2 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
3 * Copyright 2003 Andi Kleen, SuSE Labs.
5 * [ NOTE: this mechanism is now deprecated in favor of the vDSO. ]
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.
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.
17 * Note: the concept clashes with user mode linux. UML users should
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>
37 #include <asm/vsyscall.h>
38 #include <asm/pgtable.h>
39 #include <asm/compat.h>
41 #include <asm/unistd.h>
42 #include <asm/fixmap.h>
43 #include <asm/errno.h>
45 #include <asm/segment.h>
47 #include <asm/topology.h>
48 #include <asm/vgtod.h>
49 #include <asm/traps.h>
51 #define CREATE_TRACE_POINTS
52 #include "vsyscall_trace.h"
54 DEFINE_VVAR(int, vgetcpu_mode);
55 DEFINE_VVAR(struct vsyscall_gtod_data, vsyscall_gtod_data) =
57 .lock = __SEQLOCK_UNLOCKED(__vsyscall_gtod_data.lock),
60 static enum { EMULATE, NATIVE, NONE } vsyscall_mode = EMULATE;
62 static int __init vsyscall_setup(char *str)
65 if (!strcmp("emulate", str))
66 vsyscall_mode = EMULATE;
67 else if (!strcmp("native", str))
68 vsyscall_mode = NATIVE;
69 else if (!strcmp("none", str))
79 early_param("vsyscall", vsyscall_setup);
81 void update_vsyscall_tz(void)
85 write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
86 /* sys_tz has changed */
87 vsyscall_gtod_data.sys_tz = sys_tz;
88 write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
91 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
92 struct clocksource *clock, u32 mult)
96 write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
98 /* copy vsyscall data */
100 vsyscall_gtod_data.clock.vclock_mode = clock->archdata.vclock_mode;
102 vsyscall_gtod_data.clock.cycle_last = clock->cycle_last;
103 vsyscall_gtod_data.clock.mask = clock->mask;
104 vsyscall_gtod_data.clock.mult = mult;
105 vsyscall_gtod_data.clock.shift = clock->shift;
106 vsyscall_gtod_data.wall_time_sec = wall_time->tv_sec;
107 vsyscall_gtod_data.wall_time_nsec = wall_time->tv_nsec;
108 vsyscall_gtod_data.wall_to_monotonic = *wtm;
109 vsyscall_gtod_data.wall_time_coarse = __current_kernel_time();
111 write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
114 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
117 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
118 struct task_struct *tsk;
120 if (!show_unhandled_signals || !__ratelimit(&rs))
125 printk("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
126 level, tsk->comm, task_pid_nr(tsk),
127 message, regs->ip, regs->cs,
128 regs->sp, regs->ax, regs->si, regs->di);
131 static int addr_to_vsyscall_nr(unsigned long addr)
135 if ((addr & ~0xC00UL) != VSYSCALL_START)
138 nr = (addr & 0xC00UL) >> 10;
145 static bool write_ok_or_segv(unsigned long ptr, size_t size)
148 * XXX: if access_ok, get_user, and put_user handled
149 * sig_on_uaccess_error, this could go away.
152 if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
154 struct thread_struct *thread = ¤t->thread;
156 thread->error_code = 6; /* user fault, no page, write */
158 thread->trap_no = 14;
160 memset(&info, 0, sizeof(info));
161 info.si_signo = SIGSEGV;
163 info.si_code = SEGV_MAPERR;
164 info.si_addr = (void __user *)ptr;
166 force_sig_info(SIGSEGV, &info, current);
173 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
175 struct task_struct *tsk;
176 unsigned long caller;
178 int prev_sig_on_uaccess_error;
182 * No point in checking CS -- the only way to get here is a user mode
183 * trap to a high address, which means that we're in 64-bit user code.
