Previously, read() might have returned a length equal to the buffer
length, and then the subsequent store to buf[len] would write a
zero-byte one byte past the end.  This doesn't seem likely to be
a security issue, but there's some chance it could result in
pg_standby misbehaving.

Spotted by Coverity; patch by Michael Paquier, reviewed by me.