1 /*
2  *  Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
3  *  Copyright 2003 Andi Kleen, SuSE Labs.
4  *
5  *  [ NOTE: this mechanism is now deprecated in favor of the vDSO. ]
6  *
7  *  Thanks to hpa@transmeta.com for some useful hint.
8  *  Special thanks to Ingo Molnar for his early experience with
9  *  a different vsyscall implementation for Linux/IA32 and for the name.
10  *
11  *  vsyscall 1 is located at -10Mbyte, vsyscall 2 is located
12  *  at virtual address -10Mbyte+1024bytes etc... There are at max 4
13  *  vsyscalls. One vsyscall can reserve more than 1 slot to avoid
14  *  jumping out of line if necessary. We cannot add more with this
15  *  mechanism because older kernels won't return -ENOSYS.
16  *
17  *  Note: the concept clashes with user mode linux.  UML users should
18  *  use the vDSO.
19  */
20 
21 #include <linux/time.h>
22 #include <linux/init.h>
23 #include <linux/kernel.h>
24 #include <linux/timer.h>
25 #include <linux/seqlock.h>
26 #include <linux/jiffies.h>
27 #include <linux/sysctl.h>
28 #include <linux/topology.h>
29 #include <linux/clocksource.h>
30 #include <linux/getcpu.h>
31 #include <linux/cpu.h>
32 #include <linux/smp.h>
33 #include <linux/notifier.h>
34 #include <linux/syscalls.h>
35 #include <linux/ratelimit.h>
36 
37 #include <asm/vsyscall.h>
38 #include <asm/pgtable.h>
39 #include <asm/compat.h>
40 #include <asm/page.h>
41 #include <asm/unistd.h>
42 #include <asm/fixmap.h>
43 #include <asm/errno.h>
44 #include <asm/io.h>
45 #include <asm/segment.h>
46 #include <asm/desc.h>
47 #include <asm/topology.h>
48 #include <asm/vgtod.h>
49 #include <asm/traps.h>
50 
51 #define CREATE_TRACE_POINTS
52 #include "vsyscall_trace.h"
53 
54 DEFINE_VVAR(int, vgetcpu_mode);
55 DEFINE_VVAR(struct vsyscall_gtod_data, vsyscall_gtod_data);
56 
57 static enum { EMULATE, NATIVE, NONE } vsyscall_mode = EMULATE;
58 
vsyscall_setup(char * str)59 static int __init vsyscall_setup(char *str)
60 {
61 	if (str) {
62 		if (!strcmp("emulate", str))
63 			vsyscall_mode = EMULATE;
64 		else if (!strcmp("native", str))
65 			vsyscall_mode = NATIVE;
66 		else if (!strcmp("none", str))
67 			vsyscall_mode = NONE;
68 		else
69 			return -EINVAL;
70 
71 		return 0;
72 	}
73 
74 	return -EINVAL;
75 }
76 early_param("vsyscall", vsyscall_setup);
77 
update_vsyscall_tz(void)78 void update_vsyscall_tz(void)
79 {
80 	vsyscall_gtod_data.sys_tz = sys_tz;
81 }
82 
update_vsyscall(struct timespec * wall_time,struct timespec * wtm,struct clocksource * clock,u32 mult)83 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
84 			struct clocksource *clock, u32 mult)
85 {
86 	struct timespec monotonic;
87 
88 	write_seqcount_begin(&vsyscall_gtod_data.seq);
89 
90 	/* copy vsyscall data */
91 	vsyscall_gtod_data.clock.vclock_mode	= clock->archdata.vclock_mode;
92 	vsyscall_gtod_data.clock.cycle_last	= clock->cycle_last;
93 	vsyscall_gtod_data.clock.mask		= clock->mask;
94 	vsyscall_gtod_data.clock.mult		= mult;
95 	vsyscall_gtod_data.clock.shift		= clock->shift;
96 
97 	vsyscall_gtod_data.wall_time_sec	= wall_time->tv_sec;
98 	vsyscall_gtod_data.wall_time_nsec	= wall_time->tv_nsec;
99 
100 	monotonic = timespec_add(*wall_time, *wtm);
101 	vsyscall_gtod_data.monotonic_time_sec	= monotonic.tv_sec;
102 	vsyscall_gtod_data.monotonic_time_nsec	= monotonic.tv_nsec;
103 
104 	vsyscall_gtod_data.wall_time_coarse	= __current_kernel_time();
105 	vsyscall_gtod_data.monotonic_time_coarse =
106 		timespec_add(vsyscall_gtod_data.wall_time_coarse, *wtm);
