1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 1994 Linus Torvalds 4 * 5 * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86 6 * stack - Manfred Spraul <manfred@colorfullife.com> 7 * 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 11 * <kasperd@daimi.au.dk> 12 * 13 * 22 mar 2002 - Added missing clear_IF in set_vflags_* Kasper Dupont 14 * <kasperd@daimi.au.dk> 15 * 16 * ?? ??? 2002 - Fixed premature returns from handle_vm86_fault 17 * caused by Kasper Dupont's changes - Stas Sergeev 18 * 19 * 4 apr 2002 - Fixed CHECK_IF_IN_TRAP broken by Stas' changes. 20 * Kasper Dupont <kasperd@daimi.au.dk> 21 * 22 * 9 apr 2002 - Changed syntax of macros in handle_vm86_fault. 23 * Kasper Dupont <kasperd@daimi.au.dk> 24 * 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 28 * Dupont <kasperd@daimi.au.dk> 29 * 30 */ 31 32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 33 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> 43 #include <linux/mm.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> 51 52 #include <linux/uaccess.h> 53 #include <asm/io.h> 54 #include <asm/tlbflush.h> 55 #include <asm/irq.h> 56 #include <asm/traps.h> 57 #include <asm/vm86.h> 58 #include <asm/switch_to.h> 59 60 /* 61 * Known problems: 62 * 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 70 * details yet. 71 * 72 * Let's hope these problems do not actually matter for anything. 73 */ 74 75 76 /* 77 * 8- and 16-bit register defines.. 78 */ 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)) 83 84 /* 85 * virtual flags (16 and 32-bit versions) 86 */ 87 #define VFLAGS (*(unsigned short *)&(current->thread.vm86->veflags)) 88 #define VEFLAGS (current->thread.vm86->veflags) 89 90 #define set_flags(X, new, mask) \ 91 ((X) = ((X) & ~(mask)) | ((new) & (mask))) 92 93 #define SAFE_MASK (0xDD5) 94 #define RETURN_MASK (0xDFF) 95 96 void save_v86_state(struct kernel_vm86_regs *regs, int retval) 97 { 98 struct task_struct *tsk = current; 99 struct vm86plus_struct __user *user; 100 struct vm86 *vm86 = current->thread.vm86; 101 102 /* 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. 106 */ 107 local_irq_enable(); 108 109 if (!vm86 || !vm86->user_vm86) { 110 pr_alert("no user_vm86: BAD\n"); 111 do_exit(SIGSEGV); 112 } 113 set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | vm86->veflags_mask); 114 user = vm86->user_vm86; 115 116 if (!user_access_begin(user, vm86->vm86plus.is_vm86pus ? 117 sizeof(struct vm86plus_struct) : 118 sizeof(struct vm86_struct))) 119 goto Efault; 120 121 unsafe_put_user(regs->pt.bx, &user->regs.ebx, Efault_end); 122 unsafe_put_user(regs->pt.cx, &user->regs.ecx, Efault_end); 123 unsafe_put_user(regs->pt.dx, &user->regs.edx, Efault_end); 124 unsafe_put_user(regs->pt.si, &user->regs.esi, Efault_end); 125 unsafe_put_user(regs->pt.di, &user->regs.edi, Efault_end); 126 unsafe_put_user(regs->pt.bp, &user->regs.ebp, Efault_end); 127 unsafe_put_user(regs->pt.ax, &user->regs.eax, Efault_end); 128 unsafe_put_user(regs->pt.ip, &user->regs.eip, Efault_end); 129 unsafe_put_user(regs->pt.cs, &user->regs.cs, Efault_end); 130 unsafe_put_user(regs->pt.flags, &user->regs.eflags, Efault_end); 