Messages in this thread | | | From | (Andrew Walker) | Subject | Re: Problems in 1.3.93 | Date | Mon, 22 Apr 1996 16:13:16 +0200 (MET DST) |
| |
Alan Cox wrote: > > > There really shouldn't be files laying around with r-S group permissions, > > unless you really mean it. Additionally the whole issue of locks and NFS > > needs to be resolved - I know Olaf Kirch started looking at lockd. I'm > > tempted to pay X/Open the money they want for the standard and start looking > > at it myself. > > Youve missed the problem totally. Mark a file mandatory locking, lock it > (on the server), ask for it remotely - nfsd blocks. Goodbye nfs server. > > Ditto for backups and many other programs. > > Alan >
You're right - I did miss your point slightly, although I hope you agree with what I did write. Especially that marking files for mandatory locking is a serious issue and shouldn't be undertaken lightly.
I do have a couple of comments which may or may not be of interest/ relevance:
1) A critical daemon like nfsd should be coded as defensively as possibly. No harm in nfsd noticing the group-id bit and checking for locks with F_GETLK.
2) Opening files with O_NONBLOCK avoids blocking on mandatory locks.
Throw the patches out if you really. They probably won't be missed by many people. I implemented them as a small personal challenge and for a sense of completeness - but if its more trouble than its worth then okay, no problem.
-Andy
P.S. I still maintain that anybody turning on mandatory locking for a file without considering the circumstances is getting exactly what they ask for, but I'm not so blind that I can't see the situation with a horde of students who just love such devilish experiments. If the system utilities can't feasibly be bulletproofed in this respect then I suppose the locks just have to go.
-- Andy Walker Kvaerner Engineering a.s. Andrew.Walker@lysaker.kvaerner.no P.O. Box 222, N-1324 Lysaker, Norway
......if the answer isn't violence, neither is your silence......
| |