2000-02-03 Roland McGrath <roland@baalperazim.frob.com>

	Complete overhaul of pfinet based on the IPv4 networking code from the
	Linux 2.2.12 kernel sources.  This page describes a single unified set
	of interdependent changes, but there are so many changes that I have
	broken up the log entry into paragraphs based on rough topical
	divisions of the work involved.

	Subset of verbatim Linux 2.2.12 sources imported on a vendor branch.
	* linux-src: New directory, see README for details.
	* README: New file, describes linux-src layout and procedures for
	tracking Linux source updates.

	Light modifications to linux-src files to avoid really bending over
	backwards with the glue macros.  All modifications to files in
	linux-src are conditionalized by #ifdef _HURD_.
	* linux-src/include/linux/net.h [_HURD_] (struct socket): New members
	refcnt and identity; elide members fasync_list, file.
	* linux-src/include/linux/rtnetlink.h [! CONFIG_RTNETLINK]
	(rtnl_shlock, rtnl_shunlock) [! _HURD_]: Conditionalize contents on
	this, making these no-ops #ifdef _HURD_.
	* linux-src/net/core/dev.c [_HURD_] (dev_ioctl): Don't define the
	function, instead #define it to 0.
	* linux-src/net/ipv4/af_inet.c [_HURD_] (inet_ioctl): Likewise.
	* linux-src/net/ipv4/arp.c [_HURD_] (arp_ioctl): Likewise.
	* linux-src/net/ipv4/udp.c [_HURD_] (udp_ioctl): Likewise.
	* linux-src/net/ipv4/tcp.c [_HURD_] (tcp_ioctl): Likewise.
	[_HURD_] (tcp_tiocinq): New function, TIOCINQ code from tcp_ioctl.
	* linux-src/net/ipv4/devinet.c [_HURD_] (devinet_ioctl): Don't define
	the function, instead #define it to 0.
	[_HURD_] (configure_device): New function, cobbled from SIOCSIFADDR
	and SIOCSIFNETMASK code from devinet_ioctl.

	* glue-include/asm, glue-include/linux: New directories.
	These contain glue kludge headers that replace all of the
	Linux <asm/*.h> headers except checksum.h, and several of
	the Linux <linux/*.h> headers (the remainder come from
	linux-src/include/linux and are mostly unmodified).
	* glue-include/asm/atomic.h: New file, glue replacement header.
	* glue-include/asm/bitops.h: New file, glue replacement header.
	* glue-include/asm/byteorder.h: New file, glue replacement header.
	* glue-include/asm/checksum.h: New file, glue replacement header.
	* glue-include/asm/errno.h: New file, glue replacement header.
	* glue-include/asm/hardirq.h: New file, glue replacement header.
	* glue-include/asm/init.h: New file, glue replacement header.
	* glue-include/asm/segment.h: New file, glue replacement header.
	* glue-include/asm/spinlock.h: New file, glue replacement header.
	* glue-include/asm/system.h: New file, glue replacement header.
	* glue-include/asm/types.h: New file, glue replacement header.
	* glue-include/asm/uaccess.h: New file, glue replacement header.
	* glue-include/linux/autoconf.h: New file, glue replacement header.
	* glue-include/linux/binfmts.h: New file, glue replacement header.
	* glue-include/linux/config.h: New file, glue replacement header.
	* glue-include/linux/errno.h: New file, glue replacement header.
	* glue-include/linux/fcntl.h: New file, glue replacement header.
	* glue-include/linux/fs.h: New file, glue replacement header.
	* glue-include/linux/in.h: New file, glue replacement header.
	* glue-include/linux/in6.h: New file, glue replacement header.
	* glue-include/linux/interrupt.h: New file, glue replacement header.
	* glue-include/linux/ioctl.h: New file, glue replacement header.
	* glue-include/linux/ipv6.h: New file, glue replacement header.
	* glue-include/linux/kernel.h: New file, glue replacement header.
	* glue-include/linux/limits.h: New file, glue replacement header.
	* glue-include/linux/major.h: New file, glue replacement header.
	* glue-include/linux/malloc.h: New file, glue replacement header.
	* glue-include/linux/mm.h: New file, glue replacement header.
	* glue-include/linux/param.h: New file, glue replacement header.
	* glue-include/linux/personality.h: New file, glue replacement header.
	* glue-include/linux/poll.h: New file, glue replacement header.
	* glue-include/linux/proc_fs.h: New file, glue replacement header.
	* glue-include/linux/sched.h: New file, glue replacement header.
	* glue-include/linux/slab.h: New file, glue replacement header.
	* glue-include/linux/socket.h: New file, glue replacement header.
	* glue-include/linux/sockios.h: New file, glue replacement header.
	* glue-include/linux/stat.h: New file, glue replacement header.
	* glue-include/linux/string.h: New file, glue replacement header.
	* glue-include/linux/termios.h: New file, glue replacement header.
	* glue-include/linux/time.h: New file, glue replacement header.
	* glue-include/linux/timer.h: New file, glue replacement header.
	* glue-include/linux/timex.h: New file, glue replacement header.
	* glue-include/linux/types.h: New file, glue replacement header.
	* glue-include/linux/un.h: New file, glue replacement header.
	* glue-include/linux/version.h: New file, glue replacement header.
	* glue-include/linux/wait.h: New file, glue replacement header.

