vm86_32.c

linux 内核源代码

/*
 *  Copyright (C) 1994  Linus Torvalds
 *
 *  29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
 *                stack - Manfred Spraul <[e-mail]>
 *
 *  22 mar 2002 - Manfred detected the stackfaults, but didn't handle
 *                them correctly. Now the emulation will be in a
 *                consistent state after stackfaults - Kasper Dupont
 *                <[e-mail]>
 *
 *  22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
 *                <[e-mail]>
 *
 *  ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
 *                caused by Kasper Dupont's changes - Stas Sergeev
 *
 *   4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
 *                Kasper Dupont <[e-mail]>
 *
 *   9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
 *                Kasper Dupont <[e-mail]>
 *
 *   9 apr 2002 - Changed stack access macros to jump to a label
 *                instead of returning to userspace. This simplifies
 *                do_int, and is needed by handle_vm6_fault. Kasper
 *                Dupont <[e-mail]>
 *
 */

#include <linux/capability.h>
#include <linux/errno.h>
#include <linux/interrupt.h>
#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/signal.h>
#include <linux/string.h>
#include <linux/mm.h>
#include <linux/smp.h>
#include <linux/highmem.h>
#include <linux/ptrace.h>
#include <linux/audit.h>
#include <linux/stddef.h>

#include <asm/uaccess.h>
#include <asm/io.h>
#include <asm/tlbflush.h>
#include <asm/irq.h>

/*
 * Known problems:
 *
 * Interrupt handling is not guaranteed:
 * - a real x86 will disable all interrupts for one instruction
 *   after a "mov ss,xx" to make stack handling atomic even without
 *   the 'lss' instruction. We can't guarantee this in v86 mode,
 *   as the next instruction might result in a page fault or similar.
 * - a real x86 will have interrupts disabled for one instruction
 *   past the 'sti' that enables them. We don't bother with all the
 *   details yet.
 *
 * Let's hope these problems do not actually matter for anything.
 */


#define KVM86	((struct kernel_vm86_struct *)regs)
#define VMPI 	KVM86->vm86plus


/*
 * 8- and 16-bit register defines..
 */
#define AL(regs)	(((unsigned char *)&((regs)->pt.eax))[0])
#define AH(regs)	(((unsigned char *)&((regs)->pt.eax))[1])
#define IP(regs)	(*(unsigned short *)&((regs)->pt.eip))
#define SP(regs)	(*(unsigned short *)&((regs)->pt.esp))

/*
 * virtual flags (16 and 32-bit versions)
 */
#define VFLAGS	(*(unsigned short *)&(current->thread.v86flags))
#define VEFLAGS	(current->thread.v86flags)

#define set_flags(X,new,mask) \
((X) = ((X) & ~(mask)) | ((new) & (mask)))

#define SAFE_MASK	(0xDD5)
#define RETURN_MASK	(0xDFF)

/* convert kernel_vm86_regs to vm86_regs */
static int copy_vm86_regs_to_user(struct vm86_regs __user *user,
				  const struct kernel_vm86_regs *regs)
{
	int ret = 0;

	/* kernel_vm86_regs is missing xgs, so copy everything up to
	   (but not including) orig_eax, and then rest including orig_eax. */
	ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_eax));
	ret += copy_to_user(&user->orig_eax, &regs->pt.orig_eax,
			    sizeof(struct kernel_vm86_regs) -
			    offsetof(struct kernel_vm86_regs, pt.orig_eax));

	return ret;
}

/* convert vm86_regs to kernel_vm86_regs */
static int copy_vm86_regs_from_user(struct kernel_vm86_regs *regs,
				    const struct vm86_regs __user *user,
				    unsigned extra)
{
	int ret = 0;

	/* copy eax-xfs inclusive */
	ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_eax));
	/* copy orig_eax-__gsh+extra */
	ret += copy_from_user(&regs->pt.orig_eax, &user->orig_eax,
			      sizeof(struct kernel_vm86_regs) -
			      offsetof(struct kernel_vm86_regs, pt.orig_eax) +
			      extra);
	return ret;
}

struct pt_regs * FASTCALL(save_v86_state(struct kernel_vm86_regs * regs));
struct pt_regs * fastcall save_v86_state(struct kernel_vm86_regs * regs)
{
	struct tss_struct *tss;
	struct pt_regs *ret;
	unsigned long tmp;

