From: Bryan Duncan Date: Mon, 14 Nov 2011 19:34:12 +0000 (-0800) Subject: ITS #6300: Added support for using kqueue in slapd (for systems that support kqueue(2). X-Git-Url: https://git.sur5r.net/?p=openldap;a=commitdiff_plain;h=a25046535cc4f57f53b0e6952ab478593b886be6 ITS #6300: Added support for using kqueue in slapd (for systems that support kqueue(2). Patch obtained from: http://public.me.com/bryan.duncan/bryan-duncan.kqueue.090922.patch # This patch file is derived from OpenLDAP Software. All of the # modifications to OpenLDAP Software represented in the following # patch(es) were developed by Apple. I, Bryan Duncan, am authorized by # Apple, my employer, to release this work under the following terms. # # Copyright 2009 Apple Inc. All rights reserved. # # Redistribution and use in source and binary forms, with or without # modification, are permitted only as authorized by the OpenLDAP # Public License. # # A copy of this license is available in file LICENSE in the # top-level directory of the distribution or, alternatively, at # http://www.OpenLDAP.org/license.html. # --- diff --git a/configure.in b/configure.in index 24ff569cb2..b1fa415cc5 100644 --- a/configure.in +++ b/configure.in @@ -995,6 +995,18 @@ if test "${ac_cv_header_sys_epoll_h}" = yes; then AC_DEFINE(HAVE_EPOLL,1, [define if your system supports epoll])],[AC_MSG_RESULT(no)],[AC_MSG_RESULT(no)]) fi +dnl ---------------------------------------------------------------- +AC_CHECK_HEADERS( sys/event.h ) +if test "${ac_cv_header_sys_event_h}" = yes; then +AC_MSG_CHECKING(for kqueue system call) +AC_RUN_IFELSE([AC_LANG_SOURCE([[int main(int argc, char *argv) +{ + int kqfd = kqueue(); + exit (kqfd == -1 ? 1 : 0); +}]])],[AC_MSG_RESULT(yes) +AC_DEFINE(HAVE_KQUEUE,1, [define if your system supports kqueue])],[AC_MSG_RESULT(no)],[AC_MSG_RESULT(no)]) +fi + dnl ---------------------------------------------------------------- AC_CHECK_HEADERS( sys/devpoll.h ) dnl "/dev/poll" needs as well... diff --git a/include/portable.hin b/include/portable.hin index d6a1230a2b..3b17ccf046 100644 --- a/include/portable.hin +++ b/include/portable.hin @@ -283,6 +283,9 @@ /* Define to 1 if you have the header file. */ #undef HAVE_IO_H +/* define if your system supports kqueue */ +#undef HAVE_KQUEUE + /* Define to 1 if you have the `gen' library (-lgen). */ #undef HAVE_LIBGEN @@ -668,6 +671,9 @@ /* Define to 1 if you have the header file. */ #undef HAVE_SYS_ERRNO_H +/* Define to 1 if you have the header file. */ +#undef HAVE_SYS_EVENT_H + /* Define to 1 if you have the header file. */ #undef HAVE_SYS_FILE_H diff --git a/servers/slapd/daemon.c b/servers/slapd/daemon.c index 1464302c6d..2e5f2f3d2c 100644 --- a/servers/slapd/daemon.c +++ b/servers/slapd/daemon.c @@ -45,7 +45,11 @@ #include #endif -#if defined(HAVE_SYS_EPOLL_H) && defined(HAVE_EPOLL) +#if defined(HAVE_KQUEUE) +# include +# include +# include +#elif defined(HAVE_SYS_EPOLL_H) && defined(HAVE_EPOLL) # include #elif defined(SLAP_X_DEVPOLL) && defined(HAVE_SYS_DEVPOLL_H) && defined(HAVE_DEVPOLL) # include @@ -93,7 +97,7 @@ static volatile sig_atomic_t listening = 1; /* 