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 | ||
3dc33bd3 | 45 | static enum { EMULATE, NATIVE, NONE } vsyscall_mode = |
93f13a9f | 46 | #if defined(CONFIG_LEGACY_VSYSCALL_NATIVE) |
3dc33bd3 | 47 | NATIVE; |
93f13a9f | 48 | #elif defined(CONFIG_LEGACY_VSYSCALL_NONE) |
3dc33bd3 KC |
49 | NONE; |
50 | #else | |
51 | EMULATE; | |
52 | #endif | |
3ae36655 AL |
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("native", str)) | |
60 | vsyscall_mode = NATIVE; | |
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 | ||
5cec93c2 AL |
73 | static void warn_bad_vsyscall(const char *level, struct pt_regs *regs, |
74 | const char *message) | |
1da177e4 | 75 | { |
c767a54b | 76 | if (!show_unhandled_signals) |
5cec93c2 | 77 | return; |
1da177e4 | 78 | |
53b884ac AL |
79 | printk_ratelimited("%s%s[%d] %s ip:%lx cs:%lx 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); | |
c9712944 AL |
83 | } |
84 | ||
85 | static int addr_to_vsyscall_nr(unsigned long addr) | |
86 | { | |
87 | int nr; | |
88 | ||
f40c3300 | 89 | if ((addr & ~0xC00UL) != VSYSCALL_ADDR) |
c9712944 AL |
90 | return -EINVAL; |
91 | ||
92 | nr = (addr & 0xC00UL) >> 10; | |
93 | if (nr >= 3) | |
94 | return -EINVAL; | |
95 | ||
96 | return nr; | |
1da177e4 LT |
97 | } |
98 | ||
4fc34901 AL |
99 | static bool write_ok_or_segv(unsigned long ptr, size_t size) |
100 | { | |
101 | /* | |
102 | * XXX: if access_ok, get_user, and put_user handled | |
2a53ccbc | 103 | * sig_on_uaccess_err, this could go away. |
4fc34901 AL |
104 | */ |
105 | ||
106 | if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) { | |
107 | siginfo_t info; | |
108 | struct thread_struct *thread = ¤t->thread; | |
109 | ||
110 | thread->error_code = 6; /* user fault, no page, write */ | |
111 | thread->cr2 = ptr; | |
51e7dc70 | 112 | thread->trap_nr = X86_TRAP_PF; |
4fc34901 AL |
113 | |
114 | memset(&info, 0, sizeof(info)); | |
115 | info.si_signo = SIGSEGV; | |
116 | info.si_errno = 0; | |
117 | info.si_code = SEGV_MAPERR; | |
118 | info.si_addr = (void __user *)ptr; | |
119 | ||
120 | force_sig_info(SIGSEGV, &info, current); | |
121 | return false; | |
122 | } else { | |
123 | return true; | |
124 | } | |
125 | } | |
126 | ||
3ae36655 | 127 | bool emulate_vsyscall(struct pt_regs *regs, unsigned long address) |
1da177e4 | 128 | { |
5cec93c2 AL |
129 | struct task_struct *tsk; |
130 | unsigned long caller; | |
87b526d3 | 131 | int vsyscall_nr, syscall_nr, tmp; |
2a53ccbc | 132 | int prev_sig_on_uaccess_err; |
5cec93c2 AL |
133 | long ret; |
134 | ||
3ae36655 AL |
135 | /* |
136 | * No point in checking CS -- the only way to get here is a user mode | |
137 | * trap to a high address, which means that we're in 64-bit user code. | |
138 | */ | |
5cec93c2 | 139 | |
3ae36655 | 140 | WARN_ON_ONCE(address != regs->ip); |
c9712944 | 141 | |
3ae36655 AL |
142 | if (vsyscall_mode == NONE) { |
143 | warn_bad_vsyscall(KERN_INFO, regs, | |
144 | "vsyscall attempted with vsyscall=none"); | |
145 | return false; | |
c9712944 AL |
146 | } |
147 | ||
3ae36655 | 148 | vsyscall_nr = addr_to_vsyscall_nr(address); |
c149a665 AL |
149 | |
150 | trace_emulate_vsyscall(vsyscall_nr); | |
151 | ||
c9712944 AL |
152 | if (vsyscall_nr < 0) { |
153 | warn_bad_vsyscall(KERN_WARNING, regs, | |
3ae36655 | 154 | "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround"); |
5cec93c2 AL |
155 | goto sigsegv; |
156 | } | |
d0aff6e6 | 157 | |
5cec93c2 | 158 | if (get_user(caller, (unsigned long __user *)regs->sp) != 0) { |
3ae36655 AL |
159 | warn_bad_vsyscall(KERN_WARNING, regs, |
160 | "vsyscall with bad stack (exploit attempt?)"); | |
5cec93c2 AL |
161 | goto sigsegv; |
162 | } | |
8c73626a | 163 | |
5cec93c2 | 164 | tsk = current; |
4fc34901 AL |
165 | |
166 | /* | |
87b526d3 AL |
167 | * Check for access_ok violations and find the syscall nr. |
