Loading...
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#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>
38#include <linux/mm.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>
44
45#include <asm/uaccess.h>
46#include <asm/io.h>
47#include <asm/tlbflush.h>
48#include <asm/irq.h>
49#include <asm/syscalls.h>
50
51/*
52 * Known problems:
53 *
54 * Interrupt handling is not guaranteed:
55 * - a real x86 will disable all interrupts for one instruction
56 * after a "mov ss,xx" to make stack handling atomic even without
57 * the 'lss' instruction. We can't guarantee this in v86 mode,
58 * as the next instruction might result in a page fault or similar.
59 * - a real x86 will have interrupts disabled for one instruction
60 * past the 'sti' that enables them. We don't bother with all the
61 * details yet.
62 *
63 * Let's hope these problems do not actually matter for anything.
64 */
65
66
67#define KVM86 ((struct kernel_vm86_struct *)regs)
68#define VMPI KVM86->vm86plus
69
70
71/*
72 * 8- and 16-bit register defines..
73 */
74#define AL(regs) (((unsigned char *)&((regs)->pt.ax))[0])
75#define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1])
76#define IP(regs) (*(unsigned short *)&((regs)->pt.ip))
77#define SP(regs) (*(unsigned short *)&((regs)->pt.sp))
78
79/*
80 * virtual flags (16 and 32-bit versions)
81 */
82#define VFLAGS (*(unsigned short *)&(current->thread.v86flags))
83#define VEFLAGS (current->thread.v86flags)
84
85#define set_flags(X, new, mask) \
86((X) = ((X) & ~(mask)) | ((new) & (mask)))
87
88#define SAFE_MASK (0xDD5)
89#define RETURN_MASK (0xDFF)
90
91/* convert kernel_vm86_regs to vm86_regs */
92static int copy_vm86_regs_to_user(struct vm86_regs __user *user,
93 const struct kernel_vm86_regs *regs)
94{
95 int ret = 0;
96
97 /*
98 * kernel_vm86_regs is missing gs, so copy everything up to
99 * (but not including) orig_eax, and then rest including orig_eax.
100 */
101 ret += copy_to_user(user, regs, offsetof(struct kernel_vm86_regs, pt.orig_ax));
102 ret += copy_to_user(&user->orig_eax, ®s->pt.orig_ax,
103 sizeof(struct kernel_vm86_regs) -
104 offsetof(struct kernel_vm86_regs, pt.orig_ax));
105
106 return ret;
107}
108
109/* convert vm86_regs to kernel_vm86_regs */
110static int copy_vm86_regs_from_user(struct kernel_vm86_regs *regs,
111 const struct vm86_regs __user *user,
112 unsigned extra)
113{
114 int ret = 0;
115
116 /* copy ax-fs inclusive */
117 ret += copy_from_user(regs, user, offsetof(struct kernel_vm86_regs, pt.orig_ax));
118 /* copy orig_ax-__gsh+extra */
119 ret += copy_from_user(®s->pt.orig_ax, &user->orig_eax,
120 sizeof(struct kernel_vm86_regs) -
121 offsetof(struct kernel_vm86_regs, pt.orig_ax) +
122 extra);
123 return ret;
124}
125
126struct pt_regs *save_v86_state(struct kernel_vm86_regs *regs)
127{
128 struct tss_struct *tss;
129 struct pt_regs *ret;
130 unsigned long tmp;
131
132 /*
133 * This gets called from entry.S with interrupts disabled, but
134 * from process context. Enable interrupts here, before trying
135 * to access user space.
136 */
137 local_irq_enable();
138
139 if (!current->thread.vm86_info) {
140 printk("no vm86_info: BAD\n");
141 do_exit(SIGSEGV);
142 }
143 set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | current->thread.v86mask);
144 tmp = copy_vm86_regs_to_user(¤t->thread.vm86_info->regs, regs);
145 tmp += put_user(current->thread.screen_bitmap, ¤t->thread.vm86_info->screen_bitmap);
146 if (tmp) {
147 printk("vm86: could not access userspace vm86_info\n");
148 do_exit(SIGSEGV);
149 }
150
151 tss = &per_cpu(init_tss, get_cpu());
152 current->thread.sp0 = current->thread.saved_sp0;
153 current->thread.sysenter_cs = __KERNEL_CS;
154 load_sp0(tss, ¤t->thread);
155 current->thread.saved_sp0 = 0;
156 put_cpu();
157
158 ret = KVM86->regs32;
159
160 ret->fs = current->thread.saved_fs;
161 set_user_gs(ret, current->thread.saved_gs);
162
163 return ret;
164}
165
166static void mark_screen_rdonly(struct mm_struct *mm)
167{
168 pgd_t *pgd;
169 pud_t *pud;
170 pmd_t *pmd;
171 pte_t *pte;
172 spinlock_t *ptl;
173 int i;
174
175 down_write(&mm->mmap_sem);
176 pgd = pgd_offset(mm, 0xA0000);
177 if (pgd_none_or_clear_bad(pgd))
178 goto out;
179 pud = pud_offset(pgd, 0xA0000);
180 if (pud_none_or_clear_bad(pud))
181 goto out;
182 pmd = pmd_offset(pud, 0xA0000);
183 split_huge_page_pmd(mm, pmd);
184 if (pmd_none_or_clear_bad(pmd))
185 goto out;
186 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
187 for (i = 0; i < 32; i++) {
188 if (pte_present(*pte))
189 set_pte(pte, pte_wrprotect(*pte));
190 pte++;
191 }
192 pte_unmap_unlock(pte, ptl);
193out:
194 up_write(&mm->mmap_sem);
195 flush_tlb();
196}
197
198
199
200static int do_vm86_irq_handling(int subfunction, int irqnumber);
201static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk);
202
203int sys_vm86old(struct vm86_struct __user *v86, struct pt_regs *regs)
204{
205 struct kernel_vm86_struct info; /* declare this _on top_,
206 * this avoids wasting of stack space.
207 * This remains on the stack until we
208 * return to 32 bit user space.
209 */
210 struct task_struct *tsk;
211 int tmp, ret = -EPERM;
212
213 tsk = current;
214 if (tsk->thread.saved_sp0)
215 goto out;
216 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
217 offsetof(struct kernel_vm86_struct, vm86plus) -
218 sizeof(info.regs));
219 ret = -EFAULT;
220 if (tmp)
221 goto out;
222 memset(&info.vm86plus, 0, (int)&info.regs32 - (int)&info.vm86plus);
223 info.regs32 = regs;
224 tsk->thread.vm86_info = v86;
225 do_sys_vm86(&info, tsk);
226 ret = 0; /* we never return here */
227out:
228 return ret;
229}
230
231
232int sys_vm86(unsigned long cmd, unsigned long arg, struct pt_regs *regs)
233{
234 struct kernel_vm86_struct info; /* declare this _on top_,
235 * this avoids wasting of stack space.