0 when slap_listeners closed */ static ldap_pvt_thread_t *listener_tid; #ifndef SLAPD_LISTEN_BACKLOG -#define SLAPD_LISTEN_BACKLOG 1024 +#define SLAPD_LISTEN_BACKLOG 2048 #endif /* ! SLAPD_LISTEN_BACKLOG */ #define DAEMON_ID(fd) (fd & slapd_daemon_mask) @@ -140,7 +144,21 @@ typedef struct slap_daemon_st { int sd_nwriters; int sd_nfds; -#if defined(HAVE_EPOLL) +#if defined(HAVE_KQUEUE) + uint8_t* sd_fdmodes; /* indexed by fd */ + Listener** sd_l; /* indexed by fd */ + /* Double buffer the kqueue changes to avoid holding the sd_mutex \ + * during a kevent() call. \ + */ + struct kq_change { + struct kevent* sd_changes; + int sd_nchanges; + int sd_maxchanges; + } sd_kqc[2]; + int sd_changeidx; /* index to current change buffer */ + int sd_kq; +#elif defined(HAVE_EPOLL) + struct epoll_event *sd_epolls; int *sd_index; int sd_epfd; @@ -171,10 +189,219 @@ static slap_daemon_st slap_daemon[SLAPD_MAX_DAEMON_THREADS]; * with file descriptors and events respectively * * - SLAP__* for private interface; type by now is one of - * EPOLL, DEVPOLL, SELECT + * EPOLL, DEVPOLL, SELECT, KQUEUE * * private interface should not be used in the code. */ +#if defined(HAVE_KQUEUE) +# define SLAP_EVENT_FNAME "kqueue" +# define SLAP_EVENTS_ARE_INDEXED 0 +# define SLAP_EVENT_MAX (2 * dtblsize) /* each fd can have a read & a write event */ + +# define SLAP_EVENT_DECL \ + static struct kevent* events = NULL + +# define SLAP_EVENT_INIT do {\ + if (!events) { \ + events = ch_malloc(sizeof(*events) * SLAP_EVENT_MAX); \ + if (!events) { \ + Debug(LDAP_DEBUG_ANY, \ + "daemon: SLAP_EVENT_INIT: ch_malloc of events failed, wanted %d bytes\n", \ + sizeof(*events) * SLAP_EVENT_MAX, 0, 0); \ + slapd_shutdown = 2; \ + } \ + } \ +} while (0) + +# define SLAP_SOCK_INIT do { \ + int kq_i; \ + size_t kq_nbytes; \ + Debug(LDAP_DEBUG_ANY, "daemon: SLAP_SOCK_INIT: dtblsize=%d\n", dtblsize, 0, 0); \ + slap_daemon.sd_nfds = 0; \ + slap_daemon.sd_changeidx = 0; \ + for (kq_i = 0; kq_i < 2; kq_i++) { \ + struct kq_change* kqc = &slap_daemon.sd_kqc[kq_i]; \ + kqc->sd_nchanges = 0; \ + kqc->sd_maxchanges = 256; /* will grow as needed */ \ + kq_nbytes = sizeof(*kqc->sd_changes) * kqc->sd_maxchanges; \ + kqc->sd_changes = ch_calloc(1, kq_nbytes); \ + if (!kqc->sd_changes) { \ + Debug(LDAP_DEBUG_ANY, \ + "daemon: SLAP_SOCK_INIT: ch_calloc of slap_daemon.sd_changes[%d] failed, wanted %d bytes, shutting down\n", \ + kq_i, kq_nbytes, 0); \ + slapd_shutdown = 2; \ + } \ + } \ + kq_nbytes = sizeof(*slap_daemon.sd_fdmodes) * dtblsize; \ + slap_daemon.sd_fdmodes = ch_calloc(1, kq_nbytes); \ + if (!slap_daemon.sd_fdmodes) { \ + Debug(LDAP_DEBUG_ANY, \ + "daemon: SLAP_SOCK_INIT: ch_calloc of slap_daemon.sd_fdmodes failed, wanted %d bytes, shutting down\n", \ + kq_nbytes, 0, 0); \ + slapd_shutdown = 2; \ + } \ + kq_nbytes = sizeof(*slap_daemon.sd_l) * dtblsize; \ + slap_daemon.sd_l = ch_calloc(1, kq_nbytes); \ + if (!slap_daemon.sd_l) { \ + Debug(LDAP_DEBUG_ANY, \ + "daemon: SLAP_SOCK_INIT: ch_calloc of slap_daemon.sd_l failed, wanted %d bytes, shutting down\n", \ + kq_nbytes, 0, 0); \ + slapd_shutdown = 2; \ + } \ + slap_daemon.sd_kq = kqueue(); \ + if (slap_daemon.sd_kq < 0) { \ + Debug(LDAP_DEBUG_ANY, "daemon: SLAP_SOCK_INIT: kqueue() failed, errno=%d, shutting down\n", errno, 0, 0); \ + slapd_shutdown = 2; \ + } \ +} while (0) + +# define SLAP_SOCK_DESTROY do { \ + int kq_i; \ + if (slap_daemon.sd_kq > 0) { \ + close(slap_daemon.sd_kq); \ + slap_daemon.sd_kq = -1; \ + } \ + for (kq_i = 0; kq_i < 2; kq_i++) { \ + if (slap_daemon.sd_kqc[kq_i].sd_changes != NULL) { \ + ch_free(slap_daemon.sd_kqc[kq_i].sd_changes); \ + slap_daemon.sd_kqc[kq_i].sd_changes = NULL; \ + } \ + slap_daemon.sd_kqc[kq_i].sd_nchanges = 0; \ + slap_daemon.sd_kqc[kq_i].sd_maxchanges = 0; \ + } \ + if (slap_daemon.sd_l != NULL) { \ + ch_free(slap_daemon.sd_l); \ + slap_daemon.sd_l = NULL; \ + } \ + if (slap_daemon.sd_fdmodes != NULL) { \ + ch_free(slap_daemon.sd_fdmodes); \ + slap_daemon.sd_fdmodes = NULL; \ + } \ + slap_daemon.sd_nfds = 0; \ +} while (0) + +# define SLAP_KQUEUE_SOCK_ACTIVE 0x01 +# define SLAP_KQUEUE_SOCK_READ_ENABLED 0x02 +# define SLAP_KQUEUE_SOCK_WRITE_ENABLED 0x04 + +# define SLAP_SOCK_IS_ACTIVE(s) (slap_daemon.sd_fdmodes[(s)] != 0) +# define SLAP_SOCK_NOT_ACTIVE(s) (slap_daemon.sd_fdmodes[(s)] == 0) +# define SLAP_SOCK_IS_READ(s) (slap_daemon.sd_fdmodes[(s)] & SLAP_KQUEUE_SOCK_READ_ENABLED) +# define SLAP_SOCK_IS_WRITE(s) (slap_daemon.sd_fdmodes[(s)] & SLAP_KQUEUE_SOCK_WRITE_ENABLED) + +/* + * SLAP_SOCK_SET_* & SLAP_SOCK_CLR_* get called a _lot_. Since kevent() + * processes changes before it looks for events, batch up the changes which + * will get submitted the next time kevent() is called for events. + */ + +# define SLAP_KQUEUE_CHANGE(s, filter, flag) do { \ + /* If maxchanges is reached, have to realloc to make room for more. \ + * Ideally we'd call kevent(), but the daemon thread could be sitting \ + * in kevent() waiting for events. \ + */ \ + struct kq_change* kqc = &slap_daemon.sd_kqc[slap_daemon.sd_changeidx]; \ + if (kqc->sd_nchanges == kqc->sd_maxchanges) { \ + /* Don't want to do this very often. Double the size. */ \ + size_t kq_nbytes; \ + Debug(LDAP_DEBUG_CONNS, \ + "daemon: SLAP_KQUEUE_CHANGE: increasing slap_daemon.sd_kqc[%d].maxchanges from %d to %d\n", \ + slap_daemon.sd_changeidx, kqc->sd_maxchanges, 2*kqc->sd_maxchanges); \ + kqc->sd_maxchanges += kqc->sd_maxchanges; \ + kq_nbytes = sizeof(*kqc->sd_changes) * kqc->sd_maxchanges; \ + kqc->sd_changes = ch_realloc(kqc->sd_changes, kq_nbytes); \ + if (!kqc->sd_changes) { \ + Debug(LDAP_DEBUG_ANY, \ + "daemon: SLAP_KQUEUE_CHANGE: ch_realloc of slap_daemon.sd_kqc[%d].sd_changes failed, wanted %d bytes, shutting down\n", \ + slap_daemon.sd_changeidx, kq_nbytes, 0); \ + slapd_shutdown = 2; \ + break; /* Don't want to do the EV_SET if sd_changes is NULL */ \ + } \ + } \ + EV_SET(&kqc->sd_changes[kqc->sd_nchanges++], \ + (s), (filter), (flag), 0, 0, slap_daemon.sd_l[(s)]); \ +} while (0) + +# define SLAP_KQUEUE_SOCK_SET(s, filter, mode) do { \ + if ((slap_daemon.sd_fdmodes[(s)] & (mode)) != (mode)) { \ + slap_daemon.sd_fdmodes[(s)] |= (mode); \ + SLAP_KQUEUE_CHANGE((s), (filter), EV_ENABLE); \ + } \ +} while (0) + +# define