From 239a9eab1a2468b401183745a24b2fbd590a6998 Mon Sep 17 00:00:00 2001 From: "Steinar H. Gunderson" Date: Thu, 11 Apr 2013 21:06:03 +0200 Subject: [PATCH] Make Input a bit more generic, to pave the way for UDP. --- Makefile | 2 +- httpinput.cpp | 439 ++++++++++++++++++++++++++++++++++++++++++++++++++ httpinput.h | 79 +++++++++ input.cpp | 432 +------------------------------------------------ input.h | 71 +------- main.cpp | 5 +- 6 files changed, 532 insertions(+), 496 deletions(-) create mode 100644 httpinput.cpp create mode 100644 httpinput.h diff --git a/Makefile b/Makefile index 647dfe9..5367daa 100644 --- a/Makefile +++ b/Makefile @@ -4,7 +4,7 @@ PROTOC=protoc CXXFLAGS=-Wall -O2 -g LDLIBS=-lpthread -lprotobuf -OBJS=main.o server.o serverpool.o mutexlock.o input.o parse.o markpool.o acceptor.o stats.o thread.o state.pb.o +OBJS=main.o server.o serverpool.o mutexlock.o input.o httpinput.o parse.o markpool.o acceptor.o stats.o thread.o state.pb.o all: cubemap diff --git a/httpinput.cpp b/httpinput.cpp new file mode 100644 index 0000000..1db4f91 --- /dev/null +++ b/httpinput.cpp @@ -0,0 +1,439 @@ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#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 lines = split_lines(response); + if (lines.empty()) { + fprintf(stderr, "WARNING: Empty HTTP response from input.\n"); + return false; + } + + vector 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 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::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::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( + 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( + 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(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); +} + diff --git a/httpinput.h b/httpinput.h new file mode 100644 index 0000000..894d8e9 --- /dev/null +++ b/httpinput.h @@ -0,0 +1,79 @@ +#ifndef _HTTPINPUT_H +#define _HTTPINPUT_H 1 + +#include +#include + +#include "input.h" + +class InputProto; + +class HTTPInput : public Input { +public: + HTTPInput(const std::string &stream_id, const std::string &url); + + // Serialization/deserialization. + HTTPInput(const InputProto &serialized); + virtual InputProto serialize() const; + + virtual std::string get_url() const { return url; } + +private: + // Actually does the download. + virtual void do_work(); + + // Open a socket that connects to the given host and port. Does DNS resolving. + int lookup_and_connect(const std::string &host, const std::string &port); + + // Parses a HTTP response. Returns false if it not a 200. + bool parse_response(const std::string &response); + + // Stores the given data, looks for Metacube blocks (skipping data if needed), + // and calls process_block() for each one. + void process_data(char *ptr, size_t bytes); + + // Drops bytes from the head of , + // and outputs a warning. + void drop_pending_data(size_t num_bytes); + + enum State { + NOT_CONNECTED, + SENDING_REQUEST, + RECEIVING_HEADER, + RECEIVING_DATA, + CLOSING_SOCKET, // Due to error. + }; + State state; + + std::string stream_id; + + // The URL and its parsed components. + std::string url; + std::string host, port, path; + + // The HTTP request, with headers and all. + // Only relevant for SENDING_REQUEST. + std::string request; + + // How many bytes we've sent of the request so far. + // Only relevant for SENDING_REQUEST. + size_t request_bytes_sent; + + // The HTTP response we've received so far. Only relevant for RECEIVING_HEADER. + std::string response; + + // The HTTP respones headers we want to give clients for this input. + std::string http_header; + + // Data we have received but not fully processed yet. + std::vector pending_data; + + // If starts with a Metacube header, + // this is true. + bool has_metacube_header; + + // The socket we are downloading on (or -1). + int sock; +}; + +#endif // !defined(_HTTPINPUT_H) diff --git a/input.cpp b/input.cpp index a54e6fb..1e61780 100644 --- a/input.cpp +++ b/input.cpp @@ -1,40 +1,18 @@ -#include #include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include #include -#include -#include "metacube.h" -#include "mutexlock.h" #include "input.h" -#include "server.h" -#include "serverpool.h" -#include "parse.h" -#include "state.pb.h" using namespace std; -extern ServerPool *servers; - // Extremely rudimentary URL parsing. -bool parse_url(const string &url, string *host, string *port, string *path) +bool parse_url(const string &url, string *protocol, string *host, string *port, string *path) { if (url.find("http://") != 0) { return false; } + + *protocol = "http"; string rest = url.substr(strlen("http://")); size_t