236 * This remains on the stack until we
237 * return to 32 bit user space.
238 */
239 struct task_struct *tsk;
240 int tmp, ret;
241 struct vm86plus_struct __user *v86;
242
243 tsk = current;
244 switch (cmd) {
245 case VM86_REQUEST_IRQ:
246 case VM86_FREE_IRQ:
247 case VM86_GET_IRQ_BITS:
248 case VM86_GET_AND_RESET_IRQ:
249 ret = do_vm86_irq_handling(cmd, (int)arg);
250 goto out;
251 case VM86_PLUS_INSTALL_CHECK:
252 /*
253 * NOTE: on old vm86 stuff this will return the error
254 * from access_ok(), because the subfunction is
255 * interpreted as (invalid) address to vm86_struct.
256 * So the installation check works.
257 */
258 ret = 0;
259 goto out;
260 }
261
262 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
263 ret = -EPERM;
264 if (tsk->thread.saved_sp0)
265 goto out;
266 v86 = (struct vm86plus_struct __user *)arg;
267 tmp = copy_vm86_regs_from_user(&info.regs, &v86->regs,
268 offsetof(struct kernel_vm86_struct, regs32) -
269 sizeof(info.regs));
270 ret = -EFAULT;
271 if (tmp)
272 goto out;
273 info.regs32 = regs;
274 info.vm86plus.is_vm86pus = 1;
275 tsk->thread.vm86_info = (struct vm86_struct __user *)v86;
276 do_sys_vm86(&info, tsk);
277 ret = 0; /* we never return here */
278out:
279 return ret;
280}
281
282
283static void do_sys_vm86(struct kernel_vm86_struct *info, struct task_struct *tsk)
284{
285 struct tss_struct *tss;
286/*
287 * make sure the vm86() system call doesn't try to do anything silly
288 */
289 info->regs.pt.ds = 0;
290 info->regs.pt.es = 0;
291 info->regs.pt.fs = 0;
292#ifndef CONFIG_X86_32_LAZY_GS
293 info->regs.pt.gs = 0;
294#endif
295
296/*
297 * The flags register is also special: we cannot trust that the user
298 * has set it up safely, so this makes sure interrupt etc flags are
299 * inherited from protected mode.
300 */
301 VEFLAGS = info->regs.pt.flags;
302 info->regs.pt.flags &= SAFE_MASK;
303 info->regs.pt.flags |= info->regs32->flags & ~SAFE_MASK;
304 info->regs.pt.flags |= X86_VM_MASK;
305
306 switch (info->cpu_type) {
307 case CPU_286:
308 tsk->thread.v86mask = 0;
309 break;
310 case CPU_386:
311 tsk->thread.v86mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL;
312 break;
313 case CPU_486:
314 tsk->thread.v86mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
315 break;
316 default:
317 tsk->thread.v86mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
318 break;
319 }
320
321/*
322 * Save old state, set default return value (%ax) to 0 (VM86_SIGNAL)
323 */
324 info->regs32->ax = VM86_SIGNAL;
325 tsk->thread.saved_sp0 = tsk->thread.sp0;
326 tsk->thread.saved_fs = info->regs32->fs;
327 tsk->thread.saved_gs = get_user_gs(info->regs32);
328
329 tss = &per_cpu(init_tss, get_cpu());
330 tsk->thread.sp0 = (unsigned long) &info->VM86_TSS_ESP0;
331 if (cpu_has_sep)
332 tsk->thread.sysenter_cs = 0;
333 load_sp0(tss, &tsk->thread);
334 put_cpu();
335
336 tsk->thread.screen_bitmap = info->screen_bitmap;
337 if (info->flags & VM86_SCREEN_BITMAP)
338 mark_screen_rdonly(tsk->mm);
339
340 /*call __audit_syscall_exit since we do not exit via the normal paths */
341#ifdef CONFIG_AUDITSYSCALL
342 if (unlikely(current->audit_context))
343 __audit_syscall_exit(1, 0);
344#endif
345
346 __asm__ __volatile__(
347 "movl %0,%%esp\n\t"
348 "movl %1,%%ebp\n\t"
349#ifdef CONFIG_X86_32_LAZY_GS
350 "mov %2, %%gs\n\t"
351#endif
352 "jmp resume_userspace"
353 : /* no outputs */
354 :"r" (&info->regs), "r" (task_thread_info(tsk)), "r" (0));
355 /* we never return here */
356}
357
358static inline void return_to_32bit(struct kernel_vm86_regs *regs16, int retval)
359{
360 struct pt_regs *regs32;
361
362 regs32 = save_v86_state(regs16);
363 regs32->ax = retval;
364 __asm__ __volatile__("movl %0,%%esp\n\t"
365 "movl %1,%%ebp\n\t"
366 "jmp resume_userspace"
367 : : "r" (regs32), "r" (current_thread_info()));
368}
369
370static inline void set_IF(struct kernel_vm86_regs *regs)
371{
372 VEFLAGS |= X86_EFLAGS_VIF;
373 if (VEFLAGS & X86_EFLAGS_VIP)
374 return_to_32bit(regs, VM86_STI);
375}
376
377static inline void clear_IF(struct kernel_vm86_regs *regs)
378{
379 VEFLAGS &= ~X86_EFLAGS_VIF;
380}
381
382static inline void clear_TF(struct kernel_vm86_regs *regs)
383{
384 regs->pt.flags &= ~X86_EFLAGS_TF;
385}
386
387static inline void clear_AC(struct kernel_vm86_regs *regs)
388{
389 regs->pt.flags &= ~X86_EFLAGS_AC;
390}
391
392/*
393 * It is correct to call set_IF(regs) from the set_vflags_*
394 * functions. However someone forgot to call clear_IF(regs)
395 * in the opposite case.
396 * After the command sequence CLI PUSHF STI POPF you should
397 * end up with interrupts disabled, but you ended up with
398 * interrupts enabled.