SLAP_KQUEUE_SOCK_CLR(s, filter, mode) do { \ + if (slap_daemon.sd_fdmodes[(s)] & (mode)) { \ + slap_daemon.sd_fdmodes[(s)] &= ~(mode); \ + SLAP_KQUEUE_CHANGE((s), (filter), EV_DISABLE); \ + } \ +} while (0) + +# define SLAP_SOCK_SET_READ(s) SLAP_KQUEUE_SOCK_SET((s), EVFILT_READ, SLAP_KQUEUE_SOCK_READ_ENABLED) +# define SLAP_SOCK_SET_WRITE(s) SLAP_KQUEUE_SOCK_SET((s), EVFILT_WRITE, SLAP_KQUEUE_SOCK_WRITE_ENABLED) +# define SLAP_SOCK_CLR_READ(s) SLAP_KQUEUE_SOCK_CLR((s), EVFILT_READ, SLAP_KQUEUE_SOCK_READ_ENABLED) +# define SLAP_SOCK_CLR_WRITE(s) SLAP_KQUEUE_SOCK_CLR((s), EVFILT_WRITE, SLAP_KQUEUE_SOCK_WRITE_ENABLED) + +/* kqueue doesn't need to do anything to clear the event. */ +# define SLAP_EVENT_CLR_READ(i) do {} while (0) +# define SLAP_EVENT_CLR_WRITE(i) do {} while (0) + +# define SLAP_SOCK_ADD(s, l) do { \ + assert( s < dtblsize ); \ + slap_daemon.sd_l[(s)] = (l); \ + slap_daemon.sd_fdmodes[(s)] = SLAP_KQUEUE_SOCK_ACTIVE | SLAP_KQUEUE_SOCK_READ_ENABLED; \ + ++slap_daemon.sd_nfds; \ + SLAP_KQUEUE_CHANGE((s), EVFILT_READ, EV_ADD); \ + SLAP_KQUEUE_CHANGE((s), EVFILT_WRITE, EV_ADD | EV_DISABLE); \ +} while (0) + +# define SLAP_SOCK_DEL(s) do { \ + SLAP_KQUEUE_CHANGE((s), EVFILT_READ, EV_DELETE); \ + SLAP_KQUEUE_CHANGE((s), EVFILT_WRITE, EV_DELETE); \ + slap_daemon.sd_l[(s)] = NULL; \ + slap_daemon.sd_fdmodes[(s)] = 0; \ + --slap_daemon.sd_nfds; \ +} while (0) + +# define SLAP_EVENT_IS_READ(i) \ + (events[(i)].filter == EVFILT_READ && SLAP_SOCK_IS_READ(SLAP_EVENT_FD(i))) + +# define SLAP_EVENT_IS_WRITE(i) \ + (events[(i)].filter == EVFILT_WRITE && SLAP_SOCK_IS_WRITE(SLAP_EVENT_FD(i))) + +# define SLAP_EVENT_IS_LISTENER(i) \ + (events[(i)].udata && SLAP_SOCK_IS_READ(SLAP_EVENT_FD(i))) + +# define SLAP_EVENT_LISTENER(i) ((Listener*)(events[(i)].udata)) +# define SLAP_EVENT_FD(i) (events[(i)].ident) + +# define SLAP_EVENT_WAIT(tvp, nsp) do { \ + struct timespec kq_ts; \ + struct timespec* kq_tsp; \ + int kq_idx; \ + if (tvp) { \ + TIMEVAL_TO_TIMESPEC((tvp), &kq_ts); \ + kq_tsp = &kq_ts; \ + } else { \ + kq_tsp = NULL; \ + } \ + /* Save the change buffer index for use when the mutex is unlocked, \ + * then switch the index so new changes go to the other buffer. \ + */ \ + ldap_pvt_thread_mutex_lock( &slap_daemon.sd_mutex ); \ + kq_idx = slap_daemon.sd_changeidx; \ + slap_daemon.sd_changeidx ^= 1; \ + ldap_pvt_thread_mutex_unlock( &slap_daemon.sd_mutex ); \ + *(nsp) = kevent(slap_daemon.sd_kq, \ + slap_daemon.sd_kqc[kq_idx].sd_nchanges \ + ? slap_daemon.sd_kqc[kq_idx].sd_changes : NULL, \ + slap_daemon.sd_kqc[kq_idx].sd_nchanges, \ + events, SLAP_EVENT_MAX, kq_tsp); \ + slap_daemon.sd_kqc[kq_idx].sd_nchanges = 0; \ +} while(0) + +/*-------------------------------------------------------------------------------*/ + +#elif defined(HAVE_EPOLL) #if defined(HAVE_EPOLL) /*************************************** * Use epoll infrastructure - epoll(4) * @@ -2821,6 +3048,10 @@ loop: connections_shutdown(); } +#ifdef HAVE_KQUEUE + close( slap_daemon.sd_kq ); +#endif + if ( LogTest( LDAP_DEBUG_ANY )) { int t = ldap_pvt_thread_pool_backload( &connection_pool ); Debug( LDAP_DEBUG_ANY,