diff -up dhcp-4.2.4-P2/client/dhclient.c.lpf-ib dhcp-4.2.4-P2/client/dhclient.c --- dhcp-4.2.4-P2/client/dhclient.c.lpf-ib 2012-11-30 14:05:07.399394154 +0100 +++ dhcp-4.2.4-P2/client/dhclient.c 2012-11-30 14:05:07.406394070 +0100 @@ -113,6 +113,8 @@ static int check_domain_name_list(const static int check_option_values(struct universe *universe, unsigned int opt, const char *ptr, size_t len); +static void setup_ib_interface(struct interface_info *ip); + int main(int argc, char **argv) { int fd; @@ -909,6 +911,14 @@ main(int argc, char **argv) { } srandom(seed + cur_time + (unsigned)getpid()); + /* Setup specific Infiniband options */ + for (ip = interfaces; ip; ip = ip->next) { + if (ip->client && + (ip->hw_address.hbuf[0] == HTYPE_INFINIBAND)) { + setup_ib_interface(ip); + } + } + /* Start a configuration state machine for each interface. */ #ifdef DHCPv6 if (local_family == AF_INET6) { @@ -1185,6 +1195,29 @@ int find_subnet (struct subnet **sp, return 0; } +static void setup_ib_interface(struct interface_info *ip) +{ + struct group *g; + + /* Set the broadcast flag */ + ip->client->config->bootp_broadcast_always = 1; + + /* + * Find out if a dhcp-client-identifier option was specified either + * in the config file or on the command line + */ + for (g = ip->client->config->on_transmission; g != NULL; g = g->next) { + if ((g->statements != NULL) && + (strcmp(g->statements->data.option->option->name, + "dhcp-client-identifier") == 0)) { + return; + } + } + + /* No client ID specified */ + log_fatal("dhcp-client-identifier must be specified for InfiniBand"); +} + /* Individual States: * * Each routine is called from the dhclient_state_machine() in one of diff -up dhcp-4.2.4-P2/common/bpf.c.lpf-ib dhcp-4.2.4-P2/common/bpf.c --- dhcp-4.2.4-P2/common/bpf.c.lpf-ib 2012-11-30 14:05:07.394394214 +0100 +++ dhcp-4.2.4-P2/common/bpf.c 2012-11-30 14:05:07.407394058 +0100 @@ -198,11 +198,44 @@ struct bpf_insn dhcp_bpf_filter [] = { BPF_STMT(BPF_RET+BPF_K, 0), }; +/* Packet filter program for DHCP over Infiniband. + * + * XXX + * Changes to the filter program may require changes to the constant offsets + * used in lpf_gen_filter_setup to patch the port in the BPF program! + * XXX + */ +struct bpf_insn dhcp_ib_bpf_filter [] = { + /* Packet filter for Infiniband */ + /* Make sure it's a UDP packet... */ + BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 9), + BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 6), + + /* Make sure this isn't a fragment... */ + BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 6), + BPF_JUMP(BPF_JMP + BPF_JSET + BPF_K, 0x1fff, 4, 0), + + /* Get the IP header length... */ + BPF_STMT(BPF_LDX + BPF_B + BPF_MSH, 0), + + /* Make sure it's to the right port... */ + BPF_STMT(BPF_LD + BPF_H + BPF_IND, 2), + BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, 67, 0, 1), + + /* If we passed all the tests, ask for the whole packet. */ + BPF_STMT(BPF_RET + BPF_K, (u_int)-1), + + /* Otherwise, drop it. */ + BPF_STMT(BPF_RET + BPF_K, 0), +}; + #if defined (DEC_FDDI) struct bpf_insn *bpf_fddi_filter; #endif int dhcp_bpf_filter_len = sizeof dhcp_bpf_filter / sizeof (struct bpf_insn); +int dhcp_ib_bpf_filter_len = sizeof dhcp_ib_bpf_filter / sizeof (struct bpf_insn); + #if defined (HAVE_TR_SUPPORT) struct bpf_insn