399 * ( I was testing my own changes, but the only bug I
400 * could find was in a function I had not changed. )
401 * [KD]
402 */
403
404static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
405{
406 set_flags(VEFLAGS, flags, current->thread.v86mask);
407 set_flags(regs->pt.flags, flags, SAFE_MASK);
408 if (flags & X86_EFLAGS_IF)
409 set_IF(regs);
410 else
411 clear_IF(regs);
412}
413
414static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
415{
416 set_flags(VFLAGS, flags, current->thread.v86mask);
417 set_flags(regs->pt.flags, flags, SAFE_MASK);
418 if (flags & X86_EFLAGS_IF)
419 set_IF(regs);
420 else
421 clear_IF(regs);
422}
423
424static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
425{
426 unsigned long flags = regs->pt.flags & RETURN_MASK;
427
428 if (VEFLAGS & X86_EFLAGS_VIF)
429 flags |= X86_EFLAGS_IF;
430 flags |= X86_EFLAGS_IOPL;
431 return flags | (VEFLAGS & current->thread.v86mask);
432}
433
434static inline int is_revectored(int nr, struct revectored_struct *bitmap)
435{
436 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
437 :"=r" (nr)
438 :"m" (*bitmap), "r" (nr));
439 return nr;
440}
441
442#define val_byte(val, n) (((__u8 *)&val)[n])
443
444#define pushb(base, ptr, val, err_label) \
445 do { \
446 __u8 __val = val; \
447 ptr--; \
448 if (put_user(__val, base + ptr) < 0) \
449 goto err_label; \
450 } while (0)
451
452#define pushw(base, ptr, val, err_label) \
453 do { \
454 __u16 __val = val; \
455 ptr--; \
456 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
457 goto err_label; \
458 ptr--; \
459 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
460 goto err_label; \
461 } while (0)
462
463#define pushl(base, ptr, val, err_label) \
464 do { \
465 __u32 __val = val; \
466 ptr--; \
467 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
468 goto err_label; \
469 ptr--; \
470 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
471 goto err_label; \
472 ptr--; \
473 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
474 goto err_label; \
475 ptr--; \
476 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
477 goto err_label; \
478 } while (0)
479
480#define popb(base, ptr, err_label) \
481 ({ \
482 __u8 __res; \
483 if (get_user(__res, base + ptr) < 0) \
484 goto err_label; \
485 ptr++; \
486 __res; \
487 })
488
489#define popw(base, ptr, err_label) \
490 ({ \
491 __u16 __res; \
492 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
493 goto err_label; \
494 ptr++; \
495 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
496 goto err_label; \
497 ptr++; \
498 __res; \
499 })
500
501#define popl(base, ptr, err_label) \
502 ({ \
503 __u32 __res; \
504 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
505 goto err_label; \
506 ptr++; \
507 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
508 goto err_label; \
509 ptr++; \
510 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
511 goto err_label; \
512 ptr++; \
513 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
514 goto err_label; \
515 ptr++; \
516 __res; \
517 })
518
519/* There are so many possible reasons for this function to return
520 * VM86_INTx, so adding another doesn't bother me. We can expect
521 * userspace programs to be able to handle it. (Getting a problem
522 * in userspace is always better than an Oops anyway.) [KD]
523 */
524static void do_int(struct kernel_vm86_regs *regs, int i,
525 unsigned char __user *ssp, unsigned short sp)
526{
527 unsigned long __user *intr_ptr;
528 unsigned long segoffs;
529
530 if (regs->pt.cs == BIOSSEG)
531 goto cannot_handle;
532 if (is_revectored(i, &KVM86->int_revectored))
533 goto cannot_handle;
534 if (i == 0x21 && is_revectored(AH(regs), &KVM86->int21_revectored))
535 goto cannot_handle;
536 intr_ptr = (unsigned long __user *) (i << 2);
537 if (get_user(segoffs, intr_ptr))
538 goto cannot_handle;
539 if ((segoffs >> 16) == BIOSSEG)
540 goto cannot_handle;
541 pushw(ssp, sp, get_vflags(regs), cannot_handle);
542 pushw(ssp, sp, regs->pt.cs, cannot_handle);
543 pushw(ssp, sp, IP(regs), cannot_handle);
544 regs->pt.cs = segoffs >> 16;
545 SP(regs) -= 6;
546 IP(regs) = segoffs & 0xffff;
547 clear_TF(regs);
548 clear_IF(regs);
549 clear_AC(regs);
550 return;
551
552cannot_handle:
553 return_to_32bit(regs, VM86_INTx + (i << 8));
554}
555
556int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
557{
558 if (VMPI.is_vm86pus) {
559 if ((trapno == 3) || (trapno == 1)) {
560 KVM86->regs32->ax = VM86_TRAP + (trapno << 8);
561 /* setting this flag forces the code in entry_32.S to
562 call save_v86_state() and change the stack pointer
563 to KVM86->regs32 */
564 set_thread_flag(TIF_IRET);
565 return 0;
566 }
567 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
568 return 0;
569 }
570 if (trapno != 1)
571 return 1; /* we let this handle by the calling routine */
572 current->thread.trap_nr = trapno;
573 current->thread.error_code = error_code;
574 force_sig(SIGTRAP, current);
575 return 0;
576}
577
578void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
579{
580 unsigned char opcode;
581 unsigned char __user *csp;
582 unsigned char __user *ssp;
583 unsigned short ip, sp, orig_flags;
584 int data32, pref_done;
585
586#define CHECK_IF_IN_TRAP \
587 if (VMPI.vm86dbg_active && VMPI.vm86dbg_TFpendig) \
588 newflags |= X86_EFLAGS_TF
589#define VM86_FAULT_RETURN do { \
590 if (VMPI.force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) \
591 return_to_32bit(regs, VM86_PICRETURN); \
592 if (orig_flags & X86_EFLAGS_TF) \
593 handle_vm86_trap(regs, 0, 1); \
594 return; } while (0)
595
596 orig_flags = *(unsigned short *)®s->pt.flags;
597
598 csp = (unsigned char __user *) (regs->pt.cs << 4);
599 ssp = (unsigned char __user *) (regs->pt.ss << 4);
600 sp = SP(regs);
601 ip = IP(regs);
602
603 data32 = 0;
604 pref_done = 0;
605 do {
606 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
607 case 0x66: /* 32-bit data */ data32 = 1; break;
608 case 0x67: /* 32-bit address */ break;
609 case 0x2e: /* CS */ break;
610 case 0x3e: /* DS */ break;
611 case 0x26: /* ES */ break;
612 case 0x36: /* SS */ break;
613 case 0x65: /* GS */ break;
614 case 0x64: /* FS */ break;
615 case 0xf2: /* repnz */ break;
616 case 0xf3: /* rep */ break;
617 default: pref_done = 1;
618 }
619 } while (!pref_done);
620
621 switch (opcode) {
622
623 /* pushf */
624 case 0x9c:
625 if (data32) {
626 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
627 SP(regs) -= 4;
628 } else {
629 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
630 SP(regs) -= 2;
631 }
632 IP(regs) = ip;
633 VM86_FAULT_RETURN;
634
635 /* popf */
636 case 0x9d:
637 {
638 unsigned long newflags;
639 if (data32) {
640 newflags = popl(ssp, sp, simulate_sigsegv);
641 SP(regs) += 4;
642 } else {
643 newflags = popw(ssp, sp, simulate_sigsegv);
644 SP(regs) += 2;
645 }
646 IP(regs) = ip;
647 CHECK_IF_IN_TRAP;
648 if (data32)
649 set_vflags_long(newflags, regs);
650 else
651 set_vflags_short(newflags, regs);
652
653 VM86_FAULT_RETURN;
654 }
655
656 /* int xx */
657 case 0xcd: {
658 int intno = popb(csp, ip, simulate_sigsegv);
659 IP(regs) = ip;
660 if (VMPI.vm86dbg_active) {
661 if ((1 << (intno & 7)) & VMPI.vm86dbg_intxxtab[intno >> 3])
662 return_to_32bit(regs, VM86_INTx + (intno << 8));
663 }
664 do_int(regs, intno, ssp, sp);
665 return;
666 }
667
668 /* iret */
669 case 0xcf:
670 {
671 unsigned long newip;
672 unsigned long newcs;
673 unsigned long newflags;
674 if (data32) {
675 newip = popl(ssp, sp, simulate_sigsegv);
676 newcs = popl(ssp, sp, simulate_sigsegv);
677 newflags = popl(ssp, sp, simulate_sigsegv);
678 SP(regs) += 12;
679 } else {
680 newip = popw(ssp, sp, simulate_sigsegv);
681 newcs = popw(ssp, sp, simulate_sigsegv);
682 newflags = popw(ssp, sp, simulate_sigsegv);
683 SP(regs) += 6;
684 }
685 IP(regs) = newip;
686 regs->pt.cs = newcs;
687 CHECK_IF_IN_TRAP;
688 if (data32) {
689 set_vflags_long(newflags, regs);
690 } else {
691 set_vflags_short(newflags, regs);
692 }
693 VM86_FAULT_RETURN;
694 }
695
696 /* cli */
697 case 0xfa:
698 IP(regs) = ip;
699 clear_IF(regs);
700 VM86_FAULT_RETURN;
701
702 /* sti */
703 /*
704 * Damn. This is incorrect: the 'sti' instruction should actually
705 * enable interrupts after the /next/ instruction. Not good.