	* kmem_cache.c: New file.  Glue code replaces Linux kmem_cache_t et al.
	* stubs.c: New file.  No-op functions and stub variables for a few
	things the Linux networking code needs to link.

	* Makefile (core-srcs, arch-lib-srcs, ethernet-srcs, ipv4-srcs): New
	variables, listing sources used from linux-src subdirectories.
	(LINUXSRCS): Define using those.
	(SRCS): Remove devices.c; add kmem_cache.c, stubs.c.
	(UNUSEDSRC): Variable removed.
	(vpath %.c): Remove vpath for $(srcdir)/linux-inet directory.
	Add vpaths for $(srcdir)/linux-src subdirectories.
	(CPPFLAGS): Add -D_HURD_SYSTYPE defining it to $(asm_syntax) as a
	double-quoted string.  Add -I's for glue-include and linux-src/include.

	* pfinet.h: Include <sys/socket.h>, and not <linux/netdevice.h>.
	(master_device): Remove decl.
	(global_lock, packet_queue_lock): Remove common defns.
	(global_lock, net_bh_lock): Declare them as externs.
	(struct sockaddr): Remove len member, make address member just a
	struct sockaddr rather than a 0-length array.
	(setup_loopback_device, become_task_protid, become_task): Remove decls.
	(ethernet_initialize): Declare it.
	(input_work_thread): Remove decl.
	(net_bh_thread): Declare it.
	(tcp_readable): Remove decl.
	(tcp_tiocinq): Declare it.

	* config.h: Rewritten based on Linux 2.2.12 set of CONFIG_* options.
	(CONFIG_NET, CONFIG_INET, CONFIG_SKB_LARGE): These are the only
	Linux config options we set.
	(CONFIG_IP_NOSIOCRT): New macro (not a proper config option, but
	used conveniently in the code).
	* ethernet.c (ethernet_set_multi): Take only one parameter.
	Remove assert, since we always get passed IGMP_ALL_HOSTS.
	(ethernet_thread): Make static.
	(ethernet_demuxer): Use __mutex_lock in place of mutex_lock, so as to
	get cthreads instead of linux/spinlock.h glue macros.  Lock
	net_bh_lock instead of global_lock.  Set SKB->protocol with
	eth_type_trans before calling netif_rx.
	(ethernet_initialize): New function, one-time initialization broken
	out of ethernet_open.
	(ethernet_open): Ports setup moved to ethernet_initialize.
	Don't use `errno' to avoid glue conflicts.
	Use get_privileged_ports here to get the master device port, and
	deallocate it after calling device_open.
	(ethernet_xmit): Use assert_perror.  Only one arg to dev_kfree_skb now.
	(setup_ethernet_device): Change initializations for structure changes.
	Call dev_init_buffers and register_netdevice on the device.
	* timer-emul.c (all functions): Use __mutex_lock instead of mutex_lock.
	Adjust for renaming of `prevp' member to `prev' in struct timer_list.
	(mod_timer): New function.
	* socket.c (proto_ops): Variable removed.
	(net_families): New variable replaces it.
	(sock_register): Rewritten for new calling convention, set
	net_families rather than proto_ops.
	(make_sock_user, clean_socketport, sock_alloc, sock_release):
	Functions moved here from misc.c.
	* sched.c (packet_queue_lock): Variable removed.
	(net_bh_lock, net_bh_wakeup): New variables.
	(current): Variable removed (now a macro in the glue headers).
	(interruptible_sleep_on, wake_up_interruptible): Functions removed.
	They are replaced by inlines in the glue headers.
	(become_task, become_task_protid): Functions removed; they are
	replaced by macros in glue-include/linux/sched.h.
	(net_bh_worker): New function.
	* loopback.c: Completely rewritten, mostly copied from linux-2.2.12's
	drivers/net/loopback.c source file.

