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 static struct thread_info *pt_regs_to_thread_info(struct pt_regs *regs)
36 unsigned long top_of_stack =
37 (unsigned long)(regs + 1) + TOP_OF_KERNEL_STACK_PADDING;
38 return (struct thread_info *)(top_of_stack - THREAD_SIZE);
41 #ifdef CONFIG_CONTEXT_TRACKING
42 /* Called on entry from user mode with IRQs off. */
43 __visible void enter_from_user_mode(void)
45 CT_WARN_ON(ct_state() != CONTEXT_USER);
49 static inline void enter_from_user_mode(void) {}
52 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
55 if (arch == AUDIT_ARCH_X86_64) {
56 audit_syscall_entry(regs->orig_ax, regs->di,
57 regs->si, regs->dx, regs->r10);
61 audit_syscall_entry(regs->orig_ax, regs->bx,
62 regs->cx, regs->dx, regs->si);
67 * Returns the syscall nr to run (which should match regs->orig_ax) or -1
68 * to skip the syscall.
70 static long syscall_trace_enter(struct pt_regs *regs)
72 u32 arch = in_ia32_syscall() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
74 struct thread_info *ti = pt_regs_to_thread_info(regs);
75 unsigned long ret = 0;
78 if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
79 BUG_ON(regs != task_pt_regs(current));
81 work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
85 * Do seccomp first -- it should minimize exposure of other
86 * code, and keeping seccomp fast is probably more valuable
87 * than the rest of this.
89 if (work & _TIF_SECCOMP) {
90 struct seccomp_data sd;
93 sd.nr = regs->orig_ax;
94 sd.instruction_pointer = regs->ip;
96 if (arch == AUDIT_ARCH_X86_64) {
97 sd.args[0] = regs->di;
98 sd.args[1] = regs->si;
99 sd.args[2] = regs->dx;
100 sd.args[3] = regs->r10;
101 sd.args[4] = regs->r8;
102 sd.args[5] = regs->r9;
106 sd.args[0] = regs->bx;
107 sd.args[1] = regs->cx;
108 sd.args[2] = regs->dx;
109 sd.args[3] = regs->si;
110 sd.args[4] = regs->di;
111 sd.args[5] = regs->bp;
114 ret = __secure_computing(&sd);
120 if (unlikely(work & _TIF_SYSCALL_EMU))
123 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
124 tracehook_report_syscall_entry(regs))
127 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
128 trace_sys_enter(regs, regs->orig_ax);
130 do_audit_syscall_entry(regs, arch);
132 return ret ?: regs->orig_ax;
135 #define EXIT_TO_USERMODE_LOOP_FLAGS \
136 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
137 _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY)
139 static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
142 * In order to return to user mode, we need to have IRQs off with
143 * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
144 * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags
145 * can be set at any time on preemptable kernels if we have IRQs on,
146 * so we need to loop. Disabling preemption wouldn't help: doing the
147 * work to clear some of the flags can sleep.
150 /* We have work to do. */
153 if (cached_flags & _TIF_NEED_RESCHED)
156 if (cached_flags & _TIF_UPROBE)
157 uprobe_notify_resume(regs);
159 /* deal with pending signal delivery */
160 if (cached_flags & _TIF_SIGPENDING)
163 if (cached_flags & _TIF_NOTIFY_RESUME) {
164 clear_thread_flag(TIF_NOTIFY_RESUME);
165 tracehook_notify_resume(regs);
168 if (cached_flags & _TIF_USER_RETURN_NOTIFY)
169 fire_user_return_notifiers();
171 /* Disable IRQs and retry */
174 cached_flags = READ_ONCE(pt_regs_to_thread_info(regs)->flags);
176 if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
182 /* Called with IRQs disabled. */
183 __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
185 struct thread_info *ti = pt_regs_to_thread_info(regs);
188 if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
193 cached_flags = READ_ONCE(ti->flags);
195 if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
196 exit_to_usermode_loop(regs, cached_flags);
200 * Compat syscalls set TS_COMPAT. Make sure we clear it before
201 * returning to user mode. We need to clear it *after* signal
202 * handling, because syscall restart has a fixup for compat
203 * syscalls. The fixup is exercised by the ptrace_syscall_32
206 ti->status &= ~TS_COMPAT;
212 #define SYSCALL_EXIT_WORK_FLAGS \
213 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
214 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
216 static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags)
220 audit_syscall_exit(regs);
222 if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
223 trace_sys_exit(regs, regs->ax);
226 * If TIF_SYSCALL_EMU is set, we only get here because of
227 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
228 * We already reported this syscall instruction in
229 * syscall_trace_enter().
