arch/x86/kernel/vm86_32.c

   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
  59 /*
  60  * Known problems:
  61  *
  62  * Interrupt handling is not guaranteed:
  63  * - a real x86 will disable all interrupts for one instruction
  64  *   after a "mov ss,xx" to make stack handling atomic even without
  65  *   the 'lss' instruction. We can't guarantee this in v86 mode,
  66  *   as the next instruction might result in a page fault or similar.
  67  * - a real x86 will have interrupts disabled for one instruction
  68  *   past the 'sti' that enables them. We don't bother with all the
  69  *   details yet.
  70  *
  71  * Let's hope these problems do not actually matter for anything.
  72  */
  73
  74
  75 /*
  76  * 8- and 16-bit register defines..
  77  */
  78 #define AL(regs)        (((unsigned char *)&((regs)->pt.ax))[0])
  79 #define AH(regs)        (((unsigned char *)&((regs)->pt.ax))[1])
  80 #define IP(regs)        (*(unsigned short *)&((regs)->pt.ip))
  81 #define SP(regs)        (*(unsigned short *)&((regs)->pt.sp))
  82
  83 /*
  84  * virtual flags (16 and 32-bit versions)
  85  */
  86 #define VFLAGS  (*(unsigned short *)&(current->thread.vm86->veflags))
  87 #define VEFLAGS (current->thread.vm86->veflags)
  88
  89 #define set_flags(X, new, mask) \
  90 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
  91
  92 #define SAFE_MASK       (0xDD5)
  93 #define RETURN_MASK     (0xDFF)
  94
  95 void save_v86_state(struct kernel_vm86_regs *regs, int retval)
  96 {
  97         struct tss_struct *tss;
  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(VERIFY_WRITE, 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         tss = &per_cpu(cpu_tss, get_cpu());
 150         tsk->thread.sp0 = vm86->saved_sp0;
 151         tsk->thread.sysenter_cs = __KERNEL_CS;
 152         load_sp0(tss, &tsk->thread);
 153         vm86->saved_sp0 = 0;
 154         put_cpu();
 155
 156         memcpy(&regs->pt, &vm86->regs32, sizeof(struct pt_regs));
 157
 158         lazy_load_gs(vm86->regs32.gs);
 159
 160         regs->pt.ax = retval;
 161 }
 162
 163 static void mark_screen_rdonly(struct mm_struct *mm)
 164 {
 165         struct vm_area_struct *vma;
 166         spinlock_t *ptl;
 167         pgd_t *pgd;
 168         p4d_t *p4d;
 169         pud_t *pud;
 170         pmd_t *pmd;
 171         pte_t *pte;
 172         int i;
 173
 174         down_write(&mm->mmap_sem);
 175         pgd = pgd_offset(mm, 0xA0000);
 176         if (pgd_none_or_clear_bad(pgd))
 177                 goto out;
 178         p4d = p4d_offset(pgd, 0xA0000);
 179         if (p4d_none_or_clear_bad(p4d))
 180                 goto out;
 181         pud = pud_offset(p4d, 0xA0000);
 182         if (pud_none_or_clear_bad(pud))
 183                 goto out;
 184         pmd = pmd_offset(pud, 0xA0000);
 185
 186         if (pmd_trans_huge(*pmd)) {
 187                 vma = find_vma(mm, 0xA0000);
 188                 split_huge_pmd(vma, pmd, 0xA0000);
 189         }
 190         if (pmd_none_or_clear_bad(pmd))
 191                 goto out;
 192         pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
 193         for (i = 0; i < 32; i++) {
 194                 if (pte_present(*pte))
 195                         set_pte(pte, pte_wrprotect(*pte));
 196                 pte++;
 197         }
 198         pte_unmap_unlock(pte, ptl);
 199 out:
 200         up_write(&mm->mmap_sem);
 201         flush_tlb_mm_range(mm, 0xA0000, 0xA0000 + 32*PAGE_SIZE, 0UL);
 202 }
 203
 204
 205
 206 static int do_vm86_irq_handling(int subfunction, int irqnumber);
 207 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus);
 208
 209 SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86)
 210 {
 211         return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false);
 212 }
 213
 214
 215 SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg)
 216 {
 217         switch (cmd) {
 218         case VM86_REQUEST_IRQ:
 219         case VM86_FREE_IRQ:
 220         case VM86_GET_IRQ_BITS:
 221         case VM86_GET_AND_RESET_IRQ:
 222                 return do_vm86_irq_handling(cmd, (int)arg);
 223         case VM86_PLUS_INSTALL_CHECK:
 224                 /*
 225                  * NOTE: on old vm86 stuff this will return the error
 226                  *  from access_ok(), because the subfunction is
 227                  *  interpreted as (invalid) address to vm86_struct.
