]> git.sesse.net Git - cubemap/blobdiff - httpinput.cpp
Fix an issue where “http” could sneak into the Host: header.
[cubemap] / httpinput.cpp
index 68b91708e7b3059eeac08d3d62088daae2d90aa4..9dcaeaa0191dd1951fc8fdb15f34148468c563c0 100644 (file)
 #include "serverpool.h"
 #include "state.pb.h"
 #include "stream.h"
+#include "timespec.h"
 #include "util.h"
 #include "version.h"
 
 using namespace std;
 
-extern ServerPool *servers;
-
 namespace {
 
-// Compute b-a.
-timespec clock_diff(const timespec &a, const timespec &b)
+string host_header(const string &host, const string &port)
 {
-       timespec ret;
-       ret.tv_sec = b.tv_sec - a.tv_sec;
-       ret.tv_nsec = b.tv_nsec - a.tv_nsec;
-       if (ret.tv_nsec < 0) {
-               ret.tv_sec--;
-               ret.tv_nsec += 1000000000;
+       if (port == "http" || atoi(port.c_str()) == 80) {
+               return host;
+       } else {
+               return host + ":" + port;
        }
-       assert(ret.tv_nsec >= 0);
-       return ret;
 }
 
 }  // namespace
 
-HTTPInput::HTTPInput(const string &url)
+extern ServerPool *servers;
+
+HTTPInput::HTTPInput(const string &url, Input::Encoding encoding)
        : state(NOT_CONNECTED),
          url(url),
+         encoding(encoding),
          has_metacube_header(false),
          sock(-1)
 {
@@ -65,6 +62,9 @@ HTTPInput::HTTPInput(const string &url)
 HTTPInput::HTTPInput(const InputProto &serialized)
        : state(State(serialized.state())),
          url(serialized.url()),
+         encoding(serialized.is_metacube_encoded() ?
+                  Input::INPUT_ENCODING_METACUBE :
+                  Input::INPUT_ENCODING_RAW),
          request(serialized.request()),
          request_bytes_sent(serialized.request_bytes_sent()),
          response(serialized.response()),
@@ -76,8 +76,8 @@ HTTPInput::HTTPInput(const InputProto &serialized)
        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.
+       string protocol, user;
+       parse_url(url, &protocol, &user, &host, &port, &path);  // Don't care if it fails.
 
        pthread_mutex_init(&stats_mutex, NULL);
        stats.url = url;
@@ -94,6 +94,7 @@ void HTTPInput::close_socket()
 {
        if (sock != -1) {
                safe_close(sock);
+               sock = -1;
        }
 
        MutexLock lock(&stats_mutex);
@@ -116,6 +117,12 @@ InputProto HTTPInput::serialize() const
        serialized.set_bytes_received(stats.bytes_received);
        serialized.set_data_bytes_received(stats.data_bytes_received);
        serialized.set_connect_time(stats.connect_time);
+       if (encoding == Input::INPUT_ENCODING_METACUBE) {
+               serialized.set_is_metacube_encoded(true);
+       } else {
+               assert(encoding == Input::INPUT_ENCODING_RAW);
+               serialized.set_is_metacube_encoded(false);
+       }
        return serialized;
 }
 
@@ -199,7 +206,7 @@ int HTTPInput::lookup_and_connect(const string &host, const string &port)
        return -1;
 }
        
-bool HTTPInput::parse_response(const std::string &request)
+bool HTTPInput::parse_response(const string &request)
 {
        vector<string> lines = split_lines(response);
        if (lines.empty()) {
@@ -221,31 +228,14 @@ bool HTTPInput::parse_response(const std::string &request)
                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) {
-                       log(WARNING, "[%s] Ignoring malformed HTTP response line '%s'",
-                               url.c_str(), 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;
-               }
+       multimap<string, string> parameters = extract_headers(lines, url);
 
-               parameters.insert(make_pair(key, value));
+       // Remove “Content-encoding: metacube”.
+       // TODO: Make case-insensitive.
+       multimap<string, string>::iterator encoding_it =
+               parameters.find("Content-encoding");
+       if (encoding_it != parameters.end() && encoding_it->second == "metacube") {
+               parameters.erase(encoding_it);
        }
 
        // Change “Server: foo” to “Server: metacube/0.1 (reflecting: foo)”
@@ -290,7 +280,7 @@ void HTTPInput::do_work()
 
        // TODO: Make the timeout persist across restarts.
        if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
-               int err = clock_gettime(CLOCK_MONOTONIC, &last_activity);
+               int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
                assert(err != -1);
        }
 
@@ -300,7 +290,7 @@ void HTTPInput::do_work()
                        static const int timeout_secs = 30;
 
                        timespec now;
-                       int err = clock_gettime(CLOCK_MONOTONIC, &now);
+                       int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now);
                        assert(err != -1);
 
                        timespec elapsed = clock_diff(last_activity, now);
@@ -319,7 +309,7 @@ void HTTPInput::do_work()
 
                        bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, &timeout);
                        if (activity) {
-                               err = clock_gettime(CLOCK_MONOTONIC, &last_activity);
+                               err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
                                assert(err != -1);
                        } else {
                                // OK. Most likely, should_stop was set, or we have timed out.
@@ -339,11 +329,18 @@ void HTTPInput::do_work()
                        }
 
