]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
95c46b56 AL |
3 | * Copyright (c) 2012-2014 Andy Lutomirski <luto@amacapital.net> |
4 | * | |
5 | * Based on the original implementation which is: | |
1da177e4 LT |
6 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE |
7 | * Copyright 2003 Andi Kleen, SuSE Labs. | |
8 | * | |
95c46b56 AL |
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: | |
5cec93c2 | 14 | * |
95c46b56 AL |
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. | |
1da177e4 | 20 | * |
95c46b56 AL |
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. | |
1da177e4 | 24 | * |
95c46b56 AL |
25 | * The code in this file emulates vsyscalls when notified of a page |
26 | * fault to a vsyscall address. | |
1da177e4 LT |
27 | */ |
28 | ||
1da177e4 LT |
29 | #include <linux/kernel.h> |
30 | #include <linux/timer.h> | |
3f07c014 | 31 | #include <linux/sched/signal.h> |
589ee628 | 32 | #include <linux/mm_types.h> |
5cec93c2 AL |
33 | #include <linux/syscalls.h> |
34 | #include <linux/ratelimit.h> | |
1da177e4 LT |
35 | |
36 | #include <asm/vsyscall.h> | |
7460ed28 | 37 | #include <asm/unistd.h> |
1da177e4 | 38 | #include <asm/fixmap.h> |
5cec93c2 | 39 | #include <asm/traps.h> |
49275fef | 40 | #include <asm/paravirt.h> |
1da177e4 | 41 | |
c149a665 AL |
42 | #define CREATE_TRACE_POINTS |
43 | #include "vsyscall_trace.h" | |
44 | ||
076ca272 AL |
45 | static enum { EMULATE, NONE } vsyscall_mode = |
46 | #ifdef CONFIG_LEGACY_VSYSCALL_NONE | |
3dc33bd3 KC |
47 | NONE; |
48 | #else | |
49 | EMULATE; | |
50 | #endif | |
3ae36655 AL |
51 | |
52 | static int __init vsyscall_setup(char *str) | |
53 | { | |
54 | if (str) { | |
55 | if (!strcmp("emulate", str)) | |
56 | vsyscall_mode = EMULATE; | |
3ae36655 AL |
57 | else if (!strcmp("none", str)) |
58 | vsyscall_mode = NONE; | |
59 | else | |
60 | return -EINVAL; | |
61 | ||
62 | return 0; | |
63 | } | |
64 | ||
65 | return -EINVAL; | |
66 | } | |
67 | early_param("vsyscall", vsyscall_setup); | |
68 | ||
5cec93c2 AL |
69 | static void warn_bad_vsyscall(const char *level, struct pt_regs *regs, |
70 | const char *message) | |
1da177e4 | 71 | { |
c767a54b | 72 | if (!show_unhandled_signals) |
5cec93c2 | 73 | return; |
1da177e4 | 74 | |
53b884ac AL |
75 | printk_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n", |
76 | level, current->comm, task_pid_nr(current), | |
77 | message, regs->ip, regs->cs, | |
78 | regs->sp, regs->ax, regs->si, regs->di); | |
c9712944 AL |
79 | } |
80 | ||
81 | static int addr_to_vsyscall_nr(unsigned long addr) | |
82 | { | |
83 | int nr; | |
84 | ||
f40c3300 | 85 | if ((addr & ~0xC00UL) != VSYSCALL_ADDR) |
c9712944 AL |
86 | return -EINVAL; |
87 | ||
88 | nr = (addr & 0xC00UL) >> 10; | |
89 | if (nr >= 3) | |
90 | return -EINVAL; | |
91 | ||
92 | return nr; | |
1da177e4 LT |
93 | } |
94 | ||
4fc34901 AL |
95 | static bool write_ok_or_segv(unsigned long ptr, size_t size) |
96 | { | |
97 | /* | |
98 | * XXX: if access_ok, get_user, and put_user handled | |
2a53ccbc | 99 | * sig_on_uaccess_err, this could go away. |
4fc34901 AL |
100 | */ |
101 | ||
102 | if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) { | |
103 | siginfo_t info; | |
104 | struct thread_struct *thread = ¤t->thread; | |
105 | ||
