Skip to contentSkip to conversation

x86/asm: add __user on copy_user_handle_tail() pointers

PATCH In review Ben Dooks 3 participants 3 messages Mar 30, 2019 – Apr 2, 2019 x86

lore
Share & tools
lore

Ben Dooks

author
#0 lore raw

PATCH v2 1/1 x86/asm: add __user on copy_user_handle_tail() pointers

The copy_user_handle_tail() clearly uses both from and to as pointers to user-space memory. This triggers sparse warning on using the calls to get and put to user-space. This can be fixed easily by changing the call to take __user annotated pointer.s

arch/x86/lib/usercopy_64.c:68:21: warning: incorrect type in argument 1 (different address spaces)
arch/x86/lib/usercopy_64.c:68:21: expected void const volatile [noderef] <asn:1>*<noident>
arch/x86/lib/usercopy_64.c:68:21: got char *
arch/x86/lib/usercopy_64.c:70:21: warning: incorrect type in argument 1 (different address spaces)
arch/x86/lib/usercopy_64.c:70:21: expected void const volatile [noderef] <asn:1>*<noident>
arch/x86/lib/usercopy_64.c:70:21: got char *to

From Linus Torvalds:

On Thu, Mar 28, 2019 at 12:24 AM Borislav Petkov <[email protected]> wrote:

Well, but copy_user_generic() (which ends up calling the copy_user_handle_tail() eventually) casts those __user pointers to (__force void *). Converting them back to __user looks strange to me.

Linus?

Well, it does that because the x86 version of copy_user_generic() can work in either direction, so it works when either the source or destination (or both) are user pointers, but they don't _have_ to be.

So the "userness" of a pointer in that context is a bit ambiguous, and so we've picked the pointers to be just plain "void *".

That said, arguably we should have gone the other way and just made them both "__user" pointers, and do the cast the other way around.

But there's no absolutely right answer here, and nobody should ever use copy_user_generic() directly (ie it is very much meant to be only used as a internal helper for the cases that get the pointer annotations right).

---
 arch/x86/include/asm/uaccess_64.h | 2 +-
 arch/x86/lib/usercopy_64.c        | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
arch/x86/include/asm/uaccess_64.h +1 -1
index a9d637bc301d..cbca2cb28939 100644
@@ -208,7 +208,7 @@ __copy_from_user_flushcache(void *dst, const void __user *src, unsigned size)
 }
 
 unsigned long
-copy_user_handle_tail(char *to, char *from, unsigned len);
+copy_user_handle_tail(char __user *to, char __user *from, unsigned len);
 
 unsigned long
 mcsafe_handle_tail(char *to, char *from, unsigned len);
arch/x86/lib/usercopy_64.c +1 -1
index ee42bb0cbeb3..aa180424e77a 100644
@@ -60,7 +60,7 @@ EXPORT_SYMBOL(clear_user);
  * it is not necessary to optimize tail handling.
  */
 __visible unsigned long
-copy_user_handle_tail(char *to, char *from, unsigned len)
+copy_user_handle_tail(char __user *to, char __user *from, unsigned len)
 {
 	for (; len; --len, to++) {
 		char c;
Signature · 1 line
2.20.1
Reviewed-by on v2 1/1
#1 lore raw
Quoted 32 lines · Ben Dooks

The copy_user_handle_tail() clearly uses both from and to as pointers to user-space memory. This triggers sparse warning on using the calls to get and put to user-space. This can be fixed easily by changing the call to take __user annotated pointer.s

arch/x86/lib/usercopy_64.c:68:21: warning: incorrect type in argument 1 (different address spaces)
arch/x86/lib/usercopy_64.c:68:21: expected void const volatile [noderef] <asn:1>*<noident>
arch/x86/lib/usercopy_64.c:68:21: got char *
arch/x86/lib/usercopy_64.c:70:21: warning: incorrect type in argument 1 (different address spaces)
arch/x86/lib/usercopy_64.c:70:21: expected void const volatile [noderef] <asn:1>*<noident>
arch/x86/lib/usercopy_64.c:70:21: got char *to

From Linus Torvalds:

On Thu, Mar 28, 2019 at 12:24 AM Borislav Petkov <[email protected]> wrote:

Well, but copy_user_generic() (which ends up calling the copy_user_handle_tail() eventually) casts those __user pointers to (__force void *). Converting them back to __user looks strange to me.

Linus?

Well, it does that because the x86 version of copy_user_generic() can work in either direction, so it works when either the source or destination (or both) are user pointers, but they don't _have_ to be.

So the "userness" of a pointer in that context is a bit ambiguous, and so we've picked the pointers to be just plain "void *".

That said, arguably we should have gone the other way and just made them both "__user" pointers, and do the cast the other way around.

Quoted 6 lines

But there's no absolutely right answer here, and nobody should ever use copy_user_generic() directly (ie it is very much meant to be only used as a internal helper for the cases that get the pointer annotations right).

