From abe4a8942044a29acd68dc7faf4ac09563e9b921 Mon Sep 17 00:00:00 2001 From: "Alex Xu (Hello71)" Date: Sun, 26 Jun 2016 16:29:32 -0400 Subject: Initial commit --- .gitignore | 2 + Makefile | 15 + README | 17 + doc/PORTABILITY | 7 + doc/TODO | 13 + src/client.c | 460 +++++++++++++++++++++++++++ src/client.h | 1 + src/common.h | 5 + src/server.c | 327 +++++++++++++++++++ src/server.h | 1 + src/udpastcp.c | 25 ++ src/uthash.h | 966 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 12 files changed, 1839 insertions(+) create mode 100644 .gitignore create mode 100644 Makefile create mode 100644 README create mode 100644 doc/PORTABILITY create mode 100644 doc/TODO create mode 100644 src/client.c create mode 100644 src/client.h create mode 100644 src/common.h create mode 100644 src/server.c create mode 100644 src/server.h create mode 100644 src/udpastcp.c create mode 100644 src/uthash.h diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..edfb0a4 --- /dev/null +++ b/.gitignore @@ -0,0 +1,2 @@ +/src/*.o +/udpastcp diff --git a/Makefile b/Makefile new file mode 100644 index 0000000..c1b789c --- /dev/null +++ b/Makefile @@ -0,0 +1,15 @@ +CFLAGS += -Wall -Wextra + +LDLIBS := -lev + +NET_OBJS := src/client.o src/server.o +OBJS := src/udpastcp.o $(NET_OBJS) + +udpastcp: $(OBJS) + $(LINK.c) $^ $(LOADLIBES) $(LDLIBS) -o $@ + +# networking code needs aliasing to be efficient +$(NET_OBJS): CFLAGS+=-fno-strict-aliasing -Wno-sign-compare + +clean: + $(RM) $(OBJS) diff --git a/README b/README new file mode 100644 index 0000000..e90a036 --- /dev/null +++ b/README @@ -0,0 +1,17 @@ +This program hides UDP traffic as TCP traffic in order to bypass certain +firewalls. + +It is not designed to bypass rigorous deep packet inspection; for example, +there is no window scaling; all data received as UDP will be immediately +transmitted as TCP with a fixed window size. There is also no retransmission, +although that could be implemented (at a detriment to the overall network health). + +This program is intended to be C99 and POSIX compatible, and requires only +libev. If desired, it should be easily portable to use libevent instead, as it +uses only basic event-based programming concepts. + +example usage: + server# openvpn --proto udp --port 1194 + server# udpintcp server localhost 11940 localhost 1194 + client# udpintcp client localhost 1194 server 11940 + client# openvpn --remote localhost 11940 udp diff --git a/doc/PORTABILITY b/doc/PORTABILITY new file mode 100644 index 0000000..fec5deb --- /dev/null +++ b/doc/PORTABILITY @@ -0,0 +1,7 @@ +Most C99 + POSIX systems should be supported. Windows is not, since it does not +support Unix-style events, and (although this is a minor concern) it does not +support sendmsg and recvmsg. + +In addition, since this program depends on libev, the portability requirements +of libev must be satisfied: +http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod#PORTABILITY_NOTES diff --git a/doc/TODO b/doc/TODO new file mode 100644 index 0000000..1b08be3 --- /dev/null +++ b/doc/TODO @@ -0,0 +1,13 @@ +- FIN open TCP connections on exit + +- add 'TCP Fast Open' support + +- optionally return ACKs and retransmit if no ACK + +- change some of the tiny hash tables to linked lists + +- allow more than 32768 connections to multiple hosts? + +- add fake TCP options? + +- add Windows support (not likely) diff --git a/src/client.c b/src/client.c new file mode 100644 index 0000000..00f7a89 --- /dev/null +++ b/src/client.c @@ -0,0 +1,460 @@ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#ifdef CRYPTO +#include +#include +#include +#endif + +#include "common.h" +#include "client.h" +#include "uthash.h" + +struct o_c_rsock { + struct sockaddr *r_addr; + struct o_c_sock *o_socks_by_lport; + struct c_data *c_data; + ev_io io_w; + ev_timer tm_w; + UT_hash_handle hh; + int fd; + socklen_t r_addrlen; +}; + +struct o_c_sock { + struct sockaddr *c_address; + struct o_c_rsock *rsock; + char *pending_data; + size_t pending_data_size; + ev_timer tm_w; + UT_hash_handle hh_lp; + UT_hash_handle hh_ca; + uint16_t seq_num; + in_port_t l_port; + in_port_t r_port; + uint8_t status; +}; + +#define PORTS_IN_INT sizeof(int) * CHAR_BIT + +struct c_data { + const char *r_host; + const char *r_port; + struct o_c_sock *o_socks_by_caddr; + struct o_c_rsock *o_rsocks; + struct sockaddr_storage pkt_addr; + socklen_t s_addrlen; + unsigned int used_ports[32768 / PORTS_IN_INT]; + int s_sock; +}; + +static struct c_data *global_c_data; + +/* reserve a local TCP port (local addr, remote addr, remote port are usually + * fixed in the tuple) */ +static uint16_t reserve_port(unsigned int *used_ports) { + // pick a starting port for the search + uint16_t spoff = random() % 32768; + size_t smoff = spoff / PORTS_IN_INT; + unsigned int ioff; + size_t moff; + + /* two step process: + * +-----------------------------+-----------------------+ + * | 32768 32769 32770 32771 ... | 32800 32801 32802 ... | + * +-----------------------------+-----------------------+ + * 1. ^^^^^ ^^^^^ ... + * 2. ffs: ^^^^^ ^^^^^ ^^^^^ ... + */ + + // do the rest of the integer + for (ioff = spoff % PORTS_IN_INT; ioff <= PORTS_IN_INT; ioff++) { + if (used_ports[smoff] & (1 << ioff)) { + used_ports[smoff] |= 1 << ioff; + return 32768 + spoff + ioff; + } + } + + // go one integer at a time + for (moff = smoff + 1; moff != smoff; moff++) { + if ((ioff = ffs(~used_ports[moff]))) { + used_ports[moff] |= 1 << (ioff - 1); + return 32768 + smoff * PORTS_IN_INT + (ioff - 1); + } + } + + return 0; +} + +static void free_port(unsigned int *used_ports, uint16_t port_num) { + used_ports[port_num / PORTS_IN_INT] ^= 1 << (port_num % PORTS_IN_INT); +} + +static void c_tm_cb(EV_P_ ev_timer *w, int revents __attribute__((unused))) { + struct o_c_sock *sock = w->data; + DBG("timing out socket @ %p", sock); + struct tcphdr buf = { + .th_sport = sock->l_port, + .th_dport = sock->r_port, + .th_seq = htonl(sock->seq_num), + .th_off = 5, + .th_flags = TH_FIN + }; + ssize_t sz = send(sock->rsock->c_data->s_sock, &buf, sizeof(buf), 0); + if (sz < 0) { + perror("send"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } else if ((size_t)sz != sizeof(buf)) { + fprintf(stderr, "send %s our packet: tried %lu, sent %zd\n", (size_t)sz > sizeof(buf) ? "expanded" : "truncated", sizeof(buf), sz); + } + free_port(sock->rsock->c_data->used_ports, sock->l_port); + ev_timer_stop(EV_A_ w); + + HASH_DELETE(hh_lp, sock->rsock->o_socks_by_lport, sock); + + if (!sock->rsock->o_socks_by_lport) { + close(sock->rsock->fd); + + ev_io_stop(EV_A_ &sock->rsock->io_w); + ev_timer_stop(EV_A_ &sock->rsock->tm_w); + + HASH_DEL(sock->rsock->c_data->o_rsocks, sock->rsock); + + free(sock->rsock->r_addr); + free(sock->rsock); + } + + free(sock); +} + +static void cc_cb(struct ev_loop *loop __attribute__((unused)), ev_io *w, int revents __attribute__((unused))) { + DBG("-- entering cc_cb --"); + + struct o_c_rsock *rsock = w->data; + char rbuf[65536]; + socklen_t pkt_addrlen = sizeof(struct sockaddr_in6); + ssize_t should_ssz, rsz, ssz; + + if ((rsz = recvfrom(w->fd, rbuf, sizeof(rbuf), 0, (struct sockaddr *)&rsock->c_data->pkt_addr, &pkt_addrlen)) == -1) { + perror("recvfrom"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + + DBG("received %zd raw bytes on client", rsz); + + if (pkt_addrlen > sizeof(struct