+#include <stdio.h>
+#include <string.h>
+#include <stdint.h>
+#include <assert.h>
+#include <arpa/inet.h>
+#include <sys/socket.h>
+#include <pthread.h>
+#include <sys/types.h>
+#include <sys/ioctl.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <netdb.h>
+#include <poll.h>
+#include <signal.h>
+#include <errno.h>
+#include <vector>
+#include <string>
+#include <map>
+
+#include "metacube.h"
+#include "mutexlock.h"
+#include "httpinput.h"
+#include "server.h"
+#include "serverpool.h"
+#include "parse.h"
+#include "state.pb.h"
+
+using namespace std;
+
+extern ServerPool *servers;
+
+HTTPInput::HTTPInput(const string &stream_id, const string &url)
+ : state(NOT_CONNECTED),
+ stream_id(stream_id),
+ url(url),
+ has_metacube_header(false),
+ sock(-1)
+{
+}
+
+HTTPInput::HTTPInput(const InputProto &serialized)
+ : state(State(serialized.state())),
+ stream_id(serialized.stream_id()),
+ url(serialized.url()),
+ request(serialized.request()),
+ request_bytes_sent(serialized.request_bytes_sent()),
+ response(serialized.response()),
+ http_header(serialized.http_header()),
+ has_metacube_header(serialized.has_metacube_header()),
+ sock(serialized.sock())
+{
+ pending_data.resize(serialized.pending_data().size());
+ memcpy(&pending_data[0], serialized.pending_data().data(), serialized.pending_data().size());
+
+ string protocol;
+ parse_url(url, &protocol, &host, &port, &path); // Don't care if it fails.
+}
+
+InputProto HTTPInput::serialize() const
+{
+ InputProto serialized;
+ serialized.set_state(state);
+ serialized.set_stream_id(stream_id);
+ serialized.set_url(url);
+ serialized.set_request(request);
+ serialized.set_request_bytes_sent(request_bytes_sent);
+ serialized.set_response(response);
+ serialized.set_http_header(http_header);
+ serialized.set_pending_data(string(pending_data.begin(), pending_data.end()));
+ serialized.set_has_metacube_header(has_metacube_header);
+ serialized.set_sock(sock);
+ return serialized;
+}
+
+int HTTPInput::lookup_and_connect(const string &host, const string &port)
+{
+ addrinfo *ai;
+ int err = getaddrinfo(host.c_str(), port.c_str(), NULL, &ai);
+ if (err == -1) {
+ fprintf(stderr, "WARNING: Lookup of '%s' failed (%s).\n",
+ host.c_str(), gai_strerror(err));
+ freeaddrinfo(ai);
+ return -1;
+ }
+
+ // Connect to everything in turn until we have a socket.
+ while (ai && !should_stop) {
+ int sock = socket(ai->ai_family, SOCK_STREAM, IPPROTO_TCP);
+ if (sock == -1) {
+ // Could be e.g. EPROTONOSUPPORT. The show must go on.
+ continue;
+ }
+
+ do {
+ err = connect(sock, ai->ai_addr, ai->ai_addrlen);
+ } while (err == -1 && errno == EINTR);
+
+ if (err != -1) {
+ freeaddrinfo(ai);
+ return sock;
+ }
+
+ ai = ai->ai_next;
+ }
+
+ // Give the last one as error.
+ fprintf(stderr, "WARNING: Connect to '%s' failed (%s)\n",
+ host.c_str(), strerror(errno));
+ freeaddrinfo(ai);
+ return -1;
+}
+
+bool HTTPInput::parse_response(const std::string &request)
+{
+ vector<string> lines = split_lines(response);
+ if (lines.empty()) {
+ fprintf(stderr, "WARNING: Empty HTTP response from input.\n");
+ return false;
+ }
+
+ vector<string> first_line_tokens = split_tokens(lines[0]);
+ if (first_line_tokens.size() < 2) {
+ fprintf(stderr, "WARNING: Malformed response line '%s' from input.\n",
+ lines[0].c_str());
+ return false;
+ }
+
+ int response = atoi(first_line_tokens[1].c_str());
+ if (response != 200) {
+ fprintf(stderr, "WARNING: Non-200 response '%s' from input.\n",
+ lines[0].c_str());
+ return false;
+ }
+
+ multimap<string, string> parameters;
+ for (size_t i = 1; i < lines.size(); ++i) {
+ size_t split = lines[i].find(":");
+ if (split == string::npos) {
+ fprintf(stderr, "WARNING: Ignoring malformed HTTP response line '%s'\n",
+ lines[i].c_str());
+ continue;
+ }
+
+ string key(lines[i].begin(), lines[i].begin() + split);
+
+ // Skip any spaces after the colon.
