From 44c5b6768e3a1385fdf3b10893404bc5a2c1248a Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 27 Oct 2025 09:43:42 +0100 Subject: [PATCH 01/16] ARM: uaccess: Implement missing __get_user_asm_dword() When CONFIG_CPU_SPECTRE=n then get_user() is missing the 8 byte ASM variant for no real good reason. This prevents using get_user(u64) in generic code. Implement it as a sequence of two 4-byte reads with LE/BE awareness and make the unsigned long (or long long) type for the intermediate variable to read into dependend on the the target type. The __long_type() macro and idea was lifted from PowerPC. Thanks to Christophe for pointing it out. Reported-by: kernel test robot Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Mathieu Desnoyers Closes: https://lore.kernel.org/oe-kbuild-all/202509120155.pFgwfeUD-lkp@intel.com/ Link: https://patch.msgid.link/20251027083745.168468637@linutronix.de --- arch/arm/include/asm/uaccess.h | 26 +++++++++++++++++++++++++- 1 file changed, 25 insertions(+), 1 deletion(-) diff --git a/arch/arm/include/asm/uaccess.h b/arch/arm/include/asm/uaccess.h index f90be312418e..d6ae80b5df36 100644 --- a/arch/arm/include/asm/uaccess.h +++ b/arch/arm/include/asm/uaccess.h @@ -283,10 +283,17 @@ extern int __put_user_8(void *, unsigned long long); __gu_err; \ }) +/* + * This is a type: either unsigned long, if the argument fits into + * that type, or otherwise unsigned long long. + */ +#define __long_type(x) \ + __typeof__(__builtin_choose_expr(sizeof(x) > sizeof(0UL), 0ULL, 0UL)) + #define __get_user_err(x, ptr, err, __t) \ do { \ unsigned long __gu_addr = (unsigned long)(ptr); \ - unsigned long __gu_val; \ + __long_type(x) __gu_val; \ unsigned int __ua_flags; \ __chk_user_ptr(ptr); \ might_fault(); \ @@ -295,6 +302,7 @@ do { \ case 1: __get_user_asm_byte(__gu_val, __gu_addr, err, __t); break; \ case 2: __get_user_asm_half(__gu_val, __gu_addr, err, __t); break; \ case 4: __get_user_asm_word(__gu_val, __gu_addr, err, __t); break; \ + case 8: __get_user_asm_dword(__gu_val, __gu_addr, err, __t); break; \ default: (__gu_val) = __get_user_bad(); \ } \ uaccess_restore(__ua_flags); \ @@ -353,6 +361,22 @@ do { \ #define __get_user_asm_word(x, addr, err, __t) \ __get_user_asm(x, addr, err, "ldr" __t) +#ifdef __ARMEB__ +#define __WORD0_OFFS 4 +#define __WORD1_OFFS 0 +#else +#define __WORD0_OFFS 0 +#define __WORD1_OFFS 4 +#endif + +#define __get_user_asm_dword(x, addr, err, __t) \ + ({ \ + unsigned long __w0, __w1; \ + __get_user_asm(__w0, addr + __WORD0_OFFS, err, "ldr" __t); \ + __get_user_asm(__w1, addr + __WORD1_OFFS, err, "ldr" __t); \ + (x) = ((u64)__w1 << 32) | (u64) __w0; \ +}) + #define __put_user_switch(x, ptr, __err, __fn) \ do { \ const __typeof__(*(ptr)) __user *__pu_ptr = (ptr); \ From 3eb6660f26d13acdbcb9241ac3e95d44419f2284 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Wed, 29 Oct 2025 10:40:52 +0100 Subject: [PATCH 02/16] uaccess: Provide ASM GOTO safe wrappers for unsafe_*_user() ASM GOTO is miscompiled by GCC when it is used inside a auto cleanup scope: bool foo(u32 __user *p, u32 val) { scoped_guard(pagefault) unsafe_put_user(val, p, efault); return true; efault: return false; } e80: e8 00 00 00 00 call e85 e85: 65 48 8b 05 00 00 00 00 mov %gs:0x0(%rip),%rax e8d: 83 80 04 14 00 00 01 addl $0x1,0x1404(%rax) // pf_disable++ e94: 89 37 mov %esi,(%rdi) e96: 83 a8 04 14 00 00 01 subl $0x1,0x1404(%rax) // pf_disable-- e9d: b8 01 00 00 00 mov $0x1,%eax // success ea2: e9 00 00 00 00 jmp ea7 // ret ea7: 31 c0 xor %eax,%eax // fail ea9: e9 00 00 00 00 jmp eae // ret which is broken as it leaks the pagefault disable counter on failure. Clang at least fails the build. Linus suggested to add a local label into the macro scope and let that jump to the actual caller supplied error label. __label__ local_label; \ arch_unsafe_get_user(x, ptr, local_label); \ if (0) { \ local_label: \ goto label; \ That works for both GCC and clang. clang: c80: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1) c85: 65 48 8b 0c 25 00 00 00 00 mov %gs:0x0,%rcx c8e: ff 81 04 14 00 00 incl 0x1404(%rcx) // pf_disable++ c94: 31 c0 xor %eax,%eax // set retval to false c96: 89 37 mov %esi,(%rdi) // write c98: b0 01 mov $0x1,%al // set retval to true c9a: ff 89 04 14 00 00 decl 0x1404(%rcx) // pf_disable-- ca0: 2e e9 00 00 00 00 cs jmp ca6 // ret The exception table entry points correctly to c9a GCC: f70: e8 00 00 00 00 call f75 f75: 65 48 8b 05 00 00 00 00 mov %gs:0x0(%rip),%rax f7d: 83 80 04 14 00 00 01 addl $0x1,0x1404(%rax) // pf_disable++ f84: 8b 17 mov (%rdi),%edx f86: 89 16 mov %edx,(%rsi) f88: 83 a8 04 14 00 00 01 subl $0x1,0x1404(%rax) // pf_disable-- f8f: b8 01 00 00 00 mov $0x1,%eax // success f94: e9 00 00 00 00 jmp f99 // ret f99: 83 a8 04 14 00 00 01 subl $0x1,0x1404(%rax) // pf_disable-- fa0: 31 c0 xor %eax,%eax // fail fa2: e9 00 00 00 00 jmp fa7 // ret The exception table entry points correctly to f99 So both compilers optimize out the extra goto and emit correct and efficient code. Provide a generic wrapper to do that to avoid modifying all the affected architecture specific implementation with that workaround. The only change required for architectures is to rename unsafe_*_user() to arch_unsafe_*_user(). That's done in subsequent changes. Suggested-by: Linus Torvalds Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Mathieu Desnoyers Link: https://patch.msgid.link/877bweujtn.ffs@tglx --- include/linux/uaccess.h | 72 ++++++++++++++++++++++++++++++++++++++--- 1 file changed, 68 insertions(+), 4 deletions(-) diff --git a/include/linux/uaccess.h b/include/linux/uaccess.h index 1beb5b395d81..8aa82b1d6013 100644 --- a/include/linux/uaccess.h +++ b/include/linux/uaccess.h @@ -518,7 +518,34 @@ long strncpy_from_user_nofault(char *dst, const void __user *unsafe_addr, long count); long strnlen_user_nofault(const void __user *unsafe_addr, long count); -#ifndef __get_kernel_nofault +#ifdef arch_get_kernel_nofault +/* + * Wrap the architecture implementation so that @label can be outside of a + * cleanup() scope. A regular C goto works correctly, but ASM goto does + * not. Clang rejects such an attempt, but GCC silently emits buggy code. + */ +#define __get_kernel_nofault(dst, src, type, label) \ +do { \ + __label__ local_label; \ + arch_get_kernel_nofault(dst, src, type, local_label); \ + if (0) { \ + local_label: \ + goto label; \ + } \ +} while (0) + +#define __put_kernel_nofault(dst, src, type, label) \ +do { \ + __label__ local_label; \ + arch_put_kernel_nofault(dst, src, type, local_label); \ + if (0) { \ + local_label: \ + goto label; \ + } \ +} while (0) + +#elif !defined(__get_kernel_nofault) /* arch_get_kernel_nofault */ + #define __get_kernel_nofault(dst, src, type, label) \ do { \ type __user *p = (type __force __user *)(src); \ @@ -535,7 +562,8 @@ do { \ if (__put_user(data, p)) \ goto label; \ } while (0) -#endif + +#endif /* !