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
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
= pt_regs_to_thread_info(regs
);
75 unsigned long ret
= 0;
76 bool emulated
= false;
79 if (IS_ENABLED(CONFIG_DEBUG_ENTRY
))
80 BUG_ON(regs
!= task_pt_regs(current
));
82 work
= ACCESS_ONCE(ti
->flags
) & _TIF_WORK_SYSCALL_ENTRY
;
84 if (unlikely(work
& _TIF_SYSCALL_EMU
))
87 if ((emulated
|| (work
& _TIF_SYSCALL_TRACE
)) &&
88 tracehook_report_syscall_entry(regs
))
96 * Do seccomp after ptrace, to catch any tracer changes.
98 if (work
& _TIF_SECCOMP
) {
99 struct seccomp_data sd
;
102 sd
.nr
= regs
->orig_ax
;
103 sd
.instruction_pointer
= regs
->ip
;
105 if (arch
== AUDIT_ARCH_X86_64
) {
106 sd
.args
[0] = regs
->di
;
107 sd
.args
[1] = regs
->si
;
108 sd
.args
[2] = regs
->dx
;
109 sd
.args
[3] = regs
->r10
;
110 sd
.args
[4] = regs
->r8
;
111 sd
.args
[5] = regs
->r9
;
115 sd
.args
[0] = regs
->bx
;
116 sd
.args
[1] = regs
->cx
;
117 sd
.args
[2] = regs
->dx
;
118 sd
.args
[3] = regs
->si
;
119 sd
.args
[4] = regs
->di
;
120 sd
.args
[5] = regs
->bp
;
123 ret
= __secure_computing(&sd
);
129 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT
)))
130 trace_sys_enter(regs
, regs
->orig_ax
);
132 do_audit_syscall_entry(regs
, arch
);
134 return ret
?: regs
->orig_ax
;
137 #define EXIT_TO_USERMODE_LOOP_FLAGS \
138 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
139 _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY)
141 static void exit_to_usermode_loop(struct pt_regs
*regs
, u32 cached_flags
)
144 * In order to return to user mode, we need to have IRQs off with
145 * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
146 * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags
147 * can be set at any time on preemptable kernels if we have IRQs on,
148 * so we need to loop. Disabling preemption wouldn't help: doing the
149 * work to clear some of the flags can sleep.
152 /* We have work to do. */
155 if (cached_flags
& _TIF_NEED_RESCHED
)
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
);
168 rseq_handle_notify_resume(regs
);
171 if (cached_flags
& _TIF_USER_RETURN_NOTIFY
)
172 fire_user_return_notifiers();
174 /* Disable IRQs and retry */
177 cached_flags
= READ_ONCE(pt_regs_to_thread_info(regs
)->flags
);
179 if (!(cached_flags
& EXIT_TO_USERMODE_LOOP_FLAGS
))
185 /* Called with IRQs disabled. */
186 __visible
inline void prepare_exit_to_usermode(struct pt_regs
*regs
)
188 struct thread_info
*ti
= pt_regs_to_thread_info(regs
);
191 if (IS_ENABLED(CONFIG_PROVE_LOCKING
) && WARN_ON(!irqs_disabled()))
196 cached_flags
= READ_ONCE(ti
->flags
);
198 if (unlikely(cached_flags
& EXIT_TO_USERMODE_LOOP_FLAGS
))
199 exit_to_usermode_loop(regs
, cached_flags
);
203 * Compat syscalls set TS_COMPAT. Make sure we clear it before
204 * returning to user mode. We need to clear it *after* signal
205 * handling, because syscall restart has a fixup for compat
206 * syscalls. The fixup is exercised by the ptrace_syscall_32
209 * We also need to clear TS_REGS_POKED_I386: the 32-bit tracer
210 * special case only applies after poking regs and before the
211 * very next return to user mode.
213 ti
->status
&= ~(TS_COMPAT
|TS_I386_REGS_POKED
);
219 #define SYSCALL_EXIT_WORK_FLAGS \
220 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
221 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
223 static void syscall_slow_exit_work(struct pt_regs
*regs
, u32 cached_flags
)
227 audit_syscall_exit(regs
);
229 if (cached_flags
& _TIF_SYSCALL_TRACEPOINT
)
230 trace_sys_exit(regs
, regs
->ax
);
233 * If TIF_SYSCALL_EMU is set, we only get here because of
234 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
235 * We already reported this syscall instruction in
236 * syscall_trace_enter().
239 (cached_flags
& (_TIF_SINGLESTEP
| _TIF_SYSCALL_EMU
))
241 if (step
|| cached_flags
& _TIF_SYSCALL_TRACE
)
242 tracehook_report_syscall_exit(regs
, step
);
246 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
247 * state such that we can immediately switch to user mode.
249 __visible
inline void syscall_return_slowpath(struct pt_regs
*regs
)
251 struct thread_info
*ti
= pt_regs_to_thread_info(regs
);
252 u32 cached_flags
= READ_ONCE(ti
->flags
);
254 CT_WARN_ON(ct_state() != CONTEXT_KERNEL
);
256 if (IS_ENABLED(CONFIG_PROVE_LOCKING
) &&
257 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs
->orig_ax
))
261 * First do one-time work. If these work items are enabled, we
262 * want to run them exactly once per syscall exit with IRQs on.
