arch/x86/kernel/vm86_32.c

   1 /*
   2  *  Copyright (C) 1994  Linus Torvalds
   3  *
   4  *  29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
   5  *                stack - Manfred Spraul <manfred@colorfullife.com>
   6  *
   7  *  22 mar 2002 - Manfred detected the stackfaults, but didn't handle
   8  *                them correctly. Now the emulation will be in a
   9  *                consistent state after stackfaults - Kasper Dupont
  10  *                <kasperd@daimi.au.dk>
  11  *
  12  *  22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
  13  *                <kasperd@daimi.au.dk>
  14  *
  15  *  ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
  16  *                caused by Kasper Dupont's changes - Stas Sergeev
  17  *
  18  *   4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
  19  *                Kasper Dupont <kasperd@daimi.au.dk>
  20  *
  21  *   9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
  22  *                Kasper Dupont <kasperd@daimi.au.dk>
  23  *
  24  *   9 apr 2002 - Changed stack access macros to jump to a label
  25  *                instead of returning to userspace. This simplifies
  26  *                do_int, and is needed by handle_vm6_fault. Kasper
  27  *                Dupont <kasperd@daimi.au.dk>
  28  *
  29  */
  30
  31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  32
  33 #include <linux/capability.h>
  34 #include <linux/errno.h>
  35 #include <linux/interrupt.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/sched.h>
  38 #include <linux/sched/task_stack.h>
  39 #include <linux/kernel.h>
  40 #include <linux/signal.h>
  41 #include <linux/string.h>
  42 #include <linux/mm.h>
  43 #include <linux/smp.h>
  44 #include <linux/highmem.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/audit.h>
  47 #include <linux/stddef.h>
  48 #include <linux/slab.h>
  49 #include <linux/security.h>
  50
  51 #include <linux/uaccess.h>
  52 #include <asm/io.h>
  53 #include <asm/tlbflush.h>
  54 #include <asm/irq.h>
  55 #include <asm/traps.h>
  56 #include <asm/vm86.h>
  57
  58 /*
  59  * Known problems:
  60  *
  61  * Interrupt handling is not guaranteed:
  62  * - a real x86 will disable all interrupts for one instruction
  63  *   after a "mov ss,xx" to make stack handling atomic even without
  64  *   the 'lss' instruction. We can't guarantee this in v86 mode,
  65  *   as the next instruction might result in a page fault or similar.
  66  * - a real x86 will have interrupts disabled for one instruction
  67  *   past the 'sti' that enables them. We don't bother with all the
  68  *   details yet.
  69  *
  70  * Let's hope these problems do not actually matter for anything.
  71  */
  72
  73
  74 /*
  75  * 8- and 16-bit register defines..
  76  */
  77 #define AL(regs)        (((unsigned char *)&((regs)->pt.ax))[0])
  78 #define AH(regs)        (((unsigned char *)&((regs)->pt.ax))[1])
  79 #define IP(regs)        (*(unsigned short *)&((regs)->pt.ip))
  80 #define SP(regs)        (*(unsigned short *)&((regs)->pt.sp))
  81
  82 /*
  83  * virtual flags (16 and 32-bit versions)
  84  */
  85 #define VFLAGS  (*(unsigned short *)&(current->thread.vm86->veflags))
  86 #define VEFLAGS (current->thread.vm86->veflags)
  87
  88 #define set_flags(X, new, mask) \
  89 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
  90
  91 #define SAFE_MASK       (0xDD5)
  92 #define RETURN_MASK     (0xDFF)
  93
  94 void save_v86_state(struct kernel_vm86_regs *regs, int retval)
  95 {
  96         struct tss_struct *tss;
  97         struct task_struct *tsk = current;
  98         struct vm86plus_struct __user *user;
  99         struct vm86 *vm86 = current->thread.vm86;
 100         long err = 0;
 101
 102         /*
 103          * This gets called from entry.S with interrupts disabled, but
 104          * from process context. Enable interrupts here, before trying
 105          * to access user space.