sockaddr_in6)) + abort(); + + if ((size_t)rsz < sizeof(struct tcphdr)) + return; + + struct tcphdr *rhdr = (struct tcphdr *)rbuf; + + struct o_c_sock *sock; + + HASH_FIND(hh_lp, rsock->o_socks_by_lport, &rhdr->th_dport, sizeof(in_port_t), sock); + + if (!sock) { + DBG("could not find conn with lport %hu", ntohs(rhdr->th_dport)); + return; + } + + if (sock->status == TCP_SYN_SENT && rhdr->th_flags == (TH_SYN | TH_ACK)) { + DBG("SYN/ACK received, connection established"); + + sock->status = TCP_ESTABLISHED; + + struct tcphdr shdr = { + .th_sport = sock->l_port, + .th_dport = sock->r_port, + .th_seq = htonl(sock->seq_num), + .th_ack = rhdr->th_seq, + .th_win = 65535, + .th_flags = TH_ACK, + .th_off = 5 + }; + + sock->seq_num += sock->pending_data_size; + + struct iovec iovs[2] = { + { .iov_base = &shdr, .iov_len = sizeof(shdr) }, + { .iov_base = sock->pending_data, .iov_len = sock->pending_data_size } + }; + + struct msghdr msghdr = { + .msg_name = NULL, + .msg_namelen = 0, + .msg_iov = iovs, + .msg_iovlen = sizeof(iovs) / sizeof(iovs[0]) + }; + + should_ssz = sizeof(shdr) + sock->pending_data_size; + ssz = sendmsg(rsock->fd, &msghdr, 0); + + if (ssz < 0) { + perror("sendmsg"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } else if ((size_t)ssz != should_ssz) { + fprintf(stderr, "sendmsg %s our packet: tried %lu, sent %zd\n", (size_t)ssz > should_ssz ? "expanded" : "truncated", should_ssz, ssz); + } + + free(sock->pending_data); + } + + should_ssz = rsz - ntohs(rhdr->th_off) * 32 / CHAR_BIT; + if (should_ssz > 0) { + DBG("sending %zd bytes to client", should_ssz); + ssz = sendto(rsock->c_data->s_sock, rbuf + ntohs(rhdr->th_off) * 32 / CHAR_BIT, should_ssz, 0, sock->c_address, rsock->c_data->s_addrlen); + + if (ssz < 0) { + perror("sendto"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } else if ((size_t)ssz != should_ssz) { + fprintf(stderr, "sendto %s our packet: tried %lu, sent %zd\n", (size_t)ssz > should_ssz ? "expanded" : "truncated", should_ssz, ssz); + } + } +} + +static void cs_cb(EV_P_ ev_io *w, int revents __attribute__((unused))) { + DBG("-- entering cs_cb --"); + struct c_data *c_data = w->data; + socklen_t addresslen = c_data->s_addrlen; + ssize_t sz; + char rbuf[65536]; + + if ((sz = recvfrom(w->fd, rbuf, sizeof(rbuf), 0, (struct sockaddr *)&c_data->pkt_addr, &addresslen)) == -1) { + perror("recvfrom"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + + DBG("received %zd bytes on server", sz); + + if (addresslen != c_data->s_addrlen) + abort(); + + struct o_c_sock *sock; + HASH_FIND(hh_ca, c_data->o_socks_by_caddr, &c_data->pkt_addr, addresslen, sock); + + if (!sock) { + DBG("could not locate matching socket for client, initializing new connection"); + sock = calloc(1, sizeof(*sock)); + + uint16_t l_port = reserve_port(c_data->used_ports); + DBG("using port %hu", l_port); + if (!l_port) { + fputs("we ran out of ports?\n", stderr); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + sock->l_port = htons(l_port); + + struct addrinfo *res; + DBG("looking up %s:%s", c_data->r_host, c_data->r_port); + // TODO: make this asynchronous + int r = getaddrinfo(c_data->r_host, c_data->r_port, NULL, &res); + if (r) { + fprintf(stderr, "getaddrinfo: %s\n", gai_strerror(r)); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + + sock->c_address = malloc(addresslen); + memcpy(sock->c_address, &c_data->pkt_addr, addresslen); + + struct o_c_rsock *rsock; + + HASH_FIND(hh, c_data->o_rsocks, res->ai_addr, res->ai_addrlen, rsock); + + if (!rsock) { + DBG("could not locate remote socket to host, initializing new raw socket"); + rsock = malloc(sizeof(*rsock)); + rsock->r_addr = malloc(res->ai_addrlen); + + memcpy(rsock->r_addr, res->ai_addr, res->ai_addrlen); + rsock->r_addrlen = res->ai_addrlen; + freeaddrinfo(res); + rsock->o_socks_by_lport = NULL; + rsock->c_data = c_data; + + rsock->fd = socket(rsock->r_addr->sa_family, SOCK_RAW, IPPROTO_TCP); + if (!rsock->fd) { + perror("socket"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + + if (connect(rsock->fd, rsock->r_addr, rsock->r_addrlen) == -1) { + perror("connect"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + + ev_io_init(&rsock->io_w, cc_cb, rsock->fd, EV_READ); + rsock->io_w.data = rsock; + ev_io_start(EV_A_ &rsock->io_w); + + HASH_ADD_KEYPTR(hh, c_data->o_rsocks, rsock->r_addr, rsock->r_addrlen, rsock); + } + + sock->r_port = ((struct sockaddr_in *)rsock->r_addr)->sin_port; + + HASH_ADD_KEYPTR(hh_ca, c_data->o_socks_by_caddr, sock->c_address, addresslen, sock); + HASH_ADD(hh_lp, rsock->o_socks_by_lport, l_port, sizeof(in_port_t), sock); + + sock->rsock = rsock; + + sock->seq_num = random(); + + struct tcphdr buf = { + .th_sport = sock->l_port, + .th_dport = sock->r_port, + .th_seq = htonl(sock->seq_num++), + .th_flags = TH_SYN, + .th_off = 5 + }; + + sock->pending_data = malloc(sz); + memcpy(sock->pending_data, rbuf, sz); + sock->pending_data_size = sz; + + DBG("sending SYN to remote"); + sz = send(rsock->fd, &buf, sizeof(buf), 0); + if (sz < 0) { + perror("send"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } else if ((size_t)sz != sizeof(buf)) { + fprintf(stderr, "send %s our packet: tried %lu, sent %zd\n", (size_t)sz > sizeof(buf) ? "expanded" : "truncated", sizeof(buf), sz); + } + + // resend SYN + + //ev_timer_init(&sock->tm_w, c_tm_cb, 0., 60. * 60. * 3.); + //sock->tm_w.data = sock; + //ev_timer_start(EV_A_ &sock->tm_w); + + sock->status = TCP_SYN_SENT; + + return; + } + + struct tcphdr tcp_hdr = { + .th_sport = sock->l_port, + .th_dport = sock->r_port, + .th_seq = htonl(sock->seq_num), + .th_off = 5, + .th_win = 65535, + .th_flags = TH_PUSH + }; + + sock->seq_num += sz; + + struct iovec iovs[2] = { + { .iov_base = &tcp_hdr, .iov_len = sizeof(tcp_hdr) }, + { .iov_base = rbuf, .iov_len = sz } + }; + + struct msghdr msghdr = { + .msg_name = NULL, + .msg_namelen = 0, + .msg_iov = iovs, + .msg_iovlen = sizeof(iovs) / sizeof(iovs[0]) + }; + + size_t should_send_size = sizeof(tcp_hdr) + sz; + DBG("sending %zd raw bytes containing %zd bytes payload to remote", should_send_size, sz); + sz = sendmsg(sock->rsock->fd, &msghdr, 0); + if (sz < 0) { + perror("sendmsg"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } else if ((size_t)sz != should_send_size) { + fprintf(stderr, "sendmsg %s our packet: tried %lu, sent %zd\n", (size_t)sz > should_send_size ? "expanded" : "truncated", should_send_size, sz); + } + ev_timer_again(EV_A_ &sock->tm_w); +} + +static void c_cleanup() { + if (!global_c_data) + return; + + DBG("cleaning up"); + struct o_c_sock *sock; + for (sock = global_c_data->o_socks_by_caddr; sock != NULL; sock = sock->hh_ca.next) { + switch (sock->status) { + case TCP_ESTABLISHED: + // send TCP FIN + break; + case TCP_CLOSE: + break; + default: + ; + // send TCP RST + } + // don't bother freeing anything because we're about to exit anyways + } + + global_c_data = NULL; +} + +int start_client(const char *s_host, const char *s_port, const char *r_host, const char *r_port) { + struct addrinfo *res; + int r = getaddrinfo(s_host, s_port, NULL, &res); + if (r) { + fprintf(stderr, "getaddrinfo: %s\n", gai_strerror(r)); + return 3; + } + + struct c_data c_data = { 0 }; + + c_data.s_sock = socket(res->ai_family, SOCK_DGRAM, 0); + if (c_data.s_sock == -1) { + perror("socket"); + return 1; + } + + if (bind(c_data.s_sock, res->ai_addr, res->ai_addrlen) == -1) { + perror("bind"); + return 2; + } + + c_data.s_addrlen = res->ai_addrlen; + c_data.r_host = r_host; + c_data.r_port = r_port; + + freeaddrinfo(res); + + global_c_data = &c_data; + atexit(c_cleanup); + + struct ev_loop *loop = EV_DEFAULT; + ev_io s_watcher; + + s_watcher.data = &c_data; + + ev_io_init(&s_watcher, cs_cb, c_data.s_sock, EV_READ); + ev_io_start(loop, &s_watcher); + + DBG("initialization complete, starting event loop"); + r = ev_run(loop, 0); + + c_cleanup(); + return r; +} diff --git a/src/client.h b/src/client.h new file mode 100644 index 0000000..69d4c7d --- /dev/null +++ b/src/client.h @@ -0,0 +1 @@ +int start_client(const char *s_host, const char *s_port, const char *r_host, const char *r_port); diff --git a/src/common.h b/src/common.h new file mode 100644 index 0000000..f78b2ba --- /dev/null +++ b/src/common.h @@ -0,0 +1,5 @@ +#ifdef DEBUG +#define DBG(...) do { fprintf(stderr, __VA_ARGS__); putc('\n', stderr); } while (0) +#else +#define DBG(...) +#endif diff --git a/src/server.c b/src/server.c new file mode 100644 index 0000000..d5332bc --- /dev/null +++ b/src/server.c @@ -0,0 +1,327 @@ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "common.h" +#include "server.h" +#include "uthash.h" + +struct o_s_sock { + struct s_data *s_data; + struct sockaddr_storage c_addr; + struct ev_timer tm_w; + struct ev_io io_w; + UT_hash_handle hh; + int c_sock; + uint16_t seq_num; + uint8_t status; +}; + +struct s_data { + struct sockaddr *s_addr; + struct sockaddr_storage pkt_addr; + const char *r_host; + const char *r_port; + struct o_s_sock *o_socks_by_caddr; + int s_sock; + socklen_t s_addrlen; +}; + +static inline void s_prep_c_addr(struct o_s_sock *sock, struct tcphdr *hdr) { + memset(hdr, 0, sizeof(*hdr)); + hdr->th_sport = htons(((struct sockaddr_in *)sock->s_data->s_addr)->sin_port); + hdr->th_dport = htons(((struct sockaddr_in *)&sock->c_addr)->sin_port); + hdr->th_seq = htonl(sock->seq_num++); + hdr->th_off = 5; +} + +static void s_s_cleanup(EV_P_ struct o_s_sock *sock) { + DBG("cleaning up socket %p", sock); + + if (sock->status == TCP_ESTABLISHED) { + DBG("socket was ESTABLISHED, sending FIN"); + struct tcphdr buf; + s_prep_c_addr(sock, &buf); + buf.th_flags = TH_FIN; + ssize_t sz; + if ((sz = sendto(sock->s_data->s_sock, &buf, sizeof(buf), 0, (struct sockaddr *)&sock->s_data->pkt_addr, sock->s_data->s_addrlen)) == -1) { + perror("sendto"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } else if (sz != sizeof(buf)) { + fprintf(stderr, "sendto %s our packet: tried %lu, sent %zd\n", (size_t)sz > sizeof(buf) ? "expanded" : "truncated", sizeof(buf), sz); + } + } + + if (sock->c_sock != -1) { + close(sock->c_sock); + } + + ev_timer_stop(EV_A_ &sock->tm_w); + ev_io_stop(EV_A_ &sock->io_w); + + HASH_DEL(sock->s_data->o_socks_by_caddr, sock); + + free(sock); +} + +static void s_tm_cb(EV_P_ ev_timer *w, int revents __attribute__((unused))) { + DBG("timing out socket %p", w->data); + s_s_cleanup(EV_A_ w->data); +} + +static void sc_cb(EV_P_ ev_io *w, int revents __attribute__((unused))) { + struct o_s_sock *sock = w->data; + char rbuf[16384]; + ssize_t sz; + + DBG("-- entering sc_cb --"); + + if ((sz = recv(w->fd, rbuf, sizeof(rbuf), 0)) < 0) { + perror("recv"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + + DBG("received %zd bytes", sz); + + struct tcphdr hdr; + s_prep_c_addr(sock, &hdr); + hdr.th_off = 5; + + struct iovec iovs[2] = { + { .iov_base = &hdr, .iov_len = sizeof(hdr) }, + { .iov_base = rbuf, .iov_len = sz } + }; + + struct msghdr msghdr = { + .msg_name = &sock->c_addr, + .msg_namelen = sizeof(sock->c_addr), + .msg_iov = iovs, + .msg_iovlen = sizeof(iovs) / sizeof(iovs[0]) + }; + + size_t should_send_size = sizeof(hdr) + sz; + + assert(sock->status == TCP_ESTABLISHED); + + DBG("sending %zd bytes to client socket", should_send_size); + sz = sendmsg(sock->s_data->s_sock, &msghdr, 0); + if (sz < 0) { + perror("sendmsg"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } else if ((size_t)sz != should_send_size) { + fprintf(stderr, "sendmsg %s our packet: tried %lu, sent %zd\n", (size_t)sz > should_send_size ? "expanded" : "truncated", should_send_size, sz); + } + + ev_timer_again(EV_A_ &sock->tm_w); +} + +static void ss_cb(EV_P_ ev_io *w, int revents __attribute__((unused))) { + char rbuf[16384]; + ssize_t sz; + struct s_data *s_data = w->data; + socklen_t c_addrlen = s_data->s_addrlen; + int r; + + DBG("-- entering ss_cb --"); + + if ((sz = recvfrom(w->fd, rbuf, sizeof(rbuf), 0, (struct sockaddr *)&s_data->pkt_addr, &c_addrlen)) < 0) { + perror("recvfrom"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + + if (c_addrlen != s_data->s_addrlen) + abort(); + +#ifdef DEBUG + char hbuf[NI_MAXHOST]; + r = getnameinfo(&s_data->pkt_addr, c_addrlen, hbuf, sizeof(hbuf), NULL, 0, NI_NUMERICHOST); + if (r) { + fprintf(stderr, "getnameinfo: %s\n", gai_strerror(r)); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + DBG("received %zd bytes from %s", sz, hbuf); +#endif + + if ((size_t)sz < sizeof(struct tcphdr)) { + DBG("packet is smaller than TCP header, ignoring"); + return; + } + + struct tcphdr *tcphdr = (struct tcphdr *)rbuf; + + DBG("packet received on port %hu", ntohs(tcphdr->th_dport)); + + if (tcphdr->th_dport != ((struct sockaddr_in *)s_data->s_addr)->sin_port) { + DBG("packet should be on port %hu, ignoring", ntohs(((struct sockaddr_in *)s_data->s_addr)->sin_port)); + return; + } + + struct o_s_sock *sock; + + const uint8_t th_flags = tcphdr->th_flags; + + ((struct sockaddr_in *)&s_data->pkt_addr)->sin_port = tcphdr->th_sport; + + HASH_FIND(hh, s_data->o_socks_by_caddr, &s_data->pkt_addr, c_addrlen, sock); + + if (!sock) { + DBG("could not locate socket"); + + if (th_flags == TH_SYN) { + DBG("packet was SYN, initializing new connection"); + sock = malloc(sizeof(*sock)); + memcpy(&sock->c_addr, &s_data->pkt_addr, c_addrlen); + + sock->seq_num = random(); + sock->c_sock = -1; + sock->status = TCP_SYN_RECV; + + struct tcphdr buf = { + .th_sport = tcphdr->th_dport, + .th_dport = tcphdr->th_sport, + .th_seq = htonl(sock->seq_num), + .th_ack = tcphdr->th_seq, + .th_flags = TH_SYN | TH_ACK, + .th_off = 5 + }; + + HASH_ADD(hh, s_data->o_socks_by_caddr, c_addr, c_addrlen, sock); + + ((struct sockaddr_in *)&s_data->pkt_addr)->sin_port = htons(0); + + DBG("sending SYN/ACK"); + if ((sz = sendto(w->fd, &buf, sizeof(buf), 0, &s_data->pkt_addr, s_data->s_addrlen)) == -1) { + perror("sendto"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } else if (sz != sizeof(buf)) { + fprintf(stderr, "sendto %s our packet: tried %lu, sent %zd\n", (size_t)sz > sizeof(buf) ? "expanded" : "truncated", sizeof(buf), sz); + } + + ev_init(&sock->tm_w, s_tm_cb); + sock->tm_w.repeat = 10. * 60.; + sock->tm_w.data = sock; + ev_timer_again(EV_A_ &sock->tm_w); + } else { + DBG("packet was not SYN, ignoring"); + } + + return; + } + + if (tcphdr->th_off != 5) { + DBG("TCP options were specified, dropping packet"); + return; + } + + if (th_flags == TH_RST) { + DBG("RST received, cleaning up socket"); + sock->status = TCP_CLOSE; + s_s_cleanup(EV_A_ sock); + } + + if (th_flags & ~(TH_PUSH | TH_ACK)) { + DBG("TCP flags not PSH and/or ACK, dropping packet"); + return; + } + + if (sock->status == TCP_SYN_RECV) { + DBG("no UDP socket for this connection, shifting to ESTABLISHED"); + + assert(sock->c_sock == -1); + + sock->status = TCP_ESTABLISHED; + + struct addrinfo *res; + r = getaddrinfo(s_data->r_host, s_data->r_port, NULL, &res); + if (r) { + fprintf(stderr, "getaddrinfo: %s\n", gai_strerror(r)); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + + if ((sock->c_sock = socket(s_data->s_addr->sa_family, SOCK_DGRAM, 0)) == -1) { + perror("socket"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + + if (connect(sock->c_sock, res->ai_addr, res->ai_addrlen)) { + perror("connect"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } + + freeaddrinfo(res); + + ev_timer_stop(EV_A_ &sock->tm_w); + sock->tm_w.repeat = 60. * 60. * 3.; + ev_timer_start(EV_A_ &sock->tm_w); + + ev_io_init(&sock->io_w, sc_cb, sock->c_sock, EV_READ); + sock->io_w.data = sock; + ev_io_start(EV_A_ &sock->io_w); + } + + assert(sock->status == TCP_ESTABLISHED); + + DBG("sending %zu bytes to client socket", (size_t)(sz - tcphdr->th_off * 4)); + sz = send(sock->c_sock, rbuf + tcphdr->th_off * 4, sz - tcphdr->th_off * 4, 0); + if (sz < 0) { + // TODO: send TCP error? + perror("send"); + ev_break(EV_A_ EVBREAK_ONE); + return; + } +} + +int start_server(const char *s_host, const char *s_port, const char *r_host, const char *r_port) { + struct addrinfo *res; + int r = getaddrinfo(s_host, s_port, NULL, &res); + if (r) { + fprintf(stderr, "getaddrinfo: %s\n", gai_strerror(r)); + return 1; + } + + struct s_data s_data = { + .s_addr = res->ai_addr, + .s_addrlen = res->ai_addrlen, + .r_host = r_host, + .r_port = r_port + }; + + s_data.s_sock = socket(s_data.s_addr->sa_family, SOCK_RAW, IPPROTO_TCP); + if (s_data.s_sock == -1) { + perror("socket"); + freeaddrinfo(res); + return 1; + } + + struct ev_loop *loop = EV_DEFAULT; + ev_io s_watcher; + + ev_io_init(&s_watcher, ss_cb, s_data.s_sock, EV_READ); + ev_io_start(EV_A_ &s_watcher); + + s_watcher.data = &s_data; + + DBG("initialization complete, starting event loop"); + r = ev_run(loop, 0); + + freeaddrinfo(res); + + return r; +} diff --git a/src/server.h b/src/server.h new file mode 100644 index 0000000..3157173 --- /dev/null +++ b/src/server.h @@ -0,0 +1 @@ +int start_server(const char *s_addr, const char *s_port, const char *r_addr, const char *r_port); diff --git a/src/udpastcp.c b/src/udpastcp.c new file mode 100644 index 0000000..cdd73d2 --- /dev/null +++ b/src/udpastcp.c @@ -0,0 +1,25 @@ +#include +#include +#include +#include +#include +#include "server.h" +#include "client.h" + +int main(int argc, char *argv[]) { + if (argc < 6) { + puts("usage: udpintcp client|server LISTEN_HOST LISTEN_PORT REMOTE_HOST REMOTE_PORT"); + return !(argc == 2 && !strcmp(argv[1], "--help")); + } + + srandom((unsigned int)time(NULL)); + + if (!strcmp(argv[1], "client")) { + return start_client(argv[2], argv[3], argv[4], argv[5]) == 0; + } else if (!strcmp(argv[1], "server")) { + return start_server(argv[2], argv[3], argv[4], argv[5]) == 0; + } else { + fputs("invalid mode\n", stderr); + return 1; + } +} diff --git a/src/uthash.h b/src/uthash.h new file mode 100644 index 0000000..c30f9a7 --- /dev/null +++ b/src/uthash.h @@ -0,0 +1,966 @@ +/* +Copyright (c) 2003-2014, Troy D. Hanson http://troydhanson.github.com/uthash/ +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS +IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED +TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A +PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER +OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef UTHASH_H +#define UTHASH_H + +#include /* memcmp,strlen */ +#include /* ptrdiff_t */ +#include /* exit() */ + +/* These macros use decltype or the earlier __typeof GNU extension. + As decltype is only available in newer compilers (VS2010 or gcc 4.3+ + when compiling c++ source) this code uses whatever method is needed + or, for VS2008 where neither is available, uses casting workarounds. */ +#if defined(_MSC_VER) /* MS compiler */ +#if _MSC_VER >= 1600 && defined(__cplusplus) /* VS2010 or newer in C++ mode */ +#define DECLTYPE(x) (decltype(x)) +#else /* VS2008 or older (or VS2010 in C mode) */ +#define NO_DECLTYPE +#define DECLTYPE(x) +#endif +#elif defined(__BORLANDC__) || defined(__LCC__) || defined(__WATCOMC__) +#define NO_DECLTYPE +#define DECLTYPE(x) +#else /* GNU, Sun and other compilers */ +#define DECLTYPE(x) (__typeof(x)) +#endif + +#ifdef NO_DECLTYPE +#define DECLTYPE_ASSIGN(dst,src) \ +do { \ + char **_da_dst = (char**)(&(dst)); \ + *_da_dst = (char*)(src); \ +} while(0) +#else +#define DECLTYPE_ASSIGN(dst,src) \ +do { \ + (dst) = DECLTYPE(dst)(src); \ +} while(0) +#endif + +/* a number of the hash function use uint32_t which isn't defined on Pre VS2010 */ +#if defined(_WIN32) +#if defined(_MSC_VER) && _MSC_VER >= 1600 +#include +#elif defined(__WATCOMC__) || defined(__MINGW32__) || defined(__CYGWIN__) +#include +#else +typedef unsigned int uint32_t; +typedef unsigned char uint8_t; +#endif +#elif defined(__GNUC__) && !defined(__VXWORKS__) +#include +#else +typedef unsigned int uint32_t; +typedef unsigned char uint8_t; +#endif + +#define UTHASH_VERSION 1.9.9 + +#ifndef uthash_fatal +#define uthash_fatal(msg) exit(-1) /* fatal error (out of memory,etc) */ +#endif +#ifndef uthash_malloc +#define uthash_malloc(sz) malloc(sz) /* malloc fcn */ +#endif +#ifndef uthash_free +#define uthash_free(ptr,sz) free(ptr) /* free fcn */ +#endif + +#ifndef uthash_noexpand_fyi +#define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */ +#endif +#ifndef uthash_expand_fyi +#define uthash_expand_fyi(tbl) /* can be defined to log expands */ +#endif + +/* initial number of buckets */ +#define HASH_INITIAL_NUM_BUCKETS 32U /* initial number of buckets */ +#define HASH_INITIAL_NUM_BUCKETS_LOG2 5U /* lg2 of initial number of buckets */ +#define HASH_BKT_CAPACITY_THRESH 10U /* expand when bucket count reaches */ + +/* calculate the element whose hash handle address is hhe */ +#define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho))) + +#define HASH_FIND(hh,head,keyptr,keylen,out) \ +do { \ + out=NULL; \ + if (head != NULL) { \ + unsigned _hf_bkt,_hf_hashv; \ + HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \ + if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv) != 0) { \ + HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \ + keyptr,keylen,out); \ + } \ + } \ +} while (0) + +#ifdef HASH_BLOOM +#define HASH_BLOOM_BITLEN (1UL << HASH_BLOOM) +#define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8UL) + (((HASH_BLOOM_BITLEN%8UL)!=0UL) ? 1UL : 0UL) +#define HASH_BLOOM_MAKE(tbl) \ +do { \ + (tbl)->bloom_nbits = HASH_BLOOM; \ + (tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN); \ + if (!((tbl)->bloom_bv)) { uthash_fatal( "out of memory"); } \ + memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN); \ + (tbl)->bloom_sig = HASH_BLOOM_SIGNATURE; \ +} while (0) + +#define HASH_BLOOM_FREE(tbl) \ +do { \ + uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \ +} while (0) + +#define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8U] |= (1U << ((idx)%8U))) +#define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8U] & (1U << ((idx)%8U))) + +#define HASH_BLOOM_ADD(tbl,hashv) \ + HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1U))) + +#define HASH_BLOOM_TEST(tbl,hashv) \ + HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1U))) + +#else +#define HASH_BLOOM_MAKE(tbl) +#define HASH_BLOOM_FREE(tbl) +#define HASH_BLOOM_ADD(tbl,hashv) +#define HASH_BLOOM_TEST(tbl,hashv) (1) +#define HASH_BLOOM_BYTELEN 0U +#endif + +#define HASH_MAKE_TABLE(hh,head) \ +do { \ + (head)->hh.tbl = (UT_hash_table*)uthash_malloc( \ + sizeof(UT_hash_table)); \ + if (!