107 
108 	write_seqcount_end(&vsyscall_gtod_data.seq);
109 }
110 
warn_bad_vsyscall(const char * level,struct pt_regs * regs,const char * message)111 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
112 			      const char *message)
113 {
114 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
115 	struct task_struct *tsk;
116 
117 	if (!show_unhandled_signals || !__ratelimit(&rs))
118 		return;
119 
120 	tsk = current;
121 
122 	printk("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
123 	       level, tsk->comm, task_pid_nr(tsk),
124 	       message, regs->ip, regs->cs,
125 	       regs->sp, regs->ax, regs->si, regs->di);
126 }
127 
addr_to_vsyscall_nr(unsigned long addr)128 static int addr_to_vsyscall_nr(unsigned long addr)
129 {
130 	int nr;
131 
132 	if ((addr & ~0xC00UL) != VSYSCALL_START)
133 		return -EINVAL;
134 
135 	nr = (addr & 0xC00UL) >> 10;
136 	if (nr >= 3)
137 		return -EINVAL;
138 
139 	return nr;
140 }
141 
write_ok_or_segv(unsigned long ptr,size_t size)142 static bool write_ok_or_segv(unsigned long ptr, size_t size)
143 {
144 	/*
145 	 * XXX: if access_ok, get_user, and put_user handled
146 	 * sig_on_uaccess_error, this could go away.
147 	 */
148 
149 	if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
150 		siginfo_t info;
151 		struct thread_struct *thread = &current->thread;
152 
153 		thread->error_code	= 6;  /* user fault, no page, write */
154 		thread->cr2		= ptr;
155 		thread->trap_nr		= X86_TRAP_PF;
156 
157 		memset(&info, 0, sizeof(info));
158 		info.si_signo		= SIGSEGV;
159 		info.si_errno		= 0;
160 		info.si_code		= SEGV_MAPERR;
161 		info.si_addr		= (void __user *)ptr;
162 
163 		force_sig_info(SIGSEGV, &info, current);
164 		return false;
165 	} else {
166 		return true;
167 	}
168 }
169 
emulate_vsyscall(struct pt_regs * regs,unsigned long address)170 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
171 {
172 	struct task_struct *tsk;
173 	unsigned long caller;
174 	int vsyscall_nr;
175 	int prev_sig_on_uaccess_error;
176 	long ret;
177 
178 	/*
179 	 * No point in checking CS -- the only way to get here is a user mode
180 	 * trap to a high address, which means that we're in 64-bit user code.
181 	 */
182 
183 	WARN_ON_ONCE(address != regs->ip);
184 
185 	if (vsyscall_mode == NONE) {
186 		warn_bad_vsyscall(KERN_INFO, regs,
187 				  "vsyscall attempted with vsyscall=none");
188 		return false;
189 	}
190 
191 	vsyscall_nr = addr_to_vsyscall_nr(address);
192 
193 	trace_emulate_vsyscall(vsyscall_nr);
194 
195 	if (vsyscall_nr < 0) {
196 		warn_bad_vsyscall(KERN_WARNING, regs,
197 				  "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
198 		goto sigsegv;
199 	}
200 
201 	if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
202 		warn_bad_vsyscall(KERN_WARNING, regs,
203 				  "vsyscall with bad stack (exploit attempt?)");
204 		goto sigsegv;
205 	}
206 
207 	tsk = current;
208 	if (seccomp_mode(&tsk->seccomp))
209 		do_exit(SIGKILL);
210 
211 	/*
212 	 * With a real vsyscall, page faults cause SIGSEGV.  We want to
213 	 * preserve that behavior to make writing exploits harder.
214 	 */
215 	prev_sig_on_uaccess_error = current_thread_info()->sig_on_uaccess_error;
216 	current_thread_info()->sig_on_uaccess_error = 1;
217 
218 	/*
219 	 * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
220 	 * 64-bit, so we don't need to special-case it here.  For all the
221 	 * vsyscalls, NULL means "don't write anything" not "write it at
222 	 * address 0".
