1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 1994 Linus Torvalds
5 * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86
8 * 22 mar 2002 - Manfred detected the stackfaults, but didn't handle
9 * them correctly. Now the emulation will be in a
10 * consistent state after stackfaults - Kasper Dupont
13 * 22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont
16 * ?? ??? 2002 - Fixed premature returns from handle_vm86_fault
17 * caused by Kasper Dupont's changes - Stas Sergeev
19 * 4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes.
22 * 9 apr 2002 - Changed syntax of macros in handle_vm86_fault.
25 * 9 apr 2002 - Changed stack access macros to jump to a label
26 * instead of returning to userspace. This simplifies
27 * do_int, and is needed by handle_vm6_fault. Kasper
32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34 #include <linux/capability.h>
35 #include <linux/errno.h>
36 #include <linux/interrupt.h>
37 #include <linux/syscalls.h>
38 #include <linux/sched.h>
39 #include <linux/sched/task_stack.h>
40 #include <linux/kernel.h>
41 #include <linux/signal.h>
42 #include <linux/string.h>
44 #include <linux/smp.h>
45 #include <linux/highmem.h>
46 #include <linux/ptrace.h>
47 #include <linux/audit.h>
48 #include <linux/stddef.h>
49 #include <linux/slab.h>
50 #include <linux/security.h>
52 #include <linux/uaccess.h>
54 #include <asm/tlbflush.h>
56 #include <asm/traps.h>
58 #include <asm/switch_to.h>
63 * Interrupt handling is not guaranteed:
64 * - a real x86 will disable all interrupts for one instruction
65 * after a "mov ss,xx" to make stack handling atomic even without
66 * the 'lss' instruction. We can't guarantee this in v86 mode,
67 * as the next instruction might result in a page fault or similar.
68 * - a real x86 will have interrupts disabled for one instruction
69 * past the 'sti' that enables them. We don't bother with all the
72 * Let's hope these problems do not actually matter for anything.
77 * 8- and 16-bit register defines..
79 #define AL(regs) (((unsigned char *)&((regs)->pt.ax))[0])
80 #define AH(regs) (((unsigned char *)&((regs)->pt.ax))[1])
81 #define IP(regs) (*(unsigned short *)&((regs)->pt.ip))
82 #define SP(regs) (*(unsigned short *)&((regs)->pt.sp))
85 * virtual flags (16 and 32-bit versions)
87 #define VFLAGS (*(unsigned short *)&(current->thread.vm86->veflags))
88 #define VEFLAGS (current->thread.vm86->veflags)
90 #define set_flags(X, new, mask) \
91 ((X) = ((X) & ~(mask)) | ((new) & (mask)))
93 #define SAFE_MASK (0xDD5)
94 #define RETURN_MASK (0xDFF)
96 void save_v86_state(struct kernel_vm86_regs *regs, int retval)
98 struct task_struct *tsk = current;
99 struct vm86plus_struct __user *user;
100 struct vm86 *vm86 = current->thread.vm86;
103 * This gets called from entry.S with interrupts disabled, but
104 * from process context. Enable interrupts here, before trying
105 * to access user space.
111 set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | vm86->veflags_mask);
112 user = vm86->user_vm86;
114 if (!user_access_begin(user, vm86->vm86plus.is_vm86pus ?
115 sizeof(struct vm86plus_struct) :
116 sizeof(struct vm86_struct)))
119 unsafe_put_user(regs->pt.bx, &user->regs.ebx, Efault_end);
120 unsafe_put_user(regs->pt.cx, &user->regs.ecx, Efault_end);
121 unsafe_put_user(regs->pt.dx, &user->regs.edx, Efault_end);
122 unsafe_put_user(regs->pt.si, &user->regs.esi, Efault_end);
123 unsafe_put_user(regs->pt.di, &user->regs.edi, Efault_end);
124 unsafe_put_user(regs->pt.bp, &user->regs.ebp, Efault_end);
125 unsafe_put_user(regs->pt.ax, &user->regs.eax, Efault_end);
126 unsafe_put_user(regs->pt.ip, &user->regs.eip, Efault_end);
127 unsafe_put_user(regs->pt.cs, &user->regs.cs, Efault_end);
128 unsafe_put_user(regs->pt.flags, &user->regs.eflags, Efault_end);
129 unsafe_put_user(regs->pt.sp, &user->regs.esp, Efault_end);
130 unsafe_put_user(regs->pt.ss, &user->regs.ss, Efault_end);
131 unsafe_put_user(regs->es, &user->regs.es, Efault_end);
132 unsafe_put_user(regs->ds, &user->regs.ds, Efault_end);
133 unsafe_put_user(regs->fs, &user->regs.fs, Efault_end);
134 unsafe_put_user(regs->gs, &user->regs.gs, Efault_end);
137 * Don't write screen_bitmap in case some user had a value there
138 * and expected it to remain unchanged.
