Hello,
function build_dep in modprobe.c assumes that dependencies of one module
have not more than 255 chars;
that is not sufficient in kernel 2.6.7 (alsa sound modules). - Below is
a diff that solves the problem for me.
With regards, Christian Ostheimer
"There seems to be a slight problem with the "mod_strcmp" function in
modprobe.c, it scans for the first occurence of the module name in the
"mod_path" variable and expects it to be the last path element. ie
/lib/modules/2.4.22-debug/kernel/fs/vfat in my example. The comparison
will always fail if mod_path contains another substring matching the
module name."
Robert McQueen wrote
"Although William Barsse's patch fixed mod_strcmp for 2.4 kernels, there
was a remaining problem which prevented it from working for me. I've
just tracked it down - when you enable kernel 2.6 module support it
hard-wired the extension to .ko instead of checking at runtime like the
other places where 2.4 differs from 2.6. The attached patch fixes this
for me."
fixes two other issues (plus the previous as well) with a 2.4 kernel :
- should be able to modprobe an already loaded module and get 0 return
code :
# modprobe <something> && modprobe <something> && echo "ok" || echo "failed"
....
failed
Well, hope this helps and that I didn't screw up again,
- William
Support for /etc/modprobe.conf (for 2.6 kernels) should likely be added
to bb's modprobe, see attached patch.
modprobe.conf is just a (even simpler) variant of modules.conf
Hi,
There was some problem with busybox modprobe. For details see
http://www.busybox.net/lists/busybox/2004-May/011507.html
I made a patch against busybox-1.00-pre10 to fix that one.
This is a slight variant of Patrick's patch with a slightly
cleaner implementation of mod_strcmp()
-Erik
Fix parsing of all tag-value pairs (in modules.conf in particular).
Without this fix, code chokes badly on lines where either value or
both tag+value are missing, like bare
alias
line, or alias w/o the value like
alias some-module
(syntactically incorrect, but no need for coredumps either).
Initialize all fields of struct dep_t.
Without that, e.g. `busybox modprobe -v char-major-10-144' *sometimes*
fails this way (strace):
write(1, "insmod nvram `\213\f\10\n", 21) = 21
Note the garbage after module name which is taken from the m_options field,
which is not initialized in the alias reading/parsing part.
(Shell properly complains to this command, telling it can't find the
closing backtick)
I have found the problem in modprobe, so here is the promised patch
At the current stage I can use it as modprobe while switching between
2.4 and 2.6 seemlesly...(that is good!)
accept more then 1 dependency per modules.dep line. Also white space cleanup...
I think that parsing still breaks sometimes, but is mostly functional now.
Erik, I think we have met online some time ago when I was in Corel/Rebel
Netwinder project....
Anyway, I would like to use BB on 2.6.0 initrd. 1.00-pre4 works OK, if
insmod is actually presented with a full path to the module. Otherwise -
problems (not to mention conflicts when 2.4 modutil is enabled)
Here are some patches for insmod and modprobe which try to walk around
the default ".o" module format for 2.2/2.4 modules (you have probably
noticed it is now .ko in 2.6 ;-)) Trying to steal as little space as
possible if 2.6 not enabled...
The modprobe is still not perfect on 2.6 - seems to be jamming on some
dependencies, but works with some (to be debugged). Anyway after the
patches it at least tries to work....
Will there be a 1.00-pre5 coming any time soon?
Thanks, Woody
Hey guys. I've found a bug in modprobe where it generates bad strings and
makes sytem calls with them. The following patch seems to have fixed the
problem. It is rather inherited elsewhere, as there seems to be incorrect
entries in the list which results in more dependencies than really exist for
a given call to mod_process. But, this patch prevents the bad text from
going to the screen. You will notice there are cases where lcmd goes
unmodified before calling system.
Please consider the following patch.
Thanks.
-Steve
- attempting to modprobe a module that is already loaded yields "Failed
to load module", whereas modutils quietly ignores such a request.
- if a module genuinely can't be loaded due to missing symbols or
similar problems, modprobe doesn't produce any useful diagnostics
because the output from insmod has been redirected to /dev/null.
Here's a patch to address these issue
Patch by Philip Blundell
I've had some issues with modprobe which I reported a few months ago. This
is still an issue so I decided to sort it out.
The attached diff includes the changes against the unstable cvs tree that
work for me.
Changes are:
mod_process() will report success if the module at the head of the list
loads successfully. It will also report success if any module unloads
successfully.
The net result being that modprobe will succeed in the cases outlined below.
I've also added error reporting to modprobe -r. Previously it would silently
fail (but report success) if the module could not be unloaded.
Andrew
trying to call 'insmod -q', which wasn't supported. Secondly, when modprobe
was fed blank lines from modules.dep, we ended up calling xstrndup(ptr, -1),
which with suitably bad results. David provided a patch to catch the blank
lines, and I have added insmod -q support. So modprobe should work again.
-Erik