((head)->hh.tbl)) { uthash_fatal( "out of memory"); } \ + memset((head)->hh.tbl, 0, sizeof(UT_hash_table)); \ + (head)->hh.tbl->tail = &((head)->hh); \ + (head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS; \ + (head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2; \ + (head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head); \ + (head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc( \ + HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \ + if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); } \ + memset((head)->hh.tbl->buckets, 0, \ + HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \ + HASH_BLOOM_MAKE((head)->hh.tbl); \ + (head)->hh.tbl->signature = HASH_SIGNATURE; \ +} while(0) + +#define HASH_ADD(hh,head,fieldname,keylen_in,add) \ + HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add) + +#define HASH_REPLACE(hh,head,fieldname,keylen_in,add,replaced) \ +do { \ + replaced=NULL; \ + HASH_FIND(hh,head,&((add)->fieldname),keylen_in,replaced); \ + if (replaced!=NULL) { \ + HASH_DELETE(hh,head,replaced); \ + } \ + HASH_ADD(hh,head,fieldname,keylen_in,add); \ +} while(0) + +#define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \ +do { \ + unsigned _ha_bkt; \ + (add)->hh.next = NULL; \ + (add)->hh.key = (char*)(keyptr); \ + (add)->hh.keylen = (unsigned)(keylen_in); \ + if (!(head)) { \ + head = (add); \ + (head)->hh.prev = NULL; \ + HASH_MAKE_TABLE(hh,head); \ + } else { \ + (head)->hh.tbl->tail->next = (add); \ + (add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \ + (head)->hh.tbl->tail = &((add)->hh); \ + } \ + (head)->hh.tbl->num_items++; \ + (add)->hh.tbl = (head)->hh.tbl; \ + HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \ + (add)->hh.hashv, _ha_bkt); \ + HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \ + HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \ + HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \ + HASH_FSCK(hh,head); \ +} while(0) + +#define HASH_TO_BKT( hashv, num_bkts, bkt ) \ +do { \ + bkt = ((hashv) & ((num_bkts) - 1U)); \ +} while(0) + +/* delete "delptr" from the hash table. + * "the usual" patch-up process for the app-order doubly-linked-list. + * The use of _hd_hh_del below deserves special explanation. + * These used to be expressed using (delptr) but that led to a bug + * if someone used the same symbol for the head and deletee, like + * HASH_DELETE(hh,users,users); + * We want that to work, but by changing the head (users) below + * we were forfeiting our ability to further refer to the deletee (users) + * in the patch-up process. Solution: use scratch space to + * copy the deletee pointer, then the latter references are via that + * scratch pointer rather than through the repointed (users) symbol. + */ +#define HASH_DELETE(hh,head,delptr) \ +do { \ + struct UT_hash_handle *_hd_hh_del; \ + if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \ + uthash_free((head)->hh.tbl->buckets, \ + (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ + HASH_BLOOM_FREE((head)->hh.tbl); \ + uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ + head = NULL; \ + } else { \ + unsigned _hd_bkt; \ + _hd_hh_del = &((delptr)->hh); \ + if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \ + (head)->hh.tbl->tail = \ + (UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ + (head)->hh.tbl->hho); \ + } \ + if ((delptr)->hh.prev != NULL) { \ + ((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ + (head)->hh.tbl->hho))->next = (delptr)->hh.next; \ + } else { \ + DECLTYPE_ASSIGN(head,(delptr)->hh.next); \ + } \ + if (_hd_hh_del->next != NULL) { \ + ((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \ + (head)->hh.tbl->hho))->prev = \ + _hd_hh_del->prev; \ + } \ + HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \ + HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \ + (head)->hh.tbl->num_items--; \ + } \ + HASH_FSCK(hh,head); \ +} while (0) + + +/* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */ +#define HASH_FIND_STR(head,findstr,out) \ + HASH_FIND(hh,head,findstr,(unsigned)strlen(findstr),out) +#define HASH_ADD_STR(head,strfield,add) \ + HASH_ADD(hh,head,strfield[0],(unsigned int)strlen(add->strfield),add) +#define HASH_REPLACE_STR(head,strfield,add,replaced) \ + HASH_REPLACE(hh,head,strfield[0],(unsigned)strlen(add->strfield),add,replaced) +#define HASH_FIND_INT(head,findint,out) \ + HASH_FIND(hh,head,findint,sizeof(int),out) +#define HASH_ADD_INT(head,intfield,add) \ + HASH_ADD(hh,head,intfield,sizeof(int),add) +#define HASH_REPLACE_INT(head,intfield,add,replaced) \ + HASH_REPLACE(hh,head,intfield,sizeof(int),add,replaced) +#define HASH_FIND_PTR(head,findptr,out) \ + HASH_FIND(hh,head,findptr,sizeof(void *),out) +#define HASH_ADD_PTR(head,ptrfield,add) \ + HASH_ADD(hh,head,ptrfield,sizeof(void *),add) +#define HASH_REPLACE_PTR(head,ptrfield,add,replaced) \ + HASH_REPLACE(hh,head,ptrfield,sizeof(void *),add,replaced) +#define HASH_DEL(head,delptr) \ + HASH_DELETE(hh,head,delptr) + +/* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined. + * This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined. + */ +#ifdef HASH_DEBUG +#define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0) +#define HASH_FSCK(hh,head) \ +do { \ + struct UT_hash_handle *_thh; \ + if (head) { \ + unsigned _bkt_i; \ + unsigned _count; \ + char *_prev; \ + _count = 0; \ + for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \ + unsigned _bkt_count = 0; \ + _thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \ + _prev = NULL; \ + while (_thh) { \ + if (_prev != (char*)(_thh->hh_prev)) { \ + HASH_OOPS("invalid hh_prev %p, actual %p\n", \ + _thh->hh_prev, _prev ); \ + } \ + _bkt_count++; \ + _prev = (char*)(_thh); \ + _thh = _thh->hh_next; \ + } \ + _count += _bkt_count; \ + if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \ + HASH_OOPS("invalid bucket count %u, actual %u\n", \ + (head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \ + } \ + } \ + if (_count != (head)->hh.tbl->num_items) { \ + HASH_OOPS("invalid hh item count %u, actual %u\n", \ + (head)->hh.tbl->num_items, _count ); \ + } \ + /* traverse hh in app order; check next/prev integrity, count */ \ + _count = 0; \ + _prev = NULL; \ + _thh = &(head)->hh; \ + while (_thh) { \ + _count++; \ + if (_prev !