131 unsafe_put_user(regs->pt.sp, &user->regs.esp, Efault_end); 132 unsafe_put_user(regs->pt.ss, &user->regs.ss, Efault_end); 133 unsafe_put_user(regs->es, &user->regs.es, Efault_end); 134 unsafe_put_user(regs->ds, &user->regs.ds, Efault_end); 135 unsafe_put_user(regs->fs, &user->regs.fs, Efault_end); 136 unsafe_put_user(regs->gs, &user->regs.gs, Efault_end); 137 unsafe_put_user(vm86->screen_bitmap, &user->screen_bitmap, Efault_end); 138 139 user_access_end(); 140 141 preempt_disable(); 142 tsk->thread.sp0 = vm86->saved_sp0; 143 tsk->thread.sysenter_cs = __KERNEL_CS; 144 update_task_stack(tsk); 145 refresh_sysenter_cs(&tsk->thread); 146 vm86->saved_sp0 = 0; 147 preempt_enable(); 148 149 memcpy(®s->pt, &vm86->regs32, sizeof(struct pt_regs)); 150 151 lazy_load_gs(vm86->regs32.gs); 152 153 regs->pt.ax = retval; 154 return; 155 156 Efault_end: 157 user_access_end(); 158 Efault: 159 pr_alert("could not access userspace vm86 info\n"); 160 do_exit(SIGSEGV); 161 } 162 163 static void mark_screen_rdonly(struct mm_struct *mm) 164 { 165 struct vm_area_struct *vma; 166 spinlock_t *ptl; 167 pgd_t *pgd; 168 p4d_t *p4d; 169 pud_t *pud; 170 pmd_t *pmd; 171 pte_t *pte; 172 int i; 173 174 mmap_write_lock(mm); 175 pgd = pgd_offset(mm, 0xA0000); 176 if (pgd_none_or_clear_bad(pgd)) 177 goto out; 178 p4d = p4d_offset(pgd, 0xA0000); 179 if (p4d_none_or_clear_bad(p4d)) 180 goto out; 181 pud = pud_offset(p4d, 0xA0000); 182 if (pud_none_or_clear_bad(pud)) 183 goto out; 184 pmd = pmd_offset(pud, 0xA0000); 185 186 if (pmd_trans_huge(*pmd)) { 187 vma = find_vma(mm, 0xA0000); 188 split_huge_pmd(vma, pmd, 0xA0000); 189 } 190 if (pmd_none_or_clear_bad(pmd)) 191 goto out; 192 pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl); 193 for (i = 0; i < 32; i++) { 194 if (pte_present(*pte)) 195 set_pte(pte, pte_wrprotect(*pte)); 196 pte++; 197 } 198 pte_unmap_unlock(pte, ptl); 199 out: 200 mmap_write_unlock(mm); 201 flush_tlb_mm_range(mm, 0xA0000, 0xA0000 + 32*PAGE_SIZE, PAGE_SHIFT, false); 202 } 203 204 205 206 static int do_vm86_irq_handling(int subfunction, int irqnumber); 207 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus); 208 209 SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86) 210 { 211 return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false); 212 } 213 214 215 SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg) 216 { 217 switch (cmd) { 218 case VM86_REQUEST_IRQ: 219 case VM86_FREE_IRQ: 220 case VM86_GET_IRQ_BITS: 221 case VM86_GET_AND_RESET_IRQ: 222 return do_vm86_irq_handling(cmd, (int)arg); 223 case VM86_PLUS_INSTALL_CHECK: 224 /* 225 * NOTE: on old vm86 stuff this will return the error 226 * from access_ok(), because the subfunction is 227 * interpreted as (invalid) address to vm86_struct. 228 * So the installation check works. 