Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1994 Linus Torvalds |
3 | * | |
4 | * 29 dec 2001 - Fixed oopses caused by unchecked access to the vm86 | |
624dffcb | 5 | * stack - Manfred Spraul <manfred@colorfullife.com> |
1da177e4 LT |
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 | ||
c767a54b JP |
31 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
32 | ||
a9415644 | 33 | #include <linux/capability.h> |
1da177e4 LT |
34 | #include <linux/errno.h> |
35 | #include <linux/interrupt.h> | |
5522ddb3 | 36 | #include <linux/syscalls.h> |
1da177e4 LT |
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> | |
1da177e4 LT |
43 | #include <linux/highmem.h> |
44 | #include <linux/ptrace.h> | |
7e7f8a03 | 45 | #include <linux/audit.h> |
49d26b6e | 46 | #include <linux/stddef.h> |
9fda6a06 | 47 | #include <linux/slab.h> |
76fc5e7b | 48 | #include <linux/security.h> |
1da177e4 LT |
49 | |
50 | #include <asm/uaccess.h> | |
51 | #include <asm/io.h> | |
52 | #include <asm/tlbflush.h> | |
53 | #include <asm/irq.h> | |
5ed92a8a | 54 | #include <asm/traps.h> |
ba3e127e | 55 | #include <asm/vm86.h> |
1da177e4 LT |
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 | ||
1da177e4 LT |
73 | /* |
74 | * 8- and 16-bit register defines.. | |
75 | */ | |
65ea5b03 PA |
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)) | |
1da177e4 LT |
80 | |
81 | /* | |
82 | * virtual flags (16 and 32-bit versions) | |
83 | */ | |
decd275e BG |
84 | #define VFLAGS (*(unsigned short *)&(current->thread.vm86->veflags)) |
85 | #define VEFLAGS (current->thread.vm86->veflags) | |
1da177e4 | 86 | |
83e714e8 | 87 | #define set_flags(X, new, mask) \ |
1da177e4 LT |
88 | ((X) = ((X) & ~(mask)) | ((new) & (mask))) |
89 | ||
90 | #define SAFE_MASK (0xDD5) | |
91 | #define RETURN_MASK (0xDFF) | |
92 | ||
5ed92a8a | 93 | void save_v86_state(struct kernel_vm86_regs *regs, int retval) |
1da177e4 LT |
94 | { |
95 | struct tss_struct *tss; | |
ed0b2edb BG |
96 | struct task_struct *tsk = current; |
97 | struct vm86plus_struct __user *user; | |
9fda6a06 | 98 | struct vm86 *vm86 = current->thread.vm86; |
ed0b2edb | 99 | long err = 0; |
1da177e4 LT |
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 | ||
13426356 BG |
108 | if (!vm86 || !vm86->user_vm86) { |
109 | pr_alert("no user_vm86: BAD\n"); | |
1da177e4 LT |
110 | do_exit(SIGSEGV); |
111 | } | |
decd275e | 112 | set_flags(regs->pt.flags, VEFLAGS, X86_EFLAGS_VIF | vm86->veflags_mask); |
13426356 | 113 | user = vm86->user_vm86; |
ed0b2edb | 114 | |
d4ce0f26 | 115 | if (!access_ok(VERIFY_WRITE, user, vm86->vm86plus.is_vm86pus ? |
ed0b2edb BG |
116 | sizeof(struct vm86plus_struct) : |
117 | sizeof(struct vm86_struct))) { | |
13426356 | 118 | pr_alert("could not access userspace vm86 info\n"); |
ed0b2edb BG |
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 | ||
9fda6a06 | 140 | put_user_ex(vm86->screen_bitmap, &user->screen_bitmap); |
ed0b2edb BG |
141 | } put_user_catch(err); |
142 | if (err) { | |