168 | * | |
46ed99d1 | 169 | * NULL is a valid user pointer (in the access_ok sense) on 32-bit and |
4fc34901 | 170 | * 64-bit, so we don't need to special-case it here. For all the |
46ed99d1 | 171 | * vsyscalls, NULL means "don't write anything" not "write it at |
4fc34901 AL |
172 | * address 0". |
173 | */ | |
5cec93c2 AL |
174 | switch (vsyscall_nr) { |
175 | case 0: | |
4fc34901 | 176 | if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) || |
87b526d3 AL |
177 | !write_ok_or_segv(regs->si, sizeof(struct timezone))) { |
178 | ret = -EFAULT; | |
179 | goto check_fault; | |
180 | } | |
4fc34901 | 181 | |
87b526d3 AL |
182 | syscall_nr = __NR_gettimeofday; |
183 | break; | |
184 | ||
185 | case 1: | |
186 | if (!write_ok_or_segv(regs->di, sizeof(time_t))) { | |
187 | ret = -EFAULT; | |
188 | goto check_fault; | |
189 | } | |
190 | ||
191 | syscall_nr = __NR_time; | |
192 | break; | |
193 | ||
194 | case 2: | |
195 | if (!write_ok_or_segv(regs->di, sizeof(unsigned)) || | |
196 | !write_ok_or_segv(regs->si, sizeof(unsigned))) { | |
197 | ret = -EFAULT; | |
198 | goto check_fault; | |
199 | } | |
200 | ||
201 | syscall_nr = __NR_getcpu; | |
202 | break; | |
203 | } | |
204 | ||
205 | /* | |
206 | * Handle seccomp. regs->ip must be the original value. | |
207 | * See seccomp_send_sigsys and Documentation/prctl/seccomp_filter.txt. | |
208 | * | |
209 | * We could optimize the seccomp disabled case, but performance | |
210 | * here doesn't matter. | |
211 | */ | |
212 | regs->orig_ax = syscall_nr; | |
213 | regs->ax = -ENOSYS; | |
2f275de5 | 214 | tmp = secure_computing(NULL); |
87b526d3 AL |
215 | if ((!tmp && regs->orig_ax != syscall_nr) || regs->ip != address) { |
216 | warn_bad_vsyscall(KERN_DEBUG, regs, | |
217 | "seccomp tried to change syscall nr or ip"); | |
218 | do_exit(SIGSYS); | |
219 | } | |
26893107 | 220 | regs->orig_ax = -1; |
87b526d3 AL |
221 | if (tmp) |
222 | goto do_ret; /* skip requested */ | |
223 | ||
224 | /* | |
225 | * With a real vsyscall, page faults cause SIGSEGV. We want to | |
226 | * preserve that behavior to make writing exploits harder. | |
227 | */ | |
2a53ccbc IM |
228 | prev_sig_on_uaccess_err = current->thread.sig_on_uaccess_err; |
229 | current->thread.sig_on_uaccess_err = 1; | |
87b526d3 AL |
230 | |
231 | ret = -EFAULT; | |
232 | switch (vsyscall_nr) { | |
233 | case 0: | |
5cec93c2 AL |
234 | ret = sys_gettimeofday( |
235 | (struct timeval __user *)regs->di, | |
236 | (struct timezone __user *)regs->si); | |
237 | break; | |
238 | ||
239 | case 1: | |
5cec93c2 AL |
240 | ret = sys_time((time_t __user *)regs->di); |
241 | break; | |
242 | ||
243 | case 2: | |
5cec93c2 AL |
244 | ret = sys_getcpu((unsigned __user *)regs->di, |
245 | (unsigned __user *)regs->si, | |
46ed99d1 | 246 | NULL); |
5cec93c2 | 247 | break; |
5cec93c2 | 248 | } |
8c73626a | 249 | |
2a53ccbc | 250 | current->thread.sig_on_uaccess_err = prev_sig_on_uaccess_err; |
4fc34901 | 251 | |
87b526d3 | 252 | check_fault: |
5cec93c2 | 253 | if (ret == -EFAULT) { |
4fc34901 | 254 | /* Bad news -- userspace fed a bad pointer to a vsyscall. */ |
5cec93c2 AL |
255 | warn_bad_vsyscall(KERN_INFO, regs, |
256 | "vsyscall fault (exploit attempt?)"); | |
4fc34901 AL |
257 | |
258 | /* | |
259 | * If we failed to generate a signal for any reason, | |
260 | * generate one here. (This should be impossible.) | |
261 | */ | |
262 | if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) && | |
263 | !sigismember(&tsk->pending.signal, SIGSEGV))) | |
264 | goto sigsegv; | |
265 | ||
266 | return true; /* Don't emulate the ret. */ | |
5cec93c2 | 267 | } |
8c73626a | 268 | |
5cec93c2 | 269 | regs->ax = ret; |
1da177e4 | 270 | |
5651721e | 271 | do_ret: |
5cec93c2 AL |
272 | /* Emulate a ret instruction. */ |
273 | regs->ip = caller; | |
274 | regs->sp += 8; | |
