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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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/kernel.h>
39 #include <linux/signal.h>
40 #include <linux/string.h>
42 #include <linux/smp.h>
43 #include <linux/highmem.h>
44 #include <linux/ptrace.h>
45 #include <linux/audit.h>
46 #include <linux/stddef.h>
47 #include <linux/slab.h>
49 #include <asm/uaccess.h>
51 #include <asm/tlbflush.h>
53 #include <asm/traps.h>
59 * Interrupt handling is not guaranteed:
60 * - a real x86 will disable all interrupts for one instruction
61 * after a "mov ss,xx" to make stack handling atomic even without
62 * the 'lss' instruction. We can't guarantee this in v86 mode,
63 * as the next instruction might result in a page fault or similar.
64 * - a real x86 will have interrupts disabled for one instruction
65 * past the 'sti' that enables them. We don't bother with all the
68 * Let's hope these problems do not actually matter for anything.
73 * 8- and 16-bit register defines..
75 #define AL(regs) (((unsigned char *)&((regs)->pt.ax))[0])
76 #define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1])
77 #define IP(regs) (*(unsigned short *)&((regs)->pt.ip))
78 #define SP(regs) (*(unsigned short *)&((regs)->pt.sp))
81 * virtual flags (16 and 32-bit versions)
83 #define VFLAGS (*(unsigned short *)&(current->thread.vm86->veflags))
84 #define VEFLAGS (current->thread.vm86->veflags)
86 #define set_flags(X, new, mask) \
87 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
89 #define SAFE_MASK (0xDD5)
90 #define RETURN_MASK (0xDFF)
92 void save_v86_state(struct kernel_vm86_regs *regs, int retval)
94 struct tss_struct *tss;
95 struct task_struct *tsk = current;
96 struct vm86plus_struct __user *user;
97 struct vm86 *vm86 = current->thread.vm86;
101 * This gets called from entry.S with interrupts disabled, but
102 * from process context. Enable interrupts here, before trying
103 * to access user space.
107 if (!vm86 || !vm86->user_vm86) {
108 pr_alert("no user_vm86: BAD\n");
111 set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | vm86->veflags_mask);
112 user = vm86->user_vm86;
114 if (!access_ok(VERIFY_WRITE, user, vm86->vm86plus.is_vm86pus ?
115 sizeof(struct vm86plus_struct) :
116 sizeof(struct vm86_struct))) {
117 pr_alert("could not access userspace vm86 info\n");
122 put_user_ex(regs->pt.bx, &user->regs.ebx);
123 put_user_ex(regs->pt.cx, &user->regs.ecx);
124 put_user_ex(regs->pt.dx, &user->regs.edx);
125 put_user_ex(regs->pt.si, &user->regs.esi);
126 put_user_ex(regs->pt.di, &user->regs.edi);
127 put_user_ex(regs->pt.bp, &user->regs.ebp);
128 put_user_ex(regs->pt.ax, &user->regs.eax);
129 put_user_ex(regs->pt.ip, &user->regs.eip);
130 put_user_ex(regs->pt.cs, &user->regs.cs);
131 put_user_ex(regs->pt.flags, &user->regs.eflags);
132 put_user_ex(regs->pt.sp, &user->regs.esp);
133 put_user_ex(regs->pt.ss, &user->regs.ss);
134 put_user_ex(regs->es, &user->regs.es);
135 put_user_ex(regs->ds, &user->regs.ds);
136 put_user_ex(regs->fs, &user->regs.fs);
137 put_user_ex(regs->gs, &user->regs.gs);
139 put_user_ex(vm86->screen_bitmap, &user->screen_bitmap);
140 } put_user_catch(err);
142 pr_alert("could not access userspace vm86 info\n");
146 tss = &per_cpu(cpu_tss, get_cpu());
147 tsk->thread.sp0 = vm86->saved_sp0;
148 tsk->thread.sysenter_cs = __KERNEL_CS;
149 load_sp0(tss, &tsk->thread);
153 memcpy(®s->pt, &vm86->regs32, sizeof(struct pt_regs));
155 lazy_load_gs(vm86->regs32.gs);
157 regs->pt.ax = retval;
160 static void mark_screen_rdonly(struct mm_struct *mm)
169 down_write(&mm->mmap_sem);
170 pgd = pgd_offset(mm, 0xA0000);
