/*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2008 Apple Inc. All rights reserved.
*
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License"). You may not use this file except in compliance with the
- * License. Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
*
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ *
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
- * License for the specific language governing rights and limitations
- * under the License.
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
*
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
* Copyright (c) 1982, 1986, 1988, 1993
static struct tcpcb *
tcp_usrclosed(struct tcpcb *);
+__private_extern__ int tcp_win_scale = 3;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, win_scale_factor, CTLFLAG_RW,
+ &tcp_win_scale, 0, "Window scaling factor");
+
+static u_int32_t tcps_in_sw_cksum;
+SYSCTL_UINT(_net_inet_tcp, OID_AUTO, in_sw_cksum, CTLFLAG_RD,
+ &tcps_in_sw_cksum, 0,
+ "Number of received packets checksummed in software");
+
+static u_int64_t tcps_in_sw_cksum_bytes;
+SYSCTL_QUAD(_net_inet_tcp, OID_AUTO, in_sw_cksum_bytes, CTLFLAG_RD,
+ &tcps_in_sw_cksum_bytes,
+ "Amount of received data checksummed in software");
+
+static u_int32_t tcps_out_sw_cksum;
+SYSCTL_UINT(_net_inet_tcp, OID_AUTO, out_sw_cksum, CTLFLAG_RD,
+ &tcps_out_sw_cksum, 0,
+ "Number of transmitted packets checksummed in software");
+
+static u_int64_t tcps_out_sw_cksum_bytes;
+SYSCTL_QUAD(_net_inet_tcp, OID_AUTO, out_sw_cksum_bytes, CTLFLAG_RD,
+ &tcps_out_sw_cksum_bytes,
+ "Amount of transmitted data checksummed in software");
+
#if TCPDEBUG
#define TCPDEBUG0 int ostate = 0
#define TCPDEBUG1() ostate = tp ? tp->t_state : 0
#define TCPDEBUG2(req)
#endif
+#if CONFIG_USESOCKTHRESHOLD
+__private_extern__ unsigned int tcp_sockthreshold = 64;
+#else
+__private_extern__ unsigned int tcp_sockthreshold = 0;
+#endif
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, sockthreshold, CTLFLAG_RW,
+ &tcp_sockthreshold , 0, "TCP Socket size increased if less than threshold");
+
/*
* TCP attaches to socket via pru_attach(), reserving space,
* and an internet control block.
+ *
+ * Returns: 0 Success
+ * EISCONN
+ * tcp_attach:ENOBUFS
+ * tcp_attach:ENOMEM
+ * tcp_attach:??? [IPSEC specific]
*/
static int
-tcp_usr_attach(struct socket *so, int proto, struct proc *p)
+tcp_usr_attach(struct socket *so, __unused int proto, struct proc *p)
{
int error;
struct inpcb *inp = sotoinpcb(so);
/*
* Give the socket an address.
+ *
+ * Returns: 0 Success
+ * EINVAL Invalid argument [COMMON_START]
+ * EAFNOSUPPORT Address family not supported
+ * in_pcbbind:EADDRNOTAVAIL Address not available.
+ * in_pcbbind:EINVAL Invalid argument
+ * in_pcbbind:EAFNOSUPPORT Address family not supported [notdef]
+ * in_pcbbind:EACCES Permission denied
+ * in_pcbbind:EADDRINUSE Address in use
+ * in_pcbbind:EAGAIN Resource unavailable, try again
+ * in_pcbbind:EPERM Operation not permitted
*/
static int
tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct proc *p)
COMMON_START();
+ if (nam->sa_family != 0 && nam->sa_family != AF_INET) {
+ error = EAFNOSUPPORT;
+ goto out;
+ }
+
/*
* Must check for multicast addresses and disallow binding
* to them.
COMMON_START();
+ if (nam->sa_family != 0 && nam->sa_family != AF_INET6) {
+ error = EAFNOSUPPORT;
+ goto out;
+ }
+
/*
* Must check for multicast addresses and disallow binding
* to them.
/*
* Prepare to accept connections.
+ *
+ * Returns: 0 Success
+ * EINVAL [COMMON_START]
+ * in_pcbbind:EADDRNOTAVAIL Address not available.