229 */ 230 return 0; 231 } 232 233 /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */ 234 return do_sys_vm86((struct vm86plus_struct __user *) arg, true); 235 } 236 237 238 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus) 239 { 240 struct task_struct *tsk = current; 241 struct vm86 *vm86 = tsk->thread.vm86; 242 struct kernel_vm86_regs vm86regs; 243 struct pt_regs *regs = current_pt_regs(); 244 unsigned long err = 0; 245 struct vm86_struct v; 246 247 err = security_mmap_addr(0); 248 if (err) { 249 /* 250 * vm86 cannot virtualize the address space, so vm86 users 251 * need to manage the low 1MB themselves using mmap. Given 252 * that BIOS places important data in the first page, vm86 253 * is essentially useless if mmap_min_addr != 0. DOSEMU, 254 * for example, won't even bother trying to use vm86 if it 255 * can't map a page at virtual address 0. 256 * 257 * To reduce the available kernel attack surface, simply 258 * disallow vm86(old) for users who cannot mmap at va 0. 259 * 260 * The implementation of security_mmap_addr will allow 261 * suitably privileged users to map va 0 even if 262 * vm.mmap_min_addr is set above 0, and we want this 263 * behavior for vm86 as well, as it ensures that legacy 264 * tools like vbetool will not fail just because of 265 * vm.mmap_min_addr. 266 */ 267 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", 268 current->comm, task_pid_nr(current), 269 from_kuid_munged(&init_user_ns, current_uid())); 270 return -EPERM; 271 } 272 273 if (!vm86) { 274 if (!(vm86 = kzalloc(sizeof(*vm86), GFP_KERNEL))) 275 return -ENOMEM; 276 tsk->thread.vm86 = vm86; 277 } 278 if (vm86->saved_sp0) 279 return -EPERM; 280 281 if (copy_from_user(&v, user_vm86, 282 offsetof(struct vm86_struct, int_revectored))) 283 return -EFAULT; 284 285 memset(&vm86regs, 0, sizeof(vm86regs)); 286 287 vm86regs.pt.bx = v.regs.ebx; 288 vm86regs.pt.cx = v.regs.ecx; 289 vm86regs.pt.dx = v.regs.edx; 290 vm86regs.pt.si = v.regs.esi; 291 vm86regs.pt.di = v.regs.edi; 292 vm86regs.pt.bp = v.regs.ebp; 293 vm86regs.pt.ax = v.regs.eax; 294 vm86regs.pt.ip = v.regs.eip; 295 vm86regs.pt.cs = v.regs.cs; 296 vm86regs.pt.flags = v.regs.eflags; 297 vm86regs.pt.sp = v.regs.esp; 298 vm86regs.pt.ss = v.regs.ss; 299 vm86regs.es = v.regs.es; 300 vm86regs.ds = v.regs.ds; 301 vm86regs.fs = v.regs.fs; 302 vm86regs.gs = v.regs.gs; 303 304 vm86->flags = v.flags; 305 vm86->screen_bitmap = v.screen_bitmap; 306 vm86->cpu_type = v.cpu_type; 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 /* make room for real-mode segments */ 362 preempt_disable(); 363 tsk->thread.sp0 += 16; 364 365 if (boot_cpu_has(X86_FEATURE_SEP)) { 366 tsk->thread.sysenter_cs = 0; 367 refresh_sysenter_cs(&tsk->thread); 368 } 369 370 update_task_stack(tsk); 371 preempt_enable(); 372 373 if (vm86->flags & VM86_SCREEN_BITMAP) 374 mark_screen_rdonly(tsk->mm); 375 376 memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs)); 377 return regs->ax; 378 } 379 380 static inline void set_IF(struct kernel_vm86_regs *regs) 381 { 382 VEFLAGS |= X86_EFLAGS_VIF; 383 } 384 385 static inline void clear_IF(struct kernel_vm86_regs *regs) 386 { 387 VEFLAGS &= ~X86_EFLAGS_VIF; 388 } 389 390 static inline void clear_TF(struct kernel_vm86_regs *regs) 391 { 392 regs->pt.flags &= ~X86_EFLAGS_TF; 393 } 394 395 static inline void clear_AC(struct kernel_vm86_regs *regs) 396 { 397 regs->pt.flags &= ~X86_EFLAGS_AC; 398 } 399 400 /* 401 * It is correct to call set_IF(regs) from the set_vflags_* 402 * functions. However someone forgot to call clear_IF(regs) 403 * in the opposite case. 404 * After the command sequence CLI PUSHF STI POPF you should 405 * end up with interrupts disabled, but you ended up with 406 * interrupts enabled. 