2005-01-06 23:35:40 +00:00
|
|
|
/*-
|
2017-11-27 15:20:12 +00:00
|
|
|
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
|
|
|
|
*
|
2000-09-06 18:49:13 +00:00
|
|
|
* Copyright (c) 2000 Paycounter, Inc.
|
2005-03-11 21:37:45 +00:00
|
|
|
* Copyright (c) 2005 Robert N. M. Watson
|
2000-09-06 18:49:13 +00:00
|
|
|
* Author: Alfred Perlstein <alfred@paycounter.com>, <alfred@FreeBSD.org>
|
2000-06-20 01:09:23 +00:00
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
* are met:
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
* 2. 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.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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.
|
|
|
|
*/
|
|
|
|
|
2003-06-11 00:56:59 +00:00
|
|
|
#include <sys/cdefs.h>
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
2000-06-20 01:09:23 +00:00
|
|
|
#define ACCEPT_FILTER_MOD
|
|
|
|
|
2001-06-01 21:47:34 +00:00
|
|
|
#include "opt_param.h"
|
2000-06-20 01:09:23 +00:00
|
|
|
#include <sys/param.h>
|
|
|
|
#include <sys/systm.h>
|
|
|
|
#include <sys/domain.h>
|
|
|
|
#include <sys/kernel.h>
|
2004-06-01 04:08:48 +00:00
|
|
|
#include <sys/lock.h>
|
2000-06-20 01:09:23 +00:00
|
|
|
#include <sys/malloc.h>
|
|
|
|
#include <sys/mbuf.h>
|
2004-05-30 17:57:46 +00:00
|
|
|
#include <sys/module.h>
|
2004-06-01 04:08:48 +00:00
|
|
|
#include <sys/mutex.h>
|
2000-06-20 01:09:23 +00:00
|
|
|
#include <sys/protosw.h>
|
2000-09-06 18:49:13 +00:00
|
|
|
#include <sys/sysctl.h>
|
2000-06-20 01:09:23 +00:00
|
|
|
#include <sys/socket.h>
|
|
|
|
#include <sys/socketvar.h>
|
|
|
|
#include <sys/queue.h>
|
|
|
|
|
2004-06-01 04:08:48 +00:00
|
|
|
static struct mtx accept_filter_mtx;
|
|
|
|
MTX_SYSINIT(accept_filter, &accept_filter_mtx, "accept_filter_mtx",
|
|
|
|
MTX_DEF);
|
|
|
|
#define ACCEPT_FILTER_LOCK() mtx_lock(&accept_filter_mtx)
|
|
|
|
#define ACCEPT_FILTER_UNLOCK() mtx_unlock(&accept_filter_mtx)
|
|
|
|
|
2000-06-20 01:09:23 +00:00
|
|
|
static SLIST_HEAD(, accept_filter) accept_filtlsthd =
|
2009-12-28 22:56:30 +00:00
|
|
|
SLIST_HEAD_INITIALIZER(accept_filtlsthd);
|
2000-06-20 01:09:23 +00:00
|
|
|
|
|
|
|
MALLOC_DEFINE(M_ACCF, "accf", "accept filter data");
|
|
|
|
|
2000-09-06 18:49:13 +00:00
|
|
|
static int unloadable = 0;
|
|
|
|
|
2014-07-26 19:27:34 +00:00
|
|
|
SYSCTL_NODE(_net, OID_AUTO, accf, CTLFLAG_RW, 0, "Accept filters");
|
|
|
|
SYSCTL_INT(_net_accf, OID_AUTO, unloadable, CTLFLAG_RW, &unloadable, 0,
|
2000-09-06 18:49:13 +00:00
|
|
|
"Allow unload of accept filters (not recommended)");
|
|
|
|
|
2000-06-20 01:09:23 +00:00
|
|
|
/*
|
2005-02-18 18:49:44 +00:00
|
|
|
* Must be passed a malloc'd structure so we don't explode if the kld is
|
|
|
|
* unloaded, we leak the struct on deallocation to deal with this, but if a
|
|
|
|
* filter is loaded with the same name as a leaked one we re-use the entry.