3ae36655 | 275 | return true; |
5cec93c2 AL |
276 | |
277 | sigsegv: | |
5cec93c2 | 278 | force_sig(SIGSEGV, current); |
3ae36655 | 279 | return true; |
1da177e4 LT |
280 | } |
281 | ||
b9359090 AL |
282 | /* |
283 | * A pseudo VMA to allow ptrace access for the vsyscall page. This only | |
284 | * covers the 64bit vsyscall page now. 32bit has a real VMA now and does | |
285 | * not need special handling anymore: | |
286 | */ | |
287 | static const char *gate_vma_name(struct vm_area_struct *vma) | |
288 | { | |
289 | return "[vsyscall]"; | |
290 | } | |
7cbea8dc | 291 | static const struct vm_operations_struct gate_vma_ops = { |
b9359090 AL |
292 | .name = gate_vma_name, |
293 | }; | |
294 | static struct vm_area_struct gate_vma = { | |
295 | .vm_start = VSYSCALL_ADDR, | |
296 | .vm_end = VSYSCALL_ADDR + PAGE_SIZE, | |
297 | .vm_page_prot = PAGE_READONLY_EXEC, | |
298 | .vm_flags = VM_READ | VM_EXEC, | |
299 | .vm_ops = &gate_vma_ops, | |
300 | }; | |
301 | ||
302 | struct vm_area_struct *get_gate_vma(struct mm_struct *mm) | |
303 | { | |
c338867d | 304 | #ifdef CONFIG_COMPAT |
b9359090 AL |
305 | if (!mm || mm->context.ia32_compat) |
306 | return NULL; | |
307 | #endif | |
87983c66 AL |
308 | if (vsyscall_mode == NONE) |
309 | return NULL; | |
b9359090 AL |
310 | return &gate_vma; |
311 | } | |
312 | ||
313 | int in_gate_area(struct mm_struct *mm, unsigned long addr) | |
314 | { | |
315 | struct vm_area_struct *vma = get_gate_vma(mm); | |
316 | ||
317 | if (!vma) | |
318 | return 0; | |
319 | ||
320 | return (addr >= vma->vm_start) && (addr < vma->vm_end); | |
321 | } | |
322 | ||
323 | /* | |
324 | * Use this when you have no reliable mm, typically from interrupt | |
325 | * context. It is less reliable than using a task's mm and may give | |
326 | * false positives. | |
327 | */ | |
328 | int in_gate_area_no_mm(unsigned long addr) | |
329 | { | |
87983c66 | 330 | return vsyscall_mode != NONE && (addr & PAGE_MASK) == VSYSCALL_ADDR; |
b9359090 AL |
331 | } |
332 | ||
49275fef AL |
333 | /* |
334 | * The VSYSCALL page is the only user-accessible page in the kernel address | |
335 | * range. Normally, the kernel page tables can have _PAGE_USER clear, but | |
336 | * the tables covering VSYSCALL_ADDR need _PAGE_USER set if vsyscalls | |
337 | * are enabled. | |
338 | * | |
339 | * Some day we may create a "minimal" vsyscall mode in which we emulate | |
340 | * vsyscalls but leave the page not present. If so, we skip calling | |
341 | * this. | |
342 | */ | |
343 | static void __init set_vsyscall_pgtable_user_bits(void) | |
344 | { | |
345 | pgd_t *pgd; | |
346 | p4d_t *p4d; | |
347 | pud_t *pud; | |
348 | pmd_t *pmd; | |
349 | ||
350 | pgd = pgd_offset_k(VSYSCALL_ADDR); | |
351 | set_pgd(pgd, __pgd(pgd_val(*pgd) | _PAGE_USER)); | |
352 | p4d = p4d_offset(pgd, VSYSCALL_ADDR); | |
353 | #if CONFIG_PGTABLE_LEVELS >= 5 | |
354 | p4d->p4d |= _PAGE_USER; | |
355 | #endif | |
356 | pud = pud_offset(p4d, VSYSCALL_ADDR); | |
357 | set_pud(pud, __pud(pud_val(*pud) | _PAGE_USER)); | |
358 | pmd = pmd_offset(pud, VSYSCALL_ADDR); | |
359 | set_pmd(pmd, __pmd(pmd_val(*pmd) | _PAGE_USER)); | |
360 | } | |
361 | ||
e4026440 | 362 | void __init map_vsyscall(void) |
1da177e4 | 363 | { |
3ae36655 AL |
364 | extern char __vsyscall_page; |
365 | unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page); | |
1da177e4 | 366 | |
49275fef | 367 | if (vsyscall_mode != NONE) { |
87983c66 AL |
368 | __set_fixmap(VSYSCALL_PAGE, physaddr_vsyscall, |
369 | vsyscall_mode == NATIVE | |
370 | ? PAGE_KERNEL_VSYSCALL | |
371 | : PAGE_KERNEL_VVAR); | |
49275fef AL |
372 | set_vsyscall_pgtable_user_bits(); |
373 | } | |
87983c66 | 374 | |
f40c3300 AL |
375 | BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) != |
376 | (unsigned long)VSYSCALL_ADDR); | |
1da177e4 | 377 | } |