The 'lo' interface on Linux doesn't support thinks like ARP or link-layer access like we use to talk to a normal network interface. A higher-level network API must be used to access localhost.
As written, this interface is limited to not supporting ICMP since the API doesn't allow the socket to be opened for all IP traffic and be able to receive at the same time. UDP is far more useful to test with, so it was selected over ICMP. Ping won't work, but things like TFTP should work.
Signed-off-by: Joe Hershberger joe.hershberger@ni.com
---
Changes in v5: -Add details about lo support to the README -Remove socket timeout -Separate init to 2 helper static functions -Set the socket to non-blocking -Use INADDR_LOOPBACK -Use more verbose comments
Changes in v4: -Added support for the 'lo' network interface
Changes in v3: None Changes in v2: None
arch/sandbox/cpu/eth-raw-os.c | 113 +++++++++++++++++++++++++++++++++- arch/sandbox/dts/sandbox.dts | 10 +++ arch/sandbox/include/asm/eth-raw-os.h | 10 ++- board/sandbox/README.sandbox | 22 +++++++ drivers/net/sandbox-raw.c | 71 ++++++++++++++++++++- 5 files changed, 221 insertions(+), 5 deletions(-)
diff --git a/arch/sandbox/cpu/eth-raw-os.c b/arch/sandbox/cpu/eth-raw-os.c index 601205a..b76a731 100644 --- a/arch/sandbox/cpu/eth-raw-os.c +++ b/arch/sandbox/cpu/eth-raw-os.c @@ -12,6 +12,8 @@ #include <fcntl.h> #include <net/if.h> #include <netinet/in.h> +#include <netinet/ip.h> +#include <netinet/udp.h> #include <stdio.h> #include <stdlib.h> #include <string.h> @@ -20,10 +22,11 @@ #include <sys/socket.h> #include <unistd.h>
+#include <arpa/inet.h> #include <linux/if_ether.h> #include <linux/if_packet.h>
-int sandbox_eth_raw_os_start(const char *ifname, unsigned char *ethmac, +static int _raw_packet_start(const char *ifname, unsigned char *ethmac, struct eth_sandbox_raw_priv *priv) { struct sockaddr_ll *device; @@ -89,14 +92,114 @@ int sandbox_eth_raw_os_start(const char *ifname, unsigned char *ethmac, return 0; }
+static int _local_inet_start(struct eth_sandbox_raw_priv *priv) +{ + struct sockaddr_in *device; + int ret; + int flags; + int one = 1; + + /* Prepare device struct */ + priv->device = malloc(sizeof(struct sockaddr_in)); + if (priv->device == NULL) + return -ENOMEM; + device = priv->device; + memset(device, 0, sizeof(struct sockaddr_in)); + device->sin_family = AF_INET; + device->sin_addr.s_addr = htonl(INADDR_LOOPBACK); + + /** + * Open socket + * Since we specify UDP here, any incoming ICMP packets will + * not be received, so things like ping will not work on this + * localhost interface. + */ + priv->sd = socket(AF_INET, SOCK_RAW, IPPROTO_UDP); + if (priv->sd < 0) { + printf("Failed to open socket: %d %s\n", errno, + strerror(errno)); + return -errno; + } + + /* Make the socket non-blocking */ + flags = fcntl(priv->sd, F_GETFL, 0); + fcntl(priv->sd, F_SETFL, flags | O_NONBLOCK); + + /* Include the UDP/IP headers on send and receive */ + ret = setsockopt(priv->sd, IPPROTO_IP, IP_HDRINCL, &one, + sizeof(one)); + if (ret < 0) { + printf("Failed to set header include option: %d %s\n", errno, + strerror(errno)); + return -errno; + } + priv->local_bind_sd = -1; + priv->local_bind_udp_port = 0; + return 0; +} + +int sandbox_eth_raw_os_start(const char *ifname, unsigned char *ethmac, + struct eth_sandbox_raw_priv *priv) +{ + if (priv->local) + return _local_inet_start(priv); + else + return _raw_packet_start(ifname, ethmac, priv); +} + int sandbox_eth_raw_os_send(void *packet, int length, - const struct eth_sandbox_raw_priv *priv) + struct eth_sandbox_raw_priv *priv) { int retval; + struct udphdr *udph = packet + sizeof(struct iphdr);
if (!priv->sd || !priv->device) return -EINVAL;