 106          */
 107         local_irq_enable();
 108
 109         if (!vm86 || !vm86->user_vm86) {
 110                 pr_alert("no user_vm86: BAD\n");
 111                 do_exit(SIGSEGV);
 112         }
 113         set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | vm86->veflags_mask);
 114         user = vm86->user_vm86;
 115
 116         if (!access_ok(VERIFY_WRITE, user, vm86->vm86plus.is_vm86pus ?
 117                        sizeof(struct vm86plus_struct) :
 118                        sizeof(struct vm86_struct))) {
 119                 pr_alert("could not access userspace vm86 info\n");
 120                 do_exit(SIGSEGV);
 121         }
 122
 123         put_user_try {
 124                 put_user_ex(regs->pt.bx, &user->regs.ebx);
 125                 put_user_ex(regs->pt.cx, &user->regs.ecx);
 126                 put_user_ex(regs->pt.dx, &user->regs.edx);
 127                 put_user_ex(regs->pt.si, &user->regs.esi);
 128                 put_user_ex(regs->pt.di, &user->regs.edi);
 129                 put_user_ex(regs->pt.bp, &user->regs.ebp);
 130                 put_user_ex(regs->pt.ax, &user->regs.eax);
 131                 put_user_ex(regs->pt.ip, &user->regs.eip);
 132                 put_user_ex(regs->pt.cs, &user->regs.cs);
 133                 put_user_ex(regs->pt.flags, &user->regs.eflags);
 134                 put_user_ex(regs->pt.sp, &user->regs.esp);
 135                 put_user_ex(regs->pt.ss, &user->regs.ss);
 136                 put_user_ex(regs->es, &user->regs.es);
 137                 put_user_ex(regs->ds, &user->regs.ds);
 138                 put_user_ex(regs->fs, &user->regs.fs);
 139                 put_user_ex(regs->gs, &user->regs.gs);
 140
 141                 put_user_ex(vm86->screen_bitmap, &user->screen_bitmap);
 142         } put_user_catch(err);
 143         if (err) {
 144                 pr_alert("could not access userspace vm86 info\n");
 145                 do_exit(SIGSEGV);
 146         }
 147
 148         tss = &per_cpu(cpu_tss, get_cpu());
 149         tsk->thread.sp0 = vm86->saved_sp0;
 150         tsk->thread.sysenter_cs = __KERNEL_CS;
 151         load_sp0(tss, &tsk->thread);
 152         vm86->saved_sp0 = 0;
 153         put_cpu();
 154
 155         memcpy(&regs->pt, &vm86->regs32, sizeof(struct pt_regs));
 156
 157         lazy_load_gs(vm86->regs32.gs);
 158
 159         regs->pt.ax = retval;
 160 }
 161
 162 static void mark_screen_rdonly(struct mm_struct *mm)
 163 {
 164         struct vm_area_struct *vma;
 165         spinlock_t *ptl;
 166         pgd_t *pgd;
 167         p4d_t *p4d;
 168         pud_t *pud;
 169         pmd_t *pmd;
 170         pte_t *pte;
 171         int i;
 172
 173         down_write(&mm->mmap_sem);
 174         pgd = pgd_offset(mm, 0xA0000);
 175         if (pgd_none_or_clear_bad(pgd))
 176                 goto out;
 177         p4d = p4d_offset(pgd, 0xA0000);
 178         if (p4d_none_or_clear_bad(p4d))
 179                 goto out;
 180         pud = pud_offset(p4d, 0xA0000);
 181         if (pud_none_or_clear_bad(pud))
 182                 goto out;
 183         pmd = pmd_offset(pud, 0xA0000);
 184
 185         if (pmd_trans_huge(*pmd)) {
 186                 vma = find_vma(mm, 0xA0000);
 187                 split_huge_pmd(vma, pmd, 0xA0000);
 188         }
 189         if (pmd_none_or_clear_bad(pmd))
 190                 goto out;
 191         pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
 192         for (i = 0; i < 32; i++) {
 193                 if (pte_present(*pte))
 194                         set_pte(pte, pte_wrprotect(*pte));
 195                 pte++;
 196         }
 197         pte_unmap_unlock(pte, ptl);
 198 out:
 199         up_write(&mm->mmap_sem);
 200         flush_tlb_mm_range(mm, 0xA0000, 0xA0000 + 32*PAGE_SIZE, 0UL);
 201 }
 202
 203
 204
 205 static int do_vm86_irq_handling(int subfunction, int irqnumber);
 206 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus);
 207
 208 SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86)
 209 {
 210         return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false);
 211 }
 212
 213
 214 SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg)
 215 {
 216         switch (cmd) {
 217         case VM86_REQUEST_IRQ:
 218         case VM86_FREE_IRQ:
 219         case VM86_GET_IRQ_BITS:
 220         case VM86_GET_AND_RESET_IRQ:
 221                 return do_vm86_irq_handling(cmd, (int)arg);
 222         case VM86_PLUS_INSTALL_CHECK:
 223                 /*
 224                  * NOTE: on old vm86 stuff this will return the error
 225                  *  from access_ok(), because the subfunction is
 226                  *  interpreted as (invalid) address to vm86_struct.
