]> git.ipfire.org Git - thirdparty/samba.git/commitdiff
new files for updated events system
authorAndrew Tridgell <tridge@samba.org>
Tue, 1 May 2007 21:37:58 +0000 (07:37 +1000)
committerAndrew Tridgell <tridge@samba.org>
Tue, 1 May 2007 21:37:58 +0000 (07:37 +1000)
(This used to be ctdb commit 15d8308e5a0ce04351d70ac3dd25c7698931ebba)

ctdb/lib/events/events_aio.c [new file with mode: 0644]
ctdb/lib/events/events_epoll.c [new file with mode: 0644]
ctdb/lib/events/events_select.c [new file with mode: 0644]
ctdb/lib/events/events_signal.c [new file with mode: 0644]
ctdb/lib/events/events_timed.c [new file with mode: 0644]
ctdb/lib/replace/strptime.c [new file with mode: 0644]
ctdb/lib/replace/strptime.m4 [new file with mode: 0644]
ctdb/lib/util/strlist.c [new file with mode: 0644]
ctdb/lib/util/util.c [new file with mode: 0644]

diff --git a/ctdb/lib/events/events_aio.c b/ctdb/lib/events/events_aio.c
new file mode 100644 (file)
index 0000000..9f4e9c5
--- /dev/null
@@ -0,0 +1,528 @@
+/* 
+   Unix SMB/CIFS implementation.
+
+   main select loop and event handling - aio/epoll hybrid implementation
+
+   Copyright (C) Andrew Tridgell       2006
+
+   based on events_standard.c
+   
+   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., 675 Mass Ave, Cambridge, MA 02139, USA.
+*/
+/*
+  this is a very strange beast. The Linux AIO implementation doesn't
+  yet integrate properly with epoll, but there is a kernel patch that
+  allows the aio wait primitives to be used to wait for epoll events,
+  and this can be used to give us a unified event system incorporating
+  both aio events and epoll events
+
+  this is _very_ experimental code
+*/
+
+#include "includes.h"
+#include "system/filesys.h"
+#include "lib/util/dlinklist.h"
+#include "lib/events/events.h"
+#include "lib/events/events_internal.h"
+#include <sys/epoll.h>
+#include <libaio.h>
+
+#define MAX_AIO_QUEUE_DEPTH    100
+#ifndef IOCB_CMD_EPOLL_WAIT
+#define IOCB_CMD_EPOLL_WAIT    9
+#endif
+
+struct aio_event_context {
+       /* a pointer back to the generic event_context */
+       struct event_context *ev;
+
+       /* number of registered fd event handlers */
+       int num_fd_events;
+
+       uint32_t destruction_count;
+
+       io_context_t ioctx;
+
+       struct epoll_event epevent[MAX_AIO_QUEUE_DEPTH];
+
+       struct iocb *epoll_iocb;
+
+       int epoll_fd;
+       int is_epoll_set;
+};
+
+struct aio_event {
+       struct event_context *event_ctx;
+       struct iocb iocb;
+       void *private_data;
+       event_aio_handler_t handler;
+};
+
+/*
+  map from EVENT_FD_* to EPOLLIN/EPOLLOUT
+*/
+static uint32_t epoll_map_flags(uint16_t flags)
+{
+       uint32_t ret = 0;
+       if (flags & EVENT_FD_READ) ret |= (EPOLLIN | EPOLLERR | EPOLLHUP);
+       if (flags & EVENT_FD_WRITE) ret |= (EPOLLOUT | EPOLLERR | EPOLLHUP);
+       return ret;
+}
+
+/*
+ free the epoll fd
+*/
+static int aio_ctx_destructor(struct aio_event_context *aio_ev)
+{
+       io_queue_release(aio_ev->ioctx);
+       close(aio_ev->epoll_fd);
+       aio_ev->epoll_fd = -1;
+       return 0;
+}
+
+#define EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT     (1<<0)
+#define EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR  (1<<1)
+#define EPOLL_ADDITIONAL_FD_FLAG_GOT_ERROR     (1<<2)
+
+/*
+ add the epoll event to the given fd_event
+*/
+static void epoll_add_event(struct aio_event_context *aio_ev, struct fd_event *fde)
+{
+       struct epoll_event event;
+       if (aio_ev->epoll_fd == -1) return;
+
+       fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+
+       /* if we don't want events yet, don't add an aio_event */
+       if (fde->flags == 0) return;
+
+       ZERO_STRUCT(event);
+       event.events = epoll_map_flags(fde->flags);
+       event.data.ptr = fde;
+       epoll_ctl(aio_ev->epoll_fd, EPOLL_CTL_ADD, fde->fd, &event);
+       fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT;
+
+       /* only if we want to read we want to tell the event handler about errors */
+       if (fde->flags & EVENT_FD_READ) {
+               fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+       }
+}
+
+/*
+ delete the epoll event for given fd_event
+*/
+static void epoll_del_event(struct aio_event_context *aio_ev, struct fd_event *fde)
+{
+       struct epoll_event event;
+       if (aio_ev->epoll_fd == -1) return;
+
+       fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+
+       /* if there's no aio_event, we don't need to delete it */
+       if (!(fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT)) return;
+
+       ZERO_STRUCT(event);
+       event.events = epoll_map_flags(fde->flags);
+       event.data.ptr = fde;
+       epoll_ctl(aio_ev->epoll_fd, EPOLL_CTL_DEL, fde->fd, &event);
+
+       fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT;
+}
+
+/*
+ change the epoll event to the given fd_event
+*/
+static void epoll_mod_event(struct aio_event_context *aio_ev, struct fd_event *fde)
+{
+       struct epoll_event event;
+       if (aio_ev->epoll_fd == -1) return;
+
+       fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+
+       ZERO_STRUCT(event);
+       event.events = epoll_map_flags(fde->flags);
+       event.data.ptr = fde;
+       epoll_ctl(aio_ev->epoll_fd, EPOLL_CTL_MOD, fde->fd, &event);
+
+       /* only if we want to read we want to tell the event handler about errors */
+       if (fde->flags & EVENT_FD_READ) {
+               fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+       }
+}
+
+static void epoll_change_event(struct aio_event_context *aio_ev, struct fd_event *fde)
+{
+       bool got_error = (fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_GOT_ERROR);
+       bool want_read = (fde->flags & EVENT_FD_READ);
+       bool want_write= (fde->flags & EVENT_FD_WRITE);
+
+       if (aio_ev->epoll_fd == -1) return;
+
+       fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+
+       /* there's already an event */
+       if (fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT) {
+               if (want_read || (want_write && !got_error)) {
+                       epoll_mod_event(aio_ev, fde);
+                       return;
+               }
+               epoll_del_event(aio_ev, fde);
+               return;
+       }
+
+       /* there's no aio_event attached to the fde */
+       if (want_read || (want_write && !got_error)) {
+               epoll_add_event(aio_ev, fde);
+               return;
+       }
+}
+
+static int setup_epoll_wait(struct aio_event_context *aio_ev)
+{
+       if (aio_ev->is_epoll_set) {
+               return 0;
+       }
+       memset(aio_ev->epoll_iocb, 0, sizeof(*aio_ev->epoll_iocb));
+       aio_ev->epoll_iocb->aio_fildes = aio_ev->epoll_fd;
+       aio_ev->epoll_iocb->aio_lio_opcode = IOCB_CMD_EPOLL_WAIT;
+       aio_ev->epoll_iocb->aio_reqprio = 0;
+
+       aio_ev->epoll_iocb->u.c.nbytes = MAX_AIO_QUEUE_DEPTH;
+       aio_ev->epoll_iocb->u.c.offset = -1;
+       aio_ev->epoll_iocb->u.c.buf = aio_ev->epevent;
+
+       if (io_submit(aio_ev->ioctx, 1, &aio_ev->epoll_iocb) != 1) {
+               return -1;
+       }
+       aio_ev->is_epoll_set = 1;
+
+       return 0;
+}
+
+
+/*
+  event loop handling using aio/epoll hybrid
+*/
+static int aio_event_loop(struct aio_event_context *aio_ev, struct timeval *tvalp)
+{
+       int ret, i;
+       uint32_t destruction_count = ++aio_ev->destruction_count;
+       struct timespec timeout;
+       struct io_event events[8];
+
+       if (aio_ev->epoll_fd == -1) return -1;
+
+       if (aio_ev->ev->num_signal_handlers && 
+           common_event_check_signal(aio_ev->ev)) {
+               return 0;
+       }
+
+       if (tvalp) {
+               timeout.tv_sec = tvalp->tv_sec;
+               timeout.tv_nsec = tvalp->tv_usec;
+               timeout.tv_nsec *= 1000;
+       }
+
+       if (setup_epoll_wait(aio_ev) < 0) 
+               return -1;
+
+       ret = io_getevents(aio_ev->ioctx, 1, 8,
+                          events, tvalp?&timeout:NULL);
+
+       if (ret == -EINTR) {
+               if (aio_ev->ev->num_signal_handlers) {
+                       common_event_check_signal(aio_ev->ev);
+               }
+               return 0;
+       }
+
+       if (ret == 0 && tvalp) {
+               common_event_loop_timer(aio_ev->ev);
+               return 0;
+       }
+
+       for (i=0;i<ret;i++) {
+               struct io_event *event = &events[i];
+               struct iocb *finished = event->obj;
+
+               switch (finished->aio_lio_opcode) {
+               case IO_CMD_PWRITE:
+               case IO_CMD_PREAD: {
+                       struct aio_event *ae = talloc_get_type(finished->data, 
+                                                              struct aio_event);
+                       if (ae) {
+                               talloc_set_destructor(ae, NULL);
+                               ae->handler(ae->event_ctx, ae, 
+                                           event->res, ae->private_data);
+                               talloc_free(ae);
+                       }
+                       break;
+               }
+               case IOCB_CMD_EPOLL_WAIT: {
+                       struct epoll_event *ep = (struct epoll_event *)finished->u.c.buf;
+                       struct fd_event *fde;
+                       uint16_t flags = 0;
+                       int j;
+
+                       aio_ev->is_epoll_set = 0;
+
+                       for (j=0; j<event->res; j++, ep++) {
+                               fde = talloc_get_type(ep->data.ptr, 
+                                                     struct fd_event);
+                               if (fde == NULL) {
+                                       return -1;
+                               }
+                               if (ep->events & (EPOLLHUP|EPOLLERR)) {
+                                       fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_GOT_ERROR;
+                                       if (!(fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR)) {
+                                               epoll_del_event(aio_ev, fde);
+                                               continue;
+                                       }
+                                       flags |= EVENT_FD_READ;
+                               }
+                               if (ep->events & EPOLLIN) flags |= EVENT_FD_READ;
+                               if (ep->events & EPOLLOUT) flags |= EVENT_FD_WRITE;
+                               if (flags) {
+                                       fde->handler(aio_ev->ev, fde, flags, fde->private_data);
+                               }
+                       }
+                       break;
+               }
+               }
+               if (destruction_count != aio_ev->destruction_count) {
+                       return 0;
+               }
+       }
+
+       return 0;
+}
+
+/*
+  create a aio_event_context structure.