=(char*)(_thh->prev)) { \ + HASH_OOPS("invalid prev %p, actual %p\n", \ + _thh->prev, _prev ); \ + } \ + _prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \ + _thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \ + (head)->hh.tbl->hho) : NULL ); \ + } \ + if (_count != (head)->hh.tbl->num_items) { \ + HASH_OOPS("invalid app item count %u, actual %u\n", \ + (head)->hh.tbl->num_items, _count ); \ + } \ + } \ +} while (0) +#else +#define HASH_FSCK(hh,head) +#endif + +/* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to + * the descriptor to which this macro is defined for tuning the hash function. + * The app can #include to get the prototype for write(2). */ +#ifdef HASH_EMIT_KEYS +#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \ +do { \ + unsigned _klen = fieldlen; \ + write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \ + write(HASH_EMIT_KEYS, keyptr, (unsigned long)fieldlen); \ +} while (0) +#else +#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) +#endif + +/* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */ +#ifdef HASH_FUNCTION +#define HASH_FCN HASH_FUNCTION +#else +#define HASH_FCN HASH_JEN +#endif + +/* The Bernstein hash function, used in Perl prior to v5.6. Note (x<<5+x)=x*33. */ +#define HASH_BER(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _hb_keylen=(unsigned)keylen; \ + const unsigned char *_hb_key=(const unsigned char*)(key); \ + (hashv) = 0; \ + while (_hb_keylen-- != 0U) { \ + (hashv) = (((hashv) << 5) + (hashv)) + *_hb_key++; \ + } \ + bkt = (hashv) & (num_bkts-1U); \ +} while (0) + + +/* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at + * http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */ +#define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _sx_i; \ + const unsigned char *_hs_key=(const unsigned char*)(key); \ + hashv = 0; \ + for(_sx_i=0; _sx_i < keylen; _sx_i++) { \ + hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \ + } \ + bkt = hashv & (num_bkts-1U); \ +} while (0) +/* FNV-1a variation */ +#define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _fn_i; \ + const unsigned char *_hf_key=(const unsigned char*)(key); \ + hashv = 2166136261U; \ + for(_fn_i=0; _fn_i < keylen; _fn_i++) { \ + hashv = hashv ^ _hf_key[_fn_i]; \ + hashv = hashv * 16777619U; \ + } \ + bkt = hashv & (num_bkts-1U); \ +} while(0) + +#define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _ho_i; \ + const unsigned char *_ho_key=(const unsigned char*)(key); \ + hashv = 0; \ + for(_ho_i=0; _ho_i < keylen; _ho_i++) { \ + hashv += _ho_key[_ho_i]; \ + hashv += (hashv << 10); \ + hashv ^= (hashv >> 6); \ + } \ + hashv += (hashv << 3); \ + hashv ^= (hashv >> 11); \ + hashv += (hashv << 15); \ + bkt = hashv & (num_bkts-1U); \ +} while(0) + +#define HASH_JEN_MIX(a,b,c) \ +do { \ + a -= b; a -= c; a ^= ( c >> 13 ); \ + b -= c; b -= a; b ^= ( a << 8 ); \ + c -= a; c -= b; c ^= ( b >> 13 ); \ + a -= b; a -= c; a ^= ( c >> 12 ); \ + b -= c; b -= a; b ^= ( a << 16 ); \ + c -= a; c -= b; c ^= ( b >> 5 ); \ + a -= b; a -= c; a ^= ( c >> 3 ); \ + b -= c; b -= a; b ^= ( a << 10 ); \ + c -= a; c -= b; c ^= ( b >> 15 ); \ +} while (0) + +#define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _hj_i,_hj_j,_hj_k; \ + unsigned const char *_hj_key=(unsigned const char*)(key); \ + hashv = 0xfeedbeefu; \ + _hj_i = _hj_j = 0x9e3779b9u; \ + _hj_k = (unsigned)(keylen); \ + while (_hj_k >= 12U) { \ + _hj_i += (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 ) \ + + ( (unsigned)_hj_key[2] << 16 ) \ + + ( (unsigned)_hj_key[3] << 24 ) ); \ + _hj_j += (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 ) \ + + ( (unsigned)_hj_key[6] << 16 ) \ + + ( (unsigned)_hj_key[7] << 24 ) ); \ + hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 ) \ + + ( (unsigned)_hj_key[10] << 16 ) \ + + ( (unsigned)_hj_key[11] << 24 ) ); \ + \ + HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ + \ + _hj_key += 12; \ + _hj_k -= 12U; \ + } \ + hashv += (unsigned)(keylen); \ + switch ( _hj_k ) { \ + case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); /* FALLTHROUGH */ \ + case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); /* FALLTHROUGH */ \ + case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); /* FALLTHROUGH */ \ + case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); /* FALLTHROUGH */ \ + case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); /* FALLTHROUGH */ \ + case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); /* FALLTHROUGH */ \ + case 5: _hj_j += _hj_key[4]; /* FALLTHROUGH */ \ + case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); /* FALLTHROUGH */ \ + case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); /* FALLTHROUGH */ \ + case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); /* FALLTHROUGH */ \ + case 1: _hj_i += _hj_key[0]; \ + } \ + HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ + bkt = hashv & (num_bkts-1U); \ +} while(0) + +/* The Paul Hsieh hash function */ +#undef get16bits +#if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \ + || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__) +#define get16bits(d) (*((const uint16_t *) (d))) +#endif + +#if !defined (get16bits) +#define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \ + +(uint32_t)(((const uint8_t *)(d))[0]) ) +#endif +#define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned const char *_sfh_key=(unsigned const char*)(key); \ + uint32_t _sfh_tmp, _sfh_len = (uint32_t)keylen; \ + \ + unsigned _sfh_rem = _sfh_len & 3U; \ + _sfh_len >>= 2; \ + hashv = 0xcafebabeu; \ + \ + /* Main loop */ \ + for (;_sfh_len > 0U; _sfh_len--) { \ + hashv += get16bits (_sfh_key); \ + _sfh_tmp = ((uint32_t)(get16bits (_sfh_key+2)) << 11) ^ hashv; \ + hashv = (hashv << 16) ^ _sfh_tmp; \ + _sfh_key += 2U*sizeof (uint16_t); \ + hashv += hashv >> 11; \ + } \ + \ + /* Handle end cases */ \ + switch (_sfh_rem) { \ + case 3: hashv += get16bits (_sfh_key); \ + hashv ^= hashv << 16; \ + hashv ^= (uint32_t)(_sfh_key[sizeof (uint16_t)]) << 18; \ + hashv += hashv >> 11; \ + break; \ + case 2: hashv += get16bits (_sfh_key); \ + hashv ^= hashv << 11; \ + hashv += hashv >> 17; \ + break; \ + case 1: hashv += *_sfh_key; \ + hashv ^= hashv << 10; \ + hashv += hashv >> 1; \ + } \ + \ + /* Force "avalanching" of final 127 bits */ \ + hashv ^= hashv << 3; \ + hashv += hashv >> 5; \ + hashv ^= hashv << 4; \ + hashv += hashv >> 17; \ + hashv ^= hashv << 25; \ + hashv += hashv >> 6; \ + bkt = hashv & (num_bkts-1U); \ +} while(0) + +#ifdef HASH_USING_NO_STRICT_ALIASING +/* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads. + * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error. + * MurmurHash uses the faster approach only on CPU's where we know it's safe. + * + * Note the preprocessor built-in defines can be emitted using: + * + * gcc -m64 -dM -E - < /dev/null (on gcc) + * cc -## a.c (where a.c is a simple test file) (Sun Studio) + */ +#if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86)) +#define MUR_GETBLOCK(p,i) p[i] +#else /* non intel */ +#define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 3UL) == 0UL) +#define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 3UL) == 1UL) +#define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 3UL) == 2UL) +#define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 3UL) == 3UL) +#define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL)) +#if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__)) +#define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24)) +#define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16)) +#define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8)) +#else /* assume little endian non-intel */ +#define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24)) +#define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16)) +#define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8)) +#endif +#define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \ + (MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \ + (MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \ + MUR_ONE_THREE(p)))) +#endif +#define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r)))) +#define MUR_FMIX(_h) \ +do { \ + _h ^= _h >> 16; \ + _h *= 0x85ebca6bu; \ + _h ^= _h >> 13; \ + _h *= 0xc2b2ae35u; \ + _h ^= _h >> 16; \ +} while(0) + +#define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \ +do { \ + const uint8_t *_mur_data = (const uint8_t*)(key); \ + const int _mur_nblocks = (int)(keylen) / 4; \ + uint32_t _mur_h1 = 0xf88D5353u; \ + uint32_t _mur_c1 = 0xcc9e2d51u; \ + uint32_t _mur_c2 = 0x1b873593u; \ + uint32_t _mur_k1 = 0; \ + const uint8_t *_mur_tail; \ + const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+(_mur_nblocks*4)); \ + int _mur_i; \ + for(_mur_i = -_mur_nblocks; _mur_i!