dhcp_bpf_tr_filter [] = { /* accept all token ring packets due to variable length header */ diff -up dhcp-4.2.4-P2/common/lpf.c.lpf-ib dhcp-4.2.4-P2/common/lpf.c --- dhcp-4.2.4-P2/common/lpf.c.lpf-ib 2012-11-30 14:05:07.394394214 +0100 +++ dhcp-4.2.4-P2/common/lpf.c 2012-11-30 14:19:27.211031532 +0100 @@ -42,6 +42,7 @@ #include "includes/netinet/udp.h" #include "includes/netinet/if_ether.h" #include +#include #ifndef PACKET_AUXDATA #define PACKET_AUXDATA 8 @@ -59,6 +60,15 @@ struct tpacket_auxdata /* Reinitializes the specified interface after an address change. This is not required for packet-filter APIs. */ +/* Default broadcast address for IPoIB */ +static unsigned char default_ib_bcast_addr[20] = { + 0x00, 0xff, 0xff, 0xff, + 0xff, 0x12, 0x40, 0x1b, + 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, + 0xff, 0xff, 0xff, 0xff +}; + #ifdef USE_LPF_SEND void if_reinitialize_send (info) struct interface_info *info; @@ -86,10 +96,21 @@ int if_register_lpf (info) struct sockaddr common; } sa; struct ifreq ifr; + int type; + int protocol; /* Make an LPF socket. */ - if ((sock = socket(PF_PACKET, SOCK_RAW, - htons((short)ETH_P_ALL))) < 0) { + get_hw_addr(info); + + if (info->hw_address.hbuf[0] == HTYPE_INFINIBAND) { + type = SOCK_DGRAM; + protocol = ETHERTYPE_IP; + } else { + type = SOCK_RAW; + protocol = ETH_P_ALL; + } + + if ((sock = socket(PF_PACKET, type, htons((short)protocol))) < 0) { if (errno == ENOPROTOOPT || errno == EPROTONOSUPPORT || errno == ESOCKTNOSUPPORT || errno == EPFNOSUPPORT || errno == EAFNOSUPPORT || errno == EINVAL) { @@ -112,6 +133,7 @@ int if_register_lpf (info) /* Bind to the interface name */ memset (&sa, 0, sizeof sa); sa.ll.sll_family = AF_PACKET; + sa.ll.sll_protocol = htons(protocol); sa.ll.sll_ifindex = ifr.ifr_ifindex; if (bind (sock, &sa.common, sizeof sa)) { if (errno == ENOPROTOOPT || errno == EPROTONOSUPPORT || @@ -127,8 +149,6 @@ int if_register_lpf (info) log_fatal ("Bind socket to interface: %m"); } - get_hw_addr(info->name, &info->hw_address); - return sock; } #endif /* USE_LPF_SEND || USE_LPF_RECEIVE */ @@ -183,6 +203,8 @@ void if_deregister_send (info) in bpf includes... */ extern struct sock_filter dhcp_bpf_filter []; extern int dhcp_bpf_filter_len; +extern struct sock_filter dhcp_ib_bpf_filter []; +extern int dhcp_ib_bpf_filter_len; #if defined (HAVE_TR_SUPPORT) extern struct sock_filter dhcp_bpf_tr_filter []; @@ -200,11 +222,13 @@ void if_register_receive (info) /* Open a LPF device and hang it on this interface... */ info -> rfdesc = if_register_lpf (info); - val = 1; - if (setsockopt (info -> rfdesc, SOL_PACKET, PACKET_AUXDATA, &val, - sizeof val) < 0) { - if (errno != ENOPROTOOPT) - log_fatal ("Failed to set auxiliary packet data: %m"); + if (info->hw_address.hbuf[0] != HTYPE_INFINIBAND) { + val = 1; + if (setsockopt (info -> rfdesc, SOL_PACKET, PACKET_AUXDATA, + &val, sizeof val) < 0) { + if (errno != ENOPROTOOPT) + log_fatal ("Failed to set auxiliary packet data: %m"); + } } #if defined (HAVE_TR_SUPPORT) @@ -250,15 +274,28 @@ static void lpf_gen_filter_setup (info) memset(&p, 0, sizeof(p)); - /* Set up the bpf filter program structure. This is defined in - bpf.c */ - p.len = dhcp_bpf_filter_len; - p.filter = dhcp_bpf_filter; - - /* Patch the server port into the LPF program... - XXX