 228                  *  So the installation check works.
 229                  */
 230                 return 0;
 231         }
 232
 233         /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
 234         return do_sys_vm86((struct vm86plus_struct __user *) arg, true);
 235 }
 236
 237
 238 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
 239 {
 240         struct tss_struct *tss;
 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(VERIFY_READ, 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         tss = &per_cpu(cpu_tss, get_cpu());
 369         /* make room for real-mode segments */
 370         tsk->thread.sp0 += 16;
 371
 372         if (static_cpu_has(X86_FEATURE_SEP))
 373                 tsk->thread.sysenter_cs = 0;
 374
 375         load_sp0(tss, &tsk->thread);
 376         put_cpu();
 377
 378         if (vm86->flags & VM86_SCREEN_BITMAP)
 379                 mark_screen_rdonly(tsk->mm);
 380
 381         memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs));
 382         force_iret();
 383         return regs->ax;
 384 }
 385
 386 static inline void set_IF(struct kernel_vm86_regs *regs)
 387 {
 388         VEFLAGS |= X86_EFLAGS_VIF;
 389 }
 390
 391 static inline void clear_IF(struct kernel_vm86_regs *regs)
 392 {
 393         VEFLAGS &= ~X86_EFLAGS_VIF;
 394 }
 395
 396 static inline void clear_TF(struct kernel_vm86_regs *regs)
 397 {
 398         regs->pt.flags &= ~X86_EFLAGS_TF;
 399 }
 400
 401 static inline void clear_AC(struct kernel_vm86_regs *regs)
 402 {
 403         regs->pt.flags &= ~X86_EFLAGS_AC;
 404 }
 405
 406 /*
 407  * It is correct to call set_IF(regs) from the set_vflags_*
 408  * functions. However someone forgot to call clear_IF(regs)
 409  * in the opposite case.
 410  * After the command sequence CLI PUSHF STI POPF you should
 411  * end up with interrupts disabled, but you ended up with
 412  * interrupts enabled.
 413  *  ( I was testing my own changes, but the only bug I
 414  *    could find was in a function I had not changed. )
 415  * [KD]
 416  */
 417
 418 static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
 419 {
 420         set_flags(VEFLAGS, flags, current->thread.vm86->veflags_mask);
 421         set_flags(regs->pt.flags, flags, SAFE_MASK);
 422         if (flags & X86_EFLAGS_IF)
 423                 set_IF(regs);
 424         else
 425                 clear_IF(regs);
 426 }
 427
 428 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
 429 {
 430         set_flags(VFLAGS, flags, current->thread.vm86->veflags_mask);
 431         set_flags(regs->pt.flags, flags, SAFE_MASK);
 432         if (flags & X86_EFLAGS_IF)
 433                 set_IF(regs);
 434         else
 435                 clear_IF(regs);
 436 }
 437
 438 static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
 439 {
 440         unsigned long flags = regs->pt.flags & RETURN_MASK;
 441
 442         if (VEFLAGS & X86_EFLAGS_VIF)
 443                 flags |= X86_EFLAGS_IF;
 444         flags |= X86_EFLAGS_IOPL;
 445         return flags | (VEFLAGS & current->thread.vm86->veflags_mask);
 446 }
 447
 448 static inline int is_revectored(int nr, struct revectored_struct *bitmap)
 449 {
 450         return test_bit(nr, bitmap->__map);
 451 }
 452
 453 #define val_byte(val, n) (((__u8 *)&val)[n])
 454
 455 #define pushb(base, ptr, val, err_label) \
 456         do { \
 457                 __u8 __val = val; \
 458                 ptr--; \
 459                 if (put_user(__val, base + ptr) < 0) \
 460                         goto err_label; \
 461         } while (0)
 462
 463 #define pushw(base, ptr, val, err_label) \
 464         do { \
 465                 __u16 __val = val; \
 466                 ptr--; \
 467                 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 468                         goto err_label; \
 469                 ptr--; \
 470                 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 471                         goto err_label; \
 472         } while (0)
 473
 474 #define pushl(base, ptr, val, err_label) \
 475         do { \
 476                 __u32 __val = val; \
 477                 ptr--; \
 478                 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
 479                         goto err_label; \
 480                 ptr--; \
 481                 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
 482                         goto err_label; \
 483                 ptr--; \
 484                 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 485                         goto err_label; \
 486                 ptr--; \
 487                 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 488                         goto err_label; \
 489         } while (0)
 490
 491 #define popb(base, ptr, err_label) \
 492         ({ \
 493                 __u8 __res; \
 494                 if (get_user(__res, base + ptr) < 0) \