13426356 | 143 | pr_alert("could not access userspace vm86 info\n"); |
1da177e4 LT |
144 | do_exit(SIGSEGV); |
145 | } | |
146 | ||
24933b82 | 147 | tss = &per_cpu(cpu_tss, get_cpu()); |
9fda6a06 | 148 | tsk->thread.sp0 = vm86->saved_sp0; |
ed0b2edb BG |
149 | tsk->thread.sysenter_cs = __KERNEL_CS; |
150 | load_sp0(tss, &tsk->thread); | |
9fda6a06 | 151 | vm86->saved_sp0 = 0; |
1da177e4 LT |
152 | put_cpu(); |
153 | ||
5ed92a8a | 154 | memcpy(®s->pt, &vm86->regs32, sizeof(struct pt_regs)); |
49d26b6e | 155 | |
5ed92a8a | 156 | lazy_load_gs(vm86->regs32.gs); |
49d26b6e | 157 | |
5ed92a8a | 158 | regs->pt.ax = retval; |
1da177e4 LT |
159 | } |
160 | ||
60ec5585 | 161 | static void mark_screen_rdonly(struct mm_struct *mm) |
1da177e4 LT |
162 | { |
163 | pgd_t *pgd; | |
164 | pud_t *pud; | |
165 | pmd_t *pmd; | |
60ec5585 HD |
166 | pte_t *pte; |
167 | spinlock_t *ptl; | |
1da177e4 LT |
168 | int i; |
169 | ||
1a5a9906 | 170 | down_write(&mm->mmap_sem); |
60ec5585 | 171 | pgd = pgd_offset(mm, 0xA0000); |
1da177e4 LT |
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); | |
78ddc534 KS |
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 | } | |
1da177e4 LT |
183 | if (pmd_none_or_clear_bad(pmd)) |
184 | goto out; | |
60ec5585 | 185 | pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl); |
1da177e4 LT |
186 | for (i = 0; i < 32; i++) { |
187 | if (pte_present(*pte)) | |
188 | set_pte(pte, pte_wrprotect(*pte)); | |
189 | pte++; | |
190 | } | |
60ec5585 | 191 | pte_unmap_unlock(pte, ptl); |
1da177e4 | 192 | out: |
1a5a9906 | 193 | up_write(&mm->mmap_sem); |
1da177e4 LT |
194 | flush_tlb(); |
195 | } | |
196 | ||
197 | ||
198 | ||
199 | static int do_vm86_irq_handling(int subfunction, int irqnumber); | |
13426356 | 200 | static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus); |
1da177e4 | 201 | |
13426356 | 202 | SYSCALL_DEFINE1(vm86old, struct vm86_struct __user *, user_vm86) |
1da177e4 | 203 | { |
13426356 | 204 | return do_sys_vm86((struct vm86plus_struct __user *) user_vm86, false); |
1da177e4 LT |
205 | } |
206 | ||
207 | ||
5522ddb3 | 208 | SYSCALL_DEFINE2(vm86, unsigned long, cmd, unsigned long, arg) |
1da177e4 | 209 | { |
f1382f15 | 210 | switch (cmd) { |
83e714e8 PC |
211 | case VM86_REQUEST_IRQ: |
212 | case VM86_FREE_IRQ: | |
213 | case VM86_GET_IRQ_BITS: | |
214 | case VM86_GET_AND_RESET_IRQ: | |
5522ddb3 | 215 | return do_vm86_irq_handling(cmd, (int)arg); |
83e714e8 PC |
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 | */ | |
5522ddb3 | 223 | return 0; |
1da177e4 LT |
224 | } |
225 | ||
226 | /* we come here only for functions VM86_ENTER, VM86_ENTER_NO_BYPASS */ | |
90c6085a | 227 | return do_sys_vm86((struct vm86plus_struct __user *) arg, true); |
1da177e4 LT |
228 | } |
229 | ||
230 | ||
13426356 | 231 | static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus) |
1da177e4 LT |
232 | { |
233 | struct tss_struct *tss; | |
ed0b2edb | 234 | struct task_struct *tsk = current; |
