diff options
author | brain <brain@e03df62e-2008-0410-955e-edbf42e46eb7> | 2005-12-12 18:07:53 +0000 |
---|---|---|
committer | brain <brain@e03df62e-2008-0410-955e-edbf42e46eb7> | 2005-12-12 18:07:53 +0000 |
commit | 6997c6a3f0d5d858a2be779195953973a32e60bf (patch) | |
tree | 64f0c341b715048b48d89bfad42b13847709bc80 /src/inspircd.cpp | |
parent | 9ffaeadb391064fe49a794cc54b932860e0f99d6 (diff) |
Comment and tidyup
git-svn-id: http://svn.inspircd.org/repository/trunk/inspircd@2351 e03df62e-2008-0410-955e-edbf42e46eb7
Diffstat (limited to 'src/inspircd.cpp')
-rw-r--r-- | src/inspircd.cpp | 183 |
1 files changed, 122 insertions, 61 deletions
diff --git a/src/inspircd.cpp b/src/inspircd.cpp index b59a0c53c..feea64699 100644 --- a/src/inspircd.cpp +++ b/src/inspircd.cpp @@ -2434,6 +2434,16 @@ void ProcessUser(userrec* cu) } } +/** + * This function is called once a second from the mainloop. + * It is intended to do background checking on all the user structs, e.g. + * stuff like ping checks, registration timeouts, etc. + * The function returns false when it is finished, and true if + * it needs to be run again (e.g. it has processed one of a batch of + * QUIT messages, but couldnt continue iterating because the iterator + * became invalid). This function is also responsible for checking + * if InspSocket derived classes are timed out. + */ bool DoBackgroundUserStuff(time_t TIME) { unsigned int numsockets = module_sockets.size(); @@ -2451,8 +2461,11 @@ bool DoBackgroundUserStuff(time_t TIME) } if (module_sockets.size() != numsockets) break; } + /* TODO: We need a seperate hash containing only local users for this + */ for (user_hash::iterator count2 = clientlist.begin(); count2 != clientlist.end(); count2++) { + /* Sanity checks for corrupted iterators (yes, really) */ userrec* curr = NULL; if (count2->second) curr = count2->second; @@ -2677,14 +2690,22 @@ int InspIRCd(char** argv, int argc) for (;;) { #ifdef _POSIX_PRIORITY_SCHEDULING + /* If we can, yield a bit. Doesnt do us any harm, if we're busy we take back some timeslice later ;-) + * it just means that if we have absolutely NOTHING to do, we dont eat up all the cpu doing nothing. + */ sched_yield(); sched_yield(); #endif - // we only read time() once per iteration rather than tons of times! + /* time() seems to be a pretty expensive syscall, so avoid calling it too much. + * Once per loop iteration is pleanty. + */ OLDTIME = TIME; TIME = time(NULL); - // *FIX* Instead of closing sockets in kill_link when they receive the ERROR :blah line, we should queue - // them in a list, then reap the list every second or so. + /* Run background module timers every few seconds + * (the docs say modules shouldnt rely on accurate + * timing using this event, so we dont have to + * time this exactly). + */ if (((TIME % 8) == 0) && (!expire_run)) { expire_lines(); @@ -2695,84 +2716,124 @@ int InspIRCd(char** argv, int argc) if ((TIME % 8) == 1) expire_run = false; + /* Once a second, do the background processing */ if (TIME != OLDTIME) while (DoBackgroundUserStuff(TIME)); + /* Call the socket engine to wait on the active + * file descriptors. The socket engine has everything's + * descriptors in its list... dns, modules, users, + * servers... so its nice and easy, just one call. + */ SE->Wait(activefds); + /** + * Now process each of the fd's. For users, we have a fast + * lookup table which can find a user by file descriptor, so + * processing them by fd isnt expensive. If we have a lot of + * listening ports or module sockets though, things could get + * ugly. + */ unsigned int numberactive = activefds.size(); for (unsigned int activefd = 0; activefd < numberactive; activefd++) { int socket_type = SE->GetType(activefds[activefd]); - - if (socket_type == X_ESTAB_CLIENT) + switch (socket_type) { - log(DEBUG,"Got a ready socket of type X_ESTAB_CLIENT"); - userrec* cu = fd_ref_table[activefds[activefd]]; - if (cu) - { - /* It's a user */ - ProcessUser(cu); - } - } - else if (socket_type == X_ESTAB_MODULE) - { - log(DEBUG,"Got a ready socket of type X_ESTAB_MODULE"); - unsigned int numsockets = module_sockets.size(); - for (std::vector<InspSocket*>::iterator a = module_sockets.begin(); a < module_sockets.end(); a++) - { - InspSocket* s = (InspSocket*)*a; - if ((s) && (s->GetFd() == activefds[activefd])) + case X_ESTAB_CLIENT: + + userrec* cu = fd_ref_table[activefds[activefd]]; + if (cu) + ProcessUser(cu); + + break; + + case X_ESTAB_MODULE: + + /* Process module-owned sockets. + * Modules are encouraged to inherit their sockets from + * InspSocket so we can process them neatly like this. + */ + unsigned int numsockets = module_sockets.size(); + for (std::vector<InspSocket*>::iterator a = module_sockets.begin(); a < module_sockets.end(); a++) { - if ((s->Timeout(TIME)) || (!s->Poll())) + InspSocket* s = (InspSocket*)*a; + if ((s) && (s->GetFd() == activefds[activefd])) { - log(DEBUG,"Socket poll returned false, close and bail"); - SE->DelFd(s->GetFd()); - s->Close(); - module_sockets.erase(a); - delete s; - break; + if (!s->Poll()) + { + log(DEBUG,"Socket poll returned false, close and bail"); + SE->DelFd(s->GetFd()); + s->Close(); + module_sockets.erase(a); + delete s; + break; + } + if (module_sockets.size() != numsockets) break; } - if (module_sockets.size() != numsockets) break; } - } - } - else if (socket_type == X_ESTAB_DNS) - { - log(DEBUG,"Got a ready socket of type X_ESTAB_DNS"); + + break; + + case X_ESTAB_DNS: + + /* When we are using single-threaded dns, + * the sockets for dns end up in our mainloop. + * When we are using multi-threaded dns, + * each thread has its own basic poll() loop + * within it, making them 'fire and forget' + * and independent of the mainloop. + */ + log(DEBUG,"Got a ready socket of type X_ESTAB_DNS"); #ifndef THREADED_DNS - dns_poll(activefds[activefd]); + dns_poll(activefds[activefd]); #endif - } - else if (socket_type == X_LISTEN) - { - log(DEBUG,"Got a ready socket of type X_LISTEN"); - /* It maybe a listener */ - for (int count = 0; count < boundPortCount; count++) - { - if (activefds[activefd] == openSockfd[count]) + break; + + case X_LISTEN: + + /* It's a listener */ + for (int count = 0; count < boundPortCount; count++) { - char target[MAXBUF], resolved[MAXBUF]; - length = sizeof (client); - incomingSockfd = accept (openSockfd[count], (struct sockaddr *) &client, &length); - log(DEBUG,"Accepted socket %d",incomingSockfd); - strlcpy (target, (char *) inet_ntoa (client.sin_addr), MAXBUF); - strlcpy (resolved, target, MAXBUF); - if (incomingSockfd >= 0) - { - FOREACH_MOD OnRawSocketAccept(incomingSockfd, resolved, ports[count]); - statsAccept++; - AddClient(incomingSockfd, resolved, ports[count], false, inet_ntoa (client.sin_addr)); - log(DEBUG,"Adding client on port %lu fd=%lu",(unsigned long)ports[count],(unsigned long)incomingSockfd); - } - else + if (activefds[activefd] == openSockfd[count]) { - WriteOpers("*** WARNING: accept() failed on port %lu (%s)",(unsigned long)ports[count],target); - log(DEBUG,"accept failed: %lu",(unsigned long)ports[count]); - statsRefused++; + char target[MAXBUF]; + length = sizeof (client); + incomingSockfd = accept (openSockfd[count], (struct sockaddr *) &client, &length); + log(DEBUG,"Accepted socket %d",incomingSockfd); + strlcpy (target, (char *) inet_ntoa (client.sin_addr), MAXBUF); + /* Years and years ago, we used to resolve here + * using gethostbyaddr(). That is sucky and we + * don't do that any more... + */ + if (incomingSockfd >= 0) + { + FOREACH_MOD OnRawSocketAccept(incomingSockfd, target, ports[count]); + statsAccept++; + AddClient(incomingSockfd, target, ports[count], false, inet_ntoa (client.sin_addr)); + log(DEBUG,"Adding client on port %lu fd=%lu",(unsigned long)ports[count],(unsigned long)incomingSockfd); + } + else + { + WriteOpers("*** WARNING: accept() failed on port %lu (%s)",(unsigned long)ports[count],target); + log(DEBUG,"accept failed: %lu",(unsigned long)ports[count]); + statsRefused++; + } + /* We've found out what port it belongs on, + * no need to iterate the rest + */ + break; } } - } + break; + + default; + /* Something went wrong if we're in here. + * In fact, so wrong, im not quite sure + * what we would do, so for now, its going + * to safely do bugger all. + */ + break; } } |