	* io-ops.c (all functions): Use __mutex_lock in place of mutex_lock.
	(S_io_write): Call ops->sendmsg instead of ops->write,
	which no longer exists.  If O_NONBLOCK is set, set MSG_DONTWAIT in
	msg_flags.
	(S_io_read): Call ops->recvmsg instead of ops->read,
	which no longer exists  If O_NONBLOCK is set, pass MSG_DONTWAIT.
	(S_io_readable): Use USER->sock->data in place of USER->sock->sk.
	For SOCK_STREAM and SOCK_SEQPACKET types, call tcp_tiocinq.
	(S_io_set_all_openmodes, S_io_get_openmodes, S_io_set_some_openmodes,
	S_io_clear_some_openmodes): Member USER->sock->userflags is now
	renamed USER->sock->flags.
	(S_io_select): Completely rewritten using ops->poll.
	(select_wait): Function removed.
	(S_io_stat): Set st_mode to reflect S_IFSOCK.
	* socket-ops.c (all functions): Use __mutex_lock instead of mutex_lock.
	(S_socket_create): Don't set SOCK->ops or call SOCK->ops->create.
	Instead, call net_families[PF_INET]->create.
	(S_socket_listen): Remove extra checks; just call ops->listen.
	(S_socket_accept): Remove extra checks before ops->accept call.
	Avoid use of goto.
	(S_socket_connect): Remove extra checks; just call ops->connect.
	(S_socket_bind): Adjust for struct sock_addr changes.
	(S_socket_create_address): Likewise.
	(S_socket_whatis_address): Likewise.
	(S_socket_connect2): Don't diddle data structures after
	ops->socketpair call.
	(S_socket_getopt): Use sock_getsockopt if LEVEL is SOL_SOCKET.
	Accept any data size, not just sizeof (int).
	(S_socket_setopt): Use sock_setsockopt if LEVEL is SOL_SOCKET.
	(S_socket_send): Always use ops->sendmsg instead of ops->send or
	ops->sendto, which no longer exist.  If O_NONBLOCK is set, set
	MSG_DONTWAIT in msg_flags.
	(S_socket_recv): Always use ops->recvmsg instead of ops->recv, which
	no longer exists.  If O_NONBLOCK is set, set MSG_DONTWAIT in flags.
	Check for error from S_socket_create_address.

	* main.c (find_device): Don't try to set ether_dev.pa_mask (it's gone).
	(main): Don't call init_devices.  Call ethernet_initialize.
	Start net_bh_worker instead of input_work_thread.  Don't call
	setup_loopback_device.  Instead, take global_lock, do prepare_current,
	and then call sk_init, skb_init, inet_proto_init, and net_dev_init.
	Keep global_lock held while calling argp_parse.
	Call arrange_shutdown_notification only after all that.
	Fix error call for "contacting parent" to pass ERR instead of errno.
	* options.c (ADDR): #undef before defining macro.
	(parse_opt): #if 0 out EDESTADDRREQ check (I don't understand it).
	To apply settings, call configure_devices.
	(ADD_ADDR_OPT): #if 0 --address and --netmask options.  Needs fixed.
	* misc.c (make_sock_user, clean_socketport, sock_alloc, sock_release):
	Functions moved to socket.c.
	(sock_release_peer): Function removed.
	(make_sockaddr_port): Use struct sockaddr_storage to size buffer.
	Fix size calculation for new struct sock_addr layout.
	Initialize sa_family and sa_len of new struct sock_addr.

	Remove the old Linux (2.0.??) network stack and the glue code for it.
	* linux-inet, asm, linux: Directories and all files removed.
	Some of the new files in glue-include came from the old glue headers
	in the asm and linux directories, but most were substantially modified.
	* devices.c: File removed.  The equivalent glue is now elsewhere.

