linux/arch/x86/kernel/vm86_32.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  Copyright (C) 1994  Linus Torvalds
   4 *
   5 *  29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
   6 *                stack - Manfred Spraul <manfred@colorfullife.com>
   7 *
   8 *  22 mar 2002 - Manfred detected the stackfaults, but didn't handle
   9 *                them correctly. Now the emulation will be in a
  10 *                consistent state after stackfaults - Kasper Dupont
  11 *                <kasperd@daimi.au.dk>
  12 *
  13 *  22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
  14 *                <kasperd@daimi.au.dk>
  15 *
  16 *  ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
  17 *                caused by Kasper Dupont's changes - Stas Sergeev
  18 *
  19 *   4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
  20 *                Kasper Dupont <kasperd@daimi.au.dk>
  21 *
  22 *   9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
  23 *                Kasper Dupont <kasperd@daimi.au.dk>
  24 *
  25 *   9 apr 2002 - Changed stack access macros to jump to a label
  26 *                instead of returning to userspace. This simplifies
  27 *                do_int, and is needed by handle_vm6_fault. Kasper
  28 *                Dupont <kasperd@daimi.au.dk>
  29 *
  30 */
  31
  32#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  33
  34#include <linux/capability.h>
  35#include <linux/errno.h>
  36#include <linux/interrupt.h>
  37#include <linux/syscalls.h>
  38#include <linux/sched.h>
  39#include <linux/sched/task_stack.h>
  40#include <linux/kernel.h>
  41#include <linux/signal.h>
  42#include <linux/string.h>
  43#include <linux/mm.h>
  44#include <linux/smp.h>
  45#include <linux/highmem.h>
  46#include <linux/ptrace.h>
  47#include <linux/audit.h>
  48#include <linux/stddef.h>
  49#include <linux/slab.h>
  50#include <linux/security.h>
  51
  52#include <linux/uaccess.h>
  53#include <asm/io.h>
  54#include <asm/tlbflush.h>
  55#include <asm/irq.h>
  56#include <asm/traps.h>
  57#include <asm/vm86.h>
  58#include <asm/switch_to.h>
  59
  60/*
  61 * Known problems:
  62 *
  63 * Interrupt handling is not guaranteed:
  64 * - a real x86 will disable all interrupts for one instruction
  65 *   after a "mov ss,xx" to make stack handling atomic even without
  66 *   the 'lss' instruction. We can't guarantee this in v86 mode,
  67 *   as the next instruction might result in a page fault or similar.
  68 * - a real x86 will have interrupts disabled for one instruction
  69 *   past the 'sti' that enables them. We don't bother with all the
  70 *   details yet.
  71 *
  72 * Let's hope these problems do not actually matter for anything.
  73 */
  74
  75
  76/*
  77 * 8- and 16-bit register defines..
  78 */
  79#define AL(regs)        (((unsigned char *)&((regs)->pt.ax))[0])
  80#define AH(regs)        (((unsigned char *)&((regs)->pt.ax))[1])
  81#define IP(regs)        (*(unsigned short *)&((regs)->pt.ip))
  82#define SP(regs)        (*(unsigned short *)&((regs)->pt.sp))
  83
  84/*
  85 * virtual flags (16 and 32-bit versions)
  86 */
  87#define VFLAGS  (*(unsigned short *)&(current->thread.vm86->veflags))
  88#define VEFLAGS (current->thread.vm86->veflags)
  89
  90#define set_flags(X, new, mask) \
  91((X) = ((X) & ~(mask)) | ((new) & (mask)))
  92
  93#define SAFE_MASK       (0xDD5)
  94#define RETURN_MASK     (0xDFF)
  95
  96void save_v86_state(struct kernel_vm86_regs *regs, int retval)
  97{
  98        struct task_struct *tsk = current;
  99        struct vm86plus_struct __user *user;
 100        struct vm86 *vm86 = current->thread.vm86;
 101        long err = 0;
 102
 103        /*
 104         * This gets called from entry.S with interrupts disabled, but
 105         * from process context. Enable interrupts here, before trying
 106         * to access user space.
