3 Copyright (C) 2013 celeron55, Perttu Ahola <celeron55@gmail.com>
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU Lesser General Public License as published by
7 the Free Software Foundation; either version 2.1 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU Lesser General Public License for more details.
15 You should have received a copy of the GNU Lesser General Public License along
16 with this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
20 #include "clientmedia.h"
21 #include "httpfetch.h"
23 #include "filecache.h"
30 #include "network/networkprotocol.h"
32 #include "util/serialize.h"
33 #include "util/sha1.h"
34 #include "util/string.h"
36 static std::string getMediaCacheDir()
38 return porting::path_user + DIR_DELIM + "cache" + DIR_DELIM + "media";
45 ClientMediaDownloader::ClientMediaDownloader():
46 m_media_cache(getMediaCacheDir())
48 m_initial_step_done = false;
49 m_name_bound = ""; // works because "" is an invalid file name
51 m_uncached_received_count = 0;
52 m_httpfetch_caller = HTTPFETCH_DISCARD;
53 m_httpfetch_active = 0;
54 m_httpfetch_active_limit = 0;
55 m_httpfetch_next_id = 0;
56 m_httpfetch_timeout = 0;
57 m_outstanding_hash_sets = 0;
60 ClientMediaDownloader::~ClientMediaDownloader()
62 if (m_httpfetch_caller != HTTPFETCH_DISCARD)
63 httpfetch_caller_free(m_httpfetch_caller);
65 for (std::map<std::string, FileStatus*>::iterator it = m_files.begin();
66 it != m_files.end(); ++it)
69 for (u32 i = 0; i < m_remotes.size(); ++i)
73 void ClientMediaDownloader::addFile(std::string name, std::string sha1)
75 assert(!m_initial_step_done);
77 // if name was already announced, ignore the new announcement
78 if (m_files.count(name) != 0) {
79 errorstream << "Client: ignoring duplicate media announcement "
80 << "sent by server: \"" << name << "\""
85 // if name is empty or contains illegal characters, ignore the file
86 if (name.empty() || !string_allowed(name, TEXTURENAME_ALLOWED_CHARS)) {
87 errorstream << "Client: ignoring illegal file name "
88 << "sent by server: \"" << name << "\""
93 // length of sha1 must be exactly 20 (160 bits), else ignore the file
94 if (sha1.size() != 20) {
95 errorstream << "Client: ignoring illegal SHA1 sent by server: "
96 << hex_encode(sha1) << " \"" << name << "\""
101 FileStatus *filestatus = new FileStatus;
102 filestatus->received = false;
103 filestatus->sha1 = sha1;
104 filestatus->current_remote = -1;
105 m_files.insert(std::make_pair(name, filestatus));
108 void ClientMediaDownloader::addRemoteServer(std::string baseurl)
110 assert(!m_initial_step_done);
114 if (g_settings->getBool("enable_remote_media_server")) {
115 infostream << "Client: Adding remote server \""
116 << baseurl << "\" for media download" << std::endl;
118 RemoteServerStatus *remote = new RemoteServerStatus;
119 remote->baseurl = baseurl;
120 remote->active_count = 0;
121 remote->request_by_filename = false;
122 m_remotes.push_back(remote);
127 infostream << "Client: Ignoring remote server \""
128 << baseurl << "\" because cURL support is not compiled in"
134 void ClientMediaDownloader::step(Client *client)
136 if (!m_initial_step_done) {
138 m_initial_step_done = true;
141 // Remote media: check for completion of fetches
142 if (m_httpfetch_active) {
143 bool fetched_something = false;
144 HTTPFetchResult fetch_result;
146 while (httpfetch_async_get(m_httpfetch_caller, fetch_result)) {
147 m_httpfetch_active--;
148 fetched_something = true;
150 // Is this a hashset (index.mth) or a media file?
