net: Transmit zero UDP checksum as 0xFFFF

Message ID 1510701938-53743-1-git-send-email-eswierk@skyportsystems.com
State New
Headers show
Series
  • net: Transmit zero UDP checksum as 0xFFFF
Related show

Commit Message

geoff--- via Qemu-devel Nov. 14, 2017, 11:25 p.m.
The checksum algorithm used by IPv4, TCP and UDP allows a zero value
to be represented by either 0x0000 and 0xFFFF. But per RFC 768, a zero
UDP checksum must be transmitted as 0xFFFF, as 0x0000 is a special
value meaning no checksum.

Substitute 0xFFFF whenever a checksum is computed as zero on a UDP
datagram. Doing this on IPv4 packets and TCP segments is unnecessary
but legal.

(While it is tempting to make the substitution in
net_checksum_finish(), that function is also used by receivers to
verify checksums, and in that case the expected value is always
0x0000.)

Signed-off-by: Ed Swierk <eswierk@skyportsystems.com>
---
 hw/net/e1000.c      | 5 +++--
 hw/net/net_rx_pkt.c | 3 +++
 hw/net/net_tx_pkt.c | 6 ++++++
 hw/net/vmxnet3.c    | 7 +++++--
 4 files changed, 17 insertions(+), 4 deletions(-)

Comments

Jason Wang Nov. 15, 2017, 2:10 a.m. | #1
On 2017年11月15日 07:25, Ed Swierk wrote:
> The checksum algorithm used by IPv4, TCP and UDP allows a zero value
> to be represented by either 0x0000 and 0xFFFF. But per RFC 768, a zero
> UDP checksum must be transmitted as 0xFFFF, as 0x0000 is a special
> value meaning no checksum.
>
> Substitute 0xFFFF whenever a checksum is computed as zero on a UDP
> datagram. Doing this on IPv4 packets and TCP segments is unnecessary
> but legal.
>
> (While it is tempting to make the substitution in
> net_checksum_finish(), that function is also used by receivers to
> verify checksums, and in that case the expected value is always
> 0x0000.)

Then looks like you'd better have an wrapper for net_checksum_finish() 
and do things there.

>
> Signed-off-by: Ed Swierk <eswierk@skyportsystems.com>
> ---
>   hw/net/e1000.c      | 5 +++--
>   hw/net/net_rx_pkt.c | 3 +++
>   hw/net/net_tx_pkt.c | 6 ++++++
>   hw/net/vmxnet3.c    | 7 +++++--
>   4 files changed, 17 insertions(+), 4 deletions(-)
>
> diff --git a/hw/net/e1000.c b/hw/net/e1000.c
> index d642314..97242a1 100644
> --- a/hw/net/e1000.c
> +++ b/hw/net/e1000.c
> @@ -505,8 +505,9 @@ putsum(uint8_t *data, uint32_t n, uint32_t sloc, uint32_t css, uint32_t cse)
>       if (cse && cse < n)
>           n = cse + 1;
>       if (sloc < n-1) {
> -        sum = net_checksum_add(n-css, data+css);
> -        stw_be_p(data + sloc, net_checksum_finish(sum));
> +        sum = net_raw_checksum(data + css, n - css);
> +        /* For UDP, zero checksum must be sent as 0xFFFF */
> +        stw_be_p(data + sloc, sum ? sum : 0xffff);
>       }
>   }
>   
> diff --git a/hw/net/net_rx_pkt.c b/hw/net/net_rx_pkt.c
> index 1019b50..e820132 100644
> --- a/hw/net/net_rx_pkt.c
> +++ b/hw/net/net_rx_pkt.c
> @@ -588,6 +588,9 @@ bool net_rx_pkt_fix_l4_csum(struct NetRxPkt *pkt)
>   
>       /* Calculate L4 checksum */
>       csum = cpu_to_be16(_net_rx_pkt_calc_l4_csum(pkt));
> +    if (!csum) {
> +        csum = 0xFFFF; /* For UDP, zero checksum must be sent as 0xFFFF */
> +    }

I thought we should only do this for tx?

Thanks
geoff--- via Qemu-devel Nov. 15, 2017, 2:46 a.m. | #2
On Tue, Nov 14, 2017 at 6:10 PM, Jason Wang <jasowang@redhat.com> wrote:
>
>
> On 2017年11月15日 07:25, Ed Swierk wrote:
>>
>> The checksum algorithm used by IPv4, TCP and UDP allows a zero value
>> to be represented by either 0x0000 and 0xFFFF. But per RFC 768, a zero
>> UDP checksum must be transmitted as 0xFFFF, as 0x0000 is a special
>> value meaning no checksum.
>>
>> Substitute 0xFFFF whenever a checksum is computed as zero on a UDP
>> datagram. Doing this on IPv4 packets and TCP segments is unnecessary
>> but legal.
>>
>> (While it is tempting to make the substitution in
>> net_checksum_finish(), that function is also used by receivers to
>> verify checksums, and in that case the expected value is always
>> 0x0000.)
>
>
> Then looks like you'd better have an wrapper for net_checksum_finish() and
> do things there.