106 | thread->error_code = 6; /* user fault, no page, write */ | |
107 | thread->cr2 = ptr; | |
51e7dc70 | 108 | thread->trap_nr = X86_TRAP_PF; |
4fc34901 AL |
109 | |
110 | memset(&info, 0, sizeof(info)); | |
111 | info.si_signo = SIGSEGV; | |
112 | info.si_errno = 0; | |
113 | info.si_code = SEGV_MAPERR; | |
114 | info.si_addr = (void __user *)ptr; | |
115 | ||
116 | force_sig_info(SIGSEGV, &info, current); | |
117 | return false; | |
118 | } else { | |
119 | return true; | |
120 | } | |
121 | } | |
122 | ||
3ae36655 | 123 | bool emulate_vsyscall(struct pt_regs *regs, unsigned long address) |
1da177e4 | 124 | { |
5cec93c2 AL |
125 | struct task_struct *tsk; |
126 | unsigned long caller; | |
87b526d3 | 127 | int vsyscall_nr, syscall_nr, tmp; |
2a53ccbc | 128 | int prev_sig_on_uaccess_err; |
5cec93c2 AL |
129 | long ret; |
130 | ||
3ae36655 AL |
131 | /* |
132 | * No point in checking CS -- the only way to get here is a user mode | |
133 | * trap to a high address, which means that we're in 64-bit user code. | |
134 | */ | |
5cec93c2 | 135 | |
3ae36655 | 136 | WARN_ON_ONCE(address != regs->ip); |
c9712944 | 137 | |
3ae36655 AL |
138 | if (vsyscall_mode == NONE) { |
139 | warn_bad_vsyscall(KERN_INFO, regs, | |
140 | "vsyscall attempted with vsyscall=none"); | |
141 | return false; | |
c9712944 AL |
142 | } |
143 | ||
3ae36655 | 144 | vsyscall_nr = addr_to_vsyscall_nr(address); |
c149a665 AL |
145 | |
146 | trace_emulate_vsyscall(vsyscall_nr); | |
147 | ||
c9712944 AL |
148 | if (vsyscall_nr < 0) { |
149 | warn_bad_vsyscall(KERN_WARNING, regs, | |
3ae36655 | 150 | "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround"); |
5cec93c2 AL |
151 | goto sigsegv; |
152 | } | |
d0aff6e6 | 153 | |
5cec93c2 | 154 | if (get_user(caller, (unsigned long __user *)regs->sp) != 0) { |
3ae36655 AL |
155 | warn_bad_vsyscall(KERN_WARNING, regs, |
156 | "vsyscall with bad stack (exploit attempt?)"); | |
5cec93c2 AL |
157 | goto sigsegv; |
158 | } | |
8c73626a | 159 | |
5cec93c2 | 160 | tsk = current; |
4fc34901 AL |
161 | |
162 | /* | |
87b526d3 AL |
163 | * Check for access_ok violations and find the syscall nr. |
164 | * | |
46ed99d1 | 165 | * NULL is a valid user pointer (in the access_ok sense) on 32-bit and |
4fc34901 | 166 | * 64-bit, so we don't need to special-case it here. For all the |
46ed99d1 | 167 | * vsyscalls, NULL means "don't write anything" not "write it at |
4fc34901 AL |
168 | * address 0". |
169 | */ | |
5cec93c2 AL |
170 | switch (vsyscall_nr) { |
171 | case 0: | |
4fc34901 | 172 | if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) || |
87b526d3 AL |
173 | !write_ok_or_segv(regs->si, sizeof(struct timezone))) { |
174 | ret = -EFAULT; | |
175 | goto check_fault; | |
176 | } | |
4fc34901 | 177 | |
87b526d3 AL |
178 | syscall_nr = __NR_gettimeofday; |
179 | break; | |
180 | ||
181 | case 1: | |
182 | if (!write_ok_or_segv(regs->di, sizeof(time_t))) { | |
183 | ret = -EFAULT; | |
184 | goto check_fault; | |
185 | } | |
186 | ||
187 | syscall_nr = __NR_time; | |
188 | break; | |
189 | ||
190 | case 2: | |
191 | if (!write_ok_or_segv(regs->di, sizeof(unsigned)) || | |
192 | !write_ok_or_segv(regs->si, sizeof(unsigned))) { | |
193 | ret = -EFAULT; | |
194 | goto check_fault; | |
195 | } | |
196 | ||