+*/
+static int aio_event_context_init(struct event_context *ev)
+{
+       struct aio_event_context *aio_ev;
+       
+       aio_ev = talloc_zero(ev, struct aio_event_context);
+       if (!aio_ev) return -1;
+
+       aio_ev->ev = ev;
+       aio_ev->epoll_iocb = talloc(aio_ev, struct iocb);
+
+       if (io_queue_init(MAX_AIO_QUEUE_DEPTH, &aio_ev->ioctx) != 0) {
+               talloc_free(aio_ev);
+               return -1;
+       }
+
+       aio_ev->epoll_fd = epoll_create(MAX_AIO_QUEUE_DEPTH);
+       if (aio_ev->epoll_fd == -1) {
+               talloc_free(aio_ev);
+               return -1;
+       }
+
+       talloc_set_destructor(aio_ev, aio_ctx_destructor);
+
+       ev->additional_data = aio_ev;
+
+       if (setup_epoll_wait(aio_ev) < 0) {
+               talloc_free(aio_ev);
+               return -1;
+       }
+
+       return 0;
+}
+
+/*
+  destroy an fd_event
+*/
+static int aio_event_fd_destructor(struct fd_event *fde)
+{
+       struct event_context *ev = fde->event_ctx;
+       struct aio_event_context *aio_ev = talloc_get_type(ev->additional_data,
+                                                          struct aio_event_context);
+
+       aio_ev->num_fd_events--;
+       aio_ev->destruction_count++;
+
+       epoll_del_event(aio_ev, fde);
+
+       return 0;
+}
+
+/*
+  add a fd based event
+  return NULL on failure (memory allocation error)
+*/
+static struct fd_event *aio_event_add_fd(struct event_context *ev, TALLOC_CTX *mem_ctx,
+                                        int fd, uint16_t flags,
+                                        event_fd_handler_t handler,
+                                        void *private_data)
+{
+       struct aio_event_context *aio_ev = talloc_get_type(ev->additional_data,
+                                                          struct aio_event_context);
+       struct fd_event *fde;
+
+       fde = talloc(mem_ctx?mem_ctx:ev, struct fd_event);
+       if (!fde) return NULL;
+
+       fde->event_ctx          = ev;
+       fde->fd                 = fd;
+       fde->flags              = flags;
+       fde->handler            = handler;
+       fde->private_data       = private_data;
+       fde->additional_flags   = 0;
+       fde->additional_data    = NULL;
+
+       aio_ev->num_fd_events++;
+       talloc_set_destructor(fde, aio_event_fd_destructor);
+
+       epoll_add_event(aio_ev, fde);
+
+       return fde;
+}
+
+
+/*
+  return the fd event flags
+*/
+static uint16_t aio_event_get_fd_flags(struct fd_event *fde)
+{
+       return fde->flags;
+}
+
+/*
+  set the fd event flags
+*/
+static void aio_event_set_fd_flags(struct fd_event *fde, uint16_t flags)
+{
+       struct event_context *ev;
+       struct aio_event_context *aio_ev;
+
+       if (fde->flags == flags) return;
+
+       ev = fde->event_ctx;
+       aio_ev = talloc_get_type(ev->additional_data, struct aio_event_context);
+
+       fde->flags = flags;
+
+       epoll_change_event(aio_ev, fde);
+}
+
+/*
+  do a single event loop using the events defined in ev 
+*/
+static int aio_event_loop_once(struct event_context *ev)
+{
+       struct aio_event_context *aio_ev = talloc_get_type(ev->additional_data,
+                                                          struct aio_event_context);
+       struct timeval tval;
+
+       tval = common_event_loop_delay(ev);
+
+       if (timeval_is_zero(&tval)) {
+               common_event_loop_timer(ev);
+               return 0;
+       }
+
+       return aio_event_loop(aio_ev, &tval);
+}
+
+/*
+  return on failure or (with 0) if all fd events are removed
+*/
+static int aio_event_loop_wait(struct event_context *ev)
+{
+       struct aio_event_context *aio_ev = talloc_get_type(ev->additional_data,
+                                                          struct aio_event_context);
+       while (aio_ev->num_fd_events) {
+               if (aio_event_loop_once(ev) != 0) {
+                       break;
+               }
+       }
+
+       return 0;
+}
+
+/*
+  called when a disk IO event needs to be cancelled
+*/
+static int aio_destructor(struct aio_event *ae)
+{
+       struct event_context *ev = ae->event_ctx;
+       struct aio_event_context *aio_ev = talloc_get_type(ev->additional_data,
+                                                          struct aio_event_context);
+       struct io_event result;
+       io_cancel(aio_ev->ioctx, &ae->iocb, &result);
+       /* TODO: handle errors from io_cancel()! */
+       return 0;
+}
+
+/* submit an aio disk IO event */
+static struct aio_event *aio_event_add_aio(struct event_context *ev, 
+                                          TALLOC_CTX *mem_ctx,
+                                          struct iocb *iocb,
+                                          event_aio_handler_t handler,
+                                          void *private_data)
+{
+       struct aio_event_context *aio_ev = talloc_get_type(ev->additional_data,
+                                                          struct aio_event_context);
+       struct iocb *iocbp;
+       struct aio_event *ae = talloc(mem_ctx?mem_ctx:ev, struct aio_event);
+       if (ae == NULL) return NULL;
+
+       ae->event_ctx    = ev;
+       ae->iocb         = *iocb;
+       ae->handler      = handler;
+       ae->private_data = private_data;
+       iocbp = &ae->iocb;
+
+       if (io_submit(aio_ev->ioctx, 1, &iocbp) != 1) {
+               talloc_free(ae);
+               return NULL;
+       }
+       ae->iocb.data = ae;
+       talloc_set_destructor(ae, aio_destructor);
+
+       return ae;
+}
+
+static const struct event_ops aio_event_ops = {
+       .context_init   = aio_event_context_init,
+       .add_fd         = aio_event_add_fd,
+       .add_aio        = aio_event_add_aio,
+       .get_fd_flags   = aio_event_get_fd_flags,
+       .set_fd_flags   = aio_event_set_fd_flags,
+       .add_timed      = common_event_add_timed,
+       .add_signal     = common_event_add_signal,
+       .loop_once      = aio_event_loop_once,
+       .loop_wait      = aio_event_loop_wait,
+};
+
+bool events_aio_init(void)
+{
+       return event_register_backend("aio", &aio_event_ops);
+}
+
+#if _SAMBA_BUILD_
+NTSTATUS s4_events_aio_init(void)
+{
+       if (!events_aio_init()) {
+               return NT_STATUS_INTERNAL_ERROR;
+       }
+       return NT_STATUS_OK;
+}
+#endif
diff --git a/ctdb/lib/events/events_epoll.c b/ctdb/lib/events/events_epoll.c
new file mode 100644 (file)
index 0000000..41a6509
--- /dev/null
@@ -0,0 +1,429 @@
+/* 
+   Unix SMB/CIFS implementation.
+
+   main select loop and event handling - epoll implementation
+
+   Copyright (C) Andrew Tridgell       2003-2005
+   Copyright (C) Stefan Metzmacher     2005
+   
+   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., 675 Mass Ave, Cambridge, MA 02139, USA.
+*/
+
+#include "includes.h"
+#include "system/filesys.h"
+#include "lib/util/dlinklist.h"
+#include "lib/events/events.h"
+#include "lib/events/events_internal.h"
+#include <sys/epoll.h>
+
+struct epoll_event_context {
+       /* a pointer back to the generic event_context */
+       struct event_context *ev;
+
+       /* number of registered fd event handlers */
+       int num_fd_events;
+
+       /* this is changed by the destructors for the fd event
+          type. It is used to detect event destruction by event
+          handlers, which means the code that is calling the event
+          handler needs to assume that the linked list is no longer
+          valid
+       */
+       uint32_t destruction_count;
+
+       /* when using epoll this is the handle from epoll_create */
+       int epoll_fd;
+};
+
+/*
+  called when a epoll call fails, and we should fallback
+  to using select
+*/
+static void epoll_fallback_to_select(struct epoll_event_context *epoll_ev, const char *reason)
+{
+       DEBUG(0,("%s (%s) - falling back to select()\n", reason, strerror(errno)));
+       close(epoll_ev->epoll_fd);
+       epoll_ev->epoll_fd = -1;
+       talloc_set_destructor(epoll_ev, NULL);
+}
+
+/*
+  map from EVENT_FD_* to EPOLLIN/EPOLLOUT
+*/
+static uint32_t epoll_map_flags(uint16_t flags)
+{
+       uint32_t ret = 0;
+       if (flags & EVENT_FD_READ) ret |= (EPOLLIN | EPOLLERR | EPOLLHUP);
+       if (flags & EVENT_FD_WRITE) ret |= (EPOLLOUT | EPOLLERR | EPOLLHUP);
+       return ret;
+}
+
+/*
+ free the epoll fd
+*/
+static int epoll_ctx_destructor(struct epoll_event_context *epoll_ev)
+{
+       close(epoll_ev->epoll_fd);
+       epoll_ev->epoll_fd = -1;
+       return 0;
+}
+
+/*
+ init the epoll fd
+*/
+static void epoll_init_ctx(struct epoll_event_context *epoll_ev)
+{
+       epoll_ev->epoll_fd = epoll_create(64);
+       talloc_set_destructor(epoll_ev, epoll_ctx_destructor);
+}
+
+#define EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT     (1<<0)
+#define EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR  (1<<1)
+#define EPOLL_ADDITIONAL_FD_FLAG_GOT_ERROR     (1<<2)
+
+/*
+ add the epoll event to the given fd_event
+*/
+static void epoll_add_event(struct epoll_event_context *epoll_ev, struct fd_event *fde)
+{
+       struct epoll_event event;
+       if (epoll_ev->epoll_fd == -1) return;
+
+       fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+
+       /* if we don't want events yet, don't add an epoll_event */
+       if (fde->flags == 0) return;
+
+       ZERO_STRUCT(event);
+       event.events = epoll_map_flags(fde->flags);
+       event.data.ptr = fde;
+       if (epoll_ctl(epoll_ev->epoll_fd, EPOLL_CTL_ADD, fde->fd, &event) != 0) {
+               epoll_fallback_to_select(epoll_ev, "EPOLL_CTL_ADD failed");
+       }
+       fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT;
+
+       /* only if we want to read we want to tell the event handler about errors */
+       if (fde->flags & EVENT_FD_READ) {
+               fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+       }
+}
+
+/*
+ delete the epoll event for given fd_event
+*/
+static void epoll_del_event(struct epoll_event_context *epoll_ev, struct fd_event *fde)
+{
+       struct epoll_event event;
+       if (epoll_ev->epoll_fd == -1) return;
+
+       fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+
+       /* if there's no epoll_event, we don't need to delete it */
+       if (!(fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT)) return;
+
+       ZERO_STRUCT(event);
+       event.events = epoll_map_flags(fde->flags);
+       event.data.ptr = fde;
+       epoll_ctl(epoll_ev->epoll_fd, EPOLL_CTL_DEL, fde->fd, &event);
+       fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT;
+}
+
+/*
+ change the epoll event to the given fd_event
+*/
+static void epoll_mod_event(struct epoll_event_context *epoll_ev, struct fd_event *fde)
+{
+       struct epoll_event event;
+       if (epoll_ev->epoll_fd == -1) return;
+
+       fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+
+       ZERO_STRUCT(event);
+       event.events = epoll_map_flags(fde->flags);
+       event.data.ptr = fde;
+       if (epoll_ctl(epoll_ev->epoll_fd, EPOLL_CTL_MOD, fde->fd, &event) != 0) {
+               epoll_fallback_to_select(epoll_ev, "EPOLL_CTL_MOD failed");
+       }
+
+       /* only if we want to read we want to tell the event handler about errors */
+       if (fde->flags & EVENT_FD_READ) {
+               fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+       }
+}
+
+static void epoll_change_event(struct epoll_event_context *epoll_ev, struct fd_event *fde)
+{
+       bool got_error = (fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_GOT_ERROR);
+       bool want_read = (fde->flags & EVENT_FD_READ);
+       bool want_write= (fde->flags & EVENT_FD_WRITE);
+
+       if (epoll_ev->epoll_fd == -1) return;
+
+       fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
+
+       /* there's already an event */
+       if (fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT) {
+               if (want_read || (want_write && !got_error)) {
+                       epoll_mod_event(epoll_ev, fde);
+                       return;
+               }
+               /* 
+                * if we want to match the select behavior, we need to remove the epoll_event
+                * when the caller isn't interested in events.