171 if (pgd_none_or_clear_bad(pgd))
173 pud = pud_offset(pgd, 0xA0000);
174 if (pud_none_or_clear_bad(pud))
176 pmd = pmd_offset(pud, 0xA0000);
177 split_huge_page_pmd_mm(mm, 0xA0000, pmd);
178 if (pmd_none_or_clear_bad(pmd))
180 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
181 for (i = 0; i < 32; i++) {
182 if (pte_present(*pte))
183 set_pte(pte, pte_wrprotect(*pte));
186 pte_unmap_unlock(pte, ptl);
188 up_write(&mm->mmap_sem);
194 static int do_vm86_irq_handling(int subfunction, int irqnumber);
195 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus);
197 SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86)
199 return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false);
203 SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg)
206 case VM86_REQUEST_IRQ:
208 case VM86_GET_IRQ_BITS:
209 case VM86_GET_AND_RESET_IRQ:
210 return do_vm86_irq_handling(cmd, (int)arg);
211 case VM86_PLUS_INSTALL_CHECK:
213 * NOTE: on old vm86 stuff this will return the error
214 * from access_ok(), because the subfunction is
215 * interpreted as (invalid) address to vm86_struct.
216 * So the installation check works.
221 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
222 return do_sys_vm86((struct vm86plus_struct __user *) arg, true);
226 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
228 struct tss_struct *tss;
229 struct task_struct *tsk = current;
230 struct vm86 *vm86 = tsk->thread.vm86;
231 struct kernel_vm86_regs vm86regs;
232 struct pt_regs *regs = current_pt_regs();
233 unsigned long err = 0;
236 if (!(vm86 = kzalloc(sizeof(*vm86), GFP_KERNEL)))
238 tsk->thread.vm86 = vm86;
243 if (!access_ok(VERIFY_READ, user_vm86, plus ?
244 sizeof(struct vm86_struct) :
245 sizeof(struct vm86plus_struct)))
248 memset(&vm86regs, 0, sizeof(vm86regs));
251 get_user_ex(vm86regs.pt.bx, &user_vm86->regs.ebx);
252 get_user_ex(vm86regs.pt.cx, &user_vm86->regs.ecx);
253 get_user_ex(vm86regs.pt.dx, &user_vm86->regs.edx);
254 get_user_ex(vm86regs.pt.si, &user_vm86->regs.esi);
255 get_user_ex(vm86regs.pt.di, &user_vm86->regs.edi);
256 get_user_ex(vm86regs.pt.bp, &user_vm86->regs.ebp);
257 get_user_ex(vm86regs.pt.ax, &user_vm86->regs.eax);
258 get_user_ex(vm86regs.pt.ip, &user_vm86->regs.eip);
259 get_user_ex(seg, &user_vm86->regs.cs);
260 vm86regs.pt.cs = seg;
261 get_user_ex(vm86regs.pt.flags, &user_vm86->regs.eflags);
262 get_user_ex(vm86regs.pt.sp, &user_vm86->regs.esp);
263 get_user_ex(seg, &user_vm86->regs.ss);
264 vm86regs.pt.ss = seg;
265 get_user_ex(vm86regs.es, &user_vm86->regs.es);
266 get_user_ex(vm86regs.ds, &user_vm86->regs.ds);
267 get_user_ex(vm86regs.fs, &user_vm86->regs.fs);
268 get_user_ex(vm86regs.gs, &user_vm86->regs.gs);
270 get_user_ex(vm86->flags, &user_vm86->flags);
271 get_user_ex(vm86->screen_bitmap, &user_vm86->screen_bitmap);
272 get_user_ex(vm86->cpu_type, &user_vm86->cpu_type);
273 } get_user_catch(err);
277 if (copy_from_user(&vm86->int_revectored,
278 &user_vm86->int_revectored,
279 sizeof(struct revectored_struct)))
281 if (copy_from_user(&vm86->int21_revectored,
282 &user_vm86->int21_revectored,
283 sizeof(struct revectored_struct)))
286 if (copy_from_user(&vm86->vm86plus, &user_vm86->vm86plus,
287 sizeof(struct vm86plus_info_struct)))
289 vm86->vm86plus.is_vm86pus = 1;
291 memset(&vm86->vm86plus, 0,
292 sizeof(struct vm86plus_info_struct));
294 memcpy(&vm86->regs32, regs, sizeof(struct pt_regs));
295 vm86->user_vm86 = user_vm86;
298 * The flags register is also special: we cannot trust that the user
299 * has set it up safely, so this makes sure interrupt etc flags are
300 * inherited from protected mode.
