i am trying to write a custom (GUI) program installer which needs to run as a specific (system) user (without a home directory) - let's call him "installer". Things i have tried or pondered to achieve this in a couple of different ways:
launching the GUI installer app as user "installer" through (gk)su(do) (-u). the problem is that a. these commands are not available on every
distribution or do not behave the same everywhere thanks to internal aliases, and b. the x server does not allow me to connect as a different user without some potentially insecure "xhost +" modification (the old "cannot connect to X server :0.0" problem).
avoiding the X server for the installer routine and create a separate (CLI) binary for the chmod/file copy process, while the GUI runs in current user mode. the problem is that i have to give the password to the (sudo -S) command in clear text, which shows up in the logs, so not a good idea either.
several command combinations of export display, .Xauthority, xhost and anything i could find on stackoverflow, without much success.
so, what do i actually need?
a distro-independent, secure way of launching a GUI application as another (non-root, but "/home"-less system) user
an elegant way of asking the current user for the password of user "installer"
a solution that does not require the current user to enter the root password first, or install the application as root.
i'd be thankful for any thoughts on the matter.
[edit] to clarify, i do not rally want to install new programs into the established linux file structure. The whole setup is this:
I have a main program that will be installed by the root. this program can be extended by modules.
these modules will be installed into a custom folder (let's call it "/progmodules") which is owned by the "installer" user. the goal is to have authorized users to be able to install new modules, without giving them full root access.
any users who want to install new modules should be able to run the installer GUI app, enter the "installer" user password, and then have the files transferred.
Related
What exactly is --user-data-dir specifiying?
From --help parameter:
--user-data-dir <dir> Specifies the directory that user data is kept in. Can be used to open multiple distinct instances of Code.
Is it storing some temporary files there?
Is it about the access path to config files?
I am asking as I want to run VSCode (or Codium to be more exact) with sudo (I want to edit system config file that is read restricted) which requires this parameter for reasons unclear to me.
Since sudo-ing VS Code at command-line launch is only a thing on Linux, this question assumes you're on Linux, and restricts its context to Linux.
TL;DR
To answer your question directly: the user-data-dir parameter points to a folder where all personalisation except extensions resides — unique to each user.
Why does sudo-ing Code need --user-data-dir?
In typical installations of either OS and VS Code, this folder owned by the regular user cannot be accessed by the superuser.
Hence a VS Code session running with effective UID=0 tries but fails to write to the invoking user's (not the superuser's) config folder. This is what the error message prevents from happening, by forcing the user to provide an explicit root-accessible folder.
Detailed Explanation
There are two main folders that VS Code uses to store configuration data:
An extensions folder (self explanatory) — contained in ~/.vscode[1]
user-data-dir; a folder for all other personalisable things (settings, keybindings, GitHub/MS account credential caches, themes, tasks.json, you name it)[2]
On Linux the latter is located in ~/.config/Code, and has file permissions mode 0700 (unreadable and unwritable by anybody other than the owner).
This causes issues, as VS Code tries and fails to access said directory. The logical solution is to either modify the permissions (recursively) of ~/.config/Code to allow root access, or — arguably saner and objectively more privacy-respecting — to use a separate directory altogether for the sudo'ed VS Code to access.
The latter strategy is what the community decided to adopt at large; this commit from 2016 started making it compulsory to pass an explicit --user-data-dir when sudo-ing VS Code on Linux.
Should You be Doing This in the First Place?
Probably not! If your goal is to modify system config files, then you could stick to an un-elevated instance of Code, which would prompt you to Save as Admin... when you try to save. See this answer on Ask Ubuntu on why you probably want to avoid elevating VS Code without reason (unless you understand the risks and/or have to), and this one on the same thread on what you could do instead.
However, if the concerned file is read-restricted to root as well, as in the O.P’s case, then you hardly have a choice 😕; sudo away! 😀
[1] & [2]: If you want to know more about the above two folder paths on different OSes, see [1] and [2]
Hope this was helpful!
It might be helpful to easily find the default location of the user-data-dir on any OS. It can be found with this command:
Developer: Open User Data Folder
workbench.action.openUserDataFolder
which is in the Insiders Build v1.75 now, Stable soon. Opens your OS file explorer app to the location.
I'm probably just being very thick here, but it's not clear to me where I'm supposed to install 'new' user-specific programs on Windows 7 (and presumably Vista too, though I've not specifically looked at that scenario yet).
Under Windows XP (rightly or wrongly) we always installed our programs into folders under 'Program Files' and accepted that they'd be kind-of available to everyone. From what I can gather under Windows 7 I'm supposed to install my software under the user's AppData folder (possibly AppData\Local\MyApp). That makes a degree of sense, but the fact that this folder is 'hidden' by default means that we're going to have 'fun' talking our users through support stuff.
I want to install our software so that it's user specific (the Users bit in Windows 7 makes perfect sense) but I do want the user to be able to access it if required. Our program also includes a 'data' subdirectory which it needs to write into while it's running (embedded database), but as the program is intended to be single-user/standalone, the data folder being inside a user-specific folder isn't going to be a problem.