split = rest.find_first_of(":/"); @@ -72,407 +50,5 @@ bool parse_url(const string &url, string *host, string *port, string *path) return true; } -Input::Input(const string &stream_id, const string &url) - : state(NOT_CONNECTED), - stream_id(stream_id), - url(url), - has_metacube_header(false), - sock(-1) -{ -} - -Input::Input(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()); - - parse_url(url, &host, &port, &path); // Don't care if it fails. -} - -InputProto Input::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 Input::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 Input::parse_response(const std::string &request) -{ - vector lines = split_lines(response); - if (lines.empty()) { - fprintf(stderr, "WARNING: Empty HTTP response from input.\n"); - return false; - } - - vector 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 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::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::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 Input::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(); - - if (!parse_url(url, &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( - 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 Input::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( - 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(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 Input::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); -} +Input::~Input() {} diff --git a/input.h b/input.h index 14537c8..3e86510 100644 --- a/input.h +++ b/input.h @@ -1,79 +1,20 @@ #ifndef _INPUT_H #define _INPUT_H 1 -#include #include #include "thread.h" class InputProto; +// Extremely rudimentary URL parsing. +bool parse_url(const std::string &url, std::string *protocol, std::string *host, std::string *port, std::string *path); + class Input : public Thread { public: - Input(const std::string &stream_id, const std::string &url); - - // Serialization/deserialization. - Input(const InputProto &serialized); - InputProto serialize() const; - - std::string get_url() const { return url; } - -private: - // Actually does the download. - virtual void do_work(); - - // Open a socket that connects to the given host and port. Does DNS resolving. - int lookup_and_connect(const std::string &host, const std::string &port); - - // Parses a HTTP response. Returns false if it not a 200. - bool parse_response(const std::string &response); - - // Stores the given data, looks for Metacube blocks (skipping data if needed), - // and calls process_block() for each one. - void process_data(char *ptr, size_t bytes); - - // Drops bytes from the head of , - // and outputs a warning. - void drop_pending_data(size_t num_bytes); - - enum State { - NOT_CONNECTED, - SENDING_REQUEST, - RECEIVING_HEADER, - RECEIVING_DATA, - CLOSING_SOCKET, // Due to error. - }; - State state; - - std::string stream_id; - - // The URL and its parsed components. - std::string url; - std::string host, port, path; - - // The HTTP request, with headers and all. - // Only relevant for SENDING_REQUEST. - std::string request; - - // How many bytes we've sent of the request so far. - // Only relevant for SENDING_REQUEST. - size_t request_bytes_sent; - - // The HTTP response we've received so far. Only relevant for RECEIVING_HEADER. - std::string response; - - // The HTTP respones headers we want to give clients for this input. - std::string http_header; - - // Data we have received but not fully processed yet. - std::vector pending_data; - - // If starts with a Metacube header, - // this is true. - bool has_metacube_header; - - // The socket we are downloading on (or -1). - int sock; + virtual ~Input(); + virtual InputProto serialize() const = 0; + virtual std::string get_url() const = 0; }; #endif // !defined(_INPUT_H) diff --git a/main.cpp b/main.cpp index db3be63..04d0d3d 100644 --- a/main.cpp +++ b/main.cpp @@ -25,6 +25,7 @@ #include "server.h" #include "serverpool.h" #include "input.h" +#include "httpinput.h" #include "stats.h" #include "state.pb.h" @@ -267,7 +268,7 @@ vector create_inputs(const vector &config, deserialized_inputs->erase(deserialized_input_it); } if (input == NULL) { - input = new Input(stream_id, src); + input = new HTTPInput(stream_id, src); } input->run(); inputs.push_back(input); @@ -358,7 +359,7 @@ int main(int argc, char **argv) for (int i = 0; i < loaded_state.inputs_size(); ++i) { deserialized_inputs.insert(make_pair( loaded_state.inputs(i).stream_id(), - new Input(loaded_state.inputs(i)))); + new HTTPInput(loaded_state.inputs(i)))); } // Convert the acceptor from older serialized formats. -- 2.39.2