 107         */
 108        local_irq_enable();
 109
 110        if (!vm86 || !vm86->user_vm86) {
 111                pr_alert("no user_vm86: BAD\n");
 112                do_exit(SIGSEGV);
 113        }
 114        set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | vm86->veflags_mask);
 115        user = vm86->user_vm86;
 116
 117        if (!access_ok(user, vm86->vm86plus.is_vm86pus ?
 118                       sizeof(struct vm86plus_struct) :
 119                       sizeof(struct vm86_struct))) {
 120                pr_alert("could not access userspace vm86 info\n");
 121                do_exit(SIGSEGV);
 122        }
 123
 124        put_user_try {
 125                put_user_ex(regs->pt.bx, &user->regs.ebx);
 126                put_user_ex(regs->pt.cx, &user->regs.ecx);
 127                put_user_ex(regs->pt.dx, &user->regs.edx);
 128                put_user_ex(regs->pt.si, &user->regs.esi);
 129                put_user_ex(regs->pt.di, &user->regs.edi);
 130                put_user_ex(regs->pt.bp, &user->regs.ebp);
 131                put_user_ex(regs->pt.ax, &user->regs.eax);
 132                put_user_ex(regs->pt.ip, &user->regs.eip);
 133                put_user_ex(regs->pt.cs, &user->regs.cs);
 134                put_user_ex(regs->pt.flags, &user->regs.eflags);
 135                put_user_ex(regs->pt.sp, &user->regs.esp);
 136                put_user_ex(regs->pt.ss, &user->regs.ss);
 137                put_user_ex(regs->es, &user->regs.es);
 138                put_user_ex(regs->ds, &user->regs.ds);
 139                put_user_ex(regs->fs, &user->regs.fs);
 140                put_user_ex(regs->gs, &user->regs.gs);
 141
 142                put_user_ex(vm86->screen_bitmap, &user->screen_bitmap);
 143        } put_user_catch(err);
 144        if (err) {
 145                pr_alert("could not access userspace vm86 info\n");
 146                do_exit(SIGSEGV);
 147        }
 148
 149        preempt_disable();
 150        tsk->thread.sp0 = vm86->saved_sp0;
 151        tsk->thread.sysenter_cs = __KERNEL_CS;
 152        update_task_stack(tsk);
 153        refresh_sysenter_cs(&tsk->thread);
 154        vm86->saved_sp0 = 0;
 155        preempt_enable();
 156
 157        memcpy(&regs->pt, &vm86->regs32, sizeof(struct pt_regs));
 158
 159        lazy_load_gs(vm86->regs32.gs);
 160
 161        regs->pt.ax = retval;
 162}
 163
 164static void mark_screen_rdonly(struct mm_struct *mm)
 165{
 166        struct vm_area_struct *vma;
 167        spinlock_t *ptl;
 168        pgd_t *pgd;
 169        p4d_t *p4d;
 170        pud_t *pud;
 171        pmd_t *pmd;
 172        pte_t *pte;
 173        int i;
 174
 175        down_write(&mm->mmap_sem);
 176        pgd = pgd_offset(mm, 0xA0000);
 177        if (pgd_none_or_clear_bad(pgd))
 178                goto out;
 179        p4d = p4d_offset(pgd, 0xA0000);
 180        if (p4d_none_or_clear_bad(p4d))
 181                goto out;
 182        pud = pud_offset(p4d, 0xA0000);
 183        if (pud_none_or_clear_bad(pud))
 184                goto out;
 185        pmd = pmd_offset(pud, 0xA0000);
 186
 187        if (pmd_trans_huge(*pmd)) {
 188                vma = find_vma(mm, 0xA0000);
 189                split_huge_pmd(vma, pmd, 0xA0000);
 190        }
 191        if (pmd_none_or_clear_bad(pmd))
 192                goto out;
 193        pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
 194        for (i = 0; i < 32; i++) {
 195                if (pte_present(*pte))
 196                        set_pte(pte, pte_wrprotect(*pte));
 197                pte++;
 198        }
 199        pte_unmap_unlock(pte, ptl);
 200out:
 201        up_write(&mm->mmap_sem);
 202        flush_tlb_mm_range(mm, 0xA0000, 0xA0000 + 32*PAGE_SIZE, PAGE_SHIFT, false);
 203}
 204
 205
 206
 207static int do_vm86_irq_handling(int subfunction, int irqnumber);
 208static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus);
 209
 210SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86)
 211{
 212        return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false);
 213}
 214
 215
 216SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg)
 217{
 218        switch (cmd) {
 219        case VM86_REQUEST_IRQ:
 220        case VM86_FREE_IRQ:
 221        case VM86_GET_IRQ_BITS:
 222        case VM86_GET_AND_RESET_IRQ:
 223                return do_vm86_irq_handling(cmd, (int)arg);
 224        case VM86_PLUS_INSTALL_CHECK:
 225                /*
 226                 * NOTE: on old vm86 stuff this will return the error
 227                 *  from access_ok(), because the subfunction is
 228                 *  interpreted as (invalid) address to vm86_struct.