706 *
707 * Probably needs some horsing around with the TF flag. Aiee..
708 */
709 case 0xfb:
710 IP(regs) = ip;
711 set_IF(regs);
712 VM86_FAULT_RETURN;
713
714 default:
715 return_to_32bit(regs, VM86_UNKNOWN);
716 }
717
718 return;
719
720simulate_sigsegv:
721 /* FIXME: After a long discussion with Stas we finally
722 * agreed, that this is wrong. Here we should
723 * really send a SIGSEGV to the user program.
724 * But how do we create the correct context? We
725 * are inside a general protection fault handler
726 * and has just returned from a page fault handler.
727 * The correct context for the signal handler
728 * should be a mixture of the two, but how do we
729 * get the information? [KD]
730 */
731 return_to_32bit(regs, VM86_UNKNOWN);
732}
733
734/* ---------------- vm86 special IRQ passing stuff ----------------- */
735
736#define VM86_IRQNAME "vm86irq"
737
738static struct vm86_irqs {
739 struct task_struct *tsk;
740 int sig;
741} vm86_irqs[16];
742
743static DEFINE_SPINLOCK(irqbits_lock);
744static int irqbits;
745
746#define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
747 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
748 | (1 << SIGUNUSED))
749
750static irqreturn_t irq_handler(int intno, void *dev_id)
751{
752 int irq_bit;
753 unsigned long flags;
754
755 spin_lock_irqsave(&irqbits_lock, flags);
756 irq_bit = 1 << intno;
757 if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
758 goto out;
759 irqbits |= irq_bit;
760 if (vm86_irqs[intno].sig)
761 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
762 /*
763 * IRQ will be re-enabled when user asks for the irq (whether
764 * polling or as a result of the signal)
765 */
766 disable_irq_nosync(intno);
767 spin_unlock_irqrestore(&irqbits_lock, flags);
768 return IRQ_HANDLED;
769
770out:
771 spin_unlock_irqrestore(&irqbits_lock, flags);
772 return IRQ_NONE;
773}
774
775static inline void free_vm86_irq(int irqnumber)
776{
777 unsigned long flags;
778
779 free_irq(irqnumber, NULL);
780 vm86_irqs[irqnumber].tsk = NULL;
781
782 spin_lock_irqsave(&irqbits_lock, flags);
783 irqbits &= ~(1 << irqnumber);
784 spin_unlock_irqrestore(&irqbits_lock, flags);
785}
786
787void release_vm86_irqs(struct task_struct *task)
788{
789 int i;
790 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
791 if (vm86_irqs[i].tsk == task)
792 free_vm86_irq(i);
793}
794
795static inline int get_and_reset_irq(int irqnumber)
796{
797 int bit;
798 unsigned long flags;
799 int ret = 0;
800
801 if (invalid_vm86_irq(irqnumber)) return 0;
802 if (vm86_irqs[irqnumber].tsk != current) return 0;
803 spin_lock_irqsave(&irqbits_lock, flags);
804 bit = irqbits & (1 << irqnumber);
805 irqbits &= ~bit;
806 if (bit) {
807 enable_irq(irqnumber);
808 ret = 1;
809 }
810
811 spin_unlock_irqrestore(&irqbits_lock, flags);
812 return ret;
813}
814
815
816static int do_vm86_irq_handling(int subfunction, int irqnumber)
817{
818 int ret;
819 switch (subfunction) {
820 case VM86_GET_AND_RESET_IRQ: {
821 return get_and_reset_irq(irqnumber);
822 }
823 case VM86_GET_IRQ_BITS: {
824 return irqbits;
825 }
826 case VM86_REQUEST_IRQ: {
827 int sig = irqnumber >> 8;
828 int irq = irqnumber & 255;
829 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
830 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
831 if (invalid_vm86_irq(irq)) return -EPERM;
832 if (vm86_irqs[irq].tsk) return -EPERM;
833 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
834 if (ret) return ret;
835 vm86_irqs[irq].sig = sig;
836 vm86_irqs[irq].tsk = current;
837 return irq;
838 }
839 case VM86_FREE_IRQ: {
840 if (invalid_vm86_irq(irqnumber)) return -EPERM;
841 if (!vm86_irqs[irqnumber].tsk) return 0;
842 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
843 free_vm86_irq(irqnumber);
844 return 0;
845 }
846 }
847 return -EINVAL;
848}
849
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/kernel.h>
39#include <linux/signal.h>
40#include <linux/string.h>
41#include <linux/mm.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>
48#include <linux/security.h>
49
50#include <asm/uaccess.h>
51#include <asm/io.h>
52#include <asm/tlbflush.h>
53#include <asm/irq.h>
54#include <asm/traps.h>
55#include <asm/vm86.h>
56
57/*
58 * Known problems:
59 *
60 * Interrupt handling is not guaranteed:
61 * - a real x86 will disable all interrupts for one instruction
62 * after a "mov ss,xx" to make stack handling atomic even without
63 * the 'lss' instruction. We can't guarantee this in v86 mode,
64 * as the next instruction might result in a page fault or similar.
