UDP implementation and the MSG_MORE flag
Hello, we observe some unexpected behavior in the UDP implementation of the linux kernel. Some UDP packets send via the loopback interface are dropped in the kernel on the receive side when using sendto with the MSG_MORE flag. Every drop increases the InCsumErrors in /proc/self/net/snmp. Some example code to reproduce it is appended below. In the code we tracked it down to this code section. ( Even a little further but its unclear to me wy the csum() is wrong in the bad case) udpv6_recvmsg() ... if (checksum_valid || udp_skb_csum_unnecessary(skb)) { if (udp_skb_is_linear(skb)) err = copy_linear_skb(skb, copied, off, &msg->msg_iter); else err = skb_copy_datagram_msg(skb, off, msg, copied); } else { err = skb_copy_and_csum_datagram_msg(skb, off, msg); if (err == -EINVAL) { goto csum_copy_err; } } ... Perhaps someone with deeper knowledge can comment on this and can explain us the reason of this behavior. Best regards, Oliver udp-send.c #include <arpa/inet.h> #include <errno.h> #include <fcntl.h> #include <netinet/in.h> #include <poll.h> #include <signal.h> #include <stdbool.h> #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/socket.h> #include <sys/stat.h> #include <sys/types.h> #include <sys/un.h> #include <unistd.h> #define BUFFSIZE 512*1024 int main(int argc, char** argv) { int fd = 0; int port = 0; char *buffer; struct sockaddr_in addr; ssize_t addrlen = 0; if(argc == 2) { port = atoi(argv[1]); } else { port = 4711; } fd = socket(PF_INET, SOCK_DGRAM, 0); addr.sin_family = AF_INET; addr.sin_port = htons(port); addr.sin_addr.s_addr = inet_addr("127.0.0.1"); addrlen = sizeof(addr); buffer = malloc(BUFFSIZE); if (!buffer) { return 0; } printf("\nsending BROKEN segmented testdata on local port %i \n", port); snprintf(buffer, BUFFSIZE, "start-data {\n"); sendto(fd, buffer, strlen(buffer), MSG_MORE, (struct sockaddr *) &addr, addrlen); snprintf(buffer, BUFFSIZE, "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa\n"); sendto(fd, buffer, strlen(buffer), MSG_MORE, (struct sockaddr *) &addr, addrlen); snprintf(buffer, BUFFSIZE, "}\n"); sendto(fd, buffer, strlen(buffer), 0, (struct sockaddr *) &addr, addrlen); printf("\nsending VALID segmented testdata on local port %i \n", port); snprintf(buffer, BUFFSIZE, "start-data {\n"); sendto(fd, buffer, strlen(buffer), MSG_MORE, (struct sockaddr *) &addr, addrlen); snprintf(buffer, BUFFSIZE, "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa\n"); sendto(fd, buffer, strlen(buffer), MSG_MORE, (struct sockaddr *) &addr, addrlen); snprintf(buffer, BUFFSIZE, "}\n"); sendto(fd, buffer, strlen(buffer), 0, (struct sockaddr *) &addr, addrlen); printf("\nsending VALID unsegmented testdata on local port %i \n", port); snprintf(buffer, BUFFSIZE, "start-data {\n"); snprintf(buffer + strlen(buffer), BUFFSIZE - strlen(buffer), "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa\n"); snprintf(buffer+ strlen(buffer), BUFFSIZE - strlen(buffer), "}\n"); sendto(fd, buffer, strlen(buffer), 0, (struct sockaddr *) &addr, addrlen); free(buffer); return 0; } ------- udp-receive.c #include <arpa/inet.h> #include <errno.h> #include <fcntl.h> #include <netinet/in.h> #include <poll.h> #include <signal.h> #include <stdbool.h> #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/socket.h> #include <sys/stat.h> #include <sys/types.h> #include <sys/un.h> #include <unistd.h> int main(int argc, char** argv) { int fd = 0; int arg = 0; int ret = 0; struct sockaddr_in addr; ssize_t addrlen = 0; int port = 0; char *buffer; char *printbuffer; int recvlen = 0; if(argc == 2) { port = atoi(argv[1]); } else { port = 4711; } fd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP); addr.sin_family = AF_INET; addr.sin_port = htons(port); addr.sin_addr.s_addr = inet_addr("0.0.0.0"); addrlen = sizeof(addr); buffer = malloc(65536); if (!buffer) { return 0; } printbuffer = malloc(65537); if (!printbuffer) { return 0; } if(fd) { printf("\nbinding to local port %i \n", port); //bind ret = bind(fd, (struct sockaddr *)&addr, addrlen); printf("result error %i, errno %i\n", ret, errno); do { recvlen = recvfrom(fd, buffer, 65536, 0, NULL, NULL); if (recvlen >0) { printf("\nreceived %i bytes of data:\n", recvlen); memset(printbuffer, 0, 65537); memcpy(printbuffer, buffer, recvlen); printf("%s\n", printbuffer); } else if(recvlen < 0) { printf("\n receive error %i, errno %i\n", recvlen, errno); } } while(1); close(fd); } else { printf("\nerror creating socket\n"); } free(buffer); free(printbuffer); return 0; }
Hi! On 15:12 Tue 26 Jan , Oliver Graute wrote:
Some UDP packets send via the loopback interface are dropped in the kernel on the receive side when using sendto with the MSG_MORE flag. Every drop increases the InCsumErrors in /proc/self/net/snmp. Some example code to reproduce it is appended below.
In the code we tracked it down to this code section. ( Even a little further but its unclear to me wy the csum() is wrong in the bad case)
I think it is unlikely that the error is on the receiving side. I recommend to capture the data that is being sent with a sniffer and see if there is anything wrong. You may also want to try disabling hardware acceleration on your network devices, because the checksum is sometimes done in hardware. -Michi -- programing a layer 3+4 network protocol for mesh networks see http://michaelblizek.twilightparadox.com
participants (2)
-
michi1ļ¼ michaelblizek.twilightparadox.com -
Oliver Graute