407 * ( I was testing my own changes, but the only bug I 408 * could find was in a function I had not changed. ) 409 * [KD] 410 */ 411 412 static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs) 413 { 414 set_flags(VEFLAGS, flags, current->thread.vm86->veflags_mask); 415 set_flags(regs->pt.flags, flags, SAFE_MASK); 416 if (flags & X86_EFLAGS_IF) 417 set_IF(regs); 418 else 419 clear_IF(regs); 420 } 421 422 static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs) 423 { 424 set_flags(VFLAGS, flags, current->thread.vm86->veflags_mask); 425 set_flags(regs->pt.flags, flags, SAFE_MASK); 426 if (flags & X86_EFLAGS_IF) 427 set_IF(regs); 428 else 429 clear_IF(regs); 430 } 431 432 static inline unsigned long get_vflags(struct kernel_vm86_regs *regs) 433 { 434 unsigned long flags = regs->pt.flags & RETURN_MASK; 435 436 if (VEFLAGS & X86_EFLAGS_VIF) 437 flags |= X86_EFLAGS_IF; 438 flags |= X86_EFLAGS_IOPL; 439 return flags | (VEFLAGS & current->thread.vm86->veflags_mask); 440 } 441 442 static inline int is_revectored(int nr, struct revectored_struct *bitmap) 443 { 444 return test_bit(nr, bitmap->__map); 445 } 446 447 #define val_byte(val, n) (((__u8 *)&val)[n]) 448 449 #define pushb(base, ptr, val, err_label) \ 450 do { \ 451 __u8 __val = val; \ 452 ptr--; \ 453 if (put_user(__val, base + ptr) < 0) \ 454 goto err_label; \ 455 } while (0) 456 457 #define pushw(base, ptr, val, err_label) \ 458 do { \ 459 __u16 __val = val; \ 460 ptr--; \ 461 if (put_user(val_byte(__val, 1), base + ptr) < 0) \ 462 goto err_label; \ 463 ptr--; \ 464 if (put_user(val_byte(__val, 0), base + ptr) < 0) \ 465 goto err_label; \ 466 } while (0) 467 468 #define pushl(base, ptr, val, err_label) \ 469 do { \ 470 __u32 __val = val; \ 471 ptr--; \ 472 if (put_user(val_byte(__val, 3), base + ptr) < 0) \ 473 goto err_label; \ 474 ptr--; \ 475 if (put_user(val_byte(__val, 2), base + ptr) < 0) \ 476 goto err_label; \ 477 ptr--; \ 478 if (put_user(val_byte(__val, 1), base + ptr) < 0) \ 479 goto err_label; \ 480 ptr--; \ 481 if (put_user(val_byte(__val, 0), base + ptr) < 0) \ 482 goto err_label; \ 483 } while (0) 484 485 #define popb(base, ptr, err_label) \ 486 ({ \ 487 __u8 __res; \ 488 if (get_user(__res, base + ptr) < 0) \ 489 goto err_label; \ 490 ptr++; \ 491 __res; \ 492 }) 493 494 #define popw(base, ptr, err_label) \ 495 ({ \ 496 __u16 __res; \ 497 if (get_user(val_byte(__res, 0), base + ptr) < 0) \ 498 goto err_label; \ 499 ptr++; \ 500 if (get_user(val_byte(__res, 1), base + ptr) < 0) \ 501 goto err_label; \ 502 ptr++; \ 503 __res; \ 504 }) 505 506 #define popl(base, ptr, err_label) \ 507 ({ \ 508 __u32 __res; \ 509 if (get_user(val_byte(__res, 0), base + ptr) < 0) \ 510 goto err_label; \ 511 ptr++; \ 512 if (get_user(val_byte(__res, 1), base + ptr) < 0) \ 513 goto err_label; \ 514 ptr++; \ 515 if (get_user(val_byte(__res, 2), base + ptr) < 