/* * $Id$ * * DEBUG: section 05 Socket Functions * * SQUID Web Proxy Cache http://www.squid-cache.org/ * ---------------------------------------------------------- * * Squid is the result of efforts by numerous individuals from * the Internet community; see the CONTRIBUTORS file for full * details. Many organizations have provided support for Squid's * development; see the SPONSORS file for full details. Squid is * Copyrighted (C) 2001 by the Regents of the University of * California; see the COPYRIGHT file for full details. Squid * incorporates software developed and/or copyrighted by other * sources; see the CREDITS file for full details. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA. * */ /* * This is a very simple driver for Solaris /dev/poll. * * The updates are batched, one trip through the comm loop. * (like libevent.) We keep a pointer into the structs so we * can zero out an entry in the poll list if its active. * * Ported by Peter Payne from Squid 2.7.STABLE9 comm_devpoll.c * on August 11, 2010 at 3pm (GMT+0100 Europe/London). * * Last modified 2010-10-08 */ #include "squid.h" #include "CacheManager.h" #include "Store.h" #include "fde.h" #include "SquidTime.h" #ifdef USE_DEVPOLL #if HAVE_SYS_DEVPOLL_H #include /* Solaris /dev/poll support */ #endif #define DEBUG_DEVPOLL 0 /* OPEN_MAX is defined in , presumably included by sys/devpoll.h */ #define DEVPOLL_UPDATESIZE OPEN_MAX #define DEVPOLL_QUERYSIZE OPEN_MAX /* XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX */ /* define structures */ /* Current state */ struct _devpoll_state { char state; int update_offset; }; /* The update list */ struct { struct pollfd *pfds; int cur; int size; } devpoll_update; /* XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX */ /* static variables */ static int devpoll_fd; static struct timespec zero_timespec; static int max_poll_time = 1000; static struct _devpoll_state *devpoll_state; static struct dvpoll do_poll; static int dpoll_nfds; /* XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX */ /* prototypes */ static void commDevPollRegisterWithCacheManager(void); /* XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX */ /* private functions */ static void comm_flush_updates(void) { int i; if (devpoll_update.cur == -1) return; debugs(5, DEBUG_DEVPOLL ? 0 : 8, "comm_flush_updates(): " << (devpoll_update.cur + 1) << " fds queued"); i = write(devpoll_fd, devpoll_update.pfds, (devpoll_update.cur + 1) * sizeof(struct pollfd)); assert(i > 0); assert(i == sizeof(struct pollfd) * (devpoll_update.cur + 1)); devpoll_update.cur = -1; } /* * We could be "optimal" and -change- an existing entry if they * just add a bit - since the devpoll interface OR's multiple fd * updates. We'll need to POLLREMOVE entries which has a bit cleared * but for now I'll do whats "easier" and add the smart logic * later. */ static void comm_update_fd(int fd, int events) { #if DEBUG_DEVPOLL char acDebugStr[ 2048 ]; acDebugStr[0] = '\0'; if ( events & POLLIN ) strlcat( acDebugStr, " IN", sizeof(acDebugStr) ); if ( events & POLLOUT ) strlcat( acDebugStr, " OUT", sizeof(acDebugStr) ); if ( events & POLLREMOVE ) strlcat( acDebugStr, " REMOVE", sizeof(acDebugStr) ); if ( events & POLLERR ) strlcat( acDebugStr, " ERR", sizeof(acDebugStr) ); if ( events & POLLHUP ) strlcat( acDebugStr, " HUP", sizeof(acDebugStr) ); if ( events & POLLNVAL ) strlcat( acDebugStr, " NVAL", sizeof(acDebugStr) ); debugs(5, DEBUG_DEVPOLL ? 