+#include <stdio.h>
#include <assert.h>
#include <errno.h>
#include <netdb.h>
#include <netinet/in.h>
#include <poll.h>
#include <stdint.h>
-#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/ioctl.h>
#include "parse.h"
#include "serverpool.h"
#include "state.pb.h"
+#include "util.h"
#include "version.h"
using namespace std;
extern ServerPool *servers;
-HTTPInput::HTTPInput(const string &stream_id, const string &url)
+HTTPInput::HTTPInput(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()),
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);
+ }
}
void HTTPInput::close_socket()
{
- int ret;
- do {
- ret = close(sock);
- } while (ret == -1 && errno == EINTR);
-
- if (ret == -1) {
- log_perror("close()");
- }
+ safe_close(sock);
}
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);
addrinfo *ai;
int err = getaddrinfo(host.c_str(), port.c_str(), NULL, &ai);
if (err == -1) {
- log(WARNING, "Lookup of '%s' failed (%s).",
- host.c_str(), gai_strerror(err));
+ log(WARNING, "[%s] Lookup of '%s' failed (%s).",
+ url.c_str(), host.c_str(), gai_strerror(err));
freeaddrinfo(ai);
return -1;
}
+ addrinfo *base_ai = ai;
+
// Connect to everything in turn until we have a socket.
- while (ai && !should_stop) {
+ 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;
}
+ // Now do a non-blocking connect. This is important because we want to be able to be
+ // woken up, even though it's rather cumbersome.
+
+ // Set the socket as nonblocking.
+ int one = 1;
+ if (ioctl(sock, FIONBIO, &one) == -1) {
+ log_perror("ioctl(FIONBIO)");
+ safe_close(sock);
+ return -1;
+ }
+
+ // Do a non-blocking connect.
do {
err = connect(sock, ai->ai_addr, ai->ai_addrlen);
} while (err == -1 && errno == EINTR);
- if (err != -1) {
- freeaddrinfo(ai);
- return sock;
+ if (err == -1 && errno != EINPROGRESS) {
+ log_perror("connect");
+ safe_close(sock);
+ continue;
}
- do {
- err = close(sock);
- } while (err == -1 && errno == EINTR);
+ // Wait for the connect to complete, or an error to happen.
+ for ( ;; ) {
+ bool complete = wait_for_activity(sock, POLLIN | POLLOUT, NULL);
+ if (should_stop()) {
+ safe_close(sock);
+ return -1;
+ }
+ if (complete) {
+ break;
+ }
+ }
- if (err == -1) {
- log_perror("close");
- // Can still continue.
+ // Check whether it ended in an error or not.
+ socklen_t err_size = sizeof(err);
+ if (getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &err_size) == -1) {
+ log_perror("getsockopt");
+ safe_close(sock);
+ continue;
}
+ errno = err;
+
+ if (err == 0) {
+ // Successful connect.
+ freeaddrinfo(base_ai);
+ return sock;
+ }
+
+ safe_close(sock);
ai = ai->ai_next;
}
// Give the last one as error.
- log(WARNING, "Connect to '%s' failed (%s)",
- host.c_str(), strerror(errno));
- freeaddrinfo(ai);
+ log(WARNING, "[%s] Connect to '%s' failed (%s)",
+ url.c_str(), host.c_str(), strerror(errno));
+ freeaddrinfo(base_ai);
return -1;
}
{
vector<string> lines = split_lines(response);
if (lines.empty()) {
- log(WARNING, "Empty HTTP response from input.");
+ log(WARNING, "[%s] Empty HTTP response from input.", url.c_str());
return false;
}
vector<string> first_line_tokens = split_tokens(lines[0]);
if (first_line_tokens.size() < 2) {
- log(WARNING, "Malformed response line '%s' from input.",
- lines[0].c_str());
+ log(WARNING, "[%s] Malformed response line '%s' from input.",
+ url.c_str(), lines[0].c_str());
return false;
}
int response = atoi(first_line_tokens[1].c_str());
if (response != 200) {
- log(WARNING, "Non-200 response '%s' from input.",
- lines[0].c_str());
+ log(WARNING, "[%s] Non-200 response '%s' from input.",
+ url.c_str(), lines[0].c_str());
return false;
}
for (size_t i = 1; i < lines.size(); ++i) {
size_t split = lines[i].find(":");
if (split == string::npos) {
- log(WARNING, "Ignoring malformed HTTP response line '%s'",
- lines[i].c_str());
+ log(WARNING, "[%s] Ignoring malformed HTTP response line '%s'",
+ url.c_str(), lines[i].c_str());
continue;
}
}
}
+ // Set “Connection: close”.
