Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
95c46b56 AL |
2 | * Copyright (c) 2012-2014 Andy Lutomirski <luto@amacapital.net> |
3 | * | |
4 | * Based on the original implementation which is: | |
1da177e4 LT |
5 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE |
6 | * Copyright 2003 Andi Kleen, SuSE Labs. | |
7 | * | |
95c46b56 AL |
8 | * Parts of the original code have been moved to arch/x86/vdso/vma.c |
9 | * | |
10 | * This file implements vsyscall emulation. vsyscalls are a legacy ABI: | |
11 | * Userspace can request certain kernel services by calling fixed | |
12 | * addresses. This concept is problematic: | |
5cec93c2 | 13 | * |
95c46b56 AL |
14 | * - It interferes with ASLR. |
15 | * - It's awkward to write code that lives in kernel addresses but is | |
16 | * callable by userspace at fixed addresses. | |
17 | * - The whole concept is impossible for 32-bit compat userspace. | |
18 | * - UML cannot easily virtualize a vsyscall. | |
1da177e4 | 19 | * |
95c46b56 AL |
20 | * As of mid-2014, I believe that there is no new userspace code that |
21 | * will use a vsyscall if the vDSO is present. I hope that there will | |
22 | * soon be no new userspace code that will ever use a vsyscall. | |
1da177e4 | 23 | * |
95c46b56 AL |
24 | * The code in this file emulates vsyscalls when notified of a page |
25 | * fault to a vsyscall address. | |
1da177e4 LT |
26 | */ |
27 | ||
1da177e4 LT |
28 | #include <linux/kernel.h> |
29 | #include <linux/timer.h> | |
3f07c014 | 30 | #include <linux/sched/signal.h> |
5cec93c2 AL |
31 | #include <linux/syscalls.h> |
32 | #include <linux/ratelimit.h> | |
1da177e4 LT |
33 | |
34 | #include <asm/vsyscall.h> | |
7460ed28 | 35 | #include <asm/unistd.h> |
1da177e4 | 36 | #include <asm/fixmap.h> |
5cec93c2 | 37 | #include <asm/traps.h> |
1da177e4 | 38 | |
c149a665 AL |
39 | #define CREATE_TRACE_POINTS |
40 | #include "vsyscall_trace.h" | |
41 | ||
3dc33bd3 | 42 | static enum { EMULATE, NATIVE, NONE } vsyscall_mode = |
93f13a9f | 43 | #if defined(CONFIG_LEGACY_VSYSCALL_NATIVE) |
3dc33bd3 | 44 | NATIVE; |
93f13a9f | 45 | #elif defined(CONFIG_LEGACY_VSYSCALL_NONE) |
3dc33bd3 KC |
46 | NONE; |
47 | #else | |
48 | EMULATE; | |
49 | #endif | |
3ae36655 AL |
50 | |
51 | static int __init vsyscall_setup(char *str) | |
52 | { | |
53 | if (str) { | |
54 | if (!strcmp("emulate", str)) | |
55 | vsyscall_mode = EMULATE; | |
56 | else if (!strcmp("native", str)) | |
57 | vsyscall_mode = NATIVE; | |
58 | else if (!strcmp("none", str)) | |
59 | vsyscall_mode = NONE; | |
60 | else | |
61 | return -EINVAL; | |
62 | ||
63 | return 0; | |
64 | } | |
65 | ||
66 | return -EINVAL; | |
67 | } | |
68 | early_param("vsyscall", vsyscall_setup); | |
69 | ||
5cec93c2 AL |
70 | static void warn_bad_vsyscall(const char *level, struct pt_regs *regs, |
71 | const char *message) | |
1da177e4 | 72 | { |
c767a54b | 73 | if (!show_unhandled_signals) |
5cec93c2 | 74 | return; |
1da177e4 | 75 | |
53b884ac AL |
76 | printk_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n", |
77 | level, current->comm, task_pid_nr(current), | |
78 | message, regs->ip, regs->cs, | |