65 * - a real x86 will have interrupts disabled for one instruction
66 * past the 'sti' that enables them. We don't bother with all the
67 * details yet.
68 *
69 * Let's hope these problems do not actually matter for anything.
70 */
71
72
73/*
74 * 8- and 16-bit register defines..
75 */
76#define AL(regs) (((unsigned char *)&((regs)->pt.ax))[0])
77#define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1])
78#define IP(regs) (*(unsigned short *)&((regs)->pt.ip))
79#define SP(regs) (*(unsigned short *)&((regs)->pt.sp))
80
81/*
82 * virtual flags (16 and 32-bit versions)
83 */
84#define VFLAGS (*(unsigned short *)&(current->thread.vm86->veflags))
85#define VEFLAGS (current->thread.vm86->veflags)
86
87#define set_flags(X, new, mask) \
88((X) = ((X) & ~(mask)) | ((new) & (mask)))
89
90#define SAFE_MASK (0xDD5)
91#define RETURN_MASK (0xDFF)
92
93void save_v86_state(struct kernel_vm86_regs *regs, int retval)
94{
95 struct tss_struct *tss;
96 struct task_struct *tsk = current;
97 struct vm86plus_struct __user *user;
98 struct vm86 *vm86 = current->thread.vm86;
99 long err = 0;
100
101 /*
102 * This gets called from entry.S with interrupts disabled, but
103 * from process context. Enable interrupts here, before trying
104 * to access user space.
105 */
106 local_irq_enable();
107
108 if (!vm86 || !vm86->user_vm86) {
109 pr_alert("no user_vm86: BAD\n");
110 do_exit(SIGSEGV);
111 }
112 set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | vm86->veflags_mask);
113 user = vm86->user_vm86;
114
115 if (!access_ok(VERIFY_WRITE, user, vm86->vm86plus.is_vm86pus ?
116 sizeof(struct vm86plus_struct) :
117 sizeof(struct vm86_struct))) {
118 pr_alert("could not access userspace vm86 info\n");
119 do_exit(SIGSEGV);
120 }
121
122 put_user_try {
123 put_user_ex(regs->pt.bx, &user->regs.ebx);
124 put_user_ex(regs->pt.cx, &user->regs.ecx);
125 put_user_ex(regs->pt.dx, &user->regs.edx);
126 put_user_ex(regs->pt.si, &user->regs.esi);
127 put_user_ex(regs->pt.di, &user->regs.edi);
128 put_user_ex(regs->pt.bp, &user->regs.ebp);
129 put_user_ex(regs->pt.ax, &user->regs.eax);
130 put_user_ex(regs->pt.ip, &user->regs.eip);
131 put_user_ex(regs->pt.cs, &user->regs.cs);
132 put_user_ex(regs->pt.flags, &user->regs.eflags);
133 put_user_ex(regs->pt.sp, &user->regs.esp);
134 put_user_ex(regs->pt.ss, &user->regs.ss);
135 put_user_ex(regs->es, &user->regs.es);
136 put_user_ex(regs->ds, &user->regs.ds);
137 put_user_ex(regs->fs, &user->regs.fs);
138 put_user_ex(regs->gs, &user->regs.gs);
139
140 put_user_ex(vm86->screen_bitmap, &user->screen_bitmap);
141 } put_user_catch(err);
142 if (err) {
143 pr_alert("could not access userspace vm86 info\n");
144 do_exit(SIGSEGV);
145 }
146
147 tss = &per_cpu(cpu_tss, get_cpu());
148 tsk->thread.sp0 = vm86->saved_sp0;
149 tsk->thread.sysenter_cs = __KERNEL_CS;
150 load_sp0(tss, &tsk->thread);
151 vm86->saved_sp0 = 0;
152 put_cpu();
153
154 memcpy(®s->pt, &vm86->regs32, sizeof(struct pt_regs));
155
156 lazy_load_gs(vm86->regs32.gs);
157
158 regs->pt.ax = retval;
159}
160
161static void mark_screen_rdonly(struct mm_struct *mm)
162{
163 pgd_t *pgd;
164 pud_t *pud;
165 pmd_t *pmd;
166 pte_t *pte;
167 spinlock_t *ptl;
168 int i;
169
170 down_write(&mm->mmap_sem);
171 pgd = pgd_offset(mm, 0xA0000);
172 if (pgd_none_or_clear_bad(pgd))
173 goto out;
174 pud = pud_offset(pgd, 0xA0000);
175 if (pud_none_or_clear_bad(pud))
176 goto out;
177 pmd = pmd_offset(pud, 0xA0000);
178
179 if (pmd_trans_huge(*pmd)) {
180 struct vm_area_struct *vma = find_vma(mm, 0xA0000);
181 split_huge_pmd(vma, pmd, 0xA0000);
182 }
183 if (pmd_none_or_clear_bad(pmd))
184 goto out;
185 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
186 for (i = 0; i < 32; i++) {
187 if (pte_present(*pte))
188 set_pte(pte, pte_wrprotect(*pte));
189 pte++;
190 }
191 pte_unmap_unlock(pte, ptl);
192out:
193 up_write(&mm->mmap_sem);
194 flush_tlb();
195}
196
197
198
199static int do_vm86_irq_handling(int subfunction, int irqnumber);
200static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus);
201
202SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86)
203{
204 return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false);
205}
206
207
208SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg)
209{
210 switch (cmd) {
211 case VM86_REQUEST_IRQ:
212 case VM86_FREE_IRQ:
213 case VM86_GET_IRQ_BITS:
214 case VM86_GET_AND_RESET_IRQ:
215 return do_vm86_irq_handling(cmd, (int)arg);
216 case VM86_PLUS_INSTALL_CHECK:
217 /*
218 * NOTE: on old vm86 stuff this will return the error
219 * from access_ok(), because the subfunction is
220 * interpreted as (invalid) address to vm86_struct.
221 * So the installation check works.
222 */
223 return 0;
224 }
225
226 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
227 return do_sys_vm86((struct vm86plus_struct __user *) arg, true);
228}
229
230
231static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
232{
233 struct tss_struct *tss;
234 struct task_struct *tsk = current;
235 struct vm86 *vm86 = tsk->thread.vm86;
236 struct kernel_vm86_regs vm86regs;
237 struct pt_regs *regs = current_pt_regs();
238 unsigned long err = 0;
239
240 err = security_mmap_addr(0);
241 if (err) {
242 /*
243 * vm86 cannot virtualize the address space, so vm86 users
244 * need to manage the low 1MB themselves using mmap. Given
245 * that BIOS places important data in the first page, vm86
246 * is essentially useless if mmap_min_addr != 0. DOSEMU,
247 * for example, won't even bother trying to use vm86 if it
248 * can't map a page at virtual address 0.
