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 * We can return 0 to resume the syscall or anything else to go to phase
68 * 2. If we resume the syscall, we need to put something appropriate in
71 * NB: We don't have full pt_regs here, but regs->orig_ax and regs->ax
72 * are fully functional.
74 * For phase 2's benefit, our return value is:
75 * 0: resume the syscall
76 * 1: go to phase 2; no seccomp phase 2 needed
77 * anything else: go to phase 2; pass return value to seccomp
79 unsigned long syscall_trace_enter_phase1(struct pt_regs
*regs
, u32 arch
)
81 struct thread_info
*ti
= pt_regs_to_thread_info(regs
);
82 unsigned long ret
= 0;
85 if (IS_ENABLED(CONFIG_DEBUG_ENTRY
))
86 BUG_ON(regs
!= task_pt_regs(current
));
88 work
= ACCESS_ONCE(ti
->flags
) & _TIF_WORK_SYSCALL_ENTRY
;
92 * Do seccomp first -- it should minimize exposure of other
93 * code, and keeping seccomp fast is probably more valuable
94 * than the rest of this.
96 if (work
& _TIF_SECCOMP
) {
97 struct seccomp_data sd
;
100 sd
.nr
= regs
->orig_ax
;
101 sd
.instruction_pointer
= regs
->ip
;
103 if (arch
== AUDIT_ARCH_X86_64
) {
104 sd
.args
[0] = regs
->di
;
105 sd
.args
[1] = regs
->si
;
106 sd
.args
[2] = regs
->dx
;
107 sd
.args
[3] = regs
->r10
;
108 sd
.args
[4] = regs
->r8
;
109 sd
.args
[5] = regs
->r9
;
113 sd
.args
[0] = regs
->bx
;
114 sd
.args
[1] = regs
->cx
;
115 sd
.args
[2] = regs
->dx
;
116 sd
.args
[3] = regs
->si
;
117 sd
.args
[4] = regs
->di
;
118 sd
.args
[5] = regs
->bp
;
121 BUILD_BUG_ON(SECCOMP_PHASE1_OK
!= 0);
122 BUILD_BUG_ON(SECCOMP_PHASE1_SKIP
!= 1);
124 ret
= seccomp_phase1(&sd
);
125 if (ret
== SECCOMP_PHASE1_SKIP
) {
128 } else if (ret
!= SECCOMP_PHASE1_OK
) {
129 return ret
; /* Go directly to phase 2 */
132 work
&= ~_TIF_SECCOMP
;
136 /* Do our best to finish without phase 2. */
138 return ret
; /* seccomp and/or nohz only (ret == 0 here) */
140 #ifdef CONFIG_AUDITSYSCALL
141 if (work
== _TIF_SYSCALL_AUDIT
) {
143 * If there is no more work to be done except auditing,
144 * then audit in phase 1. Phase 2 always audits, so, if
145 * we audit here, then we can't go on to phase 2.
147 do_audit_syscall_entry(regs
, arch
);
152 return 1; /* Something is enabled that we can't handle in phase 1 */
155 /* Returns the syscall nr to run (which should match regs->orig_ax). */
156 long syscall_trace_enter_phase2(struct pt_regs
*regs
, u32 arch
,
157 unsigned long phase1_result
)
159 struct thread_info
*ti
= pt_regs_to_thread_info(regs
);
161 u32 work
= ACCESS_ONCE(ti
->flags
) & _TIF_WORK_SYSCALL_ENTRY
;
163 if (IS_ENABLED(CONFIG_DEBUG_ENTRY
))
164 BUG_ON(regs
!= task_pt_regs(current
));
166 #ifdef CONFIG_SECCOMP
168 * Call seccomp_phase2 before running the other hooks so that
169 * they can see any changes made by a seccomp tracer.
