Skip to contentSkip to conversation

powerpc/syscall: Fix syscall skip handling for seccomp and ptrace

PATCH Superseded Mukesh Kumar Chaurasiya (IBM) 2 participants 3 messages Jul 14 – Jul 27

Subsystems (2)

v5 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

PATCH v5 1/1 powerpc/syscall: Fix syscall skip handling for seccomp and ptrace

After enabling GENERIC_ENTRY on PowerPC, syscall_enter_from_user_mode() returns -1 as a sentinel to signal that seccomp or ptrace has intercepted the syscall and already set a return value via syscall_set_return_value(). system_call_exception() was not handling this sentinel, and since -1UL is >= NR_syscalls, the code fell into the out-of-range path and returned -ENOSYS, overwriting the errno already placed in regs->gpr[3].

The naive fix of checking r0 == -1L before the NR_syscalls bounds check is ambiguous: a user legitimately calling syscall(-1) also produces r0 == -1L, and a tracer intercepting such a call would have its injected return value silently discarded.

Fix this properly by introducing regs->entry_flags, a kernel-internal field in struct pt_regs (consuming one slot of the existing __pt_regs_pad so the ABI is preserved), with SYSCALL_ENTRY_RET_SET as an out-of-band flag. syscall_set_return_value() sets this flag whenever seccomp or ptrace injects a return value. system_call_exception() zeros entry_flags before calling syscall_enter_from_user_mode(), then checks and clears the flag
afterwards: if set, it returns regs->gpr[3] directly regardless of what syscall number the user originally requested.

This handles all seccomp actions correctly:

- SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE (no tracer), SECCOMP_RET_USER_NOTIF:
all call syscall_set_return_value(), flag is set, injected value returned.
- SECCOMP_RET_TRAP, SECCOMP_RET_KILL: call syscall_rollback() and deliver a signal; flag is not set, but the process is dying so the return value is irrelevant.

The fix covers both ppc32 and ppc64 with no #ifdefs.

---
v4 -> v5:
- Fixed build failure where BITS macro is not available
v3 -> v4:
- Use syscall_get_error (Michal)
v2 -> v3:
- Last fix is not working for -1 syscall. Fixed that with this.
v2: https://lore.kernel.org/all/[email protected]

v1 -> v2:
- Fix issues in the previous fix (Michal)
v1: https://lore.kernel.org/all/[email protected]

 arch/powerpc/include/asm/ptrace.h      | 23 ++++++++++++++++++++++-
 arch/powerpc/include/asm/syscall.h     |  6 ++++++
 arch/powerpc/include/uapi/asm/ptrace.h |  6 ++++--
 arch/powerpc/kernel/ptrace/ptrace.c    |  2 ++
 arch/powerpc/kernel/syscall.c          | 18 ++++++++++++++++++
 5 files changed, 52 insertions(+), 3 deletions(-)
arch/powerpc/include/asm/ptrace.h +22 -1
index fdeb97421785..363a89787f1c 100644
@@ -19,6 +19,7 @@
 #ifndef _ASM_POWERPC_PTRACE_H
 #define _ASM_POWERPC_PTRACE_H
 
+#include <linux/bits.h>
 #include <linux/err.h>
 #include <uapi/asm/ptrace.h>
 #include <asm/asm-const.h>
@@ -54,8 +55,9 @@ struct pt_regs
 			};
 			unsigned long result;
 			unsigned long exit_flags;
+			unsigned long entry_flags;
 			/* Maintain 16 byte interrupt stack alignment */
-			unsigned long __pt_regs_pad[3];
+			unsigned long __pt_regs_pad[2];
 		};
 	};
 #if defined(CONFIG_PPC64) || defined(CONFIG_PPC_KUAP)
@@ -233,6 +235,25 @@ static inline unsigned long frame_pointer(struct pt_regs *regs)
 #define current_pt_regs() \
 	((struct pt_regs *)((unsigned long)task_stack_page(current) + THREAD_SIZE) - 1)
 