 227                  *  So the installation check works.
 228                  */
 229                 return 0;
 230         }
 231
 232         /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
 233         return do_sys_vm86((struct vm86plus_struct __user *) arg, true);
 234 }
 235
 236
 237 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
 238 {
 239         struct tss_struct *tss;
 240         struct task_struct *tsk = current;
 241         struct vm86 *vm86 = tsk->thread.vm86;
 242         struct kernel_vm86_regs vm86regs;
 243         struct pt_regs *regs = current_pt_regs();
 244         unsigned long err = 0;
 245
 246         err = security_mmap_addr(0);
 247         if (err) {
 248                 /*
 249                  * vm86 cannot virtualize the address space, so vm86 users
 250                  * need to manage the low 1MB themselves using mmap.  Given
 251                  * that BIOS places important data in the first page, vm86
 252                  * is essentially useless if mmap_min_addr != 0.  DOSEMU,
 253                  * for example, won't even bother trying to use vm86 if it
 254                  * can't map a page at virtual address 0.
 255                  *
 256                  * To reduce the available kernel attack surface, simply
 257                  * disallow vm86(old) for users who cannot mmap at va 0.
 258                  *
 259                  * The implementation of security_mmap_addr will allow
 260                  * suitably privileged users to map va 0 even if
 261                  * vm.mmap_min_addr is set above 0, and we want this
 262                  * behavior for vm86 as well, as it ensures that legacy
 263                  * tools like vbetool will not fail just because of
 264                  * vm.mmap_min_addr.
 265                  */
 266                 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",
 267                              current->comm, task_pid_nr(current),
 268                              from_kuid_munged(&init_user_ns, current_uid()));
 269                 return -EPERM;
 270         }
 271
 272         if (!vm86) {
 273                 if (!(vm86 = kzalloc(sizeof(*vm86), GFP_KERNEL)))
 274                         return -ENOMEM;
 275                 tsk->thread.vm86 = vm86;
 276         }
 277         if (vm86->saved_sp0)
 278                 return -EPERM;
 279
 280         if (!access_ok(VERIFY_READ, user_vm86, plus ?
