xref: /linux/kernel/entry/syscall-common.c (revision dd0a04606937af5810e9117d343ee3792635bd3d)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/audit.h>
4 #include <linux/entry-common.h>
5 #include "common.h"
6 
7 #define CREATE_TRACE_POINTS
8 #include <trace/events/syscalls.h>
9 
10 static inline void syscall_enter_audit(struct pt_regs *regs, long syscall)
11 {
12 	if (unlikely(audit_context())) {
13 		unsigned long args[6];
14 
15 		syscall_get_arguments(current, regs, args);
16 		audit_syscall_entry(syscall, args[0], args[1], args[2], args[3]);
17 	}
18 }
19 
20 long syscall_trace_enter(struct pt_regs *regs, long syscall, unsigned long work)
21 {
22 	long ret = 0;
23 
24 	/*
25 	 * Handle Syscall User Dispatch.  This must comes first, since
26 	 * the ABI here can be something that doesn't make sense for
27 	 * other syscall_work features.
28 	 */
29 	if (work & SYSCALL_WORK_SYSCALL_USER_DISPATCH) {
30 		if (syscall_user_dispatch(regs))
31 			return -1L;
32 	}
33 
34 	/*
35 	 * User space got a time slice extension granted and relinquishes
36 	 * the CPU. The work stops the slice timer to avoid an extra round
37 	 * through hrtimer_interrupt().
38 	 */
39 	if (work & SYSCALL_WORK_SYSCALL_RSEQ_SLICE)
40 		rseq_syscall_enter_work(syscall);
41 
42 	/* Handle ptrace */
43 	if (work & (SYSCALL_WORK_SYSCALL_TRACE | SYSCALL_WORK_SYSCALL_EMU)) {
44 		ret = ptrace_report_syscall_entry(regs);
45 		if (ret || (work & SYSCALL_WORK_SYSCALL_EMU))
46 			return -1L;
47 	}
48 
49 	/* Do seccomp after ptrace, to catch any tracer changes. */
50 	if (work & SYSCALL_WORK_SECCOMP) {
51 		ret = __secure_computing();
52 		if (ret == -1L)
53 			return ret;
54 	}
55 
56 	/* Either of the above might have changed the syscall number */
57 	syscall = syscall_get_nr(current, regs);
58 
59 	if (unlikely(work & SYSCALL_WORK_SYSCALL_TRACEPOINT)) {
60 		trace_sys_enter(regs, syscall);
61 		/*
62 		 * Probes or BPF hooks in the tracepoint may have changed the
63 		 * system call number as well.
64 		 */
65 		syscall = syscall_get_nr(current, regs);
66 	}
67 
68 	syscall_enter_audit(regs, syscall);
69 
70 	return ret ? : syscall;
71 }
72 
73 /*
74  * If SYSCALL_EMU is set, then the only reason to report is when
75  * SINGLESTEP is set (i.e. PTRACE_SYSEMU_SINGLESTEP).  This syscall
76  * instruction has been already reported in syscall_enter_from_user_mode().
77  */
78 static inline bool report_single_step(unsigned long work)
79 {
80 	if (work & SYSCALL_WORK_SYSCALL_EMU)
81 		return false;
82 
83 	return work & SYSCALL_WORK_SYSCALL_EXIT_TRAP;
84 }
85 
86 void syscall_exit_work(struct pt_regs *regs, unsigned long work)
87 {
88 	bool step;
89 
90 	/*
91 	 * If the syscall was rolled back due to syscall user dispatching,
92 	 * then the tracers below are not invoked for the same reason as
93 	 * the entry side was not invoked in syscall_trace_enter(): The ABI
94 	 * of these syscalls is unknown.
95 	 */
96 	if (work & SYSCALL_WORK_SYSCALL_USER_DISPATCH) {
97 		if (unlikely(current->syscall_dispatch.on_dispatch)) {
98 			current->syscall_dispatch.on_dispatch = false;
99 			return;
100 		}
101 	}
102 
103 	audit_syscall_exit(regs);
104 
105 	if (work & SYSCALL_WORK_SYSCALL_TRACEPOINT)
106 		trace_sys_exit(regs, syscall_get_return_value(current, regs));
107 
108 	step = report_single_step(work);
109 	if (step || work & SYSCALL_WORK_SYSCALL_TRACE)
110 		ptrace_report_syscall_exit(regs, step);
111 }
112