-/* LibMemcached
- * Copyright (C) 2006-2009 Brian Aker
- * All rights reserved.
+/* vim:expandtab:shiftwidth=2:tabstop=2:smarttab:
+ *
+ * LibMemcached
*
- * Use and distribution licensed under the BSD license. See
- * the COPYING file in the parent directory for full text.
+ * Copyright (C) 2011 Data Differential, http://datadifferential.com/
+ * Copyright (C) 2006-2009 Brian Aker
+ * All rights reserved.
*
- * Summary: Server IO, Not public!
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ * * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *
+ * * The names of its contributors may not be used to endorse or
+ * promote products derived from this software without specific prior
+ * written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
*/
-#include "common.h"
-#include <sys/select.h>
-#include <poll.h>
+#include "libmemcached/common.h"
typedef enum {
MEM_READ,
} memc_read_or_write;
static ssize_t io_flush(memcached_server_write_instance_st ptr,
+ const bool with_flush,
memcached_return_t *error);
static void increment_udp_message_id(memcached_server_write_instance_st ptr);
};
int error;
- unlikely (read_or_write == MEM_WRITE) /* write */
+ if (read_or_write == MEM_WRITE) /* write */
+ {
fds.events= POLLOUT;
+ WATCHPOINT_SET(ptr->io_wait_count.write++);
+ }
+ else
+ {
+ WATCHPOINT_SET(ptr->io_wait_count.read++);
+ }
/*
** We are going to block on write, but at least on Solaris we might block
return MEMCACHED_FAILURE;
}
- int timeout= ptr->root->poll_timeout;
- if (ptr->root->flags.no_block == false)
- timeout= -1;
-
size_t loop_max= 5;
- while (--loop_max)
+ while (--loop_max) // While loop is for ERESTART or EINTR
{
- error= poll(&fds, 1, timeout);
+ error= poll(&fds, 1, ptr->root->poll_timeout);
switch (error)
{
- case 1:
+ case 1: // Success!
+ WATCHPOINT_IF_LABELED_NUMBER(read_or_write && loop_max < 4, "read() times we had to loop, decremented down from 5", loop_max);
+ WATCHPOINT_IF_LABELED_NUMBER(!read_or_write && loop_max < 4, "write() times we had to loop, decremented down from 5", loop_max);
+
return MEMCACHED_SUCCESS;
- case 0:
- continue;
+ case 0: // Timeout occured, we let the while() loop do its thing.
+ return MEMCACHED_TIMEOUT;
default:
- WATCHPOINT_ERRNO(errno);
+ WATCHPOINT_ERRNO(get_socket_errno());
+ switch (get_socket_errno())
{
- switch (errno)
+#ifdef TARGET_OS_LINUX
+ case ERESTART:
+#endif
+ case EINTR:
+ break;
+ default:
+ if (fds.revents & POLLERR)
{
- default:
- ptr->cached_errno= error;
- memcached_quit_server(ptr, true);
-
- return MEMCACHED_FAILURE;
+ int err;
+ socklen_t len= sizeof (err);
+ (void)getsockopt(ptr->fd, SOL_SOCKET, SO_ERROR, &err, &len);
+ ptr->cached_errno= (err == 0) ? get_socket_errno() : err;
}
+ else
+ {
+ ptr->cached_errno= get_socket_errno();
+ }
+ memcached_quit_server(ptr, true);
+
+ return MEMCACHED_FAILURE;
}
}
}
- if (loop_max == 0 && error == 0)
- return MEMCACHED_TIMEOUT;
-
/* Imposssible for anything other then -1 */
WATCHPOINT_ASSERT(error == -1);
- ptr->cached_errno= error;
+ ptr->cached_errno= get_socket_errno();
memcached_quit_server(ptr, true);
return MEMCACHED_FAILURE;
}
+memcached_return_t memcached_io_wait_for_write(memcached_server_write_instance_st ptr)
+{
+ return io_wait(ptr, MEM_WRITE);
+}
+
/**
* Try to fill the input buffer for a server with as much
* data as possible.