+ /* + * This block of code came about when testing tftp on the localhost + * interface. When using the RAW AF_INET API, the network stack is still + * in play responding to incoming traffic based on open "ports". Since + * it is raw (at the IP layer, no Ethernet) the network stack tells the + * TFTP server that the port it responded to is closed. This causes the + * TFTP transfer to be aborted. This block of code inspects the outgoing + * packet as formulated by the u-boot network stack to determine the + * source port (that the TFTP server will send packets back to) and + * opens a typical UDP socket on that port, thus preventing the network + * stack from sending that ICMP message claiming that the port has no + * bound socket. + */ + if (priv->local && (priv->local_bind_sd == -1 || + priv->local_bind_udp_port != udph->source)) { + struct iphdr *iph = packet; + struct sockaddr_in addr; + + if (priv->local_bind_sd != -1) + close(priv->local_bind_sd); + + /* A normal UDP socket is required to bind */ + priv->local_bind_sd = socket(AF_INET, SOCK_DGRAM, 0); + if (priv->local_bind_sd < 0) { + printf("Failed to open bind sd: %d %s\n", errno, + strerror(errno)); + return -errno; + } + priv->local_bind_udp_port = udph->source; + + /** + * Bind the UDP port that we intend to use as our source port + * so that the kernel will not send an ICMP port unreachable + * message to the server + */ + addr.sin_family = AF_INET; + addr.sin_port = udph->source; + addr.sin_addr.s_addr = iph->saddr; + retval = bind(priv->local_bind_sd, &addr, sizeof(addr)); + if (retval < 0) + printf("Failed to bind: %d %s\n", errno, + strerror(errno)); + } + retval = sendto(priv->sd, packet, length, 0, (struct sockaddr *)priv->device, sizeof(struct sockaddr_ll)); @@ -137,4 +240,10 @@ void sandbox_eth_raw_os_stop(struct eth_sandbox_raw_priv *priv) priv->device = NULL; close(priv->sd); priv->sd = -1; + if (priv->local) { + if (priv->local_bind_sd != -1) + close(priv->local_bind_sd); + priv->local_bind_sd = -1; + priv->local_bind_udp_port = 0; + } } diff --git a/arch/sandbox/dts/sandbox.dts b/arch/sandbox/dts/sandbox.dts index 8002196..0eaa5a5 100644 --- a/arch/sandbox/dts/sandbox.dts +++ b/arch/sandbox/dts/sandbox.dts @@ -4,6 +4,10 @@ #address-cells = <1>; #size-cells = <0>;
+ aliases { + eth5 = "/eth@90000000"; + }; + chosen { stdout-path = "/serial"; }; @@ -192,4 +196,10 @@ reg = <0x80000000 0x1000>; host-raw-interface = "eth0"; }; + + eth@90000000 { + compatible = "sandbox,eth-raw"; + reg = <0x90000000 0x1000>; + host-raw-interface = "lo"; + }; }; diff --git a/arch/sandbox/include/asm/eth-raw-os.h b/arch/sandbox/include/asm/eth-raw-os.h index df60c4f..ed4b2e2 100644 --- a/arch/sandbox/include/asm/eth-raw-os.h +++ b/arch/sandbox/include/asm/eth-raw-os.h @@ -15,16 +15,24 @@ * * sd: socket descriptor - the open socket during a session * device: struct sockaddr_ll - the host interface packets move to/from + * local: 1 or 0 to select the local interface ('lo') or not + * local_bindsd: socket descriptor to prevent the kernel from sending + * a message to the server claiming the port is + * unreachable + * local_bind_udp_port: The UDP port number that we bound to */ struct eth_sandbox_raw_priv { int sd; void *device; + int local; + int local_bind_sd; + unsigned short local_bind_udp_port; };
int sandbox_eth_raw_os_start(const char *ifname, unsigned char *ethmac, struct eth_sandbox_raw_priv *priv); int sandbox_eth_raw_os_send(void *packet, int length, - const struct eth_sandbox_raw_priv *priv); + struct eth_sandbox_raw_priv *priv); int sandbox_eth_raw_os_recv(void *packet, int *length, const struct eth_sandbox_raw_priv *priv); void sandbox_eth_raw_os_stop(struct eth_sandbox_raw_priv *priv); diff --git a/board/sandbox/README.sandbox b/board/sandbox/README.sandbox index aedf05a..73743e0 100644 --- a/board/sandbox/README.sandbox +++ b/board/sandbox/README.sandbox @@ -241,6 +241,28 @@ dhcp set serverip WWW.XXX.YYY.ZZZ tftpboot u-boot.bin