+ do {
+ ++split;
+ } while (split < lines[i].size() && lines[i][split] == ' ');
+
+ string value(lines[i].begin() + split, lines[i].end());
+
+ // Remove “Content-encoding: metacube”.
+ // TODO: Make case-insensitive.
+ if (key == "Content-encoding" && value == "metacube") {
+ continue;
+ }
+
+ parameters.insert(make_pair(key, value));
+ }
+
+ // Change “Server: foo” to “Server: metacube/0.1 (reflecting: foo)”
+ // TODO: Make case-insensitive.
+ // XXX: Use a Via: instead?
+ if (parameters.count("Server") == 0) {
+ parameters.insert(make_pair("Server", "metacube/0.1"));
+ } else {
+ for (multimap<string, string>::iterator it = parameters.begin();
+ it != parameters.end();
+ ++it) {
+ if (it->first != "Server") {
+ continue;
+ }
+ it->second = "metacube/0.1 (reflecting: " + it->second + ")";
+ }
+ }
+
+ // Construct the new HTTP header.
+ http_header = "HTTP/1.0 200 OK\r\n";
+ for (multimap<string, string>::iterator it = parameters.begin();
+ it != parameters.end();
+ ++it) {
+ http_header.append(it->first + ": " + it->second + "\r\n");
+ }
+ http_header.append("\r\n");
+ servers->set_header(stream_id, http_header);
+
+ return true;
+}
+
+void HTTPInput::do_work()
+{
+ while (!should_stop) {
+ if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
+ // Since we are non-blocking, we need to wait for the right state first.
+ // Wait up to 50 ms, then check should_stop.
+ pollfd pfd;
+ pfd.fd = sock;
+ pfd.events = (state == SENDING_REQUEST) ? POLLOUT : POLLIN;
+ pfd.events |= POLLRDHUP;
+
+ int nfds = poll(&pfd, 1, 50);
+ if (nfds == 0 || (nfds == -1 && errno == EINTR)) {
+ continue;
+ }
+ if (nfds == -1) {
+ perror("poll");
+ state = CLOSING_SOCKET;
+ }
+ }
+
+ switch (state) {
+ case NOT_CONNECTED:
+ request.clear();
+ request_bytes_sent = 0;
+ response.clear();
+
+ {
+ string protocol; // Thrown away.
+ if (!parse_url(url, &protocol, &host, &port, &path)) {
+ fprintf(stderr, "Failed to parse URL '%s'\n", url.c_str());
+ break;
+ }
+ }
+
+ sock = lookup_and_connect(host, port);
+ if (sock != -1) {
+ // Yay, successful connect. Try to set it as nonblocking.
+ int one = 1;
+ if (ioctl(sock, FIONBIO, &one) == -1) {
+ perror("ioctl(FIONBIO)");
+ state = CLOSING_SOCKET;
+ } else {
+ state = SENDING_REQUEST;
+ request = "GET " + path + " HTTP/1.0\r\nUser-Agent: cubemap\r\n\r\n";
+ request_bytes_sent = 0;
+ }
+ }
+ break;
+ case SENDING_REQUEST: {
+ size_t to_send = request.size() - request_bytes_sent;
+ int ret;
+
+ do {
+ ret = write(sock, request.data() + request_bytes_sent, to_send);
+ } while (ret == -1 && errno == EINTR);
+
+ if (ret == -1) {
+ perror("write");
+ state = CLOSING_SOCKET;
+ continue;
+ }
+
+ assert(ret >= 0);
+ request_bytes_sent += ret;
+
+ if (request_bytes_sent == request.size()) {
+ state = RECEIVING_HEADER;
+ }
+ break;
+ }
+ case RECEIVING_HEADER: {
+ char buf[4096];
+ int ret;
+
+ do {
+ ret = read(sock, buf, sizeof(buf));
+ } while (ret == -1 && errno == EINTR);
+
+ if (ret == -1) {
+ perror("read");
+ state = CLOSING_SOCKET;
+ continue;
+ }
+
+ if (ret == 0) {
+ // This really shouldn't happen...
+ fprintf(stderr, "Socket unexpectedly closed while reading header\n");
+ state = CLOSING_SOCKET;
+ continue;
+ }
+
+ RequestParseStatus status = wait_for_double_newline(&response, buf, ret);
+
+ if (status == RP_OUT_OF_SPACE) {
+ fprintf(stderr, "WARNING: fd %d sent overlong response!\n", sock);
+ state = CLOSING_SOCKET;
+ continue;
+ } else if (status == RP_NOT_FINISHED_YET) {
+ continue;
+ }
+
+ // OK, so we're fine, but there might be some of the actual data after the response.
+ // We'll need to deal with that separately.
