On a server I have a public key auth only for root account. Is there any point of logging in with a different account?
It’s rarely a good idea to log in as root, doubly so if it’s a system with sensitive data or services that could easily be disrupted accidentally. And even more important if multiple users log in. How will you know who broke things to teach them if they don’t log in first. The only time I log in to any system as root other than a test system is when I need to sftp to access files or some other system that doesn’t have a way to elevate permissions.
It’s a bad practice to log in as root even for administrative tasks. You need to run numerous commands, some of hem can be potentially dangerous while not requiring root privileges. So normally you have an
adminuser in thesudo/wheelgroup and need to login to this account. Also, this adds some protection in case your key has leaked.Is there any point of logging in with a different account?
When you edit & save a file as root, root takes ownership of that file. I personally don’t like having to run chmod or chown every time I make minor changes to something.
No, that’s not correct. If you create a new file as root, it will own that file. But editing an existing file doesn’t change the owner or group of that file.
A door with the best lock possible is still not as secure as no door at all
Yes it’s always better to login with a user and sudo so your commands are logged also having disable passwords for ssh but still using passwords for sudo gives you the best protection
Also double check that sudo is the right command, by doing
which sudo. Something I just learned to be paranoid of in this thread.Unless
whichis also compromised, my god…which sudowill check$PATHdirectories and return the first match, true. however when you typesudoand hit enter your shell will look for aliases and shell functions before searching$PATH.to see how your shell will execute ‘sudo’, say
type sudo(zsh/bash). to skip aliases/functions/builtins saycommand sudomeh nvm none of these work if your shell is compromised. you’re sending bytes to the attacker at that point. they can make you believe anything
Maybe if you escaped the command like
\\type sudo?no, if the attacker can change files in your account, they can read every byte you type in and respond with anything, including pretending to be a normal shell. im not sure how to prevent ssh from running commands in your shell
Sudo also allows for granular permissions of which commands are allowed and which aren’t.
It’s another slice of Swiss cheese. If the user has a strong enough password or other authentication method through PAM, it might stop or hinder an attacker who might only have a compromised private key, for example. If multiple users have access to the same server and one of them is compromised, the account can be disabled without completely crippling the system.
Using
sudocan also help you avoid mistakes (like accidentally rebooting a production server) by restricting which commands are available to the user.It’s just another way of minimizing your attack surface. It’s pretty much the same as hiding behind a barrier when being shot at, you stick yourself out as little as possible.
In the same way it also helps to change your SSH port to somewhere in the high numbers like 38265. This is anecdotal of course, but the amount of attacks on SSH went down by literally 99% by just changing the port like that
Then you accept only keys, you lock down root (so the username must be guessed as well) and yeah, you’re safe.
This is anecdotal
Not just anecdotal. The default SSH port gets hit by ridiculous numbers of bots because a lot of people don’t bother to change it. This will be true no matter what machine you’re on. Hell, your desktop at home has probably been scanned quite a few times even if all you do is watch porn on it
I never login with the root account. Not even on the console. You don’t want everything you do running as root unless it is required. Otherwise it is much easier for a little mistake to become a big mess.
Well, with root enabled, the SSH server at least need to verify the key, no? It’s wasting CPU power albeit tiny amount.
Audit trails
Zero-day exploits are security holes that exist and are used by bad actors, but aren’t yet known to you, or anyone capable of closing the hole. The clock to patch the hole doesn’t start running until the exploit is known: it stands at zero days until the good guys know it exists.
What zero-day exploits exist for ssh?
By definition, you don’t know. So, you block root login, and hope the bad actor doesn’t also know a zero-day for sudo.
- Swiss cheese slices: make them holes too tight.
- When you run everything as root, if you fuck your shit, your shit’s fucked.
“Best practices” tend to come from other people’s whoopsies. But it’s always good to question things, too.
Nope, not really. The only reason ppl recommend it is, because “you have then to guess the username too”. Which is just not relevant if you use strong authentication method like keys or only strong passwords.
That is absolutely not the reason ANYONE recommends it, unless you are a complete noob and entirely unfamiliar with computer security at all, and are just pulling assumptions out of your ass. Don’t fucking do that, don’t post with confidence when you’re just making shit up because you think you know better. Because you don’t.