+The bridge also support (to a lesser extent) the localhost inderface, 'lo'. + +The 'lo' interface cannot use the RAW AF_PACKET API because the lo interface +doesn't support Ethernet-level traffic. It is a higher-level interface that is +expected only to be used at the AF_INET level of the API. As such, the most raw +we can get on that interface is the RAW AF_INET API on UDP. This allows us to +set the IP_HDRINCL option to include everything except the Ethernet header in +the packets we send and receive. + +Because only UDP is supported, ICMP traffic will not work, so expect that ping +commands will time out. + +The default device tree for sandbox includes an entry for lo on the sandbox +host machine whose alias is "eth5". The following is an example of a network +operation being tested on the lo interface. + +TFTP +.... + +set ethact eth5 +tftpboot u-boot.bin +
SPI Emulation ------------- diff --git a/drivers/net/sandbox-raw.c b/drivers/net/sandbox-raw.c index 435b874..91da5f5 100644 --- a/drivers/net/sandbox-raw.c +++ b/drivers/net/sandbox-raw.c @@ -15,6 +15,8 @@
DECLARE_GLOBAL_DATA_PTR;
+static int reply_arp; +static IPaddr_t arp_ip;
static int sb_eth_raw_start(struct udevice *dev) { @@ -29,6 +31,11 @@ static int sb_eth_raw_start(struct udevice *dev) if (interface == NULL) return -EINVAL;
+ if (strcmp(interface, "lo") == 0) { + priv->local = 1; + setenv("ipaddr", "127.0.0.1"); + setenv("serverip", "127.0.0.1"); + } return sandbox_eth_raw_os_start(interface, pdata->enetaddr, priv); }
@@ -38,18 +45,78 @@ static int sb_eth_raw_send(struct udevice *dev, void *packet, int length)
debug("eth_sandbox_raw: Send packet %d\n", length);
+ if (priv->local) { + struct ethernet_hdr *eth = packet; + + if (ntohs(eth->et_protlen) == PROT_ARP) { + struct arp_hdr *arp = packet + ETHER_HDR_SIZE; + + /** + * localhost works on a higher-level API in Linux than + * ARP packets, so fake it + */ + arp_ip = NetReadIP(&arp->ar_tpa); + reply_arp = 1; + return 0; + } + packet += ETHER_HDR_SIZE; + length -= ETHER_HDR_SIZE; + } return sandbox_eth_raw_os_send(packet, length, priv); }
static int sb_eth_raw_recv(struct udevice *dev, uchar **packetp) { + struct eth_pdata *pdata = dev_get_platdata(dev); struct eth_sandbox_raw_priv *priv = dev_get_priv(dev); - int retval; + int retval = 0; int length;
- retval = sandbox_eth_raw_os_recv(net_rx_packets[0], &length, priv); + if (reply_arp) { + struct arp_hdr *arp = (void *)net_rx_packets[0] + + ETHER_HDR_SIZE; + + /* + * Fake an ARP response. The u-boot network stack is sending an + * ARP request (to find the MAC address to address the actual + * packet to) and requires an ARP response to continue. Since + * this is the localhost interface, there is no Etherent level + * traffic at all, so there is no way to send an ARP request or + * to get a response. For this reason we fake the response to + * make the u-boot network stack happy. + */ + arp->ar_hrd = htons(ARP_ETHER); + arp->ar_pro = htons(PROT_IP); + arp->ar_hln = ARP_HLEN; + arp->ar_pln = ARP_PLEN; + arp->ar_op = htons(ARPOP_REPLY); + /* Any non-zero MAC address will work */ + memset(&arp->ar_sha, 0x01, ARP_HLEN); + /* Use whatever IP we were looking for (always 127.0.0.1?) */ + NetWriteIP(&arp->ar_spa, arp_ip); + memcpy(&arp->ar_tha, pdata->enetaddr, ARP_HLEN); + NetWriteIP(&arp->ar_tpa, NetOurIP); + length = ARP_HDR_SIZE; + } else { + /* If local, the Ethernet header won't be included; skip it */ + uchar *pktptr = priv->local ? + net_rx_packets[0] + ETHER_HDR_SIZE : net_rx_packets[0]; + + retval = sandbox_eth_raw_os_recv(pktptr, &length, priv); + }
if (!retval && length) { + if (priv->local) { + struct ethernet_hdr *eth = (void *)net_rx_packets[0]; + + /* Fill in enough of the missing Ethernet header */ + memcpy(eth->et_dest, pdata->enetaddr, ARP_HLEN); + memset(eth->et_src, 0x01, ARP_HLEN); + eth->et_protlen = htons(reply_arp ? PROT_ARP : PROT_IP); + reply_arp = 0; + length += ETHER_HDR_SIZE; + } + debug("eth_sandbox_raw: received packet %d\n", length); *packetp = net_rx_packets[0];