1 // SPDX-License-Identifier: GPL-2.0-only
3 * common.c - C code for kernel entry and exit
4 * 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>
12 #include <linux/sched/task_stack.h>
14 #include <linux/smp.h>
15 #include <linux/errno.h>
16 #include <linux/ptrace.h>
17 #include <linux/tracehook.h>
18 #include <linux/audit.h>
19 #include <linux/seccomp.h>
20 #include <linux/signal.h>
21 #include <linux/export.h>
22 #include <linux/context_tracking.h>
23 #include <linux/user-return-notifier.h>
24 #include <linux/nospec.h>
25 #include <linux/uprobes.h>
26 #include <linux/livepatch.h>
27 #include <linux/syscalls.h>
28 #include <linux/uaccess.h>
31 #include <asm/traps.h>
33 #include <asm/cpufeature.h>
34 #include <asm/fpu/api.h>
35 #include <asm/nospec-branch.h>
36 #include <asm/io_bitmap.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/syscalls.h>
41 #ifdef CONFIG_CONTEXT_TRACKING
42 /* Called on entry from user mode with IRQs off. */
43 __visible inline 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 = current_thread_info();
75 unsigned long ret = 0;
78 if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
79 BUG_ON(regs != task_pt_regs(current));
81 work = READ_ONCE(ti->flags);
83 if (work & (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_EMU)) {
84 ret = tracehook_report_syscall_entry(regs);
85 if (ret || (work & _TIF_SYSCALL_EMU))
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 | _TIF_PATCH_PENDING)
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 EXIT_TO_USERMODE_LOOP_FLAGS set. Several of these flags
141 * can be set at any time on preemptible kernels if we have IRQs on,
142 * so we need to loop. Disabling preemption wouldn't help: doing the
143 * work to clear some of the flags can sleep.
146 /* We have work to do. */
149 if (cached_flags & _TIF_NEED_RESCHED)
152 if (cached_flags & _TIF_UPROBE)
153 uprobe_notify_resume(regs);
155 if (cached_flags & _TIF_PATCH_PENDING)
156 klp_update_patch_state(current);
158 /* deal with pending signal delivery */
159 if (cached_flags & _TIF_SIGPENDING)
162 if (cached_flags & _TIF_NOTIFY_RESUME) {
163 clear_thread_flag(TIF_NOTIFY_RESUME);
164 tracehook_notify_resume(regs);
165 rseq_handle_notify_resume(NULL, regs);
168 if (cached_flags & _TIF_USER_RETURN_NOTIFY)
169 fire_user_return_notifiers();
171 /* Disable IRQs and retry */
174 cached_flags = READ_ONCE(current_thread_info()->flags);
176 if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
181 /* Called with IRQs disabled. */
182 __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
184 struct thread_info *ti = current_thread_info();
187 addr_limit_user_check();
189 lockdep_assert_irqs_disabled();
192 cached_flags = READ_ONCE(ti->flags);
194 if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
195 exit_to_usermode_loop(regs, cached_flags);
197 /* Reload ti->flags; we may have rescheduled above. */
198 cached_flags = READ_ONCE(ti->flags);
200 if (unlikely(cached_flags & _TIF_IO_BITMAP))
201 tss_update_io_bitmap();
203 fpregs_assert_state_consistent();
204 if (unlikely(cached_flags & _TIF_NEED_FPU_LOAD))
209 * Compat syscalls set TS_COMPAT. Make sure we clear it before
210 * returning to user mode. We need to clear it *after* signal
211 * handling, because syscall restart has a fixup for compat
212 * syscalls. The fixup is exercised by the ptrace_syscall_32
215 * We also need to clear TS_REGS_POKED_I386: the 32-bit tracer
216 * special case only applies after poking regs and before the
217 * very next return to user mode.
219 ti->status &= ~(TS_COMPAT|TS_I386_REGS_POKED);
224 mds_user_clear_cpu_buffers();
227 #define SYSCALL_EXIT_WORK_FLAGS \
228 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
229 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
231 static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags)
235 audit_syscall_exit(regs);
237 if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
238 trace_sys_exit(regs, regs->ax);
241 * If TIF_SYSCALL_EMU is set, we only get here because of
242 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
243 * We already reported this syscall instruction in
244 * syscall_trace_enter().
247 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
249 if (step || cached_flags & _TIF_SYSCALL_TRACE)
250 tracehook_report_syscall_exit(regs, step);
254 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
255 * state such that we can immediately switch to user mode.