 495                         goto err_label; \
 496                 ptr++; \
 497                 __res; \
 498         })
 499
 500 #define popw(base, ptr, err_label) \
 501         ({ \
 502                 __u16 __res; \
 503                 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 504                         goto err_label; \
 505                 ptr++; \
 506                 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 507                         goto err_label; \
 508                 ptr++; \
 509                 __res; \
 510         })
 511
 512 #define popl(base, ptr, err_label) \
 513         ({ \
 514                 __u32 __res; \
 515                 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 516                         goto err_label; \
 517                 ptr++; \
 518                 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 519                         goto err_label; \
 520                 ptr++; \
 521                 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
 522                         goto err_label; \
 523                 ptr++; \
 524                 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
 525                         goto err_label; \
 526                 ptr++; \
 527                 __res; \
 528         })
 529
 530 /* There are so many possible reasons for this function to return
 531  * VM86_INTx, so adding another doesn't bother me. We can expect
 532  * userspace programs to be able to handle it. (Getting a problem
 533  * in userspace is always better than an Oops anyway.) [KD]
 534  */
 535 static void do_int(struct kernel_vm86_regs *regs, int i,
 536     unsigned char __user *ssp, unsigned short sp)
 537 {
 538         unsigned long __user *intr_ptr;
 539         unsigned long segoffs;
 540         struct vm86 *vm86 = current->thread.vm86;
 541
 542         if (regs->pt.cs == BIOSSEG)
 543                 goto cannot_handle;
 544         if (is_revectored(i, &vm86->int_revectored))
 545                 goto cannot_handle;
 546         if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored))
 547                 goto cannot_handle;
 548         intr_ptr = (unsigned long __user *) (i << 2);
 549         if (get_user(segoffs, intr_ptr))
 550                 goto cannot_handle;
 551         if ((segoffs >> 16) == BIOSSEG)
 552                 goto cannot_handle;
 553         pushw(ssp, sp, get_vflags(regs), cannot_handle);
 554         pushw(ssp, sp, regs->pt.cs, cannot_handle);
 555         pushw(ssp, sp, IP(regs), cannot_handle);
 556         regs->pt.cs = segoffs >> 16;
 557         SP(regs) -= 6;
 558         IP(regs) = segoffs & 0xffff;
 559         clear_TF(regs);
 560         clear_IF(regs);
 561         clear_AC(regs);
 562         return;
 563
 564 cannot_handle:
 565         save_v86_state(regs, VM86_INTx + (i << 8));
 566 }
 567
 568 int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
 569 {
 570         struct vm86 *vm86 = current->thread.vm86;
 571
 572         if (vm86->vm86plus.is_vm86pus) {
 573                 if ((trapno == 3) || (trapno == 1)) {
 574                         save_v86_state(regs, VM86_TRAP + (trapno << 8));
 575                         return 0;
 576                 }
 577                 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
 578                 return 0;
 579         }
 580         if (trapno != 1)
 581                 return 1; /* we let this handle by the calling routine */
 582         current->thread.trap_nr = trapno;
 583         current->thread.error_code = error_code;
 584         force_sig(SIGTRAP, current);
 585         return 0;
 586 }
 587
 588 void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
 589 {
 590         unsigned char opcode;
 591         unsigned char __user *csp;
 592         unsigned char __user *ssp;
 593         unsigned short ip, sp, orig_flags;
 594         int data32, pref_done;
 595         struct vm86plus_info_struct *vmpi = &current->thread.vm86->vm86plus;
 596
 597 #define CHECK_IF_IN_TRAP \
 598         if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \
 599                 newflags |= X86_EFLAGS_TF
 600
 601         orig_flags = *(unsigned short *)&regs->pt.flags;
 602
 603         csp = (unsigned char __user *) (regs->pt.cs << 4);
 604         ssp = (unsigned char __user *) (regs->pt.ss << 4);
 605         sp = SP(regs);
 606         ip = IP(regs);
 607
 608         data32 = 0;
 609         pref_done = 0;
 610         do {
 611                 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
 612                 case 0x66:      /* 32-bit data */     data32 = 1; break;
 613                 case 0x67:      /* 32-bit address */  break;
 614                 case 0x2e:      /* CS */              break;
 615                 case 0x3e:      /* DS */              break;
 616                 case 0x26:      /* ES */              break;