My problem is just that whole 'hidden folder' aspect of AppData. As much as I've trawled the MSDN, I can't work out where else I'm supposed to install user-specific programs. Taken one way it would seem to be something like AppData\Local\MyApp, and another way it would seem to be just as valid under the user's My Documents\MyApp equivalent.
Has anyone got a clear guide for where all this stuff goes? I found the MSDN docs confusing. :-)
Not really.
The directory that serves as a common
repository for application-specific
data for the current roaming user.
AppData is, surprisingly, for application data, not for installation (Click Once/Silverlight applications aside). You can, and should still install into Program Files, just don't expect to write into that folder.
You can install software into AppData if you want it to follow a user about in an Active Directory environment, which happens if you put it in AppData\Roaming (the SpecialFolder.ApplicationData location).
You can also install into AppData if you want the software to be available to just the user that installs it. This can be useful if, for example, you have multiple users on the same machine, who all want to run different versions of the software in complete isolation.
If you want settings to only apply on the local machine then you use AppData\Local, which is SpecialFolders.LocalApplicationData - this will make AD administrators very happy as the roaming profile size won't suddenly jump up 50Mb or whatever the size of your software is.
If you wanted to create settings which apply to all users then you're looking at SpecialFolders.CommonApplicationData
You should remember never to rely on the actual name of the directory - localisation issues mean this can change and the location does change with OS versions two. You should be using the special folder enumeration in your software, or the equivalent in your installer.
Could you not install into Program Files, but use AppData as it's supposed to be used, and store your database in there?
Windows 7 added the FOLDERID_UserProgramFiles known folder and by default this maps to %LOCALAPPDATA%\Programs. This is used by MSI when ALLUSERS=2 & MSIINSTALLPERUSER=1.
On Vista and earlier there is no canonical per-user application folder but just using %LOCALAPPDATA% is pretty common. Sadly MSI will just use %ProgramFiles% on these systems.
It's 2019, and I just installed Visual Studio Code (a Microsoft product) in the default folder of
%userprofile%\AppData\Local\Programs\Microsoft VS Code
This is probably for getting around the requirement to have an administrator or UAC prompt authorise the installation
Windows 7 folder structure is deeply inspired on Unix structure:
/usr/ -> C:\Program Files\ -> binaries: executables and dynamically linked
/etc/ -> C:\ProgramData\ -> global settings
/home/ -> C:\Users\ -> a folder for each user
~/.* -> C:\Users\Hikari\AppData\Roaming\ -> settings for each user
Windows has more folder, like My Documents for files with content produced by user, AppData Local and Roaming (which Unix usually handles with NFS).
It's about time for us developers to start using these structures. We must separate at least binary files that don't need to be replicated, global and user settings.
When a setup is installing an app, this setup should expect to have permission to write on Program Files. Once the setup is finished, Program Files should be writable only for other setups aiming to update binaries to other versions.
Please install executable files to the %programfiles% folder in Windows - a simple MSI based install package can perform an active setup for any new user who logs onto the machine to create the user specific files and folders in their profiles %appdata% folder. You see this behaviour for Internet Explorer, Adobe reader, etc. - It's the little MSI installer window that pops up the first time you log onto a machine which has those applications installed. - Thanks - a system admin :)
My opinion, for what it's worth, is that user-specific program files is just asking for trouble and is a damn stupid thing to do.
A much more sensible approach is to install different versions of your program to:
\Program Files\Your Program\Program_v0.1\Program.exe
\Program Files\Your Program\Program_v0.2\Program.exe
\Program Files\Your Program\Program_v0.3\Program.exe
\Program Files\Your Program\Program_v0.4\Program.exe
I would then place a bootstrapping launcher at:
\Program Files\Your Program\ProgramLauncher.exe
Then, the user application data folder will only contain data, including an INI/XML/Settings file that indicates the version of the program that this user is working with.
Such an approach satisfies the core tenant of keeping data and executing code separate, allows every user to run a specific version of the code, and offers a small amount of de-duplication by ensuring the same executable code is not copied multiple times across user folders.
Otherwise, go right ahead with installing programs to AppData and undoing the years it has taken us to achieve clean separation of code and data. I found this thread because I noticed that Chromium and DropBox are installing code to AppData. I'm going to uninstall those program, and change the permissions on my AppData folder to exclude execution to ensure I can easily spot other programs attempting the same BS.
On my website I need to give users an ability to download and easily start my executable (AppImage binary) on Linux.
For Windows version it is just .exe which works after downloading and clicking:
Download file
Click on file in browser downloads panel to start it
For Linux user now needs to do following:
Download file
Open folder containing file from browser downloads panel
Right click on it to add exec permission
Click on "Allow execution of this file" checkbox
Press Ok
Click on file to start it
It is hard to explain this flow for regular user, it makes users leave away.
Is it possible to minimize it to minimal clicks as on Windows?
Any advises appreciated to achieve minimal clicks. I can compile app for any format (its on electron but I can process it before upload)
I thought about using .deb . It will limit app for Debian-based only, but main problem in same time that I did not find ability to run post installation to exec app, and I don't want ask user to enter start menu.