249 *
250 * To reduce the available kernel attack surface, simply
251 * disallow vm86(old) for users who cannot mmap at va 0.
252 *
253 * The implementation of security_mmap_addr will allow
254 * suitably privileged users to map va 0 even if
255 * vm.mmap_min_addr is set above 0, and we want this
256 * behavior for vm86 as well, as it ensures that legacy
257 * tools like vbetool will not fail just because of
258 * vm.mmap_min_addr.
259 */
260 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",
261 current->comm, task_pid_nr(current),
262 from_kuid_munged(&init_user_ns, current_uid()));
263 return -EPERM;
264 }
265
266 if (!vm86) {
267 if (!(vm86 = kzalloc(sizeof(*vm86), GFP_KERNEL)))
268 return -ENOMEM;
269 tsk->thread.vm86 = vm86;
270 }
271 if (vm86->saved_sp0)
272 return -EPERM;
273
274 if (!access_ok(VERIFY_READ, user_vm86, plus ?
275 sizeof(struct vm86_struct) :
276 sizeof(struct vm86plus_struct)))
277 return -EFAULT;
278
279 memset(&vm86regs, 0, sizeof(vm86regs));
280 get_user_try {
281 unsigned short seg;
282 get_user_ex(vm86regs.pt.bx, &user_vm86->regs.ebx);
283 get_user_ex(vm86regs.pt.cx, &user_vm86->regs.ecx);
284 get_user_ex(vm86regs.pt.dx, &user_vm86->regs.edx);
285 get_user_ex(vm86regs.pt.si, &user_vm86->regs.esi);
286 get_user_ex(vm86regs.pt.di, &user_vm86->regs.edi);
287 get_user_ex(vm86regs.pt.bp, &user_vm86->regs.ebp);
288 get_user_ex(vm86regs.pt.ax, &user_vm86->regs.eax);
289 get_user_ex(vm86regs.pt.ip, &user_vm86->regs.eip);
290 get_user_ex(seg, &user_vm86->regs.cs);
291 vm86regs.pt.cs = seg;
292 get_user_ex(vm86regs.pt.flags, &user_vm86->regs.eflags);
293 get_user_ex(vm86regs.pt.sp, &user_vm86->regs.esp);
294 get_user_ex(seg, &user_vm86->regs.ss);
295 vm86regs.pt.ss = seg;
296 get_user_ex(vm86regs.es, &user_vm86->regs.es);
297 get_user_ex(vm86regs.ds, &user_vm86->regs.ds);
298 get_user_ex(vm86regs.fs, &user_vm86->regs.fs);
299 get_user_ex(vm86regs.gs, &user_vm86->regs.gs);
300
301 get_user_ex(vm86->flags, &user_vm86->flags);
302 get_user_ex(vm86->screen_bitmap, &user_vm86->screen_bitmap);
303 get_user_ex(vm86->cpu_type, &user_vm86->cpu_type);
304 } get_user_catch(err);
305 if (err)
306 return err;
307
308 if (copy_from_user(&vm86->int_revectored,
309 &user_vm86->int_revectored,
310 sizeof(struct revectored_struct)))
311 return -EFAULT;
312 if (copy_from_user(&vm86->int21_revectored,
313 &user_vm86->int21_revectored,
314 sizeof(struct revectored_struct)))
315 return -EFAULT;
316 if (plus) {
317 if (copy_from_user(&vm86->vm86plus, &user_vm86->vm86plus,
318 sizeof(struct vm86plus_info_struct)))
319 return -EFAULT;
320 vm86->vm86plus.is_vm86pus = 1;
321 } else
322 memset(&vm86->vm86plus, 0,
323 sizeof(struct vm86plus_info_struct));
324
325 memcpy(&vm86->regs32, regs, sizeof(struct pt_regs));
326 vm86->user_vm86 = user_vm86;
327
328/*
329 * The flags register is also special: we cannot trust that the user
330 * has set it up safely, so this makes sure interrupt etc flags are
331 * inherited from protected mode.
332 */
333 VEFLAGS = vm86regs.pt.flags;
334 vm86regs.pt.flags &= SAFE_MASK;
335 vm86regs.pt.flags |= regs->flags & ~SAFE_MASK;
336 vm86regs.pt.flags |= X86_VM_MASK;
337
338 vm86regs.pt.orig_ax = regs->orig_ax;
339
340 switch (vm86->cpu_type) {
341 case CPU_286:
342 vm86->veflags_mask = 0;
343 break;
344 case CPU_386:
345 vm86->veflags_mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL;
346 break;
347 case CPU_486:
348 vm86->veflags_mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
349 break;
350 default:
351 vm86->veflags_mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL;
352 break;
353 }
354
355/*
356 * Save old state
357 */
358 vm86->saved_sp0 = tsk->thread.sp0;
359 lazy_save_gs(vm86->regs32.gs);
360
361 tss = &per_cpu(cpu_tss, get_cpu());
362 /* make room for real-mode segments */
363 tsk->thread.sp0 += 16;
364
365 if (static_cpu_has(X86_FEATURE_SEP))
366 tsk->thread.sysenter_cs = 0;
367
368 load_sp0(tss, &tsk->thread);
369 put_cpu();
370
371 if (vm86->flags & VM86_SCREEN_BITMAP)
372 mark_screen_rdonly(tsk->mm);
373
374 memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs));
375 force_iret();
376 return regs->ax;
377}
378
379static inline void set_IF(struct kernel_vm86_regs *regs)
380{
381 VEFLAGS |= X86_EFLAGS_VIF;
382}
383
384static inline void clear_IF(struct kernel_vm86_regs *regs)
385{
386 VEFLAGS &= ~X86_EFLAGS_VIF;
387}
388
389static inline void clear_TF(struct kernel_vm86_regs *regs)
390{
391 regs->pt.flags &= ~X86_EFLAGS_TF;
392}
393
394static inline void clear_AC(struct kernel_vm86_regs *regs)
395{
396 regs->pt.flags &= ~X86_EFLAGS_AC;
397}
398
399/*
400 * It is correct to call set_IF(regs) from the set_vflags_*
401 * functions. However someone forgot to call clear_IF(regs)
402 * in the opposite case.
403 * After the command sequence CLI PUSHF STI POPF you should
404 * end up with interrupts disabled, but you ended up with
405 * interrupts enabled.