+ // TODO: Make case-insensitive.
+ parameters.erase("Connection");
+ parameters.insert(make_pair("Connection", "close"));
+
// Construct the new HTTP header.
http_header = "HTTP/1.0 200 OK\r\n";
for (multimap<string, string>::iterator it = parameters.begin();
++it) {
http_header.append(it->first + ": " + it->second + "\r\n");
}
- http_header.append("\r\n");
- servers->set_header(stream_id, http_header);
+
+ for (size_t i = 0; i < stream_ids.size(); ++i) {
+ servers->set_header(stream_ids[i], http_header, "");
+ }
return true;
}
void HTTPInput::do_work()
{
- while (!should_stop) {
+ 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)) {
+ bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, NULL);
+ if (!activity) {
+ // Most likely, should_stop was set.
continue;
}
- if (nfds == -1) {
- log_perror("poll");
- state = CLOSING_SOCKET;
- }
}
switch (state) {
request_bytes_sent = 0;
response.clear();
pending_data.clear();
- servers->set_header(stream_id, "");
+ for (size_t i = 0; i < stream_ids.size(); ++i) {
+ servers->set_header(stream_ids[i], "", "");
+ }
{
string protocol; // Thrown away.
if (!parse_url(url, &protocol, &host, &port, &path)) {
- log(WARNING, "Failed to parse URL '%s'", url.c_str());
+ log(WARNING, "[%s] Failed to parse URL '%s'", url.c_str(), url.c_str());
break;
}
}
if (ret == 0) {
// This really shouldn't happen...
- log(ERROR, "Socket unexpectedly closed while reading header");
+ log(ERROR, "[%s] Socket unexpectedly closed while reading header",
+ url.c_str());
state = CLOSING_SOCKET;
continue;
}
RequestParseStatus status = wait_for_double_newline(&response, buf, ret);
if (status == RP_OUT_OF_SPACE) {
- log(WARNING, "fd %d sent overlong response!", sock);
+ log(WARNING, "[%s] Sever sent overlong HTTP response!", url.c_str());
state = CLOSING_SOCKET;
continue;
} else if (status == RP_NOT_FINISHED_YET) {
process_data(&extra_data[0], extra_data.size());
}
+ log(INFO, "[%s] Connected to '%s', receiving data.",
+ url.c_str(), url.c_str());
state = RECEIVING_DATA;
break;
}
if (ret == 0) {
// This really shouldn't happen...
- log(ERROR, "Socket unexpectedly closed while reading header");
+ log(ERROR, "[%s] Socket unexpectedly closed while reading header",
+ url.c_str());
state = CLOSING_SOCKET;
continue;
}
break;
}
case CLOSING_SOCKET: {
- int err;
- do {
- err = close(sock);
- } while (err == -1 && errno == EINTR);
-
- if (err == -1) {
- log_perror("close");
- }
-
+ close_socket();
state = NOT_CONNECTED;
break;
}
// 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) {
- log(INFO, "Waiting 0.2 second and restarting...");
- usleep(200000);
+ if (state == NOT_CONNECTED && !should_stop()) {
+ log(INFO, "[%s] Waiting 0.2 second and restarting...", url.c_str());
+ timespec timeout_ts;
+ timeout_ts.tv_sec = 0;
+ timeout_ts.tv_nsec = 200000000;
+ wait_for_wakeup(&timeout_ts);
}
}
}
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);
+ for (size_t i = 0; i < stream_ids.size(); ++i) {
+ servers->set_header(stream_ids[i], http_header, header);
+ }
} else {
- servers->add_data(stream_id, inner_data, size);
+ for (size_t i = 0; i < stream_ids.size(); ++i) {
+ servers->add_data(stream_ids[i], inner_data, size);
+ }
}
// Consume the block. This isn't the most efficient way of dealing with things
if (num_bytes == 0) {
return;
}
- log(WARNING, "Dropping %lld junk bytes from stream, maybe it is not a Metacube stream?",
- (long long)num_bytes);
+ log(WARNING, "[%s] Dropping %lld junk bytes from stream, maybe it is not a Metacube stream?",
+ url.c_str(), (long long)num_bytes);
pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);
}