1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2012-2014 Andy Lutomirski <luto@amacapital.net>
4 *
5 * Based on the original implementation which is:
6 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
7 * Copyright 2003 Andi Kleen, SuSE Labs.
8 *
9 * Parts of the original code have been moved to arch/x86/vdso/vma.c
10 *
11 * This file implements vsyscall emulation. vsyscalls are a legacy ABI:
12 * Userspace can request certain kernel services by calling fixed
13 * addresses. This concept is problematic:
14 *
15 * - It interferes with ASLR.
16 * - It's awkward to write code that lives in kernel addresses but is
17 * callable by userspace at fixed addresses.
18 * - The whole concept is impossible for 32-bit compat userspace.
19 * - UML cannot easily virtualize a vsyscall.
20 *
21 * As of mid-2014, I believe that there is no new userspace code that
22 * will use a vsyscall if the vDSO is present. I hope that there will
23 * soon be no new userspace code that will ever use a vsyscall.
24 *
25 * The code in this file emulates vsyscalls when notified of a page
26 * fault to a vsyscall address.
27 */
28
29 #include <linux/kernel.h>
30 #include <linux/timer.h>
31 #include <linux/sched/signal.h>
32 #include <linux/mm_types.h>
33 #include <linux/syscalls.h>
34 #include <linux/ratelimit.h>
35
36 #include <asm/vsyscall.h>
37 #include <asm/unistd.h>
38 #include <asm/fixmap.h>
39 #include <asm/traps.h>
40 #include <asm/paravirt.h>
41
42 #define CREATE_TRACE_POINTS
43 #include "vsyscall_trace.h"
44
45 static enum { EMULATE, XONLY, NONE } vsyscall_mode __ro_after_init =
46 #ifdef CONFIG_LEGACY_VSYSCALL_NONE
47 NONE;
48 #elif defined(CONFIG_LEGACY_VSYSCALL_XONLY)
49 XONLY;
50 #else
51 #error VSYSCALL config is broken
52 #endif
53
54 static int __init vsyscall_setup(char *str)
55 {
56 if (str) {
57 if (!strcmp("emulate", str))
58 vsyscall_mode = EMULATE;
59 else if (!strcmp("xonly", str))
60 vsyscall_mode = XONLY;
61 else if (!strcmp("none", str))
62 vsyscall_mode = NONE;
63 else
64 return -EINVAL;
65
66 return 0;
67 }
68
69 return -EINVAL;
70 }
71 early_param("vsyscall", vsyscall_setup);
72
73 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
74 const char *message)
75 {
76 if (!show_unhandled_signals)
77 return;
78
79 printk_ratelimited("%s%s[%d] %s ip:%lx cs:%x sp:%lx ax:%lx si:%lx di:%lx\n",
80 level, current->comm, task_pid_nr(current),
81 message, regs->ip, regs->cs,
82 regs->sp, regs->ax, regs->si, regs->di);
83 }
84
85 static int addr_to_vsyscall_nr(unsigned long addr)
86 {
87 int nr;
88
89 if ((addr & ~0xC00UL) != VSYSCALL_ADDR)
90 return -EINVAL;
91
92 nr = (addr & 0xC00UL) >> 10;
93 if (nr >= 3)
94 return -EINVAL;
95
96 return nr;
97 }
98
99 static bool write_ok_or_segv(unsigned long ptr, size_t size)
100 {
101 if (!access_ok((void __user *)ptr, size)) {
102 struct thread_struct *thread = ¤t->thread;
103
104 thread->error_code = X86_PF_USER | X86_PF_WRITE;
105 thread->cr2 = ptr;
106 thread->trap_nr = X86_TRAP_PF;
107
108 force_sig_fault(SIGSEGV, SEGV_MAPERR, (void __user *)ptr);
109 return false;
110 } else {
111 return true;
112 }
113 }
114
115 bool emulate_vsyscall(unsigned long error_code,
116 struct pt_regs *regs, unsigned long address)
117 {
118 unsigned long caller;
119 int vsyscall_nr, syscall_nr, tmp;
120 long ret;
121 unsigned long orig_dx;
122
123 /* Write faults or kernel-privilege faults never get fixed up. */
124 if ((error_code & (X86_PF_WRITE | X86_PF_USER)) != X86_PF_USER)
125 return false;
126
127 if (!(error_code & X86_PF_INSTR)) {
128 /* Failed vsyscall read */
129 if (vsyscall_mode == EMULATE)
130 return false;
131
132 /*
133 * User code tried and failed to read the vsyscall page.