+ * in_pcbbind:EINVAL Invalid argument
+ * in_pcbbind:EAFNOSUPPORT Address family not supported [notdef]
+ * in_pcbbind:EACCES Permission denied
+ * in_pcbbind:EADDRINUSE Address in use
+ * in_pcbbind:EAGAIN Resource unavailable, try again
+ * in_pcbbind:EPERM Operation not permitted
*/
static int
tcp_usr_listen(struct socket *so, struct proc *p)
struct tcpcb *tp;
struct sockaddr_in *sinp;
- COMMON_START();
+ TCPDEBUG0;
+ if (inp == 0)
+ return EINVAL;
+ else if (inp->inp_state == INPCB_STATE_DEAD) {
+ if (so->so_error) {
+ error = so->so_error;
+ so->so_error = 0;
+ return error;
+ } else
+ return EINVAL;
+ }
+ tp = intotcpcb(inp);
+ TCPDEBUG1();
+ if (nam->sa_family != 0 && nam->sa_family != AF_INET) {
+ error = EAFNOSUPPORT;
+ goto out;
+ }
/*
* Must disallow TCP ``connections'' to multicast addresses.
*/
goto out;
}
-#ifndef __APPLE__
- prison_remote_ip(p, 0, &sinp->sin_addr.s_addr);
-#endif
if ((error = tcp_connect(tp, nam, p)) != 0)
goto out;
COMMON_START();
+ if (nam->sa_family != 0 && nam->sa_family != AF_INET6) {
+ error = EAFNOSUPPORT;
+ goto out;
+ }
+
/*
* Must disallow TCP ``connections'' to multicast addresses.
*/
COMMON_END(PRU_ACCEPT);
}
#endif /* INET6 */
+
/*
* Mark the connection as being incapable of further output.
+ *
+ * Returns: 0 Success
+ * EINVAL [COMMON_START]
+ * tcp_output:EADDRNOTAVAIL
+ * tcp_output:ENOBUFS
+ * tcp_output:EMSGSIZE
+ * tcp_output:EHOSTUNREACH
+ * tcp_output:ENETUNREACH
+ * tcp_output:ENETDOWN
+ * tcp_output:ENOMEM
+ * tcp_output:EACCES
+ * tcp_output:EMSGSIZE
+ * tcp_output:ENOBUFS
+ * tcp_output:??? [ignorable: mostly IPSEC/firewall/DLIL]
*/
static int
tcp_usr_shutdown(struct socket *so)
* After a receive, possibly send window update to peer.
*/
static int
-tcp_usr_rcvd(struct socket *so, int flags)
+tcp_usr_rcvd(struct socket *so, __unused int flags)
{
int error = 0;
struct inpcb *inp = sotoinpcb(so);
* pru_*() routines, the mbuf chains are our responsibility. We
* must either enqueue them or free them. The other pru_* routines
* generally are caller-frees.
+ *
+ * Returns: 0 Success
+ * ECONNRESET
+ * EINVAL
+ * ENOBUFS
+ * tcp_connect:EADDRINUSE Address in use
+ * tcp_connect:EADDRNOTAVAIL Address not available.
+ * tcp_connect:EINVAL Invalid argument
+ * tcp_connect:EAFNOSUPPORT Address family not supported [notdef]
+ * tcp_connect:EACCES Permission denied
+ * tcp_connect:EAGAIN Resource unavailable, try again
+ * tcp_connect:EPERM Operation not permitted
+ * tcp_output:EADDRNOTAVAIL
+ * tcp_output:ENOBUFS
+ * tcp_output:EMSGSIZE
+ * tcp_output:EHOSTUNREACH
+ * tcp_output:ENETUNREACH
+ * tcp_output:ENETDOWN
+ * tcp_output:ENOMEM
+ * tcp_output:EACCES
+ * tcp_output:EMSGSIZE
+ * tcp_output:ENOBUFS
+ * tcp_output:??? [ignorable: mostly IPSEC/firewall/DLIL]
+ * tcp6_connect:??? [IPV6 only]
*/
static int
tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
m_freem(control); /* empty control, just free it */
}
if(!(flags & PRUS_OOB)) {
- sbappend(&so->so_snd, m);
+ sbappendstream(&so->so_snd, m);
if (nam && tp->t_state < TCPS_SYN_SENT) {
/*
* Do implied connect if not yet connected,
if (error)
goto out;
tp->snd_wnd = TTCP_CLIENT_SND_WND;
- tcp_mss(tp, -1);
+ tcp_mss(tp, -1, IFSCOPE_NONE);
}
if (flags & PRUS_EOF) {
tp->t_flags &= ~TF_MORETOCOME;
}
} else {
- if (sbspace(&so->so_snd) < -512) {
+ if (sbspace(&so->so_snd) == 0) {
+ /* if no space is left in sockbuf,
+ * do not try to squeeze in OOB traffic */
m_freem(m);
error = ENOBUFS;
goto out;
* of data past the urgent section.