changes to filter program may require changes - to the insn number(s) used below! XXX */ - dhcp_bpf_filter [8].k = ntohs ((short)local_port); + if (info->hw_address.hbuf[0] == HTYPE_INFINIBAND) { + /* Set up the bpf filter program structure. */ + p.len = dhcp_ib_bpf_filter_len; + p.filter = dhcp_ib_bpf_filter; + + /* Patch the server port into the LPF program... + XXX + changes to filter program may require changes + to the insn number(s) used below! + XXX */ + dhcp_ib_bpf_filter[6].k = ntohs ((short)local_port); + } else { + /* Set up the bpf filter program structure. + This is defined in bpf.c */ + p.len = dhcp_bpf_filter_len; + p.filter = dhcp_bpf_filter; + + /* Patch the server port into the LPF program... + XXX changes to filter program may require changes + to the insn number(s) used below! XXX */ + dhcp_bpf_filter [8].k = ntohs ((short)local_port); + } if (setsockopt (info -> rfdesc, SOL_SOCKET, SO_ATTACH_FILTER, &p, sizeof p) < 0) { @@ -315,6 +352,54 @@ static void lpf_tr_filter_setup (info) #endif /* USE_LPF_RECEIVE */ #ifdef USE_LPF_SEND +ssize_t send_packet_ib(interface, packet, raw, len, from, to, hto) + struct interface_info *interface; + struct packet *packet; + struct dhcp_packet *raw; + size_t len; + struct in_addr from; + struct sockaddr_in *to; + struct hardware *hto; +{ + unsigned ibufp = 0; + double ih [1536 / sizeof (double)]; + unsigned char *buf = (unsigned char *)ih; + ssize_t result; + + union sockunion { + struct sockaddr sa; + struct sockaddr_ll sll; + struct sockaddr_storage ss; + } su; + + assemble_udp_ip_header (interface, buf, &ibufp, from.s_addr, + to->sin_addr.s_addr, to->sin_port, + (unsigned char *)raw, len); + memcpy (buf + ibufp, raw, len); + + memset(&su, 0, sizeof(su)); + su.sll.sll_family = AF_PACKET; + su.sll.sll_protocol = htons(ETHERTYPE_IP); + + if (!(su.sll.sll_ifindex = if_nametoindex(interface->name))) { + errno = ENOENT; + log_error ("send_packet_ib: %m - failed to get if index"); + return -1; + } + + su.sll.sll_hatype = htons(HTYPE_INFINIBAND); + su.sll.sll_halen = sizeof(interface->bcast_addr); + memcpy(&su.sll.sll_addr, interface->bcast_addr, 20); + + result = sendto(interface->wfdesc, buf, ibufp + len, 0, + &su.sa, sizeof(su)); + + if (result < 0) + log_error ("send_packet_ib: %m"); + + return result; +} + ssize_t send_packet (interface, packet, raw, len, from, to, hto) struct interface_info *interface; struct packet *packet; @@ -335,6 +420,11 @@ ssize_t send_packet (interface, packet, return send_fallback (interface, packet, raw, len, from, to, hto); + if (interface->hw_address.hbuf[0] == HTYPE_INFINIBAND) { + return send_packet_ib(interface, packet, raw, len, from, + to, hto); + } + if (hto == NULL && interface->anycast_mac_addr.hlen) hto = &interface->anycast_mac_addr; @@ -356,6 +446,42 @@ ssize_t send_packet (interface, packet, #endif /* USE_LPF_SEND */ #ifdef USE_LPF_RECEIVE +ssize_t receive_packet_ib (interface, buf, len, from, hfrom) + struct interface_info *interface; + unsigned char *buf; + size_t len; + struct sockaddr_in *from; + struct hardware *hfrom; +{ + int length = 0; + int offset = 0; + unsigned char ibuf [1536]; + unsigned bufix = 0; + unsigned paylen; + + length = read(interface->rfdesc, ibuf, sizeof(ibuf)); + + if (length <= 0) + return length; + + offset = decode_udp_ip_header(interface, ibuf, bufix, from, + (unsigned)length, &paylen, 0); + + if (offset < 0) + return 0; + + bufix += offset; + length -= offset; + + if (length < paylen) + log_fatal("Internal inconsistency at %s:%d.", MDL); + + /* Copy out the data in the packet... */ + memcpy(buf, &ibuf[bufix], paylen); + + return (ssize_t)paylen; +} + ssize_t receive_packet (interface, buf, len, from, hfrom) struct interface_info *interface; unsigned char *buf; @@ -382,6 +508,10 @@ ssize_t receive_packet (interface, buf, }; struct cmsghdr *cmsg; + if (interface->hw_address.hbuf[0] == HTYPE_INFINIBAND) { + return receive_packet_ib(interface, buf, len, from, hfrom); + } + length = recvmsg (interface -> rfdesc, &msg, 0); if (length <= 0) return length; @@ -461,11 +591,32 @@ void maybe_setup_fallback () } } -void -get_hw_addr(const char *name, struct hardware *hw) { +struct sockaddr_ll * +get_ll (struct ifaddrs *ifaddrs, struct ifaddrs **ifa, char *name) +{ + for (*ifa = ifaddrs; *ifa != NULL; *ifa = (*ifa)->ifa_next) { + if ((*ifa)->ifa_addr == NULL) + continue; + + if ((*ifa)->ifa_addr->sa_family != AF_PACKET) + continue; + + if ((*ifa)->ifa_flags & IFF_LOOPBACK) + continue; + + if (strcmp((*ifa)->ifa_name, name) == 0) + return (struct sockaddr_ll *)(void *)(*ifa)->ifa_addr; + } + return NULL; +} + +struct sockaddr_ll * +ioctl_get_ll(char *name) +{ int sock; struct ifreq tmp; - struct sockaddr *sa; + struct sockaddr *sa = NULL; + struct sockaddr_ll *sll = NULL; if (strlen(name) >= sizeof(tmp.ifr_name)) { log_fatal("Device name too long: \"%s\"", name); @@ -479,16 +630,52 @@ get_hw_addr(const char *name, struct har memset(&tmp, 0, sizeof(tmp)); strcpy(tmp.ifr_name, name); if (ioctl(sock, SIOCGIFHWADDR, &tmp) < 0) { - log_fatal("Error getting hardware address for \"%s\": %m", + log_fatal("Error getting hardware address for \"%s\": %m", name); } + close(sock); sa = &tmp.ifr_hwaddr; - switch (sa->sa_family) { + // needs to be freed outside this function + sll = dmalloc (sizeof (struct sockaddr_ll), MDL); + if (!sll) + log_fatal("Unable to allocate memory for link layer address"); + memcpy(&sll->sll_hatype, &sa->sa_family, sizeof (sll->sll_hatype)); + memcpy(sll->sll_addr, sa->sa_data, sizeof (sll->sll_addr)); + return sll; +} + +void +get_hw_addr(struct interface_info *info) +{ + struct hardware *hw = &info->hw_address; + char *name = info->name; + struct ifaddrs *ifaddrs = NULL; + struct ifaddrs *ifa = NULL; + struct sockaddr_ll *sll = NULL; + int sll_allocated = 0; + + if (getifaddrs(&ifaddrs) == -1) + log_fatal("Failed to get interfaces"); + + if ((sll = get_ll(ifaddrs, &ifa, name)) == NULL) { + /* + * We were unable to get link-layer address for name. + * Fall back to ioctl(SIOCGIFHWADDR). + */ + sll = ioctl_get_ll(name); + if (sll != NULL) + sll_allocated = 1; + else + // shouldn't happed + log_fatal("Unexpected internal error"); + } + + switch (sll->sll_hatype) { case ARPHRD_ETHER: hw->hlen = 7; hw->hbuf[0] = HTYPE_ETHER; - memcpy(&hw->hbuf[1], sa->sa_data, 6); + memcpy(&hw->hbuf[1], sll->sll_addr, 6); break; case ARPHRD_IEEE802: #ifdef ARPHRD_IEEE802_TR @@ -496,18 +683,35 @@ get_hw_addr(const char *name, struct har #endif /* ARPHRD_IEEE802_TR */ hw->hlen = 7; hw->hbuf[0] = HTYPE_IEEE802; - memcpy(&hw->hbuf[1], sa->sa_data, 