9fda6a06 | 235 | struct vm86 *vm86 = tsk->thread.vm86; |
90c6085a | 236 | struct kernel_vm86_regs vm86regs; |
5ed92a8a | 237 | struct pt_regs *regs = current_pt_regs(); |
ed0b2edb BG |
238 | unsigned long err = 0; |
239 | ||
76fc5e7b AL |
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 | ||
9fda6a06 BG |
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) | |
ed0b2edb BG |
272 | return -EPERM; |
273 | ||
13426356 | 274 | if (!access_ok(VERIFY_READ, user_vm86, plus ? |
ed0b2edb BG |
275 | sizeof(struct vm86_struct) : |
276 | sizeof(struct vm86plus_struct))) | |
277 | return -EFAULT; | |
278 | ||
90c6085a | 279 | memset(&vm86regs, 0, sizeof(vm86regs)); |
ed0b2edb BG |
280 | get_user_try { |
281 | unsigned short seg; | |
13426356 BG |
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); | |
90c6085a | 291 | vm86regs.pt.cs = seg; |
13426356 BG |
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); | |
90c6085a | 295 | vm86regs.pt.ss = seg; |
13426356 BG |
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); | |
ed0b2edb BG |
304 | } get_user_catch(err); |
305 | if (err) | |
306 | return err; | |
307 | ||
13426356 BG |
308 | if (copy_from_user(&vm86->int_revectored, |
309 | &user_vm86->int_revectored, | |
ed0b2edb BG |
310 | sizeof(struct revectored_struct))) |
311 | return -EFAULT; | |
13426356 BG |
312 | if (copy_from_user(&vm86->int21_revectored, |
313 | &user_vm86->int21_revectored, | |
ed0b2edb BG |
314 | sizeof(struct revectored_struct))) |
315 | return -EFAULT; | |
316 | if (plus) { | |
13426356 | 317 | if (copy_from_user(&vm86->vm86plus, &user_vm86->vm86plus, |
ed0b2edb BG |
318 | sizeof(struct vm86plus_info_struct))) |
319 | return -EFAULT; | |
d4ce0f26 BG |
320 | vm86->vm86plus.is_vm86pus = 1; |
321 | } else | |
322 | memset(&vm86->vm86plus, 0, | |
323 | sizeof(struct vm86plus_info_struct)); | |
5ed92a8a BG |
324 | |
325 | memcpy(&vm86->regs32, regs, sizeof(struct pt_regs)); | |
13426356 | 326 | vm86->user_vm86 = user_vm86; |
1da177e4 LT |
327 | |
328 | /* | |
65ea5b03 | 329 | * The flags register is also special: we cannot trust that the user |
1da177e4 LT |
330 | * has set it up safely, so this makes sure interrupt etc flags are |
331 | * inherited from protected mode. | |
332 | */ | |
90c6085a BG |
333 | VEFLAGS = vm86regs.pt.flags; |
334 | vm86regs.pt.flags &= SAFE_MASK; | |
5ed92a8a | 335 | vm86regs.pt.flags |= regs->flags & ~SAFE_MASK; |
90c6085a | 336 | vm86regs.pt.flags |= X86_VM_MASK; |
1da177e4 | 337 | |
5ed92a8a | 338 | vm86regs.pt.orig_ax = regs->orig_ax; |
df1ae9a5 | 339 | |
d4ce0f26 | 340 | switch (vm86->cpu_type) { |
83e714e8 | 341 | case CPU_286: |
decd275e | 342 | vm86->veflags_mask = 0; |
83e714e8 PC |
343 | break; |
344 | case CPU_386: | |
decd275e | 345 | vm86->veflags_mask = X86_EFLAGS_NT | X86_EFLAGS_IOPL; |
83e714e8 PC |
346 | break; |
347 | case CPU_486: | |
decd275e | 348 | vm86->veflags_mask = X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL; |
83e714e8 PC |
349 | break; |