* Otherwise, snd_up should be one lower.
*/
- sbappend(&so->so_snd, m);
+ sbappendstream(&so->so_snd, m);
if (nam && tp->t_state < TCPS_SYN_SENT) {
/*
* Do implied connect if not yet connected,
if (error)
goto out;
tp->snd_wnd = TTCP_CLIENT_SND_WND;
- tcp_mss(tp, -1);
+ tcp_mss(tp, -1, IFSCOPE_NONE);
}
tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
tp->t_force = 1;
/*
* Receive out-of-band data.
+ *
+ * Returns: 0 Success
+ * EINVAL [COMMON_START]
+ * EINVAL
+ * EWOULDBLOCK
*/
static int
tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
* sending CC options and if the connection duration was < MSL, then
* truncate the previous TIME-WAIT state and proceed.
* Initialize connection parameters and enter SYN-SENT state.
+ *
+ * Returns: 0 Success
+ * EADDRINUSE
+ * EINVAL
+ * in_pcbbind:EADDRNOTAVAIL Address not available.
+ * in_pcbbind:EINVAL Invalid argument
+ * in_pcbbind:EAFNOSUPPORT Address family not supported [notdef]
+ * in_pcbbind:EACCES Permission denied
+ * in_pcbbind:EADDRINUSE Address in use
+ * in_pcbbind:EAGAIN Resource unavailable, try again
+ * in_pcbbind:EPERM Operation not permitted
+ * in_pcbladdr:EINVAL Invalid argument
+ * in_pcbladdr:EAFNOSUPPORT Address family not supported
+ * in_pcbladdr:EADDRNOTAVAIL Address not available
*/
static int
tcp_connect(tp, nam, p)
if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
otp->t_state == TCPS_TIME_WAIT &&
- otp->t_starttime < tcp_msl &&
+ otp->t_starttime < (u_int32_t)tcp_msl &&
(otp->t_flags & TF_RCVD_CC))
otp = tcp_close(otp);
else {
- printf("tcp_connect: inp=%x err=EADDRINUSE\n", inp);
+ printf("tcp_connect: inp=%p err=EADDRINUSE\n", inp);
if (oinp != inp)
tcp_unlock(oinp->inp_socket, 1, 0);
return EADDRINUSE;
in_pcbrehash(inp);
lck_rw_done(inp->inp_pcbinfo->mtx);
- /* Compute window scaling to request. */
+ /* Compute window scaling to requesti according to sb_hiwat
+ * or leave us some room to increase potentially increase the window size depending
+ * on the default win scale
+ */
while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
- (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
+ (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
tp->request_r_scale++;
+ /*
+ * Inflate window size only if no setsockopt was performed on the recv sockbuf and
+ * if we're not over our number of active pcbs.
