I have been experiencing problems with the NFS server support on a Sparc
IPX under Linux.
The station (a cutting edge 16 Mb SparcStation IPX) serves an alternate
/usr directory to a dozen Sparcstations (some IPX, two LX, a few SS4), all
running under Solaris 2.5.1.
The trouble is, the user-mode nfsd is slow (old news), and the new
kernel-based nfsd (supposed to be faster) systematically refuses all mount
connections from any machine.
The /etc/exports is the following :
/snepa *.rip.ens-cachan.fr(rw,no_root_squash)
(of course, I have applied the right exportfs before using either server
daemon).
After drilling into the sources of both the kernel NFSD part and its
user-level support daemons (a first time experience to me), it has come to
me that in fs/nfsd/export.c, in the function exp_export(), the kernel half
was getting a device number for /snepa which is different from the device
number the function stat() (in the user-level part) gave (the i-node
number was OK). To me, and without learning more, that means that the
kernel uses a different device number table than what it exposes to the
user level. Of course, I'm certainly wrong, and I prefer being told that
by the Gods than having to devise a poor and ill-written workaround.
I have no convenient access to a i386 PC I could recompile Linux on, to
check whether this problem is specific or not. Has anyone succesfully used
the knfsd on recent kernel, on any platform ?
My configuration :
Hardware:
Sun Sparcstation IPX (sun4c), PROM version 2.6, 39.83 BogoMips.
16 Mb RAM, 2 Gb HD.
Software :
S/Linux 2.1.89-VGER-CVS980313
linux-nfs 0.4.22 from ftp.mathematik.tu-darmstadt.de/pub/okir
I just tried to compile 2.1.92, but to no avail : the
sparc-specific interrupt support include files don't look to have followed
the latest i386-and-generic modifs. I'm being downloading the
2.1.91-VGER-CVS snapshot from march 30th, but I'm not expecting too much
of it.
Thanks in advance.
-- Cyrille
------------------------------
Zog Zog
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu