home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!usc!nic.csu.net!nic.csu.net!nntp
- Newsgroups: comp.sys.sun.admin
- Subject: Symlink on full fs crashes SunOS 4.1.1.
- Message-ID: <1993Jan22.223720.3919@nic.csu.net>
- From: oleg@gd.cs.csufresno.edu (Oleg Kibirev)
- Date: 22 Jan 93 22:37:18 PST
- Distribution: world
- Organization: Computer Science Departement of California State University in Fresno
- Nntp-Posting-Host: gd.cs.csufresno.edu
- Lines: 34
-
- Once in a while a parition on HD of our Sun 3/60 server running SunOS 4.1.1
- fills up. I noticed that if at this point someone tries to make a symbolic
- link within this partition, they get "I/O error" instead of expected "no free
- space left on device" and the server totally messes up. Any process that tries
- to access this fs in future would lock up and ignore any signal (and will be
- marked as D in output of ps -u). This is especially unfortunate on NFS server
- since after clients do the same thing 8 times, all NFS daemons block.
-
- Freeing up a little bit of space on the parition doesn't help. If the system
- is rebooted, the problem will repeat when directory with symlink is accessed.
- unlink()ing symbolic link without trying to stat() it seams to be a solution
- (one still needs to reboot after this).
-
- I suspect SunOS has problems checking for free space before starting
- symlink(). If it is the case, it's too bad that it doesn't forget to check
- MINFREE when actually doing it. (Server's filesystems have usual 10% reserved
- for performance).
-
- Is there any NFS patch we are missing? Does it work? ;)
-
- Thanks!
- Oleg
-
- P.S: Just looked at the console of a client mounting /home from the server
- mentioned above and noticed that it's very busy repeating the same message:
-
- ...
- NFS getattr failed for server yosemite: RPC: Unable to receive
- NFS getattr failed for server yosemite: RPC: Unable to receive
- NFS getattr failed for server yosemite: RPC: Unable to receive
- ...
-
- I have never seen this message before. Probably another reboot is a fix, but
- does anyone have an idea why is this happening?
-