Re: [PATCH] Convert filldir[64]() from __put_user() to unsafe_put_user()

[Linus Torvalds]

On Sun, Oct 6, 2019 at 3:20 PM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
>
> this patch causes all my sparc64 emulations to stall during boot. It causes
> all alpha emulations to crash with [1a] and [1b] when booting from a virtual
> disk, and one of the xtensa emulations to crash with [2].

So I think your alpha emulation environment may be broken, because
Michael Cree reports that it works for him on real hardware, but he
does see the kernel unaligned count being high.

But regardless, this is my current fairly minimal patch that I think
should fix the unaligned issue, while still giving the behavior we
want on x86. I hope Al can do something nicer, but I think this is
"acceptable".

I'm running this now on x86, and I verified that x86-32 code
generation looks sane too, but it woudl be good to verify that this
makes the alignment issue go away on other architectures.

                Linus
 arch/x86/include/asm/uaccess.h | 23 ++++++++++++++++++++++
 fs/readdir.c                   | 44 ++----------------------------------------
 include/linux/uaccess.h        |  6 ++++--
 3 files changed, 29 insertions(+), 44 deletions(-)

diff --git a/arch/x86/include/asm/uaccess.h b/arch/x86/include/asm/uaccess.h
index 35c225ede0e4..61d93f062a36 100644
--- a/arch/x86/include/asm/uaccess.h
+++ b/arch/x86/include/asm/uaccess.h
@@ -734,5 +734,28 @@ do {										\
 	if (unlikely(__gu_err)) goto err_label;					\
 } while (0)
 
+/*
+ * We want the unsafe accessors to always be inlined and use
+ * the error labels - thus the macro games.
+ */
+#define unsafe_copy_loop(dst, src, len, type, label)			\
+	while (len >= sizeof(type)) {					\
+		unsafe_put_user(*(type *)src,(type __user *)dst,label);	\
+		dst += sizeof(type);					\
+		src += sizeof(type);					\
+		len -= sizeof(type);					\
+	}
+
+#define unsafe_copy_to_user(_dst,_src,_len,label)			\
+do {									\
+	char __user *__ucu_dst = (_dst);				\
+	const char *__ucu_src = (_src);					\
+	size_t __ucu_len = (_len);					\
+	unsafe_copy_loop(__ucu_dst, __ucu_src, __ucu_len, u64, label);	\
+	unsafe_copy_loop(__ucu_dst, __ucu_src, __ucu_len, u32, label);	\
+	unsafe_copy_loop(__ucu_dst, __ucu_src, __ucu_len, u16, label);	\
+	unsafe_copy_loop(__ucu_dst, __ucu_src, __ucu_len, u8, label);	\
+} while (0)
+
 #endif /* _ASM_X86_UACCESS_H */
 
diff --git a/fs/readdir.c b/fs/readdir.c
index 19bea591c3f1..6e2623e57b2e 100644
--- a/fs/readdir.c
+++ b/fs/readdir.c
@@ -27,53 +27,13 @@
 /*
  * Note the "unsafe_put_user() semantics: we goto a
  * label for errors.
- *
- * Also note how we use a "while()" loop here, even though
- * only the biggest size needs to loop. The compiler (well,
- * at least gcc) is smart enough to turn the smaller sizes
- * into just if-statements, and this way we don't need to
- * care whether 'u64' or 'u32' is the biggest size.
- */
-#define unsafe_copy_loop(dst, src, len, type, label) 		\
-	while (len >= sizeof(type)) {				\
-		unsafe_put_user(get_unaligned((type *)src),	\
-			(type __user *)dst, label);		\
-		dst += sizeof(type);				\
-		src += sizeof(type);				\
-		len -= sizeof(type);				\
-	}
-
-/*
- * We avoid doing 64-bit copies on 32-bit architectures. They
- * might be better, but the component names are mostly small,
- * and the 64-bit cases can end up being much more complex and
- * put much more register pressure on the code, so it's likely
- * not worth the pain of unaligned accesses etc.
- *
- * So limit the copies to "unsigned long" size. I did verify
- * that at least the x86-32 case is ok without this limiting,
- * but I worry about random other legacy 32-bit cases that
- * might not do as well.
- */
-#define unsafe_copy_type(dst, src, len, type, label) do {	\
-	if (sizeof(type) <= sizeof(unsigned long))		\
-		unsafe_copy_loop(dst, src, len, type, label);	\
-} while (0)
-
-/*
- * Copy the dirent name to user space, and NUL-terminate
- * it. This should not be a function call, since we're doing
- * the copy inside a "user_access_begin/end()" section.
  */
 #define unsafe_copy_dirent_name(_dst, _src, _len, label) do {	\
 	char __user *dst = (_dst);				\
 	const char *src = (_src);				\
 	size_t len = (_len);					\
-	unsafe_copy_type(dst, src, len, u64, label);	 	\
-	unsafe_copy_type(dst, src, len, u32, label);		\
-	unsafe_copy_type(dst, src, len, u16, label);		\
-	unsafe_copy_type(dst, src, len, u8,  label);		\
-	unsafe_put_user(0, dst, label);				\
+	unsafe_put_user(0, dst+len, label);			\
+	unsafe_copy_to_user(dst, src, len, label);		\
 } while (0)
 
 
diff --git a/include/linux/uaccess.h b/include/linux/uaccess.h
index e47d0522a1f4..d4ee6e942562 100644
--- a/include/linux/uaccess.h
+++ b/include/linux/uaccess.h
@@ -355,8 +355,10 @@ extern long strnlen_unsafe_user(const void __user *unsafe_addr, long count);
 #ifndef user_access_begin
 #define user_access_begin(ptr,len) access_ok(ptr, len)
 #define user_access_end() do { } while (0)
-#define unsafe_get_user(x, ptr, err) do { if (unlikely(__get_user(x, ptr))) goto err; } while (0)
-#define unsafe_put_user(x, ptr, err) do { if (unlikely(__put_user(x, ptr))) goto err; } while (0)
+#define unsafe_op_wrap(op, err) do { if (unlikely(op)) goto err; } while (0)
+#define unsafe_get_user(x,p,e) unsafe_op_wrap(__get_user(x,p),e)
+#define unsafe_put_user(x,p,e) unsafe_op_wrap(__put_user(x,p),e)
+#define unsafe_copy_to_user(d,s,l,e) unsafe_op_wrap(__copy_to_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