+ string extra_data;
+ if (status == RP_EXTRA_DATA) {
+ char *ptr = static_cast<char *>(
+ memmem(response.data(), response.size(), "\r\n\r\n", 4));
+ assert(ptr != NULL);
+ extra_data = string(ptr, &response[0] + response.size());
+ response.resize(ptr - response.data());
+ }
+
+ if (!parse_response(response)) {
+ state = CLOSING_SOCKET;
+ continue;
+ }
+
+ if (!extra_data.empty()) {
+ process_data(&extra_data[0], extra_data.size());
+ }
+
+ state = RECEIVING_DATA;
+ break;
+ }
+ case RECEIVING_DATA: {
+ char buf[4096];
+ int ret;
+
+ do {
+ ret = read(sock, buf, sizeof(buf));
+ } while (ret == -1 && errno == EINTR);
+
+ if (ret == -1) {
+ perror("read");
+ state = CLOSING_SOCKET;
+ continue;
+ }
+
+ if (ret == 0) {
+ // This really shouldn't happen...
+ fprintf(stderr, "Socket unexpectedly closed while reading header\n");
+ state = CLOSING_SOCKET;
+ continue;
+ }
+
+ process_data(buf, ret);
+ break;
+ }
+ case CLOSING_SOCKET: {
+ int err;
+ do {
+ err = close(sock);
+ } while (err == -1 && errno == EINTR);
+
+ if (err == -1) {
+ perror("close");
+ }
+
+ state = NOT_CONNECTED;
+ break;
+ }
+ default:
+ assert(false);
+ }
+
+ // If we are still in NOT_CONNECTED, either something went wrong,
+ // or the connection just got closed.
+ // The earlier steps have already given the error message, if any.
+ if (state == NOT_CONNECTED && !should_stop) {
+ fprintf(stderr, "Waiting 0.2 second and restarting...\n");
+ usleep(200000);
+ }
+ }
+}
+
+void HTTPInput::process_data(char *ptr, size_t bytes)
+{
+ pending_data.insert(pending_data.end(), ptr, ptr + bytes);
+
+ for ( ;; ) {
+ // If we don't have enough data (yet) for even the Metacube header, just return.
+ if (pending_data.size() < sizeof(metacube_block_header)) {
+ return;
+ }
+
+ // Make sure we have the Metacube sync header at the start.
+ // We may need to skip over junk data (it _should_ not happen, though).
+ if (!has_metacube_header) {
+ char *ptr = static_cast<char *>(
+ memmem(pending_data.data(), pending_data.size(),
+ METACUBE_SYNC, strlen(METACUBE_SYNC)));
+ if (ptr == NULL) {
+ // OK, so we didn't find the sync marker. We know then that
+ // we do not have the _full_ marker in the buffer, but we
+ // could have N-1 bytes. Drop everything before that,
+ // and then give up.
+ drop_pending_data(pending_data.size() - (strlen(METACUBE_SYNC) - 1));
+ return;
+ } else {
+ // Yay, we found the header. Drop everything (if anything) before it.
+ drop_pending_data(ptr - pending_data.data());
+ has_metacube_header = true;
+
+ // Re-check that we have the entire header; we could have dropped data.
+ if (pending_data.size() < sizeof(metacube_block_header)) {
+ return;
+ }
+ }
+ }
+
+ // Now it's safe to read the header.
+ metacube_block_header *hdr = reinterpret_cast<metacube_block_header *>(pending_data.data());
+ assert(memcmp(hdr->sync, METACUBE_SYNC, sizeof(hdr->sync)) == 0);
+ uint32_t size = ntohl(hdr->size);
+ uint32_t flags = ntohl(hdr->flags);
+
+ // See if we have the entire block. If not, wait for more data.
+ if (pending_data.size() < sizeof(metacube_block_header) + size) {
+ return;
+ }
+
+ // Send this block on to the data.
+ char *inner_data = pending_data.data() + sizeof(metacube_block_header);
+ if (flags & METACUBE_FLAGS_HEADER) {
+ string header(inner_data, inner_data + size);
+ servers->set_header(stream_id, http_header + header);
+ } else {
+ servers->add_data(stream_id, inner_data, size);
+ }
+
+ // Consume the block. This isn't the most efficient way of dealing with things
+ // should we have many blocks, but these routines don't need to be too efficient
+ // anyway.
+ pending_data.erase(pending_data.begin(), pending_data.begin() + sizeof(metacube_block_header) + size);
+ has_metacube_header = false;
+ }
+}
+
+void HTTPInput::drop_pending_data(size_t num_bytes)
+{
+ if (num_bytes == 0) {
+ return;
+ }
+ fprintf(stderr, "Warning: Dropping %lld junk bytes from stream, maybe it is not a Metacube stream?\n",
+ (long long)num_bytes);
+ pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);
+}
+