Re: [2.6.8-rc4-bk] NFS oops on x86-64

[Trond Myklebust <trond.myklebust@fys.uio.no>]

På fr , 13/08/2004 klokka 21:45, skreiv Trond Myklebust:

> I can see 2 problems in the NFS code alone:
> 
>   1) Replacing a test for whether or not O_APPEND and O_DIRECT are
> *both* set with one that checks whether either is set.
>   2) Adding a wonderful check in nfs_open() that causes it to return
> immediately if this new nfs_check_flags() returns 0 (i.e. OK).

The following patch should fix up those two issues, but frankly, I'm not
convinced that it is much of an improvement to be replacing the old
fcntl() interface with something that doesn't even pass down the struct
file for which the user is trying to change the open mode.

Trond

--- linux-2.6.8-rc4/fs/nfs/file.c.orig	2004-08-13 14:21:25.000000000 -0400
+++ linux-2.6.8-rc4/fs/nfs/file.c	2004-08-13 21:50:28.000000000 -0400
@@ -72,7 +72,7 @@ struct inode_operations nfs_file_inode_o
 
 static int nfs_check_flags(int flags)
 {
-	if (flags & (O_APPEND | O_DIRECT))
+	if (flags & (O_APPEND | O_DIRECT) == (O_APPEND | O_DIRECT))
 		return -EINVAL;
 
 	return 0;
@@ -89,7 +89,7 @@ nfs_file_open(struct inode *inode, struc
 	int res;
 
 	res = nfs_check_flags(filp->f_flags);
-	if (!res)
+	if (res != 0)
 		return res;
 
 	lock_kernel();