302 VEFLAGS = vm86regs.pt.flags;
303 vm86regs.pt.flags &= SAFE_MASK;
304 vm86regs.pt.flags |= regs->flags & ~SAFE_MASK;
305 vm86regs.pt.flags |= X86_VM_MASK;
307 vm86regs.pt.orig_ax = regs->orig_ax;
309 switch (vm86->cpu_type) {
311 vm86->veflags_mask = 0;
314 vm86->veflags_mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL;
317 vm86->veflags_mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
320 vm86->veflags_mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
327 vm86->saved_sp0 = tsk->thread.sp0;
328 lazy_save_gs(vm86->regs32.gs);
330 tss = &per_cpu(cpu_tss, get_cpu());
331 /* make room for real-mode segments */
332 tsk->thread.sp0 += 16;
334 tsk->thread.sysenter_cs = 0;
335 load_sp0(tss, &tsk->thread);
338 if (vm86->flags & VM86_SCREEN_BITMAP)
339 mark_screen_rdonly(tsk->mm);
341 memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs));
346 static inline void set_IF(struct kernel_vm86_regs *regs)
348 VEFLAGS |= X86_EFLAGS_VIF;
351 static inline void clear_IF(struct kernel_vm86_regs *regs)
353 VEFLAGS &= ~X86_EFLAGS_VIF;
356 static inline void clear_TF(struct kernel_vm86_regs *regs)
358 regs->pt.flags &= ~X86_EFLAGS_TF;
361 static inline void clear_AC(struct kernel_vm86_regs *regs)
363 regs->pt.flags &= ~X86_EFLAGS_AC;
367 * It is correct to call set_IF(regs) from the set_vflags_*
368 * functions. However someone forgot to call clear_IF(regs)
369 * in the opposite case.
370 * After the command sequence CLI PUSHF STI POPF you should
371 * end up with interrupts disabled, but you ended up with
372 * interrupts enabled.
373 * ( I was testing my own changes, but the only bug I
374 * could find was in a function I had not changed. )
378 static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
380 set_flags(VEFLAGS, flags, current->thread.vm86->veflags_mask);
381 set_flags(regs->pt.flags, flags, SAFE_MASK);
382 if (flags & X86_EFLAGS_IF)
388 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
390 set_flags(VFLAGS, flags, current->thread.vm86->veflags_mask);
391 set_flags(regs->pt.flags, flags, SAFE_MASK);
392 if (flags & X86_EFLAGS_IF)
398 static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
400 unsigned long flags = regs->pt.flags & RETURN_MASK;
402 if (VEFLAGS & X86_EFLAGS_VIF)
403 flags |= X86_EFLAGS_IF;
404 flags |= X86_EFLAGS_IOPL;
405 return flags | (VEFLAGS & current->thread.vm86->veflags_mask);
408 static inline int is_revectored(int nr, struct revectored_struct *bitmap)
410 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
412 :"m" (*bitmap), "r" (nr));
416 #define val_byte(val, n) (((__u8 *)&val)[n])
418 #define pushb(base, ptr, val, err_label) \
422 if (put_user(__val, base + ptr) < 0) \
426 #define pushw(base, ptr, val, err_label) \
430 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
433 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
437 #define pushl(base, ptr, val, err_label) \
441 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
444 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
447 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
450 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
454 #define popb(base, ptr, err_label) \
457 if (get_user(__res, base + ptr) < 0) \
463 #define popw(base, ptr, err_label) \
466 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
469 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
475 #define popl(base, ptr, err_label) \
478 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
481 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
484 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
487 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