79 | regs->sp, regs->ax, regs->si, regs->di); | |
c9712944 AL |
80 | } |
81 | ||
82 | static int addr_to_vsyscall_nr(unsigned long addr) | |
83 | { | |
84 | int nr; | |
85 | ||
f40c3300 | 86 | if ((addr & ~0xC00UL) != VSYSCALL_ADDR) |
c9712944 AL |
87 | return -EINVAL; |
88 | ||
89 | nr = (addr & 0xC00UL) >> 10; | |
90 | if (nr >= 3) | |
91 | return -EINVAL; | |
92 | ||
93 | return nr; | |
1da177e4 LT |
94 | } |
95 | ||
4fc34901 AL |
96 | static bool write_ok_or_segv(unsigned long ptr, size_t size) |
97 | { | |
98 | /* | |
99 | * XXX: if access_ok, get_user, and put_user handled | |
2a53ccbc | 100 | * sig_on_uaccess_err, this could go away. |
4fc34901 AL |
101 | */ |
102 | ||
103 | if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) { | |
104 | siginfo_t info; | |
105 | struct thread_struct *thread = ¤t->thread; | |
106 | ||
107 | thread->error_code = 6; /* user fault, no page, write */ | |
108 | thread->cr2 = ptr; | |
51e7dc70 | 109 | thread->trap_nr = X86_TRAP_PF; |
4fc34901 AL |
110 | |
111 | memset(&info, 0, sizeof(info)); | |
112 | info.si_signo = SIGSEGV; | |
113 | info.si_errno = 0; | |
114 | info.si_code = SEGV_MAPERR; | |
115 | info.si_addr = (void __user *)ptr; | |
116 | ||
117 | force_sig_info(SIGSEGV, &info, current); | |
118 | return false; | |
119 | } else { | |
120 | return true; | |
121 | } | |
122 | } | |
123 | ||
3ae36655 | 124 | bool emulate_vsyscall(struct pt_regs *regs, unsigned long address) |
1da177e4 | 125 | { |
5cec93c2 AL |
126 | struct task_struct *tsk; |
127 | unsigned long caller; | |
87b526d3 | 128 | int vsyscall_nr, syscall_nr, tmp; |
2a53ccbc | 129 | int prev_sig_on_uaccess_err; |
5cec93c2 AL |
130 | long ret; |
131 | ||
3ae36655 AL |
132 | /* |
133 | * No point in checking CS -- the only way to get here is a user mode | |
134 | * trap to a high address, which means that we're in 64-bit user code. | |
135 | */ | |
5cec93c2 | 136 | |
3ae36655 | 137 | WARN_ON_ONCE(address != regs->ip); |
c9712944 | 138 | |
3ae36655 AL |
139 | if (vsyscall_mode == NONE) { |
140 | warn_bad_vsyscall(KERN_INFO, regs, | |
141 | "vsyscall attempted with vsyscall=none"); | |
142 | return false; | |
c9712944 AL |
143 | } |
144 | ||
3ae36655 | 145 | vsyscall_nr = addr_to_vsyscall_nr(address); |
c149a665 AL |
146 | |
147 | trace_emulate_vsyscall(vsyscall_nr); | |
148 | ||
c9712944 AL |
149 | if (vsyscall_nr < 0) { |
150 | warn_bad_vsyscall(KERN_WARNING, regs, | |
3ae36655 | 151 | "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround"); |
5cec93c2 AL |
152 | goto sigsegv; |
153 | } | |
d0aff6e6 | 154 | |
5cec93c2 | 155 | if (get_user(caller, (unsigned long __user *)regs->sp) != 0) { |
3ae36655 AL |
156 | warn_bad_vsyscall(KERN_WARNING, regs, |
157 | "vsyscall with bad stack (exploit attempt?)"); | |
5cec93c2 AL |
158 | goto sigsegv; |
159 | } | |
8c73626a | 160 | |
5cec93c2 | 161 | tsk = current; |
4fc34901 AL |
162 | |
163 | /* | |
87b526d3 AL |
164 | * Check for access_ok violations and find the syscall nr. |
165 | * | |
46ed99d1 | 166 | * NULL is a valid user pointer (in the access_ok sense) on 32-bit and |
