[X2Go-Dev] [X2Go-Commits] [nx-libs] 51/52: glx: Length checking for RenderLarge requests (v2) [CVE-2014-8098 3/8] (v3)

Michael DePaulo mikedep333 at gmail.com
Mon Feb 16 14:34:28 CET 2015


On Sun, Feb 15, 2015 at 8:23 PM, Mihai Moldovan <ionic at ionic.de> wrote:
> On 14.02.2015 05:47 PM, git-admin at x2go.org wrote:
>> This is an automated email from the git hooks/post-receive script.
>>
>> x2go pushed a commit to branch 3.6.x
>> in repository nx-libs.
>>
>> commit 9c558f9ca2c0d4e34fa71dff272ed1c39c22cd9d
>> Author: Adam Jackson <ajax at redhat.com>
>> Date:   Mon Nov 10 12:13:43 2014 -0500
>>
>>     glx: Length checking for RenderLarge requests (v2) [CVE-2014-8098 3/8] (v3)
>>
>>     This is a half-measure until we start passing request length into the
>>     varsize function, but it's better than the nothing we had before.
>>
>>     v2: Verify that there's at least a large render header's worth of
>>     dataBytes (Julien Cristau)
>>
>>     v3: backport to RHEL5
>>
>>     v4: backport to nx-libs 3.6.x (Mike DePaulo)
>>
>>     Reviewed-by: Michal Srb <msrb at suse.com>
>>     Reviewed-by: Andy Ritger <aritger at nvidia.com>
>>     Signed-off-by: Adam Jackson <ajax at redhat.com>
>>     Signed-off-by: Alan Coopersmith <alan.coopersmith at oracle.com>
>>     Signed-off-by: Fedora X Ninjas <x at fedoraproject.org>
>>     Signed-off-by: Dave Airlie <airlied at redhat.com>
>>
>>     fixup swap
>> ---
>>  nx-X11/programs/Xserver/GL/glx/glxcmds.c     |   58 +++++++++++++++----------
>>  nx-X11/programs/Xserver/GL/glx/glxcmdsswap.c |   59 ++++++++++++++++----------
>>  2 files changed, 71 insertions(+), 46 deletions(-)
>>
>> diff --git a/nx-X11/programs/Xserver/GL/glx/glxcmds.c b/nx-X11/programs/Xserver/GL/glx/glxcmds.c
>> index 831c65b..20c12f3 100644
>> --- a/nx-X11/programs/Xserver/GL/glx/glxcmds.c
>> +++ b/nx-X11/programs/Xserver/GL/glx/glxcmds.c
>> @@ -1535,6 +1535,8 @@ int __glXRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>      ** duplicated there.
>>      */
>>
>> +    REQUEST_AT_LEAST_SIZE(xGLXRenderLargeReq);
>> +
>>      req = (xGLXRenderLargeReq *) pc;
>>      glxc = __glXForceCurrent(cl, req->contextTag, &error);
>>      if (!glxc) {
>> @@ -1542,12 +1544,15 @@ int __glXRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>       __glXResetLargeCommandStatus(cl);
>>       return error;
>>      }
>> +    if (safe_pad(req->dataBytes) < 0)
>> +        return BadLength;
>> +
>>      dataBytes = req->dataBytes;
>>
>>      /*
>>      ** Check the request length.
>>      */
>> -    if ((req->length << 2) != __GLX_PAD(dataBytes) + sz_xGLXRenderLargeReq) {
>> +    if ((req->length << 2) != safe_pad(dataBytes) + sz_xGLXRenderLargeReq) {
>>       client->errorValue = req->length;
>>       /* Reset in case this isn't 1st request. */
>>       __glXResetLargeCommandStatus(cl);
>> @@ -1557,7 +1562,7 @@ int __glXRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>
>>      if (cl->largeCmdRequestsSoFar == 0) {
>>       __GLXrenderSizeData *entry;
>> -     int extra, cmdlen;
>> +     int extra = 0, cmdlen;
>>       /*
>>       ** This is the first request of a multi request command.
>>       ** Make enough space in the buffer, then copy the entire request.
>> @@ -1567,9 +1572,13 @@ int __glXRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>           return __glXBadLargeRequest;
>>       }
>>
>> +        if (dataBytes < __GLX_RENDER_LARGE_HDR_SIZE)
>> +            return BadLength;
>> +
>>       hdr = (__GLXrenderLargeHeader *) pc;
>> -     cmdlen = hdr->length;
>>       opcode = hdr->opcode;
>> +        if ((cmdlen = safe_pad(hdr->length)) < 0)
>> +         return BadLength;
>>
>>       /*
>>       ** Check for core opcodes and grab entry data.