+ */
+
+ if (((so->so_rcv.sb_flags & SB_USRSIZE) == 0) && (inp->inp_pcbinfo->ipi_count < tcp_sockthreshold)) {
+ tp->request_r_scale = max(tcp_win_scale, tp->request_r_scale);
+ so->so_rcv.sb_hiwat = min(TCP_MAXWIN << tp->request_r_scale, (sb_max / (MSIZE+MCLBYTES)) * MCLBYTES);
+ }
+
soisconnecting(so);
tcpstat.tcps_connattempt++;
tp->t_state = TCPS_SYN_SENT;
- tp->t_timer[TCPT_KEEP] = tcp_keepinit;
+ tp->t_timer[TCPT_KEEP] = tp->t_keepinit ? tp->t_keepinit : tcp_keepinit;
tp->iss = tcp_new_isn(tp);
tcp_sendseqinit(tp);
if (oinp) {
if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
otp->t_state == TCPS_TIME_WAIT &&
- otp->t_starttime < tcp_msl &&
+ otp->t_starttime < (u_int32_t)tcp_msl &&
(otp->t_flags & TF_RCVD_CC))
otp = tcp_close(otp);
else
inp->in6p_laddr = addr6;
inp->in6p_faddr = sin6->sin6_addr;
inp->inp_fport = sin6->sin6_port;
- if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != NULL)
+ if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
inp->in6p_flowinfo = sin6->sin6_flowinfo;
in_pcbrehash(inp);
lck_rw_done(inp->inp_pcbinfo->mtx);
soisconnecting(so);
tcpstat.tcps_connattempt++;
tp->t_state = TCPS_SYN_SENT;
- tp->t_timer[TCPT_KEEP] = tcp_keepinit;
+ tp->t_timer[TCPT_KEEP] = tp->t_keepinit ? tp->t_keepinit : tcp_keepinit;
tp->iss = tcp_new_isn(tp);
tcp_sendseqinit(tp);
error = EINVAL;
break;
- case TCP_KEEPALIVE:
- error = sooptcopyin(sopt, &optval, sizeof optval,
- sizeof optval);
- if (error)
- break;
- if (optval < 0)
- error = EINVAL;
- else
- tp->t_keepidle = optval * PR_SLOWHZ;
+ case TCP_KEEPALIVE:
+ error = sooptcopyin(sopt, &optval, sizeof optval,
+ sizeof optval);
+ if (error)
+ break;
+ if (optval < 0)
+ error = EINVAL;
+ else {
+ tp->t_keepidle = optval * TCP_RETRANSHZ;
+ tp->t_timer[TCPT_KEEP] = TCP_KEEPIDLE(tp); /* reset the timer to new value */
+ }
break;
+
+ case TCP_CONNECTIONTIMEOUT:
+ error = sooptcopyin(sopt, &optval, sizeof optval,
+ sizeof optval);
+ if (error)
+ break;
+ if (optval < 0)
+ error = EINVAL;
+ else
+ tp->t_keepinit = optval * TCP_RETRANSHZ;
+ break;
default:
error = ENOPROTOOPT;
optval = tp->t_maxseg;
break;
case TCP_KEEPALIVE:
- optval = tp->t_keepidle / PR_SLOWHZ;
+ optval = tp->t_keepidle / TCP_RETRANSHZ;
break;
case TCP_NOOPT:
optval = tp->t_flags & TF_NOOPT;
case TCP_NOPUSH:
optval = tp->t_flags & TF_NOPUSH;
break;
+ case TCP_CONNECTIONTIMEOUT:
+ optval = tp->t_keepinit / TCP_RETRANSHZ;
+ break;
default:
error = ENOPROTOOPT;
break;
* sizes, respectively. These are obsolescent (this information should
* be set by the route).