186 WARN_ON_ONCE(address != regs->ip);
188 if (vsyscall_mode == NONE) {
189 warn_bad_vsyscall(KERN_INFO, regs,
190 "vsyscall attempted with vsyscall=none");
194 vsyscall_nr = addr_to_vsyscall_nr(address);
196 trace_emulate_vsyscall(vsyscall_nr);
198 if (vsyscall_nr < 0) {
199 warn_bad_vsyscall(KERN_WARNING, regs,
200 "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
204 if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
205 warn_bad_vsyscall(KERN_WARNING, regs,
206 "vsyscall with bad stack (exploit attempt?)");
211 if (seccomp_mode(&tsk->seccomp))
215 * With a real vsyscall, page faults cause SIGSEGV. We want to
216 * preserve that behavior to make writing exploits harder.
218 prev_sig_on_uaccess_error = current_thread_info()->sig_on_uaccess_error;
219 current_thread_info()->sig_on_uaccess_error = 1;
222 * 0 is a valid user pointer (in the access_ok sense) on 32-bit and
223 * 64-bit, so we don't need to special-case it here. For all the
224 * vsyscalls, 0 means "don't write anything" not "write it at
228 switch (vsyscall_nr) {
230 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
231 !write_ok_or_segv(regs->si, sizeof(struct timezone)))
234 ret = sys_gettimeofday(
235 (struct timeval __user *)regs->di,
236 (struct timezone __user *)regs->si);
240 if (!write_ok_or_segv(regs->di, sizeof(time_t)))
243 ret = sys_time((time_t __user *)regs->di);
247 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
248 !write_ok_or_segv(regs->si, sizeof(unsigned)))
251 ret = sys_getcpu((unsigned __user *)regs->di,
252 (unsigned __user *)regs->si,
260 current_thread_info()->sig_on_uaccess_error = prev_sig_on_uaccess_error;
262 if (ret == -EFAULT) {
263 /* Bad news -- userspace fed a bad pointer to a vsyscall. */
264 warn_bad_vsyscall(KERN_INFO, regs,
265 "vsyscall fault (exploit attempt?)");
268 * If we failed to generate a signal for any reason,
269 * generate one here. (This should be impossible.)
271 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
272 !sigismember(&tsk->pending.signal, SIGSEGV)))
275 return true; /* Don't emulate the ret. */
280 /* Emulate a ret instruction. */
287 force_sig(SIGSEGV, current);
292 * Assume __initcall executes before all user space. Hopefully kmod
293 * doesn't violate that. We'll find out if it does.
295 static void __cpuinit vsyscall_set_cpu(int cpu)
298 unsigned long node = 0;
300 node = cpu_to_node(cpu);
302 if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
303 write_rdtscp_aux((node << 12) | cpu);
306 * Store cpu number in limit so that it can be loaded quickly
307 * in user space in vgetcpu. (12 bits for the CPU and 8 bits for the node)
309 d = 0x0f40000000000ULL;
311 d |= (node & 0xf) << 12;
312 d |= (node >> 4) << 48;
314 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
317 static void __cpuinit cpu_vsyscall_init(void *arg)
319 /* preemption should be already off */
320 vsyscall_set_cpu(raw_smp_processor_id());
324 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
326 long cpu = (long)arg;
328 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
329 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
334 void __init map_vsyscall(void)
336 extern char __vsyscall_page;
337 unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
338 extern char __vvar_page;
339 unsigned long physaddr_vvar_page = __pa_symbol(&__vvar_page);
341 __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_vsyscall,
342 vsyscall_mode == NATIVE
343 ? PAGE_KERNEL_VSYSCALL
345 BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_FIRST_PAGE) !=
346 (unsigned long)VSYSCALL_START);
348 __set_fixmap(VVAR_PAGE, physaddr_vvar_page, PAGE_KERNEL_VVAR);
349 BUILD_BUG_ON((unsigned long)__fix_to_virt(VVAR_PAGE) !=
350 (unsigned long)VVAR_ADDRESS);
353 static int __init vsyscall_init(void)
355 BUG_ON(VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE));
357 on_each_cpu(cpu_vsyscall_init, NULL, 1);
358 /* notifier priority > KVM */
359 hotcpu_notifier(cpu_vsyscall_notifier, 30);
363 __initcall(vsyscall_init);