>> @@ -1603,16 +1612,13 @@ int __glXRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>           if (extra < 0) {
>>               return BadLength;
>>           }
>> -         /* large command's header is 4 bytes longer, so add 4 */
>> -         if (cmdlen != __GLX_PAD(entry->bytes + 4 + extra)) {
>> -             return BadLength;
>> -         }
>> -     } else {
>> -         /* constant size command */
>> -         if (cmdlen != __GLX_PAD(entry->bytes + 4)) {
>> -             return BadLength;
>> -         }
>>       }
>> +
>> +        /* the +4 is safe because we know entry.bytes is small */
>> +        if (cmdlen != safe_pad(safe_add(entry->bytes + 4, extra))) {
>> +            return BadLength;
>> +        }
>> +
>>       /*
>>       ** Make enough space in the buffer, then copy the entire request.
>>       */
>> @@ -1641,6 +1647,7 @@ int __glXRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>       ** We are receiving subsequent (i.e. not the first) requests of a
>>       ** multi request command.
>>       */
>> +        int bytesSoFar; /* including this packet */
>>
>>       /*
>>       ** Check the request number and the total request count.
>> @@ -1659,7 +1666,13 @@ int __glXRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>       /*
>>       ** Check that we didn't get too much data.
>>       */
>> -     if ((cl->largeCmdBytesSoFar + dataBytes) > cl->largeCmdBytesTotal) {
>> +        if ((bytesSoFar = safe_add(cl->largeCmdBytesSoFar, dataBytes)) < 0) {
>> +            client->errorValue = dataBytes;
>> +            __glXResetLargeCommandStatus(cl);
>> +            return __glXBadLargeRequest;
>> +        }
>> +
>> +        if (bytesSoFar > cl->largeCmdBytesTotal) {
>>           client->errorValue = dataBytes;
>>           __glXResetLargeCommandStatus(cl);
>>           return __glXBadLargeRequest;
>> @@ -1673,17 +1686,16 @@ int __glXRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>           ** This is the last request; it must have enough bytes to complete
>>           ** the command.
>>           */
>> -         /* NOTE: the two pad macros have been added below; they are needed
>> -         ** because the client library pads the total byte count, but not
>> -         ** the per-request byte counts.  The Protocol Encoding says the
>> -         ** total byte count should not be padded, so a proposal will be
>> -         ** made to the ARB to relax the padding constraint on the total
>> -         ** byte count, thus preserving backward compatibility.  Meanwhile,
>> -         ** the padding done below fixes a bug that did not allow
>> -         ** large commands of odd sizes to be accepted by the server.
>> +            /* NOTE: the pad macro below is needed because the client library
>> +             ** pads the total byte count, but not the per-request byte counts.
>> +             ** The Protocol Encoding says the total byte count should not be
>> +             ** padded, so a proposal will be made to the ARB to relax the
>> +             ** padding constraint on the total byte count, thus preserving
>> +             ** backward compatibility.  Meanwhile, the padding done below
>> +             ** fixes a bug that did not allow large commands of odd sizes to
>> +             ** be accepted by the server.
>>           */
>> -         if (__GLX_PAD(cl->largeCmdBytesSoFar) !=
>> -             __GLX_PAD(cl->largeCmdBytesTotal)) {
>> +            if (safe_pad(cl->largeCmdBytesSoFar) != cl->largeCmdBytesTotal) {
>
> cl->largeCmdBytesTotal is not wrapped in safe_pad(). This looks odd and could be a mistake.
>
>
>>               client->errorValue = dataBytes;
>>               __glXResetLargeCommandStatus(cl);
>>               return __glXBadLargeRequest;
>> diff --git a/nx-X11/programs/Xserver/GL/glx/glxcmdsswap.c b/nx-X11/programs/Xserver/GL/glx/glxcmdsswap.c
>> index 2685355..2e228c0 100644
>> --- a/nx-X11/programs/Xserver/GL/glx/glxcmdsswap.c
>> +++ b/nx-X11/programs/Xserver/GL/glx/glxcmdsswap.c
>> @@ -587,6 +587,8 @@ int __glXSwapRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>      ** duplicated there.
>>      */
>>
>> +    REQUEST_AT_LEAST_SIZE(xGLXRenderLargeReq);
>> +
>>      req = (xGLXRenderLargeReq *) pc;
>>      __GLX_SWAP_SHORT(&req->length);
>>      __GLX_SWAP_INT(&req->contextTag);
>> @@ -599,12 +601,15 @@ int __glXSwapRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>       __glXResetLargeCommandStatus(cl);
>>       return error;
>>      }
>> +    if (safe_pad(req->dataBytes) < 0)
>> +        return BadLength;
>> +
>>      dataBytes = req->dataBytes;
>>
>>      /*
>>      ** Check the request length.
>>      */
>> -    if ((req->length << 2) != __GLX_PAD(dataBytes) + sz_xGLXRenderLargeReq) {
>> +    if ((req->length << 2) != safe_pad(dataBytes) + sz_xGLXRenderLargeReq) {
>>       client->errorValue = req->length;
>>       /* Reset in case this isn't 1st request. */
>>       __glXResetLargeCommandStatus(cl);
>> @@ -614,7 +619,7 @@ int __glXSwapRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>
>>      if (cl->largeCmdRequestsSoFar == 0) {
>>       __GLXrenderSizeData *entry;
>> -     int extra;
>> +     int extra = 0;
>>       size_t cmdlen;
>>       /*
>>       ** This is the first request of a multi request command.