406 * ( I was testing my own changes, but the only bug I
407 * could find was in a function I had not changed. )
408 * [KD]
409 */
410
411static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs)
412{
413 set_flags(VEFLAGS, flags, current->thread.vm86->veflags_mask);
414 set_flags(regs->pt.flags, flags, SAFE_MASK);
415 if (flags & X86_EFLAGS_IF)
416 set_IF(regs);
417 else
418 clear_IF(regs);
419}
420
421static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs)
422{
423 set_flags(VFLAGS, flags, current->thread.vm86->veflags_mask);
424 set_flags(regs->pt.flags, flags, SAFE_MASK);
425 if (flags & X86_EFLAGS_IF)
426 set_IF(regs);
427 else
428 clear_IF(regs);
429}
430
431static inline unsigned long get_vflags(struct kernel_vm86_regs *regs)
432{
433 unsigned long flags = regs->pt.flags & RETURN_MASK;
434
435 if (VEFLAGS & X86_EFLAGS_VIF)
436 flags |= X86_EFLAGS_IF;
437 flags |= X86_EFLAGS_IOPL;
438 return flags | (VEFLAGS & current->thread.vm86->veflags_mask);
439}
440
441static inline int is_revectored(int nr, struct revectored_struct *bitmap)
442{
443 __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
444 :"=r" (nr)
445 :"m" (*bitmap), "r" (nr));
446 return nr;
447}
448
449#define val_byte(val, n) (((__u8 *)&val)[n])
450
451#define pushb(base, ptr, val, err_label) \
452 do { \
453 __u8 __val = val; \
454 ptr--; \
455 if (put_user(__val, base + ptr) < 0) \
456 goto err_label; \
457 } while (0)
458
459#define pushw(base, ptr, val, err_label) \
460 do { \
461 __u16 __val = val; \
462 ptr--; \
463 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
464 goto err_label; \
465 ptr--; \
466 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
467 goto err_label; \
468 } while (0)
469
470#define pushl(base, ptr, val, err_label) \
471 do { \
472 __u32 __val = val; \
473 ptr--; \
474 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
475 goto err_label; \
476 ptr--; \
477 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
478 goto err_label; \
479 ptr--; \
480 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
481 goto err_label; \
482 ptr--; \
483 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
484 goto err_label; \
485 } while (0)
486
487#define popb(base, ptr, err_label) \
488 ({ \
489 __u8 __res; \
490 if (get_user(__res, base + ptr) < 0) \
491 goto err_label; \
492 ptr++; \
493 __res; \
494 })
495
496#define popw(base, ptr, err_label) \
497 ({ \
498 __u16 __res; \
499 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
500 goto err_label; \
501 ptr++; \
502 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
503 goto err_label; \
504 ptr++; \
505 __res; \
506 })
507
508#define popl(base, ptr, err_label) \
509 ({ \
510 __u32 __res; \
511 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
512 goto err_label; \
513 ptr++; \
514 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
515 goto err_label; \
516 ptr++; \
517 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
518 goto err_label; \
519 ptr++; \
520 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
521 goto err_label; \
522 ptr++; \
523 __res; \
524 })
525
526/* There are so many possible reasons for this function to return
527 * VM86_INTx, so adding another doesn't bother me. We can expect
528 * userspace programs to be able to handle it. (Getting a problem
529 * in userspace is always better than an Oops anyway.) [KD]
530 */
531static void do_int(struct kernel_vm86_regs *regs, int i,
532 unsigned char __user *ssp, unsigned short sp)
533{
534 unsigned long __user *intr_ptr;
535 unsigned long segoffs;
536 struct vm86 *vm86 = current->thread.vm86;
537
538 if (regs->pt.cs == BIOSSEG)
539 goto cannot_handle;
540 if (is_revectored(i, &vm86->int_revectored))
541 goto cannot_handle;
542 if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored))
543 goto cannot_handle;
544 intr_ptr = (unsigned long __user *) (i << 2);
545 if (get_user(segoffs, intr_ptr))
546 goto cannot_handle;
547 if ((segoffs >> 16) == BIOSSEG)
548 goto cannot_handle;
549 pushw(ssp, sp, get_vflags(regs), cannot_handle);
550 pushw(ssp, sp, regs->pt.cs, cannot_handle);
551 pushw(ssp, sp, IP(regs), cannot_handle);
552 regs->pt.cs = segoffs >> 16;
553 SP(regs) -= 6;
554 IP(regs) = segoffs & 0xffff;
555 clear_TF(regs);
556 clear_IF(regs);
557 clear_AC(regs);
558 return;
559
560cannot_handle:
561 save_v86_state(regs, VM86_INTx + (i << 8));
562}
563
564int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
565{
566 struct vm86 *vm86 = current->thread.vm86;
567
568 if (vm86->vm86plus.is_vm86pus) {
569 if ((trapno == 3) || (trapno == 1)) {
570 save_v86_state(regs, VM86_TRAP + (trapno << 8));
571 return 0;
572 }
573 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
574 return 0;
575 }
576 if (trapno != 1)
577 return 1; /* we let this handle by the calling routine */
578 current->thread.trap_nr = trapno;
579 current->thread.error_code = error_code;
580 force_sig(SIGTRAP, current);
581 return 0;
582}
583
584void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
585{
586 unsigned char opcode;
587 unsigned char __user *csp;
588 unsigned char __user *ssp;
589 unsigned short ip, sp, orig_flags;
590 int data32, pref_done;
591 struct vm86plus_info_struct *vmpi = ¤t->thread.vm86->vm86plus;
592
593#define CHECK_IF_IN_TRAP \
594 if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \
595 newflags |= X86_EFLAGS_TF
596
597 orig_flags = *(unsigned short *)®s->pt.flags;
598
599 csp = (unsigned char __user *) (regs->pt.cs << 4);
600 ssp = (unsigned char __user *) (regs->pt.ss << 4);
601 sp = SP(regs);
602 ip = IP(regs);
603
604 data32 = 0;
605 pref_done = 0;
606 do {
607 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
608 case 0x66: /* 32-bit data */ data32 = 1; break;
609 case 0x67: /* 32-bit address */ break;
610 case 0x2e: /* CS */ break;
611 case 0x3e: /* DS */ break;
612 case 0x26: /* ES */ break;
613 case 0x36: /* SS */ break;
614 case 0x65: /* GS */ break;
615 case 0x64: /* FS */ break;
616 case 0xf2: /* repnz */ break;
617 case 0xf3: /* rep */ break;
618 default: pref_done = 1;
619 }
620 } while (!pref_done);
621
622 switch (opcode) {
623
624 /* pushf */
625 case 0x9c:
626 if (data32) {
627 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
628 SP(regs) -= 4;
629 } else {
630 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
631 SP(regs) -= 2;
632 }
633 IP(regs) = ip;
634 goto vm86_fault_return;
635
636 /* popf */
637 case 0x9d:
638 {
639 unsigned long newflags;
640 if (data32) {
641 newflags = popl(ssp, sp, simulate_sigsegv);
642 SP(regs) += 4;
643 } else {
644 newflags = popw(ssp, sp, simulate_sigsegv);
645 SP(regs) += 2;
646 }
647 IP(regs) = ip;
648 CHECK_IF_IN_TRAP;
649 if (data32)
650 set_vflags_long(newflags, regs);
651 else
652 set_vflags_short(newflags, regs);
653
654 goto check_vip;
655 }
656
657 /* int xx */
658 case 0xcd: {
659 int intno = popb(csp, ip, simulate_sigsegv);
660 IP(regs) = ip;
661 if (vmpi->vm86dbg_active) {
662 if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) {
663 save_v86_state(regs, VM86_INTx + (intno << 8));
664 return;
665 }
666 }
667 do_int(regs, intno, ssp, sp);
668 return;
669 }
670
671 /* iret */
672 case 0xcf:
673 {
674 unsigned long newip;
675 unsigned long newcs;
676 unsigned long newflags;
677 if (data32) {
678 newip = popl(ssp, sp, simulate_sigsegv);
679 newcs = popl(ssp, sp, simulate_sigsegv);
680 newflags = popl(ssp, sp, simulate_sigsegv);
681 SP(regs) += 12;
682 } else {
683 newip = popw(ssp, sp, simulate_sigsegv);
684 newcs = popw(ssp, sp, simulate_sigsegv);
685 newflags = popw(ssp, sp, simulate_sigsegv);
686 SP(regs) += 6;
687 }
688 IP(regs) = newip;
689 regs->pt.cs = newcs;
690 CHECK_IF_IN_TRAP;
691 if (data32) {
692 set_vflags_long(newflags, regs);
693 } else {
694 set_vflags_short(newflags, regs);
695 }
696 goto check_vip;
697 }
698
699 /* cli */
700 case 0xfa:
701 IP(regs) = ip;
702 clear_IF(regs);
703 goto vm86_fault_return;
704
705 /* sti */
706 /*
707 * Damn. This is incorrect: the 'sti' instruction should actually
708 * enable interrupts after the /next/ instruction. Not good.