+                *
+                * this is because epoll reports EPOLLERR and EPOLLHUP, even without asking for them
+                */
+               epoll_del_event(epoll_ev, fde);
+               return;
+       }
+
+       /* there's no epoll_event attached to the fde */
+       if (want_read || (want_write && !got_error)) {
+               epoll_add_event(epoll_ev, fde);
+               return;
+       }
+}
+
+/*
+  event loop handling using epoll
+*/
+static int epoll_event_loop(struct epoll_event_context *epoll_ev, struct timeval *tvalp)
+{
+       int ret, i;
+#define MAXEVENTS 8
+       struct epoll_event events[MAXEVENTS];
+       uint32_t destruction_count = ++epoll_ev->destruction_count;
+       int timeout = -1;
+
+       if (epoll_ev->epoll_fd == -1) return -1;
+
+       if (tvalp) {
+               /* it's better to trigger timed events a bit later than to early */
+               timeout = ((tvalp->tv_usec+999) / 1000) + (tvalp->tv_sec*1000);
+       }
+
+       if (epoll_ev->ev->num_signal_handlers && 
+           common_event_check_signal(epoll_ev->ev)) {
+               return 0;
+       }
+
+       ret = epoll_wait(epoll_ev->epoll_fd, events, MAXEVENTS, timeout);
+
+       if (ret == -1 && errno == EINTR && epoll_ev->ev->num_signal_handlers) {
+               if (common_event_check_signal(epoll_ev->ev)) {
+                       return 0;
+               }
+       }
+
+       if (ret == -1 && errno != EINTR) {
+               epoll_fallback_to_select(epoll_ev, "epoll_wait() failed");
+               return -1;
+       }
+
+       if (ret == 0 && tvalp) {
+               common_event_loop_timer(epoll_ev->ev);
+               return 0;
+       }
+
+       for (i=0;i<ret;i++) {
+               struct fd_event *fde = talloc_get_type(events[i].data.ptr, 
+                                                      struct fd_event);
+               uint16_t flags = 0;
+
+               if (fde == NULL) {
+                       epoll_fallback_to_select(epoll_ev, "epoll_wait() gave bad data");
+                       return -1;
+               }
+               if (events[i].events & (EPOLLHUP|EPOLLERR)) {
+                       fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_GOT_ERROR;
+                       /*
+                        * if we only wait for EVENT_FD_WRITE, we should not tell the
+                        * event handler about it, and remove the epoll_event,
+                        * as we only report errors when waiting for read events,
+                        * to match the select() behavior
+                        */
+                       if (!(fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR)) {
+                               epoll_del_event(epoll_ev, fde);
+                               continue;
+                       }
+                       flags |= EVENT_FD_READ;
+               }
+               if (events[i].events & EPOLLIN) flags |= EVENT_FD_READ;
+               if (events[i].events & EPOLLOUT) flags |= EVENT_FD_WRITE;
+               if (flags) {
+                       fde->handler(epoll_ev->ev, fde, flags, fde->private_data);
+                       if (destruction_count != epoll_ev->destruction_count) {
+                               break;
+                       }
+               }
+       }
+
+       return 0;
+}
+
+/*
+  create a epoll_event_context structure.
+*/
+static int epoll_event_context_init(struct event_context *ev)
+{
+       struct epoll_event_context *epoll_ev;
+
+       epoll_ev = talloc_zero(ev, struct epoll_event_context);
+       if (!epoll_ev) return -1;
+       epoll_ev->ev = ev;
+       epoll_ev->epoll_fd = -1;
+
+       epoll_init_ctx(epoll_ev);
+
+       ev->additional_data = epoll_ev;
+       return 0;
+}
+
+/*
+  destroy an fd_event
+*/
+static int epoll_event_fd_destructor(struct fd_event *fde)
+{
+       struct event_context *ev = fde->event_ctx;
+       struct epoll_event_context *epoll_ev = talloc_get_type(ev->additional_data,
+                                                          struct epoll_event_context);
+
+       epoll_ev->num_fd_events--;
+       epoll_ev->destruction_count++;
+
+       epoll_del_event(epoll_ev, fde);
+
+       return 0;
+}
+
+/*
+  add a fd based event
+  return NULL on failure (memory allocation error)
+*/
+static struct fd_event *epoll_event_add_fd(struct event_context *ev, TALLOC_CTX *mem_ctx,
+                                        int fd, uint16_t flags,
+                                        event_fd_handler_t handler,
+                                        void *private_data)
+{
+       struct epoll_event_context *epoll_ev = talloc_get_type(ev->additional_data,
+                                                          struct epoll_event_context);
+       struct fd_event *fde;
+
+       fde = talloc(mem_ctx?mem_ctx:ev, struct fd_event);
+       if (!fde) return NULL;
+
+       fde->event_ctx          = ev;
+       fde->fd                 = fd;
+       fde->flags              = flags;
+       fde->handler            = handler;
+       fde->private_data       = private_data;
+       fde->additional_flags   = 0;
+       fde->additional_data    = NULL;
+
+       epoll_ev->num_fd_events++;
+       talloc_set_destructor(fde, epoll_event_fd_destructor);
+
+       epoll_add_event(epoll_ev, fde);
+
+       return fde;
+}
+
+
+/*
+  return the fd event flags
+*/
+static uint16_t epoll_event_get_fd_flags(struct fd_event *fde)
+{
+       return fde->flags;
+}
+
+/*
+  set the fd event flags
+*/
+static void epoll_event_set_fd_flags(struct fd_event *fde, uint16_t flags)
+{
+       struct event_context *ev;
+       struct epoll_event_context *epoll_ev;
+
+       if (fde->flags == flags) return;
+
+       ev = fde->event_ctx;
+       epoll_ev = talloc_get_type(ev->additional_data, struct epoll_event_context);
+
+       fde->flags = flags;
+
+       epoll_change_event(epoll_ev, fde);
+}
+
+/*
+  do a single event loop using the events defined in ev 
+*/
+static int epoll_event_loop_once(struct event_context *ev)
+{
+       struct epoll_event_context *epoll_ev = talloc_get_type(ev->additional_data,
+                                                          struct epoll_event_context);
+       struct timeval tval;
+
+       tval = common_event_loop_delay(ev);
+
+       if (timeval_is_zero(&tval)) {
+               common_event_loop_timer(ev);
+               return 0;
+       }
+
+       return epoll_event_loop(epoll_ev, &tval);
+}
+
+/*
+  return on failure or (with 0) if all fd events are removed
+*/
+static int epoll_event_loop_wait(struct event_context *ev)
+{
+       struct epoll_event_context *epoll_ev = talloc_get_type(ev->additional_data,
+                                                          struct epoll_event_context);
+       while (epoll_ev->num_fd_events) {
+               if (epoll_event_loop_once(ev) != 0) {
+                       break;
+               }
+       }
+
+       return 0;
+}
+
+static const struct event_ops epoll_event_ops = {
+       .context_init   = epoll_event_context_init,
+       .add_fd         = epoll_event_add_fd,
+       .get_fd_flags   = epoll_event_get_fd_flags,
+       .set_fd_flags   = epoll_event_set_fd_flags,
+       .add_timed      = common_event_add_timed,
+       .add_signal     = common_event_add_signal,
+       .loop_once      = epoll_event_loop_once,
+       .loop_wait      = epoll_event_loop_wait,
+};
+
+bool events_epoll_init(void)
+{
+       return event_register_backend("epoll", &epoll_event_ops);
+}
+
+#if _SAMBA_BUILD_
+NTSTATUS s4_events_epoll_init(void)
+{
+       if (!events_epoll_init()) {
+               return NT_STATUS_INTERNAL_ERROR;
+       }
+       return NT_STATUS_OK;
+}
+#endif
diff --git a/ctdb/lib/events/events_select.c b/ctdb/lib/events/events_select.c
new file mode 100644 (file)
index 0000000..291ddbd
--- /dev/null
@@ -0,0 +1,307 @@
+/* 
+   Unix SMB/CIFS implementation.
+   main select loop and event handling
+   Copyright (C) Andrew Tridgell       2003-2005
+   Copyright (C) Stefan Metzmacher     2005
+   
+   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., 675 Mass Ave, Cambridge, MA 02139, USA.
+*/
+
+/*
+  This is SAMBA's default event loop code
+
+*/
+
+#include "includes.h"
+#include "system/filesys.h"
+#include "system/select.h"
+#include "lib/util/dlinklist.h"
+#include "lib/events/events.h"
+#include "lib/events/events_internal.h"
+
+struct select_event_context {
+       /* a pointer back to the generic event_context */
+       struct event_context *ev;
+
+       /* list of filedescriptor events */
+       struct fd_event *fd_events;
+
+       /* list of timed events */
+       struct timed_event *timed_events;
+
+       /* the maximum file descriptor number in fd_events */
+       int maxfd;
+
+       /* information for exiting from the event loop */
+       int exit_code;
+
+       /* this is incremented when the loop over events causes something which
+          could change the events yet to be processed */
+       uint32_t destruction_count;
+};
+
+/*
+  create a select_event_context structure.