493 /* There are so many possible reasons for this function to return
494 * VM86_INTx, so adding another doesn't bother me. We can expect
495 * userspace programs to be able to handle it. (Getting a problem
496 * in userspace is always better than an Oops anyway.) [KD]
498 static void do_int(struct kernel_vm86_regs *regs, int i,
499 unsigned char __user *ssp, unsigned short sp)
501 unsigned long __user *intr_ptr;
502 unsigned long segoffs;
503 struct vm86 *vm86 = current->thread.vm86;
505 if (regs->pt.cs == BIOSSEG)
507 if (is_revectored(i, &vm86->int_revectored))
509 if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored))
511 intr_ptr = (unsigned long __user *) (i << 2);
512 if (get_user(segoffs, intr_ptr))
514 if ((segoffs >> 16) == BIOSSEG)
516 pushw(ssp, sp, get_vflags(regs), cannot_handle);
517 pushw(ssp, sp, regs->pt.cs, cannot_handle);
518 pushw(ssp, sp, IP(regs), cannot_handle);
519 regs->pt.cs = segoffs >> 16;
521 IP(regs) = segoffs & 0xffff;
528 save_v86_state(regs, VM86_INTx + (i << 8));
531 int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
533 struct vm86 *vm86 = current->thread.vm86;
535 if (vm86->vm86plus.is_vm86pus) {
536 if ((trapno == 3) || (trapno == 1)) {
537 save_v86_state(regs, VM86_TRAP + (trapno << 8));
540 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
544 return 1; /* we let this handle by the calling routine */
545 current->thread.trap_nr = trapno;
546 current->thread.error_code = error_code;
547 force_sig(SIGTRAP, current);
551 void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
553 unsigned char opcode;
554 unsigned char __user *csp;
555 unsigned char __user *ssp;
556 unsigned short ip, sp, orig_flags;
557 int data32, pref_done;
558 struct vm86plus_info_struct *vmpi = ¤t->thread.vm86->vm86plus;
560 #define CHECK_IF_IN_TRAP \
561 if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \
562 newflags |= X86_EFLAGS_TF
564 orig_flags = *(unsigned short *)®s->pt.flags;
566 csp = (unsigned char __user *) (regs->pt.cs << 4);
567 ssp = (unsigned char __user *) (regs->pt.ss << 4);
574 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
575 case 0x66: /* 32-bit data */ data32 = 1; break;
576 case 0x67: /* 32-bit address */ break;
577 case 0x2e: /* CS */ break;
578 case 0x3e: /* DS */ break;
579 case 0x26: /* ES */ break;
580 case 0x36: /* SS */ break;
581 case 0x65: /* GS */ break;
582 case 0x64: /* FS */ break;
583 case 0xf2: /* repnz */ break;
584 case 0xf3: /* rep */ break;
585 default: pref_done = 1;
587 } while (!pref_done);
594 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
597 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
601 goto vm86_fault_return;
606 unsigned long newflags;
608 newflags = popl(ssp, sp, simulate_sigsegv);
611 newflags = popw(ssp, sp, simulate_sigsegv);
617 set_vflags_long(newflags, regs);
619 set_vflags_short(newflags, regs);
626 int intno = popb(csp, ip, simulate_sigsegv);
628 if (vmpi->vm86dbg_active) {
629 if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) {
630 save_v86_state(regs, VM86_INTx + (intno << 8));
634 do_int(regs, intno, ssp, sp);
643 unsigned long newflags;
645 newip = popl(ssp, sp, simulate_sigsegv);
646 newcs = popl(ssp, sp, simulate_sigsegv);
647 newflags = popl(ssp, sp, simulate_sigsegv);
650 newip = popw(ssp, sp, simulate_sigsegv);
651 newcs = popw(ssp, sp, simulate_sigsegv);
652 newflags = popw(ssp, sp, simulate_sigsegv);
659 set_vflags_long(newflags, regs);
661 set_vflags_short(newflags, regs);
670 goto vm86_fault_return;
674 * Damn. This is incorrect: the 'sti' instruction should actually
675 * enable interrupts after the /next/ instruction. Not good.