|
2000-06-20 01:09:23 +00:00
|
|
|
*/
|
|
|
|
int
|
|
|
|
accept_filt_add(struct accept_filter *filt)
|
|
|
|
{
|
|
|
|
struct accept_filter *p;
|
|
|
|
|
2004-06-01 04:08:48 +00:00
|
|
|
ACCEPT_FILTER_LOCK();
|
2000-06-20 01:09:23 +00:00
|
|
|
SLIST_FOREACH(p, &accept_filtlsthd, accf_next)
|
|
|
|
if (strcmp(p->accf_name, filt->accf_name) == 0) {
|
|
|
|
if (p->accf_callback != NULL) {
|
2004-06-01 04:08:48 +00:00
|
|
|
ACCEPT_FILTER_UNLOCK();
|
2000-06-20 01:09:23 +00:00
|
|
|
return (EEXIST);
|
|
|
|
} else {
|
|
|
|
p->accf_callback = filt->accf_callback;
|
2004-06-01 04:08:48 +00:00
|
|
|
ACCEPT_FILTER_UNLOCK();
|
2008-10-23 15:53:51 +00:00
|
|
|
free(filt, M_ACCF);
|
2000-06-20 01:09:23 +00:00
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (p == NULL)
|
|
|
|
SLIST_INSERT_HEAD(&accept_filtlsthd, filt, accf_next);
|
2004-06-01 04:08:48 +00:00
|
|
|
ACCEPT_FILTER_UNLOCK();
|
2000-06-20 01:09:23 +00:00
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
accept_filt_del(char *name)
|
|
|
|
{
|
|
|
|
struct accept_filter *p;
|
|
|
|
|
|
|
|
p = accept_filt_get(name);
|
|
|
|
if (p == NULL)
|
|
|
|
return (ENOENT);
|
|
|
|
|
|
|
|
p->accf_callback = NULL;
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
struct accept_filter *
|
|
|
|
accept_filt_get(char *name)
|
|
|
|
{
|
|
|
|
struct accept_filter *p;
|
|
|
|
|
2004-06-01 04:08:48 +00:00
|
|
|
ACCEPT_FILTER_LOCK();
|
2000-06-20 01:09:23 +00:00
|
|
|
SLIST_FOREACH(p, &accept_filtlsthd, accf_next)
|
|
|
|
if (strcmp(p->accf_name, name) == 0)
|
2004-06-01 04:08:48 +00:00
|
|
|
break;
|
|
|
|
ACCEPT_FILTER_UNLOCK();
|
2000-06-20 01:09:23 +00:00
|
|
|
|
2004-06-01 04:08:48 +00:00
|
|
|
return (p);
|
2000-06-20 01:09:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
accept_filt_generic_mod_event(module_t mod, int event, void *data)
|
|
|
|
{
|
|
|
|
struct accept_filter *p;
|
|
|
|
struct accept_filter *accfp = (struct accept_filter *) data;
|
2005-02-18 19:01:22 +00:00
|
|
|
int error;
|
2000-06-20 01:09:23 +00:00
|
|
|
|
|
|
|
switch (event) {
|
|
|
|
case MOD_LOAD:
|
2017-06-02 17:43:47 +00:00
|
|
|
p = malloc(sizeof(*p), M_ACCF, M_WAITOK);
|
2000-06-20 01:09:23 +00:00
|
|
|
bcopy(accfp, p, sizeof(*p));
|
|
|
|
error = accept_filt_add(p);
|
|
|
|
break;
|
|
|
|
|
|
|
|
case MOD_UNLOAD:
|
2000-07-20 12:12:41 +00:00
|
|
|
/*
|
2005-02-18 18:49:44 +00:00
|
|
|
* Do not support unloading yet. we don't keep track of
|
|
|
|
* refcounts and unloading an accept filter callback and then
|
|
|
|
* having it called is a bad thing. A simple fix would be to
|
|
|
|
* track the refcount in the struct accept_filter.