Cheers,
-Mukesh

Quoted 31 lines
---
  arch/x86/include/asm/uaccess_64.h | 2 +-
  arch/x86/lib/usercopy_64.c        | 2 +-
  2 files changed, 2 insertions(+), 2 deletions(-)
arch/x86/include/asm/uaccess_64.h +1 -1
index a9d637bc301d..cbca2cb28939 100644
@@ -208,7 +208,7 @@ __copy_from_user_flushcache(void *dst, const void __user *src, unsigned size)
  }
 
  unsigned long
-copy_user_handle_tail(char *to, char *from, unsigned len);
+copy_user_handle_tail(char __user *to, char __user *from, unsigned len);
 
  unsigned long
  mcsafe_handle_tail(char *to, char *from, unsigned len);
arch/x86/lib/usercopy_64.c +1 -1
index ee42bb0cbeb3..aa180424e77a 100644
@@ -60,7 +60,7 @@ EXPORT_SYMBOL(clear_user);
   * it is not necessary to optimize tail handling.
   */
  __visible unsigned long
-copy_user_handle_tail(char *to, char *from, unsigned len)
+copy_user_handle_tail(char __user *to, char __user *from, unsigned len)
  {
  	for (; len; --len, to++) {
  		char c;
#2 lore raw

[tip:x86/asm] x86/asm: Annotate copy_user_handle_tail() pointers with __user

In reply to Ben Dooks
x86/asm: Annotate copy_user_handle_tail() pointers with __user
Commit-ID:  b5dbb6799e3e5b8ebdce33b52b2d4ec9c66e15fe
Gitweb:     https://git.kernel.org/tip/b5dbb6799e3e5b8ebdce33b52b2d4ec9c66e15fe
Author:     Ben Dooks <[email protected]>
AuthorDate: Sat, 30 Mar 2019 11:56:24 +0000
Committer:  Borislav Petkov <[email protected]>
CommitDate: Tue, 2 Apr 2019 10:38:51 +0200

x86/asm: Annotate copy_user_handle_tail() pointers with __user

copy_user_handle_tail() clearly uses both @from and @to as pointers to user-space memory.

Currently, it triggers sparse warning on using the calls to get and put to user-space which can be fixed easily by changing the call to take __user annotated pointers:

arch/x86/lib/usercopy_64.c:68:21: warning: incorrect type in argument 1 (different address spaces)
arch/x86/lib/usercopy_64.c:68:21: expected void const volatile [noderef] <asn:1>*<noident>
arch/x86/lib/usercopy_64.c:68:21: got char *
arch/x86/lib/usercopy_64.c:70:21: warning: incorrect type in argument 1 (different address spaces)
arch/x86/lib/usercopy_64.c:70:21: expected void const volatile [noderef] <asn:1>*<noident>
arch/x86/lib/usercopy_64.c:70:21: got char *to

Linus further explains the reasoning why it was done this way:

Quoted 5 lines · Borislav Petkov

Well, but copy_user_generic() (which ends up calling the copy_user_handle_tail() eventually) casts those __user pointers to (__force void *). Converting them back to __user looks strange to me.

Linus?

Well, it does that because the x86 version of copy_user_generic() can work in either direction, so it works when either the source or destination (or both) are user pointers, but they don't _have_ to be.

So the "userness" of a pointer in that context is a bit ambiguous, and so we've picked the pointers to be just plain "void *".

That said, arguably we should have gone the other way and just made them both "__user" pointers, and do the cast the other way around.

But there's no absolutely right answer here, and nobody should ever use copy_user_generic() directly (ie it is very much meant to be only used as a internal helper for the cases that get the pointer annotations right).

[ bp: massage. ]

---
 arch/x86/include/asm/uaccess_64.h | 2 +-
 arch/x86/lib/usercopy_64.c        | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
arch/x86/include/asm/uaccess_64.h +1 -1
index a9d637bc301d..cbca2cb28939 100644
@@ -208,7 +208,7 @@ __copy_from_user_flushcache(void *dst, const void __user *src, unsigned size)
 }
 
 unsigned long
-copy_user_handle_tail(char *to, char *from, unsigned len);
+copy_user_handle_tail(char __user *to, char __user *from, unsigned len);
 
 unsigned long
 mcsafe_handle_tail(char *to, char *from, unsigned len);
arch/x86/lib/usercopy_64.c +1 -1
index ee42bb0cbeb3..aa180424e77a 100644
@@ -60,7 +60,7 @@ EXPORT_SYMBOL(clear_user);
  * it is not necessary to optimize tail handling.
  */
 __visible unsigned long
-copy_user_handle_tail(char *to, char *from, unsigned len)
+copy_user_handle_tail(char __user *to, char __user *from, unsigned len)
 {
 	for (; len; --len, to++) {
 		char c;