linux-kernel - Re: [PATCH] smb: client: Fix -Wstringop-overflow issues

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Wed, 12 Jul 2023 17:01:23 -0700
From:   Kees Cook <keescook@...omium.org>
To:     "Gustavo A. R. Silva" <gustavoars@...nel.org>
Cc:     Steve French <sfrench@...ba.org>,
        Paulo Alcantara <pc@...guebit.com>,
        Ronnie Sahlberg <lsahlber@...hat.com>,
        Shyam Prasad N <sprasad@...rosoft.com>,
        Tom Talpey <tom@...pey.com>,
        Christian Brauner <brauner@...nel.org>,
        linux-cifs@...r.kernel.org, samba-technical@...ts.samba.org,
        linux-kernel@...r.kernel.org, linux-hardening@...r.kernel.org
Subject: Re: [PATCH] smb: client: Fix -Wstringop-overflow issues

On Tue, Jul 11, 2023 at 05:12:31PM -0600, Gustavo A. R. Silva wrote:
> pSMB->hdr.Protocol is an array of size 4 bytes, hence when the compiler
> analyzes this line of code
> 
> 	parm_data = ((char *) &pSMB->hdr.Protocol) + offset;
> 
> it legitimately complains about the fact that offset points outside the
> bounds of the array. Notice that the compiler gives priority to the object
> as an array, rather than merely the address of one more byte in a structure
> to wich offset should be added (which seems to be the actual intention of
> the original implementation).
> 
> Fix this by explicitly instructing the compiler to treat the code as a
> sequence of bytes in struct smb_com_transaction2_spi_req, and not as an
> array accessed through pointer notation.
> 
> Notice that ((char *)pSMB) + sizeof(pSMB->hdr.smb_buf_length) points to
> the same address as ((char *) &pSMB->hdr.Protocol), therefore this results
> in no differences in binary output.
> 
> Fixes the following -Wstringop-overflow warnings when built s390
> architecture with defconfig (GCC 13):
>   CC [M]  fs/smb/client/cifssmb.o
> In function 'cifs_init_ace',
>     inlined from 'posix_acl_to_cifs' at fs/smb/client/cifssmb.c:3046:3,
>     inlined from 'cifs_do_set_acl' at fs/smb/client/cifssmb.c:3191:15:
> fs/smb/client/cifssmb.c:2987:31: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
>  2987 |         cifs_ace->cifs_e_perm = local_ace->e_perm;
>       |         ~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~
> In file included from fs/smb/client/cifssmb.c:27:
> fs/smb/client/cifspdu.h: In function 'cifs_do_set_acl':
> fs/smb/client/cifspdu.h:384:14: note: at offset [7, 11] into destination object 'Protocol' of size 4
>   384 |         __u8 Protocol[4];
>       |              ^~~~~~~~
> In function 'cifs_init_ace',
>     inlined from 'posix_acl_to_cifs' at fs/smb/client/cifssmb.c:3046:3,
>     inlined from 'cifs_do_set_acl' at fs/smb/client/cifssmb.c:3191:15:
> fs/smb/client/cifssmb.c:2988:30: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
>  2988 |         cifs_ace->cifs_e_tag =  local_ace->e_tag;
>       |         ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~
> fs/smb/client/cifspdu.h: In function 'cifs_do_set_acl':
> fs/smb/client/cifspdu.h:384:14: note: at offset [6, 10] into destination object 'Protocol' of size 4
>   384 |         __u8 Protocol[4];
>       |              ^~~~~~~~
> 
> This helps with the ongoing efforts to globally enable
> -Wstringop-overflow.
> 
> Link: https://github.com/KSPP/linux/issues/310
> Fixes: dc1af4c4b472 ("cifs: implement set acl method")
> Cc: stable@...r.kernel.org
> Signed-off-by: Gustavo A. R. Silva <gustavoars@...nel.org>
> ---
>  fs/smb/client/cifssmb.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/smb/client/cifssmb.c b/fs/smb/client/cifssmb.c
> index 19f7385abeec..9dee267f1893 100644
> --- a/fs/smb/client/cifssmb.c
> +++ b/fs/smb/client/cifssmb.c
> @@ -3184,7 +3184,7 @@ int cifs_do_set_acl(const unsigned int xid, struct cifs_tcon *tcon,
>  	param_offset = offsetof(struct smb_com_transaction2_spi_req,
>  				InformationLevel) - 4;
>  	offset = param_offset + params;
> -	parm_data = ((char *) &pSMB->hdr.Protocol) + offset;
> +	parm_data = ((char *)pSMB) + sizeof(pSMB->hdr.smb_buf_length) + offset;
>  	pSMB->ParameterOffset = cpu_to_le16(param_offset);
>  
>  	/* convert to on the wire format for POSIX ACL */

This looks correct, though looking at this code I think some serious
comments are needed to describe _why_ these offsets are calculated the
way the are. The only dynamic part of parm_data is name_len, and could
just as easily be calculated as:

	parm_data = pSMB->FileName + name_len;

which is MUCH more readable. But, yes, the above patch does result in
the same binary code.

Reviewed-by: Kees Cook <keescook@...omium.org>

-- 
Kees Cook