 229                 *  So the installation check works.
 230                 */
 231                return 0;
 232        }
 233
 234        /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
 235        return do_sys_vm86((struct vm86plus_struct __user *) arg, true);
 236}
 237
 238
 239static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
 240{
 241        struct task_struct *tsk = current;
 242        struct vm86 *vm86 = tsk->thread.vm86;
 243        struct kernel_vm86_regs vm86regs;
 244        struct pt_regs *regs = current_pt_regs();
 245        unsigned long err = 0;
 246
 247        err = security_mmap_addr(0);
 248        if (err) {
 249                /*
 250                 * vm86 cannot virtualize the address space, so vm86 users
 251                 * need to manage the low 1MB themselves using mmap.  Given
 252                 * that BIOS places important data in the first page, vm86
 253                 * is essentially useless if mmap_min_addr != 0.  DOSEMU,
 254                 * for example, won't even bother trying to use vm86 if it
 255                 * can't map a page at virtual address 0.
 256                 *
 257                 * To reduce the available kernel attack surface, simply
 258                 * disallow vm86(old) for users who cannot mmap at va 0.
 259                 *
 260                 * The implementation of security_mmap_addr will allow
 261                 * suitably privileged users to map va 0 even if
 262                 * vm.mmap_min_addr is set above 0, and we want this
 263                 * behavior for vm86 as well, as it ensures that legacy
 264                 * tools like vbetool will not fail just because of
 265                 * vm.mmap_min_addr.
 266                 */
 267                pr_info_once("Denied a call to vm86(old) from %s[%d] (uid: %d).  Set the vm.mmap_min_addr sysctl to 0 and/or adjust LSM mmap_min_addr policy to enable vm86 if you are using a vm86-based DOS emulator.\n",
 268                             current->comm, task_pid_nr(current),
 269                             from_kuid_munged(&init_user_ns, current_uid()));
 270                return -EPERM;
 271        }
 272
 273        if (!vm86) {
 274                if (!(vm86 = kzalloc(sizeof(*vm86), GFP_KERNEL)))
 275                        return -ENOMEM;
 276                tsk->thread.vm86 = vm86;
 277        }
 278        if (vm86->saved_sp0)
 279                return -EPERM;
 280
 281        if (!access_ok(user_vm86, plus ?