197 | syscall_nr = __NR_getcpu; | |
198 | break; | |
199 | } | |
200 | ||
201 | /* | |
202 | * Handle seccomp. regs->ip must be the original value. | |
203 | * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt. | |
204 | * | |
205 | * We could optimize the seccomp disabled case, but performance | |
206 | * here doesn't matter. | |
207 | */ | |
208 | regs->orig_ax = syscall_nr; | |
209 | regs->ax = -ENOSYS; | |
2f275de5 | 210 | tmp = secure_computing(NULL); |
87b526d3 AL |
211 | if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) { |
212 | warn_bad_vsyscall(KERN_DEBUG, regs, | |
213 | "seccomp tried to change syscall nr or ip"); | |
214 | do_exit(SIGSYS); | |
215 | } | |
26893107 | 216 | regs->orig_ax = -1; |
87b526d3 AL |
217 | if (tmp) |
218 | goto do_ret; /* skip requested */ | |
219 | ||
220 | /* | |
221 | * With a real vsyscall, page faults cause SIGSEGV. We want to | |
222 | * preserve that behavior to make writing exploits harder. | |
223 | */ | |
2a53ccbc IM |
224 | prev_sig_on_uaccess_err = current->thread.sig_on_uaccess_err; |
225 | current->thread.sig_on_uaccess_err = 1; | |
87b526d3 AL |
226 | |
227 | ret = -EFAULT; | |
228 | switch (vsyscall_nr) { | |
229 | case 0: | |
5cec93c2 AL |
230 | ret = sys_gettimeofday( |
231 | (struct timeval __user *)regs->di, | |
232 | (struct timezone __user *)regs->si); | |
233 | break; | |
234 | ||
235 | case 1: | |
5cec93c2 AL |
236 | ret = sys_time((time_t __user *)regs->di); |
237 | break; | |
238 | ||
239 | case 2: | |
5cec93c2 AL |
240 | ret = sys_getcpu((unsigned __user *)regs->di, |
241 | (unsigned __user *)regs->si, | |
46ed99d1 | 242 | NULL); |
5cec93c2 | 243 | break; |
5cec93c2 | 244 | } |
8c73626a | 245 | |
2a53ccbc | 246 | current->thread.sig_on_uaccess_err = prev_sig_on_uaccess_err; |
4fc34901 | 247 | |
87b526d3 | 248 | check_fault: |
5cec93c2 | 249 | if (ret == -EFAULT) { |
4fc34901 | 250 | /* Bad news -- userspace fed a bad pointer to a vsyscall. */ |
5cec93c2 AL |
251 | warn_bad_vsyscall(KERN_INFO, regs, |
252 | "vsyscall fault (exploit attempt?)"); | |
4fc34901 AL |
253 | |
254 | /* | |
255 | * If we failed to generate a signal for any reason, | |
256 | * generate one here. (This should be impossible.) | |
257 | */ | |
258 | if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) && | |
259 | !sigismember(&tsk->pending.signal, SIGSEGV))) | |
260 | goto sigsegv; | |
261 | ||
262 | return true; /* Don't emulate the ret. */ | |
5cec93c2 | 263 | } |
8c73626a | 264 | |
5cec93c2 | 265 | regs->ax = ret; |
1da177e4 | 266 | |
5651721e | 267 | do_ret: |
5cec93c2 AL |
268 | /* Emulate a ret instruction. */ |
269 | regs->ip = caller; | |
270 | regs->sp += 8; | |
3ae36655 | 271 | return true; |
5cec93c2 AL |
272 | |
273 | sigsegv: | |
5cec93c2 | 274 | force_sig(SIGSEGV, current); |
3ae36655 | 275 | return true; |
1da177e4 LT |
276 | } |
277 | ||
b9359090 AL |
278 | /* |
279 | * A pseudo VMA to allow ptrace access for the vsyscall page. This only | |
280 | * covers the 64bit vsyscall page now. 32bit has a real VMA now and does | |
281 | * not need special handling anymore: | |
282 | */ | |
283 | static const char *gate_vma_name(struct vm_area_struct *vma) | |
284 | { | |
285 | return "[vsyscall]"; | |
286 | } | |
7cbea8dc | 287 | static const struct vm_operations_struct gate_vma_ops = { |