145 tsk->thread.sp0 = vm86->saved_sp0;
146 tsk->thread.sysenter_cs = __KERNEL_CS;
147 update_task_stack(tsk);
148 refresh_sysenter_cs(&tsk->thread);
152 memcpy(®s->pt, &vm86->regs32, sizeof(struct pt_regs));
154 loadsegment(gs, vm86->regs32.gs);
156 regs->pt.ax = retval;
162 pr_alert("could not access userspace vm86 info\n");
163 force_exit_sig(SIGSEGV);
167 static int do_vm86_irq_handling(int subfunction, int irqnumber);
168 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus);
170 SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86)
172 return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false);
176 SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg)
179 case VM86_REQUEST_IRQ:
181 case VM86_GET_IRQ_BITS:
182 case VM86_GET_AND_RESET_IRQ:
183 return do_vm86_irq_handling(cmd, (int)arg);
184 case VM86_PLUS_INSTALL_CHECK:
186 * NOTE: on old vm86 stuff this will return the error
187 * from access_ok(), because the subfunction is
188 * interpreted as (invalid) address to vm86_struct.
189 * So the installation check works.
194 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */
195 return do_sys_vm86((struct vm86plus_struct __user *) arg, true);
199 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
201 struct task_struct *tsk = current;
202 struct vm86 *vm86 = tsk->thread.vm86;
203 struct kernel_vm86_regs vm86regs;
204 struct pt_regs *regs = current_pt_regs();
205 unsigned long err = 0;
206 struct vm86_struct v;
208 err = security_mmap_addr(0);
211 * vm86 cannot virtualize the address space, so vm86 users
212 * need to manage the low 1MB themselves using mmap. Given
213 * that BIOS places important data in the first page, vm86
214 * is essentially useless if mmap_min_addr != 0. DOSEMU,
215 * for example, won't even bother trying to use vm86 if it
216 * can't map a page at virtual address 0.
218 * To reduce the available kernel attack surface, simply
219 * disallow vm86(old) for users who cannot mmap at va 0.
221 * The implementation of security_mmap_addr will allow
222 * suitably privileged users to map va 0 even if
223 * vm.mmap_min_addr is set above 0, and we want this
224 * behavior for vm86 as well, as it ensures that legacy
225 * tools like vbetool will not fail just because of
228 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",
229 current->comm, task_pid_nr(current),
230 from_kuid_munged(&init_user_ns, current_uid()));
235 if (!(vm86 = kzalloc(sizeof(*vm86), GFP_KERNEL)))
237 tsk->thread.vm86 = vm86;
242 if (copy_from_user(&v, user_vm86,
243 offsetof(struct vm86_struct, int_revectored)))
247 /* VM86_SCREEN_BITMAP had numerous bugs and appears to have no users. */
248 if (v.flags & VM86_SCREEN_BITMAP) {
249 char comm[TASK_COMM_LEN];
251 pr_info_once("vm86: '%s' uses VM86_SCREEN_BITMAP, which is no longer supported\n", get_task_comm(comm, current));
255 memset(&vm86regs, 0, sizeof(vm86regs));
257 vm86regs.pt.bx = v.regs.ebx;
258 vm86regs.pt.cx = v.regs.ecx;
259 vm86regs.pt.dx = v.regs.edx;
260 vm86regs.pt.si = v.regs.esi;
261 vm86regs.pt.di = v.regs.edi;
262 vm86regs.pt.bp = v.regs.ebp;
263 vm86regs.pt.ax = v.regs.eax;
264 vm86regs.pt.ip = v.regs.eip;
265 vm86regs.pt.cs = v.regs.cs;
266 vm86regs.pt.flags = v.regs.eflags;
267 vm86regs.pt.sp = v.regs.esp;
268 vm86regs.pt.ss = v.regs.ss;
269 vm86regs.es = v.regs.es;
270 vm86regs.ds = v.regs.ds;
271 vm86regs.fs = v.regs.fs;
272 vm86regs.gs = v.regs.gs;
274 vm86->flags = v.flags;
275 vm86->cpu_type = v.cpu_type;
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 savesegment(gs, vm86->regs32.gs);
330 /* make room for real-mode segments */
332 tsk->thread.sp0 += 16;
334 if (boot_cpu_has(X86_FEATURE_SEP)) {
335 tsk->thread.sysenter_cs = 0;
336 refresh_sysenter_cs(&tsk->thread);
339 update_task_stack(tsk);
342 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 return test_bit(nr, bitmap->__map);
413 #define val_byte(val, n) (((__u8 *)&val)[n])
415 #define pushb(base, ptr, val, err_label) \
419 if (put_user(__val, base + ptr) < 0) \
423 #define pushw(base, ptr, val, err_label) \
427 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
430 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
434 #define pushl(base, ptr, val, err_label) \
438 if (put_user(val_byte(__val, 3), base + ptr) < 0) \