+*/
+static int select_event_context_init(struct event_context *ev)
+{
+       struct select_event_context *select_ev;
+
+       select_ev = talloc_zero(ev, struct select_event_context);
+       if (!select_ev) return -1;
+       select_ev->ev = ev;
+
+       ev->additional_data = select_ev;
+       return 0;
+}
+
+/*
+  recalculate the maxfd
+*/
+static void calc_maxfd(struct select_event_context *select_ev)
+{
+       struct fd_event *fde;
+
+       select_ev->maxfd = 0;
+       for (fde = select_ev->fd_events; fde; fde = fde->next) {
+               if (fde->fd > select_ev->maxfd) {
+                       select_ev->maxfd = fde->fd;
+               }
+       }
+}
+
+
+/* to mark the ev->maxfd invalid
+ * this means we need to recalculate it
+ */
+#define EVENT_INVALID_MAXFD (-1)
+
+/*
+  destroy an fd_event
+*/
+static int select_event_fd_destructor(struct fd_event *fde)
+{
+       struct event_context *ev = fde->event_ctx;
+       struct select_event_context *select_ev = talloc_get_type(ev->additional_data,
+                                                          struct select_event_context);
+
+       if (select_ev->maxfd == fde->fd) {
+               select_ev->maxfd = EVENT_INVALID_MAXFD;
+       }
+
+       DLIST_REMOVE(select_ev->fd_events, fde);
+       select_ev->destruction_count++;
+
+       return 0;
+}
+
+/*
+  add a fd based event
+  return NULL on failure (memory allocation error)
+*/
+static struct fd_event *select_event_add_fd(struct event_context *ev, TALLOC_CTX *mem_ctx,
+                                        int fd, uint16_t flags,
+                                        event_fd_handler_t handler,
+                                        void *private_data)
+{
+       struct select_event_context *select_ev = talloc_get_type(ev->additional_data,
+                                                          struct select_event_context);
+       struct fd_event *fde;
+
+       fde = talloc(mem_ctx?mem_ctx:ev, struct fd_event);
+       if (!fde) return NULL;
+
+       fde->event_ctx          = ev;
+       fde->fd                 = fd;
+       fde->flags              = flags;
+       fde->handler            = handler;
+       fde->private_data       = private_data;
+       fde->additional_flags   = 0;
+       fde->additional_data    = NULL;
+
+       DLIST_ADD(select_ev->fd_events, fde);
+       if (fde->fd > select_ev->maxfd) {
+               select_ev->maxfd = fde->fd;
+       }
+       talloc_set_destructor(fde, select_event_fd_destructor);
+
+       return fde;
+}
+
+
+/*
+  return the fd event flags
+*/
+static uint16_t select_event_get_fd_flags(struct fd_event *fde)
+{
+       return fde->flags;
+}
+
+/*
+  set the fd event flags
+*/
+static void select_event_set_fd_flags(struct fd_event *fde, uint16_t flags)
+{
+       struct event_context *ev;
+       struct select_event_context *select_ev;
+
+       if (fde->flags == flags) return;
+
+       ev = fde->event_ctx;
+       select_ev = talloc_get_type(ev->additional_data, struct select_event_context);
+
+       fde->flags = flags;
+}
+
+/*
+  event loop handling using select()
+*/
+static int select_event_loop_select(struct select_event_context *select_ev, struct timeval *tvalp)
+{
+       fd_set r_fds, w_fds;
+       struct fd_event *fde;
+       int selrtn;
+       uint32_t destruction_count = ++select_ev->destruction_count;
+
+       /* we maybe need to recalculate the maxfd */
+       if (select_ev->maxfd == EVENT_INVALID_MAXFD) {
+               calc_maxfd(select_ev);
+       }
+
+       FD_ZERO(&r_fds);
+       FD_ZERO(&w_fds);
+
+       /* setup any fd events */
+       for (fde = select_ev->fd_events; fde; fde = fde->next) {
+               if (fde->flags & EVENT_FD_READ) {
+                       FD_SET(fde->fd, &r_fds);
+               }
+               if (fde->flags & EVENT_FD_WRITE) {
+                       FD_SET(fde->fd, &w_fds);
+               }
+       }
+
+       if (select_ev->ev->num_signal_handlers && 
+           common_event_check_signal(select_ev->ev)) {
+               return 0;
+       }
+
+       selrtn = select(select_ev->maxfd+1, &r_fds, &w_fds, NULL, tvalp);
+
+       if (selrtn == -1 && errno == EINTR && 
+           select_ev->ev->num_signal_handlers) {
+               common_event_check_signal(select_ev->ev);
+               return 0;
+       }
+
+       if (selrtn == -1 && errno == EBADF) {
+               /* the socket is dead! this should never
+                  happen as the socket should have first been
+                  made readable and that should have removed
+                  the event, so this must be a bug. This is a
+                  fatal error. */
+               DEBUG(0,("ERROR: EBADF on select_event_loop_once\n"));
+               select_ev->exit_code = EBADF;
+               return -1;
+       }
+
+       if (selrtn == 0 && tvalp) {
+               common_event_loop_timer(select_ev->ev);
+               return 0;
+       }
+
+       if (selrtn > 0) {
+               /* at least one file descriptor is ready - check
+                  which ones and call the handler, being careful to allow
+                  the handler to remove itself when called */
+               for (fde = select_ev->fd_events; fde; fde = fde->next) {
+                       uint16_t flags = 0;
+
+                       if (FD_ISSET(fde->fd, &r_fds)) flags |= EVENT_FD_READ;
+                       if (FD_ISSET(fde->fd, &w_fds)) flags |= EVENT_FD_WRITE;
+                       if (flags) {
+                               fde->handler(select_ev->ev, fde, flags, fde->private_data);
+                               if (destruction_count != select_ev->destruction_count) {
+                                       break;
+                               }
+                       }
+               }
+       }
+
+       return 0;
+}              
+
+/*
+  do a single event loop using the events defined in ev 
+*/
+static int select_event_loop_once(struct event_context *ev)
+{
+       struct select_event_context *select_ev = talloc_get_type(ev->additional_data,
+                                                          struct select_event_context);
+       struct timeval tval;
+
+       tval = common_event_loop_delay(ev);
+
+       if (timeval_is_zero(&tval)) {
+               common_event_loop_timer(ev);
+               return 0;
+       }
+
+       return select_event_loop_select(select_ev, &tval);
+}
+
+/*
+  return on failure or (with 0) if all fd events are removed
+*/
+static int select_event_loop_wait(struct event_context *ev)
+{
+       struct select_event_context *select_ev = talloc_get_type(ev->additional_data,
+                                                          struct select_event_context);
+       select_ev->exit_code = 0;
+
+       while (select_ev->fd_events && select_ev->exit_code == 0) {
+               if (select_event_loop_once(ev) != 0) {
+                       break;
+               }
+       }
+
+       return select_ev->exit_code;
+}
+
+static const struct event_ops select_event_ops = {
+       .context_init   = select_event_context_init,
+       .add_fd         = select_event_add_fd,
+       .get_fd_flags   = select_event_get_fd_flags,
+       .set_fd_flags   = select_event_set_fd_flags,
+       .add_timed      = common_event_add_timed,
+       .add_signal     = common_event_add_signal,
+       .loop_once      = select_event_loop_once,
+       .loop_wait      = select_event_loop_wait,
+};
+
+bool events_select_init(void)
+{
+       return event_register_backend("select", &select_event_ops);
+}
+
+#if _SAMBA_BUILD_
+NTSTATUS s4_events_select_init(void)
+{
+       if (!events_select_init()) {
+               return NT_STATUS_INTERNAL_ERROR;
+       }
+       return NT_STATUS_OK;
+}
+#endif
diff --git a/ctdb/lib/events/events_signal.c b/ctdb/lib/events/events_signal.c
new file mode 100644 (file)
index 0000000..2a75a0c
--- /dev/null
@@ -0,0 +1,292 @@
+/* 
+   Unix SMB/CIFS implementation.
+
+   common events code for signal events
+
+   Copyright (C) Andrew Tridgell       2007
+   
+   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., 675 Mass Ave, Cambridge, MA 02139, USA.
+*/
+
+#include "includes.h"
+#include "system/filesys.h"
+#include "system/select.h"
+#include "system/wait.h"
+#include "lib/util/dlinklist.h"
+#include "lib/events/events.h"
+#include "lib/events/events_internal.h"
+
+#define NUM_SIGNALS 64
+
+/* maximum number of SA_SIGINFO signals to hold in the queue */
+#define SA_INFO_QUEUE_COUNT 10
+
+struct sigcounter {
+       uint32_t count;
+       uint32_t seen;
+};
+
+#define SIG_INCREMENT(s) (s).count++
+#define SIG_SEEN(s, n) (s).seen += (n)
+#define SIG_PENDING(s) ((s).seen != (s).count)
+
+
+/*
+  the poor design of signals means that this table must be static global
+*/
+static struct sig_state {
+       struct signal_event *sig_handlers[NUM_SIGNALS];
+       struct sigaction *oldact[NUM_SIGNALS];
+       struct sigcounter signal_count[NUM_SIGNALS];
+       struct sigcounter got_signal;
+       int pipe_hack[2];
+#ifdef SA_SIGINFO
+       /* with SA_SIGINFO we get quite a lot of info per signal */
+       siginfo_t *sig_info[NUM_SIGNALS];
+       struct sigcounter sig_blocked[NUM_SIGNALS];
+#endif
+} *sig_state;
+
+/*
+  return number of sigcounter events not processed yet
+*/
+static uint32_t sig_count(struct sigcounter s)
+{
+       if (s.count >= s.seen) {
+               return s.count - s.seen;
+       }
+       return 1 + (0xFFFFFFFF & ~(s.seen - s.count));
+}
+
+/*
+  signal handler - redirects to registered signals
+*/
+static void signal_handler(int signum)
+{
+       char c = 0;
+       SIG_INCREMENT(sig_state->signal_count[signum]);
+       SIG_INCREMENT(sig_state->got_signal);
+       /* doesn't matter if this pipe overflows */
+       write(sig_state->pipe_hack[1], &c, 1);
+}
+
+#ifdef SA_SIGINFO
+/*
+  signal handler with SA_SIGINFO - redirects to registered signals
+*/
+static void signal_handler_info(int signum, siginfo_t *info, void *uctx)
+{
+       uint32_t count = sig_count(sig_state->signal_count[signum]);
+       sig_state->sig_info[signum][count] = *info;
+
+       signal_handler(signum);
+
+       /* handle SA_SIGINFO */
+       if (count+1 == SA_INFO_QUEUE_COUNT) {
+               /* we've filled the info array - block this signal until
+                  these ones are delivered */
+               sigset_t set;
+               sigemptyset(&set);
+               sigaddset(&set, signum);
+               sigprocmask(SIG_BLOCK, &set, NULL);
+               SIG_INCREMENT(sig_state->sig_blocked[signum]);
+       }
+}
+#endif
+
+/*
+  destroy a signal event
+*/
+static int signal_event_destructor(struct signal_event *se)
+{
+       se->event_ctx->num_signal_handlers--;
+       DLIST_REMOVE(sig_state->sig_handlers[se->signum], se);
+       if (sig_state->sig_handlers[se->signum] == NULL) {
+               /* restore old handler, if any */
+               sigaction(se->signum, sig_state->oldact[se->signum], NULL);
+               sig_state->oldact[se->signum] = NULL;
+#ifdef SA_SIGINFO
+               if (se->sa_flags & SA_SIGINFO) {
+                       talloc_free(sig_state->sig_info[se->signum]);
+                       sig_state->sig_info[se->signum] = NULL;
+               }
+#endif
+       }
+       return 0;
+}
+
+/*
+  this is part of the pipe hack needed to avoid the signal race condition
+*/
+static void signal_pipe_handler(struct event_context *ev, struct fd_event *fde, 