 282                       sizeof(struct vm86_struct) :
 283                       sizeof(struct vm86plus_struct)))
 284                return -EFAULT;
 285
 286        memset(&vm86regs, 0, sizeof(vm86regs));
 287        get_user_try {
 288                unsigned short seg;
 289                get_user_ex(vm86regs.pt.bx, &user_vm86->regs.ebx);
 290                get_user_ex(vm86regs.pt.cx, &user_vm86->regs.ecx);
 291                get_user_ex(vm86regs.pt.dx, &user_vm86->regs.edx);
 292                get_user_ex(vm86regs.pt.si, &user_vm86->regs.esi);
 293                get_user_ex(vm86regs.pt.di, &user_vm86->regs.edi);
 294                get_user_ex(vm86regs.pt.bp, &user_vm86->regs.ebp);
 295                get_user_ex(vm86regs.pt.ax, &user_vm86->regs.eax);
 296                get_user_ex(vm86regs.pt.ip, &user_vm86->regs.eip);
 297                get_user_ex(seg, &user_vm86->regs.cs);
 298                vm86regs.pt.cs = seg;
 299                get_user_ex(vm86regs.pt.flags, &user_vm86->regs.eflags);
 300                get_user_ex(vm86regs.pt.sp, &user_vm86->regs.esp);
 301                get_user_ex(seg, &user_vm86->regs.ss);
 302                vm86regs.pt.ss = seg;
 303                get_user_ex(vm86regs.es, &user_vm86->regs.es);
 304                get_user_ex(vm86regs.ds, &user_vm86->regs.ds);
 305                get_user_ex(vm86regs.fs, &user_vm86->regs.fs);
 306                get_user_ex(vm86regs.gs, &user_vm86->regs.gs);
 307
 308                get_user_ex(vm86->flags, &user_vm86->flags);
 309                get_user_ex(vm86->screen_bitmap, &user_vm86->screen_bitmap);
 310                get_user_ex(vm86->cpu_type, &user_vm86->cpu_type);
 311        } get_user_catch(err);
 312        if (err)
 313                return err;
 314
 315        if (copy_from_user(&vm86->int_revectored,
 316                           &user_vm86->int_revectored,
 317                           sizeof(struct revectored_struct)))
 318                return -EFAULT;
 319        if (copy_from_user(&vm86->int21_revectored,
 320                           &user_vm86->int21_revectored,
 321                           sizeof(struct revectored_struct)))
 322                return -EFAULT;
 323        if (plus) {
 324                if (copy_from_user(&vm86->vm86plus, &user_vm86->vm86plus,
 325                                   sizeof(struct vm86plus_info_struct)))
 326                        return -EFAULT;
 327                vm86->vm86plus.is_vm86pus = 1;
 328        } else
 329                memset(&vm86->vm86plus, 0,
 330                       sizeof(struct vm86plus_info_struct));
 331
 332        memcpy(&vm86->regs32, regs, sizeof(struct pt_regs));
 333        vm86->user_vm86 = user_vm86;
 334
 335/*
 336 * The flags register is also special: we cannot trust that the user
 337 * has set it up safely, so this makes sure interrupt etc flags are
 338 * inherited from protected mode.
 339 */
 340        VEFLAGS = vm86regs.pt.flags;
 341        vm86regs.pt.flags &= SAFE_MASK;
 342        vm86regs.pt.flags |= regs->flags & ~SAFE_MASK;
 343        vm86regs.pt.flags |= X86_VM_MASK;
 344
 345        vm86regs.pt.orig_ax = regs->orig_ax;
 346
 347        switch (vm86->cpu_type) {
 348        case CPU_286:
 349                vm86->veflags_mask = 0;
 350                break;
 351        case CPU_386:
 352                vm86->veflags_mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL;
 353                break;
 354        case CPU_486:
 355                vm86->veflags_mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
 356                break;
 357        default:
 358                vm86->veflags_mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
 359                break;
 360        }
 361
 362/*
 363 * Save old state
 364 */
 365        vm86->saved_sp0 = tsk->thread.sp0;
 366        lazy_save_gs(vm86->regs32.gs);
 367
 368        /* make room for real-mode segments */
 369        preempt_disable();
 370        tsk->thread.sp0 += 16;
 371
 372        if (static_cpu_has(X86_FEATURE_SEP)) {
 373                tsk->thread.sysenter_cs = 0;
 374                refresh_sysenter_cs(&tsk->thread);
 375        }
 376
 377        update_task_stack(tsk);
 378        preempt_enable();
 379
 380        if (vm86->flags & VM86_SCREEN_BITMAP)
 381                mark_screen_rdonly(tsk->mm);
 382
 383        memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs));
 384        force_iret();
 385        return regs->ax;
 386}
 387
 388static inline void set_IF(struct kernel_vm86_regs *regs)
 389{
 390        VEFLAGS |= X86_EFLAGS_VIF;
 391}
 392
 393static inline void clear_IF(struct kernel_vm86_regs *regs)
 394{
 395        VEFLAGS &= ~X86_EFLAGS_VIF;
 396}
 397
 398static inline void clear_TF(struct kernel_vm86_regs *regs)
 399{
 400        regs->pt.flags &= ~X86_EFLAGS_TF;
 401}
 402
 403static inline void clear_AC(struct kernel_vm86_regs *regs)
 404{
 405        regs->pt.flags &= ~X86_EFLAGS_AC;
 406}
 407
 408/*
 409 * It is correct to call set_IF(regs) from the set_vflags_*
 410 * functions. However someone forgot to call clear_IF(regs)
 411 * in the opposite case.