+/*
+ * SYSCALL_ENTRY_RET_SET: seccomp or ptrace called syscall_set_return_value()
+ * and wants the syscall skipped; regs->gpr[3] already holds the return value.
+ */
+#define SYSCALL_ENTRY_RET_SET	BIT(0)
+
+static inline void set_syscall_entry_ret(struct pt_regs *regs)
+{
+	regs->entry_flags |= SYSCALL_ENTRY_RET_SET;
+}
+
+static inline bool test_and_clear_syscall_entry_ret(struct pt_regs *regs)
+{
+	bool set = !!(regs->entry_flags & SYSCALL_ENTRY_RET_SET);
+
+	regs->entry_flags &= ~SYSCALL_ENTRY_RET_SET;
+	return set;
+}
+
 /*
  * The 4 low bits (0xf) are available as flags to overload the trap word,
  * because interrupt vectors have minimum alignment of 0x10. TRAP_FLAGS_MASK
arch/powerpc/include/asm/syscall.h +6 -0
index 834fcc4f7b54..9ae79326abe3 100644
@@ -98,6 +98,12 @@ static inline void syscall_set_return_value(struct task_struct *task,
 			regs->gpr[3] = val;
 		}
 	}
+	/*
+	 * Mark that a return value has been explicitly set by seccomp or
+	 * ptrace so that system_call_exception() can skip the syscall
+	 * unconditionally, even when the user requested syscall(-1).
+	 */
+	set_syscall_entry_ret(regs);
 }
 
 static inline void syscall_get_arguments(struct task_struct *task,
arch/powerpc/include/uapi/asm/ptrace.h +4 -2
index a393b7f2760a..2f2a43414fe6 100644
@@ -56,7 +56,8 @@ struct pt_regs
 	unsigned long dsisr;		/* on 4xx/Book-E used for ESR */
 	unsigned long result;		/* Result of a system call */
 	unsigned long exit_flags;	/* System call exit flags */
-	unsigned long __pt_regs_pad[3];	/* Maintain 16 byte interrupt stack alignment */
+	unsigned long entry_flags;	/* System call entry flags */
+	unsigned long __pt_regs_pad[2];	/* Maintain 16 byte interrupt stack alignment */
 };
 
 #endif /* __ASSEMBLER__ */
@@ -117,7 +118,8 @@ struct pt_regs
 #define PT_DSISR 42
 #define PT_RESULT 43
 #define PT_EXIT_FLAGS 44
-#define PT_PAD 47 /* 3 times */
+#define PT_ENTRY_FLAGS 45
+#define PT_PAD 46 /* 2 times */
 #define PT_DSCR 48
 #define PT_REGS_COUNT 48
 
arch/powerpc/kernel/ptrace/ptrace.c +2 -0
index 316d4f5ead8e..440d00690cf2 100644
@@ -235,6 +235,8 @@ void __init pt_regs_check(void)
 		     offsetof(struct user_pt_regs, dsisr));
 	BUILD_BUG_ON(offsetof(struct pt_regs, result) !=
 		     offsetof(struct user_pt_regs, result));
+	BUILD_BUG_ON(offsetof(struct pt_regs, entry_flags) !=
+		     offsetof(struct user_pt_regs, entry_flags));
 
 	BUILD_BUG_ON(sizeof(struct user_pt_regs) > sizeof(struct pt_regs));
 
arch/powerpc/kernel/syscall.c +18 -0
index a9da2af6efa8..a43d3a9428cc 100644
@@ -19,9 +19,27 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
 	long ret;
 	syscall_fn f;
 
+	/*
+	 * Zero entry_flags before syscall_enter_from_user_mode() so that
+	 * syscall_set_return_value() can set SYSCALL_ENTRY_RET_SET as an
+	 * unambiguous out-of-band signal.  The field is not initialised by
+	 * the entry assembly.
+	 */
+	regs->entry_flags = 0;
 	add_random_kstack_offset();
 	r0 = syscall_enter_from_user_mode(regs, r0);
 
+	/*
+	 * Seccomp or ptrace may have set a return value and requested that
+	 * the syscall be skipped. syscall_set_return_value() sets
+	 * SYSCALL_ENTRY_RET_SET in regs->entry_flags as an
+	 * unambiguous out-of-band signal. This avoids the ambiguity of
+	 * using r0 == -1 as the skip sentinel when the user themselves
+	 * called syscall(-1).
+	 */
+	if (unlikely(test_and_clear_syscall_entry_ret(regs)))
+		return syscall_get_error(current, regs);
+
 	if (unlikely(r0 >= NR_syscalls)) {
 		if (unlikely(trap_is_unsupported_scv(regs))) {
 			/* Unsupported scv vector */
Signature · 1 line
2.55.0
#1 lore raw