677 * Probably needs some horsing around with the TF flag. Aiee..
685 save_v86_state(regs, VM86_UNKNOWN);
691 if (VEFLAGS & X86_EFLAGS_VIP) {
692 save_v86_state(regs, VM86_STI);
697 if (vmpi->force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) {
698 save_v86_state(regs, VM86_PICRETURN);
701 if (orig_flags & X86_EFLAGS_TF)
702 handle_vm86_trap(regs, 0, X86_TRAP_DB);
706 /* FIXME: After a long discussion with Stas we finally
707 * agreed, that this is wrong. Here we should
708 * really send a SIGSEGV to the user program.
709 * But how do we create the correct context? We
710 * are inside a general protection fault handler
711 * and has just returned from a page fault handler.
712 * The correct context for the signal handler
713 * should be a mixture of the two, but how do we
714 * get the information? [KD]
716 save_v86_state(regs, VM86_UNKNOWN);
719 /* ---------------- vm86 special IRQ passing stuff ----------------- */
721 #define VM86_IRQNAME "vm86irq"
723 static struct vm86_irqs {
724 struct task_struct *tsk;
728 static DEFINE_SPINLOCK(irqbits_lock);
731 #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
732 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
735 static irqreturn_t irq_handler(int intno, void *dev_id)
740 spin_lock_irqsave(&irqbits_lock, flags);
741 irq_bit = 1 << intno;
742 if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
745 if (vm86_irqs[intno].sig)
746 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
748 * IRQ will be re-enabled when user asks for the irq (whether
749 * polling or as a result of the signal)
751 disable_irq_nosync(intno);
752 spin_unlock_irqrestore(&irqbits_lock, flags);
756 spin_unlock_irqrestore(&irqbits_lock, flags);
760 static inline void free_vm86_irq(int irqnumber)
764 free_irq(irqnumber, NULL);
765 vm86_irqs[irqnumber].tsk = NULL;
767 spin_lock_irqsave(&irqbits_lock, flags);
768 irqbits &= ~(1 << irqnumber);
769 spin_unlock_irqrestore(&irqbits_lock, flags);
772 void release_vm86_irqs(struct task_struct *task)
775 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
776 if (vm86_irqs[i].tsk == task)
780 static inline int get_and_reset_irq(int irqnumber)
786 if (invalid_vm86_irq(irqnumber)) return 0;
787 if (vm86_irqs[irqnumber].tsk != current) return 0;
788 spin_lock_irqsave(&irqbits_lock, flags);
789 bit = irqbits & (1 << irqnumber);
792 enable_irq(irqnumber);
796 spin_unlock_irqrestore(&irqbits_lock, flags);
801 static int do_vm86_irq_handling(int subfunction, int irqnumber)
804 switch (subfunction) {
805 case VM86_GET_AND_RESET_IRQ: {
806 return get_and_reset_irq(irqnumber);
808 case VM86_GET_IRQ_BITS: {
811 case VM86_REQUEST_IRQ: {
812 int sig = irqnumber >> 8;
813 int irq = irqnumber & 255;
814 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
815 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
816 if (invalid_vm86_irq(irq)) return -EPERM;
817 if (vm86_irqs[irq].tsk) return -EPERM;
818 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
820 vm86_irqs[irq].sig = sig;
821 vm86_irqs[irq].tsk = current;
824 case VM86_FREE_IRQ: {
825 if (invalid_vm86_irq(irqnumber)) return -EPERM;
826 if (!vm86_irqs[irqnumber].tsk) return 0;
827 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
828 free_vm86_irq(irqnumber);