171 if (phase1_result
> 1 && seccomp_phase2(phase1_result
)) {
172 /* seccomp failures shouldn't expose any additional code. */
177 if (unlikely(work
& _TIF_SYSCALL_EMU
))
180 if ((ret
|| test_thread_flag(TIF_SYSCALL_TRACE
)) &&
181 tracehook_report_syscall_entry(regs
))
184 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT
)))
185 trace_sys_enter(regs
, regs
->orig_ax
);
187 do_audit_syscall_entry(regs
, arch
);
189 return ret
?: regs
->orig_ax
;
192 long syscall_trace_enter(struct pt_regs
*regs
)
194 u32 arch
= is_ia32_task() ? AUDIT_ARCH_I386
: AUDIT_ARCH_X86_64
;
195 unsigned long phase1_result
= syscall_trace_enter_phase1(regs
, arch
);
197 if (phase1_result
== 0)
198 return regs
->orig_ax
;
200 return syscall_trace_enter_phase2(regs
, arch
, phase1_result
);
203 #define EXIT_TO_USERMODE_LOOP_FLAGS \
204 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
205 _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY)
207 static void exit_to_usermode_loop(struct pt_regs
*regs
, u32 cached_flags
)
210 * In order to return to user mode, we need to have IRQs off with
211 * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
212 * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags
213 * can be set at any time on preemptable kernels if we have IRQs on,
214 * so we need to loop. Disabling preemption wouldn't help: doing the
215 * work to clear some of the flags can sleep.
218 /* We have work to do. */
221 if (cached_flags
& _TIF_NEED_RESCHED
)
224 if (cached_flags
& _TIF_UPROBE
)
225 uprobe_notify_resume(regs
);
227 /* deal with pending signal delivery */
228 if (cached_flags
& _TIF_SIGPENDING
)
231 if (cached_flags
& _TIF_NOTIFY_RESUME
) {
232 clear_thread_flag(TIF_NOTIFY_RESUME
);
233 tracehook_notify_resume(regs
);
236 if (cached_flags
& _TIF_USER_RETURN_NOTIFY
)
237 fire_user_return_notifiers();
239 /* Disable IRQs and retry */
242 cached_flags
= READ_ONCE(pt_regs_to_thread_info(regs
)->flags
);
244 if (!(cached_flags
& EXIT_TO_USERMODE_LOOP_FLAGS
))
250 /* Called with IRQs disabled. */
251 __visible
inline void prepare_exit_to_usermode(struct pt_regs
*regs
)
253 struct thread_info
*ti
= pt_regs_to_thread_info(regs
);
256 if (IS_ENABLED(CONFIG_PROVE_LOCKING
) && WARN_ON(!irqs_disabled()))
261 cached_flags
= READ_ONCE(ti
->flags
);
263 if (unlikely(cached_flags
& EXIT_TO_USERMODE_LOOP_FLAGS
))
264 exit_to_usermode_loop(regs
, cached_flags
);
268 * Compat syscalls set TS_COMPAT. Make sure we clear it before
269 * returning to user mode. We need to clear it *after* signal
270 * handling, because syscall restart has a fixup for compat
271 * syscalls. The fixup is exercised by the ptrace_syscall_32
274 ti
->status
&= ~TS_COMPAT
;
280 #define SYSCALL_EXIT_WORK_FLAGS \
281 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
282 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
284 static void syscall_slow_exit_work(struct pt_regs
*regs
, u32 cached_flags
)
288 audit_syscall_exit(regs
);
290 if (cached_flags
& _TIF_SYSCALL_TRACEPOINT
)
291 trace_sys_exit(regs
, regs
->ax
);
294 * If TIF_SYSCALL_EMU is set, we only get here because of
295 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
296 * We already reported this syscall instruction in
297 * syscall_trace_enter().
300 (cached_flags
& (_TIF_SINGLESTEP
| _TIF_SYSCALL_EMU
))
302 if (step
|| cached_flags
& _TIF_SYSCALL_TRACE
)
303 tracehook_report_syscall_exit(regs
, step
);
307 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
308 * state such that we can immediately switch to user mode.
310 __visible
inline void syscall_return_slowpath(struct pt_regs
*regs
)
312 struct thread_info
*ti
= pt_regs_to_thread_info(regs
);
313 u32 cached_flags
= READ_ONCE(ti
->flags
);
315 CT_WARN_ON(ct_state() != CONTEXT_KERNEL
);
317 if (IS_ENABLED(CONFIG_PROVE_LOCKING
) &&
318 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs
->orig_ax
))
322 * First do one-time work. If these work items are enabled, we
323 * want to run them exactly once per syscall exit with IRQs on.
325 if (unlikely(cached_flags
& SYSCALL_EXIT_WORK_FLAGS
))
326 syscall_slow_exit_work(regs
, cached_flags
);
329 prepare_exit_to_usermode(regs
);
333 __visible
void do_syscall_64(struct pt_regs
*regs
)
335 struct thread_info
*ti
= pt_regs_to_thread_info(regs
);
336 unsigned long nr
= regs
->orig_ax
;
338 enter_from_user_mode();
341 if (READ_ONCE(ti
->flags
) & _TIF_WORK_SYSCALL_ENTRY
)
342 nr
= syscall_trace_enter(regs
);
345 * NB: Native and x32 syscalls are dispatched from the same
346 * table. The only functional difference is the x32 bit in
347 * regs->orig_ax, which changes the behavior of some syscalls.