+                               uint16_t flags, void *private)
+{
+       char c[16];
+       /* its non-blocking, doesn't matter if we read too much */
+       read(sig_state->pipe_hack[0], c, sizeof(c));
+}
+
+/*
+  add a signal event
+  return NULL on failure (memory allocation error)
+*/
+struct signal_event *common_event_add_signal(struct event_context *ev, 
+                                            TALLOC_CTX *mem_ctx,
+                                            int signum,
+                                            int sa_flags,
+                                            event_signal_handler_t handler, 
+                                            void *private_data) 
+{
+       struct signal_event *se;
+
+       if (signum >= NUM_SIGNALS) {
+               return NULL;
+       }
+
+       /* the sig_state needs to be on a global context as it can last across
+          multiple event contexts */
+       if (sig_state == NULL) {
+               sig_state = talloc_zero(talloc_autofree_context(), struct sig_state);
+               if (sig_state == NULL) {
+                       return NULL;
+               }
+       }
+
+       se = talloc(mem_ctx?mem_ctx:ev, struct signal_event);
+       if (se == NULL) return NULL;
+
+       se->event_ctx           = ev;
+       se->handler             = handler;
+       se->private_data        = private_data;
+       se->signum              = signum;
+       se->sa_flags            = sa_flags;
+       
+       /* Ensure, no matter the destruction order, that we always have a handle on the global sig_state */
+       if (!talloc_reference(se, sig_state)) {
+               return NULL;
+       }
+
+       /* only install a signal handler if not already installed */
+       if (sig_state->sig_handlers[signum] == NULL) {
+               struct sigaction act;
+               ZERO_STRUCT(act);
+               act.sa_handler   = signal_handler;
+               act.sa_flags = sa_flags;
+#ifdef SA_SIGINFO
+               if (sa_flags & SA_SIGINFO) {
+                       act.sa_handler   = NULL;
+                       act.sa_sigaction = signal_handler_info;
+                       if (sig_state->sig_info[signum] == NULL) {
+                               sig_state->sig_info[signum] = talloc_array(sig_state, siginfo_t, SA_INFO_QUEUE_COUNT);
+                               if (sig_state->sig_info[signum] == NULL) {
+                                       talloc_free(se);
+                                       return NULL;
+                               }
+                       }
+               }
+#endif
+               sig_state->oldact[signum] = talloc(sig_state, struct sigaction);
+               if (sig_state->oldact[signum] == NULL) {
+                       talloc_free(se);
+                       return NULL;                    
+               }
+               if (sigaction(signum, &act, sig_state->oldact[signum]) == -1) {
+                       talloc_free(se);
+                       return NULL;
+               }
+       }
+
+       DLIST_ADD(sig_state->sig_handlers[signum], se);
+
+       talloc_set_destructor(se, signal_event_destructor);
+
+       /* we need to setup the pipe hack handler if not already
+          setup */
+       if (ev->pipe_fde == NULL) {
+               if (sig_state->pipe_hack[0] == 0 && 
+                   sig_state->pipe_hack[1] == 0) {
+                       pipe(sig_state->pipe_hack);
+                       set_blocking(sig_state->pipe_hack[0], False);
+                       set_blocking(sig_state->pipe_hack[1], False);
+               }
+               ev->pipe_fde = event_add_fd(ev, ev, sig_state->pipe_hack[0],
+                                           EVENT_FD_READ, signal_pipe_handler, NULL);
+       }
+       ev->num_signal_handlers++;
+
+       return se;
+}
+
+
+/*
+  check if a signal is pending
+  return != 0 if a signal was pending
+*/
+int common_event_check_signal(struct event_context *ev)
+{
+       int i;
+
+       if (!sig_state || !SIG_PENDING(sig_state->got_signal)) {
+               return 0;
+       }
+       
+       for (i=0;i<NUM_SIGNALS+1;i++) {
+               struct signal_event *se, *next;
+               struct sigcounter counter = sig_state->signal_count[i];
+               uint32_t count = sig_count(counter);
+
+               if (count == 0) {
+                       continue;
+               }
+               for (se=sig_state->sig_handlers[i];se;se=next) {
+                       next = se->next;
+#ifdef SA_SIGINFO
+                       if (se->sa_flags & SA_SIGINFO) {
+                               int j;
+                               for (j=0;j<count;j++) {
+                                       /* note the use of the sig_info array as a
+                                          ring buffer */
+                                       int ofs = (counter.count + j) % SA_INFO_QUEUE_COUNT;
+                                       se->handler(ev, se, i, 1, 
+                                                   (void*)&sig_state->sig_info[i][ofs], 
+                                                   se->private_data);
+                               }
+                               if (SIG_PENDING(sig_state->sig_blocked[i])) {
+                                       /* we'd filled the queue, unblock the
+                                          signal now */
+                                       sigset_t set;
+                                       sigemptyset(&set);
+                                       sigaddset(&set, i);
+                                       SIG_SEEN(sig_state->sig_blocked[i], 
+                                                sig_count(sig_state->sig_blocked[i]));
+                                       sigprocmask(SIG_UNBLOCK, &set, NULL);
+                               }
+                               if (se->sa_flags & SA_RESETHAND) {
+                                       talloc_free(se);
+                               }
+                               continue;
+                       }
+#endif
+                       se->handler(ev, se, i, count, NULL, se->private_data);
+                       if (se->sa_flags & SA_RESETHAND) {
+                               talloc_free(se);
+                       }
+               }
+               SIG_SEEN(sig_state->signal_count[i], count);
+               SIG_SEEN(sig_state->got_signal, count);
+       }
+
+       return 1;
+}
diff --git a/ctdb/lib/events/events_timed.c b/ctdb/lib/events/events_timed.c
new file mode 100644 (file)
index 0000000..c85f837
--- /dev/null
@@ -0,0 +1,136 @@
+/* 
+   Unix SMB/CIFS implementation.
+
+   common events code for timed events
+
+   Copyright (C) Andrew Tridgell       2003-2006
+   Copyright (C) Stefan Metzmacher     2005
+   
+   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., 675 Mass Ave, Cambridge, MA 02139, USA.
+*/
+
+#include "includes.h"
+#include "system/filesys.h"
+#include "system/select.h"
+#include "lib/util/dlinklist.h"
+#include "lib/events/events.h"
+#include "lib/events/events_internal.h"
+
+/*
+  destroy a timed event
+*/
+static int common_event_timed_destructor(struct timed_event *te)
+{
+       struct event_context *ev = talloc_get_type(te->event_ctx,
+                                                  struct event_context);
+       DLIST_REMOVE(ev->timed_events, te);
+       return 0;
+}
+
+static int common_event_timed_deny_destructor(struct timed_event *te)
+{
+       return -1;
+}
+
+/*
+  add a timed event
+  return NULL on failure (memory allocation error)
+*/
+struct timed_event *common_event_add_timed(struct event_context *ev, TALLOC_CTX *mem_ctx,
+                                          struct timeval next_event, 
+                                          event_timed_handler_t handler, 
+                                          void *private_data) 
+{
+       struct timed_event *te, *last_te, *cur_te;
+
+       te = talloc(mem_ctx?mem_ctx:ev, struct timed_event);
+       if (te == NULL) return NULL;
+
+       te->event_ctx           = ev;
+       te->next_event          = next_event;
+       te->handler             = handler;
+       te->private_data        = private_data;
+       te->additional_data     = NULL;
+
+       /* keep the list ordered */
+       last_te = NULL;
+       for (cur_te = ev->timed_events; cur_te; cur_te = cur_te->next) {
+               /* if the new event comes before the current one break */
+               if (!timeval_is_zero(&cur_te->next_event) &&
+                   timeval_compare(&te->next_event,
+                                   &cur_te->next_event) < 0) {
+                       break;
+               }
+
+               last_te = cur_te;
+       }
+
+       DLIST_ADD_AFTER(ev->timed_events, te, last_te);
+
+       talloc_set_destructor(te, common_event_timed_destructor);
+
+       return te;
+}
+
+/*
+  a timer has gone off - call it
+*/
+void common_event_loop_timer(struct event_context *ev)
+{
+       struct timeval t = timeval_current();
+       struct timed_event *te = ev->timed_events;
+
+       if (te == NULL) {
+               return;
+       }
+
+       /* deny the handler to free the event */
+       talloc_set_destructor(te, common_event_timed_deny_destructor);
+
+       /* We need to remove the timer from the list before calling the
+        * handler because in a semi-async inner event loop called from the
+        * handler we don't want to come across this event again -- vl */
+       DLIST_REMOVE(ev->timed_events, te);
+
+       te->handler(ev, te, t, te->private_data);
+
+       /* The destructor isn't necessary anymore, we've already removed the
+        * event from the list. */
+       talloc_set_destructor(te, NULL);
+
+       talloc_free(te);
+}
+
+/*
+  do a single event loop using the events defined in ev 
+*/
+struct timeval common_event_loop_delay(struct event_context *ev)
+{
+       struct timeval tval;
+
+       /* work out the right timeout for all timed events */
+       if (ev->timed_events) {
+               struct timeval t = timeval_current();
+               tval = timeval_until(&t, &ev->timed_events->next_event);
+       } else {
+               /* have a default tick time of 30 seconds. This guarantees
+                  that code that uses its own timeout checking will be
+                  able to proceeed eventually */
+               tval = timeval_set(30, 0);
+       }
+       
+       return tval;
+}
+
diff --git a/ctdb/lib/replace/strptime.c b/ctdb/lib/replace/strptime.c
new file mode 100644 (file)
index 0000000..d415b78
--- /dev/null
@@ -0,0 +1,991 @@
+/* Convert a string representation of time to a time value.
+   Copyright (C) 1996, 1997, 1998, 1999, 2000 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1996.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Library General Public License as
+   published by the Free Software Foundation; either version 2 of the
+   License, or (at your option) any later version.
+
+   The GNU C Library 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
+   Library General Public License for more details.
+
+   You should have received a copy of the GNU Library General Public
+   License along with the GNU C Library; see the file COPYING.LIB.  If not,
+   write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+   Boston, MA 02111-1307, USA.  */
+
+/* XXX This version of the implementation is not really complete.