4fc34901 | 167 | * 64-bit, so we don't need to special-case it here. For all the |
46ed99d1 | 168 | * vsyscalls, NULL means "don't write anything" not "write it at |
4fc34901 AL |
169 | * address 0". |
170 | */ | |
5cec93c2 AL |
171 | switch (vsyscall_nr) { |
172 | case 0: | |
4fc34901 | 173 | if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) || |
87b526d3 AL |
174 | !write_ok_or_segv(regs->si, sizeof(struct timezone))) { |
175 | ret = -EFAULT; | |
176 | goto check_fault; | |
177 | } | |
4fc34901 | 178 | |
87b526d3 AL |
179 | syscall_nr = __NR_gettimeofday; |
180 | break; | |
181 | ||
182 | case 1: | |
183 | if (!write_ok_or_segv(regs->di, sizeof(time_t))) { | |
184 | ret = -EFAULT; | |
185 | goto check_fault; | |
186 | } | |
187 | ||
188 | syscall_nr = __NR_time; | |
189 | break; | |
190 | ||
191 | case 2: | |
192 | if (!write_ok_or_segv(regs->di, sizeof(unsigned)) || | |
193 | !write_ok_or_segv(regs->si, sizeof(unsigned))) { | |
194 | ret = -EFAULT; | |
195 | goto check_fault; | |
196 | } | |
197 | ||
198 | syscall_nr = __NR_getcpu; | |
199 | break; | |
200 | } | |
201 | ||
202 | /* | |
203 | * Handle seccomp. regs->ip must be the original value. | |
204 | * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt. | |
205 | * | |
206 | * We could optimize the seccomp disabled case, but performance | |
207 | * here doesn't matter. | |
208 | */ | |
209 | regs->orig_ax = syscall_nr; | |
210 | regs->ax = -ENOSYS; | |
2f275de5 | 211 | tmp = secure_computing(NULL); |
87b526d3 AL |
212 | if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) { |
213 | warn_bad_vsyscall(KERN_DEBUG, regs, | |
214 | "seccomp tried to change syscall nr or ip"); | |
215 | do_exit(SIGSYS); | |
216 | } | |
26893107 | 217 | regs->orig_ax = -1; |
87b526d3 AL |
218 | if (tmp) |
219 | goto do_ret; /* skip requested */ | |
220 | ||
221 | /* | |
222 | * With a real vsyscall, page faults cause SIGSEGV. We want to | |
223 | * preserve that behavior to make writing exploits harder. | |
224 | */ | |
2a53ccbc IM |
225 | prev_sig_on_uaccess_err = current->thread.sig_on_uaccess_err; |
226 | current->thread.sig_on_uaccess_err = 1; | |
87b526d3 AL |
227 | |
228 | ret = -EFAULT; | |
229 | switch (vsyscall_nr) { | |
230 | case 0: | |
5cec93c2 AL |
231 | ret = sys_gettimeofday( |
232 | (struct timeval __user *)regs->di, | |
233 | (struct timezone __user *)regs->si); | |
234 | break; | |
235 | ||
236 | case 1: | |
5cec93c2 AL |
237 | ret = sys_time((time_t __user *)regs->di); |
238 | break; | |
239 | ||
240 | case 2: | |
5cec93c2 AL |
241 | ret = sys_getcpu((unsigned __user *)regs->di, |
242 | (unsigned __user *)regs->si, | |
46ed99d1 | 243 | NULL); |
5cec93c2 | 244 | break; |
5cec93c2 | 245 | } |
8c73626a | 246 | |
2a53ccbc | 247 | current->thread.sig_on_uaccess_err = prev_sig_on_uaccess_err; |
4fc34901 | 248 | |
87b526d3 | 249 | check_fault: |
5cec93c2 | 250 | if (ret == -EFAULT) { |
4fc34901 | 251 | /* Bad news -- userspace fed a bad pointer to a vsyscall. */ |
5cec93c2 AL |
252 | warn_bad_vsyscall(KERN_INFO, regs, |
253 | "vsyscall fault (exploit attempt?)"); | |
4fc34901 AL |
254 | |
255 | /* | |
256 | * If we failed to generate a signal for any reason, | |