|
2000-07-20 12:12:41 +00:00
|
|
|
*/
|
2000-09-06 18:49:13 +00:00
|
|
|
if (unloadable != 0) {
|
|
|
|
error = accept_filt_del(accfp->accf_name);
|
|
|
|
} else
|
|
|
|
error = EOPNOTSUPP;
|
2000-06-20 01:09:23 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MOD_SHUTDOWN:
|
|
|
|
error = 0;
|
|
|
|
break;
|
|
|
|
|
|
|
|
default:
|
|
|
|
error = EOPNOTSUPP;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return (error);
|
|
|
|
}
|
2005-02-18 18:54:42 +00:00
|
|
|
|
2005-03-12 12:57:18 +00:00
|
|
|
int
|
2017-06-02 17:49:21 +00:00
|
|
|
accept_filt_getopt(struct socket *so, struct sockopt *sopt)
|
2005-03-12 12:57:18 +00:00
|
|
|
{
|
|
|
|
struct accept_filter_arg *afap;
|
|
|
|
int error;
|
|
|
|
|
|
|
|
error = 0;
|
2017-06-02 17:43:47 +00:00
|
|
|
afap = malloc(sizeof(*afap), M_TEMP, M_WAITOK | M_ZERO);
|
2005-03-12 12:57:18 +00:00
|
|
|
SOCK_LOCK(so);
|
|
|
|
if ((so->so_options & SO_ACCEPTCONN) == 0) {
|
|
|
|
error = EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
if (so->sol_accept_filter == NULL) {
|
2005-06-11 11:59:48 +00:00
|
|
|
error = EINVAL;
|
2005-03-12 12:57:18 +00:00
|
|
|
goto out;
|
2005-06-11 11:59:48 +00:00
|
|
|
}
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
strcpy(afap->af_name, so->sol_accept_filter->accf_name);
|
|
|
|
if (so->sol_accept_filter_str != NULL)
|
|
|
|
strcpy(afap->af_arg, so->sol_accept_filter_str);
|
2005-03-12 12:57:18 +00:00
|
|
|
out:
|
|
|
|
SOCK_UNLOCK(so);
|
|
|
|
if (error == 0)
|
|
|
|
error = sooptcopyout(sopt, afap, sizeof(*afap));
|
2008-10-23 15:53:51 +00:00
|
|
|
free(afap, M_TEMP);
|
2005-03-12 12:57:18 +00:00
|
|
|
return (error);
|
|
|
|
}
|
|
|
|
|
2005-02-18 18:54:42 +00:00
|
|
|
int
|
2017-06-02 17:49:21 +00:00
|
|
|
accept_filt_setopt(struct socket *so, struct sockopt *sopt)
|
2005-02-18 18:54:42 +00:00
|
|
|
{
|
2005-02-18 19:01:22 +00:00
|
|
|
struct accept_filter_arg *afap;
|
|
|
|
struct accept_filter *afp;
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
char *accept_filter_str = NULL;
|
|
|
|
void *accept_filter_arg = NULL;
|
|
|
|
int error;
|
2005-02-18 18:54:42 +00:00
|
|
|
|
|
|
|
/*
|
2005-03-11 21:37:45 +00:00
|
|
|
* Handle the simple delete case first.
|
2005-02-18 18:54:42 +00:00
|
|
|
*/
|
2005-06-11 11:59:48 +00:00
|
|
|
if (sopt == NULL || sopt->sopt_val == NULL) {
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
struct socket *sp, *sp1;
|
|
|
|
int wakeup;
|
|
|
|
|
2005-03-11 21:37:45 +00:00
|
|
|
SOCK_LOCK(so);
|
|
|
|
if ((so->so_options & SO_ACCEPTCONN) == 0) {
|
|
|
|
SOCK_UNLOCK(so);
|
|
|
|
return (EINVAL);
|
|
|
|
}
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
if (so->sol_accept_filter == NULL) {
|
|
|
|
SOCK_UNLOCK(so);
|
|
|
|
return (0);
|
2005-02-18 18:54:42 +00:00
|
|
|
}
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
if (so->sol_accept_filter->accf_destroy != NULL)
|
|
|
|
so->sol_accept_filter->accf_destroy(so);
|
|
|
|
if (so->sol_accept_filter_str != NULL)
|
|
|
|
free(so->sol_accept_filter_str, M_ACCF);
|
|
|
|
so->sol_accept_filter = NULL;
|
|
|
|
so->sol_accept_filter_arg = NULL;
|
|
|
|
so->sol_accept_filter_str = NULL;
|
2005-02-18 18:54:42 +00:00
|
|
|
so->so_options &= ~SO_ACCEPTFILTER;
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Move from incomplete queue to complete only those
|
|
|
|
* connections, that are blocked by us.
