|  | // Copyright 2015 The Go Authors. All rights reserved. | 
|  | // Use of this source code is governed by a BSD-style | 
|  | // license that can be found in the LICENSE file. | 
|  |  | 
|  | package net | 
|  |  | 
|  | import ( | 
|  | "runtime" | 
|  | "syscall" | 
|  | "time" | 
|  | ) | 
|  |  | 
|  | func setKeepAlivePeriod(fd *netFD, d time.Duration) error { | 
|  | // The kernel expects milliseconds so round to next highest | 
|  | // millisecond. | 
|  | msecs := int(roundDurationUp(d, time.Millisecond)) | 
|  |  | 
|  | // Normally we'd do | 
|  | //	syscall.SetsockoptInt(fd.sysfd, syscall.IPPROTO_TCP, syscall.TCP_KEEPINTVL, secs) | 
|  | // here, but we can't because Solaris does not have TCP_KEEPINTVL. | 
|  | // Solaris has TCP_KEEPALIVE_ABORT_THRESHOLD, but it's not the same | 
|  | // thing, it refers to the total time until aborting (not between | 
|  | // probes), and it uses an exponential backoff algorithm instead of | 
|  | // waiting the same time between probes. We can't hope for the best | 
|  | // and do it anyway, like on Darwin, because Solaris might eventually | 
|  | // allocate a constant with a different meaning for the value of | 
|  | // TCP_KEEPINTVL on illumos. | 
|  |  | 
|  | err := fd.pfd.SetsockoptInt(syscall.IPPROTO_TCP, syscall.TCP_KEEPALIVE_THRESHOLD, msecs) | 
|  | runtime.KeepAlive(fd) | 
|  | return wrapSyscallError("setsockopt", err) | 
|  | } |