__get_kernel_nofault */ /** * get_kernel_nofault(): safely attempt to read from a location @@ -549,7 +577,42 @@ do { \ copy_from_kernel_nofault(&(val), __gk_ptr, sizeof(val));\ }) -#ifndef user_access_begin +#ifdef user_access_begin + +#ifdef arch_unsafe_get_user +/* + * Wrap the architecture implementation so that @label can be outside of a + * cleanup() scope. A regular C goto works correctly, but ASM goto does + * not. Clang rejects such an attempt, but GCC silently emits buggy code. + * + * Some architectures use internal local labels already, but this extra + * indirection here is harmless because the compiler optimizes it out + * completely in any case. This construct just ensures that the ASM GOTO + * target is always in the local scope. The C goto 'label' works correctly + * when leaving a cleanup() scope. + */ +#define unsafe_get_user(x, ptr, label) \ +do { \ + __label__ local_label; \ + arch_unsafe_get_user(x, ptr, local_label); \ + if (0) { \ + local_label: \ + goto label; \ + } \ +} while (0) + +#define unsafe_put_user(x, ptr, label) \ +do { \ + __label__ local_label; \ + arch_unsafe_put_user(x, ptr, local_label); \ + if (0) { \ + local_label: \ + goto label; \ + } \ +} while (0) +#endif /* arch_unsafe_get_user */ + +#else /* user_access_begin */ #define user_access_begin(ptr,len) access_ok(ptr, len) #define user_access_end() do { } while (0) #define unsafe_op_wrap(op, err) do { if (unlikely(op)) goto err; } while (0) @@ -559,7 +622,8 @@ do { \ #define unsafe_copy_from_user(d,s,l,e) unsafe_op_wrap(__copy_from_user(d,s,l),e) static inline unsigned long user_access_save(void) { return 0UL; } static inline void user_access_restore(unsigned long flags) { } -#endif +#endif /* !user_access_begin */ + #ifndef user_write_access_begin #define user_write_access_begin user_access_begin #define user_write_access_end user_access_end From 14219398e3e1ce774d47b2dd55852d9b693cc6e1 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 27 Oct 2025 09:43:46 +0100 Subject: [PATCH 03/16] x86/uaccess: Use unsafe wrappers for ASM GOTO ASM GOTO is miscompiled by GCC when it is used inside a auto cleanup scope: bool foo(u32 __user *p, u32 val) { scoped_guard(pagefault) unsafe_put_user(val, p, efault); return true; efault: return false; } It ends up leaking the pagefault disable counter in the fault path. clang at least fails the build. Rename unsafe_*_user() to arch_unsafe_*_user() which makes the generic uaccess header wrap it with a local label that makes both compilers emit correct code. Same for the kernel_nofault() variants. Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Mathieu Desnoyers Link: https://patch.msgid.link/20251027083745.294359925@linutronix.de --- arch/x86/include/asm/uaccess.h | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/arch/x86/include/asm/uaccess.h b/arch/x86/include/asm/uaccess.h index 91a3fb8ae7ff..367297b188c3 100644 --- a/arch/x86/include/asm/uaccess.h +++ b/arch/x86/include/asm/uaccess.h @@ -528,18 +528,18 @@ static __must_check __always_inline bool user_access_begin(const void __user *pt #define user_access_save() smap_save() #define user_access_restore(x) smap_restore(x) -#define unsafe_put_user(x, ptr, label) \ +#define arch_unsafe_put_user(x, ptr, label) \ __put_user_size((__typeof__(*(ptr)))(x), (ptr), sizeof(*(ptr)), label) #ifdef CONFIG_CC_HAS_ASM_GOTO_OUTPUT -#define unsafe_get_user(x, ptr, err_label) \ +#define arch_unsafe_get_user(x, ptr, err_label) \ do { \ __inttype(*(ptr)) __gu_val; \ __get_user_size(__gu_val, (ptr), sizeof(*(ptr)), err_label); \ (x) = (__force __typeof__(*(ptr)))__gu_val; \ } while (0) #else // !CONFIG_CC_HAS_ASM_GOTO_OUTPUT -#define unsafe_get_user(x, ptr, err_label) \ +#define arch_unsafe_get_user(x, ptr, err_label) \ do { \ int __gu_err; \ __inttype(*(ptr)) __gu_val; \ @@ -618,11 +618,11 @@ do { \ } while (0) #ifdef CONFIG_CC_HAS_ASM_GOTO_OUTPUT -#define __get_kernel_nofault(dst, src, type, err_label) \ +#define arch_get_kernel_nofault(dst, src, type, err_label) \ __get_user_size(*((type *)(dst)), (__force type __user *)(src), \ sizeof(type), err_label) #else // !CONFIG_CC_HAS_ASM_GOTO_OUTPUT -#define __get_kernel_nofault(dst, src, type, err_label) \ +#define arch_get_kernel_nofault(dst, src, type, err_label) \ do { \ int __kr_err; \ \ @@ -633,7 +633,7 @@ do { \ } while (0) #endif // CONFIG_CC_HAS_ASM_GOTO_OUTPUT -#define __put_kernel_nofault(dst, src, type, err_label) \ +#define arch_put_kernel_nofault(dst, src, type, err_label) \ __put_user_size(*((type *)(src)), (__force type __user *)(dst), \ sizeof(type), err_label) From 5002dd53144f82d43eba778c927adfa7d429c16a Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 27 Oct 2025 09:43:48 +0100 Subject: [PATCH 04/16] powerpc/uaccess: Use unsafe wrappers for ASM GOTO ASM GOTO is miscompiled by GCC when it is used inside a auto cleanup scope: bool foo(u32 __user *p, u32 val) { scoped_guard(pagefault) unsafe_put_user(val, p, efault); return true; efault: return false; } It ends up leaking the pagefault disable counter in the fault path. clang at least fails the build. Rename unsafe_*_user() to arch_unsafe_*_user() which makes the generic uaccess header wrap it with a local label that makes both compilers emit correct code. Same for the kernel_nofault() variants. Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Mathieu Desnoyers Link: https://patch.msgid.link/20251027083745.356628509@linutronix.de --- arch/powerpc/include/asm/uaccess.h | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/arch/powerpc/include/asm/uaccess.h b/arch/powerpc/include/asm/uaccess.h index 4f5a46a77fa2..784a00e681fa 100644 --- a/arch/powerpc/include/asm/uaccess.h +++ b/arch/powerpc/include/asm/uaccess.h @@ -451,7 +451,7 @@ user_write_access_begin(const void __user *ptr, size_t len) #define user_write_access_begin user_write_access_begin #define user_write_access_end prevent_current_write_to_user -#define unsafe_get_user(x, p, e) do { \ +#define arch_unsafe_get_user(x, p, e) do { \ __long_type(*(p)) __gu_val; \ __typeof__(*(p)) __user *__gu_addr = (p); \ \ @@ -459,7 +459,7 @@ user_write_access_begin(const void __user *ptr, size_t len) (x) = (__typeof__(*(p)))__gu_val; \ } while (0) -#define unsafe_put_user(x, p, e) \ +#define arch_unsafe_put_user(x, p, e) \ __put_user_size_goto((__typeof__(*(p)))(x), (p), sizeof(*(p)), e) #define unsafe_copy_from_user(d, s, l, e) \ @@ -504,11 +504,11 @@ do { \ unsafe_put_user(*(u8*)(_src + _i), (u8 __user *)(_dst + _i), e); \ } while (0) -#define __get_kernel_nofault(dst, src, type, err_label) \ +#define arch_get_kernel_nofault(dst, src, type, err_label) \ __get_user_size_goto(*((type *)(dst)), \ (__force type __user *)(src), sizeof(type), err_label) -#define __put_kernel_nofault(dst, src, type, err_label) \ +#define arch_put_kernel_nofault(dst, src, type, err_label) \ __put_user_size_goto(*((type *)(src)), \ (__force type __user *)(dst), sizeof(type), err_label) From 0988ea18c6244da3dc35cfc0ad621531d0e1508a Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 27 Oct 2025 09:43:50 +0100 Subject: [PATCH 05/16] riscv/uaccess: Use unsafe wrappers for ASM GOTO ASM GOTO is miscompiled by GCC when it is used inside a auto cleanup scope: bool foo(u32 __user *p, u32 val) { scoped_guard(pagefault) unsafe_put_user(val, p, efault); return true; efault: return false; } It ends up leaking the pagefault disable counter in the fault path. clang at least fails the build. Rename unsafe_*_user() to arch_unsafe_*_user() which makes the generic uaccess header wrap it with a local label that makes both compilers emit correct code. Same for the kernel_nofault() variants. Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Mathieu Desnoyers Link: https://patch.msgid.link/20251027083745.419351819@linutronix.de --- arch/riscv/include/asm/uaccess.h | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/arch/riscv/include/asm/uaccess.h b/arch/riscv/include/asm/uaccess.h index f5f4f7f85543..36bba6720c26 100644 --- a/arch/riscv/include/asm/uaccess.h +++ b/arch/riscv/include/asm/uaccess.h @@ -437,10 +437,10 @@ unsigned long __must_check clear_user(void __user *to, unsigned long n) __clear_user(untagged_addr(to), n) : n; } -#define __get_kernel_nofault(dst, src, type, err_label) \ +#define arch_get_kernel_nofault(dst, src, type, err_label) \ __get_user_nocheck(*((type *)(dst)), (__force __user type *)(src), err_label) -#define __put_kernel_nofault(dst, src, type, err_label) \ +#define arch_put_kernel_nofault(dst, src, type, err_label) \ __put_user_nocheck(*((type *)(src)), (__force __user type *)(dst), err_label) static __must_check __always_inline bool user_access_begin(const void __user *ptr, size_t len) @@ -460,10 +460,10 @@ static inline void user_access_restore(unsigned long enabled) { } * We want the unsafe accessors to always be inlined and use * the error labels - thus the macro games. */ -#define unsafe_put_user(x, ptr, label) \ +#define arch_unsafe_put_user(x, ptr, label) \ __put_user_nocheck(x, (ptr), label) -#define unsafe_get_user(x, ptr, label) do { \ +#define arch_unsafe_get_user(x, ptr, label) do { \ __inttype(*(ptr)) __gu_val; \ __get_user_nocheck(__gu_val, (ptr), label); \ (x) = (__force __typeof__(*(ptr)))__gu_val; \ From 43cc54d8dbe6b761bd2672bf9bb46e5290e90277 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 27 Oct 2025 09:43:52 +0100 Subject: [PATCH 06/16] s390/uaccess: Use unsafe wrappers for ASM GOTO ASM GOTO is miscompiled by GCC when it is used inside a auto cleanup scope: bool foo(u32 __user *p, u32 val) { scoped_guard(pagefault) unsafe_put_user(val, p, efault); return true; efault: return false; } It ends up leaking the pagefault disable counter in the fault path. clang at least fails the build. S390 is not affected for unsafe_*_user() as it uses its own local label already, but __get/put_kernel_nofault() lack that. Rename them to arch_*_kernel_nofault() which makes the generic uaccess header wrap it with a local label that makes both compilers emit correct code. Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Mathieu Desnoyers Acked-by: Heiko Carstens Link: https://patch.msgid.link/20251027083745.483079889@linutronix.de --- arch/s390/include/asm/uaccess.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/arch/s390/include/asm/uaccess.h b/arch/s390/include/asm/uaccess.h index 3e5b8b677057..c5e02addcd67 100644 --- a/arch/s390/include/asm/uaccess.h +++ b/arch/s390/include/asm/uaccess.h @@ -468,8 +468,8 @@ do { \ #endif /* CONFIG_CC_HAS_ASM_GOTO_OUTPUT && CONFIG_CC_HAS_ASM_AOR_FORMAT_FLAGS */ -#define __get_kernel_nofault __mvc_kernel_nofault -#define __put_kernel_nofault __mvc_kernel_nofault +#define arch_get_kernel_nofault __mvc_kernel_nofault +#define arch_put_kernel_nofault __mvc_kernel_nofault void __cmpxchg_user_key_called_with_bad_pointer(void); From 2db48d8bf87d3cb9d968e73623efc1c5a02523e7 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 31 Oct 2025 10:37:09 +0100 Subject: [PATCH 07/16] arm64: uaccess: Use unsafe wrappers for ASM GOTO Clang propagates a provided label, which is outside of a cleanup scope to ASM GOTO despite the fact that __raw_get_mem() has a local label for that purpose: "error: cannot jump from this asm goto statement to one of its possible targets" Using the unsafe wrapper with the extra local label indirection cures that. Reported-by: Stephen Rothwell Reported-by: kernel test robot Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar --- arch/arm64/include/asm/uaccess.