2 * Copyright (C) 1994 Linus Torvalds
4 * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
5 * stack - Manfred Spraul <manfred@colorfullife.com>
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>
12 * 22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
13 * <kasperd@daimi.au.dk>
15 * ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
16 * caused by Kasper Dupont's changes - Stas Sergeev
18 * 4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
19 * Kasper Dupont <kasperd@daimi.au.dk>
21 * 9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
22 * Kasper Dupont <kasperd@daimi.au.dk>
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>
31 #include <linux/capability.h>
32 #include <linux/errno.h>
33 #include <linux/interrupt.h>
34 #include <linux/sched.h>
35 #include <linux/kernel.h>
36 #include <linux/signal.h>
37 #include <linux/string.h>
39 #include <linux/smp.h>
40 #include <linux/highmem.h>
41 #include <linux/ptrace.h>
42 #include <linux/audit.h>
43 #include <linux/stddef.h>
45 #include <asm/uaccess.h>
47 #include <asm/tlbflush.h>
53 * Interrupt handling is not guaranteed:
54 * - a real x86 will disable all interrupts for one instruction
55 * after a "mov ss,xx" to make stack handling atomic even without
56 * the 'lss' instruction. We can't guarantee this in v86 mode,
57 * as the next instruction might result in a page fault or similar.
58 * - a real x86 will have interrupts disabled for one instruction
59 * past the 'sti' that enables them. We don't bother with all the
62 * Let's hope these problems do not actually matter for anything.
66 #define KVM86 ((struct kernel_vm86_struct *)regs)
67 #define VMPI KVM86->vm86plus
71 * 8- and 16-bit register defines..
73 #define AL(regs) (((unsigned char *)&((regs)->pt.ax))[0])
74 #define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1])
75 #define IP(regs) (*(unsigned short *)&((regs)->pt.ip))
76 #define SP(regs) (*(unsigned short *)&((regs)->pt.sp))
79 * virtual flags (16 and 32-bit versions)
81 #define VFLAGS (*(unsigned short *)&(current->thread.v86flags))
82 #define VEFLAGS (current->thread.v86flags)
84 #define set_flags(X, new, mask) \
85 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
87 #define SAFE_MASK (0xDD5)
88 #define RETURN_MASK (0xDFF)
90 /* convert kernel_vm86_regs to vm86_regs */
91 static int copy_vm86_regs_to_user(struct vm86_regs __user *user,
92 const struct kernel_vm86_regs *regs)
97 * kernel_vm86_regs is missing gs, so copy everything up to
98 * (but not including) orig_eax, and then rest including orig_eax.
100 ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_ax));
101 ret += copy_to_user(&user->orig_eax, ®s->pt.orig_ax,
102 sizeof(struct kernel_vm86_regs) -
103 offsetof(struct kernel_vm86_regs, pt.orig_ax));
108 /* convert vm86_regs to kernel_vm86_regs */
109 static int copy_vm86_regs_from_user(struct kernel_vm86_regs *regs,
110 const struct vm86_regs __user *user,
115 /* copy ax-fs inclusive */
116 ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_ax));
117 /* copy orig_ax-__gsh+extra */
118 ret += copy_from_user(®s->pt.orig_ax, &user->orig_eax,
119 sizeof(struct kernel_vm86_regs) -
120 offsetof(struct kernel_vm86_regs, pt.orig_ax) +
125 struct pt_regs *save_v86_state(struct kernel_vm86_regs *regs)
127 struct tss_struct *tss;
132 * This gets called from entry.S with interrupts disabled, but
133 * from process context. Enable interrupts here, before trying
134 * to access user space.
