X-Git-Url: https://git.sesse.net/?p=cubemap;a=blobdiff_plain;f=httpinput.cpp;h=807abacc87e0f466377b9a3153247454aac1f710;hp=2e6e617bad9905c78e1b3b7427cd53f2e5a6bb3f;hb=1f9e2a199957bd4c42bf56873d80c4c71381c3b4;hpb=cbdce14899459aca2e5331b6e1a969c359d28880 diff --git a/httpinput.cpp b/httpinput.cpp index 2e6e617..807abac 100644 --- a/httpinput.cpp +++ b/httpinput.cpp @@ -8,6 +8,7 @@ #include #include #include +#include #include #include #include @@ -17,18 +18,37 @@ #include "httpinput.h" #include "log.h" -#include "metacube.h" +#include "metacube2.h" #include "mutexlock.h" #include "parse.h" #include "serverpool.h" #include "state.pb.h" +#include "stream.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) +{ + 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; + } + assert(ret.tv_nsec >= 0); + return ret; +} + +} // namespace + HTTPInput::HTTPInput(const string &url) : state(NOT_CONNECTED), url(url), @@ -56,15 +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. - - // Older versions stored the extra \r\n in the HTTP header. - // Strip it if we find it. - if (http_header.size() >= 4 && - memcmp(http_header.data() + http_header.size() - 4, "\r\n\r\n", 4) == 0) { - http_header.resize(http_header.size() - 2); - } + 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; @@ -81,6 +94,7 @@ void HTTPInput::close_socket() { if (sock != -1) { safe_close(sock); + sock = -1; } MutexLock lock(&stats_mutex); @@ -208,31 +222,14 @@ bool HTTPInput::parse_response(const std::string &request) return false; } - multimap 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 parameters = extract_headers(lines, url); - parameters.insert(make_pair(key, value)); + // Remove “Content-encoding: metacube”. + // TODO: Make case-insensitive. + multimap::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)” @@ -273,11 +270,43 @@ bool HTTPInput::parse_response(const std::string &request) void HTTPInput::do_work() { + timespec last_activity; + + // 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); + assert(err != -1); + } + while (!should_stop()) { if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) { - bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, NULL); - if (!activity) { - // Most likely, should_stop was set. + // Give the socket 30 seconds since last activity before we time out. + static const int timeout_secs = 30; + + timespec now; + int err = clock_gettime(CLOCK_MONOTONIC, &now); + assert(err != -1); + + timespec elapsed = clock_diff(last_activity, now); + if (elapsed.tv_sec >= timeout_secs) { + // Timeout! + log(ERROR, "[%s] Timeout after %d seconds, closing.", url.c_str(), elapsed.tv_sec); + state = CLOSING_SOCKET; + continue; + } + + // Basically calculate (30 - (now - last_activity)) = (30 + (last_activity - now)). + // Add a second of slack to account for differences between clocks. + timespec timeout = clock_diff(now, last_activity); + timeout.tv_sec += timeout_secs + 1; + assert(timeout.tv_sec > 0 || (timeout.tv_sec >= 0 && timeout.tv_nsec > 0)); + + bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, &timeout); + if (activity) { + err = clock_gettime(CLOCK_MONOTONIC, &last_activity); + assert(err != -1); + } else { + // OK. Most likely, should_stop was set, or we have timed out. continue; } } @@ -294,8 +323,8 @@ 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; } @@ -316,6 +345,7 @@ void HTTPInput::do_work() MutexLock lock(&stats_mutex); stats.connect_time = time(NULL); + clock_gettime(CLOCK_MONOTONIC, &last_activity); } break; case SENDING_REQUEST: { @@ -365,7 +395,7 @@ 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) { @@ -454,7 +484,7 @@ void HTTPInput::process_data(char *ptr, size_t 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)) { + if (pending_data.size() < sizeof(metacube2_block_header)) { return; } @@ -463,13 +493,13 @@ void HTTPInput::process_data(char *ptr, size_t bytes) if (!has_metacube_header) { char *ptr = static_cast( memmem(pending_data.data(), pending_data.size(), - METACUBE_SYNC, strlen(METACUBE_SYNC))); + METACUBE2_SYNC, strlen(METACUBE2_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)); + drop_pending_data(pending_data.size() - (strlen(METACUBE2_SYNC) - 1)); return; } else { // Yay, we found the header. Drop everything (if anything) before it. @@ -477,25 +507,38 @@ void HTTPInput::process_data(char *ptr, size_t bytes) 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)) { + if (pending_data.size() < sizeof(metacube2_block_header)) { return; } } } // Now it's safe to read the header. - metacube_block_header *hdr = reinterpret_cast(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); - - if (size > 262144) { + metacube2_block_header hdr; + memcpy(&hdr, pending_data.data(), sizeof(hdr)); + assert(memcmp(hdr.sync, METACUBE2_SYNC, sizeof(hdr.sync)) == 0); + uint32_t size = ntohl(hdr.size); + uint16_t flags = ntohs(hdr.flags); + uint16_t expected_csum = metacube2_compute_crc(&hdr); + + if (expected_csum != ntohs(hdr.csum)) { + log(WARNING, "[%s] Metacube checksum failed (expected 0x%x, got 0x%x), " + "not reading block claiming to be %d bytes (flags=%x).", + url.c_str(), expected_csum, ntohs(hdr.csum), + size, flags); + + // Drop only the first byte, and let the rest of the code handle resync. + pending_data.erase(pending_data.begin(), pending_data.begin() + 1); + has_metacube_header = false; + continue; + } + if (size > 1048576) { log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header?", url.c_str(), size, flags); } // See if we have the entire block. If not, wait for more data. - if (pending_data.size() < sizeof(metacube_block_header) + size) { + if (pending_data.size() < sizeof(metacube2_block_header) + size) { return; } @@ -504,7 +547,7 @@ void HTTPInput::process_data(char *ptr, size_t bytes) MutexLock lock(&stats_mutex); stats.data_bytes_received += size; } - char *inner_data = pending_data.data() + sizeof(metacube_block_header); + char *inner_data = pending_data.data() + sizeof(metacube2_block_header); if (flags & METACUBE_FLAGS_HEADER) { stream_header = string(inner_data, inner_data + size); for (size_t i = 0; i < stream_indices.size(); ++i) { @@ -525,7 +568,7 @@ void HTTPInput::process_data(char *ptr, size_t bytes) // 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); + pending_data.erase(pending_data.begin(), pending_data.begin() + sizeof(metacube2_block_header) + size); has_metacube_header = false; } } @@ -535,7 +578,7 @@ 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 Metacube 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);