151 if (fetch_result.request_id < m_remotes.size())
152 remoteHashSetReceived(fetch_result);
154 remoteMediaReceived(fetch_result, client);
157 if (fetched_something)
158 startRemoteMediaTransfers();
160 // Did all remote transfers end and no new ones can be started?
161 // If so, request still missing files from the minetest server
162 // (Or report that we have all files.)
163 if (m_httpfetch_active == 0) {
164 if (m_uncached_received_count < m_uncached_count) {
165 infostream << "Client: Failed to remote-fetch "
166 << (m_uncached_count-m_uncached_received_count)
167 << " files. Requesting them"
168 << " the usual way." << std::endl;
170 startConventionalTransfers(client);
175 void ClientMediaDownloader::initialStep(Client *client)
178 m_uncached_count = m_files.size();
179 for (std::map<std::string, FileStatus*>::iterator
180 it = m_files.begin();
181 it != m_files.end(); ++it) {
182 std::string name = it->first;
183 FileStatus *filestatus = it->second;
184 const std::string &sha1 = filestatus->sha1;
186 std::ostringstream tmp_os(std::ios_base::binary);
187 bool found_in_cache = m_media_cache.load(hex_encode(sha1), tmp_os);
189 // If found in cache, try to load it from there
190 if (found_in_cache) {
191 bool success = checkAndLoad(name, sha1,
192 tmp_os.str(), true, client);
194 filestatus->received = true;
200 assert(m_uncached_received_count == 0);
202 // Create the media cache dir if we are likely to write to it
203 if (m_uncached_count != 0) {
204 bool did = fs::CreateAllDirs(getMediaCacheDir());
206 errorstream << "Client: "
207 << "Could not create media cache directory: "
208 << getMediaCacheDir()
213 // If we found all files in the cache, report this fact to the server.
214 // If the server reported no remote servers, immediately start
215 // conventional transfers. Note: if cURL support is not compiled in,
216 // m_remotes is always empty, so "!USE_CURL" is redundant but may
217 // reduce the size of the compiled code
218 if (!USE_CURL || m_uncached_count == 0 || m_remotes.empty()) {
219 startConventionalTransfers(client);
222 // Otherwise start off by requesting each server's sha1 set
224 // This is the first time we use httpfetch, so alloc a caller ID
225 m_httpfetch_caller = httpfetch_caller_alloc();
226 m_httpfetch_timeout = g_settings->getS32("curl_timeout");
228 // Set the active fetch limit to curl_parallel_limit or 84,
229 // whichever is greater. This gives us some leeway so that
230 // inefficiencies in communicating with the httpfetch thread
231 // don't slow down fetches too much. (We still want some limit
232 // so that when the first remote server returns its hash set,
233 // not all files are requested from that server immediately.)
234 // One such inefficiency is that ClientMediaDownloader::step()
235 // is only called a couple times per second, while httpfetch
236 // might return responses much faster than that.
237 // Note that httpfetch strictly enforces curl_parallel_limit
238 // but at no inter-thread communication cost. This however
239 // doesn't help with the aforementioned inefficiencies.
240 // The signifance of 84 is that it is 2*6*9 in base 13.
241 m_httpfetch_active_limit = g_settings->getS32("curl_parallel_limit");
242 m_httpfetch_active_limit = MYMAX(m_httpfetch_active_limit, 84);
244 // Write a list of hashes that we need. This will be POSTed
245 // to the server using Content-Type: application/octet-stream
246 std::string required_hash_set = serializeRequiredHashSet();
248 // minor fixme: this loop ignores m_httpfetch_active_limit
250 // another minor fixme, unlikely to matter in normal usage:
251 // these index.mth fetches do (however) count against
252 // m_httpfetch_active_limit when starting actual media file
253 // requests, so if there are lots of remote servers that are
254 // not responding, those will stall new media file transfers.
