summaryrefslogtreecommitdiff
path: root/trackerlogic.h
AgeCommit message (Collapse)Author
2024-04-01Move further towards dual stack capable opentracker, name more constantsDirk Engling
2024-04-01Tidy up v6 handlerDirk Engling
2024-03-30Limit ipv6 udp replies to an amount that does not create too large UDP packets. Credits to anonymous donorDirk Engling
2022-11-24Add support for dynamic accesslistsDirk Engling
2021-04-20Use arc4random whereever we need strong entropyDirk Engling
2021-04-20Turn random() to nrand48 in inner loop where it is not cryptographically relevant to avoid futex storm. Thanks and credits to Bart SmienkDirk Engling
2017-04-28Get rid of some warnings regarding header guardsDirk Engling
2015-06-08Increase number of locks to accomodate for more cores on contemporary CPUsDirk Engling
2010-04-22** struct ot_workstruct gets ritcher (and will become even ritcher soon).erdgeist
This is where we encapsulate all per-request data from peer to hash to peer_id, so that it is available everywhere without passing hundreds of pointers down the stack. Most functions that do work down the stack now accept an ot_workstruct and some flags. So it can end up in the stats/event-handler where it will be the default parameter in the future. ** peer_id is now being copied by default and moved to ot_workstruct So it is available in stats and subsequent functions. ** sync scrape madness is gone SYNC_SCRAPE was intended to sync tracker state that would normally be lost on restarts i.e. downloaded counts per torrent. The way was to push it in the tracker cloud after finding all neighbouring trackers. This is madness. It never was tested and can be done per tracker by fetching stats/mode=statedump from time to time and starting opentracker with the -l option later. ** livesync thread has its own ot_workstruct now So it can behave like ot_udp and ot_http against trackerlogic.c and get rid of the first half of the embarrassing global variables. The sending half will be fixed soon [tm]. ** stats can log completed events The author recognizes the needs of original content distributors to keep track of the amount of times a work has been downloaded. While not feasible and used on openbittorrent and other open and anonymous tracker installations, a tracker user can now choose to send those events to syslog.
2009-11-18Preparing more efficient header parsingerdgeist
2009-08-29Enable limiting fullscrapes to only every 5 minutes per IPerdgeist
2009-08-26Introduce the concept of an ot_net into opentracker.erdgeist
2009-06-15Log networks in its old form has been disabled. Live logging may come back soon.erdgeist
2009-03-17Fix white spaceserdgeist
Introduce loading tracker states with -l Alter tracker state to a human readable form
2009-03-04Add a stat option the count bucket stallserdgeist
Add an iterator for all torrents. It's slower but for stats it's okay. Move some stats collection stuff to the new iterator. More to come. Start a "report all stats" page. Start fixing the code to identify "busy" networks. Add the concept of hosts allowed to proxy. Add a parser for the X-Forwarded-For: HTTP header. Clean up HTTP Header handling code. (Remove some left overs of now vanished sync code).
2009-03-03Remove superfluous test for == NULL, when variable has been used, anywayerdgeist
2009-01-16Use a self pipe to wake from io_wait when a thread finished his work.erdgeist
2009-01-15We sent an incorrect reply dictionary vor v6 announceserdgeist
2009-01-13V6erdgeist
2009-01-05* http and udp routines now use thread local buffers passed in workstruct containers. In other words they do not use static_buffer anymore and are ↵erdgeist
considered to be thread safe. * the new workstruct also introduces a well defined buffer and result passing path * a new function scan_find_keywords is a wrapper around scan_urlencoded_query that maps keys in url to values passed in an array of ot_keywords structs * this new function cleans up much of url parameter parsing work, where read_ptr and write_ptr have been introduced rather than the confusing char *c, *data variables * I now use memcmp instead of byte_diff to allow compiler to optimize constant size string compares * got rid of UTORRENT_1600_WORKAROUND * livesync_ticker is now only called from one (currently main) thread to avoid race conditions
2009-01-02* opentracker now drops permissions in correct order and really chroots() if ran as rooterdgeist
* lock passing between add_peer_to_torrent and return_peers_for_torrent is now avoided by providing a more general add_peer_to_torrent_and_return_peers function that can be used with NULL parameters to not return any peers (in sync case) * in order to keep a fast overview how many torrents opentracker maintains, every mutex_bucket_unlock operation expects an additional integer parameter that tells ot_mutex.c how many torrents have been added or removed. A function mutex_get_torrent_count has been introduced.