0) \ 516 goto err_label; \ 517 ptr++; \ 518 if (get_user(val_byte(__res, 3), base + ptr) < 0) \ 519 goto err_label; \ 520 ptr++; \ 521 __res; \ 522 }) 523 524 /* There are so many possible reasons for this function to return 525 * VM86_INTx, so adding another doesn't bother me. We can expect 526 * userspace programs to be able to handle it. (Getting a problem 527 * in userspace is always better than an Oops anyway.) [KD] 528 */ 529 static void do_int(struct kernel_vm86_regs *regs, int i, 530 unsigned char __user *ssp, unsigned short sp) 531 { 532 unsigned long __user *intr_ptr; 533 unsigned long segoffs; 534 struct vm86 *vm86 = current->thread.vm86; 535 536 if (regs->pt.cs == BIOSSEG) 537 goto cannot_handle; 538 if (is_revectored(i, &vm86->int_revectored)) 539 goto cannot_handle; 540 if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored)) 541 goto cannot_handle; 542 intr_ptr = (unsigned long __user *) (i << 2); 543 if (get_user(segoffs, intr_ptr)) 544 goto cannot_handle; 545 if ((segoffs >> 16) == BIOSSEG) 546 goto cannot_handle; 547 pushw(ssp, sp, get_vflags(regs), cannot_handle); 548 pushw(ssp, sp, regs->pt.cs, cannot_handle); 549 pushw(ssp, sp, IP(regs), cannot_handle); 550 regs->pt.cs = segoffs >> 16; 551 SP(regs) -= 6; 552 IP(regs) = segoffs & 0xffff; 553 clear_TF(regs); 554 clear_IF(regs); 555 clear_AC(regs); 556 return; 557 558 cannot_handle: 559 save_v86_state(regs, VM86_INTx + (i << 8)); 560 } 561 562 int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno) 563 { 564 struct vm86 *vm86 = current->thread.vm86; 565 566 if (vm86->vm86plus.is_vm86pus) { 567 if ((trapno == 3) || (trapno == 1)) { 568 save_v86_state(regs, VM86_TRAP + (trapno << 8)); 569 return 0; 570 } 571 do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs)); 572 return 0; 573 } 574 if (trapno != 1) 575 return 1; /* we let this handle by the calling routine */ 576 current->thread.trap_nr = trapno; 577 current->thread.error_code = error_code; 578 force_sig(SIGTRAP); 579 return 0; 580 } 581 582 void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code) 583 { 584 unsigned char opcode; 585 unsigned char __user *csp; 586 unsigned char __user *ssp; 587 unsigned short ip, sp, orig_flags; 588 int data32, pref_done; 589 struct vm86plus_info_struct *vmpi = ¤t->thread.vm86->vm86plus; 590 591 #define CHECK_IF_IN_TRAP \ 592 if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \ 593 newflags |= X86_EFLAGS_TF 594 595 orig_flags = *(unsigned short *)®s->pt.flags; 596 597 csp = (unsigned char __user *) (regs->pt.cs << 4); 598 ssp = (unsigned char __user *) (regs->pt.ss << 4); 599 sp = SP(regs); 600 ip = IP(regs); 601 602 data32 = 0; 603 pref_done = 0; 604 do { 605 switch (opcode = popb(csp, ip, simulate_sigsegv)) { 606 case 0x66: /* 32-bit data */ data32 = 1; break; 607 case 0x67: /* 32-bit address */ break; 608 case 0x2e: /* CS */ break; 609 case 0x3e: /* DS */ break; 610 case 0x26: /* ES */ break; 611 case 0x36: /* SS */ break; 612 case 0x65: /* GS */ break; 613 case 0x64: /* FS */ break; 614 case 0xf2: /* repnz */ break; 615 case 0xf3: /* rep */ break; 616 default: pref_done = 1; 617 } 618 } while (!pref_done); 619 620 switch (opcode) { 621 622 /* pushf */ 623 case 0x9c: 624 if (data32) { 625 pushl(ssp, sp, get_vflags(regs), simulate_sigsegv); 626 SP(regs) -= 4; 627 } else { 628 pushw(ssp, sp, get_vflags(regs), simulate_sigsegv); 629 SP(regs) -= 2; 630 } 631 IP(regs) = ip; 632 goto vm86_fault_return; 633 634 /* popf */ 635 case 0x9d: 636 { 637 unsigned long newflags; 638 if (data32) { 639 newflags = popl(ssp, sp, simulate_sigsegv); 640 SP(regs) += 4; 641 } else { 642 newflags = popw(ssp, sp, simulate_sigsegv); 643 SP(regs) += 2; 644 } 645 IP(regs) = ip; 646 CHECK_IF_IN_TRAP; 647 if (data32) 648 set_vflags_long(newflags, regs); 649 else 650 set_vflags_short(newflags, regs); 651 652 goto check_vip; 653 } 654 655 /* int xx */ 656 case 0xcd: { 657 int intno = popb(csp, ip, simulate_sigsegv); 658 IP(regs) = ip; 659 if (vmpi->vm86dbg_active) { 660 if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) { 661 save_v86_state(regs, VM86_INTx + (intno << 8)); 662 return; 663 } 664 } 665 do_int(regs, intno, ssp, sp); 666 return; 667 } 668 669 /* iret */ 670 case 0xcf: 671 { 672 unsigned long newip; 673 unsigned long newcs; 674 unsigned long newflags; 675 if (data32) { 676 newip = popl(ssp, sp, simulate_sigsegv); 677 newcs = popl(ssp, sp, simulate_sigsegv); 678 newflags = popl(ssp, sp, simulate_sigsegv); 679 SP(regs) += 12; 680 } else { 681 newip = popw(ssp, sp, simulate_sigsegv); 682 newcs = popw(ssp, sp, simulate_sigsegv); 683 newflags = popw(ssp, sp, simulate_sigsegv); 684 SP(regs) += 6; 685 } 686 IP(regs) = newip; 687 regs->pt.cs = newcs; 688 CHECK_IF_IN_TRAP; 689 if (data32) { 690 set_vflags_long(newflags, regs); 691 } else { 692 set_vflags_short(newflags, regs); 693 } 694 goto check_vip; 695 } 696 697 /* cli */ 698 case 0xfa: 699 IP(regs) = ip; 700 clear_IF(regs); 701 goto vm86_fault_return; 702 703 /* sti */ 704 /* 705 * Damn. This is incorrect: the 'sti' instruction should actually 706 * enable interrupts after the /next/ instruction. Not good. 707 * 708 * Probably needs some horsing around with the TF flag. Aiee.. 709 */ 710 case 0xfb: 711 IP(regs) = ip; 712 set_IF(regs); 713 goto check_vip; 714 715 default: 716 save_v86_state(regs, VM86_UNKNOWN); 717 } 718 719 return; 720 721 check_vip: 722 if ((VEFLAGS & (X86_EFLAGS_VIP | X86_EFLAGS_VIF)) == 723 (X86_EFLAGS_VIP | X86_EFLAGS_VIF)) { 724 save_v86_state(regs, VM86_STI); 725 return; 726 } 727 728 vm86_fault_return: 729 if (vmpi->force_return_for_pic && (VEFLAGS & (X86_EFLAGS_IF | X86_EFLAGS_VIF))) { 730 save_v86_state(regs, VM86_PICRETURN); 731 return; 732 } 733 if (orig_flags & X86_EFLAGS_TF) 734 handle_vm86_trap(regs, 0, X86_TRAP_DB); 735 return; 736 737 simulate_sigsegv: 738 /* FIXME: After a long discussion with Stas we finally 739 * agreed, that this is wrong. Here we should 740 * really send a SIGSEGV to the user program. 741 * But how do we create the correct context? We 742 * are inside a general protection fault handler 743 * and has just returned from a page fault handler. 