Hello,

this should be probably adjusted to match

https://lore.kernel.org/linuxppc-dev/[email protected]/

Thanks

Michal

Quoted 196 lines · Mukesh Kumar Chaurasiya (IBM)

After enabling GENERIC_ENTRY on PowerPC, syscall_enter_from_user_mode() returns -1 as a sentinel to signal that seccomp or ptrace has intercepted the syscall and already set a return value via syscall_set_return_value(). system_call_exception() was not handling this sentinel, and since -1UL is >= NR_syscalls, the code fell into the out-of-range path and returned -ENOSYS, overwriting the errno already placed in regs->gpr[3].

The naive fix of checking r0 == -1L before the NR_syscalls bounds check is ambiguous: a user legitimately calling syscall(-1) also produces r0 == -1L, and a tracer intercepting such a call would have its injected return value silently discarded.

Fix this properly by introducing regs->entry_flags, a kernel-internal field in struct pt_regs (consuming one slot of the existing __pt_regs_pad so the ABI is preserved), with SYSCALL_ENTRY_RET_SET as an out-of-band flag. syscall_set_return_value() sets this flag whenever seccomp or ptrace injects a return value. system_call_exception() zeros entry_flags before calling syscall_enter_from_user_mode(), then checks and clears the flag
afterwards: if set, it returns regs->gpr[3] directly regardless of what syscall number the user originally requested.

This handles all seccomp actions correctly:

- SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE (no tracer), SECCOMP_RET_USER_NOTIF:
all call syscall_set_return_value(), flag is set, injected value returned.
- SECCOMP_RET_TRAP, SECCOMP_RET_KILL: call syscall_rollback() and deliver a signal; flag is not set, but the process is dying so the return value is irrelevant.

The fix covers both ppc32 and ppc64 with no #ifdefs.

---
v4 -> v5:
- Fixed build failure where BITS macro is not available
v3 -> v4:
- Use syscall_get_error (Michal)
v2 -> v3:
- Last fix is not working for -1 syscall. Fixed that with this.
v2: https://lore.kernel.org/all/[email protected]

v1 -> v2:
- Fix issues in the previous fix (Michal)
v1: https://lore.kernel.org/all/[email protected]

 arch/powerpc/include/asm/ptrace.h      | 23 ++++++++++++++++++++++-
 arch/powerpc/include/asm/syscall.h     |  6 ++++++
 arch/powerpc/include/uapi/asm/ptrace.h |  6 ++++--
 arch/powerpc/kernel/ptrace/ptrace.c    |  2 ++
 arch/powerpc/kernel/syscall.c          | 18 ++++++++++++++++++
 5 files changed, 52 insertions(+), 3 deletions(-)
arch/powerpc/include/asm/ptrace.h +22 -1
index fdeb97421785..363a89787f1c 100644
@@ -19,6 +19,7 @@
 #ifndef _ASM_POWERPC_PTRACE_H
 #define _ASM_POWERPC_PTRACE_H
 
+#include <linux/bits.h>
 #include <linux/err.h>
 #include <uapi/asm/ptrace.h>
 #include <asm/asm-const.h>
@@ -54,8 +55,9 @@ struct pt_regs
 			};
 			unsigned long result;
 			unsigned long exit_flags;
+			unsigned long entry_flags;
 			/* Maintain 16 byte interrupt stack alignment */
-			unsigned long __pt_regs_pad[3];
+			unsigned long __pt_regs_pad[2];
 		};
 	};
 #if defined(CONFIG_PPC64) || defined(CONFIG_PPC_KUAP)
@@ -233,6 +235,25 @@ static inline unsigned long frame_pointer(struct pt_regs *regs)
 #define current_pt_regs() \
 	((struct pt_regs *)((unsigned long)task_stack_page(current) + THREAD_SIZE) - 1)
 