256 for (u32 i = 0; i < m_remotes.size(); ++i) {
257 assert(m_httpfetch_next_id == i);
259 RemoteServerStatus *remote = m_remotes[i];
260 actionstream << "Client: Contacting remote server \""
261 << remote->baseurl << "\"" << std::endl;
263 HTTPFetchRequest fetch_request;
265 remote->baseurl + MTHASHSET_FILE_NAME;
266 fetch_request.caller = m_httpfetch_caller;
267 fetch_request.request_id = m_httpfetch_next_id; // == i
268 fetch_request.timeout = m_httpfetch_timeout;
269 fetch_request.connect_timeout = m_httpfetch_timeout;
270 fetch_request.post_data = required_hash_set;
271 fetch_request.extra_headers.push_back(
272 "Content-Type: application/octet-stream");
273 httpfetch_async(fetch_request);
275 m_httpfetch_active++;
276 m_httpfetch_next_id++;
277 m_outstanding_hash_sets++;
282 void ClientMediaDownloader::remoteHashSetReceived(
283 const HTTPFetchResult &fetch_result)
285 u32 remote_id = fetch_result.request_id;
286 assert(remote_id < m_remotes.size());
287 RemoteServerStatus *remote = m_remotes[remote_id];
289 m_outstanding_hash_sets--;
291 if (fetch_result.succeeded) {
293 // Server sent a list of file hashes that are
294 // available on it, try to parse the list
296 std::set<std::string> sha1_set;
297 deSerializeHashSet(fetch_result.data, sha1_set);
299 // Parsing succeeded: For every file that is
300 // available on this server, add this server
301 // to the available_remotes array
303 for(std::map<std::string, FileStatus*>::iterator
304 it = m_files.upper_bound(m_name_bound);
305 it != m_files.end(); ++it) {
306 FileStatus *f = it->second;
307 if (!f->received && sha1_set.count(f->sha1))
308 f->available_remotes.push_back(remote_id);
311 catch (SerializationError &e) {
312 infostream << "Client: Remote server \""
313 << remote->baseurl << "\" sent invalid hash set: "
314 << e.what() << std::endl;
318 // For compatibility: If index.mth is not found, assume that the
319 // server contains files named like the original files (not their sha1)
321 // Do NOT check for any particular response code (e.g. 404) here,
322 // because different servers respond differently
324 if (!fetch_result.succeeded && !fetch_result.timeout) {
325 infostream << "Client: Enabling compatibility mode for remote "
326 << "server \"" << remote->baseurl << "\"" << std::endl;
327 remote->request_by_filename = true;
329 // Assume every file is available on this server
331 for(std::map<std::string, FileStatus*>::iterator
332 it = m_files.upper_bound(m_name_bound);
333 it != m_files.end(); ++it) {
334 FileStatus *f = it->second;
336 f->available_remotes.push_back(remote_id);
341 void ClientMediaDownloader::remoteMediaReceived(
342 const HTTPFetchResult &fetch_result,
345 // Some remote server sent us a file.