>> @@ -624,12 +629,17 @@ int __glXSwapRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>           client->errorValue = req->requestNumber;
>>           return __glXBadLargeRequest;
>>       }
>> +     if (dataBytes < __GLX_RENDER_LARGE_HDR_SIZE)
>> +         return BadLength;
>> +
>>       hdr = (__GLXrenderLargeHeader *) pc;
>>       __GLX_SWAP_INT(&hdr->length);
>>       __GLX_SWAP_INT(&hdr->opcode);
>> -     cmdlen = hdr->length;
>>       opcode = hdr->opcode;
>>
>> +     if ((cmdlen = safe_pad(hdr->length)) < 0)
>> +         return BadLength;
>> +
>>       if ( (opcode >= __GLX_MIN_RENDER_OPCODE) &&
>>            (opcode <= __GLX_MAX_RENDER_OPCODE) ) {
>>           entry = &__glXRenderSizeTable[opcode];
>> @@ -661,16 +671,12 @@ int __glXSwapRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>           if (extra < 0) {
>>               return BadLength;
>>           }
>> -         /* large command's header is 4 bytes longer, so add 4 */
>> -         if (cmdlen != __GLX_PAD(entry->bytes + 4 + extra)) {
>> -             return BadLength;
>> -         }
>> -     } else {
>> -         /* constant size command */
>> -         if (cmdlen != __GLX_PAD(entry->bytes + 4)) {
>> -             return BadLength;
>> -         }
>>       }
>> +        /* the +4 is safe because we know entry->bytes is small */
>> +        if (cmdlen != safe_pad(safe_add(entry->bytes + 4, extra))) {
>> +            return BadLength;
>> +        }
>> +
>>       /*
>>       ** Make enough space in the buffer, then copy the entire request.
>>       */
>> @@ -698,6 +704,7 @@ int __glXSwapRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>       ** We are receiving subsequent (i.e. not the first) requests of a
>>       ** multi request command.
>>       */
>> +        int bytesSoFar; /* including this packet */
>>
>>       /*
>>       ** Check the request number and the total request count.
>> @@ -716,7 +723,13 @@ int __glXSwapRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>       /*
>>       ** Check that we didn't get too much data.
>>       */
>> -     if ((cl->largeCmdBytesSoFar + dataBytes) > cl->largeCmdBytesTotal) {
>> +        if ((bytesSoFar = safe_add(cl->largeCmdBytesSoFar, dataBytes)) < 0) {
>> +            client->errorValue = dataBytes;
>> +            __glXResetLargeCommandStatus(cl);
>> +            return __glXBadLargeRequest;
>> +        }
>> +
>> +        if (bytesSoFar > cl->largeCmdBytesTotal) {
>>           client->errorValue = dataBytes;
>>           __glXResetLargeCommandStatus(cl);
>>           return __glXBadLargeRequest;
>> @@ -730,17 +743,17 @@ int __glXSwapRenderLarge(__GLXclientState *cl, GLbyte *pc)
>>           ** This is the last request; it must have enough bytes to complete
>>           ** the command.
>>           */
>> -         /* NOTE: the two pad macros have been added below; they are needed
>> -         ** because the client library pads the total byte count, but not
>> -         ** the per-request byte counts.  The Protocol Encoding says the
>> -         ** total byte count should not be padded, so a proposal will be
>> -         ** made to the ARB to relax the padding constraint on the total
>> -         ** byte count, thus preserving backward compatibility.  Meanwhile,
>> -         ** the padding done below fixes a bug that did not allow
>> -         ** large commands of odd sizes to be accepted by the server.
>> +            /* NOTE: the pad macro below is needed because the client library
>> +             ** pads the total byte count, but not the per-request byte counts.
>> +             ** The Protocol Encoding says the total byte count should not be
>> +             ** padded, so a proposal will be made to the ARB to relax the
>> +             ** padding constraint on the total byte count, thus preserving
>> +             ** backward compatibility.  Meanwhile, the padding done below
>> +             ** fixes a bug that did not allow large commands of odd sizes to
>> +             ** be accepted by the server.
>>           */
>> -         if (__GLX_PAD(cl->largeCmdBytesSoFar) !=
>> -             __GLX_PAD(cl->largeCmdBytesTotal)) {
>> +
>> +         if (safe_pad(cl->largeCmdBytesSoFar) != cl->largeCmdBytesTotal) {
>
> Similarly not wrapped in safe_pad() here. Not sure whether this happened
> on purpose or is an error.

I will look into this later today, or tomorrow. In the meantime, feel
free to use your judgment. (And feel free to contact upstream.)


More information about the x2go-dev mailing list