350 | default: | |
decd275e | 351 | vm86->veflags_mask = X86_EFLAGS_ID | X86_EFLAGS_AC | X86_EFLAGS_NT | X86_EFLAGS_IOPL; |
83e714e8 | 352 | break; |
1da177e4 LT |
353 | } |
354 | ||
355 | /* | |
5ed92a8a | 356 | * Save old state |
1da177e4 | 357 | */ |
9fda6a06 | 358 | vm86->saved_sp0 = tsk->thread.sp0; |
5ed92a8a | 359 | lazy_save_gs(vm86->regs32.gs); |
1da177e4 | 360 | |
24933b82 | 361 | tss = &per_cpu(cpu_tss, get_cpu()); |
5ed92a8a BG |
362 | /* make room for real-mode segments */ |
363 | tsk->thread.sp0 += 16; | |
362f924b BP |
364 | |
365 | if (static_cpu_has_safe(X86_FEATURE_SEP)) | |
1da177e4 | 366 | tsk->thread.sysenter_cs = 0; |
362f924b | 367 | |
faca6227 | 368 | load_sp0(tss, &tsk->thread); |
1da177e4 LT |
369 | put_cpu(); |
370 | ||
d4ce0f26 | 371 | if (vm86->flags & VM86_SCREEN_BITMAP) |
60ec5585 | 372 | mark_screen_rdonly(tsk->mm); |
7e7f8a03 | 373 | |
5ed92a8a BG |
374 | memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs)); |
375 | force_iret(); | |
376 | return regs->ax; | |
1da177e4 LT |
377 | } |
378 | ||
83e714e8 | 379 | static inline void set_IF(struct kernel_vm86_regs *regs) |
1da177e4 | 380 | { |
a5c15d41 | 381 | VEFLAGS |= X86_EFLAGS_VIF; |
1da177e4 LT |
382 | } |
383 | ||
83e714e8 | 384 | static inline void clear_IF(struct kernel_vm86_regs *regs) |
1da177e4 | 385 | { |
a5c15d41 | 386 | VEFLAGS &= ~X86_EFLAGS_VIF; |
1da177e4 LT |
387 | } |
388 | ||
83e714e8 | 389 | static inline void clear_TF(struct kernel_vm86_regs *regs) |
1da177e4 | 390 | { |
a5c15d41 | 391 | regs->pt.flags &= ~X86_EFLAGS_TF; |
1da177e4 LT |
392 | } |
393 | ||
83e714e8 | 394 | static inline void clear_AC(struct kernel_vm86_regs *regs) |
1da177e4 | 395 | { |
a5c15d41 | 396 | regs->pt.flags &= ~X86_EFLAGS_AC; |
1da177e4 LT |
397 | } |
398 | ||
83e714e8 PC |
399 | /* |
400 | * It is correct to call set_IF(regs) from the set_vflags_* | |
1da177e4 LT |
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 | |
ab4a574e | 404 | * end up with interrupts disabled, but you ended up with |
1da177e4 LT |
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 | ||
83e714e8 | 411 | static inline void set_vflags_long(unsigned long flags, struct kernel_vm86_regs *regs) |
1da177e4 | 412 | { |
decd275e | 413 | set_flags(VEFLAGS, flags, current->thread.vm86->veflags_mask); |
65ea5b03 | 414 | set_flags(regs->pt.flags, flags, SAFE_MASK); |
a5c15d41 | 415 | if (flags & X86_EFLAGS_IF) |
1da177e4 LT |
416 | set_IF(regs); |
417 | else | |
418 | clear_IF(regs); | |
419 | } | |
420 | ||
83e714e8 | 421 | static inline void set_vflags_short(unsigned short flags, struct kernel_vm86_regs *regs) |
1da177e4 | 422 | { |
decd275e | 423 | set_flags(VFLAGS, flags, current->thread.vm86->veflags_mask); |
65ea5b03 | 424 | set_flags(regs->pt.flags, flags, SAFE_MASK); |
a5c15d41 | 425 | if (flags & X86_EFLAGS_IF) |
1da177e4 LT |
426 | set_IF(regs); |
427 | else | |
428 | clear_IF(regs); | |
429 | } | |
430 | ||
83e714e8 | 431 | static inline unsigned long get_vflags(struct kernel_vm86_regs *regs) |