 412 * After the command sequence CLI PUSHF STI POPF you should
 413 * end up with interrupts disabled, but you ended up with
 414 * interrupts enabled.
 415 *  ( I was testing my own changes, but the only bug I
 416 *    could find was in a function I had not changed. )
 417 * [KD]
 418 */
 419
 420static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
 421{
 422        set_flags(VEFLAGS, flags, current->thread.vm86->veflags_mask);
 423        set_flags(regs->pt.flags, flags, SAFE_MASK);
 424        if (flags & X86_EFLAGS_IF)
 425                set_IF(regs);
 426        else
 427                clear_IF(regs);
 428}
 429
 430static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
 431{
 432        set_flags(VFLAGS, flags, current->thread.vm86->veflags_mask);
 433        set_flags(regs->pt.flags, flags, SAFE_MASK);
 434        if (flags & X86_EFLAGS_IF)
 435                set_IF(regs);
 436        else
 437                clear_IF(regs);
 438}
 439
 440static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
 441{
 442        unsigned long flags = regs->pt.flags & RETURN_MASK;
 443
 444        if (VEFLAGS & X86_EFLAGS_VIF)
 445                flags |= X86_EFLAGS_IF;
 446        flags |= X86_EFLAGS_IOPL;
 447        return flags | (VEFLAGS & current->thread.vm86->veflags_mask);
 448}
 449
 450static inline int is_revectored(int nr, struct revectored_struct *bitmap)
 451{
 452        return test_bit(nr, bitmap->__map);
 453}
 454
 455#define val_byte(val, n) (((__u8 *)&val)[n])
 456
 457#define pushb(base, ptr, val, err_label) \
 458        do { \
 459                __u8 __val = val; \
 460                ptr--; \
 461                if (put_user(__val, base + ptr) < 0) \
 462                        goto err_label; \
 463        } while (0)
 464
 465#define pushw(base, ptr, val, err_label) \
 466        do { \
 467                __u16 __val = val; \
 468                ptr--; \
 469                if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 470                        goto err_label; \
 471                ptr--; \
 472                if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 473                        goto err_label; \
 474        } while (0)
 475
 476#define pushl(base, ptr, val, err_label) \
 477        do { \
 478                __u32 __val = val; \
 479                ptr--; \
 480                if (put_user(val_byte(__val, 3), base + ptr) < 0) \
 481                        goto err_label; \
 482                ptr--; \
 483                if (put_user(val_byte(__val, 2), base + ptr) < 0) \
 484                        goto err_label; \
 485                ptr--; \
 486                if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 487                        goto err_label; \
 488                ptr--; \
 489                if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 490                        goto err_label; \
 491        } while (0)
 492
 493#define popb(base, ptr, err_label) \
 494        ({ \
 495                __u8 __res; \
 496                if (get_user(__res, base + ptr) < 0) \
 497                        goto err_label; \
 498                ptr++; \
 499                __res; \
 500        })
 501
 502#define popw(base, ptr, err_label) \
 503        ({ \
 504                __u16 __res; \
 505                if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 506                        goto err_label; \
 507                ptr++; \
 508                if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 509                        goto err_label; \
 510                ptr++; \
 511                __res; \
 512        })
 513
 514#define popl(base, ptr, err_label) \
 515        ({ \
 516                __u32 __res; \
 517                if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 518                        goto err_label; \
 519                ptr++; \
 520                if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 521                        goto err_label; \
 522                ptr++; \
 523                if (get_user(val_byte(__res, 2), base + ptr) < 0) \
 524                        goto err_label; \
 525                ptr++; \
 526                if (get_user(val_byte(__res, 3), base + ptr) < 0) \
 527                        goto err_label; \
 528                ptr++; \