138 if (!current->thread.vm86_info) {
139 printk("no vm86_info: BAD\n");
142 set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | current->thread.v86mask);
143 tmp = copy_vm86_regs_to_user(¤t->thread.vm86_info->regs, regs);
144 tmp += put_user(current->thread.screen_bitmap, ¤t->thread.vm86_info->screen_bitmap);
146 printk("vm86: could not access userspace vm86_info\n");
150 tss = &per_cpu(init_tss, get_cpu());
151 current->thread.sp0 = current->thread.saved_sp0;
152 current->thread.sysenter_cs = __KERNEL_CS;
153 load_sp0(tss, ¤t->thread);
154 current->thread.saved_sp0 = 0;
159 ret->fs = current->thread.saved_fs;
160 loadsegment(gs, current->thread.saved_gs);
165 static void mark_screen_rdonly(struct mm_struct *mm)
174 pgd = pgd_offset(mm, 0xA0000);
175 if (pgd_none_or_clear_bad(pgd))
177 pud = pud_offset(pgd, 0xA0000);
178 if (pud_none_or_clear_bad(pud))
180 pmd = pmd_offset(pud, 0xA0000);
181 if (pmd_none_or_clear_bad(pmd))
183 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
184 for (i = 0; i < 32; i++) {
185 if (pte_present(*pte))
186 set_pte(pte, pte_wrprotect(*pte));
189 pte_unmap_unlock(pte, ptl);
196 static int do_vm86_irq_handling(int subfunction, int irqnumber);
197 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
199 asmlinkage int sys_vm86old(struct pt_regs regs)
201 struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.bx;
202 struct kernel_vm86_struct info; /* declare this _on top_,
203 * this avoids wasting of stack space.
204 * This remains on the stack until we
205 * return to 32 bit user space.
207 struct task_struct *tsk;
208 int tmp, ret = -EPERM;
211 if (tsk->thread.saved_sp0)
213 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
214 offsetof(struct kernel_vm86_struct, vm86plus) -
219 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
221 tsk->thread.vm86_info = v86;
222 do_sys_vm86(&info, tsk);
223 ret = 0; /* we never return here */
229 asmlinkage int sys_vm86(struct pt_regs regs)
231 struct kernel_vm86_struct info; /* declare this _on top_,
232 * this avoids wasting of stack space.
233 * This remains on the stack until we
234 * return to 32 bit user space.
236 struct task_struct *tsk;
238 struct vm86plus_struct __user *v86;
242 case VM86_REQUEST_IRQ:
244 case VM86_GET_IRQ_BITS:
245 case VM86_GET_AND_RESET_IRQ:
246 ret = do_vm86_irq_handling(regs.bx, (int)regs.cx);
248 case VM86_PLUS_INSTALL_CHECK:
250 * NOTE: on old vm86 stuff this will return the error
251 * from access_ok(), because the subfunction is
252 * interpreted as (invalid) address to vm86_struct.
253 * So the installation check works.
259 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
261 if (tsk->thread.saved_sp0)
263 v86 = (struct vm86plus_struct __user *)regs.cx;
264 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
265 offsetof(struct kernel_vm86_struct, regs32) -
271 info.vm86plus.is_vm86pus = 1;
272 tsk->thread.vm86_info = (struct vm86_struct __user *)v86;
273 do_sys_vm86(&info, tsk);
274 ret = 0; /* we never return here */
280 static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
282 struct tss_struct *tss;
284 * make sure the vm86() system call doesn't try to do anything silly
286 info->regs.pt.ds = 0;
287 info->regs.pt.es = 0;
288 info->regs.pt.fs = 0;
290 /* we are clearing gs later just before "jmp resume_userspace",
291 * because it is not saved/restored.
295 * The flags register is also special: we cannot trust that the user
296 * has set it up safely, so this makes sure interrupt etc flags are
297 * inherited from protected mode.