+   Some of the fields cannot add information alone.  But if seeing
+   some of them in the same format (such as year, week and weekday)
+   this is enough information for determining the date.  */
+
+#include "replace.h"
+#include "system/locale.h"
+#include "system/time.h"
+
+#ifndef __P
+# if defined (__GNUC__) || (defined (__STDC__) && __STDC__)
+#  define __P(args) args
+# else
+#  define __P(args) ()
+# endif  /* GCC.  */
+#endif  /* Not __P.  */
+
+#if ! HAVE_LOCALTIME_R && ! defined localtime_r
+# ifdef _LIBC
+#  define localtime_r __localtime_r
+# else
+/* Approximate localtime_r as best we can in its absence.  */
+#  define localtime_r my_localtime_r
+static struct tm *localtime_r __P ((const time_t *, struct tm *));
+static struct tm *
+localtime_r (t, tp)
+     const time_t *t;
+     struct tm *tp;
+{
+  struct tm *l = localtime (t);
+  if (! l)
+    return 0;
+  *tp = *l;
+  return tp;
+}
+# endif /* ! _LIBC */
+#endif /* ! HAVE_LOCALTIME_R && ! defined (localtime_r) */
+
+
+#define match_char(ch1, ch2) if (ch1 != ch2) return NULL
+#if defined __GNUC__ && __GNUC__ >= 2
+# define match_string(cs1, s2) \
+  ({ size_t len = strlen (cs1);                                                      \
+     int result = strncasecmp ((cs1), (s2), len) == 0;                       \
+     if (result) (s2) += len;                                                \
+     result; })
+#else
+/* Oh come on.  Get a reasonable compiler.  */
+# define match_string(cs1, s2) \
+  (strncasecmp ((cs1), (s2), strlen (cs1)) ? 0 : ((s2) += strlen (cs1), 1))
+#endif
+/* We intentionally do not use isdigit() for testing because this will
+   lead to problems with the wide character version.  */
+#define get_number(from, to, n) \
+  do {                                                                       \
+    int __n = n;                                                             \
+    val = 0;                                                                 \
+    while (*rp == ' ')                                                       \
+      ++rp;                                                                  \
+    if (*rp < '0' || *rp > '9')                                                      \
+      return NULL;                                                           \
+    do {                                                                     \
+      val *= 10;                                                             \
+      val += *rp++ - '0';                                                    \
+    } while (--__n > 0 && val * 10 <= to && *rp >= '0' && *rp <= '9');       \
+    if (val < from || val > to)                                                      \
+      return NULL;                                                           \
+  } while (0)
+#ifdef _NL_CURRENT
+# define get_alt_number(from, to, n) \
+  ({                                                                         \
+    __label__ do_normal;                                                     \
+    if (*decided != raw)                                                     \
+      {                                                                              \
+       const char *alts = _NL_CURRENT (LC_TIME, ALT_DIGITS);                 \
+       int __n = n;                                                          \
+       int any = 0;                                                          \
+       while (*rp == ' ')                                                    \
+         ++rp;                                                               \
+       val = 0;                                                              \
+       do {                                                                  \
+         val *= 10;                                                          \
+         while (*alts != '\0')                                               \
+           {                                                                 \
+             size_t len = strlen (alts);                                     \
+             if (strncasecmp (alts, rp, len) == 0)                           \
+               break;                                                        \
+             alts += len + 1;                                                \
+             ++val;                                                          \
+           }                                                                 \
+         if (*alts == '\0')                                                  \
+           {                                                                 \
+             if (*decided == not && ! any)                                   \
+               goto do_normal;                                               \
+             /* If we haven't read anything it's an error.  */               \
+             if (! any)                                                      \
+               return NULL;                                                  \
+             /* Correct the premature multiplication.  */                    \
+             val /= 10;                                                      \
+             break;                                                          \
+           }                                                                 \
+         else                                                                \
+           *decided = loc;                                                   \
+       } while (--__n > 0 && val * 10 <= to);                                \
+       if (val < from || val > to)                                           \
+         return NULL;                                                        \
+      }                                                                              \
+    else                                                                     \
+      {                                                                              \
+       do_normal:                                                            \
+        get_number (from, to, n);                                            \
+      }                                                                              \
+    0;                                                                       \
+  })
+#else
+# define get_alt_number(from, to, n) \
+  /* We don't have the alternate representation.  */                         \
+  get_number(from, to, n)
+#endif
+#define recursive(new_fmt) \
+  (*(new_fmt) != '\0'                                                        \
+   && (rp = strptime_internal (rp, (new_fmt), tm, decided, era_cnt)) != NULL)
+
+
+#ifdef _LIBC
+/* This is defined in locale/C-time.c in the GNU libc.  */
+extern const struct locale_data _nl_C_LC_TIME;
+extern const unsigned short int __mon_yday[2][13];
+
+# define weekday_name (&_nl_C_LC_TIME.values[_NL_ITEM_INDEX (DAY_1)].string)
+# define ab_weekday_name \
+  (&_nl_C_LC_TIME.values[_NL_ITEM_INDEX (ABDAY_1)].string)
+# define month_name (&_nl_C_LC_TIME.values[_NL_ITEM_INDEX (MON_1)].string)
+# define ab_month_name (&_nl_C_LC_TIME.values[_NL_ITEM_INDEX (ABMON_1)].string)
+# define HERE_D_T_FMT (_nl_C_LC_TIME.values[_NL_ITEM_INDEX (D_T_FMT)].string)
+# define HERE_D_FMT (_nl_C_LC_TIME.values[_NL_ITEM_INDEX (D_FMT)].string)
+# define HERE_AM_STR (_nl_C_LC_TIME.values[_NL_ITEM_INDEX (AM_STR)].string)
+# define HERE_PM_STR (_nl_C_LC_TIME.values[_NL_ITEM_INDEX (PM_STR)].string)
+# define HERE_T_FMT_AMPM \
+  (_nl_C_LC_TIME.values[_NL_ITEM_INDEX (T_FMT_AMPM)].string)
+# define HERE_T_FMT (_nl_C_LC_TIME.values[_NL_ITEM_INDEX (T_FMT)].string)
+
+# define strncasecmp(s1, s2, n) __strncasecmp (s1, s2, n)
+#else
+static char const weekday_name[][10] =
+  {
+    "Sunday", "Monday", "Tuesday", "Wednesday",
+    "Thursday", "Friday", "Saturday"
+  };
+static char const ab_weekday_name[][4] =
+  {
+    "Sun", "Mon", "Tue", "Wed", "Thu", "Fri", "Sat"
+  };
+static char const month_name[][10] =
+  {
+    "January", "February", "March", "April", "May", "June",
+    "July", "August", "September", "October", "November", "December"
+  };
+static char const ab_month_name[][4] =
+  {
+    "Jan", "Feb", "Mar", "Apr", "May", "Jun",
+    "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
+  };
+# define HERE_D_T_FMT "%a %b %e %H:%M:%S %Y"
+# define HERE_D_FMT "%m/%d/%y"
+# define HERE_AM_STR "AM"
+# define HERE_PM_STR "PM"
+# define HERE_T_FMT_AMPM "%I:%M:%S %p"
+# define HERE_T_FMT "%H:%M:%S"
+
+static const unsigned short int __mon_yday[2][13] =
+  {
+    /* Normal years.  */
+    { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334, 365 },
+    /* Leap years.  */
+    { 0, 31, 60, 91, 121, 152, 182, 213, 244, 274, 305, 335, 366 }
+  };
+#endif
+
+/* Status of lookup: do we use the locale data or the raw data?  */
+enum locale_status { not, loc, raw };
+
+
+#ifndef __isleap
+/* Nonzero if YEAR is a leap year (every 4 years,
+   except every 100th isn't, and every 400th is).  */
+# define __isleap(year)        \
+  ((year) % 4 == 0 && ((year) % 100 != 0 || (year) % 400 == 0))
+#endif
+
+/* Compute the day of the week.  */
+static void
+day_of_the_week (struct tm *tm)
+{
+  /* We know that January 1st 1970 was a Thursday (= 4).  Compute the
+     the difference between this data in the one on TM and so determine
+     the weekday.  */
+  int corr_year = 1900 + tm->tm_year - (tm->tm_mon < 2);
+  int wday = (-473
+             + (365 * (tm->tm_year - 70))
+             + (corr_year / 4)
+             - ((corr_year / 4) / 25) + ((corr_year / 4) % 25 < 0)
+             + (((corr_year / 4) / 25) / 4)
+             + __mon_yday[0][tm->tm_mon]
+             + tm->tm_mday - 1);
+  tm->tm_wday = ((wday % 7) + 7) % 7;
+}
+
+/* Compute the day of the year.  */
+static void
+day_of_the_year (struct tm *tm)
+{
+  tm->tm_yday = (__mon_yday[__isleap (1900 + tm->tm_year)][tm->tm_mon]
+                + (tm->tm_mday - 1));
+}
+
+static char *
+#ifdef _LIBC
+internal_function
+#endif
+strptime_internal __P ((const char *rp, const char *fmt, struct tm *tm,
+                       enum locale_status *decided, int era_cnt));
+
+static char *
+#ifdef _LIBC
+internal_function
+#endif
+strptime_internal (rp, fmt, tm, decided, era_cnt)
+     const char *rp;
+     const char *fmt;
+     struct tm *tm;
+     enum locale_status *decided;
+     int era_cnt;
+{
+  const char *rp_backup;
+  int cnt;
+  size_t val;
+  int have_I, is_pm;
+  int century, want_century;
+  int want_era;
+  int have_wday, want_xday;
+  int have_yday;
+  int have_mon, have_mday;
+#ifdef _NL_CURRENT
+  size_t num_eras;
+#endif
+  struct era_entry *era;
+
+  have_I = is_pm = 0;
+  century = -1;
+  want_century = 0;
+  want_era = 0;
+  era = NULL;
+
+  have_wday = want_xday = have_yday = have_mon = have_mday = 0;
+
+  while (*fmt != '\0')
+    {
+      /* A white space in the format string matches 0 more or white
+        space in the input string.  */
+      if (isspace (*fmt))
+       {
+         while (isspace (*rp))
+           ++rp;
+         ++fmt;
+         continue;
+       }
+
+      /* Any character but `%' must be matched by the same character
+        in the iput string.  */
+      if (*fmt != '%')
+       {
+         match_char (*fmt++, *rp++);
+         continue;
+       }
+
+      ++fmt;
+#ifndef _NL_CURRENT
+      /* We need this for handling the `E' modifier.  */
+    start_over:
+#endif
+
+      /* Make back up of current processing pointer.  */
+      rp_backup = rp;
+
+      switch (*fmt++)
+       {
+       case '%':
+         /* Match the `%' character itself.  */
+         match_char ('%', *rp++);
+         break;
+       case 'a':
+       case 'A':
+         /* Match day of week.  */
+         for (cnt = 0; cnt < 7; ++cnt)
+           {
+#ifdef _NL_CURRENT