I'll do that in v2.

>> index 1019b50..e820132 100644
>> --- a/hw/net/net_rx_pkt.c
>> +++ b/hw/net/net_rx_pkt.c
>> @@ -588,6 +588,9 @@ bool net_rx_pkt_fix_l4_csum(struct NetRxPkt *pkt)
>>         /* Calculate L4 checksum */
>>       csum = cpu_to_be16(_net_rx_pkt_calc_l4_csum(pkt));
>> +    if (!csum) {
>> +        csum = 0xFFFF; /* For UDP, zero checksum must be sent as 0xFFFF
>> */
>> +    }
>
>
> I thought we should only do this for tx?

We need to do this any time we modify the checksum field in a UDP
datagram header for someone else to verify. Normally this happens on
the tx path, and that someone is a remote system. But here
net_rx_pkt_fix_l4_csum() is used to fill in the checksum on packets
received with the NEEDS_CSUM vhdr flag before passing them along to
the guest.

--Ed

Patch

diff --git a/hw/net/e1000.c b/hw/net/e1000.c
index d642314..97242a1 100644
--- a/hw/net/e1000.c
+++ b/hw/net/e1000.c
@@ -505,8 +505,9 @@  putsum(uint8_t *data, uint32_t n, uint32_t sloc, uint32_t css, uint32_t cse)
     if (cse && cse < n)
         n = cse + 1;
     if (sloc < n-1) {
-        sum = net_checksum_add(n-css, data+css);
-        stw_be_p(data + sloc, net_checksum_finish(sum));
+        sum = net_raw_checksum(data + css, n - css);
+        /* For UDP, zero checksum must be sent as 0xFFFF */
+        stw_be_p(data + sloc, sum ? sum : 0xffff);
     }
 }
 
diff --git a/hw/net/net_rx_pkt.c b/hw/net/net_rx_pkt.c
index 1019b50..e820132 100644
--- a/hw/net/net_rx_pkt.c
+++ b/hw/net/net_rx_pkt.c
@@ -588,6 +588,9 @@  bool net_rx_pkt_fix_l4_csum(struct NetRxPkt *pkt)
 
     /* Calculate L4 checksum */
     csum = cpu_to_be16(_net_rx_pkt_calc_l4_csum(pkt));
+    if (!csum) {
+        csum = 0xFFFF; /* For UDP, zero checksum must be sent as 0xFFFF */
+    }
 
     /* Set calculated checksum to checksum word */
     iov_from_buf(pkt->vec, pkt->vec_len,
diff --git a/hw/net/net_tx_pkt.c b/hw/net/net_tx_pkt.c
index 20b2549..21194e7 100644
--- a/hw/net/net_tx_pkt.c
+++ b/hw/net/net_tx_pkt.c
@@ -136,6 +136,9 @@  void net_tx_pkt_update_ip_checksums(struct NetTxPkt *pkt)
         return;
     }
 
+    if (!csum) {
+        csum = 0xFFFF; /* For UDP, zero checksum must be sent as 0xFFFF */
+    }
     iov_from_buf(&pkt->vec[NET_TX_PKT_PL_START_FRAG], pkt->payload_frags,
                  pkt->virt_hdr.csum_offset, &csum, sizeof(csum));
 }
@@ -487,6 +490,9 @@  static void net_tx_pkt_do_sw_csum(struct NetTxPkt *pkt)
 
     /* Put the checksum obtained into the packet */
     csum = cpu_to_be16(net_checksum_finish(csum_cntr));
+    if (!csum) {
+        csum = 0xFFFF; /* For UDP, zero checksum must be sent as 0xFFFF */
+    }
     iov_from_buf(iov, iov_len, csum_offset, &csum, sizeof csum);
 }
 
diff --git a/hw/net/vmxnet3.c b/hw/net/vmxnet3.c
index 90f6943..de9c40e 100644
--- a/hw/net/vmxnet3.c
+++ b/hw/net/vmxnet3.c
@@ -942,6 +942,7 @@  static void vmxnet3_rx_need_csum_calculate(struct NetRxPkt *pkt,
     bool isip4, isip6, istcp, isudp;
     uint8_t *data;
     int len;
+    uint16_t sum;
 
     if (!net_rx_pkt_has_virt_hdr(pkt)) {
         return;
@@ -969,8 +970,10 @@  static void vmxnet3_rx_need_csum_calculate(struct NetRxPkt *pkt,
 
     data = (uint8_t *)pkt_data + vhdr->csum_start;
     len = pkt_len - vhdr->csum_start;
-    /* Put the checksum obtained into the packet */
-    stw_be_p(data + vhdr->csum_offset, net_raw_checksum(data, len));
+    sum = net_raw_checksum(data, len);
+    /* Put the checksum obtained into the packet; for UDP, zero checksum */
+    /* must be sent as 0xFFFF */
+    stw_be_p(data + vhdr->csum_offset, sum ? sum : 0xFFFF);
 
     vhdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
     vhdr->flags |= VIRTIO_NET_HDR_F_DATA_VALID;