if (ptr->read_buffer_length != MEMCACHED_MAX_BUFFER)
{
/* Just try a single read to grab what's available */
- ssize_t nr= read(ptr->fd,
+ ssize_t nr= recv(ptr->fd,
ptr->read_ptr + ptr->read_data_length,
- MEMCACHED_MAX_BUFFER - ptr->read_data_length);
+ MEMCACHED_MAX_BUFFER - ptr->read_data_length,
+ 0);
if (nr > 0)
{
return false;
}
-static inline void memcached_io_cork_push(memcached_server_st *ptr)
-{
- (void)ptr;
-#ifdef CORK
- if (ptr->root->flags.cork == false || ptr->state.is_corked)
- return;
-
- int enable= 1;
- int err= setsockopt(ptr->fd, IPPROTO_TCP, CORK,
- &enable, (socklen_t)sizeof(int));
- if (! err)
- ptr->state.is_corked= true;
-
- WATCHPOINT_ASSERT(ptr->state.is_corked == true);
-#endif
-}
-
-static inline void memcached_io_cork_pop(memcached_server_st *ptr)
-{
- (void)ptr;
-#ifdef CORK
- if (ptr->root->flags.cork == false || ptr->state.is_corked == false)
- return;
-
- int enable= 0;
- int err= setsockopt(ptr->fd, IPPROTO_TCP, CORK,
- &enable, (socklen_t)sizeof(int));
- if (! err)
- ptr->state.is_corked= false;
-
- WATCHPOINT_ASSERT(ptr->state.is_corked == false);
-#endif
-}
-
#if 0 // Dead code, this should be removed.
void memcached_io_preread(memcached_st *ptr)
{
{
size_t data_read;
- data_read= read(ptr->hosts[x].fd,
+ data_read= recv(ptr->hosts[x].fd,
ptr->hosts[x].read_ptr + ptr->hosts[x].read_data_length,
- MEMCACHED_MAX_BUFFER - ptr->hosts[x].read_data_length);
- if (data_read == -1)
+ MEMCACHED_MAX_BUFFER - ptr->hosts[x].read_data_length, 0);
+ if (data_read == SOCKET_ERROR)
continue;
ptr->hosts[x].read_buffer_length+= data_read;
while (1)
{
- data_read= read(ptr->fd, ptr->read_buffer, MEMCACHED_MAX_BUFFER);
+ data_read= recv(ptr->fd, ptr->read_buffer, MEMCACHED_MAX_BUFFER, 0);
if (data_read > 0)
+ {
break;
- else if (data_read == -1)
+ }
+ else if (data_read == SOCKET_ERROR)
{
- ptr->cached_errno= errno;
- memcached_return_t rc= MEMCACHED_UNKNOWN_READ_FAILURE;
- switch (errno)
+ ptr->cached_errno= get_socket_errno();
+ memcached_return_t rc= MEMCACHED_ERRNO;
+ switch (get_socket_errno())
{
+ case EWOULDBLOCK:
+#ifdef USE_EAGAIN
case EAGAIN:
+#endif
case EINTR:
#ifdef TARGET_OS_LINUX
case ERESTART:
for blocking I/O we do not return 0 and for non-blocking case
it will return EGAIN if data is not immediatly available.
*/
+ WATCHPOINT_STRING("We had a zero length recv()");
memcached_quit_server(ptr, true);
*nread= -1;
return MEMCACHED_UNKNOWN_READ_FAILURE;
size_t original_length;
const char* buffer_ptr;
- WATCHPOINT_ASSERT(ptr->fd != -1);
+ WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
original_length= length;
buffer_ptr= buffer;
- /* more writable data is coming if a flush isn't required, so delay send */
- if (! with_flush)
- {
- memcached_io_cork_push(ptr);
- }
-
while (length)
{
char *write_ptr;
buffer_end= MAX_UDP_DATAGRAM_LENGTH;
should_write= length;
if (ptr->write_buffer_offset + should_write > buffer_end)
+ {
return -1;
+ }
}
else
{
memcached_return_t rc;
ssize_t sent_length;
- WATCHPOINT_ASSERT(ptr->fd != -1);
- sent_length= io_flush(ptr, &rc);
+ WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
+ sent_length= io_flush(ptr, with_flush, &rc);
if (sent_length == -1)
+ {
return -1;
+ }
/* If io_flush calls memcached_purge, sent_length may be 0 */
unlikely (sent_length != 0)
if (with_flush)
{
memcached_return_t rc;
- WATCHPOINT_ASSERT(ptr->fd != -1);
- if (io_flush(ptr, &rc) == -1)
+ WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
+ if (io_flush(ptr, with_flush, &rc) == -1)
{
return -1;
}
-
- memcached_io_cork_pop(ptr);
}
return (ssize_t) original_length;
}
ssize_t memcached_io_writev(memcached_server_write_instance_st ptr,
- const struct __write_vector_st *vector,
+ const struct libmemcached_io_vector_st *vector,
size_t number_of, bool with_flush)
{
ssize_t total= 0;
memcached_return_t memcached_io_close(memcached_server_write_instance_st ptr)
{
- if (ptr->fd == -1)
+ if (ptr->fd == INVALID_SOCKET)
{
return MEMCACHED_SUCCESS;
}
/* in case of death shutdown to avoid blocking at close() */
- if (shutdown(ptr->fd, SHUT_RDWR) == -1 && errno != ENOTCONN)
+ if (shutdown(ptr->fd, SHUT_RDWR) == SOCKET_ERROR && get_socket_errno() != ENOTCONN)
{
WATCHPOINT_NUMBER(ptr->fd);
- WATCHPOINT_ERRNO(errno);
- WATCHPOINT_ASSERT(errno);
+ WATCHPOINT_ERRNO(get_socket_errno());
+ WATCHPOINT_ASSERT(get_socket_errno());
}
- if (close(ptr->fd) == -1)
+ if (closesocket(ptr->fd) == SOCKET_ERROR)
{
- WATCHPOINT_ERRNO(errno);
+ WATCHPOINT_ERRNO(get_socket_errno());
}
return MEMCACHED_SUCCESS;
int err= poll(fds, host_index, memc->poll_timeout);
switch (err) {
case -1:
- memc->cached_errno = errno;
+ memcached_set_errno(memc, get_socket_errno(), NULL);
/* FALLTHROUGH */
case 0:
break;
}
static ssize_t io_flush(memcached_server_write_instance_st ptr,
+ const bool with_flush,
memcached_return_t *error)
{
/*
*/
{
memcached_return_t rc;
- WATCHPOINT_ASSERT(ptr->fd != -1);
+ WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
rc= memcached_purge(ptr);
if (rc != MEMCACHED_SUCCESS && rc != MEMCACHED_STORED)
+ {
return -1;
+ }
}
ssize_t sent_length;
size_t return_length;
*error= MEMCACHED_SUCCESS;
- WATCHPOINT_ASSERT(ptr->fd != -1);
+ WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
// UDP Sanity check, make sure that we are not sending somthing too big
if (ptr->type == MEMCACHED_CONNECTION_UDP && write_length > MAX_UDP_DATAGRAM_LENGTH)
+ {
return -1;
+ }
if (ptr->write_buffer_offset == 0 || (ptr->type == MEMCACHED_CONNECTION_UDP
&& ptr->write_buffer_offset == UDP_DATAGRAM_HEADER_LENGTH))
return_length= 0;
while (write_length)
{
- WATCHPOINT_ASSERT(ptr->fd != -1);
+ WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
WATCHPOINT_ASSERT(write_length > 0);
sent_length= 0;
if (ptr->type == MEMCACHED_CONNECTION_UDP)
increment_udp_message_id(ptr);
- sent_length= write(ptr->fd, local_write_ptr, write_length);
- if (sent_length == -1)
+ WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
+ if (with_flush)
+ {
+ sent_length= send(ptr->fd, local_write_ptr, write_length, MSG_NOSIGNAL|MSG_DONTWAIT);
+ }
+ else
+ {
+ sent_length= send(ptr->fd, local_write_ptr, write_length, MSG_NOSIGNAL|MSG_DONTWAIT|MSG_MORE);
+ }
+
+ if (sent_length == SOCKET_ERROR)
{
- ptr->cached_errno= errno;
- WATCHPOINT_ERRNO(errno);
- switch (errno)
+ ptr->cached_errno= get_socket_errno();
+#if 0 // @todo I should look at why we hit this bit of code hard frequently
+ WATCHPOINT_ERRNO(get_socket_errno());
+ WATCHPOINT_NUMBER(get_socket_errno());
+#endif
+ switch (get_socket_errno())
{
case ENOBUFS:
continue;
+ case EWOULDBLOCK:
+#ifdef USE_EAGAIN
case EAGAIN:
+#endif
{
/*
* We may be blocked on write because the input buffer
memcached_quit_server(ptr, true);
return -1;
}
+ case ENOTCONN:
+ case EPIPE:
default:
memcached_quit_server(ptr, true);
*error= MEMCACHED_ERRNO;
+ WATCHPOINT_ASSERT(ptr->fd == -1);
return -1;
}
}