diff --git a/share/man/man4/tcp.4 b/share/man/man4/tcp.4 index 0ee57dcb0594..16c9e0ce84df 100644 --- a/share/man/man4/tcp.4 +++ b/share/man/man4/tcp.4 @@ -1,1099 +1,1093 @@ .\" Copyright (c) 1983, 1991, 1993 .\" The Regents of the University of California. .\" Copyright (c) 2010-2011 The FreeBSD Foundation .\" All rights reserved. .\" .\" Portions of this documentation were written at the Centre for Advanced .\" Internet Architectures, Swinburne University of Technology, Melbourne, .\" Australia by David Hayes under sponsorship from the FreeBSD Foundation. .\" .\" 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. .\" 3. Neither the name of the University nor the names of its contributors .\" may be used to endorse or promote products derived from this software .\" without specific prior written permission. .\" .\" THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. .\" .\" From: @(#)tcp.4 8.1 (Berkeley) 6/5/93 .\" -.Dd June 29, 2023 +.Dd November 17, 2023 .Dt TCP 4 .Os .Sh NAME .Nm tcp .Nd Internet Transmission Control Protocol .Sh SYNOPSIS .In sys/types.h .In sys/socket.h .In netinet/in.h .In netinet/tcp.h .Ft int .Fn socket AF_INET SOCK_STREAM 0 .Sh DESCRIPTION The .Tn TCP protocol provides reliable, flow-controlled, two-way transmission of data. It is a byte-stream protocol used to support the .Dv SOCK_STREAM abstraction. .Tn TCP uses the standard Internet address format and, in addition, provides a per-host collection of .Dq "port addresses" . Thus, each address is composed of an Internet address specifying the host and network, with a specific .Tn TCP port on the host identifying the peer entity. .Pp Sockets utilizing the .Tn TCP protocol are either .Dq active or .Dq passive . Active sockets initiate connections to passive sockets. By default, .Tn TCP sockets are created active; to create a passive socket, the .Xr listen 2 system call must be used after binding the socket with the .Xr bind 2 system call. Only passive sockets may use the .Xr accept 2 call to accept incoming connections. Only active sockets may use the .Xr connect 2 call to initiate connections. .Pp Passive sockets may .Dq underspecify their location to match incoming connection requests from multiple networks. This technique, termed .Dq "wildcard addressing" , allows a single server to provide service to clients on multiple networks. To create a socket which listens on all networks, the Internet address .Dv INADDR_ANY must be bound. The .Tn TCP port may still be specified at this time; if the port is not specified, the system will assign one. Once a connection has been established, the socket's address is fixed by the peer entity's location. The address assigned to the socket is the address associated with the network interface through which packets are being transmitted and received. Normally, this address corresponds to the peer entity's network. .Pp .Tn TCP supports a number of socket options which can be set with .Xr setsockopt 2 and tested with .Xr getsockopt 2 : .Bl -tag -width ".Dv TCP_FUNCTION_BLK" .It Dv TCP_INFO Information about a socket's underlying TCP session may be retrieved by passing the read-only option .Dv TCP_INFO to .Xr getsockopt 2 . It accepts a single argument: a pointer to an instance of .Vt "struct tcp_info" . .Pp This API is subject to change; consult the source to determine which fields are currently filled out by this option. .Fx specific additions include send window size, receive window size, and bandwidth-controlled window space. .It Dv TCP_CCALGOOPT Set or query congestion control algorithm specific parameters. See .Xr mod_cc 4 for details. .It Dv TCP_CONGESTION Select or query the congestion control algorithm that TCP will use for the connection. See .Xr mod_cc 4 for details. .It Dv TCP_FASTOPEN Enable or disable TCP Fast Open (TFO). To use this option, the kernel must be built with the .Dv TCP_RFC7413 option. .Pp This option can be set on the socket either before or after the .Xr listen 2 is invoked. Clearing this option on a listen socket after it has been set has no effect on existing TFO connections or TFO connections in progress; it only prevents new TFO connections from being established. .Pp For passively-created sockets, the .Dv TCP_FASTOPEN socket option can be queried to determine whether the connection was established using TFO. Note that connections that are established via a TFO .Tn SYN , but that fall back to using a non-TFO .Tn SYN|ACK will have the .Dv TCP_FASTOPEN socket option set. .Pp In addition to the facilities defined in RFC7413, this implementation supports a pre-shared key (PSK) mode of operation in which the TFO server requires the client to be in posession of a shared secret in order for the client to be able to successfully open TFO connections with the server. This is useful, for example, in environments where TFO servers are exposed to both internal and external clients and only wish to allow TFO connections from internal clients. .Pp In the PSK mode of operation, the server generates and sends TFO cookies to requesting clients as usual. However, when validating cookies received in TFO SYNs from clients, the server requires the client-supplied cookie to equal .Bd -literal -offset left SipHash24(key=\fI16-byte-psk\fP, msg=\fIcookie-sent-to-client\fP) .Ed .Pp Multiple concurrent valid pre-shared keys are supported so that time-based rolling PSK invalidation policies can be implemented in the system. The default number of concurrent pre-shared keys is 2. .Pp This can be adjusted with the .Dv TCP_RFC7413_MAX_PSKS kernel option. .It Dv TCP_FUNCTION_BLK Select or query the set of functions that TCP will use for this connection. This allows a user to select an alternate TCP stack. The alternate TCP stack must already be loaded in the kernel. To list the available TCP stacks, see .Va functions_available in the .Sx MIB (sysctl) Variables section further down. To list the default TCP stack, see .Va functions_default in the .Sx MIB (sysctl) Variables section. .It Dv TCP_KEEPINIT This .Xr setsockopt 2 option accepts a per-socket timeout argument of .Vt "u_int" in seconds, for new, non-established .Tn TCP connections. For the global default in milliseconds see .Va keepinit in the .Sx MIB (sysctl) Variables section further down. .It Dv TCP_KEEPIDLE This .Xr setsockopt 2 option accepts an argument of .Vt "u_int" for the amount of time, in seconds, that the connection must be idle before keepalive probes (if enabled) are sent for the connection of this socket. If set on a listening socket, the value is inherited by the newly created socket upon .Xr accept 2 . For the global default in milliseconds see .Va keepidle in the .Sx MIB (sysctl) Variables section further down. .It Dv TCP_KEEPINTVL This .Xr setsockopt 2 option accepts an argument of .Vt "u_int" to set the per-socket interval, in seconds, between keepalive probes sent to a peer. If set on a listening socket, the value is inherited by the newly created socket upon .Xr accept 2 . For the global default in milliseconds see .Va keepintvl in the .Sx MIB (sysctl) Variables section further down. .It Dv TCP_KEEPCNT This .Xr setsockopt 2 option accepts an argument of .Vt "u_int" and allows a per-socket tuning of the number of probes sent, with no response, before the connection will be dropped. If set on a listening socket, the value is inherited by the newly created socket upon .Xr accept 2 . For the global default see the .Va keepcnt in the .Sx MIB (sysctl) Variables section further down. .It Dv TCP_NODELAY Under most circumstances, .Tn TCP sends data when it is presented; when outstanding data has not yet been acknowledged, it gathers small amounts of output to be sent in a single packet once an acknowledgement is received. For a small number of clients, such as window systems that send a stream of mouse events which receive no replies, this packetization may cause significant delays. The boolean option .Dv TCP_NODELAY defeats this algorithm. .It Dv TCP_MAXSEG By default, a sender- and .No receiver- Ns Tn TCP will negotiate among themselves to determine the maximum segment size to be used for each connection. The .Dv TCP_MAXSEG option allows the user to determine the result of this negotiation, and to reduce it if desired. .It Dv TCP_MAXUNACKTIME This .Xr setsockopt 2 option accepts an argument of .Vt "u_int" to set the per-socket interval, in seconds, in which the connection must make progress. Progress is defined by at least 1 byte being acknowledged within the set time period. If a connection fails to make progress, then the .Tn TCP stack will terminate the connection with a reset. Note that the default value for this is zero which indicates no progress checks should be made. .It Dv TCP_NOOPT .Tn TCP usually sends a number of options in each packet, corresponding to various .Tn TCP extensions which are provided in this implementation. The boolean option .Dv TCP_NOOPT is provided to disable .Tn TCP option use on a per-connection basis. .It Dv TCP_NOPUSH By convention, the .No sender- Ns Tn TCP will set the .Dq push bit, and begin transmission immediately (if permitted) at the end of every user call to .Xr write 2 or .Xr writev 2 . When this option is set to a non-zero value, .Tn TCP will delay sending any data at all until either the socket is closed, or the internal send buffer is filled. .It Dv TCP_MD5SIG This option enables the use of MD5 digests (also known as TCP-MD5) on writes to the specified socket. Outgoing traffic is digested; digests on incoming traffic are verified. When this option is enabled on a socket, all inbound and outgoing TCP segments must be signed with MD5 digests. .Pp One common use for this in a .Fx router deployment is to enable based routers to interwork with Cisco equipment at peering points. Support for this feature conforms to RFC 2385. .Pp In order for this option to function correctly, it is necessary for the administrator to add a tcp-md5 key entry to the system's security associations database (SADB) using the .Xr setkey 8 utility. This entry can only be specified on a per-host basis at this time. .Pp If an SADB entry cannot be found for the destination, the system does not send any outgoing segments and drops any inbound segments. However, during connection negotiation, a non-signed segment will be accepted if an SADB entry does not exist between hosts. When a non-signed segment is accepted, the established connection is not protected with MD5 digests. .It Dv TCP_STATS Manage collection of connection level statistics using the .Xr stats 3 framework. .Pp Each dropped segment is taken into account in the TCP protocol statistics. .It Dv TCP_TXTLS_ENABLE Enable in-kernel Transport Layer Security (TLS) for data written to this socket. See .Xr ktls 4 for more details. .It Dv TCP_TXTLS_MODE The integer argument can be used to get or set the current TLS transmit mode of a socket. See .Xr ktls 4 for more details. .It Dv TCP_RXTLS_ENABLE Enable in-kernel TLS for data read from this socket. See .Xr ktls 4 for more details. .It Dv TCP_REUSPORT_LB_NUMA Changes NUMA affinity filtering for an established TCP listen socket. This option takes a single integer argument which specifies the NUMA domain to filter on for this listen socket. The argument can also have the follwing special values: .Bl -tag -width "Dv TCP_REUSPORT_LB_NUMA" .It Dv TCP_REUSPORT_LB_NUMA_NODOM Remove NUMA filtering for this listen socket. .It Dv TCP_REUSPORT_LB_NUMA_CURDOM Filter traffic associated with the domain where the calling thread is currently executing. This is typically used after a process or thread inherits a listen socket from its parent, and sets its CPU affinity to a particular core. .El .It Dv TCP_REMOTE_UDP_ENCAPS_PORT Set and get the remote UDP encapsulation port. It can only be set on a closed TCP socket. .El .Pp The option level for the .Xr setsockopt 2 call is the protocol number for .Tn TCP , available from .Xr getprotobyname 3 , or .Dv IPPROTO_TCP . All options are declared in .In netinet/tcp.h . .Pp Options at the .Tn IP transport level may be used with .Tn TCP ; see .Xr ip 4 . Incoming connection requests that are source-routed are noted, and the reverse source route is used in responding. .Pp The default congestion control algorithm for .Tn TCP is .Xr cc_newreno 4 . Other congestion control algorithms can be made available using the .Xr mod_cc 4 framework. .Ss MIB (sysctl) Variables The .Tn TCP protocol implements a number of variables in the .Va net.inet.tcp branch of the .Xr sysctl 3 MIB, which can also be read or modified with .Xr sysctl 8 . .Bl -tag -width ".Va v6pmtud_blackhole_mss" .It Va always_keepalive Assume that .Dv SO_KEEPALIVE is set on all .Tn TCP connections, the kernel will periodically send a packet to the remote host to verify the connection is still up. .It Va blackhole If enabled, disable sending of RST when a connection is attempted to a port where there is no socket accepting connections. See .Xr blackhole 4 . .It Va blackhole_local See .Xr blackhole 4 . .It Va cc A number of variables for congestion control are under the .Va net.inet.tcp.cc node. See .Xr mod_cc 4 . .It Va cc.newreno Variables for NewReno congestion control are under the .Va net.inet.tcp.cc.newreno node. See .Xr cc_newreno 4 . .It Va delacktime Maximum amount of time, in milliseconds, before a delayed ACK is sent. .It Va delayed_ack Delay ACK to try and piggyback it onto a data packet or another ACK. .It Va do_lrd Enable Lost Retransmission Detection for SACK-enabled sessions, disabled by default. Under severe congestion, a retransmission can be lost which then leads to a mandatory Retransmission Timeout (RTO), followed by slow-start. LRD will try to resend the repeatedly lost packet, preventing the time-consuming RTO and performance reducing slow-start. .It Va do_prr Perform SACK loss recovery using the Proportional Rate Reduction (PRR) algorithm described in RFC6937. This improves the effectiveness of retransmissions particular in environments with ACK thinning or burst loss events, as chances to run out of the ACK clock are reduced, preventing lengthy and performance reducing RTO based loss recovery (default is true). -.It Va do_prr_conservative -While doing Proportional Rate Reduction, remain strictly in a packet conserving -mode, sending only one new packet for each ACK received. -Helpful when a misconfigured token bucket traffic policer causes persistent -high losses leading to RTO, but reduces PRR effectiveness in more common settings -(default is false). .It Va do_tcpdrain Flush packets in the .Tn TCP reassembly queue if the system is low on mbufs. .It Va drop_synfin Drop TCP packets with both SYN and FIN set. .It Va ecn.enable Enable support for TCP Explicit Congestion Notification (ECN). ECN allows a TCP sender to reduce the transmission rate in order to avoid packet drops. .Bl -tag -compact .It 0 Disable ECN. .It 1 Allow incoming connections to request ECN. Outgoing connections will request ECN. .It 2 Allow incoming connections to request ECN. Outgoing connections will not request ECN. (default) .It 3 Negotiate on incoming connection for Accurate ECN, ECN, or no ECN. Outgoing connections will request Accurate ECN and fall back to ECN depending on the capabilities of the server. .It 4 Negotiate on incoming connection for Accurate ECN, ECN, or no ECN. Outgoing connections will not request ECN. .El .It Va ecn.maxretries Number of retries (SYN or SYN/ACK retransmits) before disabling ECN on a specific connection. This is needed to help with connection establishment when a broken firewall is in the network path. .It Va fast_finwait2_recycle Recycle .Tn TCP .Dv FIN_WAIT_2 connections faster when the socket is marked as .Dv SBS_CANTRCVMORE (no user process has the socket open, data received on the socket cannot be read). The timeout used here is .Va finwait2_timeout . .It Va fastopen.acceptany When non-zero, all client-supplied TFO cookies will be considered to be valid. The default is 0. .It Va fastopen.autokey When this and .Va net.inet.tcp.fastopen.server_enable are non-zero, a new key will be automatically generated after this specified seconds. The default is 120. .It Va fastopen.ccache_bucket_limit The maximum number of entries in a client cookie cache bucket. The default value can be tuned with the .Dv TCP_FASTOPEN_CCACHE_BUCKET_LIMIT_DEFAULT kernel option or by setting .Va net.inet.tcp.fastopen_ccache_bucket_limit in the .Xr loader 8 . .It Va fastopen.ccache_buckets The number of client cookie cache buckets. Read-only. The value can be tuned with the .Dv TCP_FASTOPEN_CCACHE_BUCKETS_DEFAULT kernel option or by setting .Va fastopen.ccache_buckets in the .Xr loader 8 . .It Va fastopen.ccache_list Print the client cookie cache. Read-only. .It Va fastopen.client_enable When zero, no new active (i.e., client) TFO connections can be created. On the transition from enabled to disabled, the client cookie cache is cleared and disabled. The transition from enabled to disabled does not affect any active TFO connections in progress; it only prevents new ones from being established. The default is 0. .It Va fastopen.keylen The key length in bytes. Read-only. .It Va fastopen.maxkeys The maximum number of keys supported. Read-only, .It Va fastopen.maxpsks The maximum number of pre-shared keys supported. Read-only. .It Va fastopen.numkeys The current number of keys installed. Read-only. .It Va fastopen.numpsks The current number of pre-shared keys installed. Read-only. .It Va fastopen.path_disable_time When a failure occurs while trying to create a new active (i.e., client) TFO connection, new active connections on the same path, as determined by the tuple .Brq client_ip, server_ip, server_port , will be forced to be non-TFO for this many seconds. Note that the path disable mechanism relies on state stored in client cookie cache entries, so it is possible for the disable time for a given path to be reduced if the corresponding client cookie cache entry is reused due to resource pressure before the disable period has elapsed. The default is .Dv TCP_FASTOPEN_PATH_DISABLE_TIME_DEFAULT . .It Va fastopen.psk_enable When non-zero, pre-shared key (PSK) mode is enabled for all TFO servers. On the transition from enabled to disabled, all installed pre-shared keys are removed. The default is 0. .It Va fastopen.server_enable When zero, no new passive (i.e., server) TFO connections can be created. On the transition from enabled to disabled, all installed keys and pre-shared keys are removed. On the transition from disabled to enabled, if .Va fastopen.autokey is non-zero and there are no keys installed, a new key will be generated immediately. The transition from enabled to disabled does not affect any passive TFO connections in progress; it only prevents new ones from being established. The default is 0. .It Va fastopen.setkey Install a new key by writing .Va net.inet.tcp.fastopen.keylen bytes to this sysctl. .It Va fastopen.setpsk Install a new pre-shared key by writing .Va net.inet.tcp.fastopen.keylen bytes to this sysctl. .It Va finwait2_timeout Timeout to use for fast recycling of .Tn TCP .Dv FIN_WAIT_2 connections .Pq Va fast_finwait2_recycle . Defaults to 60 seconds. .It Va functions_available List of available TCP function blocks (TCP stacks). .It Va functions_default The default TCP function block (TCP stack). .It Va functions_inherit_listen_socket_stack Determines whether to inherit listen socket's TCP stack or use the current system default TCP stack, as defined by .Va functions_default . Default is true. .It Va hostcache The TCP host cache is used to cache connection details and metrics to improve future performance of connections between the same hosts. At the completion of a TCP connection, a host will cache information for the connection for some defined period of time. There are a number of .Va hostcache variables under this node. See .Va hostcache.enable . .It Va hostcache.bucketlimit The maximum number of entries for the same hash. Defaults to 30. .It Va hostcache.cachelimit Overall entry limit for hostcache. Defaults to .Va hashsize * .Va bucketlimit . .It Va hostcache.count The current number of entries in the host cache. .It Va hostcache.enable Enable/disable the host cache: .Bl -tag -compact .It 0 Disable the host cache. .It 1 Enable the host cache. (default) .El .It Va hostcache.expire Time in seconds, how long a entry should be kept in the host cache since last accessed. Defaults to 3600 (1 hour). .It Va hostcache.hashsize Size of TCP hostcache hashtable. This number has to be a power of two, or will be rejected. Defaults to 512. .It Va hostcache.histo Provide a Histogram of the hostcache hash utilization. .It Va hostcache.list Provide a complete list of all current entries in the host cache. .It Va hostcache.prune Time in seconds between pruning expired host cache entries. Defaults to 300 (5 minutes). .It Va hostcache.purge Expire all entires on next pruning of host cache entries. Any non-zero setting will be reset to zero, once the purge is running. .Bl -tag -compact .It 0 Do not purge all entries when pruning the host cache (default). .It 1 Purge all entries when doing the next pruning. .It 2 Purge all entries and also reseed the hash salt. .El .It Va hostcache.purgenow Immediately purge all entries once set to any value. Setting this to 2 will also reseed the hash salt. .It Va icmp_may_rst Certain .Tn ICMP unreachable messages may abort connections in .Tn SYN-SENT state. .It Va initcwnd_segments Enable the ability to specify initial congestion window in number of segments. The default value is 10 as suggested by RFC 6928. Changing the value on the fly would not affect connections using congestion window from the hostcache. Caution: This regulates the burst of packets allowed to be sent in the first RTT. The value should be relative to the link capacity. Start with small values for lower-capacity links. Large bursts can cause buffer overruns and packet drops if routers have small buffers or the link is experiencing congestion. .It Va insecure_rst Use criteria defined in RFC793 instead of RFC5961 for accepting RST segments. Default is false. .It Va insecure_syn Use criteria defined in RFC793 instead of RFC5961 for accepting SYN segments. Default is false. .It Va isn_reseed_interval The interval (in seconds) specifying how often the secret data used in RFC 1948 initial sequence number calculations should be reseeded. By default, this variable is set to zero, indicating that no reseeding will occur. Reseeding should not be necessary, and will break .Dv TIME_WAIT recycling for a few minutes. .It Va keepcnt Number of keepalive probes sent, with no response, before a connection is dropped. The default is 8 packets. .It Va keepidle Amount of time, in milliseconds, that the connection must be idle before sending keepalive probes (if enabled). The default is 7200000 msec (7.2M msec, 2 hours). .It Va keepinit Timeout, in milliseconds, for new, non-established .Tn TCP connections. The default is 75000 msec (75K msec, 75 sec). .It Va keepintvl The interval, in milliseconds, between keepalive probes sent to remote machines, when no response is received on a .Va keepidle probe. The default is 75000 msec (75K msec, 75 sec). .It Va log_in_vain Log any connection attempts to ports where there is no socket accepting connections. The value of 1 limits the logging to .Tn SYN (connection establishment) packets only. A value of 2 results in any .Tn TCP packets to closed ports being logged. Any value not listed above disables the logging (default is 0, i.e., the logging is disabled). .It Va minmss Minimum TCP Maximum Segment Size; used to prevent a denial of service attack from an unreasonably low MSS. .It Va msl The Maximum Segment Lifetime, in milliseconds, for a packet. .It Va mssdflt The default value used for the TCP Maximum Segment Size .Pq Dq MSS for IPv4 when no advice to the contrary is received from MSS negotiation. .It Va newcwd Enable the New Congestion Window Validation mechanism as described in RFC 7661. This gently reduces the congestion window during periods, where TCP is application limited and the network bandwidth is not utilized completely. That prevents self-inflicted packet losses once the application starts to transmit data at a higher speed. .It Va nolocaltimewait Suppress creation of TCP .Dv TIME_WAIT states for connections in which both endpoints are local. .It Va path_mtu_discovery Enable Path MTU Discovery. .It Va pcbcount Number of active protocol control blocks (read-only). .It Va perconn_stats_enable Controls the default collection of statistics for all connections using the .Xr stats 3 framework. 