1 files changed, 148 insertions, 161 deletions

diff --git a/pfinet/socket-ops.c b/pfinet/socket-ops.c
index 3db4985f..bcfa7f45 100644
--- a/pfinet/socket-ops.c
+++ b/pfinet/socket-ops.c
@@ -1,5 +1,5 @@
 /* Interface functions for the socket.defs interface.
-   Copyright (C) 1995, 1996, 1997, 1999 Free Software Foundation, Inc.
+   Copyright (C) 1995,96,97,99,2000 Free Software Foundation, Inc.
    Written by Michael I. Bushnell, p/BSG.
 
    This file is part of the GNU Hurd.
@@ -20,11 +20,18 @@
 
 #include <hurd/trivfs.h>
 #include <string.h>
+#include <stddef.h>
+#include <fcntl.h>
 
 #include "pfinet.h"
 #include "socket_S.h"
 
-
+#include <linux/sched.h>
+#include <linux/socket.h>
+#include <linux/net.h>
+#include <net/sock.h>
+
+
 error_t
 S_socket_create (struct trivfs_protid *master,
 		 int sock_type,
@@ -48,16 +55,15 @@ S_socket_create (struct trivfs_protid *master,
       || protocol < 0)
     return EINVAL;
 
-  mutex_lock (&global_lock);
+  __mutex_lock (&global_lock);
 
   become_task_protid (master);
 
   sock = sock_alloc ();
 
   sock->type = sock_type;
-  sock->ops = proto_ops;
 
-  err = - (*sock->ops->create) (sock, protocol);
+  err = - (*net_families[PF_INET]->create) (sock, protocol);
   if (err)
     sock_release (sock);
   else
@@ -68,7 +74,7 @@ S_socket_create (struct trivfs_protid *master,
       ports_port_deref (user);
     }
 
-  mutex_unlock (&global_lock);
+  __mutex_unlock (&global_lock);
 
   return err;
 }
@@ -78,26 +84,17 @@ S_socket_create (struct trivfs_protid *master,
 error_t
 S_socket_listen (struct sock_user *user, int queue_limit)
 {
+  error_t err;
+
   if (!user)
     return EOPNOTSUPP;
 
-  mutex_lock (&global_lock);
-
+  __mutex_lock (&global_lock);
   become_task (user);
+  err = - (*user->sock->ops->listen) (user->sock, queue_limit);
+  __mutex_unlock (&global_lock);
 
-  if (user->sock->state == SS_UNCONNECTED)
-    {
-      if (user->sock->ops && user->sock->ops->listen)
-	(*user->sock->ops->listen) (user->sock, queue_limit);
-      user->sock->flags |= SO_ACCEPTCON;
-      mutex_unlock (&global_lock);
-      return 0;
-    }
-  else
-    {
-      mutex_unlock (&global_lock);
-      return EINVAL;
-    }
+  return err;
 }
 
 error_t
@@ -114,47 +111,43 @@ S_socket_accept (struct sock_user *user,
   if (!user)
     return EOPNOTSUPP;
 
-  mutex_lock (&global_lock);
-
-  become_task (user);
-
   sock = user->sock;
-  newsock = 0;
-  err = 0;
 
-  if ((sock->state != SS_UNCONNECTED)
-      || (!(sock->flags & SO_ACCEPTCON)))
-    err = EINVAL;
-  else if (!(newsock = sock_alloc ()))
-    err = ENOMEM;
-
-  if (err)
-    goto out;
+  __mutex_lock (&global_lock);
 
-  newsock->type = sock->type;
-  newsock->ops = sock->ops;
-
-  err = - (*sock->ops->dup) (newsock, sock);
-  if (err)
-    goto out;
-
-  err = - (*sock->ops->accept) (sock, newsock, sock->userflags);
-  if (err)
-    goto out;
+  become_task (user);
 
-  err = make_sockaddr_port (newsock, 1, addr_port, addr_port_type);
-  if (err)
-    goto out;
+  newsock = sock_alloc ();
+  if (!newsock)
+    err = ENOMEM;
+  else
+    {
+      newsock->type = sock->type;
+
+      err = - (*sock->ops->dup) (newsock, sock);
+      if (!err)
+	err = - (*sock->ops->accept) (sock, newsock, sock->flags);
+
+      if (!err)
+	/* In Linux there is a race here with the socket closing before the
+	   ops->getname call we do in make_sockaddr_port.  Since we still
+	   have the world locked, this shouldn't be an issue for us.  */
+	err = make_sockaddr_port (newsock, 1, addr_port, addr_port_type);
+
+      if (!err)
+	{
+	  newuser = make_sock_user (newsock, user->isroot, 0);
+	  *new_port = ports_get_right (newuser);
+	  *new_port_type = MACH_MSG_TYPE_MAKE_SEND;
+	  ports_port_deref (newuser);
+	}
+
+      if (err)
+	sock_release (newsock);
+    }
 
