The syncache sysctl(8) MIB is used to control the TCP SYN caching in the system, which is intended to handle SYN flood Denial of Service attacks.
When a TCP SYN segment is received on a port corresponding to a listen socket, an entry is made in the syncache, and a SYN,ACK segment is returned to the peer. The syncache entry holds the TCP options from the initial SYN, enough state to perform a SYN,ACK retransmission, and takes up less space than a TCP control block endpoint. An incoming segment which contains an ACK for the SYN,ACK and matches a syncache entry will cause the system to create a TCP control block with the options stored in the syncache entry, which is then released.
The syncache protects the system from SYN flood DoS attacks by minimizing the amount of state kept on the server, and by limiting the overall size of the syncache.
Syncookies provides a way to virtually expand the size of the syncache by keeping state regarding the initial SYN in the network. Enabling syncookies sends a cryptographic value in the SYN,ACK reply to the client machine, which is then returned in the clients ACK. If the corresponding entry is not found in the syncache, but the value passes specific security checks, the connection will be accepted. This is only used if the syncache is unable to handle the volume of incoming connections, and a prior entry has been evicted from the cache.
Syncookies have a certain number of disadvantages that a paranoid administrator may wish to take note of. Since the TCP options from the initial SYN are not saved, they are not applied to the connection, precluding use of features like window scale, timestamps, or exact MSS sizing. As the returning ACK establishes the connection, it may be possible for an attacker to ACK flood a machine in an attempt to create a connection. While steps have been taken to mitigate this risk, this may provide a way to bypass firewalls which filter incoming segments with the SYN bit set.
To disable the syncache and run only with syncookies, set net.inet.tcp.syncookies_only to 1.
hashsize Size of the syncache hash table, must be a power of 2. Read-only, tunable via loader(8). bucketlimit Limit on the number of entries permitted in each bucket of the hash table. This should be left at a low value to minimize search time. Read-only, tunable via loader(8). cachelimit Limit on the total number of entries in the syncache. Defaults to ( hashsize × bucketlimit), may be set lower to minimize memory consumption. Read-only, tunable via loader(8). rexmtlimit Maximum number of times a SYN,ACK is retransmitted before being discarded. The default of 3 retransmits corresponds to a 45 second timeout, this value may be increased depending on the RTT to client machines. Tunable via sysctl(3). count Number of entries present in the syncache (read-only).
Statistics on the performance of the syncache may be obtained via netstat(1), which provides the following counts:
syncache entries added Entries successfully inserted in the syncache. retransmitted SYN,ACK retransmissions due to a timeout expiring. dupsyn Incoming SYN segment matching an existing entry. dropped SYNs dropped because SYN,ACK could not be sent. completed Successfully completed connections. bucket overflow Entries dropped for exceeding per-bucket size. cache overflow Entries dropped for exceeding overall cache size. reset RST segment received. stale Entries dropped due to maximum retransmissions or listen socket disappearance. aborted New socket allocation failures. badack Entries dropped due to bad ACK reply. unreach Entries dropped due to ICMP unreachable messages. zone failures Failures to allocate new syncache entry. cookies received Connections created from segment containing ACK.
The existing syncache implementation first appeared in
.Fx 4.5 . The original concept of a syncache originally appeared in
.Bsx , and was later modified by
.Nx , then further extended here.
The syncache code and manual page were written by
.An Jonathan Lemon Aq jlemon@FreeBSD.org .