299 VEFLAGS = info->regs.pt.flags;
300 info->regs.pt.flags &= SAFE_MASK;
301 info->regs.pt.flags |= info->regs32->flags & ~SAFE_MASK;
302 info->regs.pt.flags |= X86_VM_MASK;
304 switch (info->cpu_type) {
306 tsk->thread.v86mask = 0;
309 tsk->thread.v86mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL;
312 tsk->thread.v86mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
315 tsk->thread.v86mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
320 * Save old state, set default return value (%ax) to 0
322 info->regs32->ax = 0;
323 tsk->thread.saved_sp0 = tsk->thread.sp0;
324 tsk->thread.saved_fs = info->regs32->fs;
325 savesegment(gs, tsk->thread.saved_gs);
327 tss = &per_cpu(init_tss, get_cpu());
328 tsk->thread.sp0 = (unsigned long) &info->VM86_TSS_ESP0;
330 tsk->thread.sysenter_cs = 0;
331 load_sp0(tss, &tsk->thread);
334 tsk->thread.screen_bitmap = info->screen_bitmap;
335 if (info->flags & VM86_SCREEN_BITMAP)
336 mark_screen_rdonly(tsk->mm);
338 /*call audit_syscall_exit since we do not exit via the normal paths */
339 if (unlikely(current->audit_context))
340 audit_syscall_exit(AUDITSC_RESULT(0), 0);
342 __asm__ __volatile__(
346 "jmp resume_userspace"
348 :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0));
349 /* we never return here */
352 static inline void return_to_32bit(struct kernel_vm86_regs *regs16, int retval)
354 struct pt_regs *regs32;
356 regs32 = save_v86_state(regs16);
358 __asm__ __volatile__("movl %0,%%esp\n\t"
360 "jmp resume_userspace"
361 : : "r" (regs32), "r" (current_thread_info()));
364 static inline void set_IF(struct kernel_vm86_regs *regs)
366 VEFLAGS |= X86_EFLAGS_VIF;
367 if (VEFLAGS & X86_EFLAGS_VIP)
368 return_to_32bit(regs, VM86_STI);
371 static inline void clear_IF(struct kernel_vm86_regs *regs)
373 VEFLAGS &= ~X86_EFLAGS_VIF;
376 static inline void clear_TF(struct kernel_vm86_regs *regs)
378 regs->pt.flags &= ~X86_EFLAGS_TF;
381 static inline void clear_AC(struct kernel_vm86_regs *regs)
383 regs->pt.flags &= ~X86_EFLAGS_AC;
387 * It is correct to call set_IF(regs) from the set_vflags_*
388 * functions. However someone forgot to call clear_IF(regs)
389 * in the opposite case.
390 * After the command sequence CLI PUSHF STI POPF you should
391 * end up with interrupts disabled, but you ended up with
392 * interrupts enabled.
393 * ( I was testing my own changes, but the only bug I
394 * could find was in a function I had not changed. )
398 static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
400 set_flags(VEFLAGS, flags, current->thread.v86mask);
401 set_flags(regs->pt.flags, flags, SAFE_MASK);
402 if (flags & X86_EFLAGS_IF)
408 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
410 set_flags(VFLAGS, flags, current->thread.v86mask);
411 set_flags(regs->pt.flags, flags, SAFE_MASK);
412 if (flags & X86_EFLAGS_IF)
418 static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
420 unsigned long flags = regs->pt.flags & RETURN_MASK;
422 if (VEFLAGS & X86_EFLAGS_VIF)
423 flags |= X86_EFLAGS_IF;
424 flags |= X86_EFLAGS_IOPL;
425 return flags | (VEFLAGS & current->thread.v86mask);
428 static inline int is_revectored(int nr, struct revectored_struct *bitmap)
430 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
432 :"m" (*bitmap), "r" (nr));
436 #define val_byte(val, n) (((__u8 *)&val)[n])
438 #define pushb(base, ptr, val, err_label) \
442 if (put_user(__val, base + ptr) < 0) \
446 #define pushw(base, ptr, val, err_label) \
450 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
453 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
457 #define pushl(base, ptr, val, err_label) \
461 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
464 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
467 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
470 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
474 #define popb(base, ptr, err_label) \
477 if (get_user(__res, base + ptr) < 0) \
483 #define popw(base, ptr, err_label) \
486 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