257 __visible inline void syscall_return_slowpath(struct pt_regs *regs)
259 struct thread_info *ti = current_thread_info();
260 u32 cached_flags = READ_ONCE(ti->flags);
262 CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
264 if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
265 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
271 * First do one-time work. If these work items are enabled, we
272 * want to run them exactly once per syscall exit with IRQs on.
274 if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS))
275 syscall_slow_exit_work(regs, cached_flags);
278 prepare_exit_to_usermode(regs);
282 __visible void do_syscall_64(unsigned long nr, struct pt_regs *regs)
284 struct thread_info *ti;
286 enter_from_user_mode();
288 ti = current_thread_info();
289 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
290 nr = syscall_trace_enter(regs);
292 if (likely(nr < NR_syscalls)) {
293 nr = array_index_nospec(nr, NR_syscalls);
294 regs->ax = sys_call_table[nr](regs);
295 #ifdef CONFIG_X86_X32_ABI
296 } else if (likely((nr & __X32_SYSCALL_BIT) &&
297 (nr & ~__X32_SYSCALL_BIT) < X32_NR_syscalls)) {
298 nr = array_index_nospec(nr & ~__X32_SYSCALL_BIT,
300 regs->ax = x32_sys_call_table[nr](regs);
304 syscall_return_slowpath(regs);
308 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
310 * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does
311 * all entry and exit work and returns with IRQs off. This function is
312 * extremely hot in workloads that use it, and it's usually called from
313 * do_fast_syscall_32, so forcibly inline it to improve performance.
315 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
317 struct thread_info *ti = current_thread_info();
318 unsigned int nr = (unsigned int)regs->orig_ax;
320 #ifdef CONFIG_IA32_EMULATION
321 ti->status |= TS_COMPAT;
324 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
326 * Subtlety here: if ptrace pokes something larger than
327 * 2^32-1 into orig_ax, this truncates it. This may or
328 * may not be necessary, but it matches the old asm
331 nr = syscall_trace_enter(regs);
334 if (likely(nr < IA32_NR_syscalls)) {
335 nr = array_index_nospec(nr, IA32_NR_syscalls);
336 #ifdef CONFIG_IA32_EMULATION
337 regs->ax = ia32_sys_call_table[nr](regs);
340 * It's possible that a 32-bit syscall implementation
341 * takes a 64-bit parameter but nonetheless assumes that
342 * the high bits are zero. Make sure we zero-extend all
345 regs->ax = ia32_sys_call_table[nr](
346 (unsigned int)regs->bx, (unsigned int)regs->cx,
347 (unsigned int)regs->dx, (unsigned int)regs->si,
348 (unsigned int)regs->di, (unsigned int)regs->bp);
349 #endif /* CONFIG_IA32_EMULATION */
352 syscall_return_slowpath(regs);
355 /* Handles int $0x80 */
356 __visible void do_int80_syscall_32(struct pt_regs *regs)
358 enter_from_user_mode();
360 do_syscall_32_irqs_on(regs);
363 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
364 __visible long do_fast_syscall_32(struct pt_regs *regs)
367 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
368 * convention. Adjust regs so it looks like we entered using int80.
371 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
372 vdso_image_32.sym_int80_landing_pad;
375 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
376 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
379 regs->ip = landing_pad;
381 enter_from_user_mode();
385 /* Fetch EBP from where the vDSO stashed it. */
389 * Micro-optimization: the pointer we're following is explicitly
390 * 32 bits, so it can't be out of range.
392 __get_user(*(u32 *)®s->bp,
393 (u32 __user __force *)(unsigned long)(u32)regs->sp)
395 get_user(*(u32 *)®s->bp,
396 (u32 __user __force *)(unsigned long)(u32)regs->sp)
400 /* User code screwed up. */
403 prepare_exit_to_usermode(regs);
404 return 0; /* Keep it simple: use IRET. */
407 /* Now this is just like a normal syscall. */
408 do_syscall_32_irqs_on(regs);
412 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
413 * SYSRETL is available on all 64-bit CPUs, so we don't need to
414 * bother with SYSEXIT.
416 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
417 * because the ECX fixup above will ensure that this is essentially
420 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
421 regs->ip == landing_pad &&
422 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
425 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
427 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
428 * because the ECX fixup above will ensure that this is essentially
431 * We don't allow syscalls at all from VM86 mode, but we still
432 * need to check VM, because we might be returning from sys_vm86.
434 return static_cpu_has(X86_FEATURE_SEP) &&
435 regs->cs == __USER_CS && regs->ss == __USER_DS &&
436 regs->ip == landing_pad &&
437 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;