 281                        sizeof(struct vm86_struct) :
 282                        sizeof(struct vm86plus_struct)))
 283                 return -EFAULT;
 284
 285         memset(&vm86regs, 0, sizeof(vm86regs));
 286         get_user_try {
 287                 unsigned short seg;
 288                 get_user_ex(vm86regs.pt.bx, &user_vm86->regs.ebx);
 289                 get_user_ex(vm86regs.pt.cx, &user_vm86->regs.ecx);
 290                 get_user_ex(vm86regs.pt.dx, &user_vm86->regs.edx);
 291                 get_user_ex(vm86regs.pt.si, &user_vm86->regs.esi);
 292                 get_user_ex(vm86regs.pt.di, &user_vm86->regs.edi);
 293                 get_user_ex(vm86regs.pt.bp, &user_vm86->regs.ebp);
 294                 get_user_ex(vm86regs.pt.ax, &user_vm86->regs.eax);
 295                 get_user_ex(vm86regs.pt.ip, &user_vm86->regs.eip);
 296                 get_user_ex(seg, &user_vm86->regs.cs);
 297                 vm86regs.pt.cs = seg;
 298                 get_user_ex(vm86regs.pt.flags, &user_vm86->regs.eflags);
 299                 get_user_ex(vm86regs.pt.sp, &user_vm86->regs.esp);
 300                 get_user_ex(seg, &user_vm86->regs.ss);
 301                 vm86regs.pt.ss = seg;
 302                 get_user_ex(vm86regs.es, &user_vm86->regs.es);
 303                 get_user_ex(vm86regs.ds, &user_vm86->regs.ds);
 304                 get_user_ex(vm86regs.fs, &user_vm86->regs.fs);
 305                 get_user_ex(vm86regs.gs, &user_vm86->regs.gs);
 306
 307                 get_user_ex(vm86->flags, &user_vm86->flags);
 308                 get_user_ex(vm86->screen_bitmap, &user_vm86->screen_bitmap);
 309                 get_user_ex(vm86->cpu_type, &user_vm86->cpu_type);
 310         } get_user_catch(err);
 311         if (err)
 312                 return err;
 313
 314         if (copy_from_user(&vm86->int_revectored,
 315                            &user_vm86->int_revectored,
 316                            sizeof(struct revectored_struct)))
 317                 return -EFAULT;
 318         if (copy_from_user(&vm86->int21_revectored,
 319                            &user_vm86->int21_revectored,
 320                            sizeof(struct revectored_struct)))
 321                 return -EFAULT;
 322         if (plus) {
 323                 if (copy_from_user(&vm86->vm86plus, &user_vm86->vm86plus,
 324                                    sizeof(struct vm86plus_info_struct)))
 325                         return -EFAULT;
 326                 vm86->vm86plus.is_vm86pus = 1;
 327         } else
 328                 memset(&vm86->vm86plus, 0,
 329                        sizeof(struct vm86plus_info_struct));
 330
 331         memcpy(&vm86->regs32, regs, sizeof(struct pt_regs));
 332         vm86->user_vm86 = user_vm86;
 333
 334 /*
 335  * The flags register is also special: we cannot trust that the user
 336  * has set it up safely, so this makes sure interrupt etc flags are
 337  * inherited from protected mode.
 338  */
 339         VEFLAGS = vm86regs.pt.flags;
 340         vm86regs.pt.flags &= SAFE_MASK;
 341         vm86regs.pt.flags |= regs->flags & ~SAFE_MASK;
 342         vm86regs.pt.flags |= X86_VM_MASK;
 343
 344         vm86regs.pt.orig_ax = regs->orig_ax;
 345
 346         switch (vm86->cpu_type) {
 347         case CPU_286:
 348                 vm86->veflags_mask = 0;
 349                 break;
 350         case CPU_386:
 351                 vm86->veflags_mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL;
 352                 break;
 353         case CPU_486:
 354                 vm86->veflags_mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
 355                 break;
 356         default:
 357                 vm86->veflags_mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
 358                 break;
 359         }
 360
 361 /*
 362  * Save old state
 363  */
 364         vm86->saved_sp0 = tsk->thread.sp0;
 365         lazy_save_gs(vm86->regs32.gs);
 366
 367         tss = &per_cpu(cpu_tss, get_cpu());
 368         /* make room for real-mode segments */
 369         tsk->thread.sp0 += 16;
 370
 371         if (static_cpu_has(X86_FEATURE_SEP))
 372                 tsk->thread.sysenter_cs = 0;
 373
 374         load_sp0(tss, &tsk->thread);
 375         put_cpu();
 376
 377         if (vm86->flags & VM86_SCREEN_BITMAP)
 378                 mark_screen_rdonly(tsk->mm);
 379
 380         memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs));
 381         force_iret();
 382         return regs->ax;
 383 }
 384
 385 static inline void set_IF(struct kernel_vm86_regs *regs)
 386 {
 387         VEFLAGS |= X86_EFLAGS_VIF;
 388 }
 389
 390 static inline void clear_IF(struct kernel_vm86_regs *regs)
 391 {
 392         VEFLAGS &= ~X86_EFLAGS_VIF;
 393 }
 394
 395 static inline void clear_TF(struct kernel_vm86_regs *regs)
 396 {
 397         regs->pt.flags &= ~X86_EFLAGS_TF;
 398 }
 399
 400 static inline void clear_AC(struct kernel_vm86_regs *regs)
 401 {
 402         regs->pt.flags &= ~X86_EFLAGS_AC;
 403 }
 404
 405 /*
 406  * It is correct to call set_IF(regs) from the set_vflags_*
 407  * functions. However someone forgot to call clear_IF(regs)
 408  * in the opposite case.