489 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
495 #define popl(base, ptr, err_label) \
498 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
501 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
504 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
507 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
513 /* There are so many possible reasons for this function to return
514 * VM86_INTx, so adding another doesn't bother me. We can expect
515 * userspace programs to be able to handle it. (Getting a problem
516 * in userspace is always better than an Oops anyway.) [KD]
518 static void do_int(struct kernel_vm86_regs *regs, int i,
519 unsigned char __user *ssp, unsigned short sp)
521 unsigned long __user *intr_ptr;
522 unsigned long segoffs;
524 if (regs->pt.cs == BIOSSEG)
526 if (is_revectored(i, &KVM86->int_revectored))
528 if (i == 0x21 && is_revectored(AH(regs), &KVM86->int21_revectored))
530 intr_ptr = (unsigned long __user *) (i << 2);
531 if (get_user(segoffs, intr_ptr))
533 if ((segoffs >> 16) == BIOSSEG)
535 pushw(ssp, sp, get_vflags(regs), cannot_handle);
536 pushw(ssp, sp, regs->pt.cs, cannot_handle);
537 pushw(ssp, sp, IP(regs), cannot_handle);
538 regs->pt.cs = segoffs >> 16;
540 IP(regs) = segoffs & 0xffff;
547 return_to_32bit(regs, VM86_INTx + (i << 8));
550 int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
552 if (VMPI.is_vm86pus) {
553 if ((trapno == 3) || (trapno == 1))
554 return_to_32bit(regs, VM86_TRAP + (trapno << 8));
555 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
559 return 1; /* we let this handle by the calling routine */
560 current->thread.trap_no = trapno;
561 current->thread.error_code = error_code;
562 force_sig(SIGTRAP, current);
566 void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
568 unsigned char opcode;
569 unsigned char __user *csp;
570 unsigned char __user *ssp;
571 unsigned short ip, sp, orig_flags;
572 int data32, pref_done;
574 #define CHECK_IF_IN_TRAP \
575 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
576 newflags |= X86_EFLAGS_TF
577 #define VM86_FAULT_RETURN do { \
578 if (VMPI.force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) \
579 return_to_32bit(regs, VM86_PICRETURN); \
580 if (orig_flags & X86_EFLAGS_TF) \
581 handle_vm86_trap(regs, 0, 1); \
584 orig_flags = *(unsigned short *)®s->pt.flags;
586 csp = (unsigned char __user *) (regs->pt.cs << 4);
587 ssp = (unsigned char __user *) (regs->pt.ss << 4);
594 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
595 case 0x66: /* 32-bit data */ data32 = 1; break;
596 case 0x67: /* 32-bit address */ break;
597 case 0x2e: /* CS */ break;
598 case 0x3e: /* DS */ break;
599 case 0x26: /* ES */ break;
600 case 0x36: /* SS */ break;
601 case 0x65: /* GS */ break;
602 case 0x64: /* FS */ break;
603 case 0xf2: /* repnz */ break;
604 case 0xf3: /* rep */ break;
605 default: pref_done = 1;
607 } while (!pref_done);
614 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
617 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
626 unsigned long newflags;
628 newflags = popl(ssp, sp, simulate_sigsegv);
631 newflags = popw(ssp, sp, simulate_sigsegv);
637 set_vflags_long(newflags, regs);
639 set_vflags_short(newflags, regs);
646 int intno = popb(csp, ip, simulate_sigsegv);
648 if (VMPI.vm86dbg_active) {
649 if ((1 << (intno & 7)) & VMPI.vm86dbg_intxxtab[intno >> 3])
650 return_to_32bit(regs, VM86_INTx + (intno << 8));
652 do_int(regs, intno, ssp, sp);
661 unsigned long newflags;
663 newip = popl(ssp, sp, simulate_sigsegv);
664 newcs = popl(ssp, sp, simulate_sigsegv);
665 newflags = popl(ssp, sp, simulate_sigsegv);
668 newip = popw(ssp, sp, simulate_sigsegv);
669 newcs = popw(ssp, sp, simulate_sigsegv);
670 newflags = popw(ssp, sp, simulate_sigsegv);
677 set_vflags_long(newflags, regs);
679 set_vflags_short(newflags, regs);
692 * Damn. This is incorrect: the 'sti' instruction should actually
693 * enable interrupts after the /next/ instruction. Not good.