+/*
+ * SYSCALL_ENTRY_RET_SET: seccomp or ptrace called syscall_set_return_value()
+ * and wants the syscall skipped; regs->gpr[3] already holds the return value.
+ */
+#define SYSCALL_ENTRY_RET_SET	BIT(0)
+
+static inline void set_syscall_entry_ret(struct pt_regs *regs)
+{
+	regs->entry_flags |= SYSCALL_ENTRY_RET_SET;
+}
+
+static inline bool test_and_clear_syscall_entry_ret(struct pt_regs *regs)
+{
+	bool set = !!(regs->entry_flags & SYSCALL_ENTRY_RET_SET);
+
+	regs->entry_flags &= ~SYSCALL_ENTRY_RET_SET;
+	return set;
+}
+
 /*
  * The 4 low bits (0xf) are available as flags to overload the trap word,
  * because interrupt vectors have minimum alignment of 0x10. TRAP_FLAGS_MASK
arch/powerpc/include/asm/syscall.h +6 -0
index 834fcc4f7b54..9ae79326abe3 100644
@@ -98,6 +98,12 @@ static inline void syscall_set_return_value(struct task_struct *task,
 			regs->gpr[3] = val;
 		}
 	}
+	/*
+	 * Mark that a return value has been explicitly set by seccomp or
+	 * ptrace so that system_call_exception() can skip the syscall
+	 * unconditionally, even when the user requested syscall(-1).
+	 */
+	set_syscall_entry_ret(regs);
 }
 
 static inline void syscall_get_arguments(struct task_struct *task,
arch/powerpc/include/uapi/asm/ptrace.h +4 -2
index a393b7f2760a..2f2a43414fe6 100644
@@ -56,7 +56,8 @@ struct pt_regs
 	unsigned long dsisr;		/* on 4xx/Book-E used for ESR */
 	unsigned long result;		/* Result of a system call */
 	unsigned long exit_flags;	/* System call exit flags */
-	unsigned long __pt_regs_pad[3];	/* Maintain 16 byte interrupt stack alignment */
+	unsigned long entry_flags;	/* System call entry flags */
+	unsigned long __pt_regs_pad[2];	/* Maintain 16 byte interrupt stack alignment */
 };
 
 #endif /* __ASSEMBLER__ */
@@ -117,7 +118,8 @@ struct pt_regs
 #define PT_DSISR 42
 #define PT_RESULT 43
 #define PT_EXIT_FLAGS 44
-#define PT_PAD 47 /* 3 times */
+#define PT_ENTRY_FLAGS 45
+#define PT_PAD 46 /* 2 times */
 #define PT_DSCR 48
 #define PT_REGS_COUNT 48
 
arch/powerpc/kernel/ptrace/ptrace.c +2 -0
index 316d4f5ead8e..440d00690cf2 100644
@@ -235,6 +235,8 @@ void __init pt_regs_check(void)
 		     offsetof(struct user_pt_regs, dsisr));
 	BUILD_BUG_ON(offsetof(struct pt_regs, result) !=
 		     offsetof(struct user_pt_regs, result));
+	BUILD_BUG_ON(offsetof(struct pt_regs, entry_flags) !=
+		     offsetof(struct user_pt_regs, entry_flags));
 
 	BUILD_BUG_ON(sizeof(struct user_pt_regs) > sizeof(struct pt_regs));
 
arch/powerpc/kernel/syscall.c +18 -0
index a9da2af6efa8..a43d3a9428cc 100644
@@ -19,9 +19,27 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
 	long ret;
 	syscall_fn f;
 
+	/*
+	 * Zero entry_flags before syscall_enter_from_user_mode() so that
+	 * syscall_set_return_value() can set SYSCALL_ENTRY_RET_SET as an
+	 * unambiguous out-of-band signal.  The field is not initialised by
+	 * the entry assembly.
+	 */
+	regs->entry_flags = 0;
 	add_random_kstack_offset();
 	r0 = syscall_enter_from_user_mode(regs, r0);
 
+	/*
+	 * Seccomp or ptrace may have set a return value and requested that
+	 * the syscall be skipped. syscall_set_return_value() sets
+	 * SYSCALL_ENTRY_RET_SET in regs->entry_flags as an
+	 * unambiguous out-of-band signal. This avoids the ambiguity of
+	 * using r0 == -1 as the skip sentinel when the user themselves
+	 * called syscall(-1).
+	 */
+	if (unlikely(test_and_clear_syscall_entry_ret(regs)))
+		return syscall_get_error(current, regs);
+
 	if (unlikely(r0 >= NR_syscalls)) {
 		if (unlikely(trap_is_unsupported_scv(regs))) {
 			/* Unsupported scv vector */
Signature · 1 line
2.55.0
#2 lore raw
Quoted 10 lines · Michal Suchánek

Hello,

this should be probably adjusted to match

https://lore.kernel.org/linuxppc-dev/[email protected]/

Thanks

Michal

Hey Michal,

Yeah, i am working towards this.

