Date: Tue, 27 Jan 1998 00:21:36 -0700 (MST) From: Kenneth Merry <ken@plutotech.com> To: dyson@FreeBSD.ORG Cc: wpaul@skynet.ctr.columbia.edu, grog@lemis.com, current@FreeBSD.ORG, dyson@FreeBSD.ORG Subject: Re: Another observation on -current and NFS Message-ID: <199801270721.AAA02087@panzer.plutotech.com> In-Reply-To: <199801270626.BAA27440@dyson.iquest.net> from "John S. Dyson" at "Jan 27, 98 01:26:45 am"
next in thread | previous in thread | raw e-mail | index | archive | help
John S. Dyson wrote...
> Bill Paul said:
> >
> > John Dyson made a couple of commits this weekend which I _thought_ would
> > fix this thing; I gave him a sample program that triggers the bug to help
> > him track it down. Assuming that he got my sample program to work, I'm
> > guessing there's more than one way to trigger the condition.
> >
> The fixes explicitly fixed the problem with the test code. There indeed
> must be more problems. I will look at it while it is still in my brain-cache.
Well, FWIW, I'll share my experiences with -current's NFS today. I
have noticed the 'strip' problem happening as well for the past couple of
months at least. In my experience, the client machine (i.e. the one doing
the strip) wouldn't hang, but the kernel it generated wouldn't boot.
I tried stripping a kernel today over NFS (between a client and
server that are both current as of today) and the kernel booted just fine.
I have also been having link troubles when compiling/linking
programs over NFS. (the linker wouldn't find functions that were indeed in
the libraries..) Those problems are gone as of today.
There is one interesting NFS problem I can report, though.
Apparantly there is a problem with access(2) and files on NFS mounted
filesystems. (one of my co-workers discovered it)
Here is a program to test permissions:
(I didn't write it, I just made a minor modification...)
========================================================================
#include <stdio.h>
#include <unistd.h>
void
usage
(
void
)
{
fprintf(stderr, "usage: testAccess <octal mode number> <file name>\n");
fprintf(stderr, "Where <octal mode number> the inclusive or of some of\n");
fprintf(stderr, "F_OK, R_OK, W_OK, or X_OK\n");
exit(1);
}
main
(
int argc,
char *argv[]
)
{
int indx;
int mode = 0;
char *tmpPtr;
if (3 != argc) {
usage();
}
for (tmpPtr = argv[1]; 0 != *tmpPtr; tmpPtr++) {
int digit;
if (!isdigit(*tmpPtr)) {
usage();
}
digit = *tmpPtr - '0';
if (7 < digit) {
usage();
}
mode = (mode << 3) + digit;
}
printf("Access returns %d for file %s and mode %o\n",
access(argv[2], mode),
argv[2],
mode);
}
========================================================================
And here is what happens:
On a local filesystem, the program behaves as expected:
====================================================
{roadwarrior:/usr/home/ken/src/yates:162:0} ls -la foo
---------- 1 ken wheel 0 Jan 26 18:35 foo
{roadwarrior:/usr/home/ken/src/yates:163:0} ./testAccess 777 foo
Access returns -1 for file foo and mode 777
====================================================
The file foo has no permissions, so access(2) should return -1.
When the same filesystem is NFS mounted, the program does not behave as
expected:
====================================================
{bladerunner:/usr2/home/ken/src/yates:34:0} ls -la foo
---------- 1 ken wheel 0 Jan 26 18:35 foo
{bladerunner:/usr2/home/ken/src/yates:35:0} ./testAccess 777 foo
Access returns 0 for file foo and mode 777
====================================================
In any case, this looks like a bug to me, but if it's just operator
error, please let me know.
As far as I can tell, though, the major (and in some cases show
stopping) NFS bugs in -current have been fixed. Thanks John!
Ken
--
Kenneth Merry
ken@plutotech.com
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199801270721.AAA02087>