 529                __res; \
 530        })
 531
 532/* There are so many possible reasons for this function to return
 533 * VM86_INTx, so adding another doesn't bother me. We can expect
 534 * userspace programs to be able to handle it. (Getting a problem
 535 * in userspace is always better than an Oops anyway.) [KD]
 536 */
 537static void do_int(struct kernel_vm86_regs *regs, int i,
 538    unsigned char __user *ssp, unsigned short sp)
 539{
 540        unsigned long __user *intr_ptr;
 541        unsigned long segoffs;
 542        struct vm86 *vm86 = current->thread.vm86;
 543
 544        if (regs->pt.cs == BIOSSEG)
 545                goto cannot_handle;
 546        if (is_revectored(i, &vm86->int_revectored))
 547                goto cannot_handle;
 548        if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored))
 549                goto cannot_handle;
 550        intr_ptr = (unsigned long __user *) (i << 2);
 551        if (get_user(segoffs, intr_ptr))
 552                goto cannot_handle;
 553        if ((segoffs >> 16) == BIOSSEG)
 554                goto cannot_handle;
 555        pushw(ssp, sp, get_vflags(regs), cannot_handle);
 556        pushw(ssp, sp, regs->pt.cs, cannot_handle);
 557        pushw(ssp, sp, IP(regs), cannot_handle);
 558        regs->pt.cs = segoffs >> 16;
 559        SP(regs) -= 6;
 560        IP(regs) = segoffs & 0xffff;
 561        clear_TF(regs);
 562        clear_IF(regs);
 563        clear_AC(regs);
 564        return;
 565
 566cannot_handle:
 567        save_v86_state(regs, VM86_INTx + (i << 8));
 568}
 569
 570int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
 571{
 572        struct vm86 *vm86 = current->thread.vm86;
 573
 574        if (vm86->vm86plus.is_vm86pus) {
 575                if ((trapno == 3) || (trapno == 1)) {
 576                        save_v86_state(regs, VM86_TRAP + (trapno << 8));
 577                        return 0;
 578                }
 579                do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
 580                return 0;
 581        }
 582        if (trapno != 1)
 583                return 1; /* we let this handle by the calling routine */
 584        current->thread.trap_nr = trapno;
 585        current->thread.error_code = error_code;
 586        force_sig(SIGTRAP, current);
 587        return 0;
 588}
 589
 590void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
 591{
 592        unsigned char opcode;
 593        unsigned char __user *csp;
 594        unsigned char __user *ssp;
 595        unsigned short ip, sp, orig_flags;
 596        int data32, pref_done;
 597        struct vm86plus_info_struct *vmpi = &current->thread.vm86->vm86plus;
 598
 599#define CHECK_IF_IN_TRAP \
 600        if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \
 601                newflags |= X86_EFLAGS_TF
 602
 603        orig_flags = *(unsigned short *)&regs->pt.flags;
 604
 605        csp = (unsigned char __user *) (regs->pt.cs << 4);
 606        ssp = (unsigned char __user *) (regs->pt.ss << 4);
 607        sp = SP(regs);
 608        ip = IP(regs);
 609
 610        data32 = 0;
 611        pref_done = 0;
 612        do {
 613                switch (opcode = popb(csp, ip, simulate_sigsegv)) {
 614                case 0x66:      /* 32-bit data */     data32 = 1; break;
 615                case 0x67:      /* 32-bit address */  break;
 616                case 0x2e:      /* CS */              break;
 617                case 0x3e:      /* DS */              break;
 618                case 0x26:      /* ES */              break;
 619                case 0x36:      /* SS */              break;
 620                case 0x65:      /* GS */              break;
 621                case 0x64:      /* FS */              break;
 622                case 0xf2:      /* repnz */       break;
 623                case 0xf3:      /* rep */             break;
 624                default: pref_done = 1;
 625                }
 626        } while (!pref_done);
 627
 628        switch (opcode) {
 629
 630        /* pushf */
 631        case 0x9c:
 632                if (data32) {
 633                        pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
 634                        SP(regs) -= 4;
 635                } else {
 636                        pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
