3 ms·

> Mobile OSs got this right: on a personal device, the permissions model should be applied to the applications. I cannot agree. If I want to run another browse

by ordu 8y ago

> Mobile OSs got this right: on a personal device, the permissions model should be applied to the applications. I cannot agree. If I want to run another browser instance, I cannot do that on a mobile system. Maybe it would possible to do with some support from browser's developers but with multiuser system I need no support from developers, I can create new user and run another browser that would believe that it is the only browser instance running. It is not just browsers. I can easily experiment with program configs, for example. Something doesn't work, I want to check is it due to some application configs or it's installed plugins. All I need is to create one more user and to start program instance from that user. Android relies on SELinux, and SELinux allows much more than old user system, but SELinux is much bigger headache when you are trying to use it in a way that was not supposed by Google. So in reality SELinux on Android allow me to do nothing, I even cannot run app that requires access to a contact list without allowing it to access a contact list. It would be nice to create one more user on Android with empty contact list and to run this program from that user. Moreover I'd like to create user with faked contact list, faked browser history, with faked all the private information, and to run most of the apps from that user.

AnIdiotOnTheNet 8y ago
Nothing you're describing is a problem with the model, just the implementation. All of those things are easily solvable by being able to choose how the program sees the world. You can do it today with the various namespaces in Linux.
- ordu 8y ago
  Yes, I agree. It is problem of implementation. But nevertheless it is a problem. I didn't tried namespaces myself, but I'm pretty sure that to do in in Linux it would be harder for me to get it to work. Even if I became familiar with namespaces. Multiuser model is settled, simple and transparent model, that just work. You have abstraction of user and abstraction of file access rights, and it is all that you need. namespaces have no such a simple model. How can I run program from other user but give it some special rights to access this git-repo in my home directory? Do I need to write a special C-program for that? Or maybe existing tools already can be configured with some obscure xml-file? I do not know, and maybe I'm mistaken, but knowing the general laws of linux software development, I guess that the best software for it I can find is a complex overengineered corporative tool with bells and wistles, and the easiest way to use namespaces in my case is to write C-program. (If I'm mistaken, please, correct me at least by stating my mistake aloud, or better point me to a docs, please.) And here we came into real issue. To write a good C-program for my tasks, I need to start thinking as a software designer, to invent new simple model of process separation, that allow me to solve 90% of my tasks with ease, and the rest with some headaches, but everything must be possible. The only way to do it in a week is to refuse to think, and to replicate multiuser model on top of namespaces. But I need no replica of multiuser model because I have an existing one. What the point of discarding multiuser model, just to move to another implementation of multiuser model? The only good thing I see is a lack of need to log into a root shell to create or delete users and groups. It would be nice, but I'm not ready to spend a week to write a C program and then unknown amount of time to maintain that program, just to stop using su/sudo for such tasks. So, I can agree that multiuser model is bad for a linux desktop. But we have no real alternative. And the mobile OS approach is the worst. It reminds me of DOS, where you can work with one process at a time, where you cannot run two copies of a program, where all is done in the single global namespace, and any process can do anything it wants. The only choice you have is "to run program or not to run".
  - AnIdiotOnTheNet 8y ago
    > namespaces have no such a simple model. How can I run program from other user but give it some special rights to access this git-repo in my home directory? In Linux terminology, launch the program in a new mount namespace with a rw-bind mount to your home directory. You can do this with firejail, bubblewrap, or minijail easily and without a config file.

by ordu 8y ago

AnIdiotOnTheNet 8y ago

Nothing you're describing is a problem with the model, just the implementation. All of those things are easily solvable by being able to choose how the program sees the world. You can do it today with the various namespaces in Linux.

ordu 8y ago
Yes, I agree. It is problem of implementation. But nevertheless it is a problem. I didn't tried namespaces myself, but I'm pretty sure that to do in in Linux it would be harder for me to get it to work. Even if I became familiar with namespaces. Multiuser model is settled, simple and transparent model, that just work. You have abstraction of user and abstraction of file access rights, and it is all that you need. namespaces have no such a simple model. How can I run program from other user but give it some special rights to access this git-repo in my home directory? Do I need to write a special C-program for that? Or maybe existing tools already can be configured with some obscure xml-file? I do not know, and maybe I'm mistaken, but knowing the general laws of linux software development, I guess that the best software for it I can find is a complex overengineered corporative tool with bells and wistles, and the easiest way to use namespaces in my case is to write C-program. (If I'm mistaken, please, correct me at least by stating my mistake aloud, or better point me to a docs, please.) And here we came into real issue. To write a good C-program for my tasks, I need to start thinking as a software designer, to invent new simple model of process separation, that allow me to solve 90% of my tasks with ease, and the rest with some headaches, but everything must be possible. The only way to do it in a week is to refuse to think, and to replicate multiuser model on top of namespaces. But I need no replica of multiuser model because I have an existing one. What the point of discarding multiuser model, just to move to another implementation of multiuser model? The only good thing I see is a lack of need to log into a root shell to create or delete users and groups. It would be nice, but I'm not ready to spend a week to write a C program and then unknown amount of time to maintain that program, just to stop using su/sudo for such tasks. So, I can agree that multiuser model is bad for a linux desktop. But we have no real alternative. And the mobile OS approach is the worst. It reminds me of DOS, where you can work with one process at a time, where you cannot run two copies of a program, where all is done in the single global namespace, and any process can do anything it wants. The only choice you have is "to run program or not to run".
- AnIdiotOnTheNet 8y ago
  > namespaces have no such a simple model. How can I run program from other user but give it some special rights to access this git-repo in my home directory? In Linux terminology, launch the program in a new mount namespace with a rw-bind mount to your home directory. You can do this with firejail, bubblewrap, or minijail easily and without a config file.