 409  * After the command sequence CLI PUSHF STI POPF you should
 410  * end up with interrupts disabled, but you ended up with
 411  * interrupts enabled.
 412  *  ( I was testing my own changes, but the only bug I
 413  *    could find was in a function I had not changed. )
 414  * [KD]
 415  */
 416
 417 static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
 418 {
 419         set_flags(VEFLAGS, flags, current->thread.vm86->veflags_mask);
 420         set_flags(regs->pt.flags, flags, SAFE_MASK);
 421         if (flags & X86_EFLAGS_IF)
 422                 set_IF(regs);
 423         else
 424                 clear_IF(regs);
 425 }
 426
 427 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
 428 {
 429         set_flags(VFLAGS, flags, current->thread.vm86->veflags_mask);
 430         set_flags(regs->pt.flags, flags, SAFE_MASK);
 431         if (flags & X86_EFLAGS_IF)
 432                 set_IF(regs);
 433         else
 434                 clear_IF(regs);
 435 }
 436
 437 static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
 438 {
 439         unsigned long flags = regs->pt.flags & RETURN_MASK;
 440
 441         if (VEFLAGS & X86_EFLAGS_VIF)
 442                 flags |= X86_EFLAGS_IF;
 443         flags |= X86_EFLAGS_IOPL;
 444         return flags | (VEFLAGS & current->thread.vm86->veflags_mask);
 445 }
 446
 447 static inline int is_revectored(int nr, struct revectored_struct *bitmap)
 448 {
 449         return test_bit(nr, bitmap->__map);
 450 }
 451
 452 #define val_byte(val, n) (((__u8 *)&val)[n])
 453
 454 #define pushb(base, ptr, val, err_label) \
 455         do { \
 456                 __u8 __val = val; \
 457                 ptr--; \
 458                 if (put_user(__val, base + ptr) < 0) \
 459                         goto err_label; \
 460         } while (0)
 461
 462 #define pushw(base, ptr, val, err_label) \
 463         do { \
 464                 __u16 __val = val; \
 465                 ptr--; \
 466                 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 467                         goto err_label; \
 468                 ptr--; \
 469                 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 470                         goto err_label; \
 471         } while (0)
 472
 473 #define pushl(base, ptr, val, err_label) \
 474         do { \
 475                 __u32 __val = val; \
 476                 ptr--; \
 477                 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
 478                         goto err_label; \
 479                 ptr--; \
 480                 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
 481                         goto err_label; \
 482                 ptr--; \
 483                 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
 484                         goto err_label; \
 485                 ptr--; \
 486                 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
 487                         goto err_label; \
 488         } while (0)
 489
 490 #define popb(base, ptr, err_label) \
 491         ({ \
 492                 __u8 __res; \
 493                 if (get_user(__res, base + ptr) < 0) \
 494                         goto err_label; \
 495                 ptr++; \
 496                 __res; \
 497         })
 498
 499 #define popw(base, ptr, err_label) \
 500         ({ \
 501                 __u16 __res; \
 502                 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 503                         goto err_label; \
 504                 ptr++; \
 505                 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 506                         goto err_label; \
 507                 ptr++; \
 508                 __res; \
 509         })
 510
 511 #define popl(base, ptr, err_label) \
 512         ({ \
 513                 __u32 __res; \
 514                 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
 515                         goto err_label; \
 516                 ptr++; \
 517                 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
 518                         goto err_label; \
 519                 ptr++; \
 520                 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
 521                         goto err_label; \
 522                 ptr++; \
 523                 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
 524                         goto err_label; \