232 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
234 if (step || cached_flags & _TIF_SYSCALL_TRACE)
235 tracehook_report_syscall_exit(regs, step);
239 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
240 * state such that we can immediately switch to user mode.
242 __visible inline void syscall_return_slowpath(struct pt_regs *regs)
244 struct thread_info *ti = pt_regs_to_thread_info(regs);
245 u32 cached_flags = READ_ONCE(ti->flags);
247 CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
249 if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
250 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
254 * First do one-time work. If these work items are enabled, we
255 * want to run them exactly once per syscall exit with IRQs on.
257 if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS))
258 syscall_slow_exit_work(regs, cached_flags);
261 prepare_exit_to_usermode(regs);
265 __visible void do_syscall_64(struct pt_regs *regs)
267 struct thread_info *ti = pt_regs_to_thread_info(regs);
268 unsigned long nr = regs->orig_ax;
270 enter_from_user_mode();
273 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
274 nr = syscall_trace_enter(regs);
277 * NB: Native and x32 syscalls are dispatched from the same
278 * table. The only functional difference is the x32 bit in
279 * regs->orig_ax, which changes the behavior of some syscalls.
281 if (likely((nr & __SYSCALL_MASK) < NR_syscalls)) {
282 regs->ax = sys_call_table[nr & __SYSCALL_MASK](
283 regs->di, regs->si, regs->dx,
284 regs->r10, regs->r8, regs->r9);
287 syscall_return_slowpath(regs);
291 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
293 * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does
294 * all entry and exit work and returns with IRQs off. This function is
295 * extremely hot in workloads that use it, and it's usually called from
296 * do_fast_syscall_32, so forcibly inline it to improve performance.
298 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
300 struct thread_info *ti = pt_regs_to_thread_info(regs);
301 unsigned int nr = (unsigned int)regs->orig_ax;
303 #ifdef CONFIG_IA32_EMULATION
304 ti->status |= TS_COMPAT;
307 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
309 * Subtlety here: if ptrace pokes something larger than
310 * 2^32-1 into orig_ax, this truncates it. This may or
311 * may not be necessary, but it matches the old asm
314 nr = syscall_trace_enter(regs);
317 if (likely(nr < IA32_NR_syscalls)) {
319 * It's possible that a 32-bit syscall implementation
320 * takes a 64-bit parameter but nonetheless assumes that
321 * the high bits are zero. Make sure we zero-extend all
324 regs->ax = ia32_sys_call_table[nr](
325 (unsigned int)regs->bx, (unsigned int)regs->cx,
326 (unsigned int)regs->dx, (unsigned int)regs->si,
327 (unsigned int)regs->di, (unsigned int)regs->bp);
330 syscall_return_slowpath(regs);
333 /* Handles int $0x80 */
334 __visible void do_int80_syscall_32(struct pt_regs *regs)
336 enter_from_user_mode();
338 do_syscall_32_irqs_on(regs);
341 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
342 __visible long do_fast_syscall_32(struct pt_regs *regs)
345 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
346 * convention. Adjust regs so it looks like we entered using int80.
349 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
350 vdso_image_32.sym_int80_landing_pad;
353 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
354 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
357 regs->ip = landing_pad;
359 enter_from_user_mode();
363 /* Fetch EBP from where the vDSO stashed it. */
367 * Micro-optimization: the pointer we're following is explicitly
368 * 32 bits, so it can't be out of range.
370 __get_user(*(u32 *)®s->bp,
371 (u32 __user __force *)(unsigned long)(u32)regs->sp)
373 get_user(*(u32 *)®s->bp,
374 (u32 __user __force *)(unsigned long)(u32)regs->sp)
378 /* User code screwed up. */
381 prepare_exit_to_usermode(regs);
382 return 0; /* Keep it simple: use IRET. */
385 /* Now this is just like a normal syscall. */
386 do_syscall_32_irqs_on(regs);
390 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
391 * SYSRETL is available on all 64-bit CPUs, so we don't need to
392 * bother with SYSEXIT.
394 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
395 * because the ECX fixup above will ensure that this is essentially
398 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
399 regs->ip == landing_pad &&
400 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
403 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
405 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
406 * because the ECX fixup above will ensure that this is essentially
409 * We don't allow syscalls at all from VM86 mode, but we still
410 * need to check VM, because we might be returning from sys_vm86.
412 return static_cpu_has(X86_FEATURE_SEP) &&
413 regs->cs == __USER_CS && regs->ss == __USER_DS &&
414 regs->ip == landing_pad &&
415 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;