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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
23 #include <linux/time.h>
24 #include <linux/init.h>
25 #include <linux/kernel.h>
26 #include <linux/timer.h>
27 #include <linux/seqlock.h>
28 #include <linux/jiffies.h>
29 #include <linux/sysctl.h>
30 #include <linux/topology.h>
31 #include <linux/timekeeper_internal.h>
32 #include <linux/getcpu.h>
33 #include <linux/cpu.h>
34 #include <linux/smp.h>
35 #include <linux/notifier.h>
36 #include <linux/syscalls.h>
37 #include <linux/ratelimit.h>
39 #include <asm/vsyscall.h>
40 #include <asm/pgtable.h>
41 #include <asm/compat.h>
43 #include <asm/unistd.h>
44 #include <asm/fixmap.h>
45 #include <asm/errno.h>
47 #include <asm/segment.h>
49 #include <asm/topology.h>
50 #include <asm/vgtod.h>
51 #include <asm/traps.h>
53 #define CREATE_TRACE_POINTS
54 #include "vsyscall_trace.h"
56 DEFINE_VVAR(int, vgetcpu_mode);
57 DEFINE_VVAR(struct vsyscall_gtod_data, vsyscall_gtod_data);
59 static enum { EMULATE, NATIVE, NONE } vsyscall_mode = EMULATE;
61 static int __init vsyscall_setup(char *str)
64 if (!strcmp("emulate", str))
65 vsyscall_mode = EMULATE;
66 else if (!strcmp("native", str))
67 vsyscall_mode = NATIVE;
68 else if (!strcmp("none", str))
78 early_param("vsyscall", vsyscall_setup);
80 void update_vsyscall_tz(void)
82 vsyscall_gtod_data.sys_tz = sys_tz;
85 void update_vsyscall(struct timekeeper *tk)
87 struct vsyscall_gtod_data *vdata = &vsyscall_gtod_data;
89 write_seqcount_begin(&vdata->seq);
91 /* copy vsyscall data */
92 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
93 vdata->clock.cycle_last = tk->clock->cycle_last;
94 vdata->clock.mask = tk->clock->mask;
95 vdata->clock.mult = tk->mult;
96 vdata->clock.shift = tk->shift;
98 vdata->wall_time_sec = tk->xtime_sec;
99 vdata->wall_time_snsec = tk->xtime_nsec;
101 vdata->monotonic_time_sec = tk->xtime_sec
102 + tk->wall_to_monotonic.tv_sec;
103 vdata->monotonic_time_snsec = tk->xtime_nsec
104 + (tk->wall_to_monotonic.tv_nsec
106 while (vdata->monotonic_time_snsec >=
107 (((u64)NSEC_PER_SEC) << tk->shift)) {
108 vdata->monotonic_time_snsec -=
109 ((u64)NSEC_PER_SEC) << tk->shift;
110 vdata->monotonic_time_sec++;
113 vdata->wall_time_coarse.tv_sec = tk->xtime_sec;
114 vdata->wall_time_coarse.tv_nsec = (long)(tk->xtime_nsec >> tk->shift);
116 vdata->monotonic_time_coarse = timespec_add(vdata->wall_time_coarse,
117 tk->wall_to_monotonic);
119 write_seqcount_end(&vdata->seq);
122 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
125 if (!show_unhandled_signals)
128 pr_notice_ratelimited("%s%s[%d] %s ip:%lx cs:%lx sp:%lx ax:%lx si:%lx di:%lx\n",
129 level, current->comm, task_pid_nr(current),
130 message, regs->ip, regs->cs,
131 regs->sp, regs->ax, regs->si, regs->di);
134 static int addr_to_vsyscall_nr(unsigned long addr)
138 if ((addr & ~0xC00UL) != VSYSCALL_START)
141 nr = (addr & 0xC00UL) >> 10;
148 #ifdef CONFIG_SECCOMP
149 static int vsyscall_seccomp(struct task_struct *tsk, int syscall_nr)
151 if (!seccomp_mode(&tsk->seccomp))
153 task_pt_regs(tsk)->orig_ax = syscall_nr;
154 task_pt_regs(tsk)->ax = syscall_nr;
155 return __secure_computing(syscall_nr);
158 #define vsyscall_seccomp(_tsk, _nr) 0
161 static bool write_ok_or_segv(unsigned long ptr, size_t size)
164 * XXX: if access_ok, get_user, and put_user handled
165 * sig_on_uaccess_error, this could go away.
168 if (!access_ok(VERIFY_WRITE, (void __user *)ptr, size)) {
170 struct thread_struct *thread = ¤t->thread;
172 thread->error_code = 6; /* user fault, no page, write */
174 thread->trap_nr = X86_TRAP_PF;
176 memset(&info, 0, sizeof(info));
177 info.si_signo = SIGSEGV;
179 info.si_code = SEGV_MAPERR;
180 info.si_addr = (void __user *)ptr;
182 force_sig_info(SIGSEGV, &info, current);
189 bool emulate_vsyscall(struct pt_regs *regs, unsigned long address)
191 struct task_struct *tsk;
192 unsigned long caller;
194 int prev_sig_on_uaccess_error;
199 * No point in checking CS -- the only way to get here is a user mode
200 * trap to a high address, which means that we're in 64-bit user code.