 525                 ptr++; \
 526                 __res; \
 527         })
 528
 529 /* There are so many possible reasons for this function to return
 530  * VM86_INTx, so adding another doesn't bother me. We can expect
 531  * userspace programs to be able to handle it. (Getting a problem
 532  * in userspace is always better than an Oops anyway.) [KD]
 533  */
 534 static void do_int(struct kernel_vm86_regs *regs, int i,
 535     unsigned char __user *ssp, unsigned short sp)
 536 {
 537         unsigned long __user *intr_ptr;
 538         unsigned long segoffs;
 539         struct vm86 *vm86 = current->thread.vm86;
 540
 541         if (regs->pt.cs == BIOSSEG)
 542                 goto cannot_handle;
 543         if (is_revectored(i, &vm86->int_revectored))
 544                 goto cannot_handle;
 545         if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored))
 546                 goto cannot_handle;
 547         intr_ptr = (unsigned long __user *) (i << 2);
 548         if (get_user(segoffs, intr_ptr))
 549                 goto cannot_handle;
 550         if ((segoffs >> 16) == BIOSSEG)
 551                 goto cannot_handle;
 552         pushw(ssp, sp, get_vflags(regs), cannot_handle);
 553         pushw(ssp, sp, regs->pt.cs, cannot_handle);
 554         pushw(ssp, sp, IP(regs), cannot_handle);
 555         regs->pt.cs = segoffs >> 16;
 556         SP(regs) -= 6;
 557         IP(regs) = segoffs & 0xffff;
 558         clear_TF(regs);
 559         clear_IF(regs);
 560         clear_AC(regs);
 561         return;
 562
 563 cannot_handle:
 564         save_v86_state(regs, VM86_INTx + (i << 8));
 565 }
 566
 567 int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
 568 {
 569         struct vm86 *vm86 = current->thread.vm86;
 570
 571         if (vm86->vm86plus.is_vm86pus) {
 572                 if ((trapno == 3) || (trapno == 1)) {
 573                         save_v86_state(regs, VM86_TRAP + (trapno << 8));
 574                         return 0;
 575                 }
 576                 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
 577                 return 0;
 578         }
 579         if (trapno != 1)
 580                 return 1; /* we let this handle by the calling routine */
 581         current->thread.trap_nr = trapno;
 582         current->thread.error_code = error_code;
 583         force_sig(SIGTRAP, current);
 584         return 0;
 585 }
 586
 587 void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
 588 {
 589         unsigned char opcode;
 590         unsigned char __user *csp;
 591         unsigned char __user *ssp;
 592         unsigned short ip, sp, orig_flags;
 593         int data32, pref_done;
 594         struct vm86plus_info_struct *vmpi = &current->thread.vm86->vm86plus;
 595
 596 #define CHECK_IF_IN_TRAP \
 597         if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \
 598                 newflags |= X86_EFLAGS_TF
 599
 600         orig_flags = *(unsigned short *)&regs->pt.flags;
 601
 602         csp = (unsigned char __user *) (regs->pt.cs << 4);
 603         ssp = (unsigned char __user *) (regs->pt.ss << 4);
 604         sp = SP(regs);
 605         ip = IP(regs);
 606
 607         data32 = 0;
 608         pref_done = 0;
 609         do {
 610                 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
 611                 case 0x66:      /* 32-bit data */     data32 = 1; break;
 612                 case 0x67:      /* 32-bit address */  break;
 613                 case 0x2e:      /* CS */              break;
 614                 case 0x3e:      /* DS */              break;
 615                 case 0x26:      /* ES */              break;
 616                 case 0x36:      /* SS */              break;
 617                 case 0x65:      /* GS */              break;
 618                 case 0x64:      /* FS */              break;
 619                 case 0xf2:      /* repnz */       break;
 620                 case 0xf3:      /* rep */             break;
 621                 default: pref_done = 1;
 622                 }
 623         } while (!pref_done);
 624
 625         switch (opcode) {
 626
 627         /* pushf */
 628         case 0x9c:
 629                 if (data32) {
 630                         pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
 631                         SP(regs) -= 4;
 632                 } else {