 637                        SP(regs) -= 2;
 638                }
 639                IP(regs) = ip;
 640                goto vm86_fault_return;
 641
 642        /* popf */
 643        case 0x9d:
 644                {
 645                unsigned long newflags;
 646                if (data32) {
 647                        newflags = popl(ssp, sp, simulate_sigsegv);
 648                        SP(regs) += 4;
 649                } else {
 650                        newflags = popw(ssp, sp, simulate_sigsegv);
 651                        SP(regs) += 2;
 652                }
 653                IP(regs) = ip;
 654                CHECK_IF_IN_TRAP;
 655                if (data32)
 656                        set_vflags_long(newflags, regs);
 657                else
 658                        set_vflags_short(newflags, regs);
 659
 660                goto check_vip;
 661                }
 662
 663        /* int xx */
 664        case 0xcd: {
 665                int intno = popb(csp, ip, simulate_sigsegv);
 666                IP(regs) = ip;
 667                if (vmpi->vm86dbg_active) {
 668                        if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) {
 669                                save_v86_state(regs, VM86_INTx + (intno << 8));
 670                                return;
 671                        }
 672                }
 673                do_int(regs, intno, ssp, sp);
 674                return;
 675        }
 676
 677        /* iret */
 678        case 0xcf:
 679                {
 680                unsigned long newip;
 681                unsigned long newcs;
 682                unsigned long newflags;
 683                if (data32) {
 684                        newip = popl(ssp, sp, simulate_sigsegv);
 685                        newcs = popl(ssp, sp, simulate_sigsegv);
 686                        newflags = popl(ssp, sp, simulate_sigsegv);
 687                        SP(regs) += 12;
 688                } else {
 689                        newip = popw(ssp, sp, simulate_sigsegv);
 690                        newcs = popw(ssp, sp, simulate_sigsegv);
 691                        newflags = popw(ssp, sp, simulate_sigsegv);
 692                        SP(regs) += 6;
 693                }
 694                IP(regs) = newip;
 695                regs->pt.cs = newcs;
 696                CHECK_IF_IN_TRAP;
 697                if (data32) {
 698                        set_vflags_long(newflags, regs);
 699                } else {
 700                        set_vflags_short(newflags, regs);
 701                }
 702                goto check_vip;
 703                }
 704
 705        /* cli */
 706        case 0xfa:
 707                IP(regs) = ip;
 708                clear_IF(regs);
 709                goto vm86_fault_return;
 710
 711        /* sti */
 712        /*
 713         * Damn. This is incorrect: the 'sti' instruction should actually
 714         * enable interrupts after the /next/ instruction. Not good.
 715         *
 716         * Probably needs some horsing around with the TF flag. Aiee..
 717         */
 718        case 0xfb:
 719                IP(regs) = ip;
 720                set_IF(regs);
 721                goto check_vip;
 722
 723        default:
 724                save_v86_state(regs, VM86_UNKNOWN);
 725        }
 726
 727        return;
 728
 729check_vip:
 730        if ((VEFLAGS & (X86_EFLAGS_VIP | X86_EFLAGS_VIF)) ==
 731            (X86_EFLAGS_VIP | X86_EFLAGS_VIF)) {
 732                save_v86_state(regs, VM86_STI);
 733                return;
 734        }
 735
 736vm86_fault_return:
 737        if (vmpi->force_return_for_pic  && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) {
 738                save_v86_state(regs, VM86_PICRETURN);
 739                return;
 740        }
 741        if (orig_flags & X86_EFLAGS_TF)
 742                handle_vm86_trap(regs, 0, X86_TRAP_DB);
 743        return;
 744
 745simulate_sigsegv:
 746        /* FIXME: After a long discussion with Stas we finally
 747         *        agreed, that this is wrong. Here we should
 748         *        really send a SIGSEGV to the user program.
 749         *        But how do we create the correct context? We
 750         *        are inside a general protection fault handler
 751         *        and has just returned from a page fault handler.