Executable bits are a basic UNIX security measure thus it is not really easy to work around this (for good reason). Thinking about this, for the specific case of downloaded files, Windows also applies some restrictions (special NTFS stream which tells Windows Explorer to warn about the dangers of an executable file from the Internet).
You can of course provide your application as a .tar.something archive and store executable files in there. After extraction, they will normally have the correct execution bits set.
The option with the deb-package can also solve your problem (for some users) but is a little more complicated:
User downloads deb package
User clicks on deb package and has some program installed that provides a GUI for installing packages (like gdebi). Like on Windows there will be some "security check" in form of a dialog box where the user needs to enter a (sudo) password. Afterwards, apt will install the package
If the package is created correctly, it can transport the executable bit correctly such that no explicit permission change is needed afterwards. If for some reason there is need to do something post-installation, Debian packages can provide postinst scripts which run (as root!) at the end of the package's installation.
In any case, as dealing with executable files is a common procedure on Linux, it might not scare so many users away as expected. If you want to make it comfortable for the users, provide the package as they expect/like them. On Windows I would think that to be a .msi package and on Linux I prefer a package corresponding to my distribution (.deb, .rpm).
If you want the users to update their packages regularly (good for security) then it is helpful to provide a "repository" that users can add and install your package from. Of course, "the best" is having a package as part of the distribution, but that is quite some effort and needs to pass a lot of "quality assurance gates" :)
https://github.com/mongodb/mongo/blob/master/debian/mongodb-org-server.postinst
I could see in the above link that mongodb has created a newuser named 'mongodb'.and I got to know that many bigger projects run their project as a custom user (here as a user 'mongodb').I wrote a small application in python and running it as sudo right now,I would like to create a new custom user for my application and run the application as the custom user.how does actually bigger projects acheive this .what is the command they use to make run the whole application or package as a custom user?
In a whole lot reading about this I found that we could use sudo -u <username> <command>
or by setting euid or uid by seteuid() function. but I want to know how actually real projects implement it and follow that standard way in my project.
As in the file you linked they change the owner of the executable to their custom
using chown so you could use that in a post-installation script (a script that will configure what is necessary for your application to run correctly)
I am creating a graphical installer that should run on Linux. Installing should consist of copying files to some places in /usr. Currently the installer is written in Python.
How can I escalate the privileges of my installer when I need to copy files? I looked at PolicyKit but
a) there doesn't seem to be a generic "install files" action-id for PolicyKit
b) of the action ids I can use, I don't think they are standard across distros
I also looked at PAM and I have code that uses libpam but I can't seem to do anything with it. After authenticating my user (by providing username and password) I don't have write access to /usr. I tried changing my user with os.setuid(0) after authentication but I get an error from the OS.
Also, strangely, it doesn't seem to matter what service I provide to pam_start. As long as the username and password are correct I can pass anything I want. I see I have /etc/pam.d/sudo. The below code is simplified, the password is correctly stored in a pam_conversation object and I do pass a handle object.
pam_start("my_user", "my_pass", "sudo_garbage_12345");
works just as well as
pam_start("my_user", "my_pass", "sudo");
That is, they both succeed.
As a last resort I can probably execute gksudo or kdesudo but I don't want to be tied to those programs. Requiring users to invoke my installer with sudo is a (very) last resort.
You might be better off wrapping RPM with a front end that takes the user options and invokes RPM to do the hard work. This also gives you infrastructure for managing dependencies and plays nicely with the existing package management system. If you need to run on a .deb based system (Debian or Ubuntu) you may also need to consturct a .deb and put some mechanism in the front end that works out which package management system is active.
Granting random users access to root privilege is generally viewed as bad form on Linux or Unix systems (or any multi-user system for that matter) as it is a significant security risk. However you do have the option of letting the user install it under their home directory (~/bin) if they don't have root access or sudo permissions that allow them to write to system areas. In this case you can require them to install it as root if they want to install in /usr/bin but permit them to install it under their home directory for their own use if they don't have root privileges.
For a graphical installer, stick with a graphical environment. Use gksudo or kdesudo if they are available, otherwise fail with an error dialog saying they need root. People (newbies in particular) will download your installer and double-click to launch it from their desktop, and you need a graphical way to ask them for their password. You don't want to pop open a terminal on them.
Given that, don't do sudo for them even if they are running from the terminal. Just output an error saying you need root and exit. If the user is already at the command prompt (like I most likely would be), I already know how to sudo or su myself into root if I want to do so. I promise you you will most likely ruffle some feathers if you attempt to make an experienced user root when they can do it themselves.
If you INSIST on doing a sudo yourself from within your installer, for God's sake please force a 'sudo -K' before you do to remove the previous timestamp. If you don't do this, and I have sudo'd recently, you will run your installer with me as root without my knowledge (since I don't expect that to happen). A 'sudo -K' will force a prompt that I can then decide whether I want to proceed as root or not.
The best way in this case is to use su within your program. Redirect inputs/outputs and you're good to go!