                        {
-                               string protocol;  // Thrown away.
-                               if (!parse_url(url, &protocol, &host, &port, &path)) {
+                               string protocol, user;  // Thrown away.
+                               if (!parse_url(url, &protocol, &user, &host, &port, &path)) {
                                        log(WARNING, "[%s] Failed to parse URL '%s'", url.c_str(), url.c_str());
                                        break;
                                }
+
+                               // Remove the brackets around IPv6 address literals.
+                               // TODO: See if we can join this with the code in parse_ip_address(),
+                               // or maybe even more it into parse_url().
+                               if (!host.empty() && host[0] == '[' && host[host.size() - 1] == ']') {
+                                       host = host.substr(1, host.size() - 2);
+                               }
                        }
 
                        sock = lookup_and_connect(host, port);
@@ -355,13 +352,13 @@ void HTTPInput::do_work()
                                        state = CLOSING_SOCKET;
                                } else {
                                        state = SENDING_REQUEST;
-                                       request = "GET " + path + " HTTP/1.0\r\nUser-Agent: cubemap\r\n\r\n";
+                                       request = "GET " + path + " HTTP/1.0\r\nHost: " + host_header(host, port) + "\r\nUser-Agent: cubemap\r\n\r\n";
                                        request_bytes_sent = 0;
                                }
 
                                MutexLock lock(&stats_mutex);
                                stats.connect_time = time(NULL);
-                               clock_gettime(CLOCK_MONOTONIC, &last_activity);
+                               clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
                        }
                        break;
                case SENDING_REQUEST: {
@@ -411,7 +408,7 @@ void HTTPInput::do_work()
                        RequestParseStatus status = wait_for_double_newline(&response, buf, ret);
                        
                        if (status == RP_OUT_OF_SPACE) {
-                               log(WARNING, "[%s] Sever sent overlong HTTP response!", url.c_str());
+                               log(WARNING, "[%s] Server sent overlong HTTP response!", url.c_str());
                                state = CLOSING_SOCKET;
                                continue;
                        } else if (status == RP_NOT_FINISHED_YET) {
@@ -438,8 +435,14 @@ void HTTPInput::do_work()
                                process_data(&extra_data[0], extra_data.size());
                        }
 
-                       log(INFO, "[%s] Connected to '%s', receiving data.",
-                                  url.c_str(), url.c_str());
+                       if (encoding == Input::INPUT_ENCODING_RAW) {
+                               log(INFO, "[%s] Connected to '%s', receiving raw data.",
+                                          url.c_str(), url.c_str());
+                       } else {
+                               assert(encoding == Input::INPUT_ENCODING_METACUBE);
+                               log(INFO, "[%s] Connected to '%s', receiving data.",
+                                          url.c_str(), url.c_str());
+                       }
                        state = RECEIVING_DATA;
                        break;
                }
@@ -498,6 +501,15 @@ void HTTPInput::process_data(char *ptr, size_t bytes)
                stats.bytes_received += bytes;
        }
 
+       if (encoding == Input::INPUT_ENCODING_RAW) {
+               for (size_t i = 0; i < stream_indices.size(); ++i) {
+                       servers->add_data(stream_indices[i], ptr, bytes, /*metacube_flags=*/0);
+               }
+               return;
+       }
+
+       assert(encoding == Input::INPUT_ENCODING_METACUBE);
+
        for ( ;; ) {
                // If we don't have enough data (yet) for even the Metacube header, just return.
                if (pending_data.size() < sizeof(metacube2_block_header)) {
@@ -548,8 +560,8 @@ void HTTPInput::process_data(char *ptr, size_t bytes)
                        has_metacube_header = false;
                        continue;
                }
-               if (size > 262144) {
-                       log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header?",
+               if (size > 10485760) {
+                       log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header??",
                                url.c_str(), size, flags);
                }
 
@@ -569,16 +581,9 @@ void HTTPInput::process_data(char *ptr, size_t bytes)
                        for (size_t i = 0; i < stream_indices.size(); ++i) {
                                servers->set_header(stream_indices[i], http_header, stream_header);
                        }
-               } else {
-                       StreamStartSuitability suitable_for_stream_start;
-                       if (flags & METACUBE_FLAGS_NOT_SUITABLE_FOR_STREAM_START) {
-                               suitable_for_stream_start = NOT_SUITABLE_FOR_STREAM_START;
-                       } else {
-                               suitable_for_stream_start = SUITABLE_FOR_STREAM_START;
-                       }
-                       for (size_t i = 0; i < stream_indices.size(); ++i) {
-                               servers->add_data(stream_indices[i], inner_data, size, suitable_for_stream_start);
-                       }
+               }
+               for (size_t i = 0; i < stream_indices.size(); ++i) {
+                       servers->add_data(stream_indices[i], inner_data, size, flags);
                }
 
                // Consume the block. This isn't the most efficient way of dealing with things
@@ -594,7 +599,7 @@ void HTTPInput::drop_pending_data(size_t num_bytes)
        if (num_bytes == 0) {
                return;
        }
-       log(WARNING, "[%s] Dropping %lld junk bytes from stream, maybe it is not a Metacube2 stream?",
+       log(WARNING, "[%s] Dropping %lld junk bytes; not a Metacube2 stream, or data was dropped from the middle of the stream.",
                url.c_str(), (long long)num_bytes);
        assert(pending_data.size() >= num_bytes);
        pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);