346 // -> decrement number of active fetches
347 // -> mark file as received if fetch succeeded
348 // -> try to load media
352 std::map<unsigned long, std::string>::iterator it =
353 m_remote_file_transfers.find(fetch_result.request_id);
354 assert(it != m_remote_file_transfers.end());
356 m_remote_file_transfers.erase(it);
359 assert(m_files.count(name) != 0);
361 FileStatus *filestatus = m_files[name];
362 assert(!filestatus->received);
363 assert(filestatus->current_remote >= 0);
365 RemoteServerStatus *remote = m_remotes[filestatus->current_remote];
367 filestatus->current_remote = -1;
368 remote->active_count--;
370 // If fetch succeeded, try to load media file
372 if (fetch_result.succeeded) {
373 bool success = checkAndLoad(name, filestatus->sha1,
374 fetch_result.data, false, client);
376 filestatus->received = true;
377 assert(m_uncached_received_count < m_uncached_count);
378 m_uncached_received_count++;
383 s32 ClientMediaDownloader::selectRemoteServer(FileStatus *filestatus)
385 assert(filestatus != NULL);
386 assert(!filestatus->received);
387 assert(filestatus->current_remote < 0);
389 if (filestatus->available_remotes.empty())
392 // Of all servers that claim to provide the file (and haven't
393 // been unsuccessfully tried before), find the one with the
394 // smallest number of currently active transfers
397 s32 best_remote_id = filestatus->available_remotes[best];
398 s32 best_active_count = m_remotes[best_remote_id]->active_count;
400 for (u32 i = 1; i < filestatus->available_remotes.size(); ++i) {
401 s32 remote_id = filestatus->available_remotes[i];
402 s32 active_count = m_remotes[remote_id]->active_count;
403 if (active_count < best_active_count) {
405 best_remote_id = remote_id;
406 best_active_count = active_count;
410 filestatus->available_remotes.erase(
411 filestatus->available_remotes.begin() + best);
413 return best_remote_id;
417 void ClientMediaDownloader::startRemoteMediaTransfers()
419 bool changing_name_bound = true;
421 for (std::map<std::string, FileStatus*>::iterator
422 files_iter = m_files.upper_bound(m_name_bound);
423 files_iter != m_files.end(); ++files_iter) {
425 // Abort if active fetch limit is exceeded
426 if (m_httpfetch_active >= m_httpfetch_active_limit)
429 const std::string &name = files_iter->first;
430 FileStatus *filestatus = files_iter->second;
432 if (!filestatus->received && filestatus->current_remote < 0) {
433 // File has not been received yet and is not currently
434 // being transferred. Choose a server for it.
435 s32 remote_id = selectRemoteServer(filestatus);
436 if (remote_id >= 0) {
437 // Found a server, so start fetching
438 RemoteServerStatus *remote =
439 m_remotes[remote_id];
441 std::string url = remote->baseurl +
442 (remote->request_by_filename ? name :
443 hex_encode(filestatus->sha1));
444 verbosestream << "Client: "
445 << "Requesting remote media file "
446 << "\"" << name << "\" "
447 << "\"" << url << "\"" << std::endl;
449 HTTPFetchRequest fetch_request;
450 fetch_request.url = url;
451 fetch_request.caller = m_httpfetch_caller;
452 fetch_request.request_id = m_httpfetch_next_id;
453 fetch_request.timeout = 0; // no data timeout!
454 fetch_request.connect_timeout =
456 httpfetch_async(fetch_request);
458 m_remote_file_transfers.insert(std::make_pair(
462 filestatus->current_remote = remote_id;
463 remote->active_count++;
464 m_httpfetch_active++;
465 m_httpfetch_next_id++;
469 if (filestatus->received ||
470 (filestatus->current_remote < 0 &&
471 !m_outstanding_hash_sets)) {
472 // If we arrive here, we conclusively know that we
473 // won't fetch this file from a remote server in the
474 // future. So update the name bound if possible.
