Guenter Roeck
2021-Apr-11 22:20 UTC
[PATCH net] virtio_net: Do not pull payload in skb->head
On 4/11/21 2:43 PM, Eric Dumazet wrote:> On Sun, Apr 11, 2021 at 11:32 PM Guenter Roeck <linux at roeck-us.net> wrote: >> >> On 4/11/21 2:23 PM, Eric Dumazet wrote: >>> On Sun, Apr 11, 2021 at 10:37 PM Guenter Roeck <linux at roeck-us.net> wrote: >>>> >>>> On 4/11/21 8:06 AM, Eric Dumazet wrote: >>>>> On Sun, Apr 11, 2021 at 3:43 PM Guenter Roeck <linux at roeck-us.net> wrote: >>>>> >>>>>> This patch causes a virtio-net interface failure when booting sh4 images >>>>>> in qemu. The test case is nothing special: Just try to get an IP address >>>>>> using udhcpc. If it fails, udhcpc reports: >>>>>> >>>>>> udhcpc: started, v1.33.0 >>>>>> udhcpc: sending discover >>>>>> FAIL >>>>>> >>>>> >>>>> Can you investigate where the incoming packet is dropped ? >>>>> >>>> >>>> Unless I am missing something, packets are not dropped. It looks more >>>> like udhcpc gets bad indigestion in the receive path and exits immediately. >>>> Plus, it doesn't happen all the time; sometimes it receives the discover >>>> response and is able to obtain an IP address. >>>> >>>> Overall this is quite puzzling since udhcpc exits immediately when the problem >>>> is seen, no matter which option I give it on the command line; it should not >>>> really do that. >>> >>> >>> Could you strace both cases and report differences you can spot ? >>> >>> strace -o STRACE -f -s 1000 udhcpc >>> >> >> I'll give it a try. It will take a while; I'll need to add strace to my root >> file systems first. >> >> As a quick hack, I added some debugging into the kernel; it looks like >> the data part of the dhcp discover response may get lost with your patch >> in place. > > Data is not lost, the payload is whole contained in skb frags, which > was expected from my patch. > > Maybe this sh arch does something wrong in this case. > > This could be checksuming... > > Please check > > nstat -n > <run udhcpc> > nstat >Does that tell you anything ? / # nstat -n / # udhcpc -n -q udhcpc: started, v1.33.0 udhcpc: sending discover / # nstat #kernel IpInReceives 1 0.0 IpInDelivers 1 0.0 UdpIgnoredMulti 1 0.0 IpExtInBcastPkts 1 0.0 IpExtInOctets 576 0.0 IpExtInBcastOctets 576 0.0 IpExtInNoECTPkts 1 0.0 Also, one interesting detail is that the problem is not seen all the time, even with your patch in place. Not sure if I mentioned that before. strace output in the success case (same image, with patch in place) looks as follows. 130 write(2, "udhcpc: sending discover\n", 25) = 25 130 socket(AF_PACKET, SOCK_DGRAM, htons(ETH_P_IP)) = 6 130 bind(6, {sa_family=AF_PACKET, sll_protocol=htons(ETH_P_IP), sll_ifindex=if_nametoindex("eth0"), sll_hatype=ARPHRD_NETROM, sll_pkttype=PACKET_HOST, sll_halen=6, sll_addr=[0xff, 0xff, 0xff, 0xf 130 sendto(6, "E\0\1H\0\0\0\0@\21y\246\0\0\0\0\377\377\377\377\0D\0C\0014\227r\1\1\6\0\5\16\36P\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0RT\0\0224V\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0 130 close(6) = 0 130 fcntl64(5, F_SETFD, FD_CLOEXEC) = 0 130 clock_gettime64(CLOCK_MONOTONIC, {tv_sec=25, tv_nsec=202168729}) = 0 130 poll([{fd=3, events=POLLIN}, {fd=5, events=POLLIN}], 2, 3000) = 1 ([{fd=5, revents=POLLIN}]) 130 read(3, 0x7bf47a73, 1) = -1 EAGAIN (Resource temporarily unavailable) 130 recvmsg(5, {msg_name=NULL, msg_namelen=0, msg_iov=[{iov_base="E\20\2@\0\10\0\0@\21l\224\n\0\2\2\377\377\377\377\0C\0D\2,\230\23\2\1\6\0\5\16\36P\0\0\0\0\0\0\0\0\n\0\2\17\n\0\2\2\0\0\0\0RT\0\0 130 clock_gettime64(CLOCK_MONOTONIC, {tv_sec=25, tv_nsec=205504062}) = 0 130 fcntl64(5, F_SETFD, FD_CLOEXEC) = 0 130 socket(AF_INET, SOCK_RAW, IPPROTO_RAW) = 6 130 ioctl(6, SIOCGIFINDEX, {ifr_name="eth0", }) = 0 130 ioctl(6, SIOCGIFHWADDR, {ifr_name="eth0", ifr_hwaddr={sa_family=ARPHRD_ETHER, sa_data=52:54:00:12:34:56}}) = 0 130 close(6) = 0 130 clock_gettime64(CLOCK_MONOTONIC, {tv_sec=25, tv_nsec=208676862}) = 0 130 write(2, "udhcpc: sending select for 10.0.2.15\n", 37) = 37 130 socket(AF_PACKET, SOCK_DGRAM, htons(ETH_P_IP)) = 6 130 bind(6, {sa_family=AF_PACKET, sll_protocol=htons(ETH_P_IP), sll_ifindex=if_nametoindex("eth0"), sll_hatype=ARPHRD_NETROM, sll_pkttype=PACKET_HOST, sll_halen=6, sll_addr=[0xff, 0xff, 0xff, 0xf 130 sendto(6, "E\0\1H\0\0\0\0@\21y\246\0\0\0\0\377\377\377\377\0D\0C\0014\25Y\1\1\6\0\5\16\36P\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0RT\0\0224V\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\ 130 close(6) = 0 130 fcntl64(5, F_SETFD, FD_CLOEXEC) = 0 130 clock_gettime64(CLOCK_MONOTONIC, {tv_sec=25, tv_nsec=213060729}) = 0 130 poll([{fd=3, events=POLLIN}, {fd=5, events=POLLIN}], 2, 3000) = 1 ([{fd=5, revents=POLLIN}]) 130 read(3, 0x7bf47a73, 1) = -1 EAGAIN (Resource temporarily unavailable) 130 recvmsg(5, {msg_name=NULL, msg_namelen=0, msg_iov=[{iov_base="E\20\2@\0\t\0\0@\21l\223\n\0\2\2\377\377\377\377\0C\0D\2,\225\23\2\1\6\0\5\16\36P\0\0\0\0\0\0\0\0\n\0\2\17\n\0\2\2\0\0\0\0RT\0\02 130 clock_gettime64(CLOCK_MONOTONIC, {tv_sec=25, tv_nsec=215453662}) = 0 130 write(2, "udhcpc: lease of 10.0.2.15 obtained, lease time 86400\n", 54) = 54 In that case, the output of nstat is as follows. / # nstat #kernel IpInReceives 2 0.0 IpInDelivers 2 0.0 UdpIgnoredMulti 2 0.0 IpExtInBcastPkts 2 0.0 IpExtInOctets 1152 0.0 IpExtInBcastOctets 1152 0.0 IpExtInNoECTPkts 2 0.0 Guenter