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/arch/arm64/include/asm/uaccess.h b/arch/arm64/include/asm/uaccess.h index 1aa4ecb73429..6490930deef8 100644 --- a/arch/arm64/include/asm/uaccess.h +++ b/arch/arm64/include/asm/uaccess.h @@ -422,9 +422,9 @@ static __must_check __always_inline bool user_access_begin(const void __user *pt } #define user_access_begin(a,b) user_access_begin(a,b) #define user_access_end() uaccess_ttbr0_disable() -#define unsafe_put_user(x, ptr, label) \ +#define arch_unsafe_put_user(x, ptr, label) \ __raw_put_mem("sttr", x, uaccess_mask_ptr(ptr), label, U) -#define unsafe_get_user(x, ptr, label) \ +#define arch_unsafe_get_user(x, ptr, label) \ __raw_get_mem("ldtr", x, uaccess_mask_ptr(ptr), label, U) /* From e497310b4ffb559e1149ee89470d5c518d234ddf Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 27 Oct 2025 09:43:55 +0100 Subject: [PATCH 08/16] uaccess: Provide scoped user access regions User space access regions are tedious and require similar code patterns all over the place: if (!user_read_access_begin(from, sizeof(*from))) return -EFAULT; unsafe_get_user(val, from, Efault); user_read_access_end(); return 0; Efault: user_read_access_end(); return -EFAULT; This got worse with the recent addition of masked user access, which optimizes the speculation prevention: if (can_do_masked_user_access()) from = masked_user_read_access_begin((from)); else if (!user_read_access_begin(from, sizeof(*from))) return -EFAULT; unsafe_get_user(val, from, Efault); user_read_access_end(); return 0; Efault: user_read_access_end(); return -EFAULT; There have been issues with using the wrong user_*_access_end() variant in the error path and other typical Copy&Pasta problems, e.g. using the wrong fault label in the user accessor which ends up using the wrong accesss end variant. These patterns beg for scopes with automatic cleanup. The resulting outcome is: scoped_user_read_access(from, Efault) unsafe_get_user(val, from, Efault); return 0; Efault: return -EFAULT; The scope guarantees the proper cleanup for the access mode is invoked both in the success and the failure (fault) path. The scoped_user_$MODE_access() macros are implemented as self terminating nested for() loops. Thanks to Andrew Cooper for pointing me at them. The scope can therefore be left with 'break', 'goto' and 'return'. Even 'continue' "works" due to the self termination mechanism. Both GCC and clang optimize all the convoluted macro maze out and the above results with clang in: b80: f3 0f 1e fa endbr64 b84: 48 b8 ef cd ab 89 67 45 23 01 movabs $0x123456789abcdef,%rax b8e: 48 39 c7 cmp %rax,%rdi b91: 48 0f 47 f8 cmova %rax,%rdi b95: 90 nop b96: 90 nop b97: 90 nop b98: 31 c9 xor %ecx,%ecx b9a: 8b 07 mov (%rdi),%eax b9c: 89 06 mov %eax,(%rsi) b9e: 85 c9 test %ecx,%ecx ba0: 0f 94 c0 sete %al ba3: 90 nop ba4: 90 nop ba5: 90 nop ba6: c3 ret Which looks as compact as it gets. The NOPs are placeholder for STAC/CLAC. GCC emits the fault path seperately: bf0: f3 0f 1e fa endbr64 bf4: 48 b8 ef cd ab 89 67 45 23 01 movabs $0x123456789abcdef,%rax bfe: 48 39 c7 cmp %rax,%rdi c01: 48 0f 47 f8 cmova %rax,%rdi c05: 90 nop c06: 90 nop c07: 90 nop c08: 31 d2 xor %edx,%edx c0a: 8b 07 mov (%rdi),%eax c0c: 89 06 mov %eax,(%rsi) c0e: 85 d2 test %edx,%edx c10: 75 09 jne c1b c12: 90 nop c13: 90 nop c14: 90 nop c15: b8 01 00 00 00 mov $0x1,%eax c1a: c3 ret c1b: 90 nop c1c: 90 nop c1d: 90 nop c1e: 31 c0 xor %eax,%eax c20: c3 ret The fault labels for the scoped*() macros and the fault labels for the actual user space accessors can be shared and must be placed outside of the scope. If masked user access is enabled on an architecture, then the pointer handed in to scoped_user_$MODE_access() can be modified to point to a guaranteed faulting user address. This modification is only scope local as the pointer is aliased inside the scope. When the scope is left the alias is not longer in effect. IOW the original pointer value is preserved so it can be used e.g. for fixup or diagnostic purposes in the fault path. Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Reviewed-by: Mathieu Desnoyers Link: https://patch.msgid.link/20251027083745.546420421@linutronix.de --- include/linux/uaccess.h | 192 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 192 insertions(+) diff --git a/include/linux/uaccess.h b/include/linux/uaccess.h index 8aa82b1d6013..5f142c05b0dc 100644 --- a/include/linux/uaccess.h +++ b/include/linux/uaccess.h @@ -2,6 +2,7 @@ #ifndef __LINUX_UACCESS_H__ #define __LINUX_UACCESS_H__ +#include #include #include #include @@ -35,9 +36,17 @@ #ifdef masked_user_access_begin #define can_do_masked_user_access() 1 +# ifndef masked_user_write_access_begin +# define masked_user_write_access_begin masked_user_access_begin +# endif +# ifndef masked_user_read_access_begin +# define masked_user_read_access_begin masked_user_access_begin +#endif #else #define can_do_masked_user_access() 0 #define masked_user_access_begin(src) NULL + #define masked_user_read_access_begin(src) NULL + #define masked_user_write_access_begin(src) NULL #define mask_user_address(src) (src) #endif @@ -633,6 +642,189 @@ static inline void user_access_restore(unsigned long flags) { } #define user_read_access_end user_access_end #endif +/* Define RW variant so the below _mode macro expansion works */ +#define masked_user_rw_access_begin(u) masked_user_access_begin(u) +#define user_rw_access_begin(u, s) user_access_begin(u, s) +#define user_rw_access_end() user_access_end() + +/* Scoped user access */ +#define USER_ACCESS_GUARD(_mode) \ +static __always_inline void __user * \ +class_user_##_mode##_begin(void __user *ptr) \ +{ \ + return ptr; \ +} \ + \ +static __always_inline void \ +class_user_##_mode##_end(void __user *ptr) \ +{ \ + user_##_mode##_access_end(); \ +} \ + \ +DEFINE_CLASS(user_ ##_mode## _access, void __user *, \ + class_user_##_mode##_end(_T), \ + class_user_##_mode##_begin(ptr), void __user *ptr) \ + \ +static __always_inline class_user_##_mode##_access_t \ +class_user_##_mode##_access_ptr(void __user *scope) \ +{ \ + return scope; \ +} + +USER_ACCESS_GUARD(read) +USER_ACCESS_GUARD(write) +USER_ACCESS_GUARD(rw) +#undef USER_ACCESS_GUARD + +/** + * __scoped_user_access_begin - Start a scoped user access + * @mode: The mode of the access class (read, write, rw) + * @uptr: The pointer to access user space memory + * @size: Size of the access + * @elbl: Error label to goto when the access region is rejected + * + * Internal helper for __scoped_user_access(). Don't use directly. + */ +#define __scoped_user_access_begin(mode, uptr, size, elbl) \ +({ \ + typeof(uptr) __retptr; \ + \ + if (can_do_masked_user_access()) { \ + __retptr = masked_user_##mode##_access_begin(uptr); \ + } else { \ + __retptr = uptr; \ + if (!user_##mode##_access_begin(uptr, size)) \ + goto elbl; \ + } \ + __retptr; \ +}) + +/** + * __scoped_user_access - Open a scope for user access + * @mode: The mode of the access class (read, write, rw) + * @uptr: The pointer to access user space memory + * @size: Size of the access + * @elbl: Error label to goto when the access region is rejected. It + * must be placed outside the scope + * + * If the user access function inside the scope requires a fault label, it + * can use @elbl or a different label outside the scope, which requires + * that user access which is implemented with ASM GOTO has been properly + * wrapped. See unsafe_get_user() for reference. + * + * scoped_user_rw_access(ptr, efault) { + * unsafe_get_user(rval, &ptr->rval, efault); + * unsafe_put_user(wval, &ptr->wval, efault); + * } + * return 0; + * efault: + * return -EFAULT; + * + * The scope is internally implemented as a autoterminating nested for() + * loop, which can be left with 'return', 'break' and 'goto' at any + * point. + * + * When the scope is left user_##@_mode##_access_end() is automatically + * invoked. + * + * When the architecture supports masked user access and the access region + * which is determined by @uptr and @size is not a valid user space + * address, i.e. < TASK_SIZE, the scope sets the pointer to a faulting user + * space address and does not terminate early. This optimizes for the good + * case and lets the performance uncritical bad case go through the fault. + * + * The eventual modification of the pointer is limited to the scope. + * Outside of the scope the original pointer value is unmodified, so that + * the original pointer value is available for diagnostic purposes in an + * out of scope fault path. + * + * Nesting scoped user access into a user access scope is invalid and fails + * the build. Nesting into other guards, e.g. pagefault is safe. + * + * The masked variant does not check the size of the access and relies on a + * mapping hole (e.g. guard page) to catch an out of range pointer, the + * first access to user memory inside the scope has to be within + * @uptr ... @uptr + PAGE_SIZE - 1 + * + * Don't use directly. Use scoped_masked_user_$MODE_access() instead. + */ +#define __scoped_user_access(mode, uptr, size, elbl) \ +for (bool done = false; !done; done = true) \ + for (void __user *_tmpptr = __scoped_user_access_begin(mode, uptr, size, elbl); \ + !done; done = true) \ + for (CLASS(user_##mode##_access, scope)(_tmpptr); !done; done = true) \ + /* Force modified pointer usage within the scope */ \ + for (const typeof(uptr) uptr = _tmpptr; !done; done = true) + +/** + * scoped_user_read_access_size - Start a scoped user read access with given size + * @usrc: Pointer to the user space address to read from + * @size: Size of the access starting from @usrc + * @elbl: Error label to goto when the access region is rejected + * + * For further information see __scoped_user_access() above. + */ +#define scoped_user_read_access_size(usrc, size, elbl) \ + __scoped_user_access(read, usrc, size, elbl) + +/** + * scoped_user_read_access - Start a scoped user read access + * @usrc: Pointer to the user space address to read from + * @elbl: Error label to goto when the access region is rejected + * + * The size of the access starting from @usrc is determined via sizeof(*@usrc)). + * + * For further information see __scoped_user_access() above. + */ +#define scoped_user_read_access(usrc, elbl) \ + scoped_user_read_access_size(usrc, sizeof(*(usrc)), elbl) + +/** + * scoped_user_write_access_size - Start a scoped user write access with given size + * @udst: Pointer to the user space address to write to + * @size: Size of the access starting from @udst + * @elbl: Error label to goto when the access region is rejected + * + * For further information see __scoped_user_access() above. + */ +#define scoped_user_write_access_size(udst, size, elbl) \ + __scoped_user_access(write, udst, size, elbl) + +/** + * scoped_user_write_access - Start a scoped user write access + * @udst: Pointer to the user space address to write to + * @elbl: Error label to goto when the access region is rejected + * + * The size of the access starting from @udst is determined via sizeof(*@udst)). + * + * For further information see __scoped_user_access() above. + */ +#define scoped_user_write_access(udst, elbl) \ + scoped_user_write_access_size(udst, sizeof(*(udst)), elbl) + +/** + * scoped_user_rw_access_size - Start a scoped user read/write access with given size + * @uptr Pointer to the user space address to read from and write to + * @size: Size of the access starting from @uptr + * @elbl: Error label to goto when the access region is rejected + * + * For further information see __scoped_user_access() above. + */ +#define scoped_user_rw_access_size(uptr, size, elbl) \ + __scoped_user_access(rw, uptr, size, elbl) + +/** + * scoped_user_rw_access - Start a scoped user read/write access + * @uptr Pointer to the user space address to read from and write to + * @elbl: Error label to goto when the access region is rejected + * + * The size of the access starting from @uptr is determined via sizeof(*@uptr)). + * + * For further information see __scoped_user_access() above. + */ +#define scoped_user_rw_access(uptr, elbl) \ + scoped_user_rw_access_size(uptr, sizeof(*(uptr)), elbl) + #ifdef CONFIG_HARDENED_USERCOPY void __noreturn usercopy_abort(const char *name, const char *detail, bool to_user, unsigned long offset, From b2cfc0cd68b830dde80fce2406580e258a1e976d Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 27 Oct 2025 09:43:56 +0100 Subject: [PATCH 09/16] uaccess: Provide put/get_user_inline() Provide convenience wrappers around scoped user access similar to put/get_user(), which reduce the usage sites to: if (!get_user_inline(val, ptr)) return -EFAULT; Should only be used if there is a demonstrable performance benefit. Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Reviewed-by: Mathieu Desnoyers Link: https://patch.msgid.link/20251027083745.609031602@linutronix.de --- include/linux/uaccess.h | 50 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 50 insertions(+) diff --git a/include/linux/uaccess.h b/include/linux/uaccess.h index 5f142c05b0dc..be395f5f7ee3 100644 --- a/include/linux/uaccess.h +++ b/include/linux/uaccess.h @@ -825,6 +825,56 @@ for (bool done = false; !done; done = true) \ #define scoped_user_rw_access(uptr, elbl) \ scoped_user_rw_access_size(uptr, sizeof(*(uptr)), elbl) +/** + * get_user_inline - Read user data inlined + * @val: The variable to store the value read from user memory + * @usrc: Pointer to the user space memory to read from + * + * Return: 0 if successful, -EFAULT when faulted + * + * Inlined variant of get_user(). Only use when there is a demonstrable + * performance reason. + */ +#define get_user_inline(val, usrc) \ +({ \ + __label__ efault; \ + typeof(usrc) _tmpsrc = usrc; \ + int _ret = 0; \ + \ + scoped_user_read_access(_tmpsrc, efault) \ + unsafe_get_user(val, _tmpsrc, efault); \ + if (0) { \ + efault: \ + _ret = -EFAULT; \ + } \ + _ret; \ +}) + +/** + * put_user_inline - Write to user memory inlined + * @val: The value to write + * @udst: Pointer to the user space memory to write to + * + * Return: 0 if successful, -EFAULT when faulted + * + * Inlined variant of put_user(). Only use when there is a demonstrable + * performance reason. + */ +#define put_user_inline(val, udst) \ +({ \ + __label__ efault; \ + typeof(udst) _tmpdst = udst; \ + int _ret = 0; \ + \ + scoped_user_write_access(_tmpdst, efault) \ + unsafe_put_user(val, _tmpdst, efault); \ + if (0) { \ + efault: \ + _ret = -EFAULT; \ + } \ + _ret; \ +}) + #ifdef CONFIG_HARDENED_USERCOPY void __noreturn usercopy_abort(const char *name, const char *detail, bool to_user, unsigned long offset, From e4e28fd6986e8cf963ec4137e6c0b95403f636ab Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 27 Oct 2025 09:44:00 +0100 Subject: [PATCH 10/16] futex: Convert to get/put_user_inline() Replace the open coded implementation with the new get/put_user_inline() helpers. This might be replaced by a regular get/put_user(), but that needs a proper performance evaluation. No functional change intended. Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Link: https://patch.msgid.link/20251027083745.736737934@linutronix.de --- kernel/futex/core.c | 4 +-- kernel/futex/futex.h | 58 +++----------------------------------------- 2 files changed, 5 insertions(+), 57 deletions(-) diff --git a/kernel/futex/core.c b/kernel/futex/core.c index 125804fbb5cb..ebcccb16ae0b 100644 --- a/kernel/futex/core.c +++ b/kernel/futex/core.c @@ -581,7 +581,7 @@ int get_futex_key(u32 __user *uaddr, unsigned int flags, union futex_key *key, if (flags & FLAGS_NUMA) { u32 __user *naddr = (void *)uaddr + size / 2; - if (futex_get_value(&node, naddr)) + if (get_user_inline(node, naddr)) return -EFAULT; if ((node != FUTEX_NO_NODE) && @@ -601,7 +601,7 @@ int get_futex_key(u32 __user *uaddr, unsigned int flags, union futex_key *key, node = numa_node_id(); node_updated = true; } - if (node_updated && futex_put_value(node, naddr)) + if (node_updated && put_user_inline(node, naddr)) return -EFAULT; } diff --git a/kernel/futex/futex.h b/kernel/futex/futex.h index 2cd57096c38e..30c2afa03889 100644 --- a/kernel/futex/futex.h +++ b/kernel/futex/futex.h @@ -281,63 +281,11 @@ static inline int futex_cmpxchg_value_locked(u32 *curval, u32 __user *uaddr, u32 return ret; } -/* - * This does a plain atomic user space read, and the user pointer has - * already been verified earlier by get_futex_key() to be both aligned - * and actually in user space, just like futex_atomic_cmpxchg_inatomic(). - * - * We still want to avoid any speculation, and while __get_user() is - * the traditional model for this, it's actually slower than doing - * this manually these days. - * - * We could just have a per-architecture special function for it, - * the same way we do futex_atomic_cmpxchg_inatomic(), but rather - * than force everybody to do that, write it out long-hand using - * the low-level user-access infrastructure. - * - * This looks a bit overkill, but generally just results in a couple - * of instructions. - */ -static __always_inline int futex_get_value(u32 *dest, u32 __user *from) -{ - u32 val; - - if (can_do_masked_user_access()) - from = masked_user_access_begin(from); - else if (!user_read_access_begin(from, sizeof(*from))) - return -EFAULT; - unsafe_get_user(val, from, Efault); - user_read_access_end(); - *dest = val; - return 0; -Efault: - user_read_access_end(); - return -EFAULT; -} - -static __always_inline int futex_put_value(u32 val, u32 __user *to) -{ - if (can_do_masked_user_access()) - to = masked_user_access_begin(to); - else if (!user_write_access_begin(to, sizeof(*to))) - return -EFAULT; - unsafe_put_user(val, to, Efault); - user_write_access_end(); - return 0; -Efault: - user_write_access_end(); - return -EFAULT; -} - +/* Read from user memory with pagefaults disabled */ static inline int futex_get_value_locked(u32 *dest, u32 __user *from) { - int ret; - - pagefault_disable(); - ret = futex_get_value(dest, from); - pagefault_enable(); - - return ret; + guard(pagefault)(); + return get_user_inline(*dest, from); } extern void __futex_unqueue(struct futex_q *q); From e02718c9865c7cbcb7959044a704b3dc5929640e Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 27 Oct 2025 09:44:02 +0100 Subject: [PATCH 11/16] x86/futex: Convert to scoped user access Replace the open coded implementation with the scoped user access guards No functional change intended. Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Link: https://patch.msgid.link/20251027083745.799714344@linutronix.de --- arch/x86/include/asm/futex.h | 75 ++++++++++++++++-------------------- 1 file changed, 33 insertions(+), 42 deletions(-) diff --git a/arch/x86/include/asm/futex.h b/arch/x86/include/asm/futex.h index 6e2458088800..fe5d9a10d900 100644 --- a/arch/x86/include/asm/futex.h +++ b/arch/x86/include/asm/futex.h @@ -46,38 +46,31 @@ do { \ } while(0) static __always_inline int arch_futex_atomic_op_inuser(int op, int oparg, int *oval, - u32 __user *uaddr) + u32 __user *uaddr) { - if (can_do_masked_user_access()) - uaddr = masked_user_access_begin(uaddr); - else if (!user_access_begin(uaddr, sizeof(u32))) - return -EFAULT; - - switch (op) { - case FUTEX_OP_SET: - unsafe_atomic_op1("xchgl %0, %2", oval, uaddr, oparg, Efault); - break; - case FUTEX_OP_ADD: - unsafe_atomic_op1(LOCK_PREFIX "xaddl %0, %2", oval, - uaddr, oparg, Efault); - break; - case FUTEX_OP_OR: - unsafe_atomic_op2("orl %4, %3", oval, uaddr, oparg, Efault); - break; - case FUTEX_OP_ANDN: - unsafe_atomic_op2("andl %4, %3", oval, uaddr, ~oparg, Efault); - break; - case FUTEX_OP_XOR: - unsafe_atomic_op2("xorl %4, %3", oval, uaddr, oparg, Efault); - break; - default: - user_access_end(); - return -ENOSYS; + scoped_user_rw_access(uaddr, Efault) { + switch (op) { + case FUTEX_OP_SET: + unsafe_atomic_op1("xchgl %0, %2", oval, uaddr, oparg, Efault); + break; + case FUTEX_OP_ADD: + unsafe_atomic_op1(LOCK_PREFIX "xaddl %0, %2", oval, uaddr, oparg, Efault); + break; + case FUTEX_OP_OR: + unsafe_atomic_op2("orl %4, %3", oval, uaddr, oparg, Efault); + break; + case FUTEX_OP_ANDN: + unsafe_atomic_op2("andl %4, %3", oval, uaddr, ~oparg, Efault); + break; + case FUTEX_OP_XOR: + unsafe_atomic_op2("xorl %4, %3", oval, uaddr, oparg, Efault); + break; + default: + return -ENOSYS; + } } - user_access_end(); return 0; Efault: - user_access_end(); return -EFAULT; } @@ -86,21 +79,19 @@ static inline int futex_atomic_cmpxchg_inatomic(u32 *uval, u32 __user *uaddr, { int ret = 0; - if (can_do_masked_user_access()) - uaddr = masked_user_access_begin(uaddr); - else if (!user_access_begin(uaddr, sizeof(u32))) - return -EFAULT; - asm volatile("\n" - "1:\t" LOCK_PREFIX "cmpxchgl %3, %2\n" - "2:\n" - _ASM_EXTABLE_TYPE_REG(1b, 2b, EX_TYPE_EFAULT_REG, %0) \ - : "+r" (ret), "=a" (oldval), "+m" (*uaddr) - : "r" (newval), "1" (oldval) - : "memory" - ); - user_access_end(); - *uval = oldval; + scoped_user_rw_access(uaddr, Efault) { + asm_inline volatile("\n" + "1:\t" LOCK_PREFIX "cmpxchgl %3, %2\n" + "2:\n" + _ASM_EXTABLE_TYPE_REG(1b, 2b, EX_TYPE_EFAULT_REG, %0) + : "+r" (ret), "=a" (oldval), "+m" (*uaddr) + : "r" (newval), "1" (oldval) + : "memory"); + *uval = oldval; + } return ret; +Efault: + return -EFAULT; } #endif From 3ce17e6909944b3f83b54915e36f5957f1327712 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 27 Oct 2025 09:44:04 +0100 Subject: [PATCH 12/16] select: Convert to scoped user access Replace the open coded implementation with the scoped user access guard. No functional change intended. Signed-off-by: Thomas Gleixner Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Reviewed-by: Mathieu Desnoyers Link: https://patch.msgid.link/20251027083745.862419776@linutronix.de --- fs/select.c | 12 ++++-------- 1 file changed, 4 insertions(+), 8 deletions(-) diff --git a/fs/select.c b/fs/select.c index 082cf60c7e23..65019b8ba3f7 100644 --- a/fs/select.c +++ b/fs/select.c @@ -776,17 +776,13 @@ static inline int get_sigset_argpack(struct sigset_argpack *to, { // the path is hot enough for overhead of copy_from_user() to matter if (from) { - if (can_do_masked_user_access()) - from = masked_user_access_begin(from); - else if (!user_read_access_begin(from, sizeof(*from))) - return -EFAULT; - unsafe_get_user(to->p, &from->p, Efault); - unsafe_get_user(to->size, &from->size, Efault); - user_read_access_end(); + scoped_user_read_access(from, Efault) { + unsafe_get_user(to->p, &from->p, Efault); + unsafe_get_user(to->size, &from->size, Efault); + } } return 0; Efault: - user_read_access_end(); return -EFAULT; } From 4db1df7a7217827ee7f8a3414932e250f1ac2204 Mon Sep 17 00:00:00 2001 From: Christophe Leroy Date: Mon, 17 Nov 2025 17:43:41 +0100 Subject: [PATCH 13/16] iov_iter: Convert copy_from_user_iter() to masked user access copy_from_user_iter() lacks a speculation barrier, which will degrade performance on some architecture like x86, which would be unfortunate as copy_from_user_iter() is a critical hotpath function. Convert copy_from_user_iter() to using masked user access on architecture that support it. This allows to add the speculation barrier without impacting performance. This is similar to what was done for copy_from_user() in commit 0fc810ae3ae1 ("x86/uaccess: Avoid barrier_nospec() in 64-bit copy_from_user()") [ tglx: Massage change log ] Signed-off-by: Christophe Leroy Signed-off-by: Thomas Gleixner Link: https://patch.msgid.link/58e4b07d469ca68a2b9477fe2c1ccc8a44cef131.1763396724.git.christophe.leroy@csgroup.eu --- lib/iov_iter.c | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-) diff --git a/lib/iov_iter.c b/lib/iov_iter.c index 2fe66a6b8789..a589935bf302 100644 --- a/lib/iov_iter.c +++ b/lib/iov_iter.c @@ -49,12 +49,16 @@ size_t copy_from_user_iter(void __user *iter_from, size_t progress, if (should_fail_usercopy()) return len; - if (access_ok(iter_from, len)) { - to += progress; - instrument_copy_from_user_before(to, iter_from, len); - res = raw_copy_from_user(to, iter_from, len); - instrument_copy_from_user_after(to, iter_from, len, res); - } + if (can_do_masked_user_access()) + iter_from = mask_user_address(iter_from); + else if (!access_ok(iter_from, len)) + return res; + + to += progress; + instrument_copy_from_user_before(to, iter_from, len); + res = raw_copy_from_user(to, iter_from, len); + instrument_copy_from_user_after(to, iter_from, len, res); + return res; } From 803abedbd540617f136a2c4d7066ff2e304f016d Mon Sep 17 00:00:00 2001 From: Christophe Leroy Date: Mon, 17 Nov 2025 17:43:42 +0100 Subject: [PATCH 14/16] iov_iter: Add missing speculation barrier to copy_from_user_iter() The results of "access_ok()" can be mis-speculated. The result is that the CPU can end speculatively: if (access_ok(from, size)) // Right here For the same reason as done in copy_from_user() in commit 74e19ef0ff80 ("uaccess: Add speculation barrier to copy_from_user()"), add a speculation barrier to copy_from_user_iter(). Signed-off-by: Christophe Leroy Signed-off-by: Thomas Gleixner Link: https://patch.msgid.link/6b73e69cc7168c89df4eab0a216e3ed4cca36b0a.1763396724.git.christophe.leroy@csgroup.eu --- lib/iov_iter.c | 14 +++++++++++--- 1 file changed, 11 insertions(+), 3 deletions(-) diff --git a/lib/iov_iter.c b/lib/iov_iter.c index a589935bf302..896760bad455 100644 --- a/lib/iov_iter.c +++ b/lib/iov_iter.c @@ -49,11 +49,19 @@ size_t copy_from_user_iter(void __user *iter_from, size_t progress, if (should_fail_usercopy()) return len; - if (can_do_masked_user_access()) + if (can_do_masked_user_access()) { iter_from = mask_user_address(iter_from); - else if (!access_ok(iter_from, len)) - return res; + } else { + if (!access_ok(iter_from, len)) + return res; + /* + * Ensure that bad access_ok() speculation will not + * lead to nasty side effects *after* the copy is + * finished: + */ + barrier_nospec(); + } to += progress; instrument_copy_from_user_before(to, iter_from, len); res = raw_copy_from_user(to, iter_from, len); From 1c204914bc4401623a1b242305c583060a0b7e4f Mon Sep 17 00:00:00 2001 From: Christophe Leroy Date: Mon, 17 Nov 2025 17:43:43 +0100 Subject: [PATCH 15/16] scm: Convert put_cmsg() to scoped user access Replace the open coded implementation with the scoped user access guard. That also corrects the imbalance between masked_user_access_begin() and user_write_access_end(), which would affect PowerPC when it gains masked user access support. No functional change intended. [ tglx: Amend change log ] Signed-off-by: Christophe Leroy Signed-off-by: Thomas Gleixner Link: https://patch.msgid.link/793219313f641eda09a892d06768d2837246bf9f.1763396724.git.christophe.leroy@csgroup.eu --- net/core/scm.c | 20 +++++++------------- 1 file changed, 7 insertions(+), 13 deletions(-) diff --git a/net/core/scm.c b/net/core/scm.c index 66eaee783e8b..cd87f66671aa 100644 --- a/net/core/scm.c +++ b/net/core/scm.c @@ -273,17 +273,13 @@ int put_cmsg(struct msghdr * msg, int level, int type, int len, void *data) check_object_size(data, cmlen - sizeof(*cm), true); - if (can_do_masked_user_access()) - cm = masked_user_access_begin(cm); - else if (!user_write_access_begin(cm, cmlen)) - goto efault; - - unsafe_put_user(cmlen, &cm->cmsg_len, efault_end); - unsafe_put_user(level, &cm->cmsg_level, efault_end); - unsafe_put_user(type, &cm->cmsg_type, efault_end); - unsafe_copy_to_user(CMSG_USER_DATA(cm), data, - cmlen - sizeof(*cm), efault_end); - user_write_access_end(); + scoped_user_write_access_size(cm, cmlen, efault) { + unsafe_put_user(cmlen, &cm->cmsg_len, efault); + unsafe_put_user(level, &cm->cmsg_level, efault); + unsafe_put_user(type, &cm->cmsg_type, efault); + unsafe_copy_to_user(CMSG_USER_DATA(cm), data, + cmlen - sizeof(*cm), efault); + } } else { struct cmsghdr *cm = msg->msg_control; @@ -301,8 +297,6 @@ int put_cmsg(struct msghdr * msg, int level, int type, int len, void *data) msg->msg_controllen -= cmlen; return 0; -efault_end: - user_write_access_end(); efault: return -EFAULT; } From 4322c8f81c58da493a3c46eda32f0e7534a350a0 Mon Sep 17 00:00:00 2001 From: Christophe Leroy Date: Mon, 17 Nov 2025 17:43:44 +0100 Subject: [PATCH 16/16] lib/strn*,uaccess: Use masked_user_{read/write}_access_begin when required Properly use masked_user_read_access_begin() and masked_user_write_access_begin() instead of masked_user_access_begin() in order to match user_read_access_end() and user_write_access_end(). This is important for architectures like PowerPC that enable separately user reads and user writes. That means masked_user_read_access_begin() is used when user memory is exclusively read during the window and masked_user_write_access_begin() is used when user memory is exclusively writen during the window. masked_user_access_begin() remains and is used when both reads and writes are performed during the open window. Each of them is expected to be terminated by the matching user_read_access_end(), user_write_access_end() and user_access_end(). Signed-off-by: Christophe Leroy Signed-off-by: Thomas Gleixner Link: https://patch.msgid.link/cb5e4b0fa49ea9c740570949d5e3544423389757.1763396724.git.christophe.leroy@csgroup.eu --- lib/strncpy_from_user.c | 2 +- lib/strnlen_user.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/lib/strncpy_from_user.c b/lib/strncpy_from_user.c index 6dc234913dd5..5bb752ff7c61 100644 --- a/lib/strncpy_from_user.c +++ b/lib/strncpy_from_user.c @@ -126,7 +126,7 @@ long strncpy_from_user(char *dst, const char __user *src, long count) if (can_do_masked_user_access()) { long retval; - src = masked_user_access_begin(src); + src = masked_user_read_access_begin(src); retval = do_strncpy_from_user(dst, src, count, count); user_read_access_end(); return retval; diff --git a/lib/strnlen_user.c b/lib/strnlen_user.c index 6e489f9e90f1..4a6574b67f82 100644 --- a/lib/strnlen_user.c +++ b/lib/strnlen_user.c @@ -99,7 +99,7 @@ long strnlen_user(const char __user *str, long count) if (can_do_masked_user_access()) { long retval; - str = masked_user_access_begin(str); + str = masked_user_read_access_begin(str); retval = do_strnlen_user(str, count, count); user_read_access_end(); return retval;