+             if (*decided !=raw)
+               {
+                 if (match_string (_NL_CURRENT (LC_TIME, DAY_1 + cnt), rp))
+                   {
+                     if (*decided == not
+                         && strcmp (_NL_CURRENT (LC_TIME, DAY_1 + cnt),
+                                    weekday_name[cnt]))
+                       *decided = loc;
+                     break;
+                   }
+                 if (match_string (_NL_CURRENT (LC_TIME, ABDAY_1 + cnt), rp))
+                   {
+                     if (*decided == not
+                         && strcmp (_NL_CURRENT (LC_TIME, ABDAY_1 + cnt),
+                                    ab_weekday_name[cnt]))
+                       *decided = loc;
+                     break;
+                   }
+               }
+#endif
+             if (*decided != loc
+                 && (match_string (weekday_name[cnt], rp)
+                     || match_string (ab_weekday_name[cnt], rp)))
+               {
+                 *decided = raw;
+                 break;
+               }
+           }
+         if (cnt == 7)
+           /* Does not match a weekday name.  */
+           return NULL;
+         tm->tm_wday = cnt;
+         have_wday = 1;
+         break;
+       case 'b':
+       case 'B':
+       case 'h':
+         /* Match month name.  */
+         for (cnt = 0; cnt < 12; ++cnt)
+           {
+#ifdef _NL_CURRENT
+             if (*decided !=raw)
+               {
+                 if (match_string (_NL_CURRENT (LC_TIME, MON_1 + cnt), rp))
+                   {
+                     if (*decided == not
+                         && strcmp (_NL_CURRENT (LC_TIME, MON_1 + cnt),
+                                    month_name[cnt]))
+                       *decided = loc;
+                     break;
+                   }
+                 if (match_string (_NL_CURRENT (LC_TIME, ABMON_1 + cnt), rp))
+                   {
+                     if (*decided == not
+                         && strcmp (_NL_CURRENT (LC_TIME, ABMON_1 + cnt),
+                                    ab_month_name[cnt]))
+                       *decided = loc;
+                     break;
+                   }
+               }
+#endif
+             if (match_string (month_name[cnt], rp)
+                 || match_string (ab_month_name[cnt], rp))
+               {
+                 *decided = raw;
+                 break;
+               }
+           }
+         if (cnt == 12)
+           /* Does not match a month name.  */
+           return NULL;
+         tm->tm_mon = cnt;
+         want_xday = 1;
+         break;
+       case 'c':
+         /* Match locale's date and time format.  */
+#ifdef _NL_CURRENT
+         if (*decided != raw)
+           {
+             if (!recursive (_NL_CURRENT (LC_TIME, D_T_FMT)))
+               {
+                 if (*decided == loc)
+                   return NULL;
+                 else
+                   rp = rp_backup;
+               }
+             else
+               {
+                 if (*decided == not &&
+                     strcmp (_NL_CURRENT (LC_TIME, D_T_FMT), HERE_D_T_FMT))
+                   *decided = loc;
+                 want_xday = 1;
+                 break;
+               }
+             *decided = raw;
+           }
+#endif
+         if (!recursive (HERE_D_T_FMT))
+           return NULL;
+         want_xday = 1;
+         break;
+       case 'C':
+         /* Match century number.  */
+#ifdef _NL_CURRENT
+       match_century:
+#endif
+         get_number (0, 99, 2);
+         century = val;
+         want_xday = 1;
+         break;
+       case 'd':
+       case 'e':
+         /* Match day of month.  */
+         get_number (1, 31, 2);
+         tm->tm_mday = val;
+         have_mday = 1;
+         want_xday = 1;
+         break;
+       case 'F':
+         if (!recursive ("%Y-%m-%d"))
+           return NULL;
+         want_xday = 1;
+         break;
+       case 'x':
+#ifdef _NL_CURRENT
+         if (*decided != raw)
+           {
+             if (!recursive (_NL_CURRENT (LC_TIME, D_FMT)))
+               {
+                 if (*decided == loc)
+                   return NULL;
+                 else
+                   rp = rp_backup;
+               }
+             else
+               {
+                 if (*decided == not
+                     && strcmp (_NL_CURRENT (LC_TIME, D_FMT), HERE_D_FMT))
+                   *decided = loc;
+                 want_xday = 1;
+                 break;
+               }
+             *decided = raw;
+           }
+#endif
+         /* Fall through.  */
+       case 'D':
+         /* Match standard day format.  */
+         if (!recursive (HERE_D_FMT))
+           return NULL;
+         want_xday = 1;
+         break;
+       case 'k':
+       case 'H':
+         /* Match hour in 24-hour clock.  */
+         get_number (0, 23, 2);
+         tm->tm_hour = val;
+         have_I = 0;
+         break;
+       case 'I':
+         /* Match hour in 12-hour clock.  */
+         get_number (1, 12, 2);
+         tm->tm_hour = val % 12;
+         have_I = 1;
+         break;
+       case 'j':
+         /* Match day number of year.  */
+         get_number (1, 366, 3);
+         tm->tm_yday = val - 1;
+         have_yday = 1;
+         break;
+       case 'm':
+         /* Match number of month.  */
+         get_number (1, 12, 2);
+         tm->tm_mon = val - 1;
+         have_mon = 1;
+         want_xday = 1;
+         break;
+       case 'M':
+         /* Match minute.  */
+         get_number (0, 59, 2);
+         tm->tm_min = val;
+         break;
+       case 'n':
+       case 't':
+         /* Match any white space.  */
+         while (isspace (*rp))
+           ++rp;
+         break;
+       case 'p':
+         /* Match locale's equivalent of AM/PM.  */
+#ifdef _NL_CURRENT
+         if (*decided != raw)
+           {
+             if (match_string (_NL_CURRENT (LC_TIME, AM_STR), rp))
+               {
+                 if (strcmp (_NL_CURRENT (LC_TIME, AM_STR), HERE_AM_STR))
+                   *decided = loc;
+                 break;
+               }
+             if (match_string (_NL_CURRENT (LC_TIME, PM_STR), rp))
+               {
+                 if (strcmp (_NL_CURRENT (LC_TIME, PM_STR), HERE_PM_STR))
+                   *decided = loc;
+                 is_pm = 1;
+                 break;
+               }
+             *decided = raw;
+           }
+#endif
+         if (!match_string (HERE_AM_STR, rp)) {
+           if (match_string (HERE_PM_STR, rp)) {
+             is_pm = 1;
+           } else {
+             return NULL;
+           }
+         }
+         break;
+       case 'r':
+#ifdef _NL_CURRENT
+         if (*decided != raw)
+           {
+             if (!recursive (_NL_CURRENT (LC_TIME, T_FMT_AMPM)))
+               {
+                 if (*decided == loc)
+                   return NULL;
+                 else
+                   rp = rp_backup;
+               }
+             else
+               {
+                 if (*decided == not &&
+                     strcmp (_NL_CURRENT (LC_TIME, T_FMT_AMPM),
+                             HERE_T_FMT_AMPM))
+                   *decided = loc;
+                 break;
+               }
+             *decided = raw;
+           }
+#endif
+         if (!recursive (HERE_T_FMT_AMPM))
+           return NULL;
+         break;
+       case 'R':
+         if (!recursive ("%H:%M"))
+           return NULL;
+         break;
+       case 's':
+         {
+           /* The number of seconds may be very high so we cannot use
+              the `get_number' macro.  Instead read the number
+              character for character and construct the result while
+              doing this.  */
+           time_t secs = 0;
+           if (*rp < '0' || *rp > '9')
+             /* We need at least one digit.  */
+             return NULL;
+
+           do
+             {
+               secs *= 10;
+               secs += *rp++ - '0';
+             }
+           while (*rp >= '0' && *rp <= '9');
+
+           if (localtime_r (&secs, tm) == NULL)
+             /* Error in function.  */
+             return NULL;
+         }
+         break;
+       case 'S':
+         get_number (0, 61, 2);
+         tm->tm_sec = val;
+         break;
+       case 'X':
+#ifdef _NL_CURRENT
+         if (*decided != raw)
+           {
+             if (!recursive (_NL_CURRENT (LC_TIME, T_FMT)))
+               {
+                 if (*decided == loc)
+                   return NULL;
+                 else
+                   rp = rp_backup;
+               }
+             else
+               {
+                 if (strcmp (_NL_CURRENT (LC_TIME, T_FMT), HERE_T_FMT))
+                   *decided = loc;
+                 break;
+               }
+             *decided = raw;
+           }
+#endif
+         /* Fall through.  */
+       case 'T':
+         if (!recursive (HERE_T_FMT))
+           return NULL;
+         break;
+       case 'u':
+         get_number (1, 7, 1);
+         tm->tm_wday = val % 7;
+         have_wday = 1;
+         break;
+       case 'g':
+         get_number (0, 99, 2);
+         /* XXX This cannot determine any field in TM.  */
+         break;
+       case 'G':
+         if (*rp < '0' || *rp > '9')
+           return NULL;
+         /* XXX Ignore the number since we would need some more
+            information to compute a real date.  */
+         do
+           ++rp;
+         while (*rp >= '0' && *rp <= '9');
+         break;
+       case 'U':
+       case 'V':
+       case 'W':
+         get_number (0, 53, 2);
+         /* XXX This cannot determine any field in TM without some
+            information.  */
+         break;
+       case 'w':
+         /* Match number of weekday.  */
+         get_number (0, 6, 1);
+         tm->tm_wday = val;
+         have_wday = 1;
+         break;
+       case 'y':
+#ifdef _NL_CURRENT
+       match_year_in_century:
+#endif
+         /* Match year within century.  */
+         get_number (0, 99, 2);
+         /* The "Year 2000: The Millennium Rollover" paper suggests that
+            values in the range 69-99 refer to the twentieth century.  */
+         tm->tm_year = val >= 69 ? val : val + 100;
+         /* Indicate that we want to use the century, if specified.  */
+         want_century = 1;
+         want_xday = 1;
+         break;
+       case 'Y':
+         /* Match year including century number.  */
+         get_number (0, 9999, 4);
+         tm->tm_year = val - 1900;
+         want_century = 0;
+         want_xday = 1;
+         break;
+       case 'Z':
+         /* XXX How to handle this?  */
+         break;
+       case 'E':
+#ifdef _NL_CURRENT
+         switch (*fmt++)
+           {
+           case 'c':
+             /* Match locale's alternate date and time format.  */
+             if (*decided != raw)
+               {
+                 const char *fmt = _NL_CURRENT (LC_TIME, ERA_D_T_FMT);
+
+                 if (*fmt == '\0')
+                   fmt = _NL_CURRENT (LC_TIME, D_T_FMT);
+
+                 if (!recursive (fmt))
+                   {
+                     if (*decided == loc)
+                       return NULL;
+                     else
+                       rp = rp_backup;
+                   }
+                 else
+                   {
+                     if (strcmp (fmt, HERE_D_T_FMT))
+                       *decided = loc;
+                     want_xday = 1;
+                     break;
+                   }
+                 *decided = raw;
+               }
+             /* The C locale has no era information, so use the
+                normal representation.  */
+             if (!recursive (HERE_D_T_FMT))
+               return NULL;
+             want_xday = 1;
+             break;
+           case 'C':