223 	 */
224 	ret = -EFAULT;
225 	switch (vsyscall_nr) {
226 	case 0:
227 		if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
228 		    !write_ok_or_segv(regs->si, sizeof(struct timezone)))
229 			break;
230 
231 		ret = sys_gettimeofday(
232 			(struct timeval __user *)regs->di,
233 			(struct timezone __user *)regs->si);
234 		break;
235 
236 	case 1:
237 		if (!write_ok_or_segv(regs->di, sizeof(time_t)))
238 			break;
239 
240 		ret = sys_time((time_t __user *)regs->di);
241 		break;
242 
243 	case 2:
244 		if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
245 		    !write_ok_or_segv(regs->si, sizeof(unsigned)))
246 			break;
247 
248 		ret = sys_getcpu((unsigned __user *)regs->di,
249 				 (unsigned __user *)regs->si,
250 				 NULL);
251 		break;
252 	}
253 
254 	current_thread_info()->sig_on_uaccess_error = prev_sig_on_uaccess_error;
255 
256 	if (ret == -EFAULT) {
257 		/* Bad news -- userspace fed a bad pointer to a vsyscall. */
258 		warn_bad_vsyscall(KERN_INFO, regs,
259 				  "vsyscall fault (exploit attempt?)");
260 
261 		/*
262 		 * If we failed to generate a signal for any reason,
263 		 * generate one here.  (This should be impossible.)
264 		 */
265 		if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
266 				 !sigismember(&tsk->pending.signal, SIGSEGV)))
267 			goto sigsegv;
268 
269 		return true;  /* Don't emulate the ret. */
270 	}
271 
272 	regs->ax = ret;
273 
274 	/* Emulate a ret instruction. */
275 	regs->ip = caller;
276 	regs->sp += 8;
277 
278 	return true;
279 
280 sigsegv:
281 	force_sig(SIGSEGV, current);
282 	return true;
283 }
284 
285 /*
286  * Assume __initcall executes before all user space. Hopefully kmod
287  * doesn't violate that. We'll find out if it does.
288  */
vsyscall_set_cpu(int cpu)289 static void __cpuinit vsyscall_set_cpu(int cpu)
290 {
291 	unsigned long d;
292 	unsigned long node = 0;
293 #ifdef CONFIG_NUMA
294 	node = cpu_to_node(cpu);
295 #endif
296 	if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
297 		write_rdtscp_aux((node << 12) | cpu);
298 
299 	/*
300 	 * Store cpu number in limit so that it can be loaded quickly
301 	 * in user space in vgetcpu. (12 bits for the CPU and 8 bits for the node)
302 	 */
303 	d = 0x0f40000000000ULL;
304 	d |= cpu;
305 	d |= (node & 0xf) << 12;
306 	d |= (node >> 4) << 48;
307 
308 	write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
309 }
310 
cpu_vsyscall_init(void * arg)311 static void __cpuinit cpu_vsyscall_init(void *arg)
312 {
313 	/* preemption should be already off */
314 	vsyscall_set_cpu(raw_smp_processor_id());
315 }
316 
317 static int __cpuinit
cpu_vsyscall_notifier(struct notifier_block * n,unsigned long action,void * arg)318 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
319 {
320 	long cpu = (long)arg;
321 
322 	if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
323 		smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
324 
325 	return NOTIFY_DONE;
326 }
327 
map_vsyscall(void)328 void __init map_vsyscall(void)
329 {
330 	extern char __vsyscall_page;
331 	unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
332 	extern char __vvar_page;
333 	unsigned long physaddr_vvar_page = __pa_symbol(&__vvar_page);
334 
335 	__set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_vsyscall,
336 		     vsyscall_mode == NATIVE
337 		     ? PAGE_KERNEL_VSYSCALL
338 		     : PAGE_KERNEL_VVAR);
339 	BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_FIRST_PAGE) !=
340 		     (unsigned long)VSYSCALL_START);
341 
342 	__set_fixmap(VVAR_PAGE, physaddr_vvar_page, PAGE_KERNEL_VVAR);
343 	BUILD_BUG_ON((unsigned long)__fix_to_virt(VVAR_PAGE) !=
344 		     (unsigned long)VVAR_ADDRESS);
345 }
346 
vsyscall_init(void)347 static int __init vsyscall_init(void)
348 {
349 	BUG_ON(VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE));
350 
351 	on_each_cpu(cpu_vsyscall_init, NULL, 1);
352 	/* notifier priority > KVM */
353 	hotcpu_notifier(cpu_vsyscall_notifier, 30);
354 
355 	return 0;
356 }
357 __initcall(vsyscall_init);
358