-  newuser = make_sock_user (newsock, user->isroot, 0);
-  *new_port = ports_get_right (newuser);
-  *new_port_type = MACH_MSG_TYPE_MAKE_SEND;
-  ports_port_deref (newuser);
+  __mutex_unlock (&global_lock);
 
- out:
-  if (err && newsock)
-    sock_release (newsock);
-  mutex_unlock (&global_lock);
   return err;
 }
 
@@ -170,25 +163,14 @@ S_socket_connect (struct sock_user *user,
 
   sock = user->sock;
 
-  mutex_lock (&global_lock);
+  __mutex_lock (&global_lock);
 
   become_task (user);
 
-  err = 0;
-
-  if (sock->state == SS_CONNECTED
-      && sock->type != SOCK_DGRAM)
-    err = EISCONN;
-  else if (sock->state != SS_UNCONNECTED
-	   && sock->state != SS_CONNECTING
-	   && sock->state != SS_CONNECTED)
-    err = EINVAL;
-
-  if (!err)
-    err = - (*sock->ops->connect) (sock, addr->address, addr->len,
-				   sock->userflags);
+  err = - (*sock->ops->connect) (sock, &addr->address, addr->address.sa_len,
+				 sock->flags);
 
-  mutex_unlock (&global_lock);
+  __mutex_unlock (&global_lock);
 
   /* MiG should do this for us, but it doesn't. */
   if (!err)
@@ -208,10 +190,11 @@ S_socket_bind (struct sock_user *user,
   if (! addr)
     return EADDRNOTAVAIL;
 
-  mutex_lock (&global_lock);
+  __mutex_lock (&global_lock);
   become_task (user);
-  err = - (*user->sock->ops->bind) (user->sock, addr->address, addr->len);
-  mutex_unlock (&global_lock);
+  err = - (*user->sock->ops->bind) (user->sock,
+				    &addr->address, addr->address.sa_len);
+  __mutex_unlock (&global_lock);
 
   /* MiG should do this for us, but it doesn't. */
   if (!err)
@@ -228,10 +211,10 @@ S_socket_name (struct sock_user *user,
   if (!user)
     return EOPNOTSUPP;
 
-  mutex_lock (&global_lock);
+  __mutex_lock (&global_lock);
   become_task (user);
   make_sockaddr_port (user->sock, 0, addr_port, addr_port_name);
-  mutex_unlock (&global_lock);
+  __mutex_unlock (&global_lock);
   return 0;
 }
 
@@ -245,10 +228,10 @@ S_socket_peername (struct sock_user *user,
   if (!user)
     return EOPNOTSUPP;
 
-  mutex_lock (&global_lock);
+  __mutex_lock (&global_lock);
   become_task (user);
   err = make_sockaddr_port (user->sock, 1, addr_port, addr_port_name);
-  mutex_unlock (&global_lock);
+  __mutex_unlock (&global_lock);
 
   return err;
 }
@@ -262,7 +245,7 @@ S_socket_connect2 (struct sock_user *user1,
   if (!user1 || !user2)
     return EOPNOTSUPP;
 
-  mutex_lock (&global_lock);
+  __mutex_lock (&global_lock);
 
   become_task (user1);
 
@@ -274,15 +257,7 @@ S_socket_connect2 (struct sock_user *user1,
   else
     err = - (*user1->sock->ops->socketpair) (user1->sock, user2->sock);
 
-  if (!err)
-    {
-      user1->sock->conn = user2->sock;
-      user2->sock->conn = user1->sock;
-      user1->sock->state = SS_CONNECTED;
-      user2->sock->state = SS_CONNECTED;
-    }
-
-  mutex_unlock (&global_lock);
+  __mutex_unlock (&global_lock);
 