b9359090 AL |
288 | .name = gate_vma_name, |
289 | }; | |
290 | static struct vm_area_struct gate_vma = { | |
291 | .vm_start = VSYSCALL_ADDR, | |
292 | .vm_end = VSYSCALL_ADDR + PAGE_SIZE, | |
293 | .vm_page_prot = PAGE_READONLY_EXEC, | |
294 | .vm_flags = VM_READ | VM_EXEC, | |
295 | .vm_ops = &gate_vma_ops, | |
296 | }; | |
297 | ||
298 | struct vm_area_struct *get_gate_vma(struct mm_struct *mm) | |
299 | { | |
c338867d | 300 | #ifdef CONFIG_COMPAT |
b9359090 AL |
301 | if (!mm || mm->context.ia32_compat) |
302 | return NULL; | |
303 | #endif | |
87983c66 AL |
304 | if (vsyscall_mode == NONE) |
305 | return NULL; | |
b9359090 AL |
306 | return &gate_vma; |
307 | } | |
308 | ||
309 | int in_gate_area(struct mm_struct *mm, unsigned long addr) | |
310 | { | |
311 | struct vm_area_struct *vma = get_gate_vma(mm); | |
312 | ||
313 | if (!vma) | |
314 | return 0; | |
315 | ||
316 | return (addr >= vma->vm_start) && (addr < vma->vm_end); | |
317 | } | |
318 | ||
319 | /* | |
320 | * Use this when you have no reliable mm, typically from interrupt | |
321 | * context. It is less reliable than using a task's mm and may give | |
322 | * false positives. | |
323 | */ | |
324 | int in_gate_area_no_mm(unsigned long addr) | |
325 | { | |
87983c66 | 326 | return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR; |
b9359090 AL |
327 | } |
328 | ||
49275fef AL |
329 | /* |
330 | * The VSYSCALL page is the only user-accessible page in the kernel address | |
331 | * range. Normally, the kernel page tables can have _PAGE_USER clear, but | |
332 | * the tables covering VSYSCALL_ADDR need _PAGE_USER set if vsyscalls | |
333 | * are enabled. | |
334 | * | |
335 | * Some day we may create a "minimal" vsyscall mode in which we emulate | |
336 | * vsyscalls but leave the page not present. If so, we skip calling | |
337 | * this. | |
338 | */ | |
85900ea5 | 339 | void __init set_vsyscall_pgtable_user_bits(pgd_t *root) |
49275fef AL |
340 | { |
341 | pgd_t *pgd; | |
342 | p4d_t *p4d; | |
343 | pud_t *pud; | |
344 | pmd_t *pmd; | |
345 | ||
85900ea5 | 346 | pgd = pgd_offset_pgd(root, VSYSCALL_ADDR); |
49275fef AL |
347 | set_pgd(pgd, __pgd(pgd_val(*pgd) | _PAGE_USER)); |
348 | p4d = p4d_offset(pgd, VSYSCALL_ADDR); | |
349 | #if CONFIG_PGTABLE_LEVELS >= 5 | |
31ad7f8e | 350 | set_p4d(p4d, __p4d(p4d_val(*p4d) | _PAGE_USER)); |
49275fef AL |
351 | #endif |
352 | pud = pud_offset(p4d, VSYSCALL_ADDR); | |
353 | set_pud(pud, __pud(pud_val(*pud) | _PAGE_USER)); | |
354 | pmd = pmd_offset(pud, VSYSCALL_ADDR); | |
355 | set_pmd(pmd, __pmd(pmd_val(*pmd) | _PAGE_USER)); | |
356 | } | |
357 | ||
e4026440 | 358 | void __init map_vsyscall(void) |
1da177e4 | 359 | { |
3ae36655 AL |
360 | extern char __vsyscall_page; |
361 | unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page); | |
1da177e4 | 362 | |
49275fef | 363 | if (vsyscall_mode != NONE) { |
87983c66 | 364 | __set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall, |
076ca272 | 365 | PAGE_KERNEL_VVAR); |
85900ea5 | 366 | set_vsyscall_pgtable_user_bits(swapper_pg_dir); |
49275fef | 367 | } |
87983c66 | 368 | |
f40c3300 AL |
369 | BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) != |
370 | (unsigned long)VSYSCALL_ADDR); | |
1da177e4 | 371 | } |