 633                         pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
 634                         SP(regs) -= 2;
 635                 }
 636                 IP(regs) = ip;
 637                 goto vm86_fault_return;
 638
 639         /* popf */
 640         case 0x9d:
 641                 {
 642                 unsigned long newflags;
 643                 if (data32) {
 644                         newflags = popl(ssp, sp, simulate_sigsegv);
 645                         SP(regs) += 4;
 646                 } else {
 647                         newflags = popw(ssp, sp, simulate_sigsegv);
 648                         SP(regs) += 2;
 649                 }
 650                 IP(regs) = ip;
 651                 CHECK_IF_IN_TRAP;
 652                 if (data32)
 653                         set_vflags_long(newflags, regs);
 654                 else
 655                         set_vflags_short(newflags, regs);
 656
 657                 goto check_vip;
 658                 }
 659
 660         /* int xx */
 661         case 0xcd: {
 662                 int intno = popb(csp, ip, simulate_sigsegv);
 663                 IP(regs) = ip;
 664                 if (vmpi->vm86dbg_active) {
 665                         if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) {
 666                                 save_v86_state(regs, VM86_INTx + (intno << 8));
 667                                 return;
 668                         }
 669                 }
 670                 do_int(regs, intno, ssp, sp);
 671                 return;
 672         }
 673
 674         /* iret */
 675         case 0xcf:
 676                 {
 677                 unsigned long newip;
 678                 unsigned long newcs;
 679                 unsigned long newflags;
 680                 if (data32) {
 681                         newip = popl(ssp, sp, simulate_sigsegv);
 682                         newcs = popl(ssp, sp, simulate_sigsegv);
 683                         newflags = popl(ssp, sp, simulate_sigsegv);
 684                         SP(regs) += 12;
 685                 } else {
 686                         newip = popw(ssp, sp, simulate_sigsegv);
 687                         newcs = popw(ssp, sp, simulate_sigsegv);
 688                         newflags = popw(ssp, sp, simulate_sigsegv);
 689                         SP(regs) += 6;
 690                 }
 691                 IP(regs) = newip;
 692                 regs->pt.cs = newcs;
 693                 CHECK_IF_IN_TRAP;
 694                 if (data32) {
 695                         set_vflags_long(newflags, regs);
 696                 } else {
 697                         set_vflags_short(newflags, regs);
 698                 }
 699                 goto check_vip;
 700                 }
 701
 702         /* cli */
 703         case 0xfa:
 704                 IP(regs) = ip;
 705                 clear_IF(regs);
 706                 goto vm86_fault_return;
 707
 708         /* sti */
 709         /*
 710          * Damn. This is incorrect: the 'sti' instruction should actually
 711          * enable interrupts after the /next/ instruction. Not good.
 712          *
 713          * Probably needs some horsing around with the TF flag. Aiee..
 714          */
 715         case 0xfb:
 716                 IP(regs) = ip;
 717                 set_IF(regs);
 718                 goto check_vip;
 719
 720         default:
 721                 save_v86_state(regs, VM86_UNKNOWN);
 722         }
 723
 724         return;
 725
 726 check_vip:
 727         if (VEFLAGS & X86_EFLAGS_VIP) {
 728                 save_v86_state(regs, VM86_STI);
 729                 return;
 730         }
 731
 732 vm86_fault_return:
 733         if (vmpi->force_return_for_pic  && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) {
 734                 save_v86_state(regs, VM86_PICRETURN);
 735                 return;
 736         }
 737         if (orig_flags & X86_EFLAGS_TF)
 738                 handle_vm86_trap(regs, 0, X86_TRAP_DB);
 739         return;
 740
 741 simulate_sigsegv:
 742         /* FIXME: After a long discussion with Stas we finally
 743          *        agreed, that this is wrong. Here we should
 744          *        really send a SIGSEGV to the user program.
 745          *        But how do we create the correct context? We
 746          *        are inside a general protection fault handler
 747          *        and has just returned from a page fault handler.
