long error_code,
  38                                unsigned long fault_addr)
  39{
  40        regs->ip = ex_fixup_addr(fixup);
  41        regs->ax = trapnr;
  42        return true;
  43}
  44EXPORT_SYMBOL_GPL(ex_handler_fault);
  45
  46/*
  47 * Handler for UD0 exception following a failed test against the
  48 * result of a refcount inc/dec/add/sub.
  49 */
  50__visible bool ex_handler_refcount(const struct exception_table_entry *fixup,
  51                                   struct pt_regs *regs, int trapnr,
  52                                   unsigned long error_code,
  53                                   unsigned long fault_addr)
  54{
  55        /* First unconditionally saturate the refcount. */
  56        *(int *)regs->cx = INT_MIN / 2;
  57
  58        /*
  59         * Strictly speaking, this reports the fixup destination, not
  60         * the fault location, and not the actually overflowing
  61         * instruction, which is the instruction before the "js", but
  62         * since that instruction could be a variety of lengths, just
  63         * report the location after the overflow, which should be close
  64         * enough for finding the overflow, as it's at least back in
  65         * the function, having returned from .text.unlikely.
  66         */
  67        regs->ip = ex_fixup_addr(fixup);
  68
  69        /*
  70         * This function has been called because either a negative refcount
  71         * value was seen by any of the refcount functions, or a zero
  72         * refcount value was seen by refcount_dec().
  73         *
  74         * If we crossed from INT_MAX to INT_MIN, OF (Overflow Flag: result
  75         * wrapped around) will be set. Additionally, seeing the refcount
  76         * reach 0 will set ZF (Zero Flag: result was zero). In each of
  77         * these cases we want a report, since it's a boundary condition.
  78         * The SF case is not reported since it indicates post-boundary
  79         * manipulations below zero or above INT_MAX. And if none of the
  80         * flags are set, something has gone very wrong, so report it.
  81         */
  82        if (regs->flags & (X86_EFLAGS_OF | X86_EFLAGS_ZF)) {
  83                bool zero = regs->flags & X86_EFLAGS_ZF;
  84
  85                refcount_error_report(regs, zero ? "hit zero" : "overflow");
  86        } else if ((regs->flags & X86_EFLAGS_SF) == 0) {
  87                /* Report if none of OF, ZF, nor SF are set. */
  88                refcount_error_report(regs, "unexpected saturation");
  89        }
  90
  91        return true;
  92}
  93EXPORT_SYMBOL(ex_handler_refcount);
  94
  95/*
  96 * Handler for when we fail to restore a task's FPU state.  We should never get
  97 * here because the FPU state of a task using the FPU (task->thread.fpu.state)
  98 * should always be valid.  However, past bugs have allowed userspace to set
  99 * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn().
 100 * These caused XRSTOR to fail when switching to the task, leaking the FPU
 101 * registers of the task previously executing on the CPU.  Mitigate this class
 102 * of vulnerability by restoring from the initial state (essentially, zeroing
 103 * out all the FPU registers) if we can't restore from the task's FPU state.
 104 */
 105__visible bool ex_handler_fprestore(const struct exception_table_entry *fixup,
 106                                    struct pt_regs *regs, int trapnr,
 107                                    unsigned long error_code,
 108                                    unsigned long fault_addr)
 109{
 110        regs->ip = ex_fixup_addr(fixup);
 111
 112        WARN_ONCE(1, "Bad FPU state detected at %pB, reinitializing FPU registers.",
 113                  (void *)instruction_pointer(regs));
 114
 115        __copy_kernel_to_fpregs(&init_fpstate, -1);
 116        return true;
 117}
 118EXPORT_SYMBOL_GPL(ex_handler_fprestore);
 119
 120__visible bool ex_handler_uaccess(const struct exception_table_entry *fixup,
 121                                  struct pt_regs *regs, int trapnr,
 122                                  unsigned long error_code,
 123                                  unsigned long fault_addr)
 124{
 125        regs->ip = ex_fixup_addr(fixup);
 126        return true;
 127}
 128EXPORT_SYMBOL(ex_handler_uaccess);
 129
 130__visible bool ex_handler_ext(const struct exception_table_entry *fixup,
 131                              struct pt_regs *regs, int trapnr,
 132                              unsigned long error_code,
 133                              unsigned long fault_addr)
 134{
 135        /* Special hack for uaccess_err */
 136        current->thread.uaccess_err = 1;
 137        regs->ip = ex_fixup_addr(fixup);
 138        return true;
 139}
 140EXPORT_SYMBOL(ex_handler_ext);
 141
 142__visible bool ex_handler_rdmsr_unsafe(const struct exception_table_entry *fixup,
 143                                       struct pt_regs *regs, int trapnr,
 144                                       unsigned long error_code,
 145                                       unsigned long fault_addr)
 146{
 147        if (pr_warn_once("unchecked MSR access error: RDMSR from 0x%x at rIP: 0x%lx (%pF)\n",
