[PATCH] [UFS] fs/ufs/super.c misreads the file system state

[PATCH] [UFS] fs/ufs/super.c misreads the file system state

[Leonid Kalev]

[-- Attachment #1: Type: text/plain, Size: 670 bytes --]

ufs_get_fs_state() needs the file system type to read the state from the 
correct place in the superblock. It takes the type from 
UFS_SB(sb)->s_flags, but that value is stored after the first call to 
ufs_get_fs_state(). The patch below moves the assignment of s_flags up, 
before the first call to ufs_get_fs_state().

The patch is against linux-2.6.23-rc6-git7, but it applies (with offset) 
to 2.6.22 and to earlier versions as well. It has been tested on the 
Solaris flavor of UFS (ufstype=sunx86) - with this change, the file 
system can be used in read-write mode.

* Please cc me when replying to this - I am not subscribed to the 
linux-kernel list. *

Leonid

[-- Attachment #2: ufs_state_bug_l.patch --]
[-- Type: text/x-patch, Size: 453 bytes --]

--- a/fs/ufs/super.c
+++ b/fs/ufs/super.c
@@ -897,6 +897,8 @@
 
 	ufs_print_super_stuff(sb, usb1, usb2, usb3);
 
+	sbi->s_flags = flags;
+
 	/*
 	 * Check, if file system was correctly unmounted.
 	 * If not, make it read only.
@@ -1026,8 +1028,6 @@
 		uspi->s_maxsymlinklen =
 		    fs32_to_cpu(sb, usb3->fs_un2.fs_44.fs_maxsymlinklen);
 	
-	sbi->s_flags = flags;
-
 	inode = iget(sb, UFS_ROOTINO);
 	if (!inode || is_bad_inode(inode))
 		goto failed;

Re: [PATCH] [UFS] fs/ufs/super.c misreads the file system state

[Satyam Sharma]

On Wed, 19 Sep 2007, Leonid Kalev wrote:
> 
> ufs_get_fs_state() needs the file system type to read the state from the
> correct place in the superblock. It takes the type from UFS_SB(sb)->s_flags,
> but that value is stored after the first call to ufs_get_fs_state(). The patch
> below moves the assignment of s_flags up, before the first call to
> ufs_get_fs_state().
> 
> The patch is against linux-2.6.23-rc6-git7, but it applies (with offset) to
> 2.6.22 and to earlier versions as well. It has been tested on the Solaris
> flavor of UFS (ufstype=sunx86) - with this change, the file system can be used
> in read-write mode.

The same fix is already in 2.6.23-rc6-mm1 (don't know if it's lined up
for .23 though, I think it should be).


Satyam

PS: All my mails to Evgeniy Dushistov never reach him. The mail.ru server
    is absolutely *evil* and simply rejects all mails unless the sender
    is already whitelisted by the reciever (!)