#include <string.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
+#include <sys/time.h>
#include <time.h>
#include <unistd.h>
#include <map>
#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 "timespec.h"
#include "util.h"
#include "version.h"
using namespace std;
+namespace {
+
+string host_header(const string &host, const string &port)
+{
+ if (atoi(port.c_str()) == 80) {
+ return host;
+ } else {
+ return host + ":" + port;
+ }
+}
+
+} // namespace
+
extern ServerPool *servers;
-
-HTTPInput::HTTPInput(const string &url)
+
+HTTPInput::HTTPInput(const string &url, Input::Encoding encoding)
: state(NOT_CONNECTED),
url(url),
+ encoding(encoding),
has_metacube_header(false),
sock(-1)
{
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()),
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;
{
if (sock != -1) {
safe_close(sock);
+ sock = -1;
}
MutexLock lock(&stats_mutex);
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;
}
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()) {
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] == ' ');
+ multimap<string, string> parameters = extract_headers(lines, url);
- 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));
+ // 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)”
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_COARSE, &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_COARSE, &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_COARSE, &last_activity);
+ assert(err != -1);
+ } else {
+ // OK. Most likely, should_stop was set, or we have timed out.
continue;
}
}
}
{
- 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);
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_COARSE, &last_activity);
}
break;
case SENDING_REQUEST: {
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) {
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;
}
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(metacube_block_header)) {
+ if (pending_data.size() < sizeof(metacube2_block_header)) {
return;
}
if (!has_metacube_header) {
char *ptr = static_cast<char *>(
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.
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<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);
-
- if (size > 262144) {
- log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header?",
+ 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 > 10485760) {
+ 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;
}
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) {
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
// 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;
}
}
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);