Re: [PATCH] ecryptfs: validate key payload size before using it as an auth_tok

[Tyler Hicks]

On 2026-10-11 13:19:01, Nguyen Ngoc Thang wrote:
> ecryptfs_get_key_payload_data() casts the payload of a "user" or
> "encrypted" key straight to struct ecryptfs_auth_tok. The payload size is
> chosen by whoever added the key, so a payload shorter than the struct
> makes later accesses run past the allocation. For example, a 712-byte
> "user" key makes write_tag_3_packet() read the password signature beyond
> the end of the kmalloc-1k object:
> 
>   BUG: KASAN: slab-out-of-bounds in ecryptfs_from_hex+0x18c/0x1d0
>   Read of size 1
>    write_tag_3_packet+0x145/0x1430
>    ecryptfs_generate_key_packet_set+0x7ae/0x1880
>    ecryptfs_write_metadata+0x590/0x930
>    ecryptfs_create+0x393/0x550
> 
> Reject payloads smaller than struct ecryptfs_auth_tok with -EINVAL. The
> callers already propagate errors from this helper, so mounting with such
> a key now fails cleanly.
> 
> Reported-by: syzbot+dc50906b100ac7a1c34d@xxxxxxxxxxxxxxxxxxxxxxxxx
> Closes: https://syzkaller.appspot.com/bug?extid=dc50906b100ac7a1c34d
> Fixes: 237fead61998 ("[PATCH] ecryptfs: fs/Makefile and fs/Kconfig")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Nguyen Ngoc Thang <ngocthang2710.1999@xxxxxxxxx>

Hello and thanks for the patch! I've already taken the user key payload
fix from another contributor and will take the session key encryption
key check from a separate contributor. You can find the links here:

 * https://lore.kernel.org/ecryptfs/20260917044558.41677-1-neganhat@xxxxxxxxx/
 * https://lore.kernel.org/ecryptfs/20261004123755.14642-1-adrianox@xxxxxxxxx/

> ---
> Notes (not for the commit log):
> syzbot reports "KASAN: slab-out-of-bounds Read in ecryptfs_from_hex". The
> reproducer adds a 712-byte "user" key whose description matches
> ecryptfs_sig=; the mount succeeds and the first create makes
> write_tag_3_packet() read the signature past the end of the kmalloc-1k
> object. Tested in QEMU with syzbot's KASAN config: base kernel hits the
> KASAN report, patched kernel fails the mount with -EINVAL. Fixes points at
> the initial ecryptfs commit since the history here is grafted; happy to
> retarget.
> 
>  fs/ecryptfs/ecryptfs_kernel.h | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/fs/ecryptfs/ecryptfs_kernel.h b/fs/ecryptfs/ecryptfs_kernel.h
> index 58165928ed1e..863761b3847a 100644
> --- a/fs/ecryptfs/ecryptfs_kernel.h
> +++ b/fs/ecryptfs/ecryptfs_kernel.h
> @@ -88,6 +88,9 @@ ecryptfs_get_encrypted_key_payload_data(struct key *key)
>  	if (!payload)
>  		return ERR_PTR(-EKEYREVOKED);
>  
> +	if (payload->decrypted_datalen < sizeof(struct ecryptfs_auth_tok))
> +		return ERR_PTR(-EINVAL);

I think Adriano's check on payload_datalen is more accurate. I believe
your check would block legitimate encrypted-keys.

Tyler

> +
>  	return (struct ecryptfs_auth_tok *)payload->payload_data;
>  }
>  
> @@ -124,6 +127,9 @@ ecryptfs_get_key_payload_data(struct key *key)
>  	if (!ukp)
>  		return ERR_PTR(-EKEYREVOKED);
>  
> +	if (ukp->datalen < sizeof(struct ecryptfs_auth_tok))
> +		return ERR_PTR(-EINVAL);
> +
>  	return (struct ecryptfs_auth_tok *)ukp->data;
>  }
>  
> -- 
> 2.43.0
>