0 disables, 1 enables, 2 enables random sampling across log id connection groups with all connections in a group receiving the same setting. .It Va perconn_stats_sample_rates A CSV list of template_spec=percent key-value pairs which controls the per template sampling rates when .Xr stats 3 sampling is enabled. .It Va persmax Maximum persistence interval, msec. .It Va persmin Minimum persistence interval, msec. .It Va pmtud_blackhole_detection Enable automatic path MTU blackhole detection. In case of retransmits of MSS sized segments, the OS will lower the MSS to check if it's an MTU problem. If the current MSS is greater than the configured value to try .Po Va net.inet.tcp.pmtud_blackhole_mss and .Va net.inet.tcp.v6pmtud_blackhole_mss .Pc , it will be set to this value, otherwise, the MSS will be set to the default values .Po Va net.inet.tcp.mssdflt and .Va net.inet.tcp.v6mssdflt .Pc . Settings: .Bl -tag -compact .It 0 Disable path MTU blackhole detection. .It 1 Enable path MTU blackhole detection for IPv4 and IPv6. .It 2 Enable path MTU blackhole detection only for IPv4. .It 3 Enable path MTU blackhole detection only for IPv6. .El .It Va pmtud_blackhole_mss MSS to try for IPv4 if PMTU blackhole detection is turned on. .It Va reass.cursegments The current total number of segments present in all reassembly queues. .It Va reass.maxqueuelen The maximum number of segments allowed in each reassembly queue. By default, the system chooses a limit based on each TCP connection's receive buffer size and maximum segment size (MSS). The actual limit applied to a session's reassembly queue will be the lower of the system-calculated automatic limit and the user-specified .Va reass.maxqueuelen limit. .It Va reass.maxsegments The maximum limit on the total number of segments across all reassembly queues. The limit can be adjusted as a tunable. .It Va recvbuf_auto Enable automatic receive buffer sizing as a connection progresses. .It Va recvbuf_max Maximum size of automatic receive buffer. .It Va recvspace Initial .Tn TCP receive window (buffer size). .It Va retries Maximum number of consecutive timer based retransmits sent after a data segment is lost (default and maximum is 12). .It Va rexmit_drop_options Drop TCP options from third and later retransmitted SYN segments of a connection. .It Va rexmit_initial , rexmit_min , rexmit_slop Adjust the retransmit timer calculation for .Tn TCP . The slop is typically added to the raw calculation to take into account occasional variances that the .Tn SRTT (smoothed round-trip time) is unable to accommodate, while the minimum specifies an absolute minimum. While a number of .Tn TCP RFCs suggest a 1 second minimum, these RFCs tend to focus on streaming behavior, and fail to deal with the fact that a 1 second minimum has severe detrimental effects over lossy interactive connections, such as a 802.11b wireless link, and over very fast but lossy connections for those cases not covered by the fast retransmit code. For this reason, we use 200ms of slop and a near-0 minimum, which gives us an effective minimum of 200ms (similar to .Tn Linux ) . The initial value is used before an RTT measurement has been performed. .It Va rfc1323 Implement the window scaling and timestamp options of RFC 1323/RFC 7323 (default is 1). Settings: .Bl -tag -compact .It 0 Disable window scaling and timestamp option. .It 1 Enable window scaling and timestamp option. .It 2 Enable only window scaling. .It 3 Enable only timestamp option. .El .It Va rfc3042 Enable the Limited Transmit algorithm as described in RFC 3042. It helps avoid timeouts on lossy links and also when the congestion window is small, as happens on short transfers. .It Va rfc3390 Enable support for RFC 3390, which allows for a variable-sized starting congestion window on new connections, depending on the maximum segment size. This helps throughput in general, but particularly affects short transfers and high-bandwidth large propagation-delay connections. .It Va rfc6675_pipe Deprecated and superseded by .Va sack.revised .It Va sack.enable Enable support for RFC 2018, TCP Selective Acknowledgment option, which allows the receiver to inform the sender about all successfully arrived segments, allowing the sender to retransmit the missing segments only. .It Va sack.globalholes Global number of TCP SACK holes currently allocated. .It Va sack.globalmaxholes Maximum number of SACK holes per system, across all connections. Defaults to 65536. .It Va sack.maxholes Maximum number of SACK holes per connection. Defaults to 128. .It Va sack.revised Enables three updated mechanisms from RFC6675 (default is true). Calculate the bytes in flight using the algorithm described in RFC 6675, and is also an improvement when Proportional Rate Reduction is enabled. Next, Rescue Retransmission helps timely loss recovery, when the trailing segments of a transmission are lost, while no additional data is ready to be sent. In case a partial ACK without a SACK block is received during SACK loss recovery, the trailing segment is immediately resent, rather than waiting for a Retransmission timeout. Finally, SACK loss recovery is also engaged, once two segments plus one byte are SACKed - even if no traditional duplicate ACKs were observed. .It Va sendbuf_auto Enable automatic send buffer sizing. .It Va sendbuf_auto_lowat Modify threshold for auto send buffer growth to account for .Dv SO_SNDLOWAT . .It Va sendbuf_inc Incrementor step size of automatic send buffer. .It Va sendbuf_max Maximum size of automatic send buffer. .It Va sendspace Initial .Tn TCP send window (buffer size). .It Va syncache Variables under the .Va net.inet.tcp.syncache node are documented in .Xr syncache 4 . .It Va syncookies Determines whether or not .Tn SYN cookies should be generated for outbound .Tn SYN-ACK packets. .Tn SYN cookies are a great help during .Tn SYN flood attacks, and are enabled by default. (See .Xr syncookies 4 . ) .It Va syncookies_only See .Xr syncookies 4 . .It Va tcbhashsize Size of the .Tn TCP control-block hash table (read-only). This is tuned using the kernel option .Dv TCBHASHSIZE or by setting .Va net.inet.tcp.tcbhashsize in the .Xr loader 8 . .It Va tolerate_missing_ts Tolerate the missing of timestamps (RFC 1323/RFC 7323) for .Tn TCP segments belonging to .Tn TCP connections for which support of .Tn TCP timestamps has been negotiated. As of June 2021, several TCP stacks are known to violate RFC 7323, including modern widely deployed ones. Therefore the default is 1, i.e., the missing of timestamps is tolerated. .It Va ts_offset_per_conn When initializing the TCP timestamps, use a per connection offset instead of a per host pair offset. Default is to use per connection offsets as recommended in RFC 7323. .It Va tso Enable TCP Segmentation Offload. .It Va udp_tunneling_overhead The overhead taken into account when using UDP encapsulation. Since MSS clamping by middleboxes will most likely not work, values larger than 8 (the size of the UDP header) are also supported. Supported values are between 8 and 1024. The default is 8. .It Va udp_tunneling_port The local UDP encapsulation port. A value of 0 indicates that UDP encapsulation is disabled. The default is 0. .It Va v6mssdflt The default value used for the TCP Maximum Segment Size .Pq Dq MSS for IPv6 when no advice to the contrary is received from MSS negotiation. .It Va v6pmtud_blackhole_mss MSS to try for IPv6 if PMTU blackhole detection is turned on. See .Va pmtud_blackhole_detection . .El .Sh ERRORS A socket operation may fail with one of the following errors returned: .Bl -tag -width Er .It Bq Er EISCONN when trying to establish a connection on a socket which already has one; .It Bo Er ENOBUFS Bc or Bo Er ENOMEM Bc when the system runs out of memory for an internal data structure; .It Bq Er ETIMEDOUT when a connection was dropped due to excessive retransmissions; .It Bq Er ECONNRESET when the remote peer forces the connection to be closed; .It Bq Er ECONNREFUSED when the remote peer actively refuses connection establishment (usually because no process is listening to the port); .It Bq Er EADDRINUSE when an attempt is made to create a socket with a port which has already been allocated; .It Bq Er EADDRNOTAVAIL when an attempt is made to create a socket with a network address for which no network interface exists; .It Bq Er EAFNOSUPPORT when an attempt is made to bind or connect a socket to a multicast address. .It Bq Er EINVAL when trying to change TCP function blocks at an invalid point in the session; .It Bq Er ENOENT when trying to use a TCP function block that is not available; .El .Sh SEE ALSO .Xr getsockopt 2 , .Xr socket 2 , .Xr stats 3 , .Xr sysctl 3 , .Xr blackhole 4 , .Xr inet 4 , .Xr intro 4 , .Xr ip 4 , .Xr ktls 4 , .Xr mod_cc 4 , .Xr siftr 4 , .Xr syncache 4 , .Xr tcp_bbr 4 , .Xr tcp_rack 4 , .Xr setkey 8 , .Xr sysctl 8 , .Xr tcp_functions 9 .Rs .%A "V. Jacobson" .%A "B. Braden" .%A "D. Borman" .%T "TCP Extensions for High Performance" .%O "RFC 1323" .Re .Rs .%A "D. Borman" .%A "B. Braden" .%A "V. Jacobson" .%A "R. Scheffenegger" .%T "TCP Extensions for High Performance" .%O "RFC 7323" .Re .Rs .%A "A. Heffernan" .%T "Protection of BGP Sessions via the TCP MD5 Signature Option" .%O "RFC 2385" .Re .Rs .%A "K. Ramakrishnan" .%A "S. Floyd" .%A "D. Black" .%T "The Addition of Explicit Congestion Notification (ECN) to IP" .%O "RFC 3168" .Re .Sh HISTORY The .Tn TCP protocol appeared in .Bx 4.2 . The RFC 1323 extensions for window scaling and timestamps were added in .Bx 4.4 . The .Dv TCP_INFO option was introduced in .Tn Linux 2.6 and is .Em subject to change . diff --git a/sys/netinet/tcp_input.c b/sys/netinet/tcp_input.c index ba11503955dc..3673bc1e0b98 100644 --- a/sys/netinet/tcp_input.c +++ b/sys/netinet/tcp_input.c @@ -1,4083 +1,4098 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 * The Regents of the University of California. All rights reserved. * Copyright (c) 2007-2008,2010 * Swinburne University of Technology, Melbourne, Australia. * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation * Copyright (c) 2010-2011 Juniper Networks, Inc. * All rights reserved. * * Portions of this software were developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by Lawrence Stewart, * James Healy and David Hayes, made possible in part by a grant from the Cisco * University Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by David Hayes under sponsorship from the FreeBSD Foundation. * * Portions of this software were developed by Robert N. M. Watson under * contract to Juniper Networks, Inc. * * 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. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. * * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 */ #include #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_rss.h" #include #include #include #ifdef TCP_HHOOK #include #endif #include #include #include /* for proc0 declaration */ #include #include #include #include #include #include #include #include #include #include #include /* before tcp_seq.h, for tcp_random18() */ #include #include #include #include #include #include #define TCPSTATES /* for logging */ #include #include #include #include #include #include #include /* required for icmp_var.h */ #include /* for ICMP_BANDLIM */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef TCPPCAP #include #endif #include #ifdef TCP_OFFLOAD #include #endif #include #include #include #include #include const int tcprexmtthresh = 3; VNET_DEFINE(int, tcp_log_in_vain) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_log_in_vain), 0, "Log all incoming TCP segments to closed ports"); VNET_DEFINE(int, blackhole) = 0; #define V_blackhole VNET(blackhole) SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(blackhole), 0, "Do not send RST on segments to closed ports"); VNET_DEFINE(bool, blackhole_local) = false; #define V_blackhole_local VNET(blackhole_local) SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(blackhole_local), false, "Enforce net.inet.tcp.blackhole for locally originated packets"); VNET_DEFINE(int, tcp_delack_enabled) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_delack_enabled), 0, "Delay ACK to try and piggyback it onto a data packet"); VNET_DEFINE(int, drop_synfin) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(drop_synfin), 0, "Drop TCP packets with SYN+FIN set"); -VNET_DEFINE(int, tcp_do_prr_conservative) = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, - &VNET_NAME(tcp_do_prr_conservative), 0, - "Do conservative Proportional Rate Reduction"); - VNET_DEFINE(int, tcp_do_prr) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_prr), 1, "Enable Proportional Rate Reduction per RFC 6937"); VNET_DEFINE(int, tcp_do_lrd) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_lrd), 1, "Perform Lost Retransmission Detection"); VNET_DEFINE(int, tcp_do_newcwv) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_newcwv), 0, "Enable New Congestion Window Validation per RFC7661"); VNET_DEFINE(int, tcp_do_rfc3042) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3042), 0, "Enable RFC 3042 (Limited Transmit)"); VNET_DEFINE(int, tcp_do_rfc3390) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3390), 0, "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); VNET_DEFINE(int, tcp_initcwnd_segments) = 10; SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, "Slow-start flight size (initial congestion window) in number of segments"); VNET_DEFINE(int, tcp_do_rfc3465) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3465), 0, "Enable RFC 3465 (Appropriate Byte Counting)"); VNET_DEFINE(int, tcp_abc_l_var) = 2; SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_abc_l_var), 2, "Cap the max cwnd increment during slow-start to this number of segments"); VNET_DEFINE(int, tcp_insecure_syn) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_insecure_syn), 0, "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); VNET_DEFINE(int, tcp_insecure_rst) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_insecure_rst), 0, "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); VNET_DEFINE(int, tcp_recvspace) = 1024*64; #define V_tcp_recvspace VNET(tcp_recvspace) SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_autorcvbuf), 0, "Enable automatic receive buffer sizing"); VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_autorcvbuf_max), 0, "Max size of automatic receive buffer"); VNET_DEFINE(struct inpcbinfo, tcbinfo); /* * TCP statistics are stored in an array of counter(9)s, which size matches * size of struct tcpstat. TCP running connection count is a regular array. */ VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, "TCP connection counts by TCP state"); /* * Kernel module interface for updating tcpstat. The first argument is an index * into tcpstat treated as an array. */ void kmod_tcpstat_add(int statnum, int val) { counter_u64_add(VNET(tcpstat)[statnum], val); } /* * Make sure that we only start a SACK loss recovery when * receiving a duplicate ACK with a SACK block, and also * complete SACK loss recovery in case the other end * reneges. */ static bool inline tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) { return ((tp->t_flags & TF_SACK_PERMIT) && ((to->to_flags & TOF_SACK) || (!TAILQ_EMPTY(&tp->snd_holes)))); } #ifdef TCP_HHOOK /* * Wrapper for the TCP established input helper hook. */ void hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) { struct tcp_hhook_data hhook_data; if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { hhook_data.tp = tp; hhook_data.th = th; hhook_data.to = to; hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, &tp->t_osd); } } #endif /* * CC wrapper hook functions */ void cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, uint16_t type) { #ifdef STATS int32_t gput; #endif INP_WLOCK_ASSERT(tptoinpcb(tp)); tp->t_ccv.nsegs = nsegs; tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th); if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) tp->t_ccv.flags |= CCF_CWND_LIMITED; else tp->t_ccv.flags &= ~CCF_CWND_LIMITED; if (type == CC_ACK) { #ifdef STATS stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, ((int32_t)tp->snd_cwnd) - tp->snd_wnd); if (!IN_RECOVERY(tp->t_flags)) stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs)); if ((tp->t_flags & TF_GPUTINPROG) && SEQ_GEQ(th->th_ack, tp->gput_ack)) { /* * Compute goodput in bits per millisecond. */ gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / max(1, tcp_ts_getticks() - tp->gput_ts); stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, gput); /* * XXXLAS: This is a temporary hack, and should be * chained off VOI_TCP_GPUT when stats(9) grows an API * to deal with chained VOIs. */ if (tp->t_stats_gput_prev > 0) stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_GPUT_ND, ((gput - tp->t_stats_gput_prev) * 100) / tp->t_stats_gput_prev); tp->t_flags &= ~TF_GPUTINPROG; tp->t_stats_gput_prev = gput; } #endif /* STATS */ if (tp->snd_cwnd > tp->snd_ssthresh) { tp->t_bytes_acked += tp->t_ccv.bytes_this_ack; if (tp->t_bytes_acked >= tp->snd_cwnd) { tp->t_bytes_acked -= tp->snd_cwnd; tp->t_ccv.flags |= CCF_ABC_SENTAWND; } } else { tp->t_ccv.flags &= ~CCF_ABC_SENTAWND; tp->t_bytes_acked = 0; } } if (CC_ALGO(tp)->ack_received != NULL) { /* XXXLAS: Find a way to live without this */ tp->t_ccv.curack = th->th_ack; CC_ALGO(tp)->ack_received(&tp->t_ccv, type); } #ifdef STATS stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); #endif } void cc_conn_init(struct tcpcb *tp) { struct hc_metrics_lite metrics; struct inpcb *inp = tptoinpcb(tp); u_int maxseg; int rtt; INP_WLOCK_ASSERT(inp); tcp_hc_get(&inp->inp_inc, &metrics); maxseg = tcp_maxseg(tp); if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { tp->t_srtt = rtt; TCPSTAT_INC(tcps_usedrtt); if (metrics.rmx_rttvar) { tp->t_rttvar = metrics.rmx_rttvar; TCPSTAT_INC(tcps_usedrttvar); } else { /* default variation is +- 1 rtt */ tp->t_rttvar = tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; } TCPT_RANGESET(tp->t_rxtcur, ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, tp->t_rttmin, TCPTV_REXMTMAX); } if (metrics.rmx_ssthresh) { /* * There's some sort of gateway or interface * buffer limit on the path. Use this to set * the slow start threshold, but set the * threshold to no less than 2*mss. */ tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); TCPSTAT_INC(tcps_usedssthresh); } /* * Set the initial slow-start flight size. * * If a SYN or SYN/ACK was lost and retransmitted, we have to * reduce the initial CWND to one segment as congestion is likely * requiring us to be cautious. */ if (tp->snd_cwnd == 1) tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ else tp->snd_cwnd = tcp_compute_initwnd(maxseg); if (CC_ALGO(tp)->conn_init != NULL) CC_ALGO(tp)->conn_init(&tp->t_ccv); } void inline cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) { INP_WLOCK_ASSERT(tptoinpcb(tp)); #ifdef STATS stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); #endif switch(type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(tp->t_flags)) { tp->snd_recover = tp->snd_max; if (tp->t_flags2 & TF2_ECN_PERMIT) tp->t_flags2 |= TF2_ECN_SND_CWR; } break; case CC_ECN: if (!IN_CONGRECOVERY(tp->t_flags) || /* * Allow ECN reaction on ACK to CWR, if * that data segment was also CE marked. */ SEQ_GEQ(th->th_ack, tp->snd_recover)) { EXIT_CONGRECOVERY(tp->t_flags); TCPSTAT_INC(tcps_ecn_rcwnd); tp->snd_recover = tp->snd_max + 1; if (tp->t_flags2 & TF2_ECN_PERMIT) tp->t_flags2 |= TF2_ECN_SND_CWR; } break; case CC_RTO: tp->t_dupacks = 0; tp->t_bytes_acked = 0; EXIT_RECOVERY(tp->t_flags); if (tp->t_flags2 & TF2_ECN_PERMIT) tp->t_flags2 |= TF2_ECN_SND_CWR; break; case CC_RTO_ERR: TCPSTAT_INC(tcps_sndrexmitbad); /* RTO was unnecessary, so reset everything. */ tp->snd_cwnd = tp->snd_cwnd_prev; tp->snd_ssthresh = tp->snd_ssthresh_prev; tp->snd_recover = tp->snd_recover_prev; if (tp->t_flags & TF_WASFRECOVERY) ENTER_FASTRECOVERY(tp->t_flags); if (tp->t_flags & TF_WASCRECOVERY) ENTER_CONGRECOVERY(tp->t_flags); tp->snd_nxt = tp->snd_max; tp->t_flags &= ~TF_PREVVALID; tp->t_badrxtwin = 0; break; } if (CC_ALGO(tp)->cong_signal != NULL) { if (th != NULL) tp->t_ccv.curack = th->th_ack; CC_ALGO(tp)->cong_signal(&tp->t_ccv, type); } } void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) { INP_WLOCK_ASSERT(tptoinpcb(tp)); /* XXXLAS: KASSERT that we're in recovery? */ if (CC_ALGO(tp)->post_recovery != NULL) { tp->t_ccv.curack = th->th_ack; CC_ALGO(tp)->post_recovery(&tp->t_ccv); } /* XXXLAS: EXIT_RECOVERY ? */ tp->t_bytes_acked = 0; tp->sackhint.delivered_data = 0; tp->sackhint.prr_out = 0; } /* * Indicate whether this ack should be delayed. We can delay the ack if * following conditions are met: * - There is no delayed ack timer in progress. * - Our last ack wasn't a 0-sized window. We never want to delay * the ack that opens up a 0-sized window. * - LRO wasn't used for this segment. We make sure by checking that the * segment size is not larger than the MSS. */ #define DELAY_ACK(tp, tlen) \ ((!tcp_timer_active(tp, TT_DELACK) && \ (tp->t_flags & TF_RXWIN0SENT) == 0) && \ (tlen <= tp->t_maxseg) && \ (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) void inline cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) { INP_WLOCK_ASSERT(tptoinpcb(tp)); if (CC_ALGO(tp)->ecnpkt_handler != NULL) { switch (iptos & IPTOS_ECN_MASK) { case IPTOS_ECN_CE: tp->t_ccv.flags |= CCF_IPHDR_CE; break; case IPTOS_ECN_ECT0: /* FALLTHROUGH */ case IPTOS_ECN_ECT1: /* FALLTHROUGH */ case IPTOS_ECN_NOTECT: tp->t_ccv.flags &= ~CCF_IPHDR_CE; break; } if (flags & TH_CWR) tp->t_ccv.flags |= CCF_TCPHDR_CWR; else tp->t_ccv.flags &= ~CCF_TCPHDR_CWR; CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv); if (tp->t_ccv.flags & CCF_ACKNOW) { tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); tp->t_flags |= TF_ACKNOW; } } } void inline cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) { cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); } /* * TCP input handling is split into multiple parts: * tcp6_input is a thin wrapper around tcp_input for the extended * ip6_protox[] call format in ip6_input * tcp_input handles primary segment validation, inpcb lookup and * SYN processing on listen sockets * tcp_do_segment processes the ACK and text of the segment for * establishing, established and closing connections */ #ifdef INET6 int tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) { struct mbuf *m; struct in6_ifaddr *ia6; struct ip6_hdr *ip6; m = *mp; if (m->m_len < *offp + sizeof(struct tcphdr)) { m = m_pullup(m, *offp + sizeof(struct tcphdr)); if (m == NULL) { *mp = m; TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } } /* * draft-itojun-ipv6-tcp-to-anycast * better place to put this in? */ ip6 = mtod(m, struct ip6_hdr *); ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); *mp = NULL; return (IPPROTO_DONE); } *mp = m; return (tcp_input_with_port(mp, offp, proto, port)); } int tcp6_input(struct mbuf **mp, int *offp, int proto) { return(tcp6_input_with_port(mp, offp, proto, 0)); } #endif /* INET6 */ int tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) { struct mbuf *m = *mp; struct tcphdr *th = NULL; struct ip *ip = NULL; struct inpcb *inp = NULL; struct tcpcb *tp = NULL; struct socket *so = NULL; u_char *optp = NULL; int off0; int optlen = 0; #ifdef INET int len; uint8_t ipttl; #endif int tlen = 0, off; int drop_hdrlen; int thflags; int rstreason = 0; /* For badport_bandlim accounting purposes */ int lookupflag; uint8_t iptos; struct m_tag *fwd_tag = NULL; #ifdef INET6 struct ip6_hdr *ip6 = NULL; int isipv6; #else const void *ip6 = NULL; #endif /* INET6 */ struct tcpopt to; /* options in this segment */ char *s = NULL; /* address and port logging */ NET_EPOCH_ASSERT(); #ifdef INET6 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; #endif off0 = *offp; m = *mp; *mp = NULL; to.to_flags = 0; TCPSTAT_INC(tcps_rcvtotal); #ifdef INET6 if (isipv6) { ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)((caddr_t)ip6 + off0); tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; if (port) goto skip6_csum; if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) th->th_sum = m->m_pkthdr.csum_data; else th->th_sum = in6_cksum_pseudo(ip6, tlen, IPPROTO_TCP, m->m_pkthdr.csum_data); th->th_sum ^= 0xffff; } else th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); if (th->th_sum) { TCPSTAT_INC(tcps_rcvbadsum); goto drop; } skip6_csum: /* * Be proactive about unspecified IPv6 address in source. * As we use all-zero to indicate unbounded/unconnected pcb, * unspecified IPv6 address can be used to confuse us. * * Note that packets with unspecified IPv6 destination is * already dropped in ip6_input. */ KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst), ("%s: unspecified destination v6 address", __func__)); if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { IP6STAT_INC(ip6s_badscope); /* XXX */ goto drop; } iptos = IPV6_TRAFFIC_CLASS(ip6); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { /* * Get IP and TCP header together in first mbuf. * Note: IP leaves IP header in first mbuf. */ if (off0 > sizeof (struct ip)) { ip_stripoptions(m); off0 = sizeof(struct ip); } if (m->m_len < sizeof (struct tcpiphdr)) { if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } } ip = mtod(m, struct ip *); th = (struct tcphdr *)((caddr_t)ip + off0); tlen = ntohs(ip->ip_len) - off0; iptos = ip->ip_tos; if (port) goto skip_csum; if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) th->th_sum = m->m_pkthdr.csum_data; else th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + tlen + IPPROTO_TCP)); th->th_sum ^= 0xffff; } else { struct ipovly *ipov = (struct ipovly *)ip; /* * Checksum extended TCP header and data. */ len = off0 + tlen; ipttl = ip->ip_ttl; bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); ipov->ih_len = htons(tlen); th->th_sum = in_cksum(m, len); /* Reset length for SDT probes. */ ip->ip_len = htons(len); /* Reset TOS bits */ ip->ip_tos = iptos; /* Re-initialization for later version check */ ip->ip_ttl = ipttl; ip->ip_v = IPVERSION; ip->ip_hl = off0 >> 2; } skip_csum: if (th->th_sum && (port == 0)) { TCPSTAT_INC(tcps_rcvbadsum); goto drop; } KASSERT(ip->ip_dst.s_addr != INADDR_ANY, ("%s: unspecified destination v4 address", __func__)); if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) { IPSTAT_INC(ips_badaddr); goto drop; } } #endif /* INET */ /* * Check that TCP offset makes sense, * pull out TCP options and adjust length. XXX */ off = th->th_off << 2; if (off < sizeof (struct tcphdr) || off > tlen) { TCPSTAT_INC(tcps_rcvbadoff); goto drop; } tlen -= off; /* tlen is used instead of ti->ti_len */ if (off > sizeof (struct tcphdr)) { #ifdef INET6 if (isipv6) { if (m->m_len < off0 + off) { m = m_pullup(m, off0 + off); if (m == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } } ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)((caddr_t)ip6 + off0); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { if (m->m_len < sizeof(struct ip) + off) { if ((m = m_pullup(m, sizeof (struct ip) + off)) == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); th = (struct tcphdr *)((caddr_t)ip + off0); } } #endif optlen = off - sizeof (struct tcphdr); optp = (u_char *)(th + 1); } thflags = tcp_get_flags(th); /* * Convert TCP protocol specific fields to host format. */ tcp_fields_to_host(th); /* * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. */ drop_hdrlen = off0 + off; /* * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ if ( #ifdef INET6 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) #ifdef INET || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) #endif #endif #if defined(INET) && !defined(INET6) (m->m_flags & M_IP_NEXTHOP) #endif ) fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); /* * For initial SYN packets we don't need write lock on matching * PCB, be it a listening one or a synchronized one. The packet * shall not modify its state. */ lookupflag = INPLOOKUP_WILDCARD | ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB); findpcb: #ifdef INET6 if (isipv6 && fwd_tag != NULL) { struct sockaddr_in6 *next_hop6; next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); /* * Transparently forwarded. Pretend to be the destination. * Already got one like this? */ inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); if (!inp) { /* * It's new. Try to find the ambushing socket. * Because we've rewritten the destination address, * any hardware-generated hash is ignored. */ inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &next_hop6->sin6_addr, next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : th->th_dport, lookupflag, m->m_pkthdr.rcvif); } } else if (isipv6) { inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET if (fwd_tag != NULL) { struct sockaddr_in *next_hop; next_hop = (struct sockaddr_in *)(fwd_tag+1); /* * Transparently forwarded. Pretend to be the destination. * already got one like this? */ inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); if (!inp) { /* * It's new. Try to find the ambushing socket. * Because we've rewritten the destination address, * any hardware-generated hash is ignored. */ inp = in_pcblookup(&V_tcbinfo, ip->ip_src, th->th_sport, next_hop->sin_addr, next_hop->sin_port ? ntohs(next_hop->sin_port) : th->th_dport, lookupflag, m->m_pkthdr.rcvif); } } else inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); #endif /* INET */ /* * If the INPCB does not exist then all data in the incoming * segment is discarded and an appropriate RST is sent back. * XXX MRT Send RST using which routing table? */ if (inp == NULL) { if (rstreason != 0) { /* We came here after second (safety) lookup. */ MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); goto dropwithreset; } /* * Log communication attempts to ports that are not * in use. */ if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || V_tcp_log_in_vain == 2) { if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) log(LOG_INFO, "%s; %s: Connection attempt " "to closed port\n", s, __func__); } /* * When blackholing do not respond with a RST but * completely ignore the segment and drop it. */ if (((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole == 2) && (V_blackhole_local || ( #ifdef INET6 isipv6 ? !in6_localaddr(&ip6->ip6_src) : #endif #ifdef INET !in_localip(ip->ip_src) #else true #endif ))) goto dropunlock; rstreason = BANDLIM_RST_CLOSEDPORT; goto dropwithreset; } INP_LOCK_ASSERT(inp); if ((inp->inp_flowtype == M_HASHTYPE_NONE) && !SOLISTENING(inp->inp_socket)) { if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) { inp->inp_flowid = m->m_pkthdr.flowid; inp->inp_flowtype = M_HASHTYPE_GET(m); #ifdef RSS } else { /* assign flowid by software RSS hash */ #ifdef INET6 if (isipv6) { rss_proto_software_hash_v6(&inp->in6p_faddr, &inp->in6p_laddr, inp->inp_fport, inp->inp_lport, IPPROTO_TCP, &inp->inp_flowid, &inp->inp_flowtype); } else #endif /* INET6 */ { rss_proto_software_hash_v4(inp->inp_faddr, inp->inp_laddr, inp->inp_fport, inp->inp_lport, IPPROTO_TCP, &inp->inp_flowid, &inp->inp_flowtype); } #endif /* RSS */ } } #if defined(IPSEC) || defined(IPSEC_SUPPORT) #ifdef INET6 if (isipv6 && IPSEC_ENABLED(ipv6) && IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { goto dropunlock; } #ifdef INET else #endif #endif /* INET6 */ #ifdef INET if (IPSEC_ENABLED(ipv4) && IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { goto dropunlock; } #endif /* INET */ #endif /* IPSEC */ /* * Check the minimum TTL for socket. */ if (inp->inp_ip_minttl != 0) { #ifdef INET6 if (isipv6) { if (inp->inp_ip_minttl > ip6->ip6_hlim) goto dropunlock; } else #endif if (inp->inp_ip_minttl > ip->ip_ttl) goto dropunlock; } tp = intotcpcb(inp); switch (tp->t_state) { case TCPS_TIME_WAIT: /* * A previous connection in TIMEWAIT state is supposed to catch * stray or duplicate segments arriving late. If this segment * was a legitimate new connection attempt, the old INPCB gets * removed and we can try again to find a listening socket. */ tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0); /* * tcp_twcheck unlocks the inp always, and frees the m if fails. */ if (tcp_twcheck(inp, &to, th, m, tlen)) goto findpcb; return (IPPROTO_DONE); case TCPS_CLOSED: /* * The TCPCB may no longer exist if the connection is winding * down or it is in the CLOSED state. Either way we drop the * segment and send an appropriate response. */ rstreason = BANDLIM_RST_CLOSEDPORT; goto dropwithreset; } if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { rstreason = BANDLIM_RST_CLOSEDPORT; goto dropwithreset; } #ifdef TCP_OFFLOAD if (tp->t_flags & TF_TOE) { tcp_offload_input(tp, m); m = NULL; /* consumed by the TOE driver */ goto dropunlock; } #endif #ifdef MAC if (mac_inpcb_check_deliver(inp, m)) goto dropunlock; #endif so = inp->inp_socket; KASSERT(so != NULL, ("%s: so == NULL", __func__)); /* * When the socket is accepting connections (the INPCB is in LISTEN * state) we look into the SYN cache if this is a new connection * attempt or the completion of a previous one. */ KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), ("%s: so accepting but tp %p not listening", __func__, tp)); if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { struct in_conninfo inc; bzero(&inc, sizeof(inc)); #ifdef INET6 if (isipv6) { inc.inc_flags |= INC_ISIPV6; if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) inc.inc_flags |= INC_IPV6MINMTU; inc.inc6_faddr = ip6->ip6_src; inc.inc6_laddr = ip6->ip6_dst; } else #endif { inc.inc_faddr = ip->ip_src; inc.inc_laddr = ip->ip_dst; } inc.inc_fport = th->th_sport; inc.inc_lport = th->th_dport; inc.inc_fibnum = so->so_fibnum; /* * Check for an existing connection attempt in syncache if * the flag is only ACK. A successful lookup creates a new * socket appended to the listen queue in SYN_RECEIVED state. */ if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { /* * Parse the TCP options here because * syncookies need access to the reflected * timestamp. */ tcp_dooptions(&to, optp, optlen, 0); /* * NB: syncache_expand() doesn't unlock inp. */ rstreason = syncache_expand(&inc, &to, th, &so, m, port); if (rstreason < 0) { /* * A failing TCP MD5 signature comparison * must result in the segment being dropped * and must not produce any response back * to the sender. */ goto dropunlock; } else if (rstreason == 0) { /* * No syncache entry, or ACK was not for our * SYN/ACK. Do our protection against double * ACK. If peer sent us 2 ACKs, then for the * first one syncache_expand() successfully * converted syncache entry into a socket, * while we were waiting on the inpcb lock. We * don't want to sent RST for the second ACK, * so we perform second lookup without wildcard * match, hoping to find the new socket. If * the ACK is stray indeed, rstreason would * hint the above code that the lookup was a * second attempt. * * NB: syncache did its own logging * of the failure cause. */ INP_WUNLOCK(inp); rstreason = BANDLIM_RST_OPENPORT; lookupflag &= ~INPLOOKUP_WILDCARD; goto findpcb; } tfo_socket_result: if (so == NULL) { /* * We completed the 3-way handshake * but could not allocate a socket * either due to memory shortage, * listen queue length limits or * global socket limits. Send RST * or wait and have the remote end * retransmit the ACK for another * try. */ if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Socket allocation failed due to " "limits or memory shortage, %s\n", s, __func__, V_tcp_sc_rst_sock_fail ? "sending RST" : "try again"); if (V_tcp_sc_rst_sock_fail) { rstreason = BANDLIM_UNLIMITED; goto dropwithreset; } else goto dropunlock; } /* * Socket is created in state SYN_RECEIVED. * Unlock the listen socket, lock the newly * created socket and update the tp variable. * If we came here via jump to tfo_socket_result, * then listening socket is read-locked. */ INP_UNLOCK(inp); /* listen socket */ inp = sotoinpcb(so); /* * New connection inpcb is already locked by * syncache_expand(). */ INP_WLOCK_ASSERT(inp); tp = intotcpcb(inp); KASSERT(tp->t_state == TCPS_SYN_RECEIVED, ("%s: ", __func__)); /* * Process the segment and the data it * contains. tcp_do_segment() consumes * the mbuf chain and unlocks the inpcb. */ TCP_PROBE5(receive, NULL, tp, m, tp, th); tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); return (IPPROTO_DONE); } /* * Segment flag validation for new connection attempts: * * Our (SYN|ACK) response was rejected. * Check with syncache and remove entry to prevent * retransmits. * * NB: syncache_chkrst does its own logging of failure * causes. */ if (thflags & TH_RST) { syncache_chkrst(&inc, th, m, port); goto dropunlock; } /* * We can't do anything without SYN. */ if ((thflags & TH_SYN) == 0) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN is missing, segment ignored\n", s, __func__); TCPSTAT_INC(tcps_badsyn); goto dropunlock; } /* * (SYN|ACK) is bogus on a listen socket. */ if (thflags & TH_ACK) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN|ACK invalid, segment rejected\n", s, __func__); syncache_badack(&inc, port); /* XXX: Not needed! */ TCPSTAT_INC(tcps_badsyn); rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } /* * If the drop_synfin option is enabled, drop all * segments with both the SYN and FIN bits set. * This prevents e.g. nmap from identifying the * TCP/IP stack. * XXX: Poor reasoning. nmap has other methods * and is constantly refining its stack detection * strategies. * XXX: This is a violation of the TCP specification * and was used by RFC1644. */ if ((thflags & TH_FIN) && V_drop_synfin) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN|FIN segment ignored (based on " "sysctl setting)\n", s, __func__); TCPSTAT_INC(tcps_badsyn); goto dropunlock; } /* * Segment's flags are (SYN) or (SYN|FIN). * * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored * as they do not affect the state of the TCP FSM. * The data pointed to by TH_URG and th_urp is ignored. */ KASSERT((thflags & (TH_RST|TH_ACK)) == 0, ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); KASSERT(thflags & (TH_SYN), ("%s: Listen socket: TH_SYN not set", __func__)); INP_RLOCK_ASSERT(inp); #ifdef INET6 /* * If deprecated address is forbidden, * we do not accept SYN to deprecated interface * address to prevent any new inbound connection from * getting established. * When we do not accept SYN, we send a TCP RST, * with deprecated source address (instead of dropping * it). We compromise it as it is much better for peer * to send a RST, and RST will be the final packet * for the exchange. * * If we do not forbid deprecated addresses, we accept * the SYN packet. RFC2462 does not suggest dropping * SYN in this case. * If we decipher RFC2462 5.5.4, it says like this: * 1. use of deprecated addr with existing * communication is okay - "SHOULD continue to be * used" * 2. use of it with new communication: * (2a) "SHOULD NOT be used if alternate address * with sufficient scope is available" * (2b) nothing mentioned otherwise. * Here we fall into (2b) case as we have no choice in * our source address selection - we must obey the peer. * * The wording in RFC2462 is confusing, and there are * multiple description text for deprecated address * handling - worse, they are not exactly the same. * I believe 5.5.4 is the best one, so we follow 5.5.4. */ if (isipv6 && !V_ip6_use_deprecated) { struct in6_ifaddr *ia6; ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); if (ia6 != NULL && (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt to deprecated " "IPv6 address rejected\n", s, __func__); rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } } #endif /* INET6 */ /* * Basic sanity checks on incoming SYN requests: * Don't respond if the destination is a link layer * broadcast according to RFC1122 4.2.3.10, p. 104. * If it is from this socket it must be forged. * Don't respond if the source or destination is a * global or subnet broad- or multicast address. * Note that it is quite possible to receive unicast * link-layer packets with a broadcast IP address. Use * in_broadcast() to find them. */ if (m->m_flags & (M_BCAST|M_MCAST)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from broad- or multicast " "link layer address ignored\n", s, __func__); goto dropunlock; } #ifdef INET6 if (isipv6) { if (th->th_dport == th->th_sport && IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt to/from self " "ignored\n", s, __func__); goto dropunlock; } if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to multicast " "address ignored\n", s, __func__); goto dropunlock; } } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { if (th->th_dport == th->th_sport && ip->ip_dst.s_addr == ip->ip_src.s_addr) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to self " "ignored\n", s, __func__); goto dropunlock; } if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to broad- " "or multicast address ignored\n", s, __func__); goto dropunlock; } } #endif /* * SYN appears to be valid. Create compressed TCP state * for syncache. */ TCP_PROBE3(debug__input, tp, th, m); tcp_dooptions(&to, optp, optlen, TO_SYN); if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, iptos, port)) != NULL) goto tfo_socket_result; /* * Entry added to syncache and mbuf consumed. * Only the listen socket is unlocked by syncache_add(). */ return (IPPROTO_DONE); } else if (tp->t_state == TCPS_LISTEN) { /* * When a listen socket is torn down the SO_ACCEPTCONN * flag is removed first while connections are drained * from the accept queue in a unlock/lock cycle of the * ACCEPT_LOCK, opening a race condition allowing a SYN * attempt go through unhandled. */ goto dropunlock; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) { tcp_dooptions(&to, optp, optlen, thflags); if ((to.to_flags & TOF_SIGNATURE) == 0) { TCPSTAT_INC(tcps_sig_err_nosigopt); goto dropunlock; } if (!TCPMD5_ENABLED() || TCPMD5_INPUT(m, th, to.to_signature) != 0) goto dropunlock; } #endif TCP_PROBE5(receive, NULL, tp, m, tp, th); /* * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later * state. tcp_do_segment() always consumes the mbuf chain, unlocks * the inpcb, and unlocks pcbinfo. * * XXXGL: in case of a pure SYN arriving on existing connection * TCP stacks won't need to modify the PCB, they would either drop * the segment silently, or send a challenge ACK. However, we try * to upgrade the lock, because calling convention for stacks is * write-lock on PCB. If upgrade fails, drop the SYN. */ if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) goto dropunlock; tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); return (IPPROTO_DONE); dropwithreset: TCP_PROBE5(receive, NULL, tp, m, tp, th); if (inp != NULL) { tcp_dropwithreset(m, th, tp, tlen, rstreason); INP_UNLOCK(inp); } else tcp_dropwithreset(m, th, NULL, tlen, rstreason); m = NULL; /* mbuf chain got consumed. */ goto drop; dropunlock: if (m != NULL) TCP_PROBE5(receive, NULL, tp, m, tp, th); if (inp != NULL) INP_UNLOCK(inp); drop: if (s != NULL) free(s, M_TCPLOG); if (m != NULL) m_freem(m); return (IPPROTO_DONE); } /* * Automatic sizing of receive socket buffer. Often the send * buffer size is not optimally adjusted to the actual network * conditions at hand (delay bandwidth product). Setting the * buffer size too small limits throughput on links with high * bandwidth and high delay (eg. trans-continental/oceanic links). * * On the receive side the socket buffer memory is only rarely * used to any significant extent. This allows us to be much * more aggressive in scaling the receive socket buffer. For * the case that the buffer space is actually used to a large * extent and we run out of kernel memory we can simply drop * the new segments; TCP on the sender will just retransmit it * later. Setting the buffer size too big may only consume too * much kernel memory if the application doesn't read() from * the socket or packet loss or reordering makes use of the * reassembly queue. * * The criteria to step up the receive buffer one notch are: * 1. Application has not set receive buffer size with * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. * 2. the number of bytes received during 1/2 of an sRTT * is at least 3/8 of the current socket buffer size. * 3. receive buffer size has not hit maximal automatic size; * * If all of the criteria are met we increaset the socket buffer * by a 1/2 (bounded by the max). This allows us to keep ahead * of slow-start but also makes it so our peer never gets limited * by our rwnd which we then open up causing a burst. * * This algorithm does two steps per RTT at most and only if * we receive a bulk stream w/o packet losses or reorderings. * Shrinking the buffer during idle times is not necessary as * it doesn't consume any memory when idle. * * TODO: Only step up if the application is actually serving * the buffer to better manage the socket buffer resources. */ int tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, int tlen) { int newsize = 0; if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && tp->t_srtt != 0 && tp->rfbuf_ts != 0 && TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); } TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); /* Start over with next RTT. */ tp->rfbuf_ts = 0; tp->rfbuf_cnt = 0; } else { tp->rfbuf_cnt += tlen; /* add up */ } return (newsize); } int tcp_input(struct mbuf **mp, int *offp, int proto) { return(tcp_input_with_port(mp, offp, proto, 0)); } static void tcp_handle_wakeup(struct tcpcb *tp) { INP_WLOCK_ASSERT(tptoinpcb(tp)); if (tp->t_flags & TF_WAKESOR) { struct socket *so = tptosocket(tp); tp->t_flags &= ~TF_WAKESOR; SOCKBUF_LOCK_ASSERT(&so->so_rcv); sorwakeup_locked(so); } } void tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, int drop_hdrlen, int tlen, uint8_t iptos) { uint16_t thflags; - int acked, ourfinisacked, needoutput = 0, sack_changed; + int acked, ourfinisacked, needoutput = 0; + sackstatus_t sack_changed; int rstreason, todrop, win, incforsyn = 0; uint32_t tiwin; uint16_t nsegs; char *s; struct inpcb *inp = tptoinpcb(tp); struct socket *so = tptosocket(tp); struct in_conninfo *inc = &inp->inp_inc; struct mbuf *mfree; struct tcpopt to; int tfo_syn; u_int maxseg; thflags = tcp_get_flags(th); tp->sackhint.last_sack_ack = 0; - sack_changed = 0; + sack_changed = SACK_NOCHANGE; nsegs = max(1, m->m_pkthdr.lro_nsegs); NET_EPOCH_ASSERT(); INP_WLOCK_ASSERT(inp); KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", __func__)); KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", __func__)); #ifdef TCPPCAP /* Save segment, if requested. */ tcp_pcap_add(th, m, &(tp->t_inpkts)); #endif TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, tlen, NULL, true); if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: " "SYN|FIN segment ignored (based on " "sysctl setting)\n", s, __func__); free(s, M_TCPLOG); } goto drop; } /* * If a segment with the ACK-bit set arrives in the SYN-SENT state * check SEQ.ACK first. */ if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { rstreason = BANDLIM_UNLIMITED; tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); goto dropwithreset; } /* * Segment received on connection. * Reset idle time and keep-alive timer. * XXX: This should be done after segment * validation to ignore broken/spoofed segs. */ if (tp->t_idle_reduce && (tp->snd_max == tp->snd_una) && ((ticks - tp->t_rcvtime) >= tp->t_rxtcur)) cc_after_idle(tp); tp->t_rcvtime = ticks; if (thflags & TH_FIN) tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN); /* * Scale up the window into a 32-bit value. * For the SYN_SENT state the scale is zero. */ tiwin = th->th_win << tp->snd_scale; #ifdef STATS stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); #endif /* * TCP ECN processing. */ if (tcp_ecn_input_segment(tp, thflags, tlen, tcp_packets_this_ack(tp, th->th_ack), iptos)) cc_cong_signal(tp, th, CC_ECN); /* * Parse options on any incoming segment. */ tcp_dooptions(&to, (u_char *)(th + 1), (th->th_off << 2) - sizeof(struct tcphdr), (thflags & TH_SYN) ? TO_SYN : 0); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if ((tp->t_flags & TF_SIGNATURE) != 0 && (to.to_flags & TOF_SIGNATURE) == 0) { TCPSTAT_INC(tcps_sig_err_sigopt); /* XXX: should drop? */ } #endif /* * If echoed timestamp is later than the current time, * fall back to non RFC1323 RTT calculation. Normalize * timestamp if syncookies were used when this connection * was established. */ if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { to.to_tsecr -= tp->ts_offset; if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) to.to_tsecr = 0; else if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && tp->t_badrxtwin != 0 && TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) cc_cong_signal(tp, th, CC_RTO_ERR); } /* * Process options only when we get SYN/ACK back. The SYN case * for incoming connections is handled in tcp_syncache. * According to RFC1323 the window field in a SYN (i.e., a * or ) segment itself is never scaled. * XXX this is traditional behavior, may need to be cleaned up. */ if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { /* Handle parallel SYN for ECN */ tcp_ecn_input_parallel_syn(tp, thflags, iptos); if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE) && !(tp->t_flags & TF_NOOPT)) { tp->t_flags |= TF_RCVD_SCALE; tp->snd_scale = to.to_wscale; } else tp->t_flags &= ~TF_REQ_SCALE; /* * Initial send window. It will be updated with * the next incoming segment to the scaled value. */ tp->snd_wnd = th->th_win; if ((to.to_flags & TOF_TS) && (tp->t_flags & TF_REQ_TSTMP) && !(tp->t_flags & TF_NOOPT)) { tp->t_flags |= TF_RCVD_TSTMP; tp->ts_recent = to.to_tsval; tp->ts_recent_age = tcp_ts_getticks(); } else tp->t_flags &= ~TF_REQ_TSTMP; if (to.to_flags & TOF_MSS) tcp_mss(tp, to.to_mss); if ((tp->t_flags & TF_SACK_PERMIT) && (!(to.to_flags & TOF_SACKPERM) || (tp->t_flags & TF_NOOPT))) tp->t_flags &= ~TF_SACK_PERMIT; if (IS_FASTOPEN(tp->t_flags)) { if ((to.to_flags & TOF_FASTOPEN) && !(tp->t_flags & TF_NOOPT)) { uint16_t mss; if (to.to_flags & TOF_MSS) mss = to.to_mss; else if ((inp->inp_vflag & INP_IPV6) != 0) mss = TCP6_MSS; else mss = TCP_MSS; tcp_fastopen_update_cache(tp, mss, to.to_tfo_len, to.to_tfo_cookie); } else tcp_fastopen_disable_path(tp); } } /* * If timestamps were negotiated during SYN/ACK and a * segment without a timestamp is received, silently drop * the segment, unless it is a RST segment or missing timestamps are * tolerated. * See section 3.2 of RFC 7323. */ if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp missing, " "segment processed normally\n", s, __func__); free(s, M_TCPLOG); } } else { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp missing, " "segment silently dropped\n", s, __func__); free(s, M_TCPLOG); } goto drop; } } /* * If timestamps were not negotiated during SYN/ACK and a * segment with a timestamp is received, ignore the * timestamp and process the packet normally. * See section 3.2 of RFC 7323. */ if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp not expected, " "segment processed normally\n", s, __func__); free(s, M_TCPLOG); } } /* * Header prediction: check for the two common cases * of a uni-directional data xfer. If the packet has * no control flags, is in-sequence, the window didn't * change and we're not retransmitting, it's a * candidate. If the length is zero and the ack moved * forward, we're the sender side of the xfer. Just * free the data acked & wake any higher level process * that was blocked waiting for space. If the length * is non-zero and the ack didn't move, we're the * receiver side. If we're getting packets in-order * (the reassembly queue is empty), add the data to * the socket buffer and note that we need a delayed ack. * Make sure that the hidden state-flags are also off. * Since we check for TCPS_ESTABLISHED first, it can only * be TH_NEEDSYN. */ if (tp->t_state == TCPS_ESTABLISHED && th->th_seq == tp->rcv_nxt && (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && tp->snd_nxt == tp->snd_max && tiwin && tiwin == tp->snd_wnd && ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && SEGQ_EMPTY(tp) && ((to.to_flags & TOF_TS) == 0 || TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { /* * If last ACK falls within this segment's sequence numbers, * record the timestamp. * NOTE that the test is modified according to the latest * proposal of the tcplw@cray.com list (Braden 1993/04/26). */ if ((to.to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { tp->ts_recent_age = tcp_ts_getticks(); tp->ts_recent = to.to_tsval; } if (tlen == 0) { if (SEQ_GT(th->th_ack, tp->snd_una) && SEQ_LEQ(th->th_ack, tp->snd_max) && !IN_RECOVERY(tp->t_flags) && (to.to_flags & TOF_SACK) == 0 && TAILQ_EMPTY(&tp->snd_holes)) { /* * This is a pure ack for outstanding data. */ TCPSTAT_INC(tcps_predack); /* * "bad retransmit" recovery without timestamps. */ if ((to.to_flags & TOF_TS) == 0 && tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && tp->t_badrxtwin != 0 && TSTMP_LT(ticks, tp->t_badrxtwin)) { cc_cong_signal(tp, th, CC_RTO_ERR); } /* * Recalculate the transmit timer / rtt. * * Some boxes send broken timestamp replies * during the SYN+ACK phase, ignore * timestamps of 0 or we could calculate a * huge RTT and blow up the retransmit timer. */ if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { uint32_t t; t = tcp_ts_getticks() - to.to_tsecr; if (!tp->t_rttlow || tp->t_rttlow > t) tp->t_rttlow = t; tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) tp->t_rttlow = ticks - tp->t_rtttime; tcp_xmit_timer(tp, ticks - tp->t_rtttime); } acked = BYTES_THIS_ACK(tp, th); #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ hhook_run_tcp_est_in(tp, th, &to); #endif TCPSTAT_ADD(tcps_rcvackpack, nsegs); TCPSTAT_ADD(tcps_rcvackbyte, acked); sbdrop(&so->so_snd, acked); if (SEQ_GT(tp->snd_una, tp->snd_recover) && SEQ_LEQ(th->th_ack, tp->snd_recover)) tp->snd_recover = th->th_ack - 1; /* * Let the congestion control algorithm update * congestion control related information. This * typically means increasing the congestion * window. */ cc_ack_received(tp, th, nsegs, CC_ACK); tp->snd_una = th->th_ack; /* * Pull snd_wl2 up to prevent seq wrap relative * to th_ack. */ tp->snd_wl2 = th->th_ack; tp->t_dupacks = 0; m_freem(m); /* * If all outstanding data are acked, stop * retransmit timer, otherwise restart timer * using current (possibly backed-off) value. * If process is waiting for space, * wakeup/selwakeup/signal. If data * are ready to send, let tcp_output * decide between more output or persist. */ TCP_PROBE3(debug__input, tp, th, m); /* * Clear t_acktime if remote side has ACKd * all data in the socket buffer. * Otherwise, update t_acktime if we received * a sufficiently large ACK. */ if (sbavail(&so->so_snd) == 0) tp->t_acktime = 0; else if (acked > 1) tp->t_acktime = ticks; if (tp->snd_una == tp->snd_max) tcp_timer_activate(tp, TT_REXMT, 0); else if (!tcp_timer_active(tp, TT_PERSIST)) tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); sowwakeup(so); if (sbavail(&so->so_snd)) (void) tcp_output(tp); goto check_delack; } } else if (th->th_ack == tp->snd_una && tlen <= sbspace(&so->so_rcv)) { int newsize = 0; /* automatic sockbuf scaling */ /* * This is a pure, in-sequence data packet with * nothing on the reassembly queue and we have enough * buffer space to take it. */ /* Clean receiver SACK report if present */ if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) tcp_clean_sackreport(tp); TCPSTAT_INC(tcps_preddat); tp->rcv_nxt += tlen; if (tlen && ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && (tp->t_fbyte_in == 0)) { tp->t_fbyte_in = ticks; if (tp->t_fbyte_in == 0) tp->t_fbyte_in = 1; if (tp->t_fbyte_out && tp->t_fbyte_in) tp->t_flags2 |= TF2_FBYTES_COMPLETE; } /* * Pull snd_wl1 up to prevent seq wrap relative to * th_seq. */ tp->snd_wl1 = th->th_seq; /* * Pull rcv_up up to prevent seq wrap relative to * rcv_nxt. */ tp->rcv_up = tp->rcv_nxt; TCPSTAT_ADD(tcps_rcvpack, nsegs); TCPSTAT_ADD(tcps_rcvbyte, tlen); TCP_PROBE3(debug__input, tp, th, m); newsize = tcp_autorcvbuf(m, th, so, tp, tlen); /* Add data to socket buffer. */ SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { m_freem(m); } else { /* * Set new socket buffer size. * Give up when limit is reached. */ if (newsize) if (!sbreserve_locked(so, SO_RCV, newsize, NULL)) so->so_rcv.sb_flags &= ~SB_AUTOSIZE; m_adj(m, drop_hdrlen); /* delayed header drop */ sbappendstream_locked(&so->so_rcv, m, 0); } /* NB: sorwakeup_locked() does an implicit unlock. */ sorwakeup_locked(so); if (DELAY_ACK(tp, tlen)) { tp->t_flags |= TF_DELACK; } else { tp->t_flags |= TF_ACKNOW; tcp_output(tp); } goto check_delack; } } /* * Calculate amount of space in receive window, * and then do TCP input processing. * Receive window is amount of space in rcv queue, * but not less than advertised window. */ win = sbspace(&so->so_rcv); if (win < 0) win = 0; tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); switch (tp->t_state) { /* * If the state is SYN_RECEIVED: * if seg contains an ACK, but not for our SYN/ACK, send a RST. */ case TCPS_SYN_RECEIVED: if (thflags & TH_RST) { /* Handle RST segments later. */ break; } if ((thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max))) { rstreason = BANDLIM_RST_OPENPORT; tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); goto dropwithreset; } if (IS_FASTOPEN(tp->t_flags)) { /* * When a TFO connection is in SYN_RECEIVED, the * only valid packets are the initial SYN, a * retransmit/copy of the initial SYN (possibly with * a subset of the original data), a valid ACK, a * FIN, or a RST. */ if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { rstreason = BANDLIM_RST_OPENPORT; tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); goto dropwithreset; } else if (thflags & TH_SYN) { /* non-initial SYN is ignored */ if ((tcp_timer_active(tp, TT_DELACK) || tcp_timer_active(tp, TT_REXMT))) goto drop; } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { goto drop; } } break; /* * If the state is SYN_SENT: * if seg contains a RST with valid ACK (SEQ.ACK has already * been verified), then drop the connection. * if seg contains a RST without an ACK, drop the seg. * if seg does not contain SYN, then drop the seg. * Otherwise this is an acceptable SYN segment * initialize tp->rcv_nxt and tp->irs * if seg contains ack then advance tp->snd_una * if seg contains an ECE and ECN support is enabled, the stream * is ECN capable. * if SYN has been acked change to ESTABLISHED else SYN_RCVD state * arrange for segment to be acked (eventually) * continue processing rest of data/controls, beginning with URG */ case TCPS_SYN_SENT: if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { TCP_PROBE5(connect__refused, NULL, tp, m, tp, th); tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); tp = tcp_drop(tp, ECONNREFUSED); } if (thflags & TH_RST) goto drop; if (!(thflags & TH_SYN)) goto drop; tp->irs = th->th_seq; tcp_rcvseqinit(tp); if (thflags & TH_ACK) { int tfo_partial_ack = 0; TCPSTAT_INC(tcps_connects); soisconnected(so); #ifdef MAC mac_socketpeer_set_from_mbuf(m, so); #endif /* Do window scaling on this connection? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; } tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN << tp->rcv_scale); tp->snd_una++; /* SYN is acked */ /* * If not all the data that was sent in the TFO SYN * has been acked, resend the remainder right away. */ if (IS_FASTOPEN(tp->t_flags) && (tp->snd_una != tp->snd_max)) { tp->snd_nxt = th->th_ack; tfo_partial_ack = 1; } /* * If there's data, delay ACK; if there's also a FIN * ACKNOW will be turned on later. */ if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); else tp->t_flags |= TF_ACKNOW; tcp_ecn_input_syn_sent(tp, thflags, iptos); /* * Received in SYN_SENT[*] state. * Transitions: * SYN_SENT --> ESTABLISHED * SYN_SENT* --> FIN_WAIT_1 */ tp->t_starttime = ticks; if (tp->t_flags & TF_NEEDFIN) { tp->t_acktime = ticks; tcp_state_change(tp, TCPS_FIN_WAIT_1); tp->t_flags &= ~TF_NEEDFIN; thflags &= ~TH_SYN; } else { tcp_state_change(tp, TCPS_ESTABLISHED); TCP_PROBE5(connect__established, NULL, tp, m, tp, th); cc_conn_init(tp); tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); } } else { /* * Received initial SYN in SYN-SENT[*] state => * simultaneous open. * If it succeeds, connection is * half-synchronized. * Otherwise, do 3-way handshake: * SYN-SENT -> SYN-RECEIVED * SYN-SENT* -> SYN-RECEIVED* */ tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN); tcp_timer_activate(tp, TT_REXMT, 0); tcp_state_change(tp, TCPS_SYN_RECEIVED); } /* * Advance th->th_seq to correspond to first data byte. * If data, trim to stay within window, * dropping FIN if necessary. */ th->th_seq++; if (tlen > tp->rcv_wnd) { todrop = tlen - tp->rcv_wnd; m_adj(m, -todrop); tlen = tp->rcv_wnd; thflags &= ~TH_FIN; TCPSTAT_INC(tcps_rcvpackafterwin); TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); } tp->snd_wl1 = th->th_seq - 1; tp->rcv_up = th->th_seq; /* * Client side of transaction: already sent SYN and data. * If the remote host used T/TCP to validate the SYN, * our data will be ACK'd; if so, enter normal data segment * processing in the middle of step 5, ack processing. * Otherwise, goto step 6. */ if (thflags & TH_ACK) goto process_ACK; goto step6; } /* * States other than LISTEN or SYN_SENT. * First check the RST flag and sequence number since reset segments * are exempt from the timestamp and connection count tests. This * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix * below which allowed reset segments in half the sequence space * to fall though and be processed (which gives forged reset * segments with a random sequence number a 50 percent chance of * killing a connection). * Then check timestamp, if present. * Then check the connection count, if present. * Then check that at least some bytes of segment are within * receive window. If segment begins before rcv_nxt, * drop leading data (and SYN); if nothing left, just ack. */ if (thflags & TH_RST) { /* * RFC5961 Section 3.2 * * - RST drops connection only if SEG.SEQ == RCV.NXT. * - If RST is in window, we send challenge ACK. * * Note: to take into account delayed ACKs, we should * test against last_ack_sent instead of rcv_nxt. * Note 2: we handle special case of closed window, not * covered by the RFC. */ if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { KASSERT(tp->t_state != TCPS_SYN_SENT, ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", __func__, th, tp)); if (V_tcp_insecure_rst || tp->last_ack_sent == th->th_seq) { TCPSTAT_INC(tcps_drops); /* Drop the connection. */ switch (tp->t_state) { case TCPS_SYN_RECEIVED: so->so_error = ECONNREFUSED; goto close; case TCPS_ESTABLISHED: case TCPS_FIN_WAIT_1: case TCPS_FIN_WAIT_2: case TCPS_CLOSE_WAIT: case TCPS_CLOSING: case TCPS_LAST_ACK: so->so_error = ECONNRESET; close: /* FALLTHROUGH */ default: tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST); tp = tcp_close(tp); } } else { TCPSTAT_INC(tcps_badrst); /* Send challenge ACK. */ tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, tp->snd_nxt, TH_ACK); tp->last_ack_sent = tp->rcv_nxt; m = NULL; } } goto drop; } /* * RFC5961 Section 4.2 * Send challenge ACK for any SYN in synchronized state. */ if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && tp->t_state != TCPS_SYN_RECEIVED) { TCPSTAT_INC(tcps_badsyn); if (V_tcp_insecure_syn && SEQ_GEQ(th->th_seq, tp->last_ack_sent) && SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); tp = tcp_drop(tp, ECONNRESET); rstreason = BANDLIM_UNLIMITED; } else { tcp_ecn_input_syn_sent(tp, thflags, iptos); /* Send challenge ACK. */ tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, tp->snd_nxt, TH_ACK); tp->last_ack_sent = tp->rcv_nxt; m = NULL; } goto drop; } /* * RFC 1323 PAWS: If we have a timestamp reply on this segment * and it's less than ts_recent, drop it. */ if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { /* Check to see if ts_recent is over 24 days old. */ if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { /* * Invalidate ts_recent. If this segment updates * ts_recent, the age will be reset later and ts_recent * will get a valid value. If it does not, setting * ts_recent to zero will at least satisfy the * requirement that zero be placed in the timestamp * echo reply when ts_recent isn't valid. The * age isn't reset until we get a valid ts_recent * because we don't want out-of-order segments to be * dropped when ts_recent is old. */ tp->ts_recent = 0; } else { TCPSTAT_INC(tcps_rcvduppack); TCPSTAT_ADD(tcps_rcvdupbyte, tlen); TCPSTAT_INC(tcps_pawsdrop); if (tlen) goto dropafterack; goto drop; } } /* * In the SYN-RECEIVED state, validate that the packet belongs to * this connection before trimming the data to fit the receive * window. Check the sequence number versus IRS since we know * the sequence numbers haven't wrapped. This is a partial fix * for the "LAND" DoS attack. */ if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { rstreason = BANDLIM_RST_OPENPORT; tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); goto dropwithreset; } todrop = tp->rcv_nxt - th->th_seq; if (todrop > 0) { if (thflags & TH_SYN) { thflags &= ~TH_SYN; th->th_seq++; if (th->th_urp > 1) th->th_urp--; else thflags &= ~TH_URG; todrop--; } /* * Following if statement from Stevens, vol. 2, p. 960. */ if (todrop > tlen || (todrop == tlen && (thflags & TH_FIN) == 0)) { /* * Any valid FIN must be to the left of the window. * At this point the FIN must be a duplicate or out * of sequence; drop it. */ thflags &= ~TH_FIN; /* * Send an ACK to resynchronize and drop any data. * But keep on processing for RST or ACK. */ tp->t_flags |= TF_ACKNOW; todrop = tlen; TCPSTAT_INC(tcps_rcvduppack); TCPSTAT_ADD(tcps_rcvdupbyte, todrop); } else { TCPSTAT_INC(tcps_rcvpartduppack); TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); } /* * DSACK - add SACK block for dropped range */ if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { tcp_update_sack_list(tp, th->th_seq, th->th_seq + todrop); /* * ACK now, as the next in-sequence segment * will clear the DSACK block again */ tp->t_flags |= TF_ACKNOW; } drop_hdrlen += todrop; /* drop from the top afterwards */ th->th_seq += todrop; tlen -= todrop; if (th->th_urp > todrop) th->th_urp -= todrop; else { thflags &= ~TH_URG; th->th_urp = 0; } } /* * If new data are received on a connection after the * user processes are gone, then RST the other end. */ if ((tp->t_flags & TF_CLOSED) && tlen) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " "after socket was closed, " "sending RST and removing tcpcb\n", s, __func__, tcpstates[tp->t_state], tlen); free(s, M_TCPLOG); } tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE); /* tcp_close will kill the inp pre-log the Reset */ tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); tp = tcp_close(tp); TCPSTAT_INC(tcps_rcvafterclose); rstreason = BANDLIM_UNLIMITED; goto dropwithreset; } /* * If segment ends after window, drop trailing data * (and PUSH and FIN); if nothing left, just ACK. */ todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); if (todrop > 0) { TCPSTAT_INC(tcps_rcvpackafterwin); if (todrop >= tlen) { TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); /* * If window is closed can only take segments at * window edge, and have to drop data and PUSH from * incoming segments. Continue processing, but * remember to ack. Otherwise, drop segment * and ack. */ if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { tp->t_flags |= TF_ACKNOW; TCPSTAT_INC(tcps_rcvwinprobe); } else goto dropafterack; } else TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); m_adj(m, -todrop); tlen -= todrop; thflags &= ~(TH_PUSH|TH_FIN); } /* * If last ACK falls within this segment's sequence numbers, * record its timestamp. * NOTE: * 1) That the test incorporates suggestions from the latest * proposal of the tcplw@cray.com list (Braden 1993/04/26). * 2) That updating only on newer timestamps interferes with * our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. * 3) That we modify the segment boundary check to be * Last.ACK.Sent <= SEG.SEQ + SEG.Len * instead of RFC1323's * Last.ACK.Sent < SEG.SEQ + SEG.Len, * This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated * Vol. 2 p.869. In such cases, we can still calculate the * RTT correctly when RCV.NXT == Last.ACK.Sent. */ if ((to.to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN|TH_FIN)) != 0))) { tp->ts_recent_age = tcp_ts_getticks(); tp->ts_recent = to.to_tsval; } /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN * flag is on (half-synchronized state), then queue data for * later processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (tp->t_state == TCPS_SYN_RECEIVED || (tp->t_flags & TF_NEEDSYN)) { if (tp->t_state == TCPS_SYN_RECEIVED && IS_FASTOPEN(tp->t_flags)) { tp->snd_wnd = tiwin; cc_conn_init(tp); } goto step6; } else if (tp->t_flags & TF_ACKNOW) goto dropafterack; else goto drop; } /* * Ack processing. */ switch (tp->t_state) { /* * In SYN_RECEIVED state, the ack ACKs our SYN, so enter * ESTABLISHED state and continue processing. * The ACK was checked above. */ case TCPS_SYN_RECEIVED: TCPSTAT_INC(tcps_connects); if (tp->t_flags & TF_SONOTCONN) { /* * Usually SYN_RECEIVED had been created from a LISTEN, * and solisten_enqueue() has already marked the socket * layer as connected. If it didn't, which can happen * only with an accept_filter(9), then the tp is marked * with TF_SONOTCONN. The other reason for this mark * to be set is a simultaneous open, a SYN_RECEIVED * that had been created from SYN_SENT. */ tp->t_flags &= ~TF_SONOTCONN; soisconnected(so); } /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; } tp->snd_wnd = tiwin; /* * Make transitions: * SYN-RECEIVED -> ESTABLISHED * SYN-RECEIVED* -> FIN-WAIT-1 */ tp->t_starttime = ticks; if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { tcp_fastopen_decrement_counter(tp->t_tfo_pending); tp->t_tfo_pending = NULL; } if (tp->t_flags & TF_NEEDFIN) { tp->t_acktime = ticks; tcp_state_change(tp, TCPS_FIN_WAIT_1); tp->t_flags &= ~TF_NEEDFIN; } else { tcp_state_change(tp, TCPS_ESTABLISHED); TCP_PROBE5(accept__established, NULL, tp, m, tp, th); /* * TFO connections call cc_conn_init() during SYN * processing. Calling it again here for such * connections is not harmless as it would undo the * snd_cwnd reduction that occurs when a TFO SYN|ACK * is retransmitted. */ if (!IS_FASTOPEN(tp->t_flags)) cc_conn_init(tp); tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); } /* * Account for the ACK of our SYN prior to * regular ACK processing below, except for * simultaneous SYN, which is handled later. */ if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) incforsyn = 1; /* * If segment contains data or ACK, will call tcp_reass() * later; if not, do so now to pass queued data to user. */ if (tlen == 0 && (thflags & TH_FIN) == 0) { (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, (struct mbuf *)0); tcp_handle_wakeup(tp); } tp->snd_wl1 = th->th_seq - 1; /* FALLTHROUGH */ /* * In ESTABLISHED state: drop duplicate ACKs; ACK out of range * ACKs. If the ack is in the range * tp->snd_una < th->th_ack <= tp->snd_max * then advance tp->snd_una to th->th_ack and drop * data from the retransmission queue. If this ACK reflects * more up to date window information we update our window information. */ case TCPS_ESTABLISHED: case TCPS_FIN_WAIT_1: case TCPS_FIN_WAIT_2: case TCPS_CLOSE_WAIT: case TCPS_CLOSING: case TCPS_LAST_ACK: if (SEQ_GT(th->th_ack, tp->snd_max)) { TCPSTAT_INC(tcps_rcvacktoomuch); goto dropafterack; } if (tcp_is_sack_recovery(tp, &to)) { if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) && (tp->t_flags & TF_LRD)) { tcp_sack_lost_retransmission(tp, th); } } else /* * Reset the value so that previous (valid) value * from the last ack with SACK doesn't get used. */ tp->sackhint.sacked_bytes = 0; #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ hhook_run_tcp_est_in(tp, th, &to); #endif if (SEQ_LEQ(th->th_ack, tp->snd_una)) { maxseg = tcp_maxseg(tp); if (tlen == 0 && (tiwin == tp->snd_wnd || (tp->t_flags & TF_SACK_PERMIT))) { /* * If this is the first time we've seen a * FIN from the remote, this is not a * duplicate and it needs to be processed * normally. This happens during a * simultaneous close. */ if ((thflags & TH_FIN) && (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { tp->t_dupacks = 0; break; } TCPSTAT_INC(tcps_rcvdupack); /* * If we have outstanding data (other than * a window probe), this is a completely * duplicate ack (ie, window info didn't * change and FIN isn't set), * the ack is the biggest we've * seen and we've seen exactly our rexmt * threshold of them, assume a packet * has been dropped and retransmit it. * Kludge snd_nxt & the congestion * window so we send only this one * packet. * * We know we're losing at the current * window size so do congestion avoidance * (set ssthresh to half the current window * and pull our congestion window back to * the new ssthresh). * * Dup acks mean that packets have left the * network (they're now cached at the receiver) * so bump cwnd by the amount in the receiver * to keep a constant cwnd packets in the * network. * * When using TCP ECN, notify the peer that * we reduced the cwnd. */ /* * Following 2 kinds of acks should not affect * dupack counting: * 1) Old acks * 2) Acks with SACK but without any new SACK * information in them. These could result from * any anomaly in the network like a switch * duplicating packets or a possible DoS attack. */ if (th->th_ack != tp->snd_una || (tcp_is_sack_recovery(tp, &to) && - !sack_changed)) + (sack_changed == SACK_NOCHANGE))) break; else if (!tcp_timer_active(tp, TT_REXMT)) tp->t_dupacks = 0; else if (++tp->t_dupacks > tcprexmtthresh || IN_FASTRECOVERY(tp->t_flags)) { cc_ack_received(tp, th, nsegs, CC_DUPACK); if (V_tcp_do_prr && - IN_FASTRECOVERY(tp->t_flags)) { - tcp_do_prr_ack(tp, th, &to); + IN_FASTRECOVERY(tp->t_flags) && + (tp->t_flags & TF_SACK_PERMIT)) { + tcp_do_prr_ack(tp, th, &to, sack_changed); } else if (tcp_is_sack_recovery(tp, &to) && IN_FASTRECOVERY(tp->t_flags)) { int awnd; /* * Compute the amount of data in flight first. * We can inject new data into the pipe iff * we have less than 1/2 the original window's * worth of data in flight. */ if (V_tcp_do_newsack) awnd = tcp_compute_pipe(tp); else awnd = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; if (awnd < tp->snd_ssthresh) { tp->snd_cwnd += maxseg; if (tp->snd_cwnd > tp->snd_ssthresh) tp->snd_cwnd = tp->snd_ssthresh; } } else tp->snd_cwnd += maxseg; (void) tcp_output(tp); goto drop; } else if (tp->t_dupacks == tcprexmtthresh || (tp->t_flags & TF_SACK_PERMIT && V_tcp_do_newsack && tp->sackhint.sacked_bytes > (tcprexmtthresh - 1) * maxseg)) { enter_recovery: /* * Above is the RFC6675 trigger condition of * more than (dupthresh-1)*maxseg sacked data. * If the count of holes in the * scoreboard is >= dupthresh, we could * also enter loss recovery, but don't * have that value readily available. */ tp->t_dupacks = tcprexmtthresh; tcp_seq onxt = tp->snd_nxt; /* * If we're doing sack, or prr, check * to see if we're already in sack * recovery. If we're not doing sack, * check to see if we're in newreno * recovery. */ if (V_tcp_do_prr || (tp->t_flags & TF_SACK_PERMIT)) { if (IN_FASTRECOVERY(tp->t_flags)) { tp->t_dupacks = 0; break; } } else { if (SEQ_LEQ(th->th_ack, tp->snd_recover)) { tp->t_dupacks = 0; break; } } /* Congestion signal before ack. */ cc_cong_signal(tp, th, CC_NDUPACK); cc_ack_received(tp, th, nsegs, CC_DUPACK); tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; if (V_tcp_do_prr) { /* * snd_ssthresh is already updated by * cc_cong_signal. */ if (tcp_is_sack_recovery(tp, &to)) { + /* + * Exclude Limited Transmit + * segments here + */ tp->sackhint.prr_delivered = - tp->sackhint.sacked_bytes; + maxseg; } else { tp->sackhint.prr_delivered = imin(tp->snd_max - tp->snd_una, imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); } tp->sackhint.recover_fs = max(1, tp->snd_nxt - tp->snd_una); } if (tcp_is_sack_recovery(tp, &to)) { TCPSTAT_INC( tcps_sack_recovery_episode); tp->snd_recover = tp->snd_nxt; tp->snd_cwnd = maxseg; (void) tcp_output(tp); if (SEQ_GT(th->th_ack, tp->snd_una)) goto resume_partialack; goto drop; } tp->snd_nxt = th->th_ack; tp->snd_cwnd = maxseg; (void) tcp_output(tp); KASSERT(tp->snd_limited <= 2, ("%s: tp->snd_limited too big", __func__)); tp->snd_cwnd = tp->snd_ssthresh + maxseg * (tp->t_dupacks - tp->snd_limited); if (SEQ_GT(onxt, tp->snd_nxt)) tp->snd_nxt = onxt; goto drop; } else if (V_tcp_do_rfc3042) { /* * Process first and second duplicate * ACKs. Each indicates a segment * leaving the network, creating room * for more. Make sure we can send a * packet on reception of each duplicate * ACK by increasing snd_cwnd by one * segment. Restore the original * snd_cwnd after packet transmission. */ cc_ack_received(tp, th, nsegs, CC_DUPACK); uint32_t oldcwnd = tp->snd_cwnd; tcp_seq oldsndmax = tp->snd_max; u_int sent; int avail; KASSERT(tp->t_dupacks == 1 || tp->t_dupacks == 2, ("%s: dupacks not 1 or 2", __func__)); if (tp->t_dupacks == 1) tp->snd_limited = 0; tp->snd_cwnd = (tp->snd_nxt - tp->snd_una) + (tp->t_dupacks - tp->snd_limited) * maxseg; /* * Only call tcp_output when there * is new data available to be sent * or we need to send an ACK. */ SOCKBUF_LOCK(&so->so_snd); avail = sbavail(&so->so_snd) - (tp->snd_nxt - tp->snd_una); SOCKBUF_UNLOCK(&so->so_snd); if (avail > 0 || tp->t_flags & TF_ACKNOW) (void) tcp_output(tp); sent = tp->snd_max - oldsndmax; if (sent > maxseg) { KASSERT((tp->t_dupacks == 2 && tp->snd_limited == 0) || (sent == maxseg + 1 && tp->t_flags & TF_SENTFIN), ("%s: sent too much", __func__)); tp->snd_limited = 2; } else if (sent > 0) ++tp->snd_limited; tp->snd_cwnd = oldcwnd; goto drop; } } break; } else { /* * This ack is advancing the left edge, reset the * counter. */ tp->t_dupacks = 0; /* * If this ack also has new SACK info, increment the * counter as per rfc6675. The variable * sack_changed tracks all changes to the SACK * scoreboard, including when partial ACKs without * SACK options are received, and clear the scoreboard * from the left side. Such partial ACKs should not be * counted as dupacks here. */ if (tcp_is_sack_recovery(tp, &to) && - sack_changed) { + (sack_changed != SACK_NOCHANGE)) { tp->t_dupacks++; /* limit overhead by setting maxseg last */ if (!IN_FASTRECOVERY(tp->t_flags) && (tp->sackhint.sacked_bytes > ((tcprexmtthresh - 1) * (maxseg = tcp_maxseg(tp))))) { goto enter_recovery; } } } resume_partialack: KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("%s: th_ack <= snd_una", __func__)); /* * If the congestion window was inflated to account * for the other side's cached packets, retract it. */ if (IN_FASTRECOVERY(tp->t_flags)) { if (SEQ_LT(th->th_ack, tp->snd_recover)) { if (tp->t_flags & TF_SACK_PERMIT) if (V_tcp_do_prr && to.to_flags & TOF_SACK) { tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; - tcp_do_prr_ack(tp, th, &to); + tcp_do_prr_ack(tp, th, &to, sack_changed); tp->t_flags |= TF_ACKNOW; (void) tcp_output(tp); } else tcp_sack_partialack(tp, th); else tcp_newreno_partial_ack(tp, th); } else cc_post_recovery(tp, th); } else if (IN_CONGRECOVERY(tp->t_flags)) { if (SEQ_LT(th->th_ack, tp->snd_recover)) { if (V_tcp_do_prr) { tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); tp->snd_fack = th->th_ack; - tcp_do_prr_ack(tp, th, &to); + /* + * During ECN cwnd reduction + * always use PRR-SSRB + */ + tcp_do_prr_ack(tp, th, &to, SACK_CHANGE); (void) tcp_output(tp); } } else cc_post_recovery(tp, th); } /* * If we reach this point, ACK is not a duplicate, * i.e., it ACKs something we sent. */ if (tp->t_flags & TF_NEEDSYN) { /* * T/TCP: Connection was half-synchronized, and our * SYN has been ACK'd (so connection is now fully * synchronized). Go to non-starred state, * increment snd_una for ACK of SYN, and check if * we can do window scaling. */ tp->t_flags &= ~TF_NEEDSYN; tp->snd_una++; /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; /* Send window already scaled. */ } } process_ACK: INP_WLOCK_ASSERT(inp); /* * Adjust for the SYN bit in sequence space, * but don't account for it in cwnd calculations. * This is for the SYN_RECEIVED, non-simultaneous * SYN case. SYN_SENT and simultaneous SYN are * treated elsewhere. */ if (incforsyn) tp->snd_una++; acked = BYTES_THIS_ACK(tp, th); KASSERT(acked >= 0, ("%s: acked unexepectedly negative " "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, tp->snd_una, th->th_ack, tp, m)); TCPSTAT_ADD(tcps_rcvackpack, nsegs); TCPSTAT_ADD(tcps_rcvackbyte, acked); /* * If we just performed our first retransmit, and the ACK * arrives within our recovery window, then it was a mistake * to do the retransmit in the first place. Recover our * original cwnd and ssthresh, and proceed to transmit where * we left off. */ if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && tp->t_badrxtwin != 0 && to.to_flags & TOF_TS && to.to_tsecr != 0 && TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) cc_cong_signal(tp, th, CC_RTO_ERR); /* * If we have a timestamp reply, update smoothed * round trip time. If no timestamp is present but * transmit timer is running and timed sequence * number was acked, update smoothed round trip time. * Since we now have an rtt measurement, cancel the * timer backoff (cf., Phil Karn's retransmit alg.). * Recompute the initial retransmit timer. * * Some boxes send broken timestamp replies * during the SYN+ACK phase, ignore * timestamps of 0 or we could calculate a * huge RTT and blow up the retransmit timer. */ if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { uint32_t t; t = tcp_ts_getticks() - to.to_tsecr; if (!tp->t_rttlow || tp->t_rttlow > t) tp->t_rttlow = t; tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) tp->t_rttlow = ticks - tp->t_rtttime; tcp_xmit_timer(tp, ticks - tp->t_rtttime); } SOCKBUF_LOCK(&so->so_snd); /* * Clear t_acktime if remote side has ACKd all data in the * socket buffer and FIN (if applicable). * Otherwise, update t_acktime if we received a sufficiently * large ACK. */ if ((tp->t_state <= TCPS_CLOSE_WAIT && acked == sbavail(&so->so_snd)) || acked > sbavail(&so->so_snd)) tp->t_acktime = 0; else if (acked > 1) tp->t_acktime = ticks; /* * If all outstanding data is acked, stop retransmit * timer and remember to restart (more output or persist). * If there is more data to be acked, restart retransmit * timer, using current (possibly backed-off) value. */ if (th->th_ack == tp->snd_max) { tcp_timer_activate(tp, TT_REXMT, 0); needoutput = 1; } else if (!tcp_timer_active(tp, TT_PERSIST)) tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); /* * If no data (only SYN) was ACK'd, * skip rest of ACK processing. */ if (acked == 0) { SOCKBUF_UNLOCK(&so->so_snd); goto step6; } /* * Let the congestion control algorithm update congestion * control related information. This typically means increasing * the congestion window. */ cc_ack_received(tp, th, nsegs, CC_ACK); if (acked > sbavail(&so->so_snd)) { if (tp->snd_wnd >= sbavail(&so->so_snd)) tp->snd_wnd -= sbavail(&so->so_snd); else tp->snd_wnd = 0; mfree = sbcut_locked(&so->so_snd, (int)sbavail(&so->so_snd)); ourfinisacked = 1; } else { mfree = sbcut_locked(&so->so_snd, acked); if (tp->snd_wnd >= (uint32_t) acked) tp->snd_wnd -= acked; else tp->snd_wnd = 0; ourfinisacked = 0; } /* NB: sowwakeup_locked() does an implicit unlock. */ sowwakeup_locked(so); m_freem(mfree); /* Detect una wraparound. */ if (!IN_RECOVERY(tp->t_flags) && SEQ_GT(tp->snd_una, tp->snd_recover) && SEQ_LEQ(th->th_ack, tp->snd_recover)) tp->snd_recover = th->th_ack - 1; /* XXXLAS: Can this be moved up into cc_post_recovery? */ if (IN_RECOVERY(tp->t_flags) && SEQ_GEQ(th->th_ack, tp->snd_recover)) { EXIT_RECOVERY(tp->t_flags); } tp->snd_una = th->th_ack; if (tp->t_flags & TF_SACK_PERMIT) { if (SEQ_GT(tp->snd_una, tp->snd_recover)) tp->snd_recover = tp->snd_una; } if (SEQ_LT(tp->snd_nxt, tp->snd_una)) tp->snd_nxt = tp->snd_una; switch (tp->t_state) { /* * In FIN_WAIT_1 STATE in addition to the processing * for the ESTABLISHED state if our FIN is now acknowledged * then enter FIN_WAIT_2. */ case TCPS_FIN_WAIT_1: if (ourfinisacked) { /* * If we can't receive any more * data, then closing user can proceed. * Starting the timer is contrary to the * specification, but if we don't get a FIN * we'll hang forever. */ if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { soisdisconnected(so); tcp_timer_activate(tp, TT_2MSL, (tcp_fast_finwait2_recycle ? tcp_finwait2_timeout : TP_MAXIDLE(tp))); } tcp_state_change(tp, TCPS_FIN_WAIT_2); } break; /* * In CLOSING STATE in addition to the processing for * the ESTABLISHED state if the ACK acknowledges our FIN * then enter the TIME-WAIT state, otherwise ignore * the segment. */ case TCPS_CLOSING: if (ourfinisacked) { tcp_twstart(tp); m_freem(m); return; } break; /* * In LAST_ACK, we may still be waiting for data to drain * and/or to be acked, as well as for the ack of our FIN. * If our FIN is now acknowledged, delete the TCB, * enter the closed state and return. */ case TCPS_LAST_ACK: if (ourfinisacked) { tp = tcp_close(tp); goto drop; } break; } } step6: INP_WLOCK_ASSERT(inp); /* * Update window information. * Don't look at window if no ACK: TAC's send garbage on first SYN. */ if ((thflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { /* keep track of pure window updates */ if (tlen == 0 && tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) TCPSTAT_INC(tcps_rcvwinupd); tp->snd_wnd = tiwin; tp->snd_wl1 = th->th_seq; tp->snd_wl2 = th->th_ack; if (tp->snd_wnd > tp->max_sndwnd) tp->max_sndwnd = tp->snd_wnd; needoutput = 1; } /* * Process segments with URG. */ if ((thflags & TH_URG) && th->th_urp && TCPS_HAVERCVDFIN(tp->t_state) == 0) { /* * This is a kludge, but if we receive and accept * random urgent pointers, we'll crash in * soreceive. It's hard to imagine someone * actually wanting to send this much urgent data. */ SOCKBUF_LOCK(&so->so_rcv); if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { th->th_urp = 0; /* XXX */ thflags &= ~TH_URG; /* XXX */ SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ goto dodata; /* XXX */ } /* * If this segment advances the known urgent pointer, * then mark the data stream. This should not happen * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since * a FIN has been received from the remote side. * In these states we ignore the URG. * * According to RFC961 (Assigned Protocols), * the urgent pointer points to the last octet * of urgent data. We continue, however, * to consider it to indicate the first octet * of data past the urgent section as the original * spec states (in one of two places). */ if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { tp->rcv_up = th->th_seq + th->th_urp; so->so_oobmark = sbavail(&so->so_rcv) + (tp->rcv_up - tp->rcv_nxt) - 1; if (so->so_oobmark == 0) so->so_rcv.sb_state |= SBS_RCVATMARK; sohasoutofband(so); tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); } SOCKBUF_UNLOCK(&so->so_rcv); /* * Remove out of band data so doesn't get presented to user. * This can happen independent of advancing the URG pointer, * but if two URG's are pending at once, some out-of-band * data may creep in... ick. */ if (th->th_urp <= (uint32_t)tlen && !