Ssh – Equivalent of scp -l bandwidth_cap for .ssh/config

bandwidthconfigurationscpssh

Short form:

You can limit the bandwidth the scp uses with the -l switch, you pass a number that's in kbits/sec.

I'd rather set this in my .ssh/config file for certain names machines.

What's the equivalent named setting for -l? I haven't been able to find it.

Followup question:

Generally, not sure how to map back and forth between ssh command line options and config names, short of doing Google searches or manually comparing man pages on a case by case basis. Is there a table that directly equates the two?

Longer form of first question, with context:

I've started using ssh config quite a bit, especially now that I need to go through a proxy and do lots of port mappings. I even define the same machine more than once depending on what type of tunneling I need.

However, when uploading a large file, it's difficult to do anything else on my machine. Even though I have more download bandwidth than up, I think that scp saturates the link so even my small requests can't reach the Internet.

There's a fix for this, using the -l bandwidth command line switch for scp.

scp -l 1000 bigfile.zip titan:

I'd like to use this in my config instead, so I'd create an additional named entry called "titan-upload" and I'd use that as the target whenever I upload.

So instead of:

scp bigfile.zip titan:

I'd say:

scp bigfile.zip titan-upload:

Or even set different caps depending on where I am:

scp bigfile.zip titan-upload-from-work:
  vs.
scp bigfile.zip titan-upload-from-home:

I'm generally on Mac and Linux.

Best Answer

Alas, as was mentioned, there doesn't see to be a config option to limit bandwidth. (I checked source code!)

Some possible solutions are to use an alias for scp, or perhaps a function. Bash is typically the default shell on both mac & linux, so this could work:

alias scp='scp -l 1000 '
     -or-
alias scp-throttle='scp -l 1000 '

(note trailing space inside quotes!¹) This would cause EVERY scp command you use to throttle bandwidth. Considering your situation, perhaps the best solution overall.

The second might be a good choice, since you could use scp for 'normal' transfers, and scp-throttle for slower transfers.

Or a function, with a bit more brains:

function scp { if [ "$*" =~ "-upload" ]; then command scp -l 1000 "$@"; else command scp  "$@"; fi; }

Basically, if we find '-upload' anywhere in the arguments, we perform the transfer with the bw limit, otherwise, a normal transfer occurs.

This would allow you to continue using your multiple names/aliases to denote actions.

scp aaa titan: - would upload normally
scp aaa titan-upload: - would throttle
scp titan:aaa . - normal
scp titan-upload-from-home:aaa . - throttled
scp a-file-to-upload titan: - oops, throttled, not intentional!

EDIT:

¹ - The trailing space INSIDE the alias allows further alias expansion after the aliased command. VERY helpful/useful. Bash Man Page, __ALIASES__ section

Related Solutions

Ssh – recursively print all current remote shells

You could write a script that just walks up the process tree printing out unique usernames/pids/executables/whatever as it went up the list, stopping when it finally left the terminal. sshd usually starts listening on a TTY after it has already done the setuid() to the user so the earliest process connected to the TTY will be running as the user you logged in with.

Here's a Linux-specific Bash script version of the concept.

script output:

Simple sudo to root:

[root@ditirlns04 ~]# ./test.sh
Current User: 0
Previous UID: 504

UID 504 is how I logged into the box so it worked:

[root@ditirlns04 ~]# logname
jadavis6
[root@ditirlns04 ~]# id jadavis6
uid=504(jadavis6) gid=504(jadavis6) groups=504(jadavis6),16777216(enterprise admins),16777217(ncatsys),16777218(sms admins),16777219(domain admins),16777220(aggieanywhere),16777221(group policy creator owners),16777222(tlh-test),16777223(dba admins),16777224(domain controllers),16777225(cupsadmin),16777226(linuxusers),16777227(da admins),16777228(webdev),16777229(schema admins),16777230,16777231(changeauditor operators - ncat),16777232(configmgr remote control users),16777233(changeauditor administrators - ncat),16777234(telnetclients),16777235(denied rodc password replication group)

Same as above, only going through an intermediary user (i.e: logged in 504, sudo'd to root, sudo'd to regular UID of 501, then back to root):

[root@ditirlns04 ~]# ./test.sh
Current User: 0
Previous UID: 501
Previous UID: 0
Previous UID: 504
[root@ditirlns04 ~]#

So each UID transition is caught in reverse order that it happened in. If you need it in the same order, just put the script into a function where you tac the output:

[root@ditirlns04 ~]# ./test.sh | tac
Previous UID: 504
Previous UID: 0
Previous UID: 501
Current User: 0
[root@ditirlns04 ~]#

script body:

#!/bin/bash

currentPID=$$
currentUser=$(id -u)
currentTTY=$(awk '{print $7}' /proc/$$/stat)

echo "Current User: $currentUser"

while /bin/true; do

  parentPID=$(awk '{print $4}' /proc/$currentPID/stat)
  parentTTY=$(awk '{print $7}' /proc/$currentPID/stat)

  if [ $parentTTY -ne $currentTTY ]; then
     break
  fi

  parentUser=$(awk '/^Uid:/ {print $2}' /proc/$currentPID/status)

  if [[ "x$parentUser" != "x$currentUser" ]]; then
     echo "Previous UID: $parentUser"
  fi

  currentPID=$parentPID
  currentUser=$parentUser

done

Basically, it just makes use of the per-pid stat and status files. You could make more platform independent versions wrapping around ps output, but I don't trust parsing command output when the datum is readily available in a format meant to be parsed, so I went the procfs route.

Basically, any time it finds a new UID it prints it out. They're guaranteed to be in the same order the user sudo'd so it should effectively cover your entire login session on that machine. You should be able to easily modify the above to print PID's or command names instead if you want that, but usually people thinking in this area are more concerned with who they'll be running as rather than the shell program that will be running.

Bash SSH Autocomplete – Autocomplete Server Names for SSH and SCP

Found it!!

It seems that in Ubuntu the entries in ~/.ssh/known_hosts are hashed, so SSH completion cannot read them. This is a feature, not a bug. Even by adding HashKnownHosts no to ~/.ssh/config and /etc/ssh/ssh_config I was unable to prevent the host hashing.

However, the hosts that I am interested in are also found in ~/.ssh/config. Here is a script for Bash Completion that reads the entries from that file:

_ssh() 
{
    local cur prev opts
    COMPREPLY=()
    cur="${COMP_WORDS[COMP_CWORD]}"
    prev="${COMP_WORDS[COMP_CWORD-1]}"
    opts=$(grep '^Host' ~/.ssh/config ~/.ssh/config.d/* 2>/dev/null | grep -v '[?*]' | cut -d ' ' -f 2-)

    COMPREPLY=( $(compgen -W "$opts" -- ${cur}) )
    return 0
}
complete -F _ssh ssh

Put that script in /etc/bash_completion.d/ssh and then source it with the following command:

$ . /etc/bash_completion.d/ssh

I found this guide invaluable and I would not have been able to script this without it. Thank you Steve Kemp for writing that terrific guide!

Best Answer

Related Solutions

Ssh – recursively print all current remote shells

Bash SSH Autocomplete – Autocomplete Server Names for SSH and SCP

Related Question