|
|
|
|
*/
|
|
|
|
wakeup = 0;
|
|
|
|
TAILQ_FOREACH_SAFE(sp, &so->sol_incomp, so_list, sp1) {
|
|
|
|
SOCK_LOCK(sp);
|
|
|
|
if (sp->so_options & SO_ACCEPTFILTER) {
|
|
|
|
TAILQ_REMOVE(&so->sol_incomp, sp, so_list);
|
|
|
|
TAILQ_INSERT_TAIL(&so->sol_comp, sp, so_list);
|
|
|
|
sp->so_qstate = SQ_COMP;
|
|
|
|
sp->so_options &= ~SO_ACCEPTFILTER;
|
|
|
|
so->sol_incqlen--;
|
|
|
|
so->sol_qlen++;
|
|
|
|
wakeup = 1;
|
|
|
|
}
|
|
|
|
SOCK_UNLOCK(sp);
|
|
|
|
}
|
|
|
|
if (wakeup)
|
|
|
|
solisten_wakeup(so); /* unlocks */
|
|
|
|
else
|
|
|
|
SOLISTEN_UNLOCK(so);
|
2005-02-18 18:54:42 +00:00
|
|
|
return (0);
|
|
|
|
}
|
2005-03-11 21:37:45 +00:00
|
|
|
|
|
|
|
/*
|
2005-03-12 12:27:47 +00:00
|
|
|
* Pre-allocate any memory we may need later to avoid blocking at
|
|
|
|
* untimely moments. This does not optimize for invalid arguments.
|
2005-02-18 18:54:42 +00:00
|
|
|
*/
|
2017-06-02 17:43:47 +00:00
|
|
|
afap = malloc(sizeof(*afap), M_TEMP, M_WAITOK);
|
2005-02-18 18:54:42 +00:00
|
|
|
error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
|
|
|
|
afap->af_name[sizeof(afap->af_name)-1] = '\0';
|
|
|
|
afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
|
|
|
|
if (error) {
|
2008-10-23 15:53:51 +00:00
|
|
|
free(afap, M_TEMP);
|
2005-02-18 18:54:42 +00:00
|
|
|
return (error);
|
|
|
|
}
|
|
|
|
afp = accept_filt_get(afap->af_name);
|
|
|
|
if (afp == NULL) {
|
2008-10-23 15:53:51 +00:00
|
|
|
free(afap, M_TEMP);
|
2005-02-18 18:54:42 +00:00
|
|
|
return (ENOENT);
|
|
|
|
}
|
|
|
|
if (afp->accf_create != NULL && afap->af_name[0] != '\0') {
|
2016-10-12 00:56:49 +00:00
|
|
|
size_t len = strlen(afap->af_name) + 1;
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
accept_filter_str = malloc(len, M_ACCF, M_WAITOK);
|
|
|
|
strcpy(accept_filter_str, afap->af_name);
|
2005-02-18 18:54:42 +00:00
|
|
|
}
|
|
|
|
|
2005-03-12 12:27:47 +00:00
|
|
|
/*
|
|
|
|
* Require a listen socket; don't try to replace an existing filter
|
|
|
|
* without first removing it.
|
|
|
|
*/
|
2005-02-18 18:54:42 +00:00
|
|
|
SOCK_LOCK(so);
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
if ((so->so_options & SO_ACCEPTCONN) == 0 ||
|
|
|
|
so->sol_accept_filter != NULL) {
|
2005-02-18 18:54:42 +00:00
|
|
|
error = EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
2005-03-12 12:27:47 +00:00
|
|
|
|
2005-02-18 18:54:42 +00:00
|
|
|
/*
|
2005-03-12 12:27:47 +00:00
|
|
|
* Invoke the accf_create() method of the filter if required. The
|
|
|
|
* socket mutex is held over this call, so create methods for filters
|
|
|
|
* can't block.
|
2005-02-18 18:54:42 +00:00
|
|
|
*/
|
|
|
|
if (afp->accf_create != NULL) {
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
accept_filter_arg = afp->accf_create(so, afap->af_arg);
|
|
|
|
if (accept_filter_arg == NULL) {
|
2005-02-18 18:54:42 +00:00
|
|
|
error = EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
so->sol_accept_filter = afp;
|
|
|
|
so->sol_accept_filter_arg = accept_filter_arg;
|
|
|
|
so->sol_accept_filter_str = accept_filter_str;
|
2005-02-18 18:54:42 +00:00
|
|
|
so->so_options |= SO_ACCEPTFILTER;
|
|
|
|
out:
|
|
|
|
SOCK_UNLOCK(so);
|
Listening sockets improvements.
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
|
|
|
if (accept_filter_str != NULL)
|
|
|
|
free(accept_filter_str, M_ACCF);
|
|
|
|
free(afap, M_TEMP);
|
2005-02-18 18:54:42 +00:00
|
|
|
return (error);
|
|
|
|
}
|