(so->so_options & SO_OOBINLINE)) { /* hdr drop is delayed */ tcp_pulloutofband(so, th, m, drop_hdrlen); } } else { /* * If no out of band data is expected, * pull receive urgent pointer along * with the receive window. */ if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) tp->rcv_up = tp->rcv_nxt; } dodata: /* XXX */ INP_WLOCK_ASSERT(inp); /* * Process the segment text, merging it into the TCP sequencing queue, * and arranging for acknowledgment of receipt if necessary. * This process logically involves adjusting tp->rcv_wnd as data * is presented to the user (this happens in tcp_usrreq.c, * case PRU_RCVD). If a FIN has already been received on this * connection then we just ignore the text. */ tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && IS_FASTOPEN(tp->t_flags)); if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && TCPS_HAVERCVDFIN(tp->t_state) == 0) { tcp_seq save_start = th->th_seq; tcp_seq save_rnxt = tp->rcv_nxt; int save_tlen = tlen; m_adj(m, drop_hdrlen); /* delayed header drop */ /* * Insert segment which includes th into TCP reassembly queue * with control block tp. Set thflags to whether reassembly now * includes a segment with FIN. This handles the common case * inline (segment is the next to be received on an established * connection, and the queue is empty), avoiding linkage into * and removal from the queue and repetition of various * conversions. * Set DELACK for segments received in order, but ack * immediately when segments are out of order (so * fast retransmit can work). */ if (th->th_seq == tp->rcv_nxt && SEGQ_EMPTY(tp) && (TCPS_HAVEESTABLISHED(tp->t_state) || tfo_syn)) { if (DELAY_ACK(tp, tlen) || tfo_syn) tp->t_flags |= TF_DELACK; else tp->t_flags |= TF_ACKNOW; tp->rcv_nxt += tlen; if (tlen && ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && (tp->t_fbyte_in == 0)) { tp->t_fbyte_in = ticks; if (tp->t_fbyte_in == 0) tp->t_fbyte_in = 1; if (tp->t_fbyte_out && tp->t_fbyte_in) tp->t_flags2 |= TF2_FBYTES_COMPLETE; } thflags = tcp_get_flags(th) & TH_FIN; TCPSTAT_INC(tcps_rcvpack); TCPSTAT_ADD(tcps_rcvbyte, tlen); SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) m_freem(m); else sbappendstream_locked(&so->so_rcv, m, 0); tp->t_flags |= TF_WAKESOR; } else { /* * XXX: Due to the header drop above "th" is * theoretically invalid by now. Fortunately * m_adj() doesn't actually frees any mbufs * when trimming from the head. */ tcp_seq temp = save_start; thflags = tcp_reass(tp, th, &temp, &tlen, m); tp->t_flags |= TF_ACKNOW; } if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0) && TCPS_HAVEESTABLISHED(tp->t_state)) { if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { /* * DSACK actually handled in the fastpath * above. */ tcp_update_sack_list(tp, save_start, save_start + save_tlen); } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { if ((tp->rcv_numsacks >= 1) && (tp->sackblks[0].end == save_start)) { /* * Partial overlap, recorded at todrop * above. */ tcp_update_sack_list(tp, tp->sackblks[0].start, tp->sackblks[0].end); } else { tcp_update_dsack_list(tp, save_start, save_start + save_tlen); } } else if (tlen >= save_tlen) { /* Update of sackblks. */ tcp_update_dsack_list(tp, save_start, save_start + save_tlen); } else if (tlen > 0) { tcp_update_dsack_list(tp, save_start, save_start + tlen); } } tcp_handle_wakeup(tp); #if 0 /* * Note the amount of data that peer has sent into * our window, in order to estimate the sender's * buffer size. * XXX: Unused. */ if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); else len = so->so_rcv.sb_hiwat; #endif } else { m_freem(m); thflags &= ~TH_FIN; } /* * If FIN is received ACK the FIN and let the user know * that the connection is closing. */ if (thflags & TH_FIN) { if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { /* The socket upcall is handled by socantrcvmore. */ socantrcvmore(so); /* * If connection is half-synchronized * (ie NEEDSYN flag on) then delay ACK, * so it may be piggybacked when SYN is sent. * Otherwise, since we received a FIN then no * more input can be expected, send ACK now. */ if (tp->t_flags & TF_NEEDSYN) tp->t_flags |= TF_DELACK; else tp->t_flags |= TF_ACKNOW; tp->rcv_nxt++; } switch (tp->t_state) { /* * In SYN_RECEIVED and ESTABLISHED STATES * enter the CLOSE_WAIT state. */ case TCPS_SYN_RECEIVED: tp->t_starttime = ticks; /* FALLTHROUGH */ case TCPS_ESTABLISHED: tcp_state_change(tp, TCPS_CLOSE_WAIT); break; /* * If still in FIN_WAIT_1 STATE FIN has not been acked so * enter the CLOSING state. */ case TCPS_FIN_WAIT_1: tcp_state_change(tp, TCPS_CLOSING); break; /* * In FIN_WAIT_2 state enter the TIME_WAIT state, * starting the time-wait timer, turning off the other * standard timers. */ case TCPS_FIN_WAIT_2: tcp_twstart(tp); return; } } TCP_PROBE3(debug__input, tp, th, m); /* * Return any desired output. */ if (needoutput || (tp->t_flags & TF_ACKNOW)) (void) tcp_output(tp); check_delack: INP_WLOCK_ASSERT(inp); if (tp->t_flags & TF_DELACK) { tp->t_flags &= ~TF_DELACK; tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); } INP_WUNLOCK(inp); return; dropafterack: /* * Generate an ACK dropping incoming segment if it occupies * sequence space, where the ACK reflects our state. * * We can now skip the test for the RST flag since all * paths to this code happen after packets containing * RST have been dropped. * * In the SYN-RECEIVED state, don't send an ACK unless the * segment we received passes the SYN-RECEIVED ACK test. * If it fails send a RST. This breaks the loop in the * "LAND" DoS attack, and also prevents an ACK storm * between two listening ports that have been sent forged * SYN segments, each with the source address of the other. */ if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && (SEQ_GT(tp->snd_una, th->th_ack) || SEQ_GT(th->th_ack, tp->snd_max)) ) { rstreason = BANDLIM_RST_OPENPORT; tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); goto dropwithreset; } TCP_PROBE3(debug__input, tp, th, m); tp->t_flags |= TF_ACKNOW; (void) tcp_output(tp); INP_WUNLOCK(inp); m_freem(m); return; dropwithreset: if (tp != NULL) { tcp_dropwithreset(m, th, tp, tlen, rstreason); INP_WUNLOCK(inp); } else tcp_dropwithreset(m, th, NULL, tlen, rstreason); return; drop: /* * Drop space held by incoming segment and return. */ TCP_PROBE3(debug__input, tp, th, m); if (tp != NULL) { INP_WUNLOCK(inp); } m_freem(m); } /* * Issue RST and make ACK acceptable to originator of segment. * The mbuf must still include the original packet header. * tp may be NULL. */ void tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, int tlen, int rstreason) { #ifdef INET struct ip *ip; #endif #ifdef INET6 struct ip6_hdr *ip6; #endif if (tp != NULL) { INP_LOCK_ASSERT(tptoinpcb(tp)); } /* Don't bother if destination was broadcast/multicast. */ if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) goto drop; #ifdef INET6 if (mtod(m, struct ip *)->ip_v == 6) { ip6 = mtod(m, struct ip6_hdr *); if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) goto drop; /* IPv6 anycast check is done at tcp6_input() */ } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { ip = mtod(m, struct ip *); if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) goto drop; } #endif /* Perform bandwidth limiting. */ if (badport_bandlim(rstreason) < 0) goto drop; /* tcp_respond consumes the mbuf chain. */ if (tcp_get_flags(th) & TH_ACK) { tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, TH_RST); } else { if (tcp_get_flags(th) & TH_SYN) tlen++; if (tcp_get_flags(th) & TH_FIN) tlen++; tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, (tcp_seq)0, TH_RST|TH_ACK); } return; drop: m_freem(m); } /* * Parse TCP options and place in tcpopt. */ void tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) { int opt, optlen; to->to_flags = 0; for (; cnt > 0; cnt -= optlen, cp += optlen) { opt = cp[0]; if (opt == TCPOPT_EOL) break; if (opt == TCPOPT_NOP) optlen = 1; else { if (cnt < 2) break; optlen = cp[1]; if (optlen < 2 || optlen > cnt) break; } switch (opt) { case TCPOPT_MAXSEG: if (optlen != TCPOLEN_MAXSEG) continue; if (!(flags & TO_SYN)) continue; to->to_flags |= TOF_MSS; bcopy((char *)cp + 2, (char *)&to->to_mss, sizeof(to->to_mss)); to->to_mss = ntohs(to->to_mss); break; case TCPOPT_WINDOW: if (optlen != TCPOLEN_WINDOW) continue; if (!(flags & TO_SYN)) continue; to->to_flags |= TOF_SCALE; to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); break; case TCPOPT_TIMESTAMP: if (optlen != TCPOLEN_TIMESTAMP) continue; to->to_flags |= TOF_TS; bcopy((char *)cp + 2, (char *)&to->to_tsval, sizeof(to->to_tsval)); to->to_tsval = ntohl(to->to_tsval); bcopy((char *)cp + 6, (char *)&to->to_tsecr, sizeof(to->to_tsecr)); to->to_tsecr = ntohl(to->to_tsecr); break; case TCPOPT_SIGNATURE: /* * In order to reply to a host which has set the * TCP_SIGNATURE option in its initial SYN, we have * to record the fact that the option was observed * here for the syncache code to perform the correct * response. */ if (optlen != TCPOLEN_SIGNATURE) continue; to->to_flags |= TOF_SIGNATURE; to->to_signature = cp + 2; break; case TCPOPT_SACK_PERMITTED: if (optlen != TCPOLEN_SACK_PERMITTED) continue; if (!(flags & TO_SYN)) continue; if (!V_tcp_do_sack) continue; to->to_flags |= TOF_SACKPERM; break; case TCPOPT_SACK: if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) continue; if (flags & TO_SYN) continue; to->to_flags |= TOF_SACK; to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; to->to_sacks = cp + 2; TCPSTAT_INC(tcps_sack_rcv_blocks); break; case TCPOPT_FAST_OPEN: /* * Cookie length validation is performed by the * server side cookie checking code or the client * side cookie cache update code. */ if (!(flags & TO_SYN)) continue; if (!V_tcp_fastopen_client_enable && !V_tcp_fastopen_server_enable) continue; to->to_flags |= TOF_FASTOPEN; to->to_tfo_len = optlen - 2; to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; break; default: continue; } } } /* * Pull out of band byte out of a segment so * it doesn't appear in the user's data queue. * It is still reflected in the segment length for * sequencing purposes. */ void tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off) { int cnt = off + th->th_urp - 1; while (cnt >= 0) { if (m->m_len > cnt) { char *cp = mtod(m, caddr_t) + cnt; struct tcpcb *tp = sototcpcb(so); INP_WLOCK_ASSERT(tptoinpcb(tp)); tp->t_iobc = *cp; tp->t_oobflags |= TCPOOB_HAVEDATA; bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); m->m_len--; if (m->m_flags & M_PKTHDR) m->m_pkthdr.len--; return; } cnt -= m->m_len; m = m->m_next; if (m == NULL) break; } panic("tcp_pulloutofband"); } /* * Collect new round-trip time estimate * and update averages and current timeout. */ void tcp_xmit_timer(struct tcpcb *tp, int rtt) { int delta; INP_WLOCK_ASSERT(tptoinpcb(tp)); TCPSTAT_INC(tcps_rttupdated); if (tp->t_rttupdated < UCHAR_MAX) tp->t_rttupdated++; #ifdef STATS stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, imax(0, rtt * 1000 / hz)); #endif if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { /* * srtt is stored as fixed point with 5 bits after the * binary point (i.e., scaled by 8). The following magic * is equivalent to the smoothing algorithm in rfc793 with * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed * point). Adjust rtt to origin 0. */ delta = ((rtt - 1) << TCP_DELTA_SHIFT) - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); if ((tp->t_srtt += delta) <= 0) tp->t_srtt = 1; /* * We accumulate a smoothed rtt variance (actually, a * smoothed mean difference), then set the retransmit * timer to smoothed rtt + 4 times the smoothed variance. * rttvar is stored as fixed point with 4 bits after the * binary point (scaled by 16). The following is * equivalent to rfc793 smoothing with an alpha of .75 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces * rfc793's wired-in beta. */ if (delta < 0) delta = -delta; delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); if ((tp->t_rttvar += delta) <= 0) tp->t_rttvar = 1; } else { /* * No rtt measurement yet - use the unsmoothed rtt. * Set the variance to half the rtt (so our first * retransmit happens at 3*rtt). */ tp->t_srtt = rtt << TCP_RTT_SHIFT; tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); } tp->t_rtttime = 0; tp->t_rxtshift = 0; /* * the retransmit should happen at rtt + 4 * rttvar. * Because of the way we do the smoothing, srtt and rttvar * will each average +1/2 tick of bias. When we compute * the retransmit timer, we want 1/2 tick of rounding and * 1 extra tick because of +-1/2 tick uncertainty in the * firing of the timer. The bias will give us exactly the * 1.5 tick we need. But, because the bias is * statistical, we have to test that we don't drop below * the minimum feasible timer (which is 2 ticks). */ TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); /* * We received an ack for a packet that wasn't retransmitted; * it is probably safe to discard any error indications we've * received recently. This isn't quite right, but close enough * for now (a route might have failed after we sent a segment, * and the return path might not be symmetrical). */ tp->t_softerror = 0; } /* * Determine a reasonable value for maxseg size. * If the route is known, check route for mtu. * If none, use an mss that can be handled on the outgoing interface * without forcing IP to fragment. If no route is found, route has no mtu, * or the destination isn't local, use a default, hopefully conservative * size (usually 512 or the default IP max size, but no more than the mtu * of the interface), as we can't discover anything about intervening * gateways or networks. We also initialize the congestion/slow start * window to be a single segment if the destination isn't local. * While looking at the routing entry, we also initialize other path-dependent * parameters from pre-set or cached values in the routing entry. * * NOTE that resulting t_maxseg doesn't include space for TCP options or * IP options, e.g. IPSEC data, since length of this data may vary, and * thus it is calculated for every segment separately in tcp_output(). * * NOTE that this routine is only called when we process an incoming * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS * settings are handled in tcp_mssopt(). */ void tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) { int mss = 0; uint32_t maxmtu = 0; struct inpcb *inp = tptoinpcb(tp); struct hc_metrics_lite metrics; #ifdef INET6 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; size_t min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : sizeof (struct tcpiphdr); #else size_t min_protoh = sizeof(struct tcpiphdr); #endif INP_WLOCK_ASSERT(inp); if (tp->t_port) min_protoh += V_tcp_udp_tunneling_overhead; if (mtuoffer != -1) { KASSERT(offer == -1, ("%s: conflict", __func__)); offer = mtuoffer - min_protoh; } /* Initialize. */ #ifdef INET6 if (isipv6) { maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); tp->t_maxseg = V_tcp_v6mssdflt; } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { maxmtu = tcp_maxmtu(&inp->inp_inc, cap); tp->t_maxseg = V_tcp_mssdflt; } #endif /* * No route to sender, stay with default mss and return. */ if (maxmtu == 0) { /* * In case we return early we need to initialize metrics * to a defined state as tcp_hc_get() would do for us * if there was no cache hit. */ if (metricptr != NULL) bzero(metricptr, sizeof(struct hc_metrics_lite)); return; } /* What have we got? */ switch (offer) { case 0: /* * Offer == 0 means that there was no MSS on the SYN * segment, in this case we use tcp_mssdflt as * already assigned to t_maxseg above. */ offer = tp->t_maxseg; break; case -1: /* * Offer == -1 means that we didn't receive SYN yet. */ /* FALLTHROUGH */ default: /* * Prevent DoS attack with too small MSS. Round up * to at least minmss. */ offer = max(offer, V_tcp_minmss); } /* * rmx information is now retrieved from tcp_hostcache. */ tcp_hc_get(&inp->inp_inc, &metrics); if (metricptr != NULL) bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); /* * If there's a discovered mtu in tcp hostcache, use it. * Else, use the link mtu. */ if (metrics.rmx_mtu) mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; else { #ifdef INET6 if (isipv6) { mss = maxmtu - min_protoh; if (!V_path_mtu_discovery && !in6_localaddr(&inp->in6p_faddr)) mss = min(mss, V_tcp_v6mssdflt); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { mss = maxmtu - min_protoh; if (!V_path_mtu_discovery && !in_localaddr(inp->inp_faddr)) mss = min(mss, V_tcp_mssdflt); } #endif /* * XXX - The above conditional (mss = maxmtu - min_protoh) * probably violates the TCP spec. * The problem is that, since we don't know the * other end's MSS, we are supposed to use a conservative * default. But, if we do that, then MTU discovery will * never actually take place, because the conservative * default is much less than the MTUs typically seen * on the Internet today. For the moment, we'll sweep * this under the carpet. * * The conservative default might not actually be a problem * if the only case this occurs is when sending an initial * SYN with options and data to a host we've never talked * to before. Then, they will reply with an MSS value which * will get recorded and the new parameters should get * recomputed. For Further Study. */ } mss = min(mss, offer); /* * Sanity check: make sure that maxseg will be large * enough to allow some data on segments even if the * all the option space is used (40bytes). Otherwise * funny things may happen in tcp_output. * * XXXGL: shouldn't we reserve space for IP/IPv6 options? */ mss = max(mss, 64); tp->t_maxseg = mss; } void tcp_mss(struct tcpcb *tp, int offer) { int mss; uint32_t bufsize; struct inpcb *inp = tptoinpcb(tp); struct socket *so; struct hc_metrics_lite metrics; struct tcp_ifcap cap; KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); bzero(&cap, sizeof(cap)); tcp_mss_update(tp, offer, -1, &metrics, &cap); mss = tp->t_maxseg; /* * If there's a pipesize, change the socket buffer to that size, * don't change if sb_hiwat is different than default (then it * has been changed on purpose with setsockopt). * Make the socket buffers an integral number of mss units; * if the mss is larger than the socket buffer, decrease the mss. */ so = inp->inp_socket; SOCKBUF_LOCK(&so->so_snd); if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) bufsize = metrics.rmx_sendpipe; else bufsize = so->so_snd.sb_hiwat; if (bufsize < mss) mss = bufsize; else { bufsize = roundup(bufsize, mss); if (bufsize > sb_max) bufsize = sb_max; if (bufsize > so->so_snd.sb_hiwat) (void)sbreserve_locked(so, SO_SND, bufsize, NULL); } SOCKBUF_UNLOCK(&so->so_snd); /* * Sanity check: make sure that maxseg will be large * enough to allow some data on segments even if the * all the option space is used (40bytes). Otherwise * funny things may happen in tcp_output. * * XXXGL: shouldn't we reserve space for IP/IPv6 options? */ tp->t_maxseg = max(mss, 64); SOCKBUF_LOCK(&so->so_rcv); if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) bufsize = metrics.rmx_recvpipe; else bufsize = so->so_rcv.sb_hiwat; if (bufsize > mss) { bufsize = roundup(bufsize, mss); if (bufsize > sb_max) bufsize = sb_max; if (bufsize > so->so_rcv.sb_hiwat) (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); } SOCKBUF_UNLOCK(&so->so_rcv); /* Check the interface for TSO capabilities. */ if (cap.ifcap & CSUM_TSO) { tp->t_flags |= TF_TSO; tp->t_tsomax = cap.tsomax; tp->t_tsomaxsegcount = cap.tsomaxsegcount; tp->t_tsomaxsegsize = cap.tsomaxsegsize; } } /* * Determine the MSS option to send on an outgoing SYN. */ int tcp_mssopt(struct in_conninfo *inc) { int mss = 0; uint32_t thcmtu = 0; uint32_t maxmtu = 0; size_t min_protoh; KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); #ifdef INET6 if (inc->inc_flags & INC_ISIPV6) { mss = V_tcp_v6mssdflt; maxmtu = tcp_maxmtu6(inc, NULL); min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { mss = V_tcp_mssdflt; maxmtu = tcp_maxmtu(inc, NULL); min_protoh = sizeof(struct tcpiphdr); } #endif #if defined(INET6) || defined(INET) thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ #endif if (maxmtu && thcmtu) mss = min(maxmtu, thcmtu) - min_protoh; else if (maxmtu || thcmtu) mss = max(maxmtu, thcmtu) - min_protoh; return (mss); } void -tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) +tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, sackstatus_t sack_changed) { int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; int maxseg = tcp_maxseg(tp); INP_WLOCK_ASSERT(tptoinpcb(tp)); /* * Compute the amount of data that this ACK is indicating * (del_data) and an estimate of how many bytes are in the * network. */ if (tcp_is_sack_recovery(tp, to) || (IN_CONGRECOVERY(tp->t_flags) && !IN_FASTRECOVERY(tp->t_flags))) { del_data = tp->sackhint.delivered_data; if (V_tcp_do_newsack) pipe = tcp_compute_pipe(tp); else pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; } else { if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + tp->snd_recover - tp->snd_una)) del_data = maxseg; pipe = imax(0, tp->snd_max - tp->snd_una - imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); } tp->sackhint.prr_delivered += del_data; /* * Proportional Rate Reduction */ if (pipe >= tp->snd_ssthresh) { if (tp->sackhint.recover_fs == 0) tp->sackhint.recover_fs = imax(1, tp->snd_nxt - tp->snd_una); snd_cnt = howmany((long)tp->sackhint.prr_delivered * tp->snd_ssthresh, tp->sackhint.recover_fs) - tp->sackhint.prr_out; } else { - if (V_tcp_do_prr_conservative || (del_data == 0)) + /* + * PRR 6937bis heuristic: + * - A partial ack without SACK block beneath snd_recover + * indicates further loss. + * - An SACK scoreboard update adding a new hole indicates + * further loss, so be conservative and send at most one + * segment. + * - Prevent ACK splitting attacks, by being conservative + * when no new data is acked. + */ + if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) limit = tp->sackhint.prr_delivered - tp->sackhint.prr_out; else limit = imax(tp->sackhint.prr_delivered - tp->sackhint.prr_out, del_data) + maxseg; snd_cnt = imin((tp->snd_ssthresh - pipe), limit); } snd_cnt = imax(snd_cnt, 0) / maxseg; /* * Send snd_cnt new data into the network in response to this ack. * If there is going to be a SACK retransmission, adjust snd_cwnd * accordingly. */ if (IN_FASTRECOVERY(tp->t_flags)) { if (tcp_is_sack_recovery(tp, to)) { tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg); } else { tp->snd_cwnd = (tp->snd_max - tp->snd_una) + (snd_cnt * maxseg); } } else if (IN_CONGRECOVERY(tp->t_flags)) tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); } /* * On a partial ack arrives, force the retransmission of the * next unacknowledged segment. Do not clear tp->t_dupacks. * By setting snd_nxt to ti_ack, this forces retransmission timer to * be started again. */ void tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) { tcp_seq onxt = tp->snd_nxt; uint32_t ocwnd = tp->snd_cwnd; u_int maxseg = tcp_maxseg(tp); INP_WLOCK_ASSERT(tptoinpcb(tp)); tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; tp->snd_nxt = th->th_ack; /* * Set snd_cwnd to one segment beyond acknowledged offset. * (tp->snd_una has not yet been updated when this function is called.) */ tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); tp->t_flags |= TF_ACKNOW; (void) tcp_output(tp); tp->snd_cwnd = ocwnd; if (SEQ_GT(onxt, tp->snd_nxt)) tp->snd_nxt = onxt; /* * Partial window deflation. Relies on fact that tp->snd_una * not updated yet. */ if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); else tp->snd_cwnd = 0; tp->snd_cwnd += maxseg; } int tcp_compute_pipe(struct tcpcb *tp) { if (tp->t_fb->tfb_compute_pipe == NULL) { return (tp->snd_max - tp->snd_una + tp->sackhint.sack_bytes_rexmit - tp->sackhint.sacked_bytes - tp->sackhint.lost_bytes); } else { return((*tp->t_fb->tfb_compute_pipe)(tp)); } } uint32_t tcp_compute_initwnd(uint32_t maxseg) { /* * Calculate the Initial Window, also used as Restart Window * * RFC5681 Section 3.1 specifies the default conservative values. * RFC3390 specifies slightly more aggressive values. * RFC6928 increases it to ten segments. * Support for user specified value for initial flight size. */ if (V_tcp_initcwnd_segments) return min(V_tcp_initcwnd_segments * maxseg, max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); else if (V_tcp_do_rfc3390) return min(4 * maxseg, max(2 * maxseg, 4380)); else { /* Per RFC5681 Section 3.1 */ if (maxseg > 2190) return (2 * maxseg); else if (maxseg > 1095) return (3 * maxseg); else return (4 * maxseg); } } diff --git a/sys/netinet/tcp_sack.c b/sys/netinet/tcp_sack.c index 589b0c424acb..891053c872dd 100644 --- a/sys/netinet/tcp_sack.c +++ b/sys/netinet/tcp_sack.c @@ -1,1135 +1,1143 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 * The Regents of the University of California. * 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. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. * * @(#)tcp_sack.c 8.12 (Berkeley) 5/24/95 */ /*- * @@(#)COPYRIGHT 1.1 (NRL) 17 January 1995 * * NRL grants permission for redistribution and use in source and binary * forms, with or without modification, of the software and documentation * created at NRL 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. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgements: * This product includes software developed by the University of * California, Berkeley and its contributors. * This product includes software developed at the Information * Technology Division, US Naval Research Laboratory. * 4. Neither the name of the NRL nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL 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 NRL 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. * * The views and conclusions contained in the software and documentation * are those of the authors and should not be interpreted as representing * official policies, either expressed or implied, of the US Naval * Research Laboratory (NRL). */ #include #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include /* for proc0 declaration */ #include #include #include #include #include #include /* before tcp_seq.h, for tcp_random18() */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include VNET_DECLARE(struct uma_zone *, sack_hole_zone); #define V_sack_hole_zone VNET(sack_hole_zone) SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "TCP SACK"); VNET_DEFINE(int, tcp_do_sack) = 1; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_sack), 0, "Enable/Disable TCP SACK support"); VNET_DEFINE(int, tcp_do_newsack) = 1; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, revised, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_newsack), 0, "Use revised SACK loss recovery per RFC 6675"); VNET_DEFINE(int, tcp_sack_maxholes) = 128; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, maxholes, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_sack_maxholes), 0, "Maximum number of TCP SACK holes allowed per connection"); VNET_DEFINE(int, tcp_sack_globalmaxholes) = 65536; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalmaxholes, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_sack_globalmaxholes), 0, "Global maximum number of TCP SACK holes"); VNET_DEFINE(int, tcp_sack_globalholes) = 0; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalholes, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcp_sack_globalholes), 0, "Global number of TCP SACK holes currently allocated"); int tcp_dsack_block_exists(struct tcpcb *tp) { /* Return true if a DSACK block exists */ if (tp->rcv_numsacks == 0) return (0); if (SEQ_LEQ(tp->sackblks[0].end, tp->rcv_nxt)) return(1); return (0); } /* * This function will find overlaps with the currently stored sackblocks * and add any overlap as a dsack block upfront */ void tcp_update_dsack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end) { struct sackblk head_blk,mid_blk,saved_blks[MAX_SACK_BLKS]; int i, j, n, identical; tcp_seq start, end; INP_WLOCK_ASSERT(tptoinpcb(tp)); KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end")); if (SEQ_LT(rcv_end, tp->rcv_nxt) || ((rcv_end == tp->rcv_nxt) && (tp->rcv_numsacks > 0 ) && (tp->sackblks[0].end == tp->rcv_nxt))) { saved_blks[0].start = rcv_start; saved_blks[0].end = rcv_end; } else { saved_blks[0].start = saved_blks[0].end = 0; } head_blk.start = head_blk.end = 0; mid_blk.start = rcv_start; mid_blk.end = rcv_end; identical = 0; for (i = 0; i < tp->rcv_numsacks; i++) { start = tp->sackblks[i].start; end = tp->sackblks[i].end; if (SEQ_LT(rcv_end, start)) { /* pkt left to sack blk */ continue; } if (SEQ_GT(rcv_start, end)) { /* pkt right to sack blk */ continue; } if (SEQ_GT(tp->rcv_nxt, end)) { if ((SEQ_MAX(rcv_start, start) != SEQ_MIN(rcv_end, end)) && (SEQ_GT(head_blk.start, SEQ_MAX(rcv_start, start)) || (head_blk.start == head_blk.end))) { head_blk.start = SEQ_MAX(rcv_start, start); head_blk.end = SEQ_MIN(rcv_end, end); } continue; } if (((head_blk.start == head_blk.end) || SEQ_LT(start, head_blk.start)) && (SEQ_GT(end, rcv_start) && SEQ_LEQ(start, rcv_end))) { head_blk.start = start; head_blk.end = end; } mid_blk.start = SEQ_MIN(mid_blk.start, start); mid_blk.end = SEQ_MAX(mid_blk.end, end); if ((mid_blk.start == start) && (mid_blk.end == end)) identical = 1; } if (SEQ_LT(head_blk.start, head_blk.end)) { /* store overlapping range */ saved_blks[0].start = SEQ_MAX(rcv_start, head_blk.start); saved_blks[0].end = SEQ_MIN(rcv_end, head_blk.end); } n = 1; /* * Second, if not ACKed, store the SACK block that * overlaps with the DSACK block unless it is identical */ if ((SEQ_LT(tp->rcv_nxt, mid_blk.end) && !