264 if (unlikely(cached_flags
& SYSCALL_EXIT_WORK_FLAGS
))
265 syscall_slow_exit_work(regs
, cached_flags
);
268 prepare_exit_to_usermode(regs
);
272 __visible
void do_syscall_64(struct pt_regs
*regs
)
274 struct thread_info
*ti
= pt_regs_to_thread_info(regs
);
275 unsigned long nr
= regs
->orig_ax
;
277 enter_from_user_mode();
280 if (READ_ONCE(ti
->flags
) & _TIF_WORK_SYSCALL_ENTRY
)
281 nr
= syscall_trace_enter(regs
);
284 * NB: Native and x32 syscalls are dispatched from the same
285 * table. The only functional difference is the x32 bit in
286 * regs->orig_ax, which changes the behavior of some syscalls.
288 if (likely((nr
& __SYSCALL_MASK
) < NR_syscalls
)) {
289 regs
->ax
= sys_call_table
[nr
& __SYSCALL_MASK
](
290 regs
->di
, regs
->si
, regs
->dx
,
291 regs
->r10
, regs
->r8
, regs
->r9
);
294 syscall_return_slowpath(regs
);
298 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
300 * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does
301 * all entry and exit work and returns with IRQs off. This function is
302 * extremely hot in workloads that use it, and it's usually called from
303 * do_fast_syscall_32, so forcibly inline it to improve performance.
305 static __always_inline
void do_syscall_32_irqs_on(struct pt_regs
*regs
)
307 struct thread_info
*ti
= pt_regs_to_thread_info(regs
);
308 unsigned int nr
= (unsigned int)regs
->orig_ax
;
310 #ifdef CONFIG_IA32_EMULATION
311 ti
->status
|= TS_COMPAT
;
314 if (READ_ONCE(ti
->flags
) & _TIF_WORK_SYSCALL_ENTRY
) {
316 * Subtlety here: if ptrace pokes something larger than
317 * 2^32-1 into orig_ax, this truncates it. This may or
318 * may not be necessary, but it matches the old asm
321 nr
= syscall_trace_enter(regs
);
324 if (likely(nr
< IA32_NR_syscalls
)) {
326 * It's possible that a 32-bit syscall implementation
327 * takes a 64-bit parameter but nonetheless assumes that
328 * the high bits are zero. Make sure we zero-extend all
331 regs
->ax
= ia32_sys_call_table
[nr
](
332 (unsigned int)regs
->bx
, (unsigned int)regs
->cx
,
333 (unsigned int)regs
->dx
, (unsigned int)regs
->si
,
334 (unsigned int)regs
->di
, (unsigned int)regs
->bp
);
337 syscall_return_slowpath(regs
);
340 /* Handles int $0x80 */
341 __visible
void do_int80_syscall_32(struct pt_regs
*regs
)
343 enter_from_user_mode();
345 do_syscall_32_irqs_on(regs
);
348 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
349 __visible
long do_fast_syscall_32(struct pt_regs
*regs
)
352 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
353 * convention. Adjust regs so it looks like we entered using int80.
356 unsigned long landing_pad
= (unsigned long)current
->mm
->context
.vdso
+
357 vdso_image_32
.sym_int80_landing_pad
;
360 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
361 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
364 regs
->ip
= landing_pad
;
366 enter_from_user_mode();
370 /* Fetch EBP from where the vDSO stashed it. */
374 * Micro-optimization: the pointer we're following is explicitly
375 * 32 bits, so it can't be out of range.
377 __get_user(*(u32
*)®s
->bp
,
378 (u32 __user __force
*)(unsigned long)(u32
)regs
->sp
)
380 get_user(*(u32
*)®s
->bp
,
381 (u32 __user __force
*)(unsigned long)(u32
)regs
->sp
)
385 /* User code screwed up. */
388 prepare_exit_to_usermode(regs
);
389 return 0; /* Keep it simple: use IRET. */
392 /* Now this is just like a normal syscall. */
393 do_syscall_32_irqs_on(regs
);
397 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
398 * SYSRETL is available on all 64-bit CPUs, so we don't need to
399 * bother with SYSEXIT.
401 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
402 * because the ECX fixup above will ensure that this is essentially
405 return regs
->cs
== __USER32_CS
&& regs
->ss
== __USER_DS
&&
406 regs
->ip
== landing_pad
&&
407 (regs
->flags
& (X86_EFLAGS_RF
| X86_EFLAGS_TF
)) == 0;
410 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
412 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
413 * because the ECX fixup above will ensure that this is essentially
416 * We don't allow syscalls at all from VM86 mode, but we still
417 * need to check VM, because we might be returning from sys_vm86.
419 return static_cpu_has(X86_FEATURE_SEP
) &&
420 regs
->cs
== __USER_CS
&& regs
->ss
== __USER_DS
&&
421 regs
->ip
== landing_pad
&&
422 (regs
->flags
& (X86_EFLAGS_RF
| X86_EFLAGS_TF
| X86_EFLAGS_VM
)) == 0;