   /* MiG should do this for us, but it doesn't. */
   if (!err)
@@ -299,23 +274,30 @@ S_socket_create_address (mach_port_t server,
 			 mach_port_t *addr_port,
 			 mach_msg_type_name_t *addr_port_type)
 {
-  struct sock_addr *addr;
   error_t err;
+  struct sock_addr *addrstruct;
+  const struct sockaddr *const sa = (void *) data;
 
   if (sockaddr_type != AF_INET)
     return EAFNOSUPPORT;
+  if (sa->sa_family != sockaddr_type
+      || data_len < offsetof (struct sockaddr, sa_data))
+    return EINVAL;
 
   err = ports_create_port (addrport_class, pfinet_bucket,
-			   sizeof (struct sock_addr) + data_len, &addr);
+			   (offsetof (struct sock_addr, address)
+			    + data_len), &addrstruct);
   if (err)
     return err;
 
-  addr->len = data_len;
-  bcopy (data, addr->address, data_len);
+  memcpy (&addrstruct->address, data, data_len);
+
+  /* BSD does not require incoming sa_len to be set, so we don't either.  */
+  addrstruct->address.sa_len = data_len;
 
-  *addr_port = ports_get_right (addr);
+  *addr_port = ports_get_right (addrstruct);
   *addr_port_type = MACH_MSG_TYPE_MAKE_SEND;
-  ports_port_deref (addr);
+  ports_port_deref (addrstruct);
   return 0;
 }
 
@@ -337,11 +319,12 @@ S_socket_whatis_address (struct sock_addr *addr,
   if (!addr)
     return EOPNOTSUPP;
 
-  *type = AF_INET;
-  if (*datalen < addr->len)
-    *data = mmap (0, addr->len, PROT_READ|PROT_WRITE, MAP_ANON, 0, 0);
-  bcopy (addr->address, *data, addr->len);
-  *datalen = addr->len;
+  *type = addr->address.sa_family;
+  if (*datalen < addr->address.sa_len)
+    *data = mmap (0, addr->address.sa_len,
+		  PROT_READ|PROT_WRITE, MAP_ANON, 0, 0);
+  *datalen = addr->address.sa_len;
+  memcpy (*data, &addr->address, addr->address.sa_len);
 
   return 0;
 }
@@ -355,10 +338,10 @@ S_socket_shutdown (struct sock_user *user,
   if (!user)
     return EOPNOTSUPP;
 
-  mutex_lock (&global_lock);
+  __mutex_lock (&global_lock);
   become_task (user);
   err = - (*user->sock->ops->shutdown) (user->sock, direction);
-  mutex_unlock (&global_lock);
+  __mutex_unlock (&global_lock);
 
   return err;
 }
@@ -375,18 +358,17 @@ S_socket_getopt (struct sock_user *user,
   if (! user)
     return EOPNOTSUPP;
 
-  /* XXX all options supported in the linux code are in fact ints.  */
-  *datalen = sizeof (int);
-
-  mutex_lock (&global_lock);
+  __mutex_lock (&global_lock);
   become_task (user);
 
-  err =
-    - (user->sock->ops->getsockopt)(user->sock, level, option, *data, datalen);
+  err = - (level == SOL_SOCKET ? sock_getsockopt
+	   : *user->sock->ops->getsockopt)
+    (user->sock, level, option, *data, datalen);
 
-  assert (*datalen == sizeof (int));
+  __mutex_unlock (&global_lock);
 
-  mutex_unlock (&global_lock);
+  /* XXX option data not properly typed, needs byte-swapping for netmsgserver.
+     Most options are ints, some like IP_OPTIONS are bytesex-neutral.  */
 
   return err;
 }
@@ -403,13 +385,17 @@ S_socket_setopt (struct sock_user *user,
   if (! user)
     return EOPNOTSUPP;
 
-  mutex_lock (&global_lock);
+  /* XXX option data not properly typed, needs byte-swapping for netmsgserver.
+     Most options are ints, some like IP_OPTIONS are bytesex-neutral.  */
+
+  __mutex_lock (&global_lock);
   become_task (user);
 
-  err =
-    - (user->sock->ops->setsockopt)(user->sock, level, option, data, datalen);
+  err = - (level == SOL_SOCKET ? sock_setsockopt
+	   : *user->sock->ops->setsockopt)
+    (user->sock, level, option, data, datalen);
 
-  mutex_unlock (&global_lock);
+  __mutex_unlock (&global_lock);
 
   return err;
 }
@@ -427,6 +413,11 @@ S_socket_send (struct sock_user *user,
 	       mach_msg_type_number_t *amount)
 {
   int sent;
+  struct iovec iov = { data, datalen };
+  struct msghdr m = { msg_name: addr ? &addr->address : 0,
+		      msg_namelen: addr ? addr->address.sa_len : 0,
+		      msg_flags: flags,
+		      msg_controllen: 0, msg_iov: &iov, msg_iovlen: 1 };
 
   if (!user)
     return EOPNOTSUPP;
@@ -435,19 +426,12 @@ S_socket_send (struct sock_user *user,
   if (nports != 0 || controllen != 0)
     return EINVAL;
 