441 if (put_user(val_byte(__val, 2), base + ptr) < 0) \
444 if (put_user(val_byte(__val, 1), base + ptr) < 0) \
447 if (put_user(val_byte(__val, 0), base + ptr) < 0) \
451 #define popb(base, ptr, err_label) \
454 if (get_user(__res, base + ptr) < 0) \
460 #define popw(base, ptr, err_label) \
463 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
466 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
472 #define popl(base, ptr, err_label) \
475 if (get_user(val_byte(__res, 0), base + ptr) < 0) \
478 if (get_user(val_byte(__res, 1), base + ptr) < 0) \
481 if (get_user(val_byte(__res, 2), base + ptr) < 0) \
484 if (get_user(val_byte(__res, 3), base + ptr) < 0) \
490 /* There are so many possible reasons for this function to return
491 * VM86_INTx, so adding another doesn't bother me. We can expect
492 * userspace programs to be able to handle it. (Getting a problem
493 * in userspace is always better than an Oops anyway.) [KD]
495 static void do_int(struct kernel_vm86_regs *regs, int i,
496 unsigned char __user *ssp, unsigned short sp)
498 unsigned long __user *intr_ptr;
499 unsigned long segoffs;
500 struct vm86 *vm86 = current->thread.vm86;
502 if (regs->pt.cs == BIOSSEG)
504 if (is_revectored(i, &vm86->int_revectored))
506 if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored))
508 intr_ptr = (unsigned long __user *) (i << 2);
509 if (get_user(segoffs, intr_ptr))
511 if ((segoffs >> 16) == BIOSSEG)
513 pushw(ssp, sp, get_vflags(regs), cannot_handle);
514 pushw(ssp, sp, regs->pt.cs, cannot_handle);
515 pushw(ssp, sp, IP(regs), cannot_handle);
516 regs->pt.cs = segoffs >> 16;
518 IP(regs) = segoffs & 0xffff;
525 save_v86_state(regs, VM86_INTx + (i << 8));
528 int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno)
530 struct vm86 *vm86 = current->thread.vm86;
532 if (vm86->vm86plus.is_vm86pus) {
533 if ((trapno == 3) || (trapno == 1)) {
534 save_v86_state(regs, VM86_TRAP + (trapno << 8));
537 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs));
541 return 1; /* we let this handle by the calling routine */
542 current->thread.trap_nr = trapno;
543 current->thread.error_code = error_code;
548 void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code)
550 unsigned char opcode;
551 unsigned char __user *csp;
552 unsigned char __user *ssp;
553 unsigned short ip, sp, orig_flags;
554 int data32, pref_done;
555 struct vm86plus_info_struct *vmpi = ¤t->thread.vm86->vm86plus;
557 #define CHECK_IF_IN_TRAP \
558 if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \
559 newflags |= X86_EFLAGS_TF
561 orig_flags = *(unsigned short *)®s->pt.flags;
563 csp = (unsigned char __user *) (regs->pt.cs << 4);
564 ssp = (unsigned char __user *) (regs->pt.ss << 4);
571 switch (opcode = popb(csp, ip, simulate_sigsegv)) {
572 case 0x66: /* 32-bit data */ data32 = 1; break;
573 case 0x67: /* 32-bit address */ break;
574 case 0x2e: /* CS */ break;
575 case 0x3e: /* DS */ break;
576 case 0x26: /* ES */ break;
577 case 0x36: /* SS */ break;
578 case 0x65: /* GS */ break;
579 case 0x64: /* FS */ break;
580 case 0xf2: /* repnz */ break;
581 case 0xf3: /* rep */ break;
582 default: pref_done = 1;
584 } while (!pref_done);
591 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv);
594 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv);
598 goto vm86_fault_return;
603 unsigned long newflags;
605 newflags = popl(ssp, sp, simulate_sigsegv);
608 newflags = popw(ssp, sp, simulate_sigsegv);
614 set_vflags_long(newflags, regs);
616 set_vflags_short(newflags, regs);
623 int intno = popb(csp, ip, simulate_sigsegv);
625 if (vmpi->vm86dbg_active) {
626 if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) {
627 save_v86_state(regs, VM86_INTx + (intno << 8));
631 do_int(regs, intno, ssp, sp);
640 unsigned long newflags;
642 newip = popl(ssp, sp, simulate_sigsegv);
643 newcs = popl(ssp, sp, simulate_sigsegv);
644 newflags = popl(ssp, sp, simulate_sigsegv);
647 newip = popw(ssp, sp, simulate_sigsegv);
648 newcs = popw(ssp, sp, simulate_sigsegv);
649 newflags = popw(ssp, sp, simulate_sigsegv);
656 set_vflags_long(newflags, regs);
658 set_vflags_short(newflags, regs);
667 goto vm86_fault_return;
671 * Damn. This is incorrect: the 'sti' instruction should actually
672 * enable interrupts after the /next/ instruction. Not good.