475 if (changing_name_bound)
479 changing_name_bound = false;
484 void ClientMediaDownloader::startConventionalTransfers(Client *client)
486 assert(m_httpfetch_active == 0);
488 if (m_uncached_received_count != m_uncached_count) {
489 // Some media files have not been received yet, use the
490 // conventional slow method (minetest protocol) to get them
491 std::list<std::string> file_requests;
492 for (std::map<std::string, FileStatus*>::iterator
493 it = m_files.begin();
494 it != m_files.end(); ++it) {
495 if (!it->second->received)
496 file_requests.push_back(it->first);
498 assert((s32) file_requests.size() ==
499 m_uncached_count - m_uncached_received_count);
500 client->request_media(file_requests);
504 void ClientMediaDownloader::conventionalTransferDone(
505 const std::string &name,
506 const std::string &data,
509 // Check that file was announced
510 std::map<std::string, FileStatus*>::iterator
511 file_iter = m_files.find(name);
512 if (file_iter == m_files.end()) {
513 errorstream << "Client: server sent media file that was"
514 << "not announced, ignoring it: \"" << name << "\""
518 FileStatus *filestatus = file_iter->second;
519 assert(filestatus != NULL);
521 // Check that file hasn't already been received
522 if (filestatus->received) {
523 errorstream << "Client: server sent media file that we already"
524 << "received, ignoring it: \"" << name << "\""
529 // Mark file as received, regardless of whether loading it works and
530 // whether the checksum matches (because at this point there is no
531 // other server that could send a replacement)
532 filestatus->received = true;
533 assert(m_uncached_received_count < m_uncached_count);
534 m_uncached_received_count++;
536 // Check that received file matches announced checksum
538 checkAndLoad(name, filestatus->sha1, data, false, client);
541 bool ClientMediaDownloader::checkAndLoad(
542 const std::string &name, const std::string &sha1,
543 const std::string &data, bool is_from_cache, Client *client)
545 const char *cached_or_received = is_from_cache ? "cached" : "received";
546 const char *cached_or_received_uc = is_from_cache ? "Cached" : "Received";
547 std::string sha1_hex = hex_encode(sha1);
549 // Compute actual checksum of data
550 std::string data_sha1;
552 SHA1 data_sha1_calculator;
553 data_sha1_calculator.addBytes(data.c_str(), data.size());
554 unsigned char *data_tmpdigest = data_sha1_calculator.getDigest();
555 data_sha1.assign((char*) data_tmpdigest, 20);
556 free(data_tmpdigest);
559 // Check that received file matches announced checksum
560 if (data_sha1 != sha1) {
561 std::string data_sha1_hex = hex_encode(data_sha1);
562 infostream << "Client: "
563 << cached_or_received_uc << " media file "
564 << sha1_hex << " \"" << name << "\" "
565 << "mismatches actual checksum " << data_sha1_hex
570 // Checksum is ok, try loading the file
571 bool success = client->loadMedia(data, name);
573 infostream << "Client: "
574 << "Failed to load " << cached_or_received << " media: "
575 << sha1_hex << " \"" << name << "\""
580 verbosestream << "Client: "
581 << "Loaded " << cached_or_received << " media: "
582 << sha1_hex << " \"" << name << "\""
585 // Update cache (unless we just loaded the file from the cache)
587 m_media_cache.update(sha1_hex, data);
594 Minetest Hashset File Format
596 All values are stored in big-endian byte order.
597 [u32] signature: 'MTHS'
599 For each hash in set:
606 std::string ClientMediaDownloader::serializeRequiredHashSet()
608 std::ostringstream os(std::ios::binary);
610 writeU32(os, MTHASHSET_FILE_SIGNATURE); // signature
611 writeU16(os, 1); // version
613 // Write list of hashes of files that have not been
614 // received (found in cache) yet
615 for (std::map<std::string, FileStatus*>::iterator
616 it = m_files.begin();
617 it != m_files.end(); ++it) {
618 if (!it->second->received) {
619 assert(it->second->sha1.size() == 20);
620 os << it->second->sha1;
627 void ClientMediaDownloader::deSerializeHashSet(const std::string &data,
628 std::set<std::string> &result)
630 if (data.size() < 6 || data.size() % 20 != 6) {
631 throw SerializationError(
632 "ClientMediaDownloader::deSerializeHashSet: "
633 "invalid hash set file size");
636 const u8 *data_cstr = (const u8*) data.c_str();
638 u32 signature = readU32(&data_cstr[0]);
639 if (signature != MTHASHSET_FILE_SIGNATURE) {
640 throw SerializationError(
641 "ClientMediaDownloader::deSerializeHashSet: "
642 "invalid hash set file signature");
645 u16 version = readU16(&data_cstr[4]);
647 throw SerializationError(
648 "ClientMediaDownloader::deSerializeHashSet: "
649 "unsupported hash set file version");
652 for (u32 pos = 6; pos < data.size(); pos += 20) {
653 result.insert(data.substr(pos, 20));