257 | * generate one here. (This should be impossible.) | |
258 | */ | |
259 | if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) && | |
260 | !sigismember(&tsk->pending.signal, SIGSEGV))) | |
261 | goto sigsegv; | |
262 | ||
263 | return true; /* Don't emulate the ret. */ | |
5cec93c2 | 264 | } |
8c73626a | 265 | |
5cec93c2 | 266 | regs->ax = ret; |
1da177e4 | 267 | |
5651721e | 268 | do_ret: |
5cec93c2 AL |
269 | /* Emulate a ret instruction. */ |
270 | regs->ip = caller; | |
271 | regs->sp += 8; | |
3ae36655 | 272 | return true; |
5cec93c2 AL |
273 | |
274 | sigsegv: | |
5cec93c2 | 275 | force_sig(SIGSEGV, current); |
3ae36655 | 276 | return true; |
1da177e4 LT |
277 | } |
278 | ||
b9359090 AL |
279 | /* |
280 | * A pseudo VMA to allow ptrace access for the vsyscall page. This only | |
281 | * covers the 64bit vsyscall page now. 32bit has a real VMA now and does | |
282 | * not need special handling anymore: | |
283 | */ | |
284 | static const char *gate_vma_name(struct vm_area_struct *vma) | |
285 | { | |
286 | return "[vsyscall]"; | |
287 | } | |
7cbea8dc | 288 | static const struct vm_operations_struct gate_vma_ops = { |
b9359090 AL |
289 | .name = gate_vma_name, |
290 | }; | |
291 | static struct vm_area_struct gate_vma = { | |
292 | .vm_start = VSYSCALL_ADDR, | |
293 | .vm_end = VSYSCALL_ADDR + PAGE_SIZE, | |
294 | .vm_page_prot = PAGE_READONLY_EXEC, | |
295 | .vm_flags = VM_READ | VM_EXEC, | |
296 | .vm_ops = &gate_vma_ops, | |
297 | }; | |
298 | ||
299 | struct vm_area_struct *get_gate_vma(struct mm_struct *mm) | |
300 | { | |
c338867d | 301 | #ifdef CONFIG_COMPAT |
b9359090 AL |
302 | if (!mm || mm->context.ia32_compat) |
303 | return NULL; | |
304 | #endif | |
87983c66 AL |
305 | if (vsyscall_mode == NONE) |
306 | return NULL; | |
b9359090 AL |
307 | return &gate_vma; |
308 | } | |
309 | ||
310 | int in_gate_area(struct mm_struct *mm, unsigned long addr) | |
311 | { | |
312 | struct vm_area_struct *vma = get_gate_vma(mm); | |
313 | ||
314 | if (!vma) | |
315 | return 0; | |
316 | ||
317 | return (addr >= vma->vm_start) && (addr < vma->vm_end); | |
318 | } | |
319 | ||
320 | /* | |
321 | * Use this when you have no reliable mm, typically from interrupt | |
322 | * context. It is less reliable than using a task's mm and may give | |
323 | * false positives. | |
324 | */ | |
325 | int in_gate_area_no_mm(unsigned long addr) | |
326 | { | |
87983c66 | 327 | return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR; |
b9359090 AL |
328 | } |
329 | ||
e4026440 | 330 | void __init map_vsyscall(void) |
1da177e4 | 331 | { |
3ae36655 AL |
332 | extern char __vsyscall_page; |
333 | unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page); | |
1da177e4 | 334 | |
87983c66 AL |
335 | if (vsyscall_mode != NONE) |
336 | __set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall, | |
337 | vsyscall_mode == NATIVE | |
338 | ? PAGE_KERNEL_VSYSCALL | |
339 | : PAGE_KERNEL_VVAR); | |
340 | ||
f40c3300 AL |
341 | BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) != |
342 | (unsigned long)VSYSCALL_ADDR); | |
1da177e4 | 343 | } |