1da177e4 | 432 | { |
65ea5b03 | 433 | unsigned long flags = regs->pt.flags & RETURN_MASK; |
1da177e4 | 434 | |
a5c15d41 | 435 | if (VEFLAGS & X86_EFLAGS_VIF) |
436 | flags |= X86_EFLAGS_IF; | |
437 | flags |= X86_EFLAGS_IOPL; | |
decd275e | 438 | return flags | (VEFLAGS & current->thread.vm86->veflags_mask); |
1da177e4 LT |
439 | } |
440 | ||
83e714e8 | 441 | static inline int is_revectored(int nr, struct revectored_struct *bitmap) |
1da177e4 LT |
442 | { |
443 | __asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0" | |
444 | :"=r" (nr) | |
83e714e8 | 445 | :"m" (*bitmap), "r" (nr)); |
1da177e4 LT |
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; \ | |
83e714e8 | 457 | } while (0) |
1da177e4 LT |
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; \ | |
83e714e8 | 468 | } while (0) |
1da177e4 LT |
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; \ | |
83e714e8 | 485 | } while (0) |
1da177e4 LT |
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 | */ | |
531 | static void do_int(struct kernel_vm86_regs *regs, int i, | |
83e714e8 | 532 | unsigned char __user *ssp, unsigned short sp) |
1da177e4 LT |
533 | { |
534 | unsigned long __user *intr_ptr; | |
535 | unsigned long segoffs; | |
90c6085a | 536 | struct vm86 *vm86 = current->thread.vm86; |
1da177e4 | 537 | |
65ea5b03 | 538 | if (regs->pt.cs == BIOSSEG) |
1da177e4 | 539 | goto cannot_handle; |
d4ce0f26 | 540 | if (is_revectored(i, &vm86->int_revectored)) |
1da177e4 | 541 | goto cannot_handle; |
d4ce0f26 | 542 | if (i == 0x21 && is_revectored(AH(regs), &vm86->int21_revectored)) |
1da177e4 LT |
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); | |
65ea5b03 | 550 | pushw(ssp, sp, regs->pt.cs, cannot_handle); |
1da177e4 | 551 | pushw(ssp, sp, IP(regs), cannot_handle); |
65ea5b03 | 552 | regs->pt.cs = segoffs >> 16; |
1da177e4 LT |
553 | SP(regs) -= 6; |
554 | IP(regs) = segoffs & 0xffff; | |
555 | clear_TF(regs); | |
556 | clear_IF(regs); | |
557 | clear_AC(regs); | |
558 | return; | |
559 | ||
560 | cannot_handle: | |
5ed92a8a | 561 | save_v86_state(regs, VM86_INTx + (i << 8)); |
1da177e4 LT |
562 | } |
563 | ||
83e714e8 | 564 | int handle_vm86_trap(struct kernel_vm86_regs *regs, long error_code, int trapno) |
1da177e4 | 565 | { |
90c6085a BG |
566 | struct vm86 *vm86 = current->thread.vm86; |
567 | ||
568 | if (vm86->vm86plus.is_vm86pus) { | |
6554287b | 569 | if ((trapno == 3) || (trapno == 1)) { |
5ed92a8a | 570 | save_v86_state(regs, VM86_TRAP + (trapno << 8)); |
6554287b BO |
571 | return 0; |
572 | } | |
65ea5b03 | 573 | do_int(regs, trapno, (unsigned char __user *) (regs->pt.ss << 4), SP(regs)); |
1da177e4 LT |
574 | return 0; |
575 | } | |
83e714e8 | 576 | if (trapno != 1) |
1da177e4 | 577 | return 1; /* we let this handle by the calling routine */ |
51e7dc70 | 578 | current->thread.trap_nr = trapno; |
1da177e4 | 579 | current->thread.error_code = error_code; |
0f540910 | 580 | force_sig(SIGTRAP, current); |