Regards,
Mukesh

Quoted 197 lines · Mukesh Kumar Chaurasiya (IBM)
Quoted 196 lines

After enabling GENERIC_ENTRY on PowerPC, syscall_enter_from_user_mode() returns -1 as a sentinel to signal that seccomp or ptrace has intercepted the syscall and already set a return value via syscall_set_return_value(). system_call_exception() was not handling this sentinel, and since -1UL is >= NR_syscalls, the code fell into the out-of-range path and returned -ENOSYS, overwriting the errno already placed in regs->gpr[3].

The naive fix of checking r0 == -1L before the NR_syscalls bounds check is ambiguous: a user legitimately calling syscall(-1) also produces r0 == -1L, and a tracer intercepting such a call would have its injected return value silently discarded.

Fix this properly by introducing regs->entry_flags, a kernel-internal field in struct pt_regs (consuming one slot of the existing __pt_regs_pad so the ABI is preserved), with SYSCALL_ENTRY_RET_SET as an out-of-band flag. syscall_set_return_value() sets this flag whenever seccomp or ptrace injects a return value. system_call_exception() zeros entry_flags before calling syscall_enter_from_user_mode(), then checks and clears the flag
afterwards: if set, it returns regs->gpr[3] directly regardless of what syscall number the user originally requested.

This handles all seccomp actions correctly:

- SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE (no tracer), SECCOMP_RET_USER_NOTIF:
all call syscall_set_return_value(), flag is set, injected value returned.
- SECCOMP_RET_TRAP, SECCOMP_RET_KILL: call syscall_rollback() and deliver a signal; flag is not set, but the process is dying so the return value is irrelevant.

The fix covers both ppc32 and ppc64 with no #ifdefs.

---
v4 -> v5:
- Fixed build failure where BITS macro is not available
v3 -> v4:
- Use syscall_get_error (Michal)
v2 -> v3:
- Last fix is not working for -1 syscall. Fixed that with this.
v2: https://lore.kernel.org/all/[email protected]

v1 -> v2:
- Fix issues in the previous fix (Michal)
v1: https://lore.kernel.org/all/[email protected]

 arch/powerpc/include/asm/ptrace.h      | 23 ++++++++++++++++++++++-
 arch/powerpc/include/asm/syscall.h     |  6 ++++++
 arch/powerpc/include/uapi/asm/ptrace.h |  6 ++++--
 arch/powerpc/kernel/ptrace/ptrace.c    |  2 ++
 arch/powerpc/kernel/syscall.c          | 18 ++++++++++++++++++
 5 files changed, 52 insertions(+), 3 deletions(-)
arch/powerpc/include/asm/ptrace.h +22 -1
index fdeb97421785..363a89787f1c 100644
@@ -19,6 +19,7 @@
 #ifndef _ASM_POWERPC_PTRACE_H
 #define _ASM_POWERPC_PTRACE_H
 
+#include <linux/bits.h>
 #include <linux/err.h>
 #include <uapi/asm/ptrace.h>
 #include <asm/asm-const.h>
@@ -54,8 +55,9 @@ struct pt_regs
 			};
 			unsigned long result;
 			unsigned long exit_flags;
+			unsigned long entry_flags;
 			/* Maintain 16 byte interrupt stack alignment */
-			unsigned long __pt_regs_pad[3];
+			unsigned long __pt_regs_pad[2];
 		};
 	};
 #if defined(CONFIG_PPC64) || defined(CONFIG_PPC_KUAP)
@@ -233,6 +235,25 @@ static inline unsigned long frame_pointer(struct pt_regs *regs)
 #define current_pt_regs() \
 	((struct pt_regs *)((unsigned long)task_stack_page(current) + THREAD_SIZE) - 1)
 