674 * Probably needs some horsing around with the TF flag. Aiee..
682 save_v86_state(regs, VM86_UNKNOWN);
688 if ((VEFLAGS & (X86_EFLAGS_VIP | X86_EFLAGS_VIF)) ==
689 (X86_EFLAGS_VIP | X86_EFLAGS_VIF)) {
690 save_v86_state(regs, VM86_STI);
695 if (vmpi->force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) {
696 save_v86_state(regs, VM86_PICRETURN);
699 if (orig_flags & X86_EFLAGS_TF)
700 handle_vm86_trap(regs, 0, X86_TRAP_DB);
704 /* FIXME: After a long discussion with Stas we finally
705 * agreed, that this is wrong. Here we should
706 * really send a SIGSEGV to the user program.
707 * But how do we create the correct context? We
708 * are inside a general protection fault handler
709 * and has just returned from a page fault handler.
710 * The correct context for the signal handler
711 * should be a mixture of the two, but how do we
712 * get the information? [KD]
714 save_v86_state(regs, VM86_UNKNOWN);
717 /* ---------------- vm86 special IRQ passing stuff ----------------- */
719 #define VM86_IRQNAME "vm86irq"
721 static struct vm86_irqs {
722 struct task_struct *tsk;
726 static DEFINE_SPINLOCK(irqbits_lock);
729 #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \
730 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \
733 static irqreturn_t irq_handler(int intno, void *dev_id)
738 spin_lock_irqsave(&irqbits_lock, flags);
739 irq_bit = 1 << intno;
740 if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk)
743 if (vm86_irqs[intno].sig)
744 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1);
746 * IRQ will be re-enabled when user asks for the irq (whether
747 * polling or as a result of the signal)
749 disable_irq_nosync(intno);
750 spin_unlock_irqrestore(&irqbits_lock, flags);
754 spin_unlock_irqrestore(&irqbits_lock, flags);
758 static inline void free_vm86_irq(int irqnumber)
762 free_irq(irqnumber, NULL);
763 vm86_irqs[irqnumber].tsk = NULL;
765 spin_lock_irqsave(&irqbits_lock, flags);
766 irqbits &= ~(1 << irqnumber);
767 spin_unlock_irqrestore(&irqbits_lock, flags);
770 void release_vm86_irqs(struct task_struct *task)
773 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++)
774 if (vm86_irqs[i].tsk == task)
778 static inline int get_and_reset_irq(int irqnumber)
784 if (invalid_vm86_irq(irqnumber)) return 0;
785 if (vm86_irqs[irqnumber].tsk != current) return 0;
786 spin_lock_irqsave(&irqbits_lock, flags);
787 bit = irqbits & (1 << irqnumber);
790 enable_irq(irqnumber);
794 spin_unlock_irqrestore(&irqbits_lock, flags);
799 static int do_vm86_irq_handling(int subfunction, int irqnumber)
802 switch (subfunction) {
803 case VM86_GET_AND_RESET_IRQ: {
804 return get_and_reset_irq(irqnumber);
806 case VM86_GET_IRQ_BITS: {
809 case VM86_REQUEST_IRQ: {
810 int sig = irqnumber >> 8;
811 int irq = irqnumber & 255;
812 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
813 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM;
814 if (invalid_vm86_irq(irq)) return -EPERM;
815 if (vm86_irqs[irq].tsk) return -EPERM;
816 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL);
818 vm86_irqs[irq].sig = sig;
819 vm86_irqs[irq].tsk = current;
822 case VM86_FREE_IRQ: {
823 if (invalid_vm86_irq(irqnumber)) return -EPERM;
824 if (!vm86_irqs[irqnumber].tsk) return 0;
825 if (vm86_irqs[irqnumber].tsk != current) return -EPERM;
826 free_vm86_irq(irqnumber);