1da177e4 LT |
581 | return 0; |
582 | } | |
583 | ||
83e714e8 | 584 | void handle_vm86_fault(struct kernel_vm86_regs *regs, long error_code) |
1da177e4 LT |
585 | { |
586 | unsigned char opcode; | |
587 | unsigned char __user *csp; | |
588 | unsigned char __user *ssp; | |
5fd75ebb | 589 | unsigned short ip, sp, orig_flags; |
1da177e4 | 590 | int data32, pref_done; |
d4ce0f26 | 591 | struct vm86plus_info_struct *vmpi = ¤t->thread.vm86->vm86plus; |
1da177e4 LT |
592 | |
593 | #define CHECK_IF_IN_TRAP \ | |
d4ce0f26 | 594 | if (vmpi->vm86dbg_active && vmpi->vm86dbg_TFpendig) \ |
a5c15d41 | 595 | newflags |= X86_EFLAGS_TF |
1da177e4 | 596 | |
65ea5b03 | 597 | orig_flags = *(unsigned short *)®s->pt.flags; |
5fd75ebb | 598 | |
65ea5b03 PA |
599 | csp = (unsigned char __user *) (regs->pt.cs << 4); |
600 | ssp = (unsigned char __user *) (regs->pt.ss << 4); | |
1da177e4 LT |
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)) { | |
83e714e8 PC |
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; | |
1da177e4 LT |
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; | |
5ed92a8a | 634 | goto vm86_fault_return; |
1da177e4 LT |
635 | |
636 | /* popf */ | |
637 | case 0x9d: | |
638 | { | |
639 | unsigned long newflags; | |
640 | if (data32) { | |
83e714e8 | 641 | newflags = popl(ssp, sp, simulate_sigsegv); |
1da177e4 LT |
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; | |
83e714e8 | 649 | if (data32) |
1da177e4 | 650 | set_vflags_long(newflags, regs); |
83e714e8 | 651 | else |
1da177e4 | 652 | set_vflags_short(newflags, regs); |
83e714e8 | 653 | |
5ed92a8a | 654 | goto check_vip; |
1da177e4 LT |
655 | } |
656 | ||
657 | /* int xx */ | |
658 | case 0xcd: { | |
83e714e8 | 659 | int intno = popb(csp, ip, simulate_sigsegv); |
1da177e4 | 660 | IP(regs) = ip; |
d4ce0f26 | 661 | if (vmpi->vm86dbg_active) { |
5ed92a8a BG |
662 | if ((1 << (intno & 7)) & vmpi->vm86dbg_intxxtab[intno >> 3]) { |
663 | save_v86_state(regs, VM86_INTx + (intno << 8)); | |
664 | return; | |
665 | } | |
1da177e4 LT |
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) { | |
83e714e8 PC |
678 | newip = popl(ssp, sp, simulate_sigsegv); |
679 | newcs = popl(ssp, sp, simulate_sigsegv); | |
680 | newflags = popl(ssp, sp, simulate_sigsegv); | |
1da177e4 LT |
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; | |
65ea5b03 | 689 | regs->pt.cs = newcs; |
1da177e4 LT |
690 | CHECK_IF_IN_TRAP; |
691 | if (data32) { | |
692 | set_vflags_long(newflags, regs); | |
693 | } else { | |
694 | set_vflags_short(newflags, regs); | |
695 | } | |
5ed92a8a | 696 | goto check_vip; |
1da177e4 LT |
697 | } |
698 | ||
699 | /* cli */ | |
700 | case 0xfa: | |
701 | IP(regs) = ip; | |
702 | clear_IF(regs); | |
5ed92a8a | 703 | goto vm86_fault_return; |
1da177e4 LT |
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); | |
5ed92a8a | 715 | goto check_vip; |
1da177e4 LT |
716 | |
717 | default: | |
5ed92a8a | 718 | save_v86_state(regs, VM86_UNKNOWN); |
1da177e4 LT |
719 | } |