134 */
135 warn_bad_vsyscall(KERN_INFO, regs, "vsyscall read attempt denied -- look up the vsyscall kernel parameter if you need a workaround");
136 return false;
137 }
138
139 /*
140 * No point in checking CS -- the only way to get here is a user mode
141 * trap to a high address, which means that we're in 64-bit user code.
142 */
143
144 WARN_ON_ONCE(address != regs->ip);
145
146 if (vsyscall_mode == NONE) {
147 warn_bad_vsyscall(KERN_INFO, regs,
148 "vsyscall attempted with vsyscall=none");
149 return false;
150 }
151
152 vsyscall_nr = addr_to_vsyscall_nr(address);
153
154 trace_emulate_vsyscall(vsyscall_nr);
155
156 if (vsyscall_nr < 0) {
157 warn_bad_vsyscall(KERN_WARNING, regs,
158 "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
159 goto sigsegv;
160 }
161
162 if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
163 warn_bad_vsyscall(KERN_WARNING, regs,
164 "vsyscall with bad stack (exploit attempt?)");
165 goto sigsegv;
166 }
167
168 /*
169 * Check for access_ok violations and find the syscall nr.
170 *
171 * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
172 * 64-bit, so we don't need to special-case it here. For all the
173 * vsyscalls, NULL means "don't write anything" not "write it at
174 * address 0".
175 */
176 switch (vsyscall_nr) {
177 case 0:
178 if (!write_ok_or_segv(regs->di, sizeof(struct __kernel_old_timeval)) ||
179 !write_ok_or_segv(regs->si, sizeof(struct timezone))) {
180 ret = -EFAULT;
181 goto check_fault;
182 }
183
184 syscall_nr = __NR_gettimeofday;
185 break;
186
187 case 1:
188 if (!write_ok_or_segv(regs->di, sizeof(__kernel_old_time_t))) {
189 ret = -EFAULT;
190 goto check_fault;
191 }
192
193 syscall_nr = __NR_time;
194 break;
195
196 case 2:
197 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
198 !write_ok_or_segv(regs->si, sizeof(unsigned))) {
199 ret = -EFAULT;
200 goto check_fault;
201 }
202
203 syscall_nr = __NR_getcpu;
204 break;
205 }
206
207 /*
208 * Handle seccomp. regs->ip must be the original value.
209 * See seccomp_send_sigsys and Documentation/userspace-api/seccomp_filter.rst.
210 *
211 * We could optimize the seccomp disabled case, but performance
212 * here doesn't matter.
213 */
214 regs->orig_ax = syscall_nr;
215 regs->ax = -ENOSYS;
216 tmp = secure_computing();
217 if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) {
218 warn_bad_vsyscall(KERN_DEBUG, regs,
219 "seccomp tried to change syscall nr or ip");
220 force_exit_sig(SIGSYS);
221 return true;
222 }
223 regs->orig_ax = -1;
224 if (tmp)
225 goto do_ret; /* skip requested */
226
227 /*
228 * With a real vsyscall, page faults cause SIGSEGV.