695 * Probably needs some horsing around with the TF flag. Aiee..
703 return_to_32bit(regs, VM86_UNKNOWN);
709 /* FIXME: After a long discussion with Stas we finally
710 * agreed, that this is wrong. Here we should
711 * really send a SIGSEGV to the user program.
712 * But how do we create the correct context? We
713 * are inside a general protection fault handler
714 * and has just returned from a page fault handler.
715 * The correct context for the signal handler
716 * should be a mixture of the two, but how do we
717 * get the information? [KD]
719 return_to_32bit(regs, VM86_UNKNOWN);
722 /* ---------------- vm86 special IRQ passing stuff ----------------- */
724 #define VM86_IRQNAME "vm86irq"
726 static struct vm86_irqs {
727 struct task_struct *tsk;
731 static DEFINE_SPINLOCK(irqbits_lock);
734 #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
735 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
738 static irqreturn_t irq_handler(int intno, void *dev_id)
743 spin_lock_irqsave(&irqbits_lock, flags);
744 irq_bit = 1 << intno;
745 if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
748 if (vm86_irqs[intno].sig)
749 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
751 * IRQ will be re-enabled when user asks for the irq (whether
752 * polling or as a result of the signal)
754 disable_irq_nosync(intno);
755 spin_unlock_irqrestore(&irqbits_lock, flags);
759 spin_unlock_irqrestore(&irqbits_lock, flags);
763 static inline void free_vm86_irq(int irqnumber)
767 free_irq(irqnumber, NULL);
768 vm86_irqs[irqnumber].tsk = NULL;
770 spin_lock_irqsave(&irqbits_lock, flags);
771 irqbits &= ~(1 << irqnumber);
772 spin_unlock_irqrestore(&irqbits_lock, flags);
775 void release_vm86_irqs(struct task_struct *task)
778 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
779 if (vm86_irqs[i].tsk == task)
783 static inline int get_and_reset_irq(int irqnumber)
789 if (invalid_vm86_irq(irqnumber)) return 0;
790 if (vm86_irqs[irqnumber].tsk != current) return 0;
791 spin_lock_irqsave(&irqbits_lock, flags);
792 bit = irqbits & (1 << irqnumber);
795 enable_irq(irqnumber);
799 spin_unlock_irqrestore(&irqbits_lock, flags);
804 static int do_vm86_irq_handling(int subfunction, int irqnumber)
807 switch (subfunction) {
808 case VM86_GET_AND_RESET_IRQ: {
809 return get_and_reset_irq(irqnumber);
811 case VM86_GET_IRQ_BITS: {
814 case VM86_REQUEST_IRQ: {
815 int sig = irqnumber >> 8;
816 int irq = irqnumber & 255;
817 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
818 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
819 if (invalid_vm86_irq(irq)) return -EPERM;
820 if (vm86_irqs[irq].tsk) return -EPERM;
821 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
823 vm86_irqs[irq].sig = sig;
824 vm86_irqs[irq].tsk = current;
827 case VM86_FREE_IRQ: {
828 if (invalid_vm86_irq(irqnumber)) return -EPERM;
829 if (!vm86_irqs[irqnumber].tsk) return 0;
830 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
831 free_vm86_irq(irqnumber);