2 * common.c - C code for kernel entry and exit
3 * Copyright (c) 2015 Andrew Lutomirski
6 * Based on asm and ptrace code by many authors. The code here originated
7 * in ptrace.c and signal.c.
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
13 #include <linux/smp.h>
14 #include <linux/errno.h>
15 #include <linux/ptrace.h>
16 #include <linux/tracehook.h>
17 #include <linux/audit.h>
18 #include <linux/seccomp.h>
19 #include <linux/signal.h>
20 #include <linux/export.h>
21 #include <linux/context_tracking.h>
22 #include <linux/user-return-notifier.h>
23 #include <linux/uprobes.h>
26 #include <asm/traps.h>
28 #include <asm/uaccess.h>
29 #include <asm/cpufeature.h>
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/syscalls.h>
34 #ifdef CONFIG_CONTEXT_TRACKING
35 /* Called on entry from user mode with IRQs off. */
36 __visible inline void enter_from_user_mode(void)
38 CT_WARN_ON(ct_state() != CONTEXT_USER);
42 static inline void enter_from_user_mode(void) {}
45 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
48 if (arch == AUDIT_ARCH_X86_64) {
49 audit_syscall_entry(regs->orig_ax, regs->di,
50 regs->si, regs->dx, regs->r10);
54 audit_syscall_entry(regs->orig_ax, regs->bx,
55 regs->cx, regs->dx, regs->si);
60 * Returns the syscall nr to run (which should match regs->orig_ax) or -1
61 * to skip the syscall.
63 static long syscall_trace_enter(struct pt_regs *regs)
65 u32 arch = in_ia32_syscall() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
67 struct thread_info *ti = current_thread_info();
68 unsigned long ret = 0;
69 bool emulated = false;
72 if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
73 BUG_ON(regs != task_pt_regs(current));
75 work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
77 if (unlikely(work & _TIF_SYSCALL_EMU))
80 if ((emulated || (work & _TIF_SYSCALL_TRACE)) &&
81 tracehook_report_syscall_entry(regs))
89 * Do seccomp after ptrace, to catch any tracer changes.
91 if (work & _TIF_SECCOMP) {
92 struct seccomp_data sd;
95 sd.nr = regs->orig_ax;
96 sd.instruction_pointer = regs->ip;
98 if (arch == AUDIT_ARCH_X86_64) {
99 sd.args[0] = regs->di;
100 sd.args[1] = regs->si;
101 sd.args[2] = regs->dx;
102 sd.args[3] = regs->r10;
103 sd.args[4] = regs->r8;
104 sd.args[5] = regs->r9;
108 sd.args[0] = regs->bx;
109 sd.args[1] = regs->cx;
110 sd.args[2] = regs->dx;
111 sd.args[3] = regs->si;
112 sd.args[4] = regs->di;
113 sd.args[5] = regs->bp;
116 ret = __secure_computing(&sd);
122 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
123 trace_sys_enter(regs, regs->orig_ax);
125 do_audit_syscall_entry(regs, arch);
127 return ret ?: regs->orig_ax;
130 #define EXIT_TO_USERMODE_LOOP_FLAGS \
131 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
132 _TIF_NEED_RESCHED_MASK | _TIF_USER_RETURN_NOTIFY)
134 static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
137 * In order to return to user mode, we need to have IRQs off with
138 * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
139 * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags
140 * can be set at any time on preemptable kernels if we have IRQs on,
141 * so we need to loop. Disabling preemption wouldn't help: doing the
142 * work to clear some of the flags can sleep.
145 /* We have work to do. */
148 if (cached_flags & _TIF_NEED_RESCHED_MASK)
151 #ifdef ARCH_RT_DELAYS_SIGNAL_SEND
152 if (unlikely(current->forced_info.si_signo)) {
153 struct task_struct *t = current;
154 force_sig_info(t->forced_info.si_signo, &t->forced_info, t);
155 t->forced_info.si_signo = 0;
158 if (cached_flags & _TIF_UPROBE)
159 uprobe_notify_resume(regs);
161 /* deal with pending signal delivery */
162 if (cached_flags & _TIF_SIGPENDING)
165 if (cached_flags & _TIF_NOTIFY_RESUME) {
166 clear_thread_flag(TIF_NOTIFY_RESUME);
167 tracehook_notify_resume(regs);
170 if (cached_flags & _TIF_USER_RETURN_NOTIFY)
171 fire_user_return_notifiers();
173 /* Disable IRQs and retry */
176 cached_flags = READ_ONCE(current_thread_info()->flags);
178 if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
183 /* Called with IRQs disabled. */
184 __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
186 struct thread_info *ti = current_thread_info();
189 if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
194 cached_flags = READ_ONCE(ti->flags);
196 if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
197 exit_to_usermode_loop(regs, cached_flags);
201 * Compat syscalls set TS_COMPAT. Make sure we clear it before
202 * returning to user mode. We need to clear it *after* signal
203 * handling, because syscall restart has a fixup for compat
204 * syscalls. The fixup is exercised by the ptrace_syscall_32
207 * We also need to clear TS_REGS_POKED_I386: the 32-bit tracer
208 * special case only applies after poking regs and before the
209 * very next return to user mode.
