Re: [PATCH] ecryptfs: avoid to access NULL pointer when write metadata in xattr

[Tyler Hicks <tyhicks@canonical.com>]

[-- Attachment #1: Type: text/plain, Size: 4900 bytes --]

Hello and thanks for the patch!

On 2014-07-24 17:25:42, Chao Yu wrote:
> https://bugzilla.kernel.org/show_bug.cgi?id=41692

This actually isn't the bug that this patch fixes. It is a different bug
(that I don't think exists anymore) and someone happened to test for the
bug on a newer kernel and hit the oops that you're fixing here.

> 
> Christopher Head 2014-06-28 05:26:20 UTC described:
> "I tried to reproduce this on 3.12.21. Instead, when I do "echo hello > foo"
> in an ecryptfs mount with ecryptfs_xattr specified, I get a kernel crash:
> 
> BUG: unable to handle kernel NULL pointer dereference at           (null)
> IP: [<ffffffff8110eb39>] fsstack_copy_attr_all+0x2/0x61
> PGD d7840067 PUD b2c3c067 PMD 0 
> Oops: 0002 [#1] SMP 
> Modules linked in: nvidia(PO)
> CPU: 3 PID: 3566 Comm: bash Tainted: P           O 3.12.21-gentoo-r1 #2
> Hardware name: ASUSTek Computer Inc. G60JX/G60JX, BIOS 206 03/15/2010
> task: ffff8801948944c0 ti: ffff8800bad70000 task.ti: ffff8800bad70000
> RIP: 0010:[<ffffffff8110eb39>]  [<ffffffff8110eb39>] fsstack_copy_attr_all+0x2/0x61
> RSP: 0018:ffff8800bad71c10  EFLAGS: 00010246
> RAX: 00000000000181a4 RBX: ffff880198648480 RCX: 0000000000000000
> RDX: 0000000000000004 RSI: ffff880172010450 RDI: 0000000000000000
> RBP: ffff880198490e40 R08: 0000000000000000 R09: 0000000000000000
> R10: ffff880172010450 R11: ffffea0002c51e80 R12: 0000000000002000
> R13: 000000000000001a R14: 0000000000000000 R15: ffff880198490e40
> FS:  00007ff224caa700(0000) GS:ffff88019fcc0000(0000) knlGS:0000000000000000
> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 0000000000000000 CR3: 00000000bb07f000 CR4: 00000000000007e0
> Stack:
> ffffffff811826e8 ffff8800a39d8000 0000000000000000 000000000000001a
> ffff8800a01d0000 ffff8800a39d8000 ffffffff81185fd5 ffffffff81082c2c
> 00000001a39d8000 53d0abbc98490e40 0000000000000037 ffff8800a39d8220
> Call Trace:
> [<ffffffff811826e8>] ? ecryptfs_setxattr+0x40/0x52
> [<ffffffff81185fd5>] ? ecryptfs_write_metadata+0x1b3/0x223
> [<ffffffff81082c2c>] ? should_resched+0x5/0x23
> [<ffffffff8118322b>] ? ecryptfs_initialize_file+0xaf/0xd4
> [<ffffffff81183344>] ? ecryptfs_create+0xf4/0x142
> [<ffffffff810f8c0d>] ? vfs_create+0x48/0x71
> [<ffffffff810f9c86>] ? do_last.isra.68+0x559/0x952
> [<ffffffff810f7ce7>] ? link_path_walk+0xbd/0x458
> [<ffffffff810fa2a3>] ? path_openat+0x224/0x472
> [<ffffffff810fa7bd>] ? do_filp_open+0x2b/0x6f
> [<ffffffff81103606>] ? __alloc_fd+0xd6/0xe7
> [<ffffffff810ee6ab>] ? do_sys_open+0x65/0xe9
> [<ffffffff8157d022>] ? system_call_fastpath+0x16/0x1b
> RIP  [<ffffffff8110eb39>] fsstack_copy_attr_all+0x2/0x61
> RSP <ffff8800bad71c10>
> CR2: 0000000000000000
> ---[ end trace df9dba5f1ddb8565 ]---"
> 
> If we create a file when we mount with ecryptfs_xattr_metadata option, we will
> encounter a crash in this path:
> ->ecryptfs_create
>   ->ecryptfs_initialize_file
>     ->ecryptfs_write_metadata
>       ->ecryptfs_write_metadata_to_xattr
>         ->ecryptfs_setxattr
>           ->fsstack_copy_attr_all
> It's because our dentry->d_inode used in fsstack_copy_attr_all is NULL, and it
> will be initialized when ecryptfs_initialize_file finish.
> 
> So we should skip copying attr from lower inode when the value of ->d_inode is
> invalid.
> 
> Signed-off-by: Chao Yu <chao2.yu@samsung.com>
> ---

I wrote a patch to fix this bug last week, made the changes necessary to
run through all of the eCryptfs tests with the ecryptfs_xattr_metadata
mount option, tested the fix, and then forgot to send out the patch.
(doh!) But it worked out well because I think I like your patch better.

I moved the innards of ecryptfs_setxattr() into a new function that
accepts the lower_dentry and ecryptfs inode as separate arguments. That
way it can work regardless of the eCryptfs dentry not yet being
d_initialized().

After looking at your patch, I don't think that's necessary. Skipping
the fsstack_copy_attr_all() should be fine for brand new eCryptfs
inodes since the attr copy just happened in ecryptfs_inode_set().

I'll test your patch and then push it to the eCryptfs -next branch. I'll
also update the bug link to point to the specific comment rather than
the unrelated bug itself. Thanks!

Tyler

>  fs/ecryptfs/inode.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/ecryptfs/inode.c b/fs/ecryptfs/inode.c
> index e67f9f0..436b7d8 100644
> --- a/fs/ecryptfs/inode.c
> +++ b/fs/ecryptfs/inode.c
> @@ -1037,7 +1037,7 @@ ecryptfs_setxattr(struct dentry *dentry, const char *name, const void *value,
>  	}
>  
>  	rc = vfs_setxattr(lower_dentry, name, value, size, flags);
> -	if (!rc)
> +	if (!rc && dentry->d_inode)
>  		fsstack_copy_attr_all(dentry->d_inode, lower_dentry->d_inode);
>  out:
>  	return rc;
> -- 
> 2.0.1.474.g72c7794
> 
> 

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]