TCP retransmits an unacknowledged packet up to tcp_retries2 sysctl setting times
(defaults to 15) using an exponential backoff timeout for which each
retransmission timeout is between TCP_RTO_MIN (200 ms) and TCP_RTO_MAX (120
seconds). Once the 15th retry expires (by default), the TCP stack will notify
the layers above (ie. app) of a broken connection.

The value of TCP_RTO_MIN and TCP_RTO_MAX is hardcoded in the Linux kernel and
defined by the following constants:

define TCP_RTO_MAX ((unsigned)(120*HZ)) #define TCP_RTO_MIN ((unsigned)(HZ/5))

Linux 2.6+ uses HZ of 1000ms, so TCP_RTO_MIN is ~200 ms and TCP_RTO_MAX is ~120
seconds. Given a default value of tcp_retries set to 15, it means that it takes
924.6 seconds before a broken network link is notified to the upper layer (ie.
application), since the connection is detected as broken when the last (15th)
retry expires.

The tcp_retries2 sysctl can be tuned via /proc/sys/net/ipv4/tcp_retries2 or the
sysctl net.ipv4.tcp_retries2.

From the Linux kernel doc tcp_retries2 – INTEGER This value influences the
timeout of an alive TCP connection, when RTO retransmissions remain
unacknowledged. Given a value of N, a hypothetical TCP connection following
exponential backoff with an initial RTO of TCP_RTO_MIN would retransmit N
times before killing the connection at the (N+1)th RTO.

The default value of 15 yields a hypothetical timeout of 924.6 seconds and
is a lower bound for the effective timeout. TCP will effectively time out at
the first RTO which exceeds the hypothetical timeout.

RFC 1122 recommends at least 100 seconds for the timeout, which corresponds
to a value of at least 8.

Source: https://pracucci.com/linux-tcp-rto-min-max-and-tcp-retries2.html