709 *
710 * Probably needs some horsing around with the TF flag. Aiee..
711 */
712 case 0xfb:
713 IP(regs) = ip;
714 set_IF(regs);
715 goto check_vip;
716
717 default:
718 save_v86_state(regs, VM86_UNKNOWN);
719 }
720
721 return;
722
723check_vip:
724 if (VEFLAGS & X86_EFLAGS_VIP) {
725 save_v86_state(regs, VM86_STI);
726 return;
727 }
728
729vm86_fault_return:
730 if (vmpi->force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) {
731 save_v86_state(regs, VM86_PICRETURN);
732 return;
733 }
734 if (orig_flags & X86_EFLAGS_TF)
735 handle_vm86_trap(regs, 0, X86_TRAP_DB);
736 return;
737
738simulate_sigsegv:
739 /* FIXME: After a long discussion with Stas we finally
740 * agreed, that this is wrong. Here we should
741 * really send a SIGSEGV to the user program.
742 * But how do we create the correct context? We
743 * are inside a general protection fault handler
744 * and has just returned from a page fault handler.
745 * The correct context for the signal handler
746 * should be a mixture of the two, but how do we
747 * get the information? [KD]
748 */
749 save_v86_state(regs, VM86_UNKNOWN);
750}
751
752/* ---------------- vm86 special IRQ passing stuff ----------------- */
753
754#define VM86_IRQNAME "vm86irq"
755
756static struct vm86_irqs {
757 struct task_struct *tsk;
758 int sig;
759} vm86_irqs[16];
760
761static DEFINE_SPINLOCK(irqbits_lock);
762static int irqbits;
763
764#define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
765 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
766 | (1 << SIGUNUSED))
767
768static irqreturn_t irq_handler(int intno, void *dev_id)
769{
770 int irq_bit;
771 unsigned long flags;
772
773 spin_lock_irqsave(&irqbits_lock, flags);
774 irq_bit = 1 << intno;
775 if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
776 goto out;
777 irqbits |= irq_bit;
778 if (vm86_irqs[intno].sig)
779 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
780 /*
781 * IRQ will be re-enabled when user asks for the irq (whether
782 * polling or as a result of the signal)
783 */
784 disable_irq_nosync(intno);
785 spin_unlock_irqrestore(&irqbits_lock, flags);
786 return IRQ_HANDLED;
787
788out:
789 spin_unlock_irqrestore(&irqbits_lock, flags);
790 return IRQ_NONE;
791}
792
793static inline void free_vm86_irq(int irqnumber)
794{
795 unsigned long flags;
796
797 free_irq(irqnumber, NULL);
798 vm86_irqs[irqnumber].tsk = NULL;
799
800 spin_lock_irqsave(&irqbits_lock, flags);
801 irqbits &= ~(1 << irqnumber);
802 spin_unlock_irqrestore(&irqbits_lock, flags);
803}
804
805void release_vm86_irqs(struct task_struct *task)
806{
807 int i;
808 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
809 if (vm86_irqs[i].tsk == task)
810 free_vm86_irq(i);
811}
812
813static inline int get_and_reset_irq(int irqnumber)
814{
815 int bit;
816 unsigned long flags;
817 int ret = 0;
818
819 if (invalid_vm86_irq(irqnumber)) return 0;
820 if (vm86_irqs[irqnumber].tsk != current) return 0;
821 spin_lock_irqsave(&irqbits_lock, flags);
822 bit = irqbits & (1 << irqnumber);
823 irqbits &= ~bit;
824 if (bit) {
825 enable_irq(irqnumber);
826 ret = 1;
827 }
828
829 spin_unlock_irqrestore(&irqbits_lock, flags);
830 return ret;
831}
832
833
834static int do_vm86_irq_handling(int subfunction, int irqnumber)
835{
836 int ret;
837 switch (subfunction) {
838 case VM86_GET_AND_RESET_IRQ: {
839 return get_and_reset_irq(irqnumber);
840 }
841 case VM86_GET_IRQ_BITS: {
842 return irqbits;
843 }
844 case VM86_REQUEST_IRQ: {
845 int sig = irqnumber >> 8;
846 int irq = irqnumber & 255;
847 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
848 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
849 if (invalid_vm86_irq(irq)) return -EPERM;
850 if (vm86_irqs[irq].tsk) return -EPERM;
851 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
852 if (ret) return ret;
853 vm86_irqs[irq].sig = sig;
854 vm86_irqs[irq].tsk = current;
855 return irq;
856 }
857 case VM86_FREE_IRQ: {
858 if (invalid_vm86_irq(irqnumber)) return -EPERM;
859 if (!vm86_irqs[irqnumber].tsk) return 0;
860 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
861 free_vm86_irq(irqnumber);
862 return 0;
863 }
864 }
865 return -EINVAL;
866}
867