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/nospec.h>
24 #include <linux/uprobes.h>
27 #include <asm/traps.h>
29 #include <asm/uaccess.h>
30 #include <asm/cpufeature.h>
31 #include <asm/nospec-branch.h>
33 #define CREATE_TRACE_POINTS
34 #include <trace/events/syscalls.h>
36 #ifdef CONFIG_CONTEXT_TRACKING
37 /* Called on entry from user mode with IRQs off. */
38 __visible inline void enter_from_user_mode(void)
40 CT_WARN_ON(ct_state() != CONTEXT_USER);
44 static inline void enter_from_user_mode(void) {}
47 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
50 if (arch == AUDIT_ARCH_X86_64) {
51 audit_syscall_entry(regs->orig_ax, regs->di,
52 regs->si, regs->dx, regs->r10);
56 audit_syscall_entry(regs->orig_ax, regs->bx,
57 regs->cx, regs->dx, regs->si);
62 * Returns the syscall nr to run (which should match regs->orig_ax) or -1
63 * to skip the syscall.
65 static long syscall_trace_enter(struct pt_regs *regs)
67 u32 arch = in_ia32_syscall() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
69 struct thread_info *ti = current_thread_info();
70 unsigned long ret = 0;
71 bool emulated = false;
74 if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
75 BUG_ON(regs != task_pt_regs(current));
77 work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
79 if (unlikely(work & _TIF_SYSCALL_EMU))
82 if ((emulated || (work & _TIF_SYSCALL_TRACE)) &&
83 tracehook_report_syscall_entry(regs))
91 * Do seccomp after ptrace, to catch any tracer changes.
93 if (work & _TIF_SECCOMP) {
94 struct seccomp_data sd;
97 sd.nr = regs->orig_ax;
98 sd.instruction_pointer = regs->ip;
100 if (arch == AUDIT_ARCH_X86_64) {
101 sd.args[0] = regs->di;
102 sd.args[1] = regs->si;
103 sd.args[2] = regs->dx;
104 sd.args[3] = regs->r10;
105 sd.args[4] = regs->r8;
106 sd.args[5] = regs->r9;
110 sd.args[0] = regs->bx;
111 sd.args[1] = regs->cx;
112 sd.args[2] = regs->dx;
113 sd.args[3] = regs->si;
114 sd.args[4] = regs->di;
115 sd.args[5] = regs->bp;
118 ret = __secure_computing(&sd);
124 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
125 trace_sys_enter(regs, regs->orig_ax);
127 do_audit_syscall_entry(regs, arch);
129 return ret ?: regs->orig_ax;
132 #define EXIT_TO_USERMODE_LOOP_FLAGS \
133 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
134 _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY)
136 static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
139 * In order to return to user mode, we need to have IRQs off with
140 * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
141 * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags
142 * can be set at any time on preemptable kernels if we have IRQs on,
143 * so we need to loop. Disabling preemption wouldn't help: doing the
144 * work to clear some of the flags can sleep.
147 /* We have work to do. */
150 if (cached_flags & _TIF_NEED_RESCHED)
153 if (cached_flags & _TIF_UPROBE)
154 uprobe_notify_resume(regs);
156 /* deal with pending signal delivery */
157 if (cached_flags & _TIF_SIGPENDING)
160 if (cached_flags & _TIF_NOTIFY_RESUME) {
161 clear_thread_flag(TIF_NOTIFY_RESUME);
162 tracehook_notify_resume(regs);
165 if (cached_flags & _TIF_USER_RETURN_NOTIFY)
166 fire_user_return_notifiers();
168 /* Disable IRQs and retry */
171 cached_flags = READ_ONCE(current_thread_info()->flags);
173 if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
178 /* Called with IRQs disabled. */
179 __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
181 struct thread_info *ti = current_thread_info();
184 if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
189 cached_flags = READ_ONCE(ti->flags);
191 if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
192 exit_to_usermode_loop(regs, cached_flags);
196 * Compat syscalls set TS_COMPAT. Make sure we clear it before
197 * returning to user mode. We need to clear it *after* signal
198 * handling, because syscall restart has a fixup for compat
199 * syscalls. The fixup is exercised by the ptrace_syscall_32
202 * We also need to clear TS_REGS_POKED_I386: the 32-bit tracer
203 * special case only applies after poking regs and before the
204 * very next return to user mode.
206 ti->status &= ~(TS_COMPAT|TS_I386_REGS_POKED);
211 mds_user_clear_cpu_buffers();
214 #define SYSCALL_EXIT_WORK_FLAGS \
215 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
216 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
218 static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags)
222 audit_syscall_exit(regs);
224 if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
225 trace_sys_exit(regs, regs->ax);
228 * If TIF_SYSCALL_EMU is set, we only get here because of
229 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
230 * We already reported this syscall instruction in
231 * syscall_trace_enter().