-  mutex_lock (&global_lock);
-
+  __mutex_lock (&global_lock);
   become_task (user);
-
-  if (addr)
-    sent = (*user->sock->ops->sendto) (user->sock, data, datalen,
-				       user->sock->userflags, flags,
-				       addr->address, addr->len);
-  else
-    sent = (*user->sock->ops->send) (user->sock, data, datalen,
-				     user->sock->userflags, flags);
-
-  mutex_unlock (&global_lock);
+  if (user->sock->flags & O_NONBLOCK)
+    m.msg_flags |= MSG_DONTWAIT;
+  sent = (*user->sock->ops->sendmsg) (user->sock, &m, datalen, 0);
+  __mutex_unlock (&global_lock);
 
   /* MiG should do this for us, but it doesn't. */
   if (addr && sent >= 0)
@@ -477,50 +461,53 @@ S_socket_recv (struct sock_user *user,
 	       int *outflags,
 	       mach_msg_type_number_t amount)
 {
-  int recvd;
-  char addr[128];
-  size_t addrlen = sizeof addr;
-  int didalloc = 0;
+  error_t err;
+  union { struct sockaddr_storage storage; struct sockaddr sa; } addr;
+  int alloced = 0;
+  struct iovec iov;
+  struct msghdr m = { msg_name: &addr.sa, msg_namelen: sizeof addr,
+		      msg_controllen: 0, msg_iov: &iov, msg_iovlen: 1 };
 
   if (!user)
     return EOPNOTSUPP;
 
-  /* For unused recvmsg interface */
-  *nports = 0;
-  *portstype = MACH_MSG_TYPE_COPY_SEND;
-  *controllen = 0;
-  *outflags = 0;
-
-  /* Instead of this, we should peek at the socket and only allocate
-     as much as necessary. */
-  if (*datalen < amount)
+  /* Instead of this, we should peek and the socket and only
+     allocate as much as necessary. */
+  if (amount > *datalen)
     {
-      vm_allocate (mach_task_self (), (vm_address_t *) data, amount, 1);
-      didalloc = 1;
+      *data = mmap (0, amount, PROT_READ|PROT_WRITE, MAP_ANON, 0, 0);
+      alloced = 1;
     }
 
-  mutex_lock (&global_lock);
-  become_task (user);
-
-  recvd =  (*user->sock->ops->recvfrom) (user->sock, *data, amount,
-					 user->sock->userflags, flags,
-					 (struct sockaddr *)addr, &addrlen);
-
-  mutex_unlock (&global_lock);
-
-  if (recvd < 0)
-    return (error_t)-recvd;
-
-  *datalen = recvd;
-
-  if (didalloc && round_page (*datalen) < round_page (amount))
-    vm_deallocate (mach_task_self (),
-		   (vm_address_t) (*data + round_page (*datalen)),
-		   round_page (amount) - round_page (*datalen));
+  iov.iov_base = *data;
+  iov.iov_len = amount;
 
+  __mutex_lock (&global_lock);
+  become_task (user);
+  if (user->sock->flags & O_NONBLOCK)
+    flags |= MSG_DONTWAIT;
+  err = (*user->sock->ops->recvmsg) (user->sock, &m, amount, flags, 0);
+  __mutex_unlock (&global_lock);
 
-  S_socket_create_address (0, AF_INET, addr, addrlen, addrport,
-			   addrporttype);
+  if (err < 0)
+    err = -err;
+  else
+    {
+      *datalen = err;
+      if (alloced && round_page (*datalen) < round_page (amount))
+	munmap (*data + round_page (*datalen),
+		round_page (amount) - round_page (*datalen));
+      err = S_socket_create_address (0, addr.sa.sa_family,
+				     (void *) &addr.sa, m.msg_namelen,
+				     addrport, addrporttype);
+      if (err && alloced)
+	munmap (*data, *datalen);
+
+      *outflags = m.msg_flags;
+      *nports = 0;
+      *portstype = MACH_MSG_TYPE_COPY_SEND;
+      *controllen = 0;
+    }
 
-  return 0;
+  return err;
 }