6); + memcpy(&hw->hbuf[1], sll->sll_addr, 6); break; case ARPHRD_FDDI: hw->hlen = 7; hw->hbuf[0] = HTYPE_FDDI; - memcpy(&hw->hbuf[1], sa->sa_data, 6); + memcpy(&hw->hbuf[1], sll->sll_addr, 6); + break; + case ARPHRD_INFINIBAND: + /* For Infiniband, save the broadcast address and store + * the port GUID into the hardware address. + */ + if (ifa->ifa_flags & IFF_BROADCAST) { + struct sockaddr_ll *bll; + + bll = (struct sockaddr_ll *)ifa->ifa_broadaddr; + memcpy(&info->bcast_addr, bll->sll_addr, 20); + } else { + memcpy(&info->bcast_addr, default_ib_bcast_addr, + 20); + } + + hw->hlen = 1; + hw->hbuf[0] = HTYPE_INFINIBAND; break; #if defined(ARPHRD_PPP) case ARPHRD_PPP: if (local_family != AF_INET6) - log_fatal("Unsupported device type %d for \"%s\"", - sa->sa_family, name); + log_fatal("local_family != AF_INET6 for \"%s\"", + name); hw->hlen = 0; hw->hbuf[0] = HTYPE_RESERVED; /* 0xdeadbeef should never occur on the wire, @@ -520,10 +724,13 @@ get_hw_addr(const char *name, struct har break; #endif default: - log_fatal("Unsupported device type %ld for \"%s\"", - (long int)sa->sa_family, name); + freeifaddrs(ifaddrs); + log_fatal("Unsupported device type %h for \"%s\"", + sll->sll_hatype, name); } - close(sock); + if (sll_allocated) + dfree(sll, MDL); + freeifaddrs(ifaddrs); } #endif diff -up dhcp-4.2.4-P2/common/socket.c.lpf-ib dhcp-4.2.4-P2/common/socket.c --- dhcp-4.2.4-P2/common/socket.c.lpf-ib 2012-08-24 21:11:21.000000000 +0200 +++ dhcp-4.2.4-P2/common/socket.c 2012-11-30 14:05:07.408394046 +0100 @@ -325,7 +325,7 @@ void if_register_send (info) info->wfdesc = if_register_socket(info, AF_INET, 0); /* If this is a normal IPv4 address, get the hardware address. */ if (strcmp(info->name, "fallback") != 0) - get_hw_addr(info->name, &info->hw_address); + get_hw_addr(info); #if defined (USE_SOCKET_FALLBACK) /* Fallback only registers for send, but may need to receive as well. */ @@ -388,7 +388,7 @@ void if_register_receive (info) #endif /* IP_PKTINFO... */ /* If this is a normal IPv4 address, get the hardware address. */ if (strcmp(info->name, "fallback") != 0) - get_hw_addr(info->name, &info->hw_address); + get_hw_addr(info); if (!quiet_interface_discovery) log_info ("Listening on Socket/%s%s%s", @@ -498,7 +498,7 @@ if_register6(struct interface_info *info if (req_multi) if_register_multicast(info); - get_hw_addr(info->name, &info->hw_address); + get_hw_addr(info); if (!quiet_interface_discovery) { if (info->shared_network != NULL) { diff -up dhcp-4.2.4-P2/includes/dhcpd.h.lpf-ib dhcp-4.2.4-P2/includes/dhcpd.h --- dhcp-4.2.4-P2/includes/dhcpd.h.lpf-ib 2012-11-30 14:05:07.400394142 +0100 +++ dhcp-4.2.4-P2/includes/dhcpd.h 2012-11-30 14:05:07.409394034 +0100 @@ -1249,6 +1249,7 @@ struct interface_info { struct shared_network *shared_network; /* Networks connected to this interface. */ struct hardware hw_address; /* Its physical address. */ + u_int8_t bcast_addr[20]; /* Infiniband broadcast address */ struct in_addr *addresses; /* Addresses associated with this * interface. */ @@ -2368,7 +2369,7 @@ void print_dns_status (int, struct dhcp_ #endif const char *print_time(TIME); -void get_hw_addr(const char *name, struct hardware *hw); +void get_hw_addr(struct interface_info *info); /* socket.c */ #if defined (USE_SOCKET_SEND) || defined (USE_SOCKET_RECEIVE) \