349 if (likely((nr
& __SYSCALL_MASK
) < NR_syscalls
)) {
350 regs
->ax
= sys_call_table
[nr
& __SYSCALL_MASK
](
351 regs
->di
, regs
->si
, regs
->dx
,
352 regs
->r10
, regs
->r8
, regs
->r9
);
355 syscall_return_slowpath(regs
);
359 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
361 * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does
362 * all entry and exit work and returns with IRQs off. This function is
363 * extremely hot in workloads that use it, and it's usually called from
364 * do_fast_syscall_32, so forcibly inline it to improve performance.
366 static __always_inline
void do_syscall_32_irqs_on(struct pt_regs
*regs
)
368 struct thread_info
*ti
= pt_regs_to_thread_info(regs
);
369 unsigned int nr
= (unsigned int)regs
->orig_ax
;
371 #ifdef CONFIG_IA32_EMULATION
372 ti
->status
|= TS_COMPAT
;
375 if (READ_ONCE(ti
->flags
) & _TIF_WORK_SYSCALL_ENTRY
) {
377 * Subtlety here: if ptrace pokes something larger than
378 * 2^32-1 into orig_ax, this truncates it. This may or
379 * may not be necessary, but it matches the old asm
382 nr
= syscall_trace_enter(regs
);
385 if (likely(nr
< IA32_NR_syscalls
)) {
387 * It's possible that a 32-bit syscall implementation
388 * takes a 64-bit parameter but nonetheless assumes that
389 * the high bits are zero. Make sure we zero-extend all
392 regs
->ax
= ia32_sys_call_table
[nr
](
393 (unsigned int)regs
->bx
, (unsigned int)regs
->cx
,
394 (unsigned int)regs
->dx
, (unsigned int)regs
->si
,
395 (unsigned int)regs
->di
, (unsigned int)regs
->bp
);
398 syscall_return_slowpath(regs
);
401 /* Handles int $0x80 */
402 __visible
void do_int80_syscall_32(struct pt_regs
*regs
)
404 enter_from_user_mode();
406 do_syscall_32_irqs_on(regs
);
409 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
410 __visible
long do_fast_syscall_32(struct pt_regs
*regs
)
413 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
414 * convention. Adjust regs so it looks like we entered using int80.
417 unsigned long landing_pad
= (unsigned long)current
->mm
->context
.vdso
+
418 vdso_image_32
.sym_int80_landing_pad
;
421 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
422 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
425 regs
->ip
= landing_pad
;
427 enter_from_user_mode();
431 /* Fetch EBP from where the vDSO stashed it. */
435 * Micro-optimization: the pointer we're following is explicitly
436 * 32 bits, so it can't be out of range.
438 __get_user(*(u32
*)®s
->bp
,
439 (u32 __user __force
*)(unsigned long)(u32
)regs
->sp
)
441 get_user(*(u32
*)®s
->bp
,
442 (u32 __user __force
*)(unsigned long)(u32
)regs
->sp
)
446 /* User code screwed up. */
449 prepare_exit_to_usermode(regs
);
450 return 0; /* Keep it simple: use IRET. */
453 /* Now this is just like a normal syscall. */
454 do_syscall_32_irqs_on(regs
);
458 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
459 * SYSRETL is available on all 64-bit CPUs, so we don't need to
460 * bother with SYSEXIT.
462 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
463 * because the ECX fixup above will ensure that this is essentially
466 return regs
->cs
== __USER32_CS
&& regs
->ss
== __USER_DS
&&
467 regs
->ip
== landing_pad
&&
468 (regs
->flags
& (X86_EFLAGS_RF
| X86_EFLAGS_TF
)) == 0;
471 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
473 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
474 * because the ECX fixup above will ensure that this is essentially
477 * We don't allow syscalls at all from VM86 mode, but we still
478 * need to check VM, because we might be returning from sys_vm86.
480 return static_cpu_has(X86_FEATURE_SEP
) &&
481 regs
->cs
== __USER_CS
&& regs
->ss
== __USER_DS
&&
482 regs
->ip
== landing_pad
&&
483 (regs
->flags
& (X86_EFLAGS_RF
| X86_EFLAGS_TF
| X86_EFLAGS_VM
)) == 0;