0 : 8, "comm_update_fd(FD " << fd << ", events=[" << acDebugStr << " ])"); #endif /* DEBUG_DEVPOLL */ if (devpoll_update.cur != -1 && (devpoll_update.cur == devpoll_update.size)) comm_flush_updates(); devpoll_update.cur++; #if DEBUG_DEVPOLL debugs(5, DEBUG_DEVPOLL ? 0 : 8, "comm_update_fd: -> new slot (" << devpoll_update.cur << ")"); #endif /* DEBUG_DEVPOLL */ devpoll_state[fd].update_offset = devpoll_update.cur; devpoll_update.pfds[devpoll_update.cur].fd = fd; devpoll_update.pfds[devpoll_update.cur].events = events; devpoll_update.pfds[devpoll_update.cur].revents = 0; } static void commIncomingStats(StoreEntry *sentry) { StatCounters *f = &statCounter; storeAppendPrintf(sentry, "Total number of devpoll loops: %ld\n", statCounter.select_loops); storeAppendPrintf(sentry, "Histogram of returned filedescriptors\n"); statHistDump(&f->select_fds_hist, sentry, statHistIntDumper); } static void commDevPollRegisterWithCacheManager(void) { CacheManager::GetInstance()-> registerAction("comm_devpoll_incoming", "comm_incoming() stats", commIncomingStats, 0, 1); } /* XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX */ /* public functions */ /* * comm_select_init * * This is a needed exported function which will be called to initialise * the network loop code. */ void comm_select_init(void) { /* set up static structure */ zero_timespec.tv_sec = 0; zero_timespec.tv_nsec = 0; /* allocate memory first before attempting to open poll device */ /* This tracks the FD devpoll offset+state */ devpoll_state = (struct _devpoll_state *)xcalloc( SQUID_MAXFD, sizeof(struct _devpoll_state) ); /* And this is the stuff we use to read events */ do_poll.dp_fds = (struct pollfd *)xcalloc( DEVPOLL_QUERYSIZE, sizeof(struct pollfd) ); dpoll_nfds = DEVPOLL_QUERYSIZE; devpoll_update.pfds = (struct pollfd *)xcalloc( DEVPOLL_UPDATESIZE, sizeof(struct pollfd) ); devpoll_update.cur = -1; devpoll_update.size = DEVPOLL_UPDATESIZE; /* attempt to open /dev/poll device */ devpoll_fd = open("/dev/poll", O_RDWR); if (devpoll_fd < 0) fatalf("comm_select_init: can't open /dev/poll: %s\n", xstrerror()); fd_open(devpoll_fd, FD_UNKNOWN, "devpoll ctl"); commDevPollRegisterWithCacheManager(); } /* * comm_setselect * * This is a needed exported function which will be called to register * and deregister interest in a pending IO state for a given FD. * * If handler=NULL then: * if type=COMM_SELECT_READ then STOP polling for input awaiting processing * if type=COMM_SELECT_WRITE then STOP polling for output buffer ready * If handler != NULL then: * if type=COMM_SELECT_READ then START polling for input awaiting processing * if type=COMM_SELECT_WRITE then START polling for output buffer ready * */ void commSetSelect(int fd, unsigned int type, PF * handler, void *client_data, time_t timeout) { assert(fd >= 0); debugs(5, DEBUG_DEVPOLL ? 0 : 8, "commSetSelect(FD " << fd << ",type=" << type << ",handler=" << handler << ",client_data=" << client_data << ",timeout=" << timeout << ")"); /* POLLIN/POLLOUT are defined in */ fde *F = &fd_table[fd]; if (!F->flags.open) { /* remove from poll set */ comm_update_fd( fd, POLLREMOVE ); devpoll_state[fd].state = 0; return; } int st_new = 0; /* new state (derive from old state and function args) */ if ( type & COMM_SELECT_READ ) { if ( handler != NULL ) { /* we want to POLLIN */ st_new |= POLLIN; } else { ; /* we want to clear POLLIN because handler is NULL */ } F->read_handler = handler; F->read_data = client_data; } else if ( devpoll_state[fd].state & POLLIN ) { /* we're not changing reading state so take from existing */ st_new |= POLLIN; } if ( type & COMM_SELECT_WRITE ) { if ( handler != NULL ) { /* we want to POLLOUT */ st_new |= POLLOUT; } else { ; /* we want to clear POLLOUT because handler is NULL */ } F->write_handler = handler; F->write_data = client_data; } else if ( devpoll_state[fd].state & POLLOUT ) { /* we're not changing writing state so take from existing */ st_new |= POLLOUT; } if ( devpoll_state[fd].state ^ st_new ) { /* something has changed, update /dev/poll of what to listen for */ comm_update_fd( fd, POLLREMOVE ); if ( st_new ) comm_update_fd( fd, st_new ); devpoll_state[fd].state = st_new; } if (timeout) F->timeout = squid_curtime + timeout; } void commResetSelect(int fd) { commSetSelect(fd, COMM_SELECT_WRITE, NULL, NULL, 0); commSetSelect(fd, COMM_SELECT_READ, NULL, NULL, 0); } /* * comm_select * Check all connections for new connections and input data that is to be * processed. Also check for connections with data queued and whether we can * write it out. * * Called to do the new-style IO, courtesy of of squid (like most of this * new IO code). This routine handles the stuff we've hidden in * comm_setselect and fd_table[] and calls callbacks for IO ready * events. */ comm_err_t comm_select(int msec) { int num, i; fde *F; PF *hdl; PROF_start(comm_check_incoming); if (msec > max_poll_time) msec = max_poll_time; for (;;) { do_poll.dp_timeout = msec; do_poll.dp_nfds = dpoll_nfds; comm_flush_updates(); num = ioctl(devpoll_fd, DP_POLL, &do_poll); ++statCounter.select_loops; if (num >= 0) break; if (ignoreErrno(errno)) break; getCurrentTime(); PROF_stop(comm_check_incoming); return COMM_ERROR; } PROF_stop(comm_check_incoming); getCurrentTime(); statHistCount(&statCounter.select_fds_hist, num); if (num == 0) return COMM_TIMEOUT; /* No error.. */ PROF_start(comm_handle_ready_fd); for (i = 0; i < num; i++) { int fd = (int)do_poll.dp_fds[i].fd; F = &fd_table[fd]; debugs(5, DEBUG_DEVPOLL ? 0 : 8, "comm_select(): got FD " << fd << " events=" << std::hex << do_poll.dp_fds[i].revents << " monitoring=" << devpoll_state[fd].state << " F->read_handler=" << F->read_handler << " F->write_handler=" << F->write_handler); /* handle errors */ if (do_poll.dp_fds[i].revents & (POLLERR | POLLHUP | POLLNVAL)) { debugs(5, DEBUG_DEVPOLL ? 0 : 8, "comm_select(): devpoll event error: fd " << fd); continue; /* XXX! */ } if (do_poll.dp_fds[i].revents & POLLIN || F->flags.read_pending) { if ( (hdl = F->read_handler) != NULL ) { debugs(5, DEBUG_DEVPOLL ? 0 : 8, "comm_select(): Calling read handler on FD " << fd); PROF_start(comm_read_handler); F->flags.read_pending = 0; F->read_handler = NULL; hdl(fd, F->read_data); PROF_stop(comm_read_handler); statCounter.select_fds++; } else { debugs(5, DEBUG_DEVPOLL ? 0 : 8, "comm_select(): no read handler for FD " << fd); // remove interest since no handler exist for this event. commSetSelect(fd, COMM_SELECT_READ, NULL, NULL, 0); } } if (do_poll.dp_fds[i].revents & POLLOUT) { if ((hdl = F->write_handler) != NULL) { debugs(5, DEBUG_DEVPOLL ? 0 : 8, "comm_select(): Calling write handler on FD " << fd); PROF_start(comm_write_handler); F->write_handler = NULL; hdl(fd, F->write_data); PROF_stop(comm_write_handler); statCounter.select_fds++; } else { debugs(5, DEBUG_DEVPOLL ? 0 : 8, "comm_select(): no write handler for FD " << fd); // remove interest since no handler exist for this event. commSetSelect(fd, COMM_SELECT_WRITE, NULL, NULL, 0); } } } PROF_stop(comm_handle_ready_fd); return COMM_OK; } void comm_quick_poll_required(void) { max_poll_time = 10; } #endif /* USE_DEVPOLL */