You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Justin Erenkrantz <ju...@erenkrantz.com> on 2003/10/24 23:57:49 UTC
[PATCH] Fix integer overflow in svn_auth.h
My compiler (Sun Forte) gave me:
"../../subversion/subversion/libsvn_ra_dav/session.c", line 138: warning:
integer overflow detected: op "<<" (E_INTEGER_OVERFLOW_DETECTED)
session.c line 138 is:
svn_failures |= SVN_AUTH_SSL_OTHER;
SVN_AUTH_SSL_OTHER is defined as (1<<31), which is -2147483648. I think
the or will lose/obscure all of the other status values - which based on
my reading, isn't quite right (and is an overflow condition). So, I think
it should be (1<<30).
But, svn_auth.h has this above the #define's:
* Note that these values are stored in the on disk auth cache by the
* SSL server certificate auth provider, so the meaning of these bits
* must not be changed without careful consideration.
So, what to do? Yet, I propose to change it as follows. -- justin
Index: subversion/include/svn_auth.h
===================================================================
--- subversion/include/svn_auth.h (revision 7517)
+++ subversion/include/svn_auth.h (working copy)
@@ -277,7 +277,7 @@
#define SVN_AUTH_SSL_EXPIRED (1<<1)
#define SVN_AUTH_SSL_CNMISMATCH (1<<2)
#define SVN_AUTH_SSL_UNKNOWNCA (1<<3)
-#define SVN_AUTH_SSL_OTHER (1<<31)
+#define SVN_AUTH_SSL_OTHER (1<<30)
typedef svn_error_t *(*svn_auth_ssl_server_trust_prompt_func_t) (
svn_auth_cred_ssl_server_trust_t **cred,
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: [PATCH] Fix integer overflow in svn_auth.h
Posted by Branko Čibej <br...@xbc.nu>.
Tobias Ringström wrote:
> Justin Erenkrantz wrote:
>
>> So, what to do? Yet, I propose to change it as follows. -- justin
>
>
> Whoups, my mistake. I added that code yesterday. I really intended
> (1<<30) to avoid the sign bit, but I obviously cannot count. I doubt
> that anyone has had time to depend on that bit position yet, so I've
> fixed it in revision 7518. Thanks for spotting this! I think I have
> seen gcc warn about this, but it did not this time... :-(
>
> Actually it would have been enough to say (1U << 31), but I think it's
> still wise to stay away from the sign bit.
No, Greg Hudson is right. Use apr_uint32_t for the bitmask. And please
define the constants as constants, not expressions; e.g.,
#define SVN_AUTH_SSL_OTHER 0x80000000
Then the compiler won't complain.
--
Brane Čibej <br...@xbc.nu> http://www.xbc.nu/brane/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: [PATCH] Fix integer overflow in svn_auth.h
Posted by Tobias Ringström <to...@ringstrom.mine.nu>.
Justin Erenkrantz wrote:
> So, what to do? Yet, I propose to change it as follows. -- justin
Whoups, my mistake. I added that code yesterday. I really intended
(1<<30) to avoid the sign bit, but I obviously cannot count. I doubt
that anyone has had time to depend on that bit position yet, so I've
fixed it in revision 7518. Thanks for spotting this! I think I have
seen gcc warn about this, but it did not this time... :-(
Actually it would have been enough to say (1U << 31), but I think it's
still wise to stay away from the sign bit.
/Tobias
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: [PATCH] Fix integer overflow in svn_auth.h
Posted by Greg Hudson <gh...@MIT.EDU>.
On Fri, 2003-10-24 at 19:57, Justin Erenkrantz wrote:
> "../../subversion/subversion/libsvn_ra_dav/session.c", line 138: warning:
> integer overflow detected: op "<<" (E_INTEGER_OVERFLOW_DETECTED)
By my reading, if we want to use 1 << 31, we need to cast the 1 to
unsigned apr_uint32_t, and hold the status in an apr_uint32_t instead of
an int. That also has implications for ssl_server_trust_providers.c and
possibly other code. It's a lot more work to fix this up, but we really
shouldn't be using a signed type for bit containers even if we aren't
using 1<<31; it's cleaner to use an unsigned type.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org