+/*
+ * SYSCALL_ENTRY_RET_SET: seccomp or ptrace called syscall_set_return_value()
+ * and wants the syscall skipped; regs->gpr[3] already holds the return value.
+ */
+#define SYSCALL_ENTRY_RET_SET	BIT(0)
+
+static inline void set_syscall_entry_ret(struct pt_regs *regs)
+{
+	regs->entry_flags |= SYSCALL_ENTRY_RET_SET;
+}
+
+static inline bool test_and_clear_syscall_entry_ret(struct pt_regs *regs)
+{
+	bool set = !!(regs->entry_flags & SYSCALL_ENTRY_RET_SET);
+
+	regs->entry_flags &= ~SYSCALL_ENTRY_RET_SET;
+	return set;
+}
+
 /*
  * The 4 low bits (0xf) are available as flags to overload the trap word,
  * because interrupt vectors have minimum alignment of 0x10. TRAP_FLAGS_MASK
arch/powerpc/include/asm/syscall.h +6 -0
index 834fcc4f7b54..9ae79326abe3 100644
@@ -98,6 +98,12 @@ static inline void syscall_set_return_value(struct task_struct *task,
 			regs->gpr[3] = val;
 		}
 	}
+	/*
+	 * Mark that a return value has been explicitly set by seccomp or
+	 * ptrace so that system_call_exception() can skip the syscall
+	 * unconditionally, even when the user requested syscall(-1).
+	 */
+	set_syscall_entry_ret(regs);
 }
 
 static inline void syscall_get_arguments(struct task_struct *task,
arch/powerpc/include/uapi/asm/ptrace.h +4 -2
index a393b7f2760a..2f2a43414fe6 100644
@@ -56,7 +56,8 @@ struct pt_regs
 	unsigned long dsisr;		/* on 4xx/Book-E used for ESR */
 	unsigned long result;		/* Result of a system call */
 	unsigned long exit_flags;	/* System call exit flags */
-	unsigned long __pt_regs_pad[3];	/* Maintain 16 byte interrupt stack alignment */
+	unsigned long entry_flags;	/* System call entry flags */
+	unsigned long __pt_regs_pad[2];	/* Maintain 16 byte interrupt stack alignment */
 };
 
 #endif /* __ASSEMBLER__ */
@@ -117,7 +118,8 @@ struct pt_regs
 #define PT_DSISR 42
 #define PT_RESULT 43
 #define PT_EXIT_FLAGS 44
-#define PT_PAD 47 /* 3 times */
+#define PT_ENTRY_FLAGS 45
+#define PT_PAD 46 /* 2 times */
 #define PT_DSCR 48
 #define PT_REGS_COUNT 48
 
arch/powerpc/kernel/ptrace/ptrace.c +2 -0
index 316d4f5ead8e..440d00690cf2 100644
@@ -235,6 +235,8 @@ void __init pt_regs_check(void)
 		     offsetof(struct user_pt_regs, dsisr));
 	BUILD_BUG_ON(offsetof(struct pt_regs, result) !=
 		     offsetof(struct user_pt_regs, result));
+	BUILD_BUG_ON(offsetof(struct pt_regs, entry_flags) !=
+		     offsetof(struct user_pt_regs, entry_flags));
 
 	BUILD_BUG_ON(sizeof(struct user_pt_regs) > sizeof(struct pt_regs));
 
arch/powerpc/kernel/syscall.c +18 -0
index a9da2af6efa8..a43d3a9428cc 100644
@@ -19,9 +19,27 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
 	long ret;
 	syscall_fn f;
 
+	/*
+	 * Zero entry_flags before syscall_enter_from_user_mode() so that
+	 * syscall_set_return_value() can set SYSCALL_ENTRY_RET_SET as an
+	 * unambiguous out-of-band signal.  The field is not initialised by
+	 * the entry assembly.
+	 */
+	regs->entry_flags = 0;
 	add_random_kstack_offset();
 	r0 = syscall_enter_from_user_mode(regs, r0);
 
+	/*
+	 * Seccomp or ptrace may have set a return value and requested that
+	 * the syscall be skipped. syscall_set_return_value() sets
+	 * SYSCALL_ENTRY_RET_SET in regs->entry_flags as an
+	 * unambiguous out-of-band signal. This avoids the ambiguity of
+	 * using r0 == -1 as the skip sentinel when the user themselves
+	 * called syscall(-1).
+	 */
+	if (unlikely(test_and_clear_syscall_entry_ret(regs)))
+		return syscall_get_error(current, regs);
+
 	if (unlikely(r0 >= NR_syscalls)) {
 		if (unlikely(trap_is_unsupported_scv(regs))) {
 			/* Unsupported scv vector */
Signature · 1 line
2.55.0