Re: [9fans] 9n
2018-05-02 19:24 GMT+02:00 Fran. J Ballesteros : > I just learned to love absolute paths. > Actually they kind of emerge from my design by themselves. > IIRC, there was no deadlock caused that you should be aware of. > I'ts been a long time and quite a few protocols since then, I can look for > the source; there must be also some docs in the web. > Well, actually I'm pretty curious about the implementation. I'd like to see how you did isolated the changes, since to me they seem rather huge (but my protocol diverge more from 9P2000 than 9P2000.ix). Also, I welcome any suggestion for further documents to read about the topic. > Also, I'm more in favor of prefix mount tables, that they are very > different from what 9 does and they would lead to a very different system. > Can you elaborate? What differences this approach would produce? I can foresee some (eg bind semantics) but maybe I'm missing some of them. > Good luck and have fun. > Thanks! :-) Giacomo > > > On 2 May 2018, at 19:14, Giacomo Tesio wrote: > > > > 2013-06-17 21:06 GMT+02:00 Nemo : > > You should ask if anyone else did that before doing it, instead of saying > > they are un-spined life forms. > > > > Here I am, finally! :-) > > > > I'm designing yet another file protocol for my toy/research os (whose > kernel is derived from Charles Forsyth's Plan9-9k), and I'd like to give a > look at your prior art. > > > > Some of my design decisions lead to a management of mount tables that is > pretty similar to what you describe in your paper about the integration of > 9P2000.ix. > > > > Given you already walked this path, I'd like to know what you have > learnt and if you faced issues I should be aware. > > For example, the slight difference in bind semantics seems to reduce the > risk of accidental loops in the namespace, but I would expect it would > break related userspace assumptions. > > Also, resolving the dot of each process in the Pgrp each time a mount is > done, seems pretty complex and prone to deadlocks. > > > > > > Don't you have a tricorder? > > > > No... but usually I can get away with my sonic screwdriver... :-) > > > > > > Giacomo > > > >
[9fans] gVisor - user space kernel in Go
Just saw this today; might be of interest to some 9fans. Apache open source from Google: https://github.com/google/gvisor
Re: [9fans] 9n
I just learned to love absolute paths. IIRC, there was no deadlock caused that you should be aware of. I'ts been a long time and quite a few protocols since then, I can look for the source; there must be also some docs in the web. Also, I'm more in favor of prefix mount tables, that they are very different from what 9 does and they would lead to a very different system. Good luck and have fun. > On 2 May 2018, at 19:14, Giacomo Tesio wrote: > > 2013-06-17 21:06 GMT+02:00 Nemo : > You should ask if anyone else did that before doing it, instead of saying > they are un-spined life forms. > > Here I am, finally! :-) > > I'm designing yet another file protocol for my toy/research os (whose kernel > is derived from Charles Forsyth's Plan9-9k), and I'd like to give a look at > your prior art. > > Some of my design decisions lead to a management of mount tables that is > pretty similar to what you describe in your paper about the integration of > 9P2000.ix. > > Given you already walked this path, I'd like to know what you have learnt and > if you faced issues I should be aware. > For example, the slight difference in bind semantics seems to reduce the risk > of accidental loops in the namespace, but I would expect it would break > related userspace assumptions. > Also, resolving the dot of each process in the Pgrp each time a mount is > done, seems pretty complex and prone to deadlocks. > > > Don't you have a tricorder? > > No... but usually I can get away with my sonic screwdriver... :-) > > > Giacomo >
Re: [9fans] 9n
2013-06-17 21:06 GMT+02:00 Nemo : > You should ask if anyone else did that before doing it, instead of saying > they are un-spined life forms. > Here I am, finally! :-) I'm designing yet another file protocol for my toy/research os (whose kernel is derived from Charles Forsyth's Plan9-9k), and I'd like to give a look at your prior art. Some of my design decisions lead to a management of mount tables that is pretty similar to what you describe in your paper about the integration of 9P2000.ix. Given you already walked this path, I'd like to know what you have learnt and if you faced issues I should be aware. For example, the slight difference in bind semantics seems to reduce the risk of accidental loops in the namespace, but I would expect it would break related userspace assumptions. Also, resolving the dot of each process in the Pgrp each time a mount is done, seems pretty complex and prone to deadlocks. > Don't you have a tricorder? > No... but usually I can get away with my sonic screwdriver... :-) Giacomo