211 current->thread.status &= ~(TS_COMPAT|TS_I386_REGS_POKED);
217 #define SYSCALL_EXIT_WORK_FLAGS \
218 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
219 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
221 static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags)
225 audit_syscall_exit(regs);
227 if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
228 trace_sys_exit(regs, regs->ax);
231 * If TIF_SYSCALL_EMU is set, we only get here because of
232 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
233 * We already reported this syscall instruction in
234 * syscall_trace_enter().
237 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
239 if (step || cached_flags & _TIF_SYSCALL_TRACE)
240 tracehook_report_syscall_exit(regs, step);
244 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
245 * state such that we can immediately switch to user mode.
247 __visible inline void syscall_return_slowpath(struct pt_regs *regs)
249 struct thread_info *ti = current_thread_info();
250 u32 cached_flags = READ_ONCE(ti->flags);
252 CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
254 if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
255 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
259 * First do one-time work. If these work items are enabled, we
260 * want to run them exactly once per syscall exit with IRQs on.
262 if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS))
263 syscall_slow_exit_work(regs, cached_flags);
266 prepare_exit_to_usermode(regs);
270 __visible void do_syscall_64(struct pt_regs *regs)
272 struct thread_info *ti = current_thread_info();
273 unsigned long nr = regs->orig_ax;
275 enter_from_user_mode();
278 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
279 nr = syscall_trace_enter(regs);
282 * NB: Native and x32 syscalls are dispatched from the same
283 * table. The only functional difference is the x32 bit in
284 * regs->orig_ax, which changes the behavior of some syscalls.
286 if (likely((nr & __SYSCALL_MASK) < NR_syscalls)) {
287 regs->ax = sys_call_table[nr & __SYSCALL_MASK](
288 regs->di, regs->si, regs->dx,
289 regs->r10, regs->r8, regs->r9);
292 syscall_return_slowpath(regs);
296 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
298 * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does
299 * all entry and exit work and returns with IRQs off. This function is
300 * extremely hot in workloads that use it, and it's usually called from
301 * do_fast_syscall_32, so forcibly inline it to improve performance.
303 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
305 struct thread_info *ti = current_thread_info();
306 unsigned int nr = (unsigned int)regs->orig_ax;
308 #ifdef CONFIG_IA32_EMULATION
309 current->thread.status |= TS_COMPAT;
312 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
314 * Subtlety here: if ptrace pokes something larger than
315 * 2^32-1 into orig_ax, this truncates it. This may or
316 * may not be necessary, but it matches the old asm
319 nr = syscall_trace_enter(regs);
322 if (likely(nr < IA32_NR_syscalls)) {
324 * It's possible that a 32-bit syscall implementation
325 * takes a 64-bit parameter but nonetheless assumes that
326 * the high bits are zero. Make sure we zero-extend all
329 regs->ax = ia32_sys_call_table[nr](
330 (unsigned int)regs->bx, (unsigned int)regs->cx,
331 (unsigned int)regs->dx, (unsigned int)regs->si,
332 (unsigned int)regs->di, (unsigned int)regs->bp);
335 syscall_return_slowpath(regs);
338 /* Handles int $0x80 */
339 __visible void do_int80_syscall_32(struct pt_regs *regs)
341 enter_from_user_mode();
343 do_syscall_32_irqs_on(regs);
346 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
347 __visible long do_fast_syscall_32(struct pt_regs *regs)
350 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
351 * convention. Adjust regs so it looks like we entered using int80.
354 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
355 vdso_image_32.sym_int80_landing_pad;
358 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
359 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
362 regs->ip = landing_pad;
364 enter_from_user_mode();
368 /* Fetch EBP from where the vDSO stashed it. */
372 * Micro-optimization: the pointer we're following is explicitly
373 * 32 bits, so it can't be out of range.
375 __get_user(*(u32 *)®s->bp,
376 (u32 __user __force *)(unsigned long)(u32)regs->sp)
378 get_user(*(u32 *)®s->bp,
379 (u32 __user __force *)(unsigned long)(u32)regs->sp)
383 /* User code screwed up. */
386 prepare_exit_to_usermode(regs);
387 return 0; /* Keep it simple: use IRET. */
390 /* Now this is just like a normal syscall. */
391 do_syscall_32_irqs_on(regs);
395 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
396 * SYSRETL is available on all 64-bit CPUs, so we don't need to
397 * bother with SYSEXIT.
399 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
400 * because the ECX fixup above will ensure that this is essentially
403 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
404 regs->ip == landing_pad &&
405 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
408 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
410 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
411 * because the ECX fixup above will ensure that this is essentially
414 * We don't allow syscalls at all from VM86 mode, but we still
415 * need to check VM, because we might be returning from sys_vm86.
417 return static_cpu_has(X86_FEATURE_SEP) &&
418 regs->cs == __USER_CS && regs->ss == __USER_DS &&
419 regs->ip == landing_pad &&
420 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;