 748          *        The correct context for the signal handler
 749          *        should be a mixture of the two, but how do we
 750          *        get the information? [KD]
 751          */
 752         save_v86_state(regs, VM86_UNKNOWN);
 753 }
 754
 755 /* ---------------- vm86 special IRQ passing stuff ----------------- */
 756
 757 #define VM86_IRQNAME            "vm86irq"
 758
 759 static struct vm86_irqs {
 760         struct task_struct *tsk;
 761         int sig;
 762 } vm86_irqs[16];
 763
 764 static DEFINE_SPINLOCK(irqbits_lock);
 765 static int irqbits;
 766
 767 #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
 768         | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO)  | (1 << SIGURG) \
 769         | (1 << SIGUNUSED))
 770
 771 static irqreturn_t irq_handler(int intno, void *dev_id)
 772 {
 773         int irq_bit;
 774         unsigned long flags;
 775
 776         spin_lock_irqsave(&irqbits_lock, flags);
 777         irq_bit = 1 << intno;
 778         if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
 779                 goto out;
 780         irqbits |= irq_bit;
 781         if (vm86_irqs[intno].sig)
 782                 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
 783         /*
 784          * IRQ will be re-enabled when user asks for the irq (whether
 785          * polling or as a result of the signal)
 786          */
 787         disable_irq_nosync(intno);
 788         spin_unlock_irqrestore(&irqbits_lock, flags);
 789         return IRQ_HANDLED;
 790
 791 out:
 792         spin_unlock_irqrestore(&irqbits_lock, flags);
 793         return IRQ_NONE;
 794 }
 795
 796 static inline void free_vm86_irq(int irqnumber)
 797 {
 798         unsigned long flags;
 799
 800         free_irq(irqnumber, NULL);
 801         vm86_irqs[irqnumber].tsk = NULL;
 802
 803         spin_lock_irqsave(&irqbits_lock, flags);
 804         irqbits &= ~(1 << irqnumber);
 805         spin_unlock_irqrestore(&irqbits_lock, flags);
 806 }
 807
 808 void release_vm86_irqs(struct task_struct *task)
 809 {
 810         int i;
 811         for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
 812             if (vm86_irqs[i].tsk == task)
 813                 free_vm86_irq(i);
 814 }
 815
 816 static inline int get_and_reset_irq(int irqnumber)
 817 {
 818         int bit;
 819         unsigned long flags;
 820         int ret = 0;
 821
 822         if (invalid_vm86_irq(irqnumber)) return 0;
 823         if (vm86_irqs[irqnumber].tsk != current) return 0;
 824         spin_lock_irqsave(&irqbits_lock, flags);
 825         bit = irqbits & (1 << irqnumber);
 826         irqbits &= ~bit;
 827         if (bit) {
 828                 enable_irq(irqnumber);
 829                 ret = 1;
 830         }
 831
 832         spin_unlock_irqrestore(&irqbits_lock, flags);
 833         return ret;
 834 }
 835
 836
 837 static int do_vm86_irq_handling(int subfunction, int irqnumber)
 838 {
 839         int ret;
 840         switch (subfunction) {
 841                 case VM86_GET_AND_RESET_IRQ: {
 842                         return get_and_reset_irq(irqnumber);
 843                 }
 844                 case VM86_GET_IRQ_BITS: {
 845                         return irqbits;
 846                 }
 847                 case VM86_REQUEST_IRQ: {
 848                         int sig = irqnumber >> 8;
 849                         int irq = irqnumber & 255;
 850                         if (!capable(CAP_SYS_ADMIN)) return -EPERM;
 851                         if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
 852                         if (invalid_vm86_irq(irq)) return -EPERM;
 853                         if (vm86_irqs[irq].tsk) return -EPERM;
 854                         ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
 855                         if (ret) return ret;
 856                         vm86_irqs[irq].sig = sig;
 857                         vm86_irqs[irq].tsk = current;
 858                         return irq;
 859                 }
 860                 case  VM86_FREE_IRQ: {
 861                         if (invalid_vm86_irq(irqnumber)) return -EPERM;
 862                         if (!vm86_irqs[irqnumber].tsk) return 0;
 863                         if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
 864                         free_vm86_irq(irqnumber);
 865                         return 0;
 866                 }
 867         }
 868         return -EINVAL;
 869 }
 870