 617                 case 0x36:      /* SS */              break;
 618                 case 0x65:      /* GS */              break;
 619                 case 0x64:      /* FS */              break;
 620                 case 0xf2:      /* repnz */       break;
 621                 case 0xf3:      /* rep */             break;
 622                 default: pref_done = 1;
 623                 }
 624         } while (!pref_done);
 625
 626         switch (opcode) {
 627
 628         /* pushf */
 629         case 0x9c:
 630                 if (data32) {
 631                         pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
 632                         SP(regs) -= 4;
 633                 } else {
 634                         pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
 635                         SP(regs) -= 2;
 636                 }
 637                 IP(regs) = ip;
 638                 goto vm86_fault_return;
 639
 640         /* popf */
 641         case 0x9d:
 642                 {
 643                 unsigned long newflags;
 644                 if (data32) {
 645                         newflags = popl(ssp, sp, simulate_sigsegv);
 646                         SP(regs) += 4;
 647                 } else {
 648                         newflags = popw(ssp, sp, simulate_sigsegv);
 649                         SP(regs) += 2;
 650                 }
 651                 IP(regs) = ip;
 652                 CHECK_IF_IN_TRAP;
 653                 if (data32)
 654                         set_vflags_long(newflags, regs);
 655                 else
 656                         set_vflags_short(newflags, regs);
 657
 658                 goto check_vip;
 659                 }
 660
 661         /* int xx */
 662         case 0xcd: {
 663                 int intno = popb(csp, ip, simulate_sigsegv);
 664                 IP(regs) = ip;
 665                 if (vmpi->vm86dbg_active) {
 666                         if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) {
 667                                 save_v86_state(regs, VM86_INTx + (intno << 8));
 668                                 return;
 669                         }
 670                 }
 671                 do_int(regs, intno, ssp, sp);
 672                 return;
 673         }
 674
 675         /* iret */
 676         case 0xcf:
 677                 {
 678                 unsigned long newip;
 679                 unsigned long newcs;
 680                 unsigned long newflags;
 681                 if (data32) {
 682                         newip = popl(ssp, sp, simulate_sigsegv);
 683                         newcs = popl(ssp, sp, simulate_sigsegv);
 684                         newflags = popl(ssp, sp, simulate_sigsegv);
 685                         SP(regs) += 12;
 686                 } else {
 687                         newip = popw(ssp, sp, simulate_sigsegv);
 688                         newcs = popw(ssp, sp, simulate_sigsegv);
 689                         newflags = popw(ssp, sp, simulate_sigsegv);
 690                         SP(regs) += 6;
 691                 }
 692                 IP(regs) = newip;
 693                 regs->pt.cs = newcs;
 694                 CHECK_IF_IN_TRAP;
 695                 if (data32) {
 696                         set_vflags_long(newflags, regs);
 697                 } else {
 698                         set_vflags_short(newflags, regs);
 699                 }
 700                 goto check_vip;
 701                 }
 702
 703         /* cli */
 704         case 0xfa:
 705                 IP(regs) = ip;
 706                 clear_IF(regs);
 707                 goto vm86_fault_return;
 708
 709         /* sti */
 710         /*
 711          * Damn. This is incorrect: the 'sti' instruction should actually
 712          * enable interrupts after the /next/ instruction. Not good.
 713          *
 714          * Probably needs some horsing around with the TF flag. Aiee..
 715          */
 716         case 0xfb:
 717                 IP(regs) = ip;
 718                 set_IF(regs);
 719                 goto check_vip;
 720
 721         default:
 722                 save_v86_state(regs, VM86_UNKNOWN);
 723         }
 724
 725         return;
 726
 727 check_vip:
 728         if (VEFLAGS & X86_EFLAGS_VIP) {
 729                 save_v86_state(regs, VM86_STI);
 730                 return;
 731         }
 732
 733 vm86_fault_return:
 734         if (vmpi->force_return_for_pic  && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) {
 735                 save_v86_state(regs, VM86_PICRETURN);
 736                 return;
 737         }
 738         if (orig_flags & X86_EFLAGS_TF)
 739                 handle_vm86_trap(regs, 0, X86_TRAP_DB);
 740         return;
 741
 742 simulate_sigsegv:
 743         /* FIXME: After a long discussion with Stas we finally
 744          *        agreed, that this is wrong. Here we should
 745          *        really send a SIGSEGV to the user program.