	/*
	 * This gets called from entry.S with interrupts disabled, but
	 * from process context. Enable interrupts here, before trying
	 * to access user space.
	 */
	local_irq_enable();

	if (!current->thread.vm86_info) {
		printk("no vm86_info: BAD\n");
		do_exit(SIGSEGV);
	}
	set_flags(regs->pt.eflags, VEFLAGS, VIF_MASK | current->thread.v86mask);
	tmp = copy_vm86_regs_to_user(&current->thread.vm86_info->regs,regs);
	tmp += put_user(current->thread.screen_bitmap,&current->thread.vm86_info->screen_bitmap);
	if (tmp) {
		printk("vm86: could not access userspace vm86_info\n");
		do_exit(SIGSEGV);
	}

	tss = &per_cpu(init_tss, get_cpu());
	current->thread.esp0 = current->thread.saved_esp0;
	current->thread.sysenter_cs = __KERNEL_CS;
	load_esp0(tss, &current->thread);
	current->thread.saved_esp0 = 0;
	put_cpu();

	ret = KVM86->regs32;

	ret->xfs = current->thread.saved_fs;
	loadsegment(gs, current->thread.saved_gs);

	return ret;
}

static void mark_screen_rdonly(struct mm_struct *mm)
{
	pgd_t *pgd;
	pud_t *pud;
	pmd_t *pmd;
	pte_t *pte;
	spinlock_t *ptl;
	int i;

	pgd = pgd_offset(mm, 0xA0000);
	if (pgd_none_or_clear_bad(pgd))
		goto out;
	pud = pud_offset(pgd, 0xA0000);
	if (pud_none_or_clear_bad(pud))
		goto out;
	pmd = pmd_offset(pud, 0xA0000);
	if (pmd_none_or_clear_bad(pmd))
		goto out;
	pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
	for (i = 0; i < 32; i++) {
		if (pte_present(*pte))
			set_pte(pte, pte_wrprotect(*pte));
		pte++;
	}
	pte_unmap_unlock(pte, ptl);
out:
	flush_tlb();
}



static int do_vm86_irq_handling(int subfunction, int irqnumber);
static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);

asmlinkage int sys_vm86old(struct pt_regs regs)
{
	struct vm86_struct __user *v86 = (struct vm86_struct __user *)regs.ebx;
	struct kernel_vm86_struct info; /* declare this _on top_,
					 * this avoids wasting of stack space.
					 * This remains on the stack until we
					 * return to 32 bit user space.
					 */
	struct task_struct *tsk;
	int tmp, ret = -EPERM;

	tsk = current;
	if (tsk->thread.saved_esp0)
		goto out;
	tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
				       offsetof(struct kernel_vm86_struct, vm86plus) -
				       sizeof(info.regs));
	ret = -EFAULT;
	if (tmp)
		goto out;
	memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
	info.regs32 = &regs;
	tsk->thread.vm86_info = v86;
	do_sys_vm86(&info, tsk);
	ret = 0;	/* we never return here */
out:
	return ret;
}


asmlinkage int sys_vm86(struct pt_regs regs)
{
	struct kernel_vm86_struct info; /* declare this _on top_,
					 * this avoids wasting of stack space.
					 * This remains on the stack until we
					 * return to 32 bit user space.
					 */
	struct task_struct *tsk;
	int tmp, ret;
	struct vm86plus_struct __user *v86;

	tsk = current;
	switch (regs.ebx) {
		case VM86_REQUEST_IRQ:
		case VM86_FREE_IRQ:
		case VM86_GET_IRQ_BITS:
		case VM86_GET_AND_RESET_IRQ:
			ret = do_vm86_irq_handling(regs.ebx, (int)regs.ecx);
			goto out;
		case VM86_PLUS_INSTALL_CHECK:
			/* NOTE: on old vm86 stuff this will return the error
			   from access_ok(), because the subfunction is
			   interpreted as (invalid) address to vm86_struct.
			   So the installation check works.
			 */
			ret = 0;
			goto out;
	}

	/* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
	ret = -EPERM;
	if (tsk->thread.saved_esp0)
		goto out;
	v86 = (struct vm86plus_struct __user *)regs.ecx;
	tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
				       offsetof(struct kernel_vm86_struct, regs32) -
				       sizeof(info.regs));
	ret = -EFAULT;
	if (tmp)
		goto out;
	info.regs32 = &regs;
	info.vm86plus.is_vm86pus = 1;
	tsk->thread.vm86_info = (struct vm86_struct __user *)v86;
	do_sys_vm86(&info, tsk);
	ret = 0;	/* we never return here */
out:
	return ret;
}


static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
{
	struct tss_struct *tss;
/*
 * make sure the vm86() system call doesn't try to do anything silly
 */
	info->regs.pt.xds = 0;
	info->regs.pt.xes = 0;
	info->regs.pt.xfs = 0;

