Skip to contentSkip to conversation

Fix __copy_{to,from}_user_inatomic

PATCH In review Franck Bui-Huu 3 participants 5 messages Feb 16, 2007 – Feb 19, 2007 MIPS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Fix __copy_{to,from}_user_inatomic

From: Franck Bui-Huu <[email protected]>

These functions are aliases to __copy_{to,from}_user resp but they are not allowed to sleep. Therefore might_sleep() must not be used by their implementions.

---
 include/asm-mips/uaccess.h |    6 ++----
 1 files changed, 2 insertions(+), 4 deletions(-)
include/asm-mips/uaccess.h +2 -4
index 825fcbd..fd01939 100644
@@ -407,7 +407,7 @@ extern size_t __copy_user(void *__to, const void *__from, size_t __n);
  * @from: Source address, in kernel space.
  * @n:    Number of bytes to copy.
  *
- * Context: User context only.  This function may sleep.
+ * Context: User context only.
  *
  * Copy data from kernel space to user space.  Caller must check
  * the specified block with access_ok() before calling this function.
@@ -421,7 +421,6 @@ extern size_t __copy_user(void *__to, const void *__from, size_t __n);
 	const void *__cu_from;						\
 	long __cu_len;							\
 									\
-	might_sleep();							\
 	__cu_to = (to);							\
 	__cu_from = (from);						\
 	__cu_len = (n);							\
@@ -490,7 +489,7 @@ extern size_t __copy_user(void *__to, const void *__from, size_t __n);
  * @from: Source address, in user space.
  * @n:    Number of bytes to copy.
  *
- * Context: User context only.  This function may sleep.
+ * Context: User context only.
  *
  * Copy data from user space to kernel space.  Caller must check
  * the specified block with access_ok() before calling this function.
@@ -507,7 +506,6 @@ extern size_t __copy_user(void *__to, const void *__from, size_t __n);
 	const void __user *__cu_from;					\
 	long __cu_len;							\
 									\
-	might_sleep();							\
 	__cu_to = (to);							\
 	__cu_from = (from);						\
 	__cu_len = (n);							\
Signature · 1 line
1.4.4.3.ge6d4
#1 lore raw
Quoted 3 lines · Franck Bui-Huu

These functions are aliases to __copy_{to,from}_user resp but they are not allowed to sleep. Therefore might_sleep() must not be used by their implementions.

The _inatomic functions are know to buggy but this doesn't quite fix the whole issues with them. On error __copy_from_user_inatomic should not clear the non-copied part of the destination buffer. See 01408c4939479ec46c15aa7ef6e2406be50eeeca and 7c12d81134b130ccd4c286b434ca48c4cda71a2f for the rationale.

Ralf

#2 lore raw
Quoted 8 lines · Ralf Baechle
Quoted 3 lines · Franck Bui-Huu

These functions are aliases to __copy_{to,from}_user resp but they are not allowed to sleep. Therefore might_sleep() must not be used by their implementions.

The _inatomic functions are know to buggy but this doesn't quite fix the whole issues with them.

ok so the patch's should have been: "Fix console warnings triggered by __copy_{to,from}_user_inatomic()"

Quoted 4 lines

On error __copy_from_user_inatomic should not clear the non-copied part of the destination buffer. See 01408c4939479ec46c15aa7ef6e2406be50eeeca and 7c12d81134b130ccd4c286b434ca48c4cda71a2f for the rationale.

I'll try to take a look at it.

Thanks

Signature · 1 line
              Franck
#3 lore raw
In reply to Franck Bui-Huu
Quoted 3 lines · Franck Bui-Huu

These functions are aliases to __copy_{to,from}_user resp but they are not allowed to sleep. Therefore might_sleep() must not be used by their implementions.

This changes CONFIG_PREEMPT_VOLUNTARY behavior. In this case might_sleep() is not just an annotation.

Well, so currently CONFIG_PREEMPT_VOLUNTARY seems broken. Maybe we need separete functions anyway.
---
Atsushi Nemoto

#4 lore raw
Quoted 12 lines · Atsushi Nemoto

Date: Sat, 17 Feb 2007 01:12:20 +0900 (JST)
To: [email protected]
Cc: [email protected], [email protected]
Subject: Re: [PATCH] Fix __copy_{to,from}_user_inatomic
From: Atsushi Nemoto <[email protected]>
Content-Type: Text/Plain; charset=us-ascii

Quoted 3 lines · Franck Bui-Huu

These functions are aliases to __copy_{to,from}_user resp but they are not allowed to sleep. Therefore might_sleep() must not be used by their implementions.

Quoted 5 lines

This changes CONFIG_PREEMPT_VOLUNTARY behavior. In this case might_sleep() is not just an annotation.

Well, so currently CONFIG_PREEMPT_VOLUNTARY seems broken. Maybe we need separete functions anyway.

The issue needed to be tackled so while suboptimal I've added a chainsawed version of memcpy to provide __copy_from_user_inatomic. Not elegant but hopefully that will be cleaned soon. memcpy could use a general ovehaul anyway.

Ralf