=0; _mur_i++) { \ + _mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \ + _mur_k1 *= _mur_c1; \ + _mur_k1 = MUR_ROTL32(_mur_k1,15); \ + _mur_k1 *= _mur_c2; \ + \ + _mur_h1 ^= _mur_k1; \ + _mur_h1 = MUR_ROTL32(_mur_h1,13); \ + _mur_h1 = (_mur_h1*5U) + 0xe6546b64u; \ + } \ + _mur_tail = (const uint8_t*)(_mur_data + (_mur_nblocks*4)); \ + _mur_k1=0; \ + switch((keylen) & 3U) { \ + case 3: _mur_k1 ^= (uint32_t)_mur_tail[2] << 16; /* FALLTHROUGH */ \ + case 2: _mur_k1 ^= (uint32_t)_mur_tail[1] << 8; /* FALLTHROUGH */ \ + case 1: _mur_k1 ^= (uint32_t)_mur_tail[0]; \ + _mur_k1 *= _mur_c1; \ + _mur_k1 = MUR_ROTL32(_mur_k1,15); \ + _mur_k1 *= _mur_c2; \ + _mur_h1 ^= _mur_k1; \ + } \ + _mur_h1 ^= (uint32_t)(keylen); \ + MUR_FMIX(_mur_h1); \ + hashv = _mur_h1; \ + bkt = hashv & (num_bkts-1U); \ +} while(0) +#endif /* HASH_USING_NO_STRICT_ALIASING */ + +/* key comparison function; return 0 if keys equal */ +#define HASH_KEYCMP(a,b,len) memcmp(a,b,(unsigned long)(len)) + +/* iterate over items in a known bucket to find desired item */ +#define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \ +do { \ + if (head.hh_head != NULL) { DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); } \ + else { out=NULL; } \ + while (out != NULL) { \ + if ((out)->hh.keylen == (keylen_in)) { \ + if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) { break; } \ + } \ + if ((out)->hh.hh_next != NULL) { DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); } \ + else { out = NULL; } \ + } \ +} while(0) + +/* add an item to a bucket */ +#define HASH_ADD_TO_BKT(head,addhh) \ +do { \ + head.count++; \ + (addhh)->hh_next = head.hh_head; \ + (addhh)->hh_prev = NULL; \ + if (head.hh_head != NULL) { (head).hh_head->hh_prev = (addhh); } \ + (head).hh_head=addhh; \ + if ((head.count >= ((head.expand_mult+1U) * HASH_BKT_CAPACITY_THRESH)) \ + && ((addhh)->tbl->noexpand != 1U)) { \ + HASH_EXPAND_BUCKETS((addhh)->tbl); \ + } \ +} while(0) + +/* remove an item from a given bucket */ +#define HASH_DEL_IN_BKT(hh,head,hh_del) \ + (head).count--; \ + if ((head).hh_head == hh_del) { \ + (head).hh_head = hh_del->hh_next; \ + } \ + if (hh_del->hh_prev) { \ + hh_del->hh_prev->hh_next = hh_del->hh_next; \ + } \ + if (hh_del->hh_next) { \ + hh_del->hh_next->hh_prev = hh_del->hh_prev; \ + } + +/* Bucket expansion has the effect of doubling the number of buckets + * and redistributing the items into the new buckets. Ideally the + * items will distribute more or less evenly into the new buckets + * (the extent to which this is true is a measure of the quality of + * the hash function as it applies to the key domain). + * + * With the items distributed into more buckets, the chain length + * (item count) in each bucket is reduced. Thus by expanding buckets + * the hash keeps a bound on the chain length. This bounded chain + * length is the essence of how a hash provides constant time lookup. + * + * The calculation of tbl->ideal_chain_maxlen below deserves some + * explanation. First, keep in mind that we're calculating the ideal + * maximum chain length based on the *new* (doubled) bucket count. + * In fractions this is just n/b (n=number of items,b=new num buckets). + * Since the ideal chain length is an integer, we want to calculate + * ceil(n/b). We don't depend on floating point arithmetic in this + * hash, so to calculate ceil(n/b) with integers we could write + * + * ceil(n/b) = (n/b) + ((n%b)?1:0) + * + * and in fact a previous version of this hash did just that. + * But now we have improved things a bit by recognizing that b is + * always a power of two. We keep its base 2 log handy (call it lb), + * so now we can write this with a bit shift and logical AND: + * + * ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0) + * + */ +#define HASH_EXPAND_BUCKETS(tbl) \ +do { \ + unsigned _he_bkt; \ + unsigned _he_bkt_i; \ + struct UT_hash_handle *_he_thh, *_he_hh_nxt; \ + UT_hash_bucket *_he_new_buckets, *_he_newbkt; \ + _he_new_buckets = (UT_hash_bucket*)uthash_malloc( \ + 2UL * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ + if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \ + memset(_he_new_buckets, 0, \ + 2UL * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ + tbl->ideal_chain_maxlen = \ + (tbl->num_items >> (tbl->log2_num_buckets+1U)) + \ + (((tbl->num_items & ((tbl->num_buckets*2U)-1U)) != 0U) ? 1U : 0U); \ + tbl->nonideal_items = 0; \ + for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \ + { \ + _he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \ + while (_he_thh != NULL) { \ + _he_hh_nxt = _he_thh->hh_next; \ + HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2U, _he_bkt); \ + _he_newbkt = &(_he_new_buckets[ _he_bkt ]); \ + if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \ + tbl->nonideal_items++; \ + _he_newbkt->expand_mult = _he_newbkt->count / \ + tbl->ideal_chain_maxlen; \ + } \ + _he_thh->hh_prev = NULL; \ + _he_thh->hh_next = _he_newbkt->hh_head; \ + if (_he_newbkt->hh_head != NULL) { _he_newbkt->hh_head->hh_prev = \ + _he_thh; } \ + _he_newbkt->hh_head = _he_thh; \ + _he_thh = _he_hh_nxt; \ + } \ + } \ + uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ + tbl->num_buckets *= 2U; \ + tbl->log2_num_buckets++; \ + tbl->buckets = _he_new_buckets; \ + tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \ + (tbl->ineff_expands+1U) : 0U; \ + if (tbl->ineff_expands > 1U) { \ + tbl->noexpand=1; \ + uthash_noexpand_fyi(tbl); \ + } \ + uthash_expand_fyi(tbl); \ +} while(0) + + +/* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */ +/* Note that HASH_SORT assumes the hash handle name to be hh. + * HASH_SRT was added to allow the hash handle name to be passed in. */ +#define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn) +#define HASH_SRT(hh,head,cmpfcn) \ +do { \ + unsigned _hs_i; \ + unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \ + struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \ + if (head != NULL) { \ + _hs_insize = 1; \ + _hs_looping = 1; \ + _hs_list = &((head)->hh); \ + while (_hs_looping != 0U) { \ + _hs_p = _hs_list; \ + _hs_list = NULL; \ + _hs_tail = NULL; \ + _hs_nmerges = 0; \ + while (_hs_p != NULL) { \ + _hs_nmerges++; \ + _hs_q = _hs_p; \ + _hs_psize = 0; \ + for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \ + _hs_psize++; \ + _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ? \ + ((void*)((char*)(_hs_q->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + if (! (_hs_q) ) { break; } \ + } \ + _hs_qsize = _hs_insize; \ + while ((_hs_psize > 0U) || ((_hs_qsize > 0U) && (_hs_q != NULL))) {\ + if (_hs_psize == 0U) { \ + _hs_e = _hs_q; \ + _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ? \ + ((void*)((char*)(_hs_q->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + _hs_qsize--; \ + } else if ( (_hs_qsize == 0U) || (_hs_q == NULL) ) { \ + _hs_e = _hs_p; \ + if (_hs_p != NULL){ \ + _hs_p = (UT_hash_handle*)((_hs_p->next != NULL) ? \ + ((void*)((char*)(_hs_p->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + } \ + _hs_psize--; \ + } else if (( \ + cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \ + DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \ + ) <= 0) { \ + _hs_e = _hs_p; \ + if (_hs_p != NULL){ \ + _hs_p = (UT_hash_handle*)((_hs_p->next != NULL) ? \ + ((void*)((char*)(_hs_p->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + } \ + _hs_psize--; \ + } else { \ + _hs_e = _hs_q; \ + _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ? \ + ((void*)((char*)(_hs_q->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + _hs_qsize--; \ + } \ + if ( _hs_tail != NULL ) { \ + _hs_tail->next = ((_hs_e != NULL) ? \ + ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \ + } else { \ + _hs_list = _hs_e; \ + } \ + if (_hs_e != NULL) { \ + _hs_e->prev = ((_hs_tail != NULL) ? \ + ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \ + } \ + _hs_tail = _hs_e; \ + } \ + _hs_p = _hs_q; \ + } \ + if (_hs_tail != NULL){ \ + _hs_tail->next = NULL; \ + } \ + if ( _hs_nmerges <= 1U ) { \ + _hs_looping=0; \ + (head)->hh.tbl->tail = _hs_tail; \ + DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \ + } \ + _hs_insize *= 2U; \ + } \ + HASH_FSCK(hh,head); \ + } \ +} while (0) + +/* This function selects items from one hash into another hash. + * The end result is that the selected items have dual presence + * in both hashes. There is no copy of the items made; rather + * they are added into the new hash through a secondary hash + * hash handle that must be present in the structure. */ +#define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \ +do { \ + unsigned _src_bkt, _dst_bkt; \ + void *_last_elt=NULL, *_elt; \ + UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \ + ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \ + if (src != NULL) { \ + for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \ + for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \ + _src_hh != NULL; \ + _src_hh = _src_hh->hh_next) { \ + _elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \ + if (cond(_elt)) { \ + _dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \ + _dst_hh->key = _src_hh->key; \ + _dst_hh->keylen = _src_hh->keylen; \ + _dst_hh->hashv = _src_hh->hashv; \ + _dst_hh->prev = _last_elt; \ + _dst_hh->next = NULL; \ + if (_last_elt_hh != NULL) { _last_elt_hh->next = _elt; } \ + if (dst == NULL) { \ + DECLTYPE_ASSIGN(dst,_elt); \ + HASH_MAKE_TABLE(hh_dst,dst); \ + } else { \ + _dst_hh->tbl = (dst)->hh_dst.tbl; \ + } \ + HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \ + HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \ + (dst)->hh_dst.tbl->num_items++; \ + _last_elt = _elt; \ + _last_elt_hh = _dst_hh; \ + } \ + } \ + } \ + } \ + HASH_FSCK(hh_dst,dst); \ +} while (0) + +#define HASH_CLEAR(hh,head) \ +do { \ + if (head != NULL) { \ + uthash_free((head)->hh.tbl->buckets, \ + (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \ + HASH_BLOOM_FREE((head)->hh.tbl); \ + uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ + (head)=NULL; \ + } \ +} while(0) + +#define HASH_OVERHEAD(hh,head) \ + ((head != NULL) ? ( \ + (size_t)(((head)->hh.tbl->num_items * sizeof(UT_hash_handle)) + \ + ((head)->hh.tbl->num_buckets * sizeof(UT_hash_bucket)) + \ + sizeof(UT_hash_table) + \ + (HASH_BLOOM_BYTELEN))) : 0U) + +#ifdef NO_DECLTYPE +#define HASH_ITER(hh,head,el,tmp) \ +for(((el)=(head)), ((*(char**)(&(tmp)))=(char*)((head!=NULL)?(head)->hh.next:NULL)); \ + (el) != NULL; ((el)=(tmp)), ((*(char**)(&(tmp)))=(char*)((tmp!=NULL)?(tmp)->hh.next:NULL))) +#else +#define HASH_ITER(hh,head,el,tmp) \ +for(((el)=(head)), ((tmp)=DECLTYPE(el)((head!=NULL)?(head)->hh.next:NULL)); \ + (el) != NULL; ((el)=(tmp)), ((tmp)=DECLTYPE(el)((tmp!=NULL)?(tmp)->hh.next:NULL))) +#endif + +/* obtain a count of items in the hash */ +#define HASH_COUNT(head) HASH_CNT(hh,head) +#define HASH_CNT(hh,head) ((head != NULL)?((head)->hh.tbl->num_items):0U) + +typedef struct UT_hash_bucket { + struct UT_hash_handle *hh_head; + unsigned count; + + /* expand_mult is normally set to 0. In this situation, the max chain length + * threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If + * the bucket's chain exceeds this length, bucket expansion is triggered). + * However, setting expand_mult to a non-zero value delays bucket expansion + * (that would be triggered by additions to this particular bucket) + * until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH. + * (The multiplier is simply expand_mult+1). The whole idea of this + * multiplier is to reduce bucket expansions, since they are expensive, in + * situations where we know that a particular bucket tends to be overused. + * It is better to let its chain length grow to a longer yet-still-bounded + * value, than to do an O(n) bucket expansion too often. + */ + unsigned expand_mult; + +} UT_hash_bucket; + +/* random signature used only to find hash tables in external analysis */ +#define HASH_SIGNATURE 0xa0111fe1u +#define HASH_BLOOM_SIGNATURE 0xb12220f2u + +typedef struct UT_hash_table { + UT_hash_bucket *buckets; + struct UT_hash_handle *tail; /* tail hh in app order, for fast append */ + unsigned num_buckets, log2_num_buckets; + unsigned num_items; + ptrdiff_t hho; /* hash handle offset (byte pos of hash handle in element */ + + /* in an ideal situation (all buckets used equally), no bucket would have + * more than ceil(#items/#buckets) items. that's the ideal chain length. */ + unsigned ideal_chain_maxlen; + + /* nonideal_items is the number of items in the hash whose chain position + * exceeds the ideal chain maxlen. these items pay the penalty for an uneven + * hash distribution; reaching them in a chain traversal takes >ideal steps */ + unsigned nonideal_items; + + /* ineffective expands occur when a bucket doubling was performed, but + * afterward, more than half the items in the hash had nonideal chain + * positions. If this happens on two consecutive expansions we inhibit any + * further expansion, as it's not helping; this happens when the hash + * function isn't a good fit for the key domain. When expansion is inhibited + * the hash will still work, albeit no longer in constant time. */ + unsigned ineff_expands, noexpand; + + uint32_t signature; /* used only to find hash tables in external analysis */ +#ifdef HASH_BLOOM + uint32_t bloom_sig; /* used only to test bloom exists in external analysis */ + uint8_t *bloom_bv; + uint8_t bloom_nbits; +#endif + +} UT_hash_table; + +typedef struct UT_hash_handle { + struct UT_hash_table *tbl; + void *prev; /* prev element in app order */ + void *next; /* next element in app order */ + struct UT_hash_handle *hh_prev; /* previous hh in bucket order */ + struct UT_hash_handle *hh_next; /* next hh in bucket order */ + void *key; /* ptr to enclosing struct's key */ + unsigned keylen; /* enclosing struct's key len */ + unsigned hashv; /* result of hash-fcn(key) */ +} UT_hash_handle; + +#endif /* UTHASH_H */ -- cgit v1.2.3-54-g00ecf