229 */
230 ret = -EFAULT;
231 switch (vsyscall_nr) {
232 case 0:
233 /* this decodes regs->di and regs->si on its own */
234 ret = __x64_sys_gettimeofday(regs);
235 break;
236
237 case 1:
238 /* this decodes regs->di on its own */
239 ret = __x64_sys_time(regs);
240 break;
241
242 case 2:
243 /* while we could clobber regs->dx, we didn't in the past... */
244 orig_dx = regs->dx;
245 regs->dx = 0;
246 /* this decodes regs->di, regs->si and regs->dx on its own */
247 ret = __x64_sys_getcpu(regs);
248 regs->dx = orig_dx;
249 break;
250 }
251
252 check_fault:
253 if (ret == -EFAULT) {
254 /* Bad news -- userspace fed a bad pointer to a vsyscall. */
255 warn_bad_vsyscall(KERN_INFO, regs,
256 "vsyscall fault (exploit attempt?)");
257 goto sigsegv;
258 }
259
260 regs->ax = ret;
261
262 do_ret:
263 /* Emulate a ret instruction. */
264 regs->ip = caller;
265 regs->sp += 8;
266 return true;
267
268 sigsegv:
269 force_sig(SIGSEGV);
270 return true;
271 }
272
273 /*
274 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
275 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
276 * not need special handling anymore:
277 */
278 static const char *gate_vma_name(struct vm_area_struct *vma)
279 {
280 return "[vsyscall]";
281 }
282 static const struct vm_operations_struct gate_vma_ops = {
283 .name = gate_vma_name,
284 };
285 static struct vm_area_struct gate_vma __ro_after_init = {
286 .vm_start = VSYSCALL_ADDR,
287 .vm_end = VSYSCALL_ADDR + PAGE_SIZE,
288 .vm_page_prot = PAGE_READONLY_EXEC,
289 .vm_flags = VM_READ | VM_EXEC,
290 .vm_ops = &gate_vma_ops,
291 };
292
293 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
294 {
295 #ifdef CONFIG_COMPAT
296 if (!mm || !test_bit(MM_CONTEXT_HAS_VSYSCALL, &mm->context.flags))
297 return NULL;
298 #endif
299 if (vsyscall_mode == NONE)
300 return NULL;
301 return &gate_vma;
302 }
303
304 int in_gate_area(struct mm_struct *mm, unsigned long addr)
305 {
306 struct vm_area_struct *vma = get_gate_vma(mm);
307
308 if (!vma)
309 return 0;
310
311 return (addr >= vma->vm_start) && (addr < vma->vm_end);
312 }
313
314 /*
315 * Use this when you have no reliable mm, typically from interrupt
316 * context. It is less reliable than using a task's mm and may give
317 * false positives.
318 */
319 int in_gate_area_no_mm(unsigned long addr)
320 {
321 return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR;
322 }
323
324 /*
325 * The VSYSCALL page is the only user-accessible page in the kernel address
326 * range. Normally, the kernel page tables can have _PAGE_USER clear, but
327 * the tables covering VSYSCALL_ADDR need _PAGE_USER set if vsyscalls
328 * are enabled.
329 *
330 * Some day we may create a "minimal" vsyscall mode in which we emulate
331 * vsyscalls but leave the page not present. If so, we skip calling
332 * this.
333 */
334 void __init set_vsyscall_pgtable_user_bits(pgd_t *root)
335 {
336 pgd_t *pgd;
337 p4d_t *p4d;
338 pud_t *pud;
339 pmd_t *pmd;
340
341 pgd = pgd_offset_pgd(root, VSYSCALL_ADDR);
342 set_pgd(pgd, __pgd(pgd_val(*pgd) | _PAGE_USER));
343 p4d = p4d_offset(pgd, VSYSCALL_ADDR);
344 #if CONFIG_PGTABLE_LEVELS >= 5
345 set_p4d(p4d, __p4d(p4d_val(*p4d) | _PAGE_USER));
346 #endif
347 pud = pud_offset(p4d, VSYSCALL_ADDR);
348 set_pud(pud, __pud(pud_val(*pud) | _PAGE_USER));
349 pmd = pmd_offset(pud, VSYSCALL_ADDR);
350 set_pmd(pmd, __pmd(pmd_val(*pmd) | _PAGE_USER));
351 }
352
353 void __init map_vsyscall(void)
354 {
355 extern char __vsyscall_page;
356 unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
357
358 /*
359 * For full emulation, the page needs to exist for real. In
360 * execute-only mode, there is no PTE at all backing the vsyscall
361 * page.
362 */
363 if (vsyscall_mode == EMULATE) {
364 __set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall,
365 PAGE_KERNEL_VVAR);
366 set_vsyscall_pgtable_user_bits(swapper_pg_dir);
367 }
368
369 if (vsyscall_mode == XONLY)
370 vm_flags_init(&gate_vma, VM_EXEC);
371
372 BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) !=
373 (unsigned long)VSYSCALL_ADDR);
374 }
375
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.