/* we are clearing gs later just before "jmp resume_userspace",
 * because it is not saved/restored.
 */

/*
 * The eflags register is also special: we cannot trust that the user
 * has set it up safely, so this makes sure interrupt etc flags are
 * inherited from protected mode.
 */
 	VEFLAGS = info->regs.pt.eflags;
	info->regs.pt.eflags &= SAFE_MASK;
	info->regs.pt.eflags |= info->regs32->eflags & ~SAFE_MASK;
	info->regs.pt.eflags |= VM_MASK;

	switch (info->cpu_type) {
		case CPU_286:
			tsk->thread.v86mask = 0;
			break;
		case CPU_386:
			tsk->thread.v86mask = NT_MASK | IOPL_MASK;
			break;
		case CPU_486:
			tsk->thread.v86mask = AC_MASK | NT_MASK | IOPL_MASK;
			break;
		default:
			tsk->thread.v86mask = ID_MASK | AC_MASK | NT_MASK | IOPL_MASK;
			break;
	}

/*
 * Save old state, set default return value (%eax) to 0
 */
	info->regs32->eax = 0;
	tsk->thread.saved_esp0 = tsk->thread.esp0;
	tsk->thread.saved_fs = info->regs32->xfs;
	savesegment(gs, tsk->thread.saved_gs);

	tss = &per_cpu(init_tss, get_cpu());
	tsk->thread.esp0 = (unsigned long) &info->VM86_TSS_ESP0;
	if (cpu_has_sep)
		tsk->thread.sysenter_cs = 0;
	load_esp0(tss, &tsk->thread);
	put_cpu();

	tsk->thread.screen_bitmap = info->screen_bitmap;
	if (info->flags & VM86_SCREEN_BITMAP)
		mark_screen_rdonly(tsk->mm);

	/*call audit_syscall_exit since we do not exit via the normal paths */
	if (unlikely(current->audit_context))
		audit_syscall_exit(AUDITSC_RESULT(0), 0);

	__asm__ __volatile__(
		"movl %0,%%esp\n\t"
		"movl %1,%%ebp\n\t"
		"mov  %2, %%gs\n\t"
		"jmp resume_userspace"
		: /* no outputs */
		:"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0));
	/* we never return here */
}

static inline void return_to_32bit(struct kernel_vm86_regs * regs16, int retval)
{
	struct pt_regs * regs32;

	regs32 = save_v86_state(regs16);
	regs32->eax = retval;
	__asm__ __volatile__("movl %0,%%esp\n\t"
		"movl %1,%%ebp\n\t"
		"jmp resume_userspace"
		: : "r" (regs32), "r" (current_thread_info()));
}

static inline void set_IF(struct kernel_vm86_regs * regs)
{
	VEFLAGS |= VIF_MASK;
	if (VEFLAGS & VIP_MASK)
		return_to_32bit(regs, VM86_STI);
}

static inline void clear_IF(struct kernel_vm86_regs * regs)
{
	VEFLAGS &= ~VIF_MASK;
}

static inline void clear_TF(struct kernel_vm86_regs * regs)
{
	regs->pt.eflags &= ~TF_MASK;
}

static inline void clear_AC(struct kernel_vm86_regs * regs)
{
	regs->pt.eflags &= ~AC_MASK;
}

/* It is correct to call set_IF(regs) from the set_vflags_*
 * functions. However someone forgot to call clear_IF(regs)
 * in the opposite case.
 * After the command sequence CLI PUSHF STI POPF you should
 * end up with interrups disabled, but you ended up with
 * interrupts enabled.
 *  ( I was testing my own changes, but the only bug I
 *    could find was in a function I had not changed. )
 * [KD]
 */

static inline void set_vflags_long(unsigned long eflags, struct kernel_vm86_regs * regs)
{
	set_flags(VEFLAGS, eflags, current->thread.v86mask);
	set_flags(regs->pt.eflags, eflags, SAFE_MASK);
	if (eflags & IF_MASK)
		set_IF(regs);
	else
		clear_IF(regs);
}

static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs * regs)
{
	set_flags(VFLAGS, flags, current->thread.v86mask);
	set_flags(regs->pt.eflags, flags, SAFE_MASK);
	if (flags & IF_MASK)
		set_IF(regs);
	else
		clear_IF(regs);
}

static inline unsigned long get_vflags(struct kernel_vm86_regs * regs)
{
	unsigned long flags = regs->pt.eflags & RETURN_MASK;

	if (VEFLAGS & VIF_MASK)
		flags |= IF_MASK;
	flags |= IOPL_MASK;