((mid_blk.start == saved_blks[0].start) && (mid_blk.end == saved_blks[0].end))) || identical == 1) { saved_blks[n].start = mid_blk.start; saved_blks[n++].end = mid_blk.end; } for (j = 0; (j < tp->rcv_numsacks) && (n < MAX_SACK_BLKS); j++) { if (((SEQ_LT(tp->sackblks[j].end, mid_blk.start) || SEQ_GT(tp->sackblks[j].start, mid_blk.end)) && (SEQ_GT(tp->sackblks[j].start, tp->rcv_nxt)))) saved_blks[n++] = tp->sackblks[j]; } j = 0; for (i = 0; i < n; i++) { /* we can end up with a stale initial entry */ if (SEQ_LT(saved_blks[i].start, saved_blks[i].end)) { tp->sackblks[j++] = saved_blks[i]; } } tp->rcv_numsacks = j; } /* * This function is called upon receipt of new valid data (while not in * header prediction mode), and it updates the ordered list of sacks. */ void tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end) { /* * First reported block MUST be the most recent one. Subsequent * blocks SHOULD be in the order in which they arrived at the * receiver. These two conditions make the implementation fully * compliant with RFC 2018. */ struct sackblk head_blk, saved_blks[MAX_SACK_BLKS]; int num_head, num_saved, i; INP_WLOCK_ASSERT(tptoinpcb(tp)); /* Check arguments. */ KASSERT(SEQ_LEQ(rcv_start, rcv_end), ("rcv_start <= rcv_end")); if ((rcv_start == rcv_end) && (tp->rcv_numsacks >= 1) && (rcv_end == tp->sackblks[0].end)) { /* retaining DSACK block below rcv_nxt (todrop) */ head_blk = tp->sackblks[0]; } else { /* SACK block for the received segment. */ head_blk.start = rcv_start; head_blk.end = rcv_end; } /* * Merge updated SACK blocks into head_blk, and save unchanged SACK * blocks into saved_blks[]. num_saved will have the number of the * saved SACK blocks. */ num_saved = 0; for (i = 0; i < tp->rcv_numsacks; i++) { tcp_seq start = tp->sackblks[i].start; tcp_seq end = tp->sackblks[i].end; if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) { /* * Discard this SACK block. */ } else if (SEQ_LEQ(head_blk.start, end) && SEQ_GEQ(head_blk.end, start)) { /* * Merge this SACK block into head_blk. This SACK * block itself will be discarded. */ /* * |-| * |---| merge * * |-| * |---| merge * * |-----| * |-| DSACK smaller * * |-| * |-----| DSACK smaller */ if (head_blk.start == end) head_blk.start = start; else if (head_blk.end == start) head_blk.end = end; else { if (SEQ_LT(head_blk.start, start)) { tcp_seq temp = start; start = head_blk.start; head_blk.start = temp; } if (SEQ_GT(head_blk.end, end)) { tcp_seq temp = end; end = head_blk.end; head_blk.end = temp; } if ((head_blk.start != start) || (head_blk.end != end)) { if ((num_saved >= 1) && SEQ_GEQ(saved_blks[num_saved-1].start, start) && SEQ_LEQ(saved_blks[num_saved-1].end, end)) num_saved--; saved_blks[num_saved].start = start; saved_blks[num_saved].end = end; num_saved++; } } } else { /* * This block supercedes the prior block */ if ((num_saved >= 1) && SEQ_GEQ(saved_blks[num_saved-1].start, start) && SEQ_LEQ(saved_blks[num_saved-1].end, end)) num_saved--; /* * Save this SACK block. */ saved_blks[num_saved].start = start; saved_blks[num_saved].end = end; num_saved++; } } /* * Update SACK list in tp->sackblks[]. */ num_head = 0; if (SEQ_LT(rcv_start, rcv_end)) { /* * The received data segment is an out-of-order segment. Put * head_blk at the top of SACK list. */ tp->sackblks[0] = head_blk; num_head = 1; /* * If the number of saved SACK blocks exceeds its limit, * discard the last SACK block. */ if (num_saved >= MAX_SACK_BLKS) num_saved--; } if ((rcv_start == rcv_end) && (rcv_start == tp->sackblks[0].end)) { num_head = 1; } if (num_saved > 0) { /* * Copy the saved SACK blocks back. */ bcopy(saved_blks, &tp->sackblks[num_head], sizeof(struct sackblk) * num_saved); } /* Save the number of SACK blocks. */ tp->rcv_numsacks = num_head + num_saved; } void tcp_clean_dsack_blocks(struct tcpcb *tp) { struct sackblk saved_blks[MAX_SACK_BLKS]; int num_saved, i; INP_WLOCK_ASSERT(tptoinpcb(tp)); /* * Clean up any DSACK blocks that * are in our queue of sack blocks. * */ num_saved = 0; for (i = 0; i < tp->rcv_numsacks; i++) { tcp_seq start = tp->sackblks[i].start; tcp_seq end = tp->sackblks[i].end; if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) { /* * Discard this D-SACK block. */ continue; } /* * Save this SACK block. */ saved_blks[num_saved].start = start; saved_blks[num_saved].end = end; num_saved++; } if (num_saved > 0) { /* * Copy the saved SACK blocks back. */ bcopy(saved_blks, &tp->sackblks[0], sizeof(struct sackblk) * num_saved); } tp->rcv_numsacks = num_saved; } /* * Delete all receiver-side SACK information. */ void tcp_clean_sackreport(struct tcpcb *tp) { int i; INP_WLOCK_ASSERT(tptoinpcb(tp)); tp->rcv_numsacks = 0; for (i = 0; i < MAX_SACK_BLKS; i++) tp->sackblks[i].start = tp->sackblks[i].end=0; } /* * Allocate struct sackhole. */ static struct sackhole * tcp_sackhole_alloc(struct tcpcb *tp, tcp_seq start, tcp_seq end) { struct sackhole *hole; if (tp->snd_numholes >= V_tcp_sack_maxholes || V_tcp_sack_globalholes >= V_tcp_sack_globalmaxholes) { TCPSTAT_INC(tcps_sack_sboverflow); return NULL; } hole = (struct sackhole *)uma_zalloc(V_sack_hole_zone, M_NOWAIT); if (hole == NULL) return NULL; hole->start = start; hole->end = end; hole->rxmit = start; tp->snd_numholes++; atomic_add_int(&V_tcp_sack_globalholes, 1); return hole; } /* * Free struct sackhole. */ static void tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole) { uma_zfree(V_sack_hole_zone, hole); tp->snd_numholes--; atomic_subtract_int(&V_tcp_sack_globalholes, 1); KASSERT(tp->snd_numholes >= 0, ("tp->snd_numholes >= 0")); KASSERT(V_tcp_sack_globalholes >= 0, ("tcp_sack_globalholes >= 0")); } /* * Insert new SACK hole into scoreboard. */ static struct sackhole * tcp_sackhole_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end, struct sackhole *after) { struct sackhole *hole; /* Allocate a new SACK hole. */ hole = tcp_sackhole_alloc(tp, start, end); if (hole == NULL) return NULL; /* Insert the new SACK hole into scoreboard. */ if (after != NULL) TAILQ_INSERT_AFTER(&tp->snd_holes, after, hole, scblink); else TAILQ_INSERT_TAIL(&tp->snd_holes, hole, scblink); /* Update SACK hint. */ if (tp->sackhint.nexthole == NULL) tp->sackhint.nexthole = hole; return hole; } /* * Remove SACK hole from scoreboard. */ static void tcp_sackhole_remove(struct tcpcb *tp, struct sackhole *hole) { /* Update SACK hint. */ if (tp->sackhint.nexthole == hole) tp->sackhint.nexthole = TAILQ_NEXT(hole, scblink); /* Remove this SACK hole. */ TAILQ_REMOVE(&tp->snd_holes, hole, scblink); /* Free this SACK hole. */ tcp_sackhole_free(tp, hole); } /* * Process cumulative ACK and the TCP SACK option to update the scoreboard. * tp->snd_holes is an ordered list of holes (oldest to newest, in terms of * the sequence space). - * Returns 1 if incoming ACK has previously unknown SACK information, - * 0 otherwise. + * Returns SACK_NEWLOSS if incoming ACK indicates ongoing loss (hole split, new hole), + * SACK_CHANGE if incoming ACK has previously unknown SACK information, + * SACK_NOCHANGE otherwise. */ -int +sackstatus_t tcp_sack_doack(struct tcpcb *tp, struct tcpopt *to, tcp_seq th_ack) { struct sackhole *cur, *temp; struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1], *sblkp; - int i, j, num_sack_blks, sack_changed; + int i, j, num_sack_blks; + sackstatus_t sack_changed; int delivered_data, left_edge_delta; tcp_seq loss_hiack = 0; int loss_thresh = 0; int loss_sblks = 0; int notlost_bytes = 0; INP_WLOCK_ASSERT(tptoinpcb(tp)); num_sack_blks = 0; - sack_changed = 0; + sack_changed = SACK_NOCHANGE; delivered_data = 0; left_edge_delta = 0; /* * If SND.UNA will be advanced by SEG.ACK, and if SACK holes exist, * treat [SND.UNA, SEG.ACK) as if it is a SACK block. * Account changes to SND.UNA always in delivered data. */ if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) { left_edge_delta = th_ack - tp->snd_una; sack_blocks[num_sack_blks].start = tp->snd_una; sack_blocks[num_sack_blks++].end = th_ack; /* * Pulling snd_fack forward if we got here * due to DSACK blocks */ if (SEQ_LT(tp->snd_fack, th_ack)) { delivered_data += th_ack - tp->snd_una; tp->snd_fack = th_ack; - sack_changed = 1; + sack_changed = SACK_CHANGE; } } /* * Append received valid SACK blocks to sack_blocks[], but only if we * received new blocks from the other side. */ if (to->to_flags & TOF_SACK) { for (i = 0; i < to->to_nsacks; i++) { bcopy((to->to_sacks + i * TCPOLEN_SACK), &sack, sizeof(sack)); sack.start = ntohl(sack.start); sack.end = ntohl(sack.end); if (SEQ_GT(sack.end, sack.start) && SEQ_GT(sack.start, tp->snd_una) && SEQ_GT(sack.start, th_ack) && SEQ_LT(sack.start, tp->snd_max) && SEQ_GT(sack.end, tp->snd_una) && SEQ_LEQ(sack.end, tp->snd_max)) { sack_blocks[num_sack_blks++] = sack; } else if (SEQ_LEQ(sack.start, th_ack) && SEQ_LEQ(sack.end, th_ack)) { /* * Its a D-SACK block. */ tcp_record_dsack(tp, sack.start, sack.end, 0); } } } /* * Return if SND.UNA is not advanced and no valid SACK block is * received. */ if (num_sack_blks == 0) return (sack_changed); /* * Sort the SACK blocks so we can update the scoreboard with just one * pass. The overhead of sorting up to 4+1 elements is less than * making up to 4+1 passes over the scoreboard. */ for (i = 0; i < num_sack_blks; i++) { for (j = i + 1; j < num_sack_blks; j++) { if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) { sack = sack_blocks[i]; sack_blocks[i] = sack_blocks[j]; sack_blocks[j] = sack; } } } if (TAILQ_EMPTY(&tp->snd_holes)) { /* * Empty scoreboard. Need to initialize snd_fack (it may be * uninitialized or have a bogus value). Scoreboard holes * (from the sack blocks received) are created later below * (in the logic that adds holes to the tail of the * scoreboard). */ tp->snd_fack = SEQ_MAX(tp->snd_una, th_ack); tp->sackhint.sacked_bytes = 0; /* reset */ tp->sackhint.hole_bytes = 0; } /* * In the while-loop below, incoming SACK blocks (sack_blocks[]) and * SACK holes (snd_holes) are traversed from their tails with just * one pass in order to reduce the number of compares especially when * the bandwidth-delay product is large. * * Note: Typically, in the first RTT of SACK recovery, the highest * three or four SACK blocks with the same ack number are received. * In the second RTT, if retransmitted data segments are not lost, * the highest three or four SACK blocks with ack number advancing * are received. */ sblkp = &sack_blocks[num_sack_blks - 1]; /* Last SACK block */ tp->sackhint.last_sack_ack = sblkp->end; if (SEQ_LT(tp->snd_fack, sblkp->start)) { /* * The highest SACK block is beyond fack. First, * check if there was a successful Rescue Retransmission, * and move this hole left. With normal holes, snd_fack * is always to the right of the end. */ if (((temp = TAILQ_LAST(&tp->snd_holes, sackhole_head)) != NULL) && SEQ_LEQ(tp->snd_fack,temp->end)) { tp->sackhint.hole_bytes -= temp->end - temp->start; temp->start = SEQ_MAX(tp->snd_fack, SEQ_MAX(tp->snd_una, th_ack)); temp->end = sblkp->start; temp->rxmit = temp->start; delivered_data += sblkp->end - sblkp->start; tp->sackhint.hole_bytes += temp->end - temp->start; KASSERT(tp->sackhint.hole_bytes >= 0, ("sackhint hole bytes >= 0")); tp->snd_fack = sblkp->end; - sack_changed = 1; sblkp--; + sack_changed = SACK_NEWLOSS; } else { /* * Append a new SACK hole at the tail. If the * second or later highest SACK blocks are also * beyond the current fack, they will be inserted * by way of hole splitting in the while-loop below. */ temp = tcp_sackhole_insert(tp, tp->snd_fack,sblkp->start,NULL); if (temp != NULL) { delivered_data += sblkp->end - sblkp->start; tp->sackhint.hole_bytes += temp->end - temp->start; tp->snd_fack = sblkp->end; - sack_changed = 1; /* Go to the previous sack block. */ sblkp--; + sack_changed = SACK_CHANGE; } else { /* * We failed to add a new hole based on the current * sack block. Skip over all the sack blocks that * fall completely to the right of snd_fack and * proceed to trim the scoreboard based on the * remaining sack blocks. This also trims the * scoreboard for th_ack (which is sack_blocks[0]). */ while (sblkp >= sack_blocks && SEQ_LT(tp->snd_fack, sblkp->start)) sblkp--; if (sblkp >= sack_blocks && SEQ_LT(tp->snd_fack, sblkp->end)) { delivered_data += sblkp->end - tp->snd_fack; tp->snd_fack = sblkp->end; - sack_changed = 1; + /* + * While the Scoreboard didn't change in + * size, we only ended up here because + * some SACK data had to be dismissed. + */ + sack_changed = SACK_NEWLOSS; } } } } else if (SEQ_LT(tp->snd_fack, sblkp->end)) { /* fack is advanced. */ delivered_data += sblkp->end - tp->snd_fack; tp->snd_fack = sblkp->end; - sack_changed = 1; + sack_changed = SACK_CHANGE; } cur = TAILQ_LAST(&tp->snd_holes, sackhole_head); /* Last SACK hole. */ loss_hiack = tp->snd_fack; /* * Since the incoming sack blocks are sorted, we can process them * making one sweep of the scoreboard. */ while (cur != NULL) { if (!(sblkp >= sack_blocks)) { if (((loss_sblks >= tcprexmtthresh) || (loss_thresh > (tcprexmtthresh-1)*tp->t_maxseg))) break; loss_thresh += loss_hiack - cur->end; loss_hiack = cur->start; loss_sblks++; if (!((loss_sblks >= tcprexmtthresh) || (loss_thresh > (tcprexmtthresh-1)*tp->t_maxseg))) { notlost_bytes += cur->end - cur->start; } else { break; } cur = TAILQ_PREV(cur, sackhole_head, scblink); continue; } if (SEQ_GEQ(sblkp->start, cur->end)) { /* * SACKs data beyond the current hole. Go to the * previous sack block. */ sblkp--; continue; } if (SEQ_LEQ(sblkp->end, cur->start)) { /* * SACKs data before the current hole. Go to the * previous hole. */ loss_thresh += loss_hiack - cur->end; loss_hiack = cur->start; loss_sblks++; if (!((loss_sblks >= tcprexmtthresh) || (loss_thresh > (tcprexmtthresh-1)*tp->t_maxseg))) notlost_bytes += cur->end - cur->start; cur = TAILQ_PREV(cur, sackhole_head, scblink); continue; } tp->sackhint.sack_bytes_rexmit -= (SEQ_MIN(cur->rxmit, cur->end) - cur->start); KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, ("sackhint bytes rtx >= 0")); - sack_changed = 1; + sack_changed = SACK_CHANGE; if (SEQ_LEQ(sblkp->start, cur->start)) { /* Data acks at least the beginning of hole. */ if (SEQ_GEQ(sblkp->end, cur->end)) { /* Acks entire hole, so delete hole. */ delivered_data += (cur->end - cur->start); temp = cur; cur = TAILQ_PREV(cur, sackhole_head, scblink); tp->sackhint.hole_bytes -= temp->end - temp->start; tcp_sackhole_remove(tp, temp); /* * The sack block may ack all or part of the * next hole too, so continue onto the next * hole. */ continue; } else { /* Move start of hole forward. */ delivered_data += (sblkp->end - cur->start); tp->sackhint.hole_bytes -= sblkp->end - cur->start; cur->start = sblkp->end; cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); } } else { /* Data acks at least the end of hole. */ if (SEQ_GEQ(sblkp->end, cur->end)) { /* Move end of hole backward. */ delivered_data += (cur->end - sblkp->start); tp->sackhint.hole_bytes -= cur->end - sblkp->start; cur->end = sblkp->start; cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); if ((tp->t_flags & TF_LRD) && SEQ_GEQ(cur->rxmit, cur->end)) cur->rxmit = tp->snd_recover; } else { /* * ACKs some data in middle of a hole; need * to split current hole */ temp = tcp_sackhole_insert(tp, sblkp->end, cur->end, cur); + sack_changed = SACK_NEWLOSS; if (temp != NULL) { if (SEQ_GT(cur->rxmit, temp->rxmit)) { temp->rxmit = cur->rxmit; tp->sackhint.sack_bytes_rexmit += (SEQ_MIN(temp->rxmit, temp->end) - temp->start); } tp->sackhint.hole_bytes -= sblkp->end - sblkp->start; loss_thresh += loss_hiack - temp->end; loss_hiack = temp->start; loss_sblks++; if (!((loss_sblks >= tcprexmtthresh) || (loss_thresh > (tcprexmtthresh-1)*tp->t_maxseg))) notlost_bytes += temp->end - temp->start; cur->end = sblkp->start; cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); if ((tp->t_flags & TF_LRD) && SEQ_GEQ(cur->rxmit, cur->end)) cur->rxmit = tp->snd_recover; delivered_data += (sblkp->end - sblkp->start); } } } tp->sackhint.sack_bytes_rexmit += (SEQ_MIN(cur->rxmit, cur->end) - cur->start); /* * Testing sblkp->start against cur->start tells us whether * we're done with the sack block or the sack hole. * Accordingly, we advance one or the other. */ if (SEQ_LEQ(sblkp->start, cur->start)) { loss_thresh += loss_hiack - cur->end; loss_hiack = cur->start; loss_sblks++; if (!((loss_sblks >= tcprexmtthresh) || (loss_thresh > (tcprexmtthresh-1)*tp->t_maxseg))) notlost_bytes += cur->end - cur->start; cur = TAILQ_PREV(cur, sackhole_head, scblink); } else { sblkp--; } } KASSERT(!(TAILQ_EMPTY(&tp->snd_holes) && (tp->sackhint.hole_bytes != 0)), ("SACK scoreboard empty, but accounting non-zero\n")); KASSERT(notlost_bytes <= tp->sackhint.hole_bytes, ("SACK: more bytes marked notlost than in scoreboard holes")); if (!(to->to_flags & TOF_SACK)) /* * If this ACK did not contain any * SACK blocks, any only moved the * left edge right, it is a pure * cumulative ACK. Do not count * DupAck for this. Also required * for RFC6675 rescue retransmission. */ - sack_changed = 0; + sack_changed = SACK_NOCHANGE; tp->sackhint.delivered_data = delivered_data; tp->sackhint.sacked_bytes += delivered_data - left_edge_delta; tp->sackhint.lost_bytes = tp->sackhint.hole_bytes - notlost_bytes; KASSERT((delivered_data >= 0), ("delivered_data < 0")); KASSERT((tp->sackhint.sacked_bytes >= 0), ("sacked_bytes < 0")); return (sack_changed); } /* * Free all SACK holes to clear the scoreboard. */ void tcp_free_sackholes(struct tcpcb *tp) { struct sackhole *q; INP_WLOCK_ASSERT(tptoinpcb(tp)); while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL) tcp_sackhole_remove(tp, q); tp->sackhint.sack_bytes_rexmit = 0; tp->sackhint.delivered_data = 0; tp->sackhint.sacked_bytes = 0; tp->sackhint.hole_bytes = 0; tp->sackhint.lost_bytes = 0; KASSERT(tp->snd_numholes == 0, ("tp->snd_numholes == 0")); KASSERT(tp->sackhint.nexthole == NULL, ("tp->sackhint.nexthole == NULL")); } /* * Partial ack handling within a sack recovery episode. Keeping this very * simple for now. When a partial ack is received, force snd_cwnd to a value * that will allow the sender to transmit no more than 2 segments. If * necessary, a better scheme can be adopted at a later point, but for now, * the goal is to prevent the sender from bursting a large amount of data in * the midst of sack recovery. */ void tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th) { struct sackhole *temp; int num_segs = 1; u_int maxseg = tcp_maxseg(tp); INP_WLOCK_ASSERT(tptoinpcb(tp)); tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; /* Send one or 2 segments based on how much new data was acked. */ if ((BYTES_THIS_ACK(tp, th) / maxseg) >= 2) num_segs = 2; tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit + (tp->snd_nxt - tp->snd_recover) + num_segs * maxseg); if (tp->snd_cwnd > tp->snd_ssthresh) tp->snd_cwnd = tp->snd_ssthresh; tp->t_flags |= TF_ACKNOW; /* * RFC6675 rescue retransmission * Add a hole between th_ack (snd_una is not yet set) and snd_max, * if this was a pure cumulative ACK and no data was send beyond * recovery point. Since the data in the socket has not been freed * at this point, we check if the scoreboard is empty, and the ACK * delivered some new data, indicating a full ACK. Also, if the * recovery point is still at snd_max, we are probably application * limited. However, this inference might not always be true. The * rescue retransmission may rarely be slightly premature * compared to RFC6675. * The corresponding ACK+SACK will cause any further outstanding * segments to be retransmitted. This addresses a corner case, when * the trailing packets of a window are lost and no further data * is available for sending. */ if ((V_tcp_do_newsack) && SEQ_LT(th->th_ack, tp->snd_recover) && TAILQ_EMPTY(&tp->snd_holes) && (tp->sackhint.delivered_data > 0)) { /* * Exclude FIN sequence space in * the hole for the rescue retransmission, * and also don't create a hole, if only * the ACK for a FIN is outstanding. */ tcp_seq highdata = tp->snd_max; if (tp->t_flags & TF_SENTFIN) highdata--; highdata = SEQ_MIN(highdata, tp->snd_recover); if (th->th_ack != highdata) { tp->snd_fack = th->th_ack; if ((temp = tcp_sackhole_insert(tp, SEQ_MAX(th->th_ack, highdata - maxseg), highdata, NULL)) != NULL) tp->sackhint.hole_bytes += temp->end - temp->start; } } (void) tcp_output(tp); } #if 0 /* * Debug version of tcp_sack_output() that walks the scoreboard. Used for * now to sanity check the hint. */ static struct sackhole * tcp_sack_output_debug(struct tcpcb *tp, int *sack_bytes_rexmt) { struct sackhole *p; INP_WLOCK_ASSERT(tptoinpcb(tp)); *sack_bytes_rexmt = 0; TAILQ_FOREACH(p, &tp->snd_holes, scblink) { if (SEQ_LT(p->rxmit, p->end)) { if (SEQ_LT(p->rxmit, tp->snd_una)) {/* old SACK hole */ continue; } *sack_bytes_rexmt += (p->rxmit - p->start); break; } *sack_bytes_rexmt += (SEQ_MIN(p->rxmit, p->end) - p->start); } return (p); } #endif /* * Returns the next hole to retransmit and the number of retransmitted bytes * from the scoreboard. We store both the next hole and the number of * retransmitted bytes as hints (and recompute these on the fly upon SACK/ACK * reception). This avoids scoreboard traversals completely. * * The loop here will traverse *at most* one link. Here's the argument. For * the loop to traverse more than 1 link before finding the next hole to * retransmit, we would need to have at least 1 node following the current * hint with (rxmit == end). But, for all holes following the current hint, * (start == rxmit), since we have not yet retransmitted from them. * Therefore, in order to traverse more 1 link in the loop below, we need to * have at least one node following the current hint with (start == rxmit == * end). But that can't happen, (start == end) means that all the data in * that hole has been sacked, in which case, the hole would have been removed * from the scoreboard. */ struct sackhole * tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) { struct sackhole *hole = NULL; INP_WLOCK_ASSERT(tptoinpcb(tp)); *sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit; hole = tp->sackhint.nexthole; if (hole == NULL) return (hole); if (SEQ_GEQ(hole->rxmit, hole->end)) { for (;;) { hole = TAILQ_NEXT(hole, scblink); if (hole == NULL) return (hole); if (SEQ_LT(hole->rxmit, hole->end)) { tp->sackhint.nexthole = hole; break; } } } KASSERT(SEQ_LT(hole->start, hole->end), ("%s: hole.start >= hole.end", __func__)); if (!(V_tcp_do_newsack)) { KASSERT(SEQ_LT(hole->start, tp->snd_fack), ("%s: hole.start >= snd.fack", __func__)); KASSERT(SEQ_LT(hole->end, tp->snd_fack), ("%s: hole.end >= snd.fack", __func__)); KASSERT(SEQ_LT(hole->rxmit, tp->snd_fack), ("%s: hole.rxmit >= snd.fack", __func__)); if (SEQ_GEQ(hole->start, hole->end) || SEQ_GEQ(hole->start, tp->snd_fack) || SEQ_GEQ(hole->end, tp->snd_fack) || SEQ_GEQ(hole->rxmit, tp->snd_fack)) { log(LOG_CRIT,"tcp: invalid SACK hole (%u-%u,%u) vs fwd ack %u, ignoring.\n", hole->start, hole->end, hole->rxmit, tp->snd_fack); return (NULL); } } return (hole); } /* * After a timeout, the SACK list may be rebuilt. This SACK information * should be used to avoid retransmitting SACKed data. This function * traverses the SACK list to see if snd_nxt should be moved forward. */ void tcp_sack_adjust(struct tcpcb *tp) { struct sackhole *p, *cur = TAILQ_FIRST(&tp->snd_holes); INP_WLOCK_ASSERT(tptoinpcb(tp)); if (cur == NULL) return; /* No holes */ if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack)) return; /* We're already beyond any SACKed blocks */ /*- * Two cases for which we want to advance snd_nxt: * i) snd_nxt lies between end of one hole and beginning of another * ii) snd_nxt lies between end of last hole and snd_fack */ while ((p = TAILQ_NEXT(cur, scblink)) != NULL) { if (SEQ_LT(tp->snd_nxt, cur->end)) return; if (SEQ_GEQ(tp->snd_nxt, p->start)) cur = p; else { tp->snd_nxt = p->start; return; } } if (SEQ_LT(tp->snd_nxt, cur->end)) return; tp->snd_nxt = tp->snd_fack; } /* * Lost Retransmission Detection * Check is FACK is beyond the rexmit of the leftmost hole. * If yes, we restart sending from still existing holes, * and adjust cwnd via the congestion control module. */ void tcp_sack_lost_retransmission(struct tcpcb *tp, struct tcphdr *th) { struct sackhole *temp; if (IN_RECOVERY(tp->t_flags) && SEQ_GT(tp->snd_fack, tp->snd_recover) && ((temp = TAILQ_FIRST(&tp->snd_holes)) != NULL) && SEQ_GEQ(temp->rxmit, temp->end) && SEQ_GEQ(tp->snd_fack, temp->rxmit)) { TCPSTAT_INC(tcps_sack_lostrexmt); /* * Start retransmissions from the first hole, and * subsequently all other remaining holes, including * those, which had been sent completely before. */ tp->sackhint.nexthole = temp; TAILQ_FOREACH(temp, &tp->snd_holes, scblink) { if (SEQ_GEQ(tp->snd_fack, temp->rxmit) && SEQ_GEQ(temp->rxmit, temp->end)) temp->rxmit = temp->start; } /* * Remember the old ssthresh, to deduct the beta factor used * by the CC module. Finally, set cwnd to ssthresh just * prior to invoking another cwnd reduction by the CC * module, to not shrink it excessively. */ tp->snd_cwnd = tp->snd_ssthresh; /* * Formally exit recovery, and let the CC module adjust * ssthresh as intended. */ EXIT_RECOVERY(tp->t_flags); cc_cong_signal(tp, th, CC_NDUPACK); /* * For PRR, adjust recover_fs as if this new reduction * initialized this variable. * cwnd will be adjusted by SACK or PRR processing * subsequently, only set it to a safe value here. */ tp->snd_cwnd = tcp_maxseg(tp); tp->sackhint.recover_fs = (tp->snd_max - tp->snd_una) - tp->sackhint.recover_fs; } } diff --git a/sys/netinet/tcp_var.h b/sys/netinet/tcp_var.h index c6e24b187e0f..2236a1385b44 100644 --- a/sys/netinet/tcp_var.h +++ b/sys/netinet/tcp_var.h @@ -1,1613 +1,1620 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1993, 1994, 1995 * The Regents of the University of California. 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. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. * * @(#)tcp_var.h 8.4 (Berkeley) 5/24/95 */ #ifndef _NETINET_TCP_VAR_H_ #define _NETINET_TCP_VAR_H_ #include #include #ifdef _KERNEL #include "opt_kern_tls.h" #include #include #include #endif #define TCP_END_BYTE_INFO 8 /* Bytes that makeup the "end information array" */ /* Types of ending byte info */ #define TCP_EI_EMPTY_SLOT 0 #define TCP_EI_STATUS_CLIENT_FIN 0x1 #define TCP_EI_STATUS_CLIENT_RST 0x2 #define TCP_EI_STATUS_SERVER_FIN 0x3 #define TCP_EI_STATUS_SERVER_RST 0x4 #define TCP_EI_STATUS_RETRAN 0x5 #define TCP_EI_STATUS_PROGRESS 0x6 #define TCP_EI_STATUS_PERSIST_MAX 0x7 #define TCP_EI_STATUS_KEEP_MAX 0x8 #define TCP_EI_STATUS_DATA_A_CLOSE 0x9 #define TCP_EI_STATUS_RST_IN_FRONT 0xa #define TCP_EI_STATUS_2MSL 0xb #define TCP_EI_STATUS_MAX_VALUE 0xb #define TCP_TRK_REQ_LOG_NEW 0x01 #define TCP_TRK_REQ_LOG_COMPLETE 0x02 #define TCP_TRK_REQ_LOG_FREED 0x03 #define TCP_TRK_REQ_LOG_ALLOCFAIL 0x04 #define TCP_TRK_REQ_LOG_MOREYET 0x05 #define TCP_TRK_REQ_LOG_FORCEFREE 0x06 #define TCP_TRK_REQ_LOG_STALE 0x07 #define TCP_TRK_REQ_LOG_SEARCH 0x08 /************************************************/ /* Status bits we track to assure no duplicates, * the bits here are not used by the code but * for human representation. To check a bit we * take and shift over by 1 minus the value (1-8). */ /************************************************/ #define TCP_EI_BITS_CLIENT_FIN 0x001 #define TCP_EI_BITS_CLIENT_RST 0x002 #define TCP_EI_BITS_SERVER_FIN 0x004 #define TCP_EI_BITS_SERVER_RST 0x008 #define TCP_EI_BITS_RETRAN 0x010 #define TCP_EI_BITS_PROGRESS 0x020 #define TCP_EI_BITS_PRESIST_MAX 0x040 #define TCP_EI_BITS_KEEP_MAX 0x080 #define TCP_EI_BITS_DATA_A_CLO 0x100 #define TCP_EI_BITS_RST_IN_FR 0x200 /* a front state reset */ #define TCP_EI_BITS_2MS_TIMER 0x400 /* 2 MSL timer expired */ #if defined(_KERNEL) || defined(_WANT_TCPCB) #include /* TCP segment queue entry */ struct tseg_qent { TAILQ_ENTRY(tseg_qent) tqe_q; struct mbuf *tqe_m; /* mbuf contains packet */ struct mbuf *tqe_last; /* last mbuf in chain */ tcp_seq tqe_start; /* TCP Sequence number start */ int tqe_len; /* TCP segment data length */ uint32_t tqe_flags; /* The flags from tcp_get_flags() */ uint32_t tqe_mbuf_cnt; /* Count of mbuf overhead */ }; TAILQ_HEAD(tsegqe_head, tseg_qent); struct sackblk { tcp_seq start; /* start seq no. of sack block */ tcp_seq end; /* end seq no. */ }; struct sackhole { tcp_seq start; /* start seq no. of hole */ tcp_seq end; /* end seq no. */ tcp_seq rxmit; /* next seq. no in hole to be retransmitted */ TAILQ_ENTRY(sackhole) scblink; /* scoreboard linkage */ }; struct sackhint { struct sackhole *nexthole; int32_t sack_bytes_rexmit; tcp_seq last_sack_ack; /* Most recent/largest sacked ack */ int32_t delivered_data; /* Newly acked data from last SACK */ int32_t sacked_bytes; /* Total sacked bytes reported by the * receiver via sack option */ uint32_t recover_fs; /* Flight Size at the start of Loss recovery */ uint32_t prr_delivered; /* Total bytes delivered using PRR */ uint32_t prr_out; /* Bytes sent during IN_RECOVERY */ int32_t hole_bytes; /* current number of bytes in scoreboard holes */ int32_t lost_bytes; /* number of rfc6675 IsLost() bytes */ }; #define SEGQ_EMPTY(tp) TAILQ_EMPTY(&(tp)->t_segq) STAILQ_HEAD(tcp_log_stailq, tcp_log_mem); #define TCP_TRK_TRACK_FLG_EMPTY 0x00 /* Available */ #define TCP_TRK_TRACK_FLG_USED 0x01 /* In use */ #define TCP_TRK_TRACK_FLG_OPEN 0x02 /* End is not valid (open range request) */ #define TCP_TRK_TRACK_FLG_SEQV 0x04 /* We had a sendfile that touched it */ #define TCP_TRK_TRACK_FLG_COMP 0x08 /* Sendfile as placed the last bits (range req only) */ #define TCP_TRK_TRACK_FLG_FSND 0x10 /* First send has been done into the seq space */ #define MAX_TCP_TRK_REQ 5 /* Max we will have at once */ struct tcp_sendfile_track { uint64_t timestamp; /* User sent timestamp */ uint64_t start; /* Start of sendfile offset */ uint64_t end; /* End if not open-range req */ uint64_t localtime; /* Time we actually got the req */ uint64_t deadline; /* If in CU mode, deadline to delivery */ uint64_t first_send; /* Time of first send in the range */ uint64_t cspr; /* Client suggested pace rate */ uint64_t sent_at_fs; /* What was t_sndbytes as we begun sending */ uint64_t rxt_at_fs; /* What was t_snd_rxt_bytes as we begun sending */ tcp_seq start_seq; /* First TCP Seq assigned */ tcp_seq end_seq; /* If range req last seq */ uint32_t flags; /* Type of request open etc */ uint32_t sbcc_at_s; /* When we allocate what is the sb_cc */ uint32_t hint_maxseg; /* Client hinted maxseg */ uint32_t hybrid_flags; /* Hybrid flags on this request */ }; /* * Change Query responses for a stack switch we create a structure * that allows query response from the new stack to the old, if * supported. * * There are three queries currently defined. * - sendmap * - timers * - rack_times * * For the sendmap query the caller fills in the * req and the req_param as the first seq (usually * snd_una). When the response comes back indicating * that there was data (return value 1), then the caller * can build a sendmap entry based on the range and the * times. The next query would then be done at the * newly created sendmap_end. Repeated until sendmap_end == snd_max. * * Flags in sendmap_flags are defined below as well. * * For timers the standard PACE_TMR_XXXX flags are returned indicating * a pacing timer (possibly) and one other timer. If pacing timer then * the expiration timeout time in microseconds is in timer_pacing_to. * And the value used with whatever timer (if a flag is set) is in * timer_rxt. If no timers are running a 0 is returned and of * course no flags are set in timer_hpts_flags. * * The rack_times are a misc collection of information that * the old stack might possibly fill in. Of course its possible * that an old stack may not have a piece of information. If so * then setting that value to zero is advised. Setting any * timestamp passed should only place a zero in it when it * is unfilled. This may mean that a time is off by a micro-second * but this is ok in the grand scheme of things. * * When switching stacks it is desireable to get as much information * from the old stack to the new stack as possible. Though not always * will the stack be compatible in the types of information. The * init() function needs to take care when it begins changing * things such as inp_flags2 and the timer units to position these * changes at a point where it is unlikely they will fail after * making such changes. A stack optionally can have an "undo" * function * * To transfer information to the old stack from the new in * respect to LRO and the inp_flags2, the new stack should set * the inp_flags2 to what it supports. The old stack in its * fini() function should call the tcp_handle_orphaned_packets() * to clean up any packets. Note that a new stack should attempt */ /* Query types */ #define TCP_QUERY_SENDMAP 1 #define TCP_QUERY_TIMERS_UP 2 #define TCP_QUERY_RACK_TIMES 3 /* Flags returned in sendmap_flags */ #define SNDMAP_ACKED 0x000001/* The remote endpoint acked this */ #define SNDMAP_OVERMAX 0x000008/* We have more retran's then we can fit */ #define SNDMAP_SACK_PASSED 0x000010/* A sack was done above this block */ #define SNDMAP_HAS_FIN 0x000040/* segment is sent with fin */ #define SNDMAP_TLP 0x000080/* segment sent as tail-loss-probe */ #define SNDMAP_HAS_SYN 0x000800/* SYN is on this guy */ #define SNDMAP_HAD_PUSH 0x008000/* Push was sent on original send */ #define SNDMAP_MASK (SNDMAP_ACKED|SNDMAP_OVERMAX|SNDMAP_SACK_PASSED|SNDMAP_HAS_FIN\ |SNDMAP_TLP|SNDMAP_HAS_SYN|SNDMAP_HAD_PUSH) #define SNDMAP_NRTX 3 struct tcp_query_resp { int req; uint32_t req_param; union { struct { tcp_seq sendmap_start; tcp_seq sendmap_end; int sendmap_send_cnt; uint64_t sendmap_time[SNDMAP_NRTX]; uint64_t sendmap_ack_arrival; int sendmap_flags; uint32_t sendmap_r_rtr_bytes; /* If FAS is available if not 0 */ uint32_t sendmap_fas; uint8_t sendmap_dupacks; }; struct { uint32_t timer_hpts_flags; uint32_t timer_pacing_to; uint32_t timer_timer_exp; }; struct { /* Timestamps and rtt's */ uint32_t rack_reorder_ts; /* Last uscts that reordering was seen */ uint32_t rack_num_dsacks; /* Num of dsacks seen */ uint32_t rack_rxt_last_time; /* Last time a RXT/TLP or rack tmr went off */ uint32_t rack_min_rtt; /* never 0 smallest rtt seen */ uint32_t rack_rtt; /* Last rtt used by rack */ uint32_t rack_tmit_time; /* The time the rtt seg was tmited */ uint32_t rack_time_went_idle; /* If in persist the time we went idle */ /* Prr data */ uint32_t rack_sacked; uint32_t rack_holes_rxt; uint32_t rack_prr_delivered; uint32_t rack_prr_recovery_fs; uint32_t rack_prr_out; uint32_t rack_prr_sndcnt; /* TLP data */ uint16_t rack_tlp_cnt_out; /* How many tlp's have been sent */ /* Various bits */ uint8_t rack_tlp_out; /* Is a TLP outstanding */ uint8_t rack_srtt_measured; /* The previous stack has measured srtt */ uint8_t rack_in_persist; /* Is the old stack in persists? */ uint8_t rack_wanted_output; /* Did the prevous stack have a want output set */ }; }; }; #define TCP_TMR_GRANULARITY_TICKS 1 /* TCP timers are in ticks (msec if hz=1000) */ #define TCP_TMR_GRANULARITY_USEC 2 /* TCP timers are in microseconds */ typedef enum { TT_REXMT = 0, TT_PERSIST, TT_KEEP, TT_2MSL, TT_DELACK, TT_N, } tt_which; typedef enum { TT_PROCESSING = 0, TT_PROCESSED, TT_STARTING, TT_STOPPING, } tt_what; /* * Tcp control block, one per tcp connection. */ struct tcpcb { struct inpcb t_inpcb; /* embedded protocol independent cb */ #define t_start_zero t_fb #define t_zero_size (sizeof(struct tcpcb) - \ offsetof(struct tcpcb, t_start_zero)) struct tcp_function_block *t_fb;/* TCP function call block */ void *t_fb_ptr; /* Pointer to t_fb specific data */ struct callout t_callout; sbintime_t t_timers[TT_N]; sbintime_t t_precisions[TT_N]; /* HPTS. Used by BBR and Rack stacks. See tcp_hpts.c for more info. */ TAILQ_ENTRY(tcpcb) t_hpts; /* linkage to HPTS ring */ STAILQ_HEAD(, mbuf) t_inqueue; /* HPTS input packets queue */ uint32_t t_hpts_request; /* Current hpts request, zero if * fits in the pacing window. */ uint32_t t_hpts_slot; /* HPTS wheel slot this tcb is. */ uint32_t t_hpts_drop_reas; /* Reason we are dropping the pcb. */ uint32_t t_hpts_gencnt; uint16_t t_hpts_cpu; /* CPU chosen by hpts_cpuid(). */ uint16_t t_lro_cpu; /* CPU derived from LRO. */ #define HPTS_CPU_NONE ((uint16_t)-1) enum { IHPTS_NONE = 0, IHPTS_ONQUEUE, IHPTS_MOVING, } t_in_hpts; /* Is it linked into HPTS? */ uint32_t t_maxseg:24, /* maximum segment size */ _t_logstate:8; /* State of "black box" logging */ uint32_t t_port:16, /* Tunneling (over udp) port */ t_state:4, /* state of this connection */ t_idle_reduce : 1, t_delayed_ack: 7, /* Delayed ack variable */ t_fin_is_rst: 1, /* Are fin's treated as resets */ t_log_state_set: 1, bits_spare : 2; u_int t_flags; tcp_seq snd_una; /* sent but unacknowledged */ tcp_seq snd_max; /* highest sequence number sent; * used to recognize retransmits */ tcp_seq snd_nxt; /* send next */ tcp_seq snd_up; /* send urgent pointer */ uint32_t snd_wnd; /* send window */ uint32_t snd_cwnd; /* congestion-controlled window */ uint32_t ts_offset; /* our timestamp offset */ uint32_t rfbuf_ts; /* recv buffer autoscaling timestamp */ int rcv_numsacks; /* # distinct sack blks present */ u_int t_tsomax; /* TSO total burst length limit */ u_int t_tsomaxsegcount; /* TSO maximum segment count */ u_int t_tsomaxsegsize; /* TSO maximum segment size in bytes */ tcp_seq rcv_nxt; /* receive next */ tcp_seq rcv_adv; /* advertised window */ uint32_t rcv_wnd; /* receive window */ u_int t_flags2; /* More tcpcb flags storage */ int t_srtt; /* smoothed round-trip time */ int t_rttvar; /* variance in round-trip time */ uint32_t ts_recent; /* timestamp echo data */ u_char snd_scale; /* window scaling for send window */ u_char rcv_scale; /* window scaling for recv window */ u_char snd_limited; /* segments limited transmitted */ u_char request_r_scale; /* pending window scaling */ tcp_seq last_ack_sent; u_int t_rcvtime; /* inactivity time */ tcp_seq rcv_up; /* receive urgent pointer */ int t_segqlen; /* segment reassembly queue length */ uint32_t t_segqmbuflen; /* total reassembly queue byte length */ struct tsegqe_head t_segq; /* segment reassembly queue */ uint32_t snd_ssthresh; /* snd_cwnd size threshold for * for slow start exponential to * linear switch */ tcp_seq snd_wl1; /* window update seg seq number */ tcp_seq snd_wl2; /* window update seg ack number */ tcp_seq irs; /* initial receive sequence number */ tcp_seq iss; /* initial send sequence number */ u_int t_acktime; /* RACK and BBR incoming new data was acked */ u_int t_sndtime; /* time last data was sent */ u_int ts_recent_age; /* when last updated */ tcp_seq snd_recover; /* for use in NewReno Fast Recovery */ char t_oobflags; /* have some */ char t_iobc; /* input character */ uint8_t t_nic_ktls_xmit:1, /* active nic ktls xmit sessions */ t_nic_ktls_xmit_dis:1, /* disabled nic xmit ktls? */ t_nic_ktls_spare:6; /* spare nic ktls */ int t_rxtcur; /* current retransmit value (ticks) */ int t_rxtshift; /* log(2) of rexmt exp. backoff */ u_int t_rtttime; /* RTT measurement start time */ tcp_seq t_rtseq; /* sequence number being timed */ u_int t_starttime; /* time connection was established */ u_int t_fbyte_in; /* ticks time first byte queued in */ u_int t_fbyte_out; /* ticks time first byte queued out */ u_int t_pmtud_saved_maxseg; /* pre-blackhole MSS */ int t_blackhole_enter; /* when to enter blackhole detection */ int t_blackhole_exit; /* when to exit blackhole detection */ u_int t_rttmin; /* minimum rtt allowed */ int t_softerror; /* possible error not yet reported */ uint32_t max_sndwnd; /* largest window peer has offered */ uint32_t snd_cwnd_prev; /* cwnd prior to retransmit */ uint32_t snd_ssthresh_prev; /* ssthresh prior to retransmit */ tcp_seq snd_recover_prev; /* snd_recover prior to retransmit */ int t_sndzerowin; /* zero-window updates sent */ int snd_numholes; /* number of holes seen by sender */ u_int t_badrxtwin; /* window for retransmit recovery */ TAILQ_HEAD(sackhole_head, sackhole) snd_holes; /* SACK scoreboard (sorted) */ tcp_seq snd_fack; /* last seq number(+1) sack'd by rcv'r*/ struct sackblk sackblks[MAX_SACK_BLKS]; /* seq nos. of sack blocks */ struct sackhint sackhint; /* SACK scoreboard hint */ int t_rttlow; /* smallest observerved RTT */ int rfbuf_cnt; /* recv buffer autoscaling byte count */ struct toedev *tod; /* toedev handling this connection */ int t_sndrexmitpack; /* retransmit packets sent */ int t_rcvoopack; /* out-of-order packets received */ void *t_toe; /* TOE pcb pointer */ struct cc_algo *t_cc; /* congestion control algorithm */ struct cc_var t_ccv; /* congestion control specific vars */ int t_bytes_acked; /* # bytes acked during current RTT */ u_int t_maxunacktime; u_int t_keepinit; /* time to establish connection */ u_int t_keepidle; /* time before keepalive probes begin */ u_int t_keepintvl; /* interval between keepalives */ u_int t_keepcnt; /* number of keepalives before close */ int t_dupacks; /* consecutive dup acks recd */ int t_lognum; /* Number of log entries */ int t_loglimit; /* Maximum number of log entries */ uint32_t t_rcep; /* Number of received CE marked pkts */ uint32_t t_scep; /* Synced number of delivered CE pkts */ int64_t t_pacing_rate; /* bytes / sec, -1 => unlimited */ struct tcp_log_stailq t_logs; /* Log buffer */ struct tcp_log_id_node *t_lin; struct tcp_log_id_bucket *t_lib; const char *t_output_caller; /* Function that called tcp_output */ struct statsblob *t_stats; /* Per-connection stats */ /* Should these be a pointer to the arrays or an array? */ uint32_t t_logsn; /* Log "serial number" */ uint32_t gput_ts; /* Time goodput measurement started */ tcp_seq gput_seq; /* Outbound measurement seq */ tcp_seq gput_ack; /* Inbound measurement ack */ int32_t t_stats_gput_prev; /* XXXLAS: Prev gput measurement */ uint32_t t_maxpeakrate; /* max peak rate set by user, bytes/s */ uint32_t t_sndtlppack; /* tail loss probe packets sent */ uint64_t t_sndtlpbyte; /* total tail loss probe bytes sent */ uint64_t t_sndbytes; /* total bytes sent */ uint64_t t_snd_rxt_bytes; /* total bytes retransmitted */ uint32_t t_dsack_bytes; /* dsack bytes received */ uint32_t t_dsack_tlp_bytes; /* dsack bytes received for TLPs sent */ uint32_t t_dsack_pack; /* dsack packets we have eceived */ uint8_t t_tmr_granularity; /* Granularity of all timers srtt etc */ uint8_t t_rttupdated; /* number of times rtt sampled */ /* TCP Fast Open */ uint8_t t_tfo_client_cookie_len; /* TFO client cookie length */ uint32_t t_end_info_status; /* Status flag of end info */ unsigned int *t_tfo_pending; /* TFO server pending counter */ union { uint8_t client[TCP_FASTOPEN_MAX_COOKIE_LEN]; uint64_t server; } t_tfo_cookie; /* TCP Fast Open cookie to send */ union { uint8_t t_end_info_bytes[TCP_END_BYTE_INFO]; uint64_t t_end_info; }; struct osd t_osd; /* storage for Khelp module data */ uint8_t _t_logpoint; /* Used when a BB log points is enabled */ /* * Keep all #ifdef'ed components at the end of the structure! * This is important to minimize problems when compiling modules * using this structure from within the modules' directory. */ #ifdef TCP_REQUEST_TRK /* Response tracking addons. */ uint8_t t_tcpreq_req; /* Request count */ uint8_t t_tcpreq_open; /* Number of open range requests */ uint8_t t_tcpreq_closed; /* Number of closed range requests */ uint32_t tcp_hybrid_start; /* Num of times we started hybrid pacing */ uint32_t tcp_hybrid_stop; /* Num of times we stopped hybrid pacing */ uint32_t tcp_hybrid_error; /* Num of times we failed to start hybrid pacing */ struct tcp_sendfile_track t_tcpreq_info[MAX_TCP_TRK_REQ]; #endif #ifdef TCP_ACCOUNTING uint64_t tcp_cnt_counters[TCP_NUM_CNT_COUNTERS]; uint64_t tcp_proc_time[TCP_NUM_CNT_COUNTERS]; #endif #ifdef TCPPCAP struct mbufq t_inpkts; /* List of saved input packets. */ struct mbufq t_outpkts; /* List of saved output packets. */ #endif }; #endif /* _KERNEL || _WANT_TCPCB */ #ifdef _KERNEL struct tcptemp { u_char tt_ipgen[40]; /* the size must be of max ip header, now IPv6 */ struct tcphdr tt_t; }; +/* SACK scoreboard update status */ +typedef enum { + SACK_NOCHANGE = 0, + SACK_CHANGE, + SACK_NEWLOSS +} sackstatus_t; + /* Enable TCP/UDP tunneling port */ #define TCP_TUNNELING_PORT_MIN 0 #define TCP_TUNNELING_PORT_MAX 65535 #define TCP_TUNNELING_PORT_DEFAULT 0 /* Enable TCP/UDP tunneling port */ #define TCP_TUNNELING_OVERHEAD_MIN sizeof(struct udphdr) #define TCP_TUNNELING_OVERHEAD_MAX 1024 #define TCP_TUNNELING_OVERHEAD_DEFAULT TCP_TUNNELING_OVERHEAD_MIN /* Minimum map entries limit value, if set */ #define TCP_MIN_MAP_ENTRIES_LIMIT 128 /* * TODO: We yet need to brave plowing in * to tcp_input() and the pru_usrreq() block. * Right now these go to the old standards which * are somewhat ok, but in the long term may * need to be changed. If we do tackle tcp_input() * then we need to get rid of the tcp_do_segment() * function below. */ /* Flags for tcp functions */ #define TCP_FUNC_BEING_REMOVED 0x01 /* Can no longer be referenced */ #define TCP_FUNC_OUTPUT_CANDROP 0x02 /* tfb_tcp_output may ask tcp_drop */ /** * If defining the optional tcp_timers, in the * tfb_tcp_timer_stop call you must use the * callout_async_drain() function with the * tcp_timer_discard callback. You should check * the return of callout_async_drain() and if 0 * increment tt_draincnt. Since the timer sub-system * does not know your callbacks you must provide a * stop_all function that loops through and calls * tcp_timer_stop() with each of your defined timers. * * Adding a tfb_tcp_handoff_ok function allows the socket * option to change stacks to query you even if the * connection is in a later stage. You return 0 to * say you can take over and run your stack, you return * non-zero (an error number) to say no you can't. * If the function is undefined you can only change * in the early states (before connect or listen). * * tfb_tcp_fb_init is used to allow the new stack to * setup its control block. Among the things it must * do is: * a) Make sure that the inp_flags2 is setup correctly * for LRO. There are two flags that the previous * stack may have set INP_MBUF_ACKCMP and * INP_SUPPORTS_MBUFQ. If the new stack does not * support these it *should* clear the flags. * b) Make sure that the timers are in the proper * granularity that the stack wants. The stack * should check the t_tmr_granularity field. Currently * there are two values that it may hold * TCP_TMR_GRANULARITY_TICKS and TCP_TMR_GRANULARITY_USEC. * Use the functions tcp_timer_convert(tp, granularity); * to move the timers to the correct format for your stack. * * The new stack may also optionally query the tfb_chg_query * function if the old stack has one. The new stack may ask * for one of three entries and can also state to the old * stack its support for the INP_MBUF_ACKCMP and * INP_SUPPORTS_MBUFQ. This is important since if there are * queued ack's without that statement the old stack will * be forced to discard the queued acks. The requests that * can be made for information by the new stacks are: * * Note also that the tfb_tcp_fb_init() when called can * determine if a query is needed by looking at the * value passed in the ptr. The ptr is designed to be * set in with any allocated memory, but the address * of the condtion (ptr == &tp->t_fb_ptr) will be * true if this is not a stack switch but the initial * setup of a tcb (which means no query would be needed). * If, however, the value is not t_fb_ptr, then the caller * is in the middle of a stack switch and is the new stack. * A query would be appropriate (if the new stack support * the query mechanism). * * TCP_QUERY_SENDMAP - Query of outstanding data. * TCP_QUERY_TIMERS_UP - Query about running timers. * TCP_SUPPORTED_LRO - Declaration in req_param of * the inp_flags2 supported by * the new stack. * TCP_QUERY_RACK_TIMES - Enquire about various timestamps * and states the old stack may be in. * * tfb_tcp_fb_fini is changed to add a flag to tell * the old stack if the tcb is being destroyed or * not. A one in the flag means the TCB is being * destroyed, a zero indicates its transitioning to * another stack (via socket option). The * tfb_tcp_fb_fini() function itself should not change timers * or inp_flags2 (the tfb_tcp_fb_init() must do that). However * if the old stack supports the LRO mbuf queuing, and the new * stack does not communicate via chg messages that it too does, * it must assume it does not and free any queued mbufs. * */ struct tcp_function_block { char tfb_tcp_block_name[TCP_FUNCTION_NAME_LEN_MAX]; int (*tfb_tcp_output)(struct tcpcb *); void (*tfb_tcp_do_segment)(struct tcpcb *, struct mbuf *, struct tcphdr *, int, int, uint8_t); int (*tfb_do_segment_nounlock)(struct tcpcb *, struct mbuf *, struct tcphdr *, int, int, uint8_t, int, struct timeval *); int (*tfb_do_queued_segments)(struct tcpcb *, int); int (*tfb_tcp_ctloutput)(struct tcpcb *, struct sockopt *); /* Optional memory allocation/free routine */ int (*tfb_tcp_fb_init)(struct tcpcb *, void **); void (*tfb_tcp_fb_fini)(struct tcpcb *, int); /* Optional timers, must define all if you define one */ int (*tfb_tcp_timer_stop_all)(struct tcpcb *); void (*tfb_tcp_rexmit_tmr)(struct tcpcb *); int (*tfb_tcp_handoff_ok)(struct tcpcb *); void (*tfb_tcp_mtu_chg)(struct tcpcb *tp); int (*tfb_pru_options)(struct tcpcb *, int); void (*tfb_hwtls_change)(struct tcpcb *, int); int (*tfb_chg_query)(struct tcpcb *, struct tcp_query_resp *); void (*tfb_switch_failed)(struct tcpcb *); bool (*tfb_early_wake_check)(struct tcpcb *); int (*tfb_compute_pipe)(struct tcpcb *tp); volatile uint32_t tfb_refcnt; uint32_t tfb_flags; uint8_t tfb_id; }; struct tcp_function { TAILQ_ENTRY(tcp_function) tf_next; char tf_name[TCP_FUNCTION_NAME_LEN_MAX]; struct tcp_function_block *tf_fb; }; TAILQ_HEAD(tcp_funchead, tcp_function); struct tcpcb * tcp_drop(struct tcpcb *, int); #ifdef _NETINET_IN_PCB_H_ #define intotcpcb(inp) __containerof((inp), struct tcpcb, t_inpcb) #define sototcpcb(so) intotcpcb(sotoinpcb(so)) #define tptoinpcb(tp) (&(tp)->t_inpcb) #define tptosocket(tp) (tp)->t_inpcb.inp_socket /* * tcp_output() * Handles tcp_drop request from advanced stacks and reports that inpcb is * gone with negative return code. * Drop in replacement for the default stack. */ static inline int tcp_output(struct tcpcb *tp) { struct inpcb *inp = tptoinpcb(tp); int rv; INP_WLOCK_ASSERT(inp); rv = tp->t_fb->tfb_tcp_output(tp); if (rv < 0) { KASSERT(tp->t_fb->tfb_flags & TCP_FUNC_OUTPUT_CANDROP, ("TCP stack %s requested tcp_drop(%p)", tp->t_fb->tfb_tcp_block_name, tp)); tp = tcp_drop(tp, -rv); if (tp) INP_WUNLOCK(inp); } return (rv); } /* * tcp_output_unlock() * Always returns unlocked, handles drop request from advanced stacks. * Always returns positive error code. */ static inline int tcp_output_unlock(struct tcpcb *tp) { struct inpcb *inp = tptoinpcb(tp); int rv; INP_WLOCK_ASSERT(inp); rv = tp->t_fb->tfb_tcp_output(tp); if (rv < 0) { KASSERT(tp->t_fb->tfb_flags & TCP_FUNC_OUTPUT_CANDROP, ("TCP stack %s requested tcp_drop(%p)", tp->t_fb->tfb_tcp_block_name, tp)); rv = -rv; tp = tcp_drop(tp, rv); if (tp) INP_WUNLOCK(inp); } else INP_WUNLOCK(inp); return (rv); } /* * tcp_output_nodrop() * Always returns locked. It is caller's responsibility to run tcp_drop()! * Useful in syscall implementations, when we want to perform some logging * and/or tracing with tcpcb before calling tcp_drop(). To be used with * tcp_unlock_or_drop() later. * * XXXGL: maybe don't allow stacks to return a drop request at certain * TCP states? Why would it do in connect(2)? In recv(2)? */ static inline int tcp_output_nodrop(struct tcpcb *tp) { int rv; INP_WLOCK_ASSERT(tptoinpcb(tp)); rv = tp->t_fb->tfb_tcp_output(tp); KASSERT(rv >= 0 || tp->t_fb->tfb_flags & TCP_FUNC_OUTPUT_CANDROP, ("TCP stack %s requested tcp_drop(%p)", tp->t_fb->tfb_tcp_block_name, tp)); return (rv); } /* * tcp_unlock_or_drop() * Handle return code from tfb_tcp_output() after we have logged/traced, * to be used with tcp_output_nodrop(). */ static inline int tcp_unlock_or_drop(struct tcpcb *tp, int tcp_output_retval) { struct inpcb *inp = tptoinpcb(tp); INP_WLOCK_ASSERT(inp); if (tcp_output_retval < 0) { tcp_output_retval = -tcp_output_retval; if (tcp_drop(tp, tcp_output_retval) != NULL) INP_WUNLOCK(inp); } else INP_WUNLOCK(inp); return (tcp_output_retval); } #endif /* _NETINET_IN_PCB_H_ */ static int inline tcp_packets_this_ack(struct tcpcb *tp, tcp_seq ack) { return ((ack - tp->snd_una) / tp->t_maxseg + ((((ack - tp->snd_una) % tp->t_maxseg) != 0) ? 1 : 0)); } #endif /* _KERNEL */ /* * Flags and utility macros for the t_flags field. */ #define TF_ACKNOW 0x00000001 /* ack peer immediately */ #define TF_DELACK 0x00000002 /* ack, but try to delay it */ #define TF_NODELAY 0x00000004 /* don't delay packets to coalesce */ #define TF_NOOPT 0x00000008 /* don't use tcp options */ #define TF_SENTFIN 0x00000010 /* have sent FIN */ #define TF_REQ_SCALE 0x00000020 /* have/will request window scaling */ #define TF_RCVD_SCALE 0x00000040 /* other side has requested scaling */ #define TF_REQ_TSTMP 0x00000080 /* have/will request timestamps */ #define TF_RCVD_TSTMP 0x00000100 /* a timestamp was received in SYN */ #define TF_SACK_PERMIT 0x00000200 /* other side said I could SACK */ #define TF_NEEDSYN 0x00000400 /* send SYN (implicit state) */ #define TF_NEEDFIN 0x00000800 /* send FIN (implicit state) */ #define TF_NOPUSH 0x00001000 /* don't push */ #define TF_PREVVALID 0x00002000 /* saved values for bad rxmit valid * Note: accessing and restoring from * these may only be done in the 1st * RTO recovery round (t_rxtshift == 1) */ #define TF_WAKESOR 0x00004000 /* wake up receive socket */ #define TF_GPUTINPROG 0x00008000 /* Goodput measurement in progress */ #define TF_MORETOCOME 0x00010000 /* More data to be appended to sock */ #define TF_SONOTCONN 0x00020000 /* needs soisconnected() on ESTAB */ #define TF_LASTIDLE 0x00040000 /* connection was previously idle */ #define TF_RXWIN0SENT 0x00080000 /* sent a receiver win 0 in response */ #define TF_FASTRECOVERY 0x00100000 /* in NewReno Fast Recovery */ #define TF_WASFRECOVERY 0x00200000 /* was in NewReno Fast Recovery */ #define TF_SIGNATURE 0x00400000 /* require MD5 digests (RFC2385) */ #define TF_FORCEDATA 0x00800000 /* force out a byte */ #define TF_TSO 0x01000000 /* TSO enabled on this connection */ #define TF_TOE 0x02000000 /* this connection is offloaded */ #define TF_CLOSED 0x04000000 /* close(2) called on socket */ #define TF_UNUSED1 0x08000000 /* unused */ #define TF_LRD 0x10000000 /* Lost Retransmission Detection */ #define TF_CONGRECOVERY 0x20000000 /* congestion recovery mode */ #define TF_WASCRECOVERY 0x40000000 /* was in congestion recovery */ #define TF_FASTOPEN 0x80000000 /* TCP Fast Open indication */ #define IN_FASTRECOVERY(t_flags) (t_flags & TF_FASTRECOVERY) #define ENTER_FASTRECOVERY(t_flags) t_flags |= TF_FASTRECOVERY #define EXIT_FASTRECOVERY(t_flags) t_flags &= ~TF_FASTRECOVERY #define IN_CONGRECOVERY(t_flags) (t_flags & TF_CONGRECOVERY) #define ENTER_CONGRECOVERY(t_flags) t_flags |= TF_CONGRECOVERY #define EXIT_CONGRECOVERY(t_flags) t_flags &= ~TF_CONGRECOVERY #define IN_RECOVERY(t_flags) (t_flags & (TF_CONGRECOVERY | TF_FASTRECOVERY)) #define ENTER_RECOVERY(t_flags) t_flags |= (TF_CONGRECOVERY | TF_FASTRECOVERY) #define EXIT_RECOVERY(t_flags) t_flags &= ~(TF_CONGRECOVERY | TF_FASTRECOVERY) #if defined(_KERNEL) && !defined(TCP_RFC7413) #define IS_FASTOPEN(t_flags) (false) #else #define IS_FASTOPEN(t_flags) (t_flags & TF_FASTOPEN) #endif #define BYTES_THIS_ACK(tp, th) (th->th_ack - tp->snd_una) /* * Flags for the t_oobflags field. */ #define TCPOOB_HAVEDATA 0x01 #define TCPOOB_HADDATA 0x02 /* * Flags for the extended TCP flags field, t_flags2 */ #define TF2_PLPMTU_BLACKHOLE 0x00000001 /* Possible PLPMTUD Black Hole. */ #define TF2_PLPMTU_PMTUD 0x00000002 /* Allowed to attempt PLPMTUD. */ #define TF2_PLPMTU_MAXSEGSNT 0x00000004 /* Last seg sent was full seg. */ #define TF2_LOG_AUTO 0x00000008 /* Session is auto-logging. */ #define TF2_DROP_AF_DATA 0x00000010 /* Drop after all data ack'd */ #define TF2_ECN_PERMIT 0x00000020 /* connection ECN-ready */ #define TF2_ECN_SND_CWR 0x00000040 /* ECN CWR in queue */ #define TF2_ECN_SND_ECE 0x00000080 /* ECN ECE in queue */ #define TF2_ACE_PERMIT 0x00000100 /* Accurate ECN mode */ #define TF2_HPTS_CPU_SET 0x00000200 /* t_hpts_cpu is not random */ #define TF2_FBYTES_COMPLETE 0x00000400 /* We have first bytes in and out */ #define TF2_ECN_USE_ECT1 0x00000800 /* Use ECT(1) marking on session */ #define TF2_TCP_ACCOUNTING 0x00001000 /* Do TCP accounting */ #define TF2_HPTS_CALLS 0x00002000 /* tcp_output() called via HPTS */ #define TF2_MBUF_L_ACKS 0x00004000 /* large mbufs for ack compression */ #define TF2_MBUF_ACKCMP 0x00008000 /* mbuf ack compression ok */ #define TF2_SUPPORTS_MBUFQ 0x00010000 /* Supports the mbuf queue method */ #define TF2_MBUF_QUEUE_READY 0x00020000 /* Inputs can be queued */ #define TF2_DONT_SACK_QUEUE 0x00040000 /* Don't wake on sack */ #define TF2_CANNOT_DO_ECN 0x00080000 /* The stack does not do ECN */ /* * Structure to hold TCP options that are only used during segment * processing (in tcp_input), but not held in the tcpcb. * It's basically used to reduce the number of parameters * to tcp_dooptions and tcp_addoptions. * The binary order of the to_flags is relevant for packing of the * options in tcp_addoptions. */ struct tcpopt { u_int32_t to_flags; /* which options are present */ #define TOF_MSS 0x0001 /* maximum segment size */ #define TOF_SCALE 0x0002 /* window scaling */ #define TOF_SACKPERM 0x0004 /* SACK permitted */ #define TOF_TS 0x0010 /* timestamp */ #define TOF_SIGNATURE 0x0040 /* TCP-MD5 signature option (RFC2385) */ #define TOF_SACK 0x0080 /* Peer sent SACK option */ #define TOF_FASTOPEN 0x0100 /* TCP Fast Open (TFO) cookie */ #define TOF_MAXOPT 0x0200 u_int32_t to_tsval; /* new timestamp */ u_int32_t to_tsecr; /* reflected timestamp */ u_char *to_sacks; /* pointer to the first SACK blocks */ u_char *to_signature; /* pointer to the TCP-MD5 signature */ u_int8_t *to_tfo_cookie; /* pointer to the TFO cookie */ u_int16_t to_mss; /* maximum segment size */ u_int8_t to_wscale; /* window scaling */ u_int8_t to_nsacks; /* number of SACK blocks */ u_int8_t to_tfo_len; /* TFO cookie length */ u_int32_t to_spare; /* UTO */ }; /* * Flags for tcp_dooptions. */ #define TO_SYN 0x01 /* parse SYN-only options */ struct hc_metrics_lite { /* must stay in sync with hc_metrics */ uint32_t rmx_mtu; /* MTU for this path */ uint32_t rmx_ssthresh; /* outbound gateway buffer limit */ uint32_t rmx_rtt; /* estimated round trip time */ uint32_t rmx_rttvar; /* estimated rtt variance */ uint32_t rmx_cwnd; /* congestion window */ uint32_t rmx_sendpipe; /* outbound delay-bandwidth product */ uint32_t rmx_recvpipe; /* inbound delay-bandwidth product */ }; /* * Used by tcp_maxmtu() to communicate interface specific features * and limits at the time of connection setup. */ struct tcp_ifcap { int ifcap; u_int tsomax; u_int tsomaxsegcount; u_int tsomaxsegsize; }; #ifndef _NETINET_IN_PCB_H_ struct in_conninfo; #endif /* _NETINET_IN_PCB_H_ */ /* * The smoothed round-trip time and estimated variance * are stored as fixed point numbers scaled by the values below. * For convenience, these scales are also used in smoothing the average * (smoothed = (1/scale)sample + ((scale-1)/scale)smoothed). * With these scales, srtt has 3 bits to the right of the binary point, * and thus an "ALPHA" of 0.875. rttvar has 2 bits to the right of the * binary point, and is smoothed with an ALPHA of 0.75. */ #define TCP_RTT_SCALE 32 /* multiplier for srtt; 3 bits frac. */ #define TCP_RTT_SHIFT 5 /* shift for srtt; 3 bits frac. */ #define TCP_RTTVAR_SCALE 16 /* multiplier for rttvar; 2 bits */ #define TCP_RTTVAR_SHIFT 4 /* shift for rttvar; 2 bits */ #define TCP_DELTA_SHIFT 2 /* see tcp_input.c */ /* * The initial retransmission should happen at rtt + 4 * rttvar. * Because of the way we do the smoothing, srtt and rttvar * will each average +1/2 tick of bias. When we compute * the retransmit timer, we want 1/2 tick of rounding and * 1 extra tick because of +-1/2 tick uncertainty in the * firing of the timer. The bias will give us exactly the * 1.5 tick we need. But, because the bias is * statistical, we have to test that we don't drop below * the minimum feasible timer (which is 2 ticks). * This version of the macro adapted from a paper by Lawrence * Brakmo and Larry Peterson which outlines a problem caused * by insufficient precision in the original implementation, * which results in inappropriately large RTO values for very * fast networks. */ #define TCP_REXMTVAL(tp) \ max((tp)->t_rttmin, (((tp)->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)) \ + (tp)->t_rttvar) >> TCP_DELTA_SHIFT) /* * TCP statistics. * Many of these should be kept per connection, * but that's inconvenient at the moment. */ struct tcpstat { uint64_t tcps_connattempt; /* connections initiated */ uint64_t tcps_accepts; /* connections accepted */ uint64_t tcps_connects; /* connections established */ uint64_t tcps_drops; /* connections dropped */ uint64_t tcps_conndrops; /* embryonic connections dropped */ uint64_t tcps_minmssdrops; /* average minmss too low drops */ uint64_t tcps_closed; /* conn. closed (includes drops) */ uint64_t tcps_segstimed; /* segs where we tried to get rtt */ uint64_t tcps_rttupdated; /* times we succeeded */ uint64_t tcps_delack; /* delayed acks sent */ uint64_t tcps_timeoutdrop; /* conn. dropped in rxmt timeout */ uint64_t tcps_rexmttimeo; /* retransmit timeouts */ uint64_t tcps_persisttimeo; /* persist timeouts */ uint64_t tcps_keeptimeo; /* keepalive timeouts */ uint64_t tcps_keepprobe; /* keepalive probes sent */ uint64_t tcps_keepdrops; /* connections dropped in keepalive */ uint64_t tcps_progdrops; /* drops due to no progress */ uint64_t tcps_sndtotal; /* total packets sent */ uint64_t tcps_sndpack; /* data packets sent */ uint64_t tcps_sndbyte; /* data bytes sent */ uint64_t tcps_sndrexmitpack; /* data packets retransmitted */ uint64_t tcps_sndrexmitbyte; /* data bytes retransmitted */ uint64_t tcps_sndrexmitbad; /* unnecessary packet retransmissions */ uint64_t tcps_sndacks; /* ack-only packets sent */ uint64_t tcps_sndprobe; /* window probes sent */ uint64_t tcps_sndurg; /* packets sent with URG only */ uint64_t tcps_sndwinup; /* window update-only packets sent */ uint64_t tcps_sndctrl; /* control (SYN|FIN|RST) packets sent */ uint64_t tcps_rcvtotal; /* total packets received */ uint64_t tcps_rcvpack; /* packets received in sequence */ uint64_t tcps_rcvbyte; /* bytes received in sequence */ uint64_t tcps_rcvbadsum; /* packets received with ccksum errs */ uint64_t tcps_rcvbadoff; /* packets received with bad offset */ uint64_t tcps_rcvreassfull; /* packets dropped for no reass space */ uint64_t tcps_rcvshort; /* packets received too short */ uint64_t tcps_rcvduppack; /* duplicate-only packets received */ uint64_t tcps_rcvdupbyte; /* duplicate-only bytes received */ uint64_t tcps_rcvpartduppack; /* packets with some duplicate data */ uint64_t tcps_rcvpartdupbyte; /* dup. bytes in part-dup. packets */ uint64_t tcps_rcvoopack; /* out-of-order packets received */ uint64_t tcps_rcvoobyte; /* out-of-order bytes received */ uint64_t tcps_rcvpackafterwin; /* packets with data after window */ uint64_t tcps_rcvbyteafterwin; /* bytes rcvd after window */ uint64_t tcps_rcvafterclose; /* packets rcvd after "close" */ uint64_t tcps_rcvwinprobe; /* rcvd window probe packets */ uint64_t tcps_rcvdupack; /* rcvd duplicate acks */ uint64_t tcps_rcvacktoomuch; /* rcvd acks for unsent data */ uint64_t tcps_rcvackpack; /* rcvd ack packets */ uint64_t tcps_rcvackbyte; /* bytes acked by rcvd acks */ uint64_t tcps_rcvwinupd; /* rcvd window update packets */ uint64_t tcps_pawsdrop; /* segments dropped due to PAWS */ uint64_t tcps_predack; /* times hdr predict ok for acks */ uint64_t tcps_preddat; /* times hdr predict ok for data pkts */ uint64_t tcps_pcbcachemiss; uint64_t tcps_cachedrtt; /* times cached RTT in route updated */ uint64_t tcps_cachedrttvar; /* times cached rttvar updated */ uint64_t tcps_cachedssthresh; /* times cached ssthresh updated */ uint64_t tcps_usedrtt; /* times RTT initialized from route */ uint64_t tcps_usedrttvar; /* times RTTVAR initialized from rt */ uint64_t tcps_usedssthresh; /* times ssthresh initialized from rt*/ uint64_t tcps_persistdrop; /* timeout in persist state */ uint64_t tcps_badsyn; /* bogus SYN, e.g. premature ACK */ uint64_t tcps_mturesent; /* resends due to MTU discovery */ uint64_t tcps_listendrop; /* listen queue overflows */ uint64_t tcps_badrst; /* ignored RSTs in the window */ uint64_t tcps_sc_added; /* entry added to syncache */ uint64_t tcps_sc_retransmitted; /* syncache entry was retransmitted */ uint64_t tcps_sc_dupsyn; /* duplicate SYN packet */ uint64_t tcps_sc_dropped; /* could not reply to packet */ uint64_t tcps_sc_completed; /* successful extraction of entry */ uint64_t tcps_sc_bucketoverflow;/* syncache per-bucket limit hit */ uint64_t tcps_sc_cacheoverflow; /* syncache cache limit hit */ uint64_t tcps_sc_reset; /* RST removed entry from syncache */ uint64_t tcps_sc_stale; /* timed out or listen socket gone */ uint64_t tcps_sc_aborted; /* syncache entry aborted */ uint64_t tcps_sc_badack; /* removed due to bad ACK */ uint64_t tcps_sc_unreach; /* ICMP unreachable received */ uint64_t tcps_sc_zonefail; /* zalloc() failed */ uint64_t tcps_sc_sendcookie; /* SYN cookie sent */ uint64_t tcps_sc_recvcookie; /* SYN cookie received */ uint64_t tcps_hc_added; /* entry added to hostcache */ uint64_t tcps_hc_bucketoverflow;/* hostcache per bucket limit hit */ uint64_t tcps_finwait2_drops; /* Drop FIN_WAIT_2 connection after time limit */ /* SACK related stats */ uint64_t tcps_sack_recovery_episode; /* SACK recovery episodes */ uint64_t tcps_sack_rexmits; /* SACK rexmit segments */ uint64_t tcps_sack_rexmit_bytes; /* SACK rexmit bytes */ uint64_t tcps_sack_rcv_blocks; /* SACK blocks (options) received */ uint64_t tcps_sack_send_blocks; /* SACK blocks (options) sent */ uint64_t tcps_sack_lostrexmt; /* SACK lost retransmission recovered */ uint64_t tcps_sack_sboverflow; /* times scoreboard overflowed */ /* ECN related stats */ uint64_t tcps_ecn_rcvce; /* ECN Congestion Experienced */ uint64_t tcps_ecn_rcvect0; /* ECN Capable Transport */ uint64_t tcps_ecn_rcvect1; /* ECN Capable Transport */ uint64_t tcps_ecn_shs; /* ECN successful handshakes */ uint64_t tcps_ecn_rcwnd; /* # times ECN reduced the cwnd */ /* TCP_SIGNATURE related stats */ uint64_t tcps_sig_rcvgoodsig; /* Total matching signature received */ uint64_t tcps_sig_rcvbadsig; /* Total bad signature received */ uint64_t tcps_sig_err_buildsig; /* Failed to make signature */ uint64_t tcps_sig_err_sigopt; /* No signature expected by socket */ uint64_t tcps_sig_err_nosigopt; /* No signature provided by segment */ /* Path MTU Discovery Black Hole Detection related stats */ uint64_t tcps_pmtud_blackhole_activated; /* Black Hole Count */ uint64_t tcps_pmtud_blackhole_activated_min_mss; /* BH at min MSS Count */ uint64_t tcps_pmtud_blackhole_failed; /* Black Hole Failure Count */ uint64_t tcps_tunneled_pkts; /* Packets encap's in UDP received */ uint64_t tcps_tunneled_errs; /* Packets that had errors that were UDP encaped */ /* Dsack related stats */ uint64_t tcps_dsack_count; /* Number of ACKs arriving with DSACKs */ uint64_t tcps_dsack_bytes; /* Number of bytes DSACK'ed no TLP */ uint64_t tcps_dsack_tlp_bytes; /* Number of bytes DSACK'ed due to TLPs */ /* TCPS_TIME_WAIT usage stats */ uint64_t tcps_tw_recycles; /* Times time-wait was recycled. */ uint64_t tcps_tw_resets; /* Times time-wait sent a reset. */ uint64_t tcps_tw_responds; /* Times time-wait sent a valid ack. */ /* Accurate ECN Handshake stats */ uint64_t tcps_ace_nect; /* ACE SYN packet with Non-ECT */ uint64_t tcps_ace_ect1; /* ACE SYN packet with ECT1 */ uint64_t tcps_ace_ect0; /* ACE SYN packet with ECT0 */ uint64_t tcps_ace_ce; /* ACE SYN packet with CE */ /* ECN related stats */ uint64_t tcps_ecn_sndect0; /* ECN Capable Transport */ uint64_t tcps_ecn_sndect1; /* ECN Capable Transport */ /* * BBR and Rack implement TLP's these values count TLP bytes in * two catagories, bytes that were retransmitted and bytes that * were newly transmited. Both types can serve as TLP's but they * are accounted differently. */ uint64_t tcps_tlpresends; /* number of tlp resends */ uint64_t tcps_tlpresend_bytes; /* number of bytes resent by tlp */ uint64_t _pad[4]; /* 4 TBD placeholder for STABLE */ }; #define tcps_rcvmemdrop tcps_rcvreassfull /* compat */ #ifdef _KERNEL #define TI_UNLOCKED 1 #define TI_RLOCKED 2 #include VNET_PCPUSTAT_DECLARE(struct tcpstat, tcpstat); /* tcp statistics */ /* * In-kernel consumers can use these accessor macros directly to update * stats. */ #define TCPSTAT_ADD(name, val) \ VNET_PCPUSTAT_ADD(struct tcpstat, tcpstat, name, (val)) #define TCPSTAT_INC(name) TCPSTAT_ADD(name, 1) /* * Kernel module consumers must use this accessor macro. */ void kmod_tcpstat_add(int statnum, int val); #define KMOD_TCPSTAT_ADD(name, val) \ kmod_tcpstat_add(offsetof(struct tcpstat, name) / sizeof(uint64_t), val) #define KMOD_TCPSTAT_INC(name) KMOD_TCPSTAT_ADD(name, 1) /* * Running TCP connection count by state. */ VNET_DECLARE(counter_u64_t, tcps_states[TCP_NSTATES]); #define V_tcps_states VNET(tcps_states) #define TCPSTATES_INC(state) counter_u64_add(V_tcps_states[state], 1) #define TCPSTATES_DEC(state) counter_u64_add(V_tcps_states[state], -1) /* * TCP specific helper hook point identifiers. */ #define HHOOK_TCP_EST_IN 0 #define HHOOK_TCP_EST_OUT 1 #define HHOOK_TCP_LAST HHOOK_TCP_EST_OUT struct tcp_hhook_data { struct tcpcb *tp; struct tcphdr *th; struct tcpopt *to; uint32_t len; int tso; tcp_seq curack; }; #ifdef TCP_HHOOK void hhook_run_tcp_est_out(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t len, int tso); #endif #endif /* * TCB structure exported to user-land via sysctl(3). * * Fields prefixed with "xt_" are unique to the export structure, and fields * with "t_" or other prefixes match corresponding fields of 'struct tcpcb'. * * Legend: * (s) - used by userland utilities in src * (p) - used by utilities in ports * (3) - is known to be used by third party software not in ports * (n) - no known usage * * Evil hack: declare only if in_pcb.h and sys/socketvar.h have been * included. Not all of our clients do. */ #if defined(_NETINET_IN_PCB_H_) && defined(_SYS_SOCKETVAR_H_) struct xtcpcb { ksize_t xt_len; /* length of this structure */ struct xinpcb xt_inp; char xt_stack[TCP_FUNCTION_NAME_LEN_MAX]; /* (s) */ char xt_logid[TCP_LOG_ID_LEN]; /* (s) */ char xt_cc[TCP_CA_NAME_MAX]; /* (s) */ int64_t spare64[6]; int32_t t_state; /* (s,p) */ uint32_t t_flags; /* (s,p) */ int32_t t_sndzerowin; /* (s) */ int32_t t_sndrexmitpack; /* (s) */ int32_t t_rcvoopack; /* (s) */ int32_t t_rcvtime; /* (s) */ int32_t tt_rexmt; /* (s) */ int32_t tt_persist; /* (s) */ int32_t tt_keep; /* (s) */ int32_t tt_2msl; /* (s) */ int32_t tt_delack; /* (s) */ int32_t t_logstate; /* (3) */ uint32_t t_snd_cwnd; /* (s) */ uint32_t t_snd_ssthresh; /* (s) */ uint32_t t_maxseg; /* (s) */ uint32_t t_rcv_wnd; /* (s) */ uint32_t t_snd_wnd; /* (s) */ uint32_t xt_ecn; /* (s) */ uint32_t t_dsack_bytes; /* (n) */ uint32_t t_dsack_tlp_bytes; /* (n) */ uint32_t t_dsack_pack; /* (n) */ uint16_t xt_encaps_port; /* (s) */ int16_t spare16; int32_t spare32[22]; } __aligned(8); #ifdef _KERNEL void tcp_inptoxtp(const struct inpcb *, struct xtcpcb *); #endif #endif /* * TCP function information (name-to-id mapping, aliases, and refcnt) * exported to user-land via sysctl(3). */ struct tcp_function_info { uint32_t tfi_refcnt; uint8_t tfi_id; char tfi_name[TCP_FUNCTION_NAME_LEN_MAX]; char tfi_alias[TCP_FUNCTION_NAME_LEN_MAX]; }; /* * Identifiers for TCP sysctl nodes */ #define TCPCTL_DO_RFC1323 1 /* use RFC-1323 extensions */ #define TCPCTL_MSSDFLT 3 /* MSS default */ #define TCPCTL_STATS 4 /* statistics */ #define TCPCTL_RTTDFLT 5 /* default RTT estimate */ #define TCPCTL_KEEPIDLE 6 /* keepalive idle timer */ #define TCPCTL_KEEPINTVL 7 /* interval to send keepalives */ #define TCPCTL_SENDSPACE 8 /* send buffer space */ #define TCPCTL_RECVSPACE 9 /* receive buffer space */ #define TCPCTL_KEEPINIT 10 /* timeout for establishing syn */ #define TCPCTL_PCBLIST 11 /* list of all outstanding PCBs */ #define TCPCTL_DELACKTIME 12 /* time before sending delayed ACK */ #define TCPCTL_V6MSSDFLT 13 /* MSS default for IPv6 */ #define TCPCTL_SACK 14 /* Selective Acknowledgement,rfc 2018 */ #define TCPCTL_DROP 15 /* drop tcp connection */ #define TCPCTL_STATES 16 /* connection counts by TCP state */ #ifdef _KERNEL #ifdef SYSCTL_DECL SYSCTL_DECL(_net_inet_tcp); SYSCTL_DECL(_net_inet_tcp_sack); MALLOC_DECLARE(M_TCPLOG); #endif VNET_DECLARE(int, tcp_log_in_vain); #define V_tcp_log_in_vain VNET(tcp_log_in_vain) /* * Global TCP tunables shared between different stacks. * Please keep the list sorted. */ VNET_DECLARE(int, drop_synfin); VNET_DECLARE(int, path_mtu_discovery); VNET_DECLARE(int, tcp_abc_l_var); VNET_DECLARE(int, tcp_autorcvbuf_max); VNET_DECLARE(int, tcp_autosndbuf_inc); VNET_DECLARE(int, tcp_autosndbuf_max); VNET_DECLARE(int, tcp_delack_enabled); VNET_DECLARE(int, tcp_do_autorcvbuf); VNET_DECLARE(int, tcp_do_autosndbuf); VNET_DECLARE(int, tcp_do_ecn); VNET_DECLARE(int, tcp_do_lrd); VNET_DECLARE(int, tcp_do_prr); VNET_DECLARE(int, tcp_do_prr_conservative); VNET_DECLARE(int, tcp_do_newcwv); VNET_DECLARE(int, tcp_do_rfc1323); VNET_DECLARE(int, tcp_tolerate_missing_ts); VNET_DECLARE(int, tcp_do_rfc3042); VNET_DECLARE(int, tcp_do_rfc3390); VNET_DECLARE(int, tcp_do_rfc3465); VNET_DECLARE(int, tcp_do_newsack); VNET_DECLARE(int, tcp_do_sack); VNET_DECLARE(int, tcp_do_tso); VNET_DECLARE(int, tcp_ecn_maxretries); VNET_DECLARE(int, tcp_initcwnd_segments); VNET_DECLARE(int, tcp_insecure_rst); VNET_DECLARE(int, tcp_insecure_syn); VNET_DECLARE(uint32_t, tcp_map_entries_limit); VNET_DECLARE(uint32_t, tcp_map_split_limit); VNET_DECLARE(int, tcp_minmss); VNET_DECLARE(int, tcp_mssdflt); #ifdef STATS VNET_DECLARE(int, tcp_perconn_stats_dflt_tpl); VNET_DECLARE(int, tcp_perconn_stats_enable); #endif /* STATS */ VNET_DECLARE(int, tcp_recvspace); VNET_DECLARE(int, tcp_retries); VNET_DECLARE(int, tcp_sack_globalholes); VNET_DECLARE(int, tcp_sack_globalmaxholes); VNET_DECLARE(int, tcp_sack_maxholes); VNET_DECLARE(int, tcp_sc_rst_sock_fail); VNET_DECLARE(int, tcp_sendspace); VNET_DECLARE(int, tcp_udp_tunneling_overhead); VNET_DECLARE(int, tcp_udp_tunneling_port); VNET_DECLARE(struct inpcbinfo, tcbinfo); #define V_tcp_do_lrd VNET(tcp_do_lrd) #define V_tcp_do_prr VNET(tcp_do_prr) -#define V_tcp_do_prr_conservative VNET(tcp_do_prr_conservative) #define V_tcp_do_newcwv VNET(tcp_do_newcwv) #define V_drop_synfin VNET(drop_synfin) #define V_path_mtu_discovery VNET(path_mtu_discovery) #define V_tcbinfo VNET(tcbinfo) #define V_tcp_abc_l_var VNET(tcp_abc_l_var) #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) #define V_tcp_autosndbuf_inc VNET(tcp_autosndbuf_inc) #define V_tcp_autosndbuf_max VNET(tcp_autosndbuf_max) #define V_tcp_delack_enabled VNET(tcp_delack_enabled) #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) #define V_tcp_do_autosndbuf VNET(tcp_do_autosndbuf) #define V_tcp_do_ecn VNET(tcp_do_ecn) #define V_tcp_do_rfc1323 VNET(tcp_do_rfc1323) #define V_tcp_tolerate_missing_ts VNET(tcp_tolerate_missing_ts) #define V_tcp_ts_offset_per_conn VNET(tcp_ts_offset_per_conn) #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) #define V_tcp_do_rfc3390 VNET(tcp_do_rfc3390) #define V_tcp_do_rfc3465 VNET(tcp_do_rfc3465) #define V_tcp_do_newsack VNET(tcp_do_newsack) #define V_tcp_do_sack VNET(tcp_do_sack) #define V_tcp_do_tso VNET(tcp_do_tso) #define V_tcp_ecn_maxretries VNET(tcp_ecn_maxretries) #define V_tcp_initcwnd_segments VNET(tcp_initcwnd_segments) #define V_tcp_insecure_rst VNET(tcp_insecure_rst) #define V_tcp_insecure_syn VNET(tcp_insecure_syn) #define V_tcp_map_entries_limit VNET(tcp_map_entries_limit) #define V_tcp_map_split_limit VNET(tcp_map_split_limit) #define V_tcp_minmss VNET(tcp_minmss) #define V_tcp_mssdflt VNET(tcp_mssdflt) #ifdef STATS #define V_tcp_perconn_stats_dflt_tpl VNET(tcp_perconn_stats_dflt_tpl) #define V_tcp_perconn_stats_enable VNET(tcp_perconn_stats_enable) #endif /* STATS */ #define V_tcp_recvspace VNET(tcp_recvspace) #define V_tcp_retries VNET(tcp_retries) #define V_tcp_sack_globalholes VNET(tcp_sack_globalholes) #define V_tcp_sack_globalmaxholes VNET(tcp_sack_globalmaxholes) #define V_tcp_sack_maxholes VNET(tcp_sack_maxholes) #define V_tcp_sc_rst_sock_fail VNET(tcp_sc_rst_sock_fail) #define V_tcp_sendspace VNET(tcp_sendspace) #define V_tcp_udp_tunneling_overhead VNET(tcp_udp_tunneling_overhead) #define V_tcp_udp_tunneling_port VNET(tcp_udp_tunneling_port) #ifdef TCP_HHOOK VNET_DECLARE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST + 1]); #define V_tcp_hhh VNET(tcp_hhh) #endif int tcp_addoptions(struct tcpopt *, u_char *); struct tcpcb * tcp_close(struct tcpcb *); void tcp_discardcb(struct tcpcb *); void tcp_twstart(struct tcpcb *); int tcp_ctloutput(struct socket *, struct sockopt *); void tcp_fini(void *); char *tcp_log_addrs(struct in_conninfo *, struct tcphdr *, const void *, const void *); char *tcp_log_vain(struct in_conninfo *, struct tcphdr *, const void *, const void *); int tcp_reass(struct tcpcb *, struct tcphdr *, tcp_seq *, int *, struct mbuf *); void tcp_reass_global_init(void); void tcp_reass_flush(struct tcpcb *); void tcp_dooptions(struct tcpopt *, u_char *, int, int); void tcp_dropwithreset(struct mbuf *, struct tcphdr *, struct tcpcb *, int, int); void tcp_pulloutofband(struct socket *, struct tcphdr *, struct mbuf *, int); void tcp_xmit_timer(struct tcpcb *, int); void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); void cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, uint16_t type); void cc_conn_init(struct tcpcb *tp); void cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); void cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos); void cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos); void cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type); #ifdef TCP_HHOOK void hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to); #endif int tcp_input(struct mbuf **, int *, int); int tcp_autorcvbuf(struct mbuf *, struct tcphdr *, struct socket *, struct tcpcb *, int); int tcp_input_with_port(struct mbuf **, int *, int, uint16_t); void tcp_do_segment(struct tcpcb *, struct mbuf *, struct tcphdr *, int, int, uint8_t); int register_tcp_functions(struct tcp_function_block *blk, int wait); int register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, const char *names[], int *num_names); int register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, int wait); int deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, bool force); struct tcp_function_block *find_and_ref_tcp_functions(struct tcp_function_set *fs); int find_tcp_function_alias(struct tcp_function_block *blk, struct tcp_function_set *fs); uint32_t tcp_get_srtt(struct tcpcb *tp, int granularity); void tcp_switch_back_to_default(struct tcpcb *tp); struct tcp_function_block * find_and_ref_tcp_fb(struct tcp_function_block *fs); int tcp_default_ctloutput(struct tcpcb *tp, struct sockopt *sopt); int tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt); void tcp_log_socket_option(struct tcpcb *tp, uint32_t option_num, uint32_t option_val, int err); extern counter_u64_t tcp_inp_lro_direct_queue; extern counter_u64_t tcp_inp_lro_wokeup_queue; extern counter_u64_t tcp_inp_lro_compressed; extern counter_u64_t tcp_inp_lro_locks_taken; extern counter_u64_t tcp_extra_mbuf; extern counter_u64_t tcp_would_have_but; extern counter_u64_t tcp_comp_total; extern counter_u64_t tcp_uncomp_total; extern counter_u64_t tcp_bad_csums; #ifdef TCP_SAD_DETECTION /* Various SACK attack thresholds */ extern int32_t tcp_force_detection; extern int32_t tcp_sad_limit; extern int32_t tcp_sack_to_ack_thresh; extern int32_t tcp_sack_to_move_thresh; extern int32_t tcp_restoral_thresh; extern int32_t tcp_sad_decay_val; extern int32_t tcp_sad_pacing_interval; extern int32_t tcp_sad_low_pps; extern int32_t tcp_map_minimum; extern int32_t tcp_attack_on_turns_on_logging; #endif extern uint32_t tcp_ack_war_time_window; extern uint32_t tcp_ack_war_cnt; uint32_t tcp_maxmtu(struct in_conninfo *, struct tcp_ifcap *); uint32_t tcp_maxmtu6(struct in_conninfo *, struct tcp_ifcap *); void tcp6_use_min_mtu(struct tcpcb *); u_int tcp_maxseg(const struct tcpcb *); u_int tcp_fixed_maxseg(const struct tcpcb *); void tcp_mss_update(struct tcpcb *, int, int, struct hc_metrics_lite *, struct tcp_ifcap *); void tcp_mss(struct tcpcb *, int); int tcp_mssopt(struct in_conninfo *); struct tcpcb * tcp_newtcpcb(struct inpcb *); int tcp_default_output(struct tcpcb *); void tcp_state_change(struct tcpcb *, int); void tcp_respond(struct tcpcb *, void *, struct tcphdr *, struct mbuf *, tcp_seq, tcp_seq, uint16_t); bool tcp_twcheck(struct inpcb *, struct tcpopt *, struct tcphdr *, struct mbuf *, int); void tcp_setpersist(struct tcpcb *); void tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp); struct tcptemp * tcpip_maketemplate(struct inpcb *); void tcpip_fillheaders(struct inpcb *, uint16_t, void *, void *); void tcp_timer_activate(struct tcpcb *, tt_which, u_int); bool tcp_timer_active(struct tcpcb *, tt_which); void tcp_timer_stop(struct tcpcb *); int inp_to_cpuid(struct inpcb *inp); /* * All tcp_hc_* functions are IPv4 and IPv6 (via in_conninfo) */ void tcp_hc_init(void); #ifdef VIMAGE void tcp_hc_destroy(void); #endif void tcp_hc_get(struct in_conninfo *, struct hc_metrics_lite *); uint32_t tcp_hc_getmtu(struct in_conninfo *); void tcp_hc_updatemtu(struct in_conninfo *, uint32_t); void tcp_hc_update(struct in_conninfo *, struct hc_metrics_lite *); void cc_after_idle(struct tcpcb *tp); extern struct protosw tcp_protosw; /* shared for TOE */ extern struct protosw tcp6_protosw; /* shared for TOE */ uint32_t tcp_new_ts_offset(struct in_conninfo *); tcp_seq tcp_new_isn(struct in_conninfo *); -int tcp_sack_doack(struct tcpcb *, struct tcpopt *, tcp_seq); +sackstatus_t + tcp_sack_doack(struct tcpcb *, struct tcpopt *, tcp_seq); int tcp_dsack_block_exists(struct tcpcb *); void tcp_update_dsack_list(struct tcpcb *, tcp_seq, tcp_seq); void tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_laststart, tcp_seq rcv_lastend); void tcp_clean_dsack_blocks(struct tcpcb *tp); void tcp_clean_sackreport(struct tcpcb *tp); void tcp_sack_adjust(struct tcpcb *tp); struct sackhole *tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt); -void tcp_do_prr_ack(struct tcpcb *, struct tcphdr *, struct tcpopt *); +void tcp_do_prr_ack(struct tcpcb *, struct tcphdr *, struct tcpopt *, sackstatus_t); void tcp_lost_retransmission(struct tcpcb *, struct tcphdr *); void tcp_sack_partialack(struct tcpcb *, struct tcphdr *); void tcp_free_sackholes(struct tcpcb *tp); void tcp_sack_lost_retransmission(struct tcpcb *, struct tcphdr *); int tcp_newreno(struct tcpcb *, struct tcphdr *); int tcp_compute_pipe(struct tcpcb *); uint32_t tcp_compute_initwnd(uint32_t); void tcp_sndbuf_autoscale(struct tcpcb *, struct socket *, uint32_t); int tcp_stats_sample_rollthedice(struct tcpcb *tp, void *seed_bytes, size_t seed_len); int tcp_can_enable_pacing(void); void tcp_decrement_paced_conn(void); void tcp_change_time_units(struct tcpcb *, int); void tcp_handle_orphaned_packets(struct tcpcb *); struct mbuf * tcp_m_copym(struct mbuf *m, int32_t off0, int32_t *plen, int32_t seglimit, int32_t segsize, struct sockbuf *sb, bool hw_tls); int tcp_stats_init(void); void tcp_log_end_status(struct tcpcb *tp, uint8_t status); #ifdef TCP_REQUEST_TRK void tcp_req_free_a_slot(struct tcpcb *tp, struct tcp_sendfile_track *ent); struct tcp_sendfile_track * tcp_req_find_a_req_that_is_completed_by(struct tcpcb *tp, tcp_seq th_ack, int *ip); int tcp_req_check_for_comp(struct tcpcb *tp, tcp_seq ack_point); int tcp_req_is_entry_comp(struct tcpcb *tp, struct tcp_sendfile_track *ent, tcp_seq ack_point); struct tcp_sendfile_track * tcp_req_find_req_for_seq(struct tcpcb *tp, tcp_seq seq); void tcp_req_log_req_info(struct tcpcb *tp, struct tcp_sendfile_track *req, uint16_t slot, uint8_t val, uint64_t offset, uint64_t nbytes); uint32_t tcp_estimate_tls_overhead(struct socket *so, uint64_t tls_usr_bytes); void tcp_req_alloc_req(struct tcpcb *tp, union tcp_log_userdata *user, uint64_t ts); struct tcp_sendfile_track * tcp_req_alloc_req_full(struct tcpcb *tp, struct tcp_snd_req *req, uint64_t ts, int rec_dups); #endif #ifdef TCP_ACCOUNTING int tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss); #endif static inline void tcp_lro_features_off(struct tcpcb *tp) { tp->t_flags2 &= ~(TF2_SUPPORTS_MBUFQ| TF2_MBUF_QUEUE_READY| TF2_DONT_SACK_QUEUE| TF2_MBUF_ACKCMP| TF2_MBUF_L_ACKS); } static inline void tcp_fields_to_host(struct tcphdr *th) { th->th_seq = ntohl(th->th_seq); th->th_ack = ntohl(th->th_ack); th->th_win = ntohs(th->th_win); th->th_urp = ntohs(th->th_urp); } static inline void tcp_fields_to_net(struct tcphdr *th) { th->th_seq = htonl(th->th_seq); th->th_ack = htonl(th->th_ack); th->th_win = htons(th->th_win); th->th_urp = htons(th->th_urp); } static inline uint16_t tcp_get_flags(const struct tcphdr *th) { return (((uint16_t)th->th_x2 << 8) | th->th_flags); } static inline void tcp_set_flags(struct tcphdr *th, uint16_t flags) { th->th_x2 = (flags >> 8) & 0x0f; th->th_flags = flags & 0xff; } static inline void tcp_account_for_send(struct tcpcb *tp, uint32_t len, uint8_t is_rxt, uint8_t is_tlp, bool hw_tls) { if (is_tlp) { tp->t_sndtlppack++; tp->t_sndtlpbyte += len; } /* To get total bytes sent you must add t_snd_rxt_bytes to t_sndbytes */ if (is_rxt) tp->t_snd_rxt_bytes += len; else tp->t_sndbytes += len; #ifdef KERN_TLS if (hw_tls && is_rxt && len != 0) { uint64_t rexmit_percent = (1000ULL * tp->t_snd_rxt_bytes) / (10ULL * (tp->t_snd_rxt_bytes + tp->t_sndbytes)); if (rexmit_percent > ktls_ifnet_max_rexmit_pct) ktls_disable_ifnet(tp); } #endif } #endif /* _KERNEL */ #endif /* _NETINET_TCP_VAR_H_ */