 746          *        But how do we create the correct context? We
 747          *        are inside a general protection fault handler
 748          *        and has just returned from a page fault handler.
 749          *        The correct context for the signal handler
 750          *        should be a mixture of the two, but how do we
 751          *        get the information? [KD]
 752          */
 753         save_v86_state(regs, VM86_UNKNOWN);
 754 }
 755
 756 /* ---------------- vm86 special IRQ passing stuff ----------------- */
 757
 758 #define VM86_IRQNAME            "vm86irq"
 759
 760 static struct vm86_irqs {
 761         struct task_struct *tsk;
 762         int sig;
 763 } vm86_irqs[16];
 764
 765 static DEFINE_SPINLOCK(irqbits_lock);
 766 static int irqbits;
 767
 768 #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
 769         | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO)  | (1 << SIGURG) \
 770         | (1 << SIGUNUSED))
 771
 772 static irqreturn_t irq_handler(int intno, void *dev_id)
 773 {
 774         int irq_bit;
 775         unsigned long flags;
 776
 777         spin_lock_irqsave(&irqbits_lock, flags);
 778         irq_bit = 1 << intno;
 779         if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
 780                 goto out;
 781         irqbits |= irq_bit;
 782         if (vm86_irqs[intno].sig)
 783                 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
 784         /*
 785          * IRQ will be re-enabled when user asks for the irq (whether
 786          * polling or as a result of the signal)
 787          */
 788         disable_irq_nosync(intno);
 789         spin_unlock_irqrestore(&irqbits_lock, flags);
 790         return IRQ_HANDLED;
 791
 792 out:
 793         spin_unlock_irqrestore(&irqbits_lock, flags);
 794         return IRQ_NONE;
 795 }
 796
 797 static inline void free_vm86_irq(int irqnumber)
 798 {
 799         unsigned long flags;
 800
 801         free_irq(irqnumber, NULL);
 802         vm86_irqs[irqnumber].tsk = NULL;
 803
 804         spin_lock_irqsave(&irqbits_lock, flags);
 805         irqbits &= ~(1 << irqnumber);
 806         spin_unlock_irqrestore(&irqbits_lock, flags);
 807 }
 808
 809 void release_vm86_irqs(struct task_struct *task)
 810 {
 811         int i;
 812         for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
 813             if (vm86_irqs[i].tsk == task)
 814                 free_vm86_irq(i);
 815 }
 816
 817 static inline int get_and_reset_irq(int irqnumber)
 818 {
 819         int bit;
 820         unsigned long flags;
 821         int ret = 0;
 822
 823         if (invalid_vm86_irq(irqnumber)) return 0;
 824         if (vm86_irqs[irqnumber].tsk != current) return 0;
 825         spin_lock_irqsave(&irqbits_lock, flags);
 826         bit = irqbits & (1 << irqnumber);
 827         irqbits &= ~bit;
 828         if (bit) {
 829                 enable_irq(irqnumber);
 830                 ret = 1;
 831         }
 832
 833         spin_unlock_irqrestore(&irqbits_lock, flags);
 834         return ret;
 835 }
 836
 837
 838 static int do_vm86_irq_handling(int subfunction, int irqnumber)
 839 {
 840         int ret;
 841         switch (subfunction) {
 842                 case VM86_GET_AND_RESET_IRQ: {
 843                         return get_and_reset_irq(irqnumber);
 844                 }
 845                 case VM86_GET_IRQ_BITS: {
 846                         return irqbits;
 847                 }
 848                 case VM86_REQUEST_IRQ: {
 849                         int sig = irqnumber >> 8;
 850                         int irq = irqnumber & 255;
 851                         if (!capable(CAP_SYS_ADMIN)) return -EPERM;
 852                         if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
 853                         if (invalid_vm86_irq(irq)) return -EPERM;
 854                         if (vm86_irqs[irq].tsk) return -EPERM;
 855                         ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
 856                         if (ret) return ret;
 857                         vm86_irqs[irq].sig = sig;
 858                         vm86_irqs[irq].tsk = current;
 859                         return irq;
 860                 }
 861                 case  VM86_FREE_IRQ: {
 862                         if (invalid_vm86_irq(irqnumber)) return -EPERM;
 863                         if (!vm86_irqs[irqnumber].tsk) return 0;
 864                         if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
 865                         free_vm86_irq(irqnumber);
 866                         return 0;
 867                 }
 868         }
 869         return -EINVAL;
 870 }
 871