 148                         (unsigned int)regs->cx, regs->ip, (void *)regs->ip))
 149                show_stack_regs(regs);
 150
 151        /* Pretend that the read succeeded and returned 0. */
 152        regs->ip = ex_fixup_addr(fixup);
 153        regs->ax = 0;
 154        regs->dx = 0;
 155        return true;
 156}
 157EXPORT_SYMBOL(ex_handler_rdmsr_unsafe);
 158
 159__visible bool ex_handler_wrmsr_unsafe(const struct exception_table_entry *fixup,
 160                                       struct pt_regs *regs, int trapnr,
 161                                       unsigned long error_code,
 162                                       unsigned long fault_addr)
 163{
 164        if (pr_warn_once("unchecked MSR access error: WRMSR to 0x%x (tried to write 0x%08x%08x) at rIP: 0x%lx (%pF)\n",
 165                         (unsigned int)regs->cx, (unsigned int)regs->dx,
 166                         (unsigned int)regs->ax,  regs->ip, (void *)regs->ip))
 167                show_stack_regs(regs);
 168
 169        /* Pretend that the write succeeded. */
 170        regs->ip = ex_fixup_addr(fixup);
 171        return true;
 172}
 173EXPORT_SYMBOL(ex_handler_wrmsr_unsafe);
 174
 175__visible bool ex_handler_clear_fs(const struct exception_table_entry *fixup,
 176                                   struct pt_regs *regs, int trapnr,
 177                                   unsigned long error_code,
 178                                   unsigned long fault_addr)
 179{
 180        if (static_cpu_has(X86_BUG_NULL_SEG))
 181                asm volatile ("mov %0, %%fs" : : "rm" (__USER_DS));
 182        asm volatile ("mov %0, %%fs" : : "rm" (0));
 183        return ex_handler_default(fixup, regs, trapnr, error_code, fault_addr);
 184}
 185EXPORT_SYMBOL(ex_handler_clear_fs);
 186
 187__visible bool ex_has_fault_handler(unsigned long ip)
 188{
 189        const struct exception_table_entry *e;
 190        ex_handler_t handler;
 191
 192        e = search_exception_tables(ip);
 193        if (!e)
 194                return false;
 195        handler = ex_fixup_handler(e);
 196
 197        return handler == ex_handler_fault;
 198}
 199
 200int fixup_exception(struct pt_regs *regs, int trapnr, unsigned long error_code,
 201                    unsigned long fault_addr)
 202{
 203        const struct exception_table_entry *e;
 204        ex_handler_t handler;
 205
 206#ifdef CONFIG_PNPBIOS
 207        if (unlikely(SEGMENT_IS_PNP_CODE(regs->cs))) {
 208                extern u32 pnp_bios_fault_eip, pnp_bios_fault_esp;
 209                extern u32 pnp_bios_is_utter_crap;
 210                pnp_bios_is_utter_crap = 1;
 211                printk(KERN_CRIT "PNPBIOS fault.. attempting recovery.\n");
 212                __asm__ volatile(
 213                        "movl %0, %%esp\n\t"
 214                        "jmp *%1\n\t"
 215                        : : "g" (pnp_bios_fault_esp), "g" (pnp_bios_fault_eip));
 216                panic("do_trap: can't hit this");
 217        }
 218#endif
 219
 220        e = search_exception_tables(regs->ip);
 221        if (!e)
 222                return 0;
 223
 224        handler = ex_fixup_handler(e);
 225        return handler(e, regs, trapnr, error_code, fault_addr);
 226}
 227
 228extern unsigned int early_recursion_flag;
 229
 230/* Restricted version used during very early boot */
 231void __init early_fixup_exception(struct pt_regs *regs, int trapnr)
 232{
 233        /* Ignore early NMIs. */
 234        if (trapnr == X86_TRAP_NMI)
 235                return;
 236
 237        if (early_recursion_flag > 2)
 238                goto halt_loop;
 239
 240        /*
 241         * Old CPUs leave the high bits of CS on the stack
 242         * undefined.  I'm not sure which CPUs do this, but at least
 243         * the 486 DX works this way.
 244         * Xen pv domains are not using the default __KERNEL_CS.
 245         */
 246        if (!xen_pv_domain() && regs->cs != __KERNEL_CS)
 247                goto fail;
 248
 249        /*
 250         * The full exception fixup machinery is available as soon as
 251         * the early IDT is loaded.  This means that it is the
 252         * responsibility of extable users to either function correctly
 253         * when handlers are invoked early or to simply avoid causing
 254         * exceptions before they're ready to handle them.
 255         *
 256         * This is better than filtering which handlers can be used,
 257         * because refusing to call a handler here is guaranteed to
 258         * result in a hard-to-debug panic.
 259         *
 260         * Keep in mind that not all vectors actually get here.  Early
 261         * page faults, for example, are special.
 262         */
 263        if (fixup_exception(regs, trapnr, regs->orig_ax, 0))
 264                return;
 265
 266        if (fixup_bug(regs, trapnr))
 267                return;
 268
 269fail:
 270        early_printk("PANIC: early exception 0x%02x IP %lx:%lx error %lx cr2 0x%lx\n",
 271                     (unsigned)trapnr, (unsigned long)regs->cs, regs->ip,
 272                     regs->orig_ax, read_cr2());
 273
 274        show_regs(regs);
 275
 276halt_loop:
 277        while (true)
 278                halt();
 279}
 280