203 WARN_ON_ONCE(address != regs->ip);
205 if (vsyscall_mode == NONE) {
206 warn_bad_vsyscall(KERN_INFO, regs,
207 "vsyscall attempted with vsyscall=none");
211 vsyscall_nr = addr_to_vsyscall_nr(address);
213 trace_emulate_vsyscall(vsyscall_nr);
215 if (vsyscall_nr < 0) {
216 warn_bad_vsyscall(KERN_WARNING, regs,
217 "misaligned vsyscall (exploit attempt or buggy program) -- look up the vsyscall kernel parameter if you need a workaround");
221 if (get_user(caller, (unsigned long __user *)regs->sp) != 0) {
222 warn_bad_vsyscall(KERN_WARNING, regs,
223 "vsyscall with bad stack (exploit attempt?)");
229 * With a real vsyscall, page faults cause SIGSEGV. We want to
230 * preserve that behavior to make writing exploits harder.
232 prev_sig_on_uaccess_error = current_thread_info()->sig_on_uaccess_error;
233 current_thread_info()->sig_on_uaccess_error = 1;
236 * NULL is a valid user pointer (in the access_ok sense) on 32-bit and
237 * 64-bit, so we don't need to special-case it here. For all the
238 * vsyscalls, NULL means "don't write anything" not "write it at
243 switch (vsyscall_nr) {
245 skip = vsyscall_seccomp(tsk, __NR_gettimeofday);
249 if (!write_ok_or_segv(regs->di, sizeof(struct timeval)) ||
250 !write_ok_or_segv(regs->si, sizeof(struct timezone)))
253 ret = sys_gettimeofday(
254 (struct timeval __user *)regs->di,
255 (struct timezone __user *)regs->si);
259 skip = vsyscall_seccomp(tsk, __NR_time);
263 if (!write_ok_or_segv(regs->di, sizeof(time_t)))
266 ret = sys_time((time_t __user *)regs->di);
270 skip = vsyscall_seccomp(tsk, __NR_getcpu);
274 if (!write_ok_or_segv(regs->di, sizeof(unsigned)) ||
275 !write_ok_or_segv(regs->si, sizeof(unsigned)))
278 ret = sys_getcpu((unsigned __user *)regs->di,
279 (unsigned __user *)regs->si,
284 current_thread_info()->sig_on_uaccess_error = prev_sig_on_uaccess_error;
287 if ((long)regs->ax <= 0L) /* seccomp errno emulation */
289 goto done; /* seccomp trace/trap */
292 if (ret == -EFAULT) {
293 /* Bad news -- userspace fed a bad pointer to a vsyscall. */
294 warn_bad_vsyscall(KERN_INFO, regs,
295 "vsyscall fault (exploit attempt?)");
298 * If we failed to generate a signal for any reason,
299 * generate one here. (This should be impossible.)
301 if (WARN_ON_ONCE(!sigismember(&tsk->pending.signal, SIGBUS) &&
302 !sigismember(&tsk->pending.signal, SIGSEGV)))
305 return true; /* Don't emulate the ret. */
311 /* Emulate a ret instruction. */
318 force_sig(SIGSEGV, current);
323 * Assume __initcall executes before all user space. Hopefully kmod
324 * doesn't violate that. We'll find out if it does.
326 static void __cpuinit vsyscall_set_cpu(int cpu)
329 unsigned long node = 0;
331 node = cpu_to_node(cpu);
333 if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
334 write_rdtscp_aux((node << 12) | cpu);
337 * Store cpu number in limit so that it can be loaded quickly
338 * in user space in vgetcpu. (12 bits for the CPU and 8 bits for the node)
340 d = 0x0f40000000000ULL;
342 d |= (node & 0xf) << 12;
343 d |= (node >> 4) << 48;
345 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
348 static void __cpuinit cpu_vsyscall_init(void *arg)
350 /* preemption should be already off */
351 vsyscall_set_cpu(raw_smp_processor_id());
355 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
357 long cpu = (long)arg;
359 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
360 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
365 void __init map_vsyscall(void)
367 extern char __vsyscall_page;
368 unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
369 extern char __vvar_page;
370 unsigned long physaddr_vvar_page = __pa_symbol(&__vvar_page);
372 __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_vsyscall,
373 vsyscall_mode == NATIVE
374 ? PAGE_KERNEL_VSYSCALL
376 BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_FIRST_PAGE) !=
377 (unsigned long)VSYSCALL_START);
379 __set_fixmap(VVAR_PAGE, physaddr_vvar_page, PAGE_KERNEL_VVAR);
380 BUILD_BUG_ON((unsigned long)__fix_to_virt(VVAR_PAGE) !=
381 (unsigned long)VVAR_ADDRESS);
384 static int __init vsyscall_init(void)
386 BUG_ON(VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE));
388 on_each_cpu(cpu_vsyscall_init, NULL, 1);
389 /* notifier priority > KVM */
390 hotcpu_notifier(cpu_vsyscall_notifier, 30);
394 __initcall(vsyscall_init);