234 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
236 if (step || cached_flags & _TIF_SYSCALL_TRACE)
237 tracehook_report_syscall_exit(regs, step);
241 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
242 * state such that we can immediately switch to user mode.
244 __visible inline void syscall_return_slowpath(struct pt_regs *regs)
246 struct thread_info *ti = current_thread_info();
247 u32 cached_flags = READ_ONCE(ti->flags);
249 CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
251 if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
252 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
256 * First do one-time work. If these work items are enabled, we
257 * want to run them exactly once per syscall exit with IRQs on.
259 if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS))
260 syscall_slow_exit_work(regs, cached_flags);
263 prepare_exit_to_usermode(regs);
267 __visible void do_syscall_64(struct pt_regs *regs)
269 struct thread_info *ti = current_thread_info();
270 unsigned long nr = regs->orig_ax;
272 enter_from_user_mode();
275 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
276 nr = syscall_trace_enter(regs);
279 * NB: Native and x32 syscalls are dispatched from the same
280 * table. The only functional difference is the x32 bit in
281 * regs->orig_ax, which changes the behavior of some syscalls.
283 if (likely((nr & __SYSCALL_MASK) < NR_syscalls)) {
284 nr = array_index_nospec(nr & __SYSCALL_MASK, NR_syscalls);
285 regs->ax = sys_call_table[nr](
286 regs->di, regs->si, regs->dx,
287 regs->r10, regs->r8, regs->r9);
290 syscall_return_slowpath(regs);
294 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
296 * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does
297 * all entry and exit work and returns with IRQs off. This function is
298 * extremely hot in workloads that use it, and it's usually called from
299 * do_fast_syscall_32, so forcibly inline it to improve performance.
301 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
303 struct thread_info *ti = current_thread_info();
304 unsigned int nr = (unsigned int)regs->orig_ax;
306 #ifdef CONFIG_IA32_EMULATION
307 ti->status |= TS_COMPAT;
310 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
312 * Subtlety here: if ptrace pokes something larger than
313 * 2^32-1 into orig_ax, this truncates it. This may or
314 * may not be necessary, but it matches the old asm
317 nr = syscall_trace_enter(regs);
320 if (likely(nr < IA32_NR_syscalls)) {
321 nr = array_index_nospec(nr, IA32_NR_syscalls);
323 * It's possible that a 32-bit syscall implementation
324 * takes a 64-bit parameter but nonetheless assumes that
325 * the high bits are zero. Make sure we zero-extend all
328 regs->ax = ia32_sys_call_table[nr](
329 (unsigned int)regs->bx, (unsigned int)regs->cx,
330 (unsigned int)regs->dx, (unsigned int)regs->si,
331 (unsigned int)regs->di, (unsigned int)regs->bp);
334 syscall_return_slowpath(regs);
337 /* Handles int $0x80 */
338 __visible void do_int80_syscall_32(struct pt_regs *regs)
340 enter_from_user_mode();
342 do_syscall_32_irqs_on(regs);
345 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
346 __visible long do_fast_syscall_32(struct pt_regs *regs)
349 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
350 * convention. Adjust regs so it looks like we entered using int80.
353 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
354 vdso_image_32.sym_int80_landing_pad;
357 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
358 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
361 regs->ip = landing_pad;
363 enter_from_user_mode();
367 /* Fetch EBP from where the vDSO stashed it. */
371 * Micro-optimization: the pointer we're following is explicitly
372 * 32 bits, so it can't be out of range.
374 __get_user(*(u32 *)®s->bp,
375 (u32 __user __force *)(unsigned long)(u32)regs->sp)
377 get_user(*(u32 *)®s->bp,
378 (u32 __user __force *)(unsigned long)(u32)regs->sp)
382 /* User code screwed up. */
385 prepare_exit_to_usermode(regs);
386 return 0; /* Keep it simple: use IRET. */
389 /* Now this is just like a normal syscall. */
390 do_syscall_32_irqs_on(regs);
394 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
395 * SYSRETL is available on all 64-bit CPUs, so we don't need to
396 * bother with SYSEXIT.
398 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
399 * because the ECX fixup above will ensure that this is essentially
402 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
403 regs->ip == landing_pad &&
404 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
407 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
409 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
410 * because the ECX fixup above will ensure that this is essentially
413 * We don't allow syscalls at all from VM86 mode, but we still
414 * need to check VM, because we might be returning from sys_vm86.
416 return static_cpu_has(X86_FEATURE_SEP) &&
417 regs->cs == __USER_CS && regs->ss == __USER_DS &&
418 regs->ip == landing_pad &&
419 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;