744 * The correct context for the signal handler 745 * should be a mixture of the two, but how do we 746 * get the information? [KD] 747 */ 748 save_v86_state(regs, VM86_UNKNOWN); 749 } 750 751 /* ---------------- vm86 special IRQ passing stuff ----------------- */ 752 753 #define VM86_IRQNAME "vm86irq" 754 755 static struct vm86_irqs { 756 struct task_struct *tsk; 757 int sig; 758 } vm86_irqs[16]; 759 760 static DEFINE_SPINLOCK(irqbits_lock); 761 static int irqbits; 762 763 #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \ 764 | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \ 765 | (1 << SIGUNUSED)) 766 767 static irqreturn_t irq_handler(int intno, void *dev_id) 768 { 769 int irq_bit; 770 unsigned long flags; 771 772 spin_lock_irqsave(&irqbits_lock, flags); 773 irq_bit = 1 << intno; 774 if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk) 775 goto out; 776 irqbits |= irq_bit; 777 if (vm86_irqs[intno].sig) 778 send_sig(vm86_irqs[intno].sig, vm86_irqs[intno].tsk, 1); 779 /* 780 * IRQ will be re-enabled when user asks for the irq (whether 781 * polling or as a result of the signal) 782 */ 783 disable_irq_nosync(intno); 784 spin_unlock_irqrestore(&irqbits_lock, flags); 785 return IRQ_HANDLED; 786 787 out: 788 spin_unlock_irqrestore(&irqbits_lock, flags); 789 return IRQ_NONE; 790 } 791 792 static inline void free_vm86_irq(int irqnumber) 793 { 794 unsigned long flags; 795 796 free_irq(irqnumber, NULL); 797 vm86_irqs[irqnumber].tsk = NULL; 798 799 spin_lock_irqsave(&irqbits_lock, flags); 800 irqbits &= ~(1 << irqnumber); 801 spin_unlock_irqrestore(&irqbits_lock, flags); 802 } 803 804 void release_vm86_irqs(struct task_struct *task) 805 { 806 int i; 807 for (i = FIRST_VM86_IRQ ; i <= LAST_VM86_IRQ; i++) 808 if (vm86_irqs[i].tsk == task) 809 free_vm86_irq(i); 810 } 811 812 static inline int get_and_reset_irq(int irqnumber) 813 { 814 int bit; 815 unsigned long flags; 816 int ret = 0; 817 818 if (invalid_vm86_irq(irqnumber)) return 0; 819 if (vm86_irqs[irqnumber].tsk != current) return 0; 820 spin_lock_irqsave(&irqbits_lock, flags); 821 bit = irqbits & (1 << irqnumber); 822 irqbits &= ~bit; 823 if (bit) { 824 enable_irq(irqnumber); 825 ret = 1; 826 } 827 828 spin_unlock_irqrestore(&irqbits_lock, flags); 829 return ret; 830 } 831 832 833 static int do_vm86_irq_handling(int subfunction, int irqnumber) 834 { 835 int ret; 836 switch (subfunction) { 837 case VM86_GET_AND_RESET_IRQ: { 838 return get_and_reset_irq(irqnumber); 839 } 840 case VM86_GET_IRQ_BITS: { 841 return irqbits; 842 } 843 case VM86_REQUEST_IRQ: { 844 int sig = irqnumber >> 8; 845 int irq = irqnumber & 255; 846 if (!capable(CAP_SYS_ADMIN)) return -EPERM; 847 if (!((1 << sig) & ALLOWED_SIGS)) return -EPERM; 848 if (invalid_vm86_irq(irq)) return -EPERM; 849 if (vm86_irqs[irq].tsk) return -EPERM; 850 ret = request_irq(irq, &irq_handler, 0, VM86_IRQNAME, NULL); 851 if (ret) return ret; 852 vm86_irqs[irq].sig = sig; 853 vm86_irqs[irq].tsk = current; 854 return irq; 855 } 856 case VM86_FREE_IRQ: { 857 if (invalid_vm86_irq(irqnumber)) return -EPERM; 858 if (!vm86_irqs[irqnumber].tsk) return 0; 859 if (vm86_irqs[irqnumber].tsk != current) return -EPERM; 860 free_vm86_irq(irqnumber); 861 return 0; 862 } 863 } 864 return -EINVAL; 865 } 866 867