+             if (*decided != raw)
+               {
+                 if (era_cnt >= 0)
+                   {
+                     era = _nl_select_era_entry (era_cnt);
+                     if (match_string (era->era_name, rp))
+                       {
+                         *decided = loc;
+                         break;
+                       }
+                     else
+                       return NULL;
+                   }
+                 else
+                   {
+                     num_eras = _NL_CURRENT_WORD (LC_TIME,
+                                                  _NL_TIME_ERA_NUM_ENTRIES);
+                     for (era_cnt = 0; era_cnt < (int) num_eras;
+                          ++era_cnt, rp = rp_backup)
+                       {
+                         era = _nl_select_era_entry (era_cnt);
+                         if (match_string (era->era_name, rp))
+                           {
+                             *decided = loc;
+                             break;
+                           }
+                       }
+                     if (era_cnt == (int) num_eras)
+                       {
+                         era_cnt = -1;
+                         if (*decided == loc)
+                           return NULL;
+                       }
+                     else
+                       break;
+                   }
+
+                 *decided = raw;
+               }
+             /* The C locale has no era information, so use the
+                normal representation.  */
+             goto match_century;
+           case 'y':
+             if (*decided == raw)
+               goto match_year_in_century;
+
+             get_number(0, 9999, 4);
+             tm->tm_year = val;
+             want_era = 1;
+             want_xday = 1;
+             break;
+           case 'Y':
+             if (*decided != raw)
+               {
+                 num_eras = _NL_CURRENT_WORD (LC_TIME,
+                                              _NL_TIME_ERA_NUM_ENTRIES);
+                 for (era_cnt = 0; era_cnt < (int) num_eras;
+                      ++era_cnt, rp = rp_backup)
+                   {
+                     era = _nl_select_era_entry (era_cnt);
+                     if (recursive (era->era_format))
+                       break;
+                   }
+                 if (era_cnt == (int) num_eras)
+                   {
+                     era_cnt = -1;
+                     if (*decided == loc)
+                       return NULL;
+                     else
+                       rp = rp_backup;
+                   }
+                 else
+                   {
+                     *decided = loc;
+                     era_cnt = -1;
+                     break;
+                   }
+
+                 *decided = raw;
+               }
+             get_number (0, 9999, 4);
+             tm->tm_year = val - 1900;
+             want_century = 0;
+             want_xday = 1;
+             break;
+           case 'x':
+             if (*decided != raw)
+               {
+                 const char *fmt = _NL_CURRENT (LC_TIME, ERA_D_FMT);
+
+                 if (*fmt == '\0')
+                   fmt = _NL_CURRENT (LC_TIME, D_FMT);
+
+                 if (!recursive (fmt))
+                   {
+                     if (*decided == loc)
+                       return NULL;
+                     else
+                       rp = rp_backup;
+                   }
+                 else
+                   {
+                     if (strcmp (fmt, HERE_D_FMT))
+                       *decided = loc;
+                     break;
+                   }
+                 *decided = raw;
+               }
+             if (!recursive (HERE_D_FMT))
+               return NULL;
+             break;
+           case 'X':
+             if (*decided != raw)
+               {
+                 const char *fmt = _NL_CURRENT (LC_TIME, ERA_T_FMT);
+
+                 if (*fmt == '\0')
+                   fmt = _NL_CURRENT (LC_TIME, T_FMT);
+
+                 if (!recursive (fmt))
+                   {
+                     if (*decided == loc)
+                       return NULL;
+                     else
+                       rp = rp_backup;
+                   }
+                 else
+                   {
+                     if (strcmp (fmt, HERE_T_FMT))
+                       *decided = loc;
+                     break;
+                   }
+                 *decided = raw;
+               }
+             if (!recursive (HERE_T_FMT))
+               return NULL;
+             break;
+           default:
+             return NULL;
+           }
+         break;
+#else
+         /* We have no information about the era format.  Just use
+            the normal format.  */
+         if (*fmt != 'c' && *fmt != 'C' && *fmt != 'y' && *fmt != 'Y'
+             && *fmt != 'x' && *fmt != 'X')
+           /* This is an illegal format.  */
+           return NULL;
+
+         goto start_over;
+#endif
+       case 'O':
+         switch (*fmt++)
+           {
+           case 'd':
+           case 'e':
+             /* Match day of month using alternate numeric symbols.  */
+             get_alt_number (1, 31, 2);
+             tm->tm_mday = val;
+             have_mday = 1;
+             want_xday = 1;
+             break;
+           case 'H':
+             /* Match hour in 24-hour clock using alternate numeric
+                symbols.  */
+             get_alt_number (0, 23, 2);
+             tm->tm_hour = val;
+             have_I = 0;
+             break;
+           case 'I':
+             /* Match hour in 12-hour clock using alternate numeric
+                symbols.  */
+             get_alt_number (1, 12, 2);
+             tm->tm_hour = val - 1;
+             have_I = 1;
+             break;
+           case 'm':
+             /* Match month using alternate numeric symbols.  */
+             get_alt_number (1, 12, 2);
+             tm->tm_mon = val - 1;
+             have_mon = 1;
+             want_xday = 1;
+             break;
+           case 'M':
+             /* Match minutes using alternate numeric symbols.  */
+             get_alt_number (0, 59, 2);
+             tm->tm_min = val;
+             break;
+           case 'S':
+             /* Match seconds using alternate numeric symbols.  */
+             get_alt_number (0, 61, 2);
+             tm->tm_sec = val;
+             break;
+           case 'U':
+           case 'V':
+           case 'W':
+             get_alt_number (0, 53, 2);
+             /* XXX This cannot determine any field in TM without
+                further information.  */
+             break;
+           case 'w':
+             /* Match number of weekday using alternate numeric symbols.  */
+             get_alt_number (0, 6, 1);
+             tm->tm_wday = val;
+             have_wday = 1;
+             break;
+           case 'y':
+             /* Match year within century using alternate numeric symbols.  */
+             get_alt_number (0, 99, 2);
+             tm->tm_year = val >= 69 ? val : val + 100;
+             want_xday = 1;
+             break;
+           default:
+             return NULL;
+           }
+         break;
+       default:
+         return NULL;
+       }
+    }
+
+  if (have_I && is_pm)
+    tm->tm_hour += 12;
+
+  if (century != -1)
+    {
+      if (want_century)
+       tm->tm_year = tm->tm_year % 100 + (century - 19) * 100;
+      else
+       /* Only the century, but not the year.  Strange, but so be it.  */
+       tm->tm_year = (century - 19) * 100;
+    }
+
+#ifdef _NL_CURRENT
+  if (era_cnt != -1)
+    {
+      era = _nl_select_era_entry(era_cnt);
+      if (want_era)
+       tm->tm_year = (era->start_date[0]
+                      + ((tm->tm_year - era->offset)
+                         * era->absolute_direction));
+      else
+       /* Era start year assumed.  */
+       tm->tm_year = era->start_date[0];
+    }
+  else
+#endif
+    if (want_era)
+      return NULL;
+
+  if (want_xday && !have_wday)
+    {
+      if ( !(have_mon && have_mday) && have_yday)
+       {
+         /* We don't have tm_mon and/or tm_mday, compute them.  */
+         int t_mon = 0;
+         while (__mon_yday[__isleap(1900 + tm->tm_year)][t_mon] <= tm->tm_yday)
+             t_mon++;
+         if (!have_mon)
+             tm->tm_mon = t_mon - 1;
+         if (!have_mday)
+             tm->tm_mday =
+               (tm->tm_yday
+                - __mon_yday[__isleap(1900 + tm->tm_year)][t_mon - 1] + 1);
+       }
+      day_of_the_week (tm);
+    }
+  if (want_xday && !have_yday)
+    day_of_the_year (tm);
+
+  return discard_const_p(char, rp);
+}
+
+
+char *rep_strptime(const char *buf, const char *format, struct tm *tm)
+{
+  enum locale_status decided;
+
+#ifdef _NL_CURRENT
+  decided = not;
+#else
+  decided = raw;
+#endif
+  return strptime_internal (buf, format, tm, &decided, -1);
+}
diff --git a/ctdb/lib/replace/strptime.m4 b/ctdb/lib/replace/strptime.m4
new file mode 100644 (file)
index 0000000..da22fc5
--- /dev/null
@@ -0,0 +1,13 @@
+AC_CACHE_CHECK([whether strptime is available and works],libreplace_cv_STRPTIME_OK,[
+       AC_TRY_RUN([
+               #define LIBREPLACE_CONFIGURE_TEST_STRPTIME
+               #include "$libreplacedir/test/strptime.c"
+               ],
+               [libreplace_cv_STRPTIME_OK=yes],
+               [libreplace_cv_STRPTIME_OK=no],
+               [libreplace_cv_STRPTIME_OK="assuming not"])
+])
+if test x"$libreplace_cv_STRPTIME_OK" != x"yes"; then
+        AC_DEFINE(REPLACE_STRPTIME,1,[Whether strptime should be replaced])
+        LIBREPLACEOBJ="${LIBREPLACEOBJ} strptime.o"
+fi
diff --git a/ctdb/lib/util/strlist.c b/ctdb/lib/util/strlist.c
new file mode 100644 (file)
index 0000000..c103bc3
--- /dev/null
@@ -0,0 +1,53 @@
+/* 
+   Unix SMB/CIFS implementation.
+   
+   Copyright (C) Andrew Tridgell 2005
+   Copyright (C) Jelmer Vernooij 2005
+   
+   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., 675 Mass Ave, Cambridge, MA 02139, USA.
+*/
+
+#include "includes.h"
+#include "system/locale.h"
+
+/**
+  return the number of elements in a string list
+*/
+_PUBLIC_ size_t str_list_length(const char **list)
+{
+       size_t ret;
+       for (ret=0;list && list[ret];ret++) /* noop */ ;
+       return ret;
+}
+
+
+/**
+  add an entry to a string list
+*/
+_PUBLIC_ const char **str_list_add(const char **list, const char *s)
+{
+       size_t len = str_list_length(list);
+       const char **ret;
+
+       ret = talloc_realloc(NULL, list, const char *, len+2);
+       if (ret == NULL) return NULL;
+
+       ret[len] = talloc_strdup(ret, s);
+       if (ret[len] == NULL) return NULL;
+
+       ret[len+1] = NULL;
+
+       return ret;
+}
diff --git a/ctdb/lib/util/util.c b/ctdb/lib/util/util.c
new file mode 100644 (file)
index 0000000..add3839
--- /dev/null
@@ -0,0 +1,53 @@
+/* 
+   Unix SMB/CIFS implementation.
+   
+   Copyright (C) Andrew Tridgell 2005
+   
+   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., 675 Mass Ave, Cambridge, MA 02139, USA.
+*/
+
+#include "includes.h"
+#include "system/filesys.h"
+
+
+/**
+ Set a fd into blocking/nonblocking mode. Uses POSIX O_NONBLOCK if available,
+ else
+  if SYSV use O_NDELAY
+  if BSD use FNDELAY
+**/
+
+_PUBLIC_ int set_blocking(int fd, BOOL set)
+{
+       int val;
+#ifdef O_NONBLOCK
+#define FLAG_TO_SET O_NONBLOCK
+#else
+#ifdef SYSV
+#define FLAG_TO_SET O_NDELAY
+#else /* BSD */
+#define FLAG_TO_SET FNDELAY
+#endif
+#endif
+
+       if((val = fcntl(fd, F_GETFL, 0)) == -1)
+               return -1;
+       if(set) /* Turn blocking on - ie. clear nonblock flag */
+               val &= ~FLAG_TO_SET;
+       else
+               val |= FLAG_TO_SET;
+       return fcntl( fd, F_SETFL, val);
+#undef FLAG_TO_SET
+}