If there is a vulnerability in SSH (and it’s happened before), attackers could use that to get into root directly, quickly, and easily. It’s an instant own.
If root login is disabled, it’s way less likely that whatever bug it is ALSO allows them to bypass root login being disabled. Now they have to yeah, find a user account, compromise that, try to key log or session hijack or whatever they set up, be successful, and elevate to root. That’s WAY more work, way more time to detect, to install patches.
If the effort is higher, then this kind of attack isn’t going to be used to own small fry servers; it’s only be worth it for bigger targets, even if they’re more well protected.
If you leave root enabled, you’re already burnt. You’re already a bot in the DDoS network.
And why? You couldn’t be bothered to type one extra command in your terminal? One extra word at the start of each command?
Sorry bitch, eat your fucking vegetables
Don’t quit your day job.
Most comments here suggest 3 things
- least privilege: Which is ok, but on a Server any modification you do requires root anyway, there is usually very little benefit
- Additional protection through required sudo password: This is for example easily circumvented by modifying the bashrc or similar with an sudo alias to get the password
- Multiuser & audittrails: yes this is a valid point, on a system that is modified or administered by multiple ppl there are various reasons lime access logging and UAC for that
An actual person from the pen testing world: https://youtu.be/fKuqYQdqRIs
If ssh has a security issue and you permit root logins then hostiles likely have an easier time getting access to root on the machine than if they only get access to your user account—then they need multiple exploits.
Generally you also want to be root as little as possible. Hence sudo, run0, etc.
Its a concept called defense in depth. Without root login now you require the key AND sudo password.
Also, outside of self hosted you will have multiple people logging in. You want them to log in with their own users for logging and permission management.
The sudo password can be easily extracted by modifying the bashrc.
And who is going to edit your .bashrc?
The attacker that is currently with user privileges on the server?
that’s why root owns my .bash* stuff
I don’t think that actually works; the attacker could just remove .bashrc and create a new file with the same name.
you’re right. that’s something i wanted to look into. guess setfacl would do the trick?
“chattr +i” is what I use to make things immutable
If the .bashrc is immutable, the attacker can’t remove it.
That’s how it works.The home directory would need to be immutable, not bashrc.
How did the attacker gain your user’s privileges? Malware-infected user installation? A vulnerability in genuine software running as your user? In most scenarios these things only become worse when running as root instead.
The scenario OC stated is that if the attacker has access to the user on the server then the attacker would still need the sudo password in order to get root privileges, contrary to direct root login where the attack has direct access to root privileges.
So, now i am looking into this scenario where the attack is on the server with the user privileges: the attacker now modifies for example the bashrc to alias sudo to extract the password once the user runs sudo.
So the sudo password does not have any meaningful protection, other then maybe adding a time variable which is when the user accesses the server and runs sudo
Simple solution is to not use sudo.
Sorta like Slackware’s default.And what do you suggest to use otherwise to maintain a server? I am not aware of a solution that would help here? As an attacker you could easily alias any command or even start a modified shell that logs ever keystroke and simulates the default bash/zsh or whatever.
Nah just set up PAM to use TOTP or a third party MFA service to send a push to your phone for sudo privs.
This was downvoted, but is a good question.
If your account is compromised, the shell init code could be modified to install a keylogger to discover the root password. That’s correct.
Still, that capture doesn’t happen instantly. On a personal server, it could be months until the owner logs in next. On a corporate machines, there may be daily scans for signs of intrusion, malware, etc. Either way, the attacker has been slowed down and there is a chance they won’t succeed in a timeframe that’s useful to them.
It’s perhaps like a locking a bike: with right tool and enough time, a thief can steal the bike. Sometimes slowing them down sufficiently is enough to win.
Doesn’t even have to be the key necessarily. Could get in via some exploit first. Either way taking over the machine became a 2-step process.
you would need 2 different exploits for 2 different types of attack though.
its always good to have an extra layer of “oh shit i need another exploit”. unless your threat modelling includes nation-states, that is.
Unless your threat modelling includes nation-states
At which point you should have a handful of extra layers