2008-12-08Make all torrents in their buckets sorted again.erdgeist
2008-12-07Handle program end more politelyerdgeist
2008-12-06Renamed OT_FLAG to OT_PEERFLAG to make code easier to readerdgeist
Introduced READ16/32 and WRITE16/32 makros to abstract loading/storing from unaligned addresses away on cpu's that can actually load/store everywhere Removed all unnecessary memmoves, especially where it only moved 6 bytes in inner loop. I replaced them with WRITE16/32(READ16/32()) makros
2008-12-03New peer flag to see, whether peer came from announce or syncerdgeist
2008-12-01Live Sync for peers that do not come back too early.erdgeist
2008-11-28The BIG refactoring [tm]. Too many changes to count them. If it doesn't suite you, revert to last version.erdgeist
2008-10-28Whitespace fixeserdgeist
2008-10-06Make ot_try_bind local again.erdgeist
2008-10-04added live sync codeerdgeist
added a config file parser added tracker id changed WANT_CLOSED_TRACKER and WANT_BLACKLIST into WANT_ACCESS_WHITE and WANT_ACCESS_BLACK changed WANT_TRACKER_SYNC to WANT_SYNC_BATCH and added WANT_SYNC_LIVE added an option to switch off fullscrapes cleaned up many internal hardcoded values, like PROTO_FLAG,
2007-12-20Introduce some kind of versioningerdgeist
2007-12-03Drop ot_{byte,word,dword} and use uint{8,16,32}_t, also simplify includeserdgeist
2007-11-21Fixing mmap allocation went to ot_mutex.c. Clean all torrents moved to ot_clean.c. Also make static buffers staticerdgeist
2007-11-12Moving accesslist stuff to its own objecterdgeist
2007-11-12Full scrape moves to its own objecterdgeist
2007-11-06Move stats out of opentracker.c, also have an own file handle udp requests.erdgeist
2007-11-06No one can get access to buckets now without locking them. Also split up the trackerlogic.c-monster in functional sub-units. HEADS UP: this code is untested ↵erdgeist
and not considered stable.
2007-11-06Make to_hex thread safe. Get rid off old /24 counting code. Protect more bucket accesses by locks.erdgeist
2007-11-06We do now use the tools for locking buckets in order to prepare MT codeerdgeist
2007-11-03Introducing first tools to make opentracker multithreaded.erdgeist
2007-11-02Introduce configuration dependent number of parameters in a nicer wayerdgeist
2007-11-01Allow for more than one admin ip addresserdgeist
2007-10-31Add more full scrape statserdgeist
2007-10-29Reenabled syncingerdgeist
2007-10-27Number of buckets is 1024 nowerdgeist
Clean all torrents now only cleans one bucket at a time All torrents that are being worked upon in an announce are being cleaned on demoand torrent's peer lists now keep extra counts for seeds and peers to speed up scrape and announce Sync has gone for now. I will think up a new way to implement. The old one was way to slow.
2007-10-19Introducing multiscrapeerdgeist
2007-10-19After studying opentracker live data, I decided to radically reduce default allocation for vectors and making it grow faster instead: Most pools only had ↵erdgeist
one or two peers, wasting 8*15 or 8*14 bytes.
2007-10-19Increase check interval for timeouted sockets to 10 seconds. That may prevent some disrupted scrapes.erdgeist
2007-10-17Increase send timeout. Also use an own variable for that.erdgeist
2007-10-13since gettimeofday is rather expansive, we do only fetch time once in a 5 second period, when we are delivered a SIGALRM.erdgeist