*/
-u_long tcp_sendspace = 1024*16;
-SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
- &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
-u_long tcp_recvspace = 1024*16;
-SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
- &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
-
-__private_extern__ int tcp_sockthreshold = 256;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, sockthreshold, CTLFLAG_RW,
- &tcp_sockthreshold , 0, "TCP Socket size increased if less than threshold");
+u_int32_t tcp_sendspace = 1448*256;
+u_int32_t tcp_recvspace = 1448*384;
+
+/* During attach, the size of socket buffer allocated is limited to
+ * sb_max in sbreserve. Disallow setting the tcp send and recv space
+ * to be more than sb_max because that will cause tcp_attach to fail
+ * (see radar 5713060)
+ */
+static int
+sysctl_tcp_sospace(struct sysctl_oid *oidp, __unused void *arg1,
+ __unused int arg2, struct sysctl_req *req) {
+ u_int32_t new_value = 0, *space_p = NULL;
+ int changed = 0, error = 0;
+ u_quad_t sb_effective_max = (sb_max / (MSIZE+MCLBYTES)) * MCLBYTES;
+
+ switch (oidp->oid_number) {
+ case TCPCTL_SENDSPACE:
+ space_p = &tcp_sendspace;
+ break;
+ case TCPCTL_RECVSPACE:
+ space_p = &tcp_recvspace;
+ break;
+ default:
+ return EINVAL;
+ }
+ error = sysctl_io_number(req, *space_p, sizeof(u_int32_t),
+ &new_value, &changed);
+ if (changed) {
+ if (new_value > 0 && new_value <= sb_effective_max) {
+ *space_p = new_value;
+ } else {
+ error = ERANGE;
+ }
+ }
+ return error;
+}
+
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLTYPE_INT | CTLFLAG_RW,
+ &tcp_sendspace , 0, &sysctl_tcp_sospace, "IU", "Maximum outgoing TCP datagram size");
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLTYPE_INT | CTLFLAG_RW,
+ &tcp_recvspace , 0, &sysctl_tcp_sospace, "IU", "Maximum incoming TCP datagram size");
+
-#define TCP_INCREASED_SPACE 65535 /* Automatically increase tcp send/rcv space to this value */
/*
* Attach TCP protocol to socket, allocating
* internet protocol control block, tcp control block,
* bufer space, and entering LISTEN state if to accept connections.
+ *
+ * Returns: 0 Success
+ * in_pcballoc:ENOBUFS
+ * in_pcballoc:ENOMEM
+ * in_pcballoc:??? [IPSEC specific]
+ * soreserve:ENOBUFS
*/
static int
tcp_attach(so, p)
register struct tcpcb *tp;
struct inpcb *inp;
int error;
+ u_long sb_effective_max;
#if INET6
- int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != NULL;
+ int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
#endif
error = in_pcballoc(so, &tcbinfo, p);
if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
/*
- * The goal is to let clients have large send/rcv default windows (TCP_INCREASED_SPACE)
- * while not hogging mbuf space for servers. This is done by watching a threshold
- * of tcpcbs in use and bumping the default send and rcvspace only if under that threshold.
- * The theory being that busy servers have a lot more active tcpcbs and don't want the potential
- * memory penalty of having much larger sockbuffs. The sysctl allows to fine tune that threshold value. */
-
- if (inp->inp_pcbinfo->ipi_count < tcp_sockthreshold)
- error = soreserve(so, MAX(TCP_INCREASED_SPACE, tcp_sendspace), MAX(TCP_INCREASED_SPACE,tcp_recvspace));
+ * The goal is to let clients machines use large send/rcv default windows to compensate for link
+ * latency and make sure the receiver is not constraining the sender window.
+ * But we doon't want to have a few connections use all our mbuf space for servers.
+ * This is done by watching a threshold of tcpcbs in use and bumping the default send and rcvspace
+ * only if that threshold isn't reached.
+ * We're also advertising a much bigger window size (tuneable by sysctl) in correlation with * the max socket buffer size if
+ * we consider that we have enough ressources for it. This window will be adjusted depending on the
+ * global socket layer buffer use with the use of tcp_sbpace
+ */
+
+ if (inp->inp_pcbinfo->ipi_count < tcp_sockthreshold) {
+ sb_effective_max = (sb_max / (MSIZE+MCLBYTES)) * MCLBYTES;
+ error = soreserve(so, max(min((TCP_MAXWIN << tcp_win_scale)/4, sb_effective_max), tcp_sendspace),
+ max(min((TCP_MAXWIN << tcp_win_scale)/2, sb_effective_max), tcp_recvspace));
+ }
else
error = soreserve(so, tcp_sendspace, tcp_recvspace);
if (error)
return (tp);
}
+void
+tcp_in_cksum_stats(u_int32_t len)
+{
+ tcps_in_sw_cksum++;
+ tcps_in_sw_cksum_bytes += len;
+}
+
+void
+tcp_out_cksum_stats(u_int32_t len)
+{
+ tcps_out_sw_cksum++;
+ tcps_out_sw_cksum_bytes += len;
+}
projects / apple / xnu.git / blobdiff