 752         *        The correct context for the signal handler
 753         *        should be a mixture of the two, but how do we
 754         *        get the information? [KD]
 755         */
 756        save_v86_state(regs, VM86_UNKNOWN);
 757}
 758
 759/* ---------------- vm86 special IRQ passing stuff ----------------- */
 760
 761#define VM86_IRQNAME            "vm86irq"
 762
 763static struct vm86_irqs {
 764        struct task_struct *tsk;
 765        int sig;
 766} vm86_irqs[16];
 767
 768static DEFINE_SPINLOCK(irqbits_lock);
 769static int irqbits;
 770
 771#define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
 772        | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO)  | (1 << SIGURG) \
 773        | (1 << SIGUNUSED))
 774
 775static irqreturn_t irq_handler(int intno, void *dev_id)
 776{
 777        int irq_bit;
 778        unsigned long flags;
 779
 780        spin_lock_irqsave(&irqbits_lock, flags);
 781        irq_bit = 1 << intno;
 782        if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
 783                goto out;
 784        irqbits |= irq_bit;
 785        if (vm86_irqs[intno].sig)
 786                send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
 787        /*
 788         * IRQ will be re-enabled when user asks for the irq (whether
 789         * polling or as a result of the signal)
 790         */
 791        disable_irq_nosync(intno);
 792        spin_unlock_irqrestore(&irqbits_lock, flags);
 793        return IRQ_HANDLED;
 794
 795out:
 796        spin_unlock_irqrestore(&irqbits_lock, flags);
 797        return IRQ_NONE;
 798}
 799
 800static inline void free_vm86_irq(int irqnumber)
 801{
 802        unsigned long flags;
 803
 804        free_irq(irqnumber, NULL);
 805        vm86_irqs[irqnumber].tsk = NULL;
 806
 807        spin_lock_irqsave(&irqbits_lock, flags);
 808        irqbits &= ~(1 << irqnumber);
 809        spin_unlock_irqrestore(&irqbits_lock, flags);
 810}
 811
 812void release_vm86_irqs(struct task_struct *task)
 813{
 814        int i;
 815        for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
 816            if (vm86_irqs[i].tsk == task)
 817                free_vm86_irq(i);
 818}
 819
 820static inline int get_and_reset_irq(int irqnumber)
 821{
 822        int bit;
 823        unsigned long flags;
 824        int ret = 0;
 825
 826        if (invalid_vm86_irq(irqnumber)) return 0;
 827        if (vm86_irqs[irqnumber].tsk != current) return 0;
 828        spin_lock_irqsave(&irqbits_lock, flags);
 829        bit = irqbits & (1 << irqnumber);
 830        irqbits &= ~bit;
 831        if (bit) {
 832                enable_irq(irqnumber);
 833                ret = 1;
 834        }
 835
 836        spin_unlock_irqrestore(&irqbits_lock, flags);
 837        return ret;
 838}
 839
 840
 841static int do_vm86_irq_handling(int subfunction, int irqnumber)
 842{
 843        int ret;
 844        switch (subfunction) {
 845                case VM86_GET_AND_RESET_IRQ: {
 846                        return get_and_reset_irq(irqnumber);
 847                }
 848                case VM86_GET_IRQ_BITS: {
 849                        return irqbits;
 850                }
 851                case VM86_REQUEST_IRQ: {
 852                        int sig = irqnumber >> 8;
 853                        int irq = irqnumber & 255;
 854                        if (!capable(CAP_SYS_ADMIN)) return -EPERM;
 855                        if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
 856                        if (invalid_vm86_irq(irq)) return -EPERM;
 857                        if (vm86_irqs[irq].tsk) return -EPERM;
 858                        ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
 859                        if (ret) return ret;
 860                        vm86_irqs[irq].sig = sig;
 861                        vm86_irqs[irq].tsk = current;
 862                        return irq;
 863                }
 864                case  VM86_FREE_IRQ: {
 865                        if (invalid_vm86_irq(irqnumber)) return -EPERM;
 866                        if (!vm86_irqs[irqnumber].tsk) return 0;
 867                        if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
 868                        free_vm86_irq(irqnumber);
 869                        return 0;
 870                }
 871        }
 872        return -EINVAL;
 873}
 874
 875