720 | ||
721 | return; | |
722 | ||
5ed92a8a BG |
723 | check_vip: |
724 | if (VEFLAGS & X86_EFLAGS_VIP) { | |
725 | save_v86_state(regs, VM86_STI); | |
726 | return; | |
727 | } | |
728 | ||
729 | vm86_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 | ||
1da177e4 LT |
738 | simulate_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 | */ | |
5ed92a8a | 749 | save_v86_state(regs, VM86_UNKNOWN); |
1da177e4 LT |
750 | } |
751 | ||
752 | /* ---------------- vm86 special IRQ passing stuff ----------------- */ | |
753 | ||
754 | #define VM86_IRQNAME "vm86irq" | |
755 | ||
756 | static struct vm86_irqs { | |
757 | struct task_struct *tsk; | |
758 | int sig; | |
759 | } vm86_irqs[16]; | |
760 | ||
761 | static DEFINE_SPINLOCK(irqbits_lock); | |
762 | static int irqbits; | |
763 | ||
83e714e8 | 764 | #define ALLOWED_SIGS (1 /* 0 = don't send a signal */ \ |
1da177e4 | 765 | | (1 << SIGUSR1) | (1 << SIGUSR2) | (1 << SIGIO) | (1 << SIGURG) \ |
83e714e8 PC |
766 | | (1 << SIGUNUSED)) |
767 | ||
7d12e780 | 768 | static irqreturn_t irq_handler(int intno, void *dev_id) |
1da177e4 LT |
769 | { |
770 | int irq_bit; | |
771 | unsigned long flags; | |
772 | ||
83e714e8 | 773 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 | 774 | irq_bit = 1 << intno; |
83e714e8 | 775 | if ((irqbits & irq_bit) || !vm86_irqs[intno].tsk) |
1da177e4 LT |
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); | |
1da177e4 LT |
780 | /* |
781 | * IRQ will be re-enabled when user asks for the irq (whether | |
782 | * polling or as a result of the signal) | |
783 | */ | |
ad671423 PP |
784 | disable_irq_nosync(intno); |
785 | spin_unlock_irqrestore(&irqbits_lock, flags); | |
1da177e4 LT |
786 | return IRQ_HANDLED; |
787 | ||
788 | out: | |
83e714e8 | 789 | spin_unlock_irqrestore(&irqbits_lock, flags); |
1da177e4 LT |
790 | return IRQ_NONE; |
791 | } | |
792 | ||
793 | static 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 | ||
83e714e8 | 800 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 | 801 | irqbits &= ~(1 << irqnumber); |
83e714e8 | 802 | spin_unlock_irqrestore(&irqbits_lock, flags); |
1da177e4 LT |
803 | } |
804 | ||
805 | void 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 | ||
813 | static inline int get_and_reset_irq(int irqnumber) | |
814 | { | |
815 | int bit; | |
816 | unsigned long flags; | |
ad671423 | 817 | int ret = 0; |
83e714e8 | 818 | |
1da177e4 LT |
819 | if (invalid_vm86_irq(irqnumber)) return 0; |
820 | if (vm86_irqs[irqnumber].tsk != current) return 0; | |
83e714e8 | 821 | spin_lock_irqsave(&irqbits_lock, flags); |
1da177e4 LT |
822 | bit = irqbits & (1 << irqnumber); |
823 | irqbits &= ~bit; | |
ad671423 PP |
824 | if (bit) { |
825 | enable_irq(irqnumber); | |
826 | ret = 1; | |
827 | } | |
828 | ||
83e714e8 | 829 | spin_unlock_irqrestore(&irqbits_lock, flags); |
ad671423 | 830 | return ret; |
1da177e4 LT |
831 | } |
832 | ||
833 | ||
834 | static 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 |