| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
|
|
If we take over the hotplug call and manage the events we don't need
to call the event fake script in dev.d/. Just set all expected values
to the new network interface name and call hotplug.d/. This way the
device renaming is completely handled inside of udev and userspace
can't get confused.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
On Sat, 2004-12-11 at 18:44 +0200, Martin Schlemmer [c] wrote:
>
> Any suggestions to determining the version of the installed udev?
> This is now during startup, to see if we can make use of using
> udevsend as hotplug agent. If the system was up, udevinfo could
> be used, but that is in /usr/bin that might be on a seperate /usr.
> I know we might move udevinfo to /bin, but that might be an issue
> for some, and adding a -V switch to /sbin/udev might be a better
> choice.
|
|
|
|
|
|
Remove udev.bus, cause it's currently unused and newer kernels will pass
it in the hotplug environment as PHYSDEVBUS.
Remove udev.action, cause it's unused.
Rename udev_set_values() to udev_init_device().
|
|
|
|
|
|
Damm, it's hard to merge a multi-line tree into one flat line at times...
|
|
The /etc/dev.d/input/input.dev was called twice for /dev/input/mouse.
Skip the execution if we get a directory named after the subsystem.
Move UDEV_NO_DEVD where it belongs.
|
|
|
|
|
|
I just noticed that the DEVNAME enviroment variable isn't being set anymore
in udev 0.046 on device removal, while it was being set in 0.042. We're using
the property tto do umount -l <devices> when a block device is removed. Afaik
there is no other way to associate a device with it's DEVNAME on removal ?
Also are there cases where doing umount -l on the removed devices is wrong?
I guess the device is gone, so there is no sense in keeping it mounted (it's
not like the filesystem is gonna come back in a sane state again)..
Attached (trivial) patch brings back the DEVNAME variable on device removal.
|
|
If /proc/sys/kernel/hotplug points to /sbin/udevsend we handle the whole
hotplug event with multiplexing /etc/hotplug.d/.
|
|
Move the wait_for_sysfs logic into the udev binary. udev is called for
every hotplug event. It also waits for /devices events.
|
|
Fix from Sjoerd Simons <sjoerd@spring.luon.net>, it got lost
on the recent reorganization of the udev processing stages.
|
|
Forked scripts and callouts may want to follow udev's configured behavior
and log only if udev is logging itself.
|
|
The slow logging facilites on some systems are a reason for
the reported slowness of udevstart. On one of my boxes udevstart
is down from 9 second to 0.3 seconds.
|
|
|
|
Move the logic when and how to call the dev.d/ scripts into the
main processing path.
|
|
Remove the overwriting of main_argv[] hack and use the values
from the udev object.
Pass the udev object to call_foreach_file().
In the udevstart case, export SUBSYSTEM and UDEVSTART to the
environment.
|
|
|
|
"udev udevstart" will run udev as udevstart. This makes it easier
to run a test in the source tree without the need to create a
symlink.
|
|
This makes the udev operation completely lockless by storing a
file for every node in /dev/.udevdb/* This solved the problem
with deadlocking concurrent udev processes waiting for each other
to release the file lock under heavy load.
|
|
|
|
|
|
This patch exposes the wait_for_sysfs functions to all possible users,
so we need to maintain only one list of exceptions. The last list is
hereby removed from udev.c.
|
|
|
|
Here we supress the dev.d/ execution if we didn't change a network
interface's name with a rule. This should solve the issue of two
running dhclients for the same interface, cause the
/etc/dev.d/net/hotplug.dev script that fakes the hotplug event runs
with every udevstart for every interface and fakes a second identical
hotplug event on bootup.
With this patch netif interfaces are no longer stored in the udevdb.
It is not needed, cause we don't have permissions or symlinks :) and
all information is available in sysfs.
This patch also moves the dev_d execution calls out of the
udev_add/udev_remove. As with the former api-cleanup-patch we have
all processed data in one udev struct and can place the execution
calls where needed.
|
|
Here is the first patch to cleanup the internal processing of the
various stages of an udev event. It should not change any behavior,
but if your system depends on udev, please always test it before reboot :)
We pass only one generic structure around between add, remove,
namedev, db and dev_d handling and make all relevant data available
to all internal stages. All udev structures are renamed to "udev".
We replace the fake parameter by a flag in the udev structure.
We open the class device in the main binaries and not in udev_add, to
make it possible to use libsysfs for udevstart directory crawling.
The last sleep parameters are removed.
|
|
Seems that we never closed the opened syslog.
Here is a patch to do this in all our binaries.
|
|
I'm using an older C compiler, and it doesn't like assignments mixed with
declarations, but this is also a style cleanup.
|
|
Here we remove all the sysfs sleep loops from udev as wait_for_sysfs
will do this for us and any other hotplug user. We still keep a small
blacklist of subsystems we don't care about but any missing entry here
will no longer lead to a spinning udev waiting for files.
|
|
|
|
|
|
This patch fixes the reintroduced bug with the
sig_handler(), if we link against a -mregparm=3 compiled
klibc on i386.
It also fixes some compiler warnings about redefined
asmlinkage on some systems.
Also some (broken?) compilers on distros throw out warnings
if asmlinkage is before "static void". This fixes it, too.
|
|
This patch fixes two
warning: assignment from incompatible pointer type
asmlinkage is the reason for the warning. We can
simply cast to avoid it.
It also fixes this warning:
warning: implicit declaration of function `umask'
|
|
Here is the patch, that should prevent all of the known deadlocks with
corrupt tdb databases we discovered.
Thanks to Frank Steiner <fsteiner-mail@bio.ifi.lmu.de>, who tested all this
endlessly with a NFS mounted /dev. The conclusion is, that udev will not work
on filesystems without proper record locking, but we should prevent the
endless loops anyway. This patch implements:
o recovery from a corrupted udev database. udev will continue
without database support now, instead of doing nothing. So the node should
be generated in any case, remove will obviously not work for custom names.
o added iteration limits to the tdb-code at the places we discovered endless
loops. In the case tdb tries to find more than 100.000 entries with the
same hash, we better give up :)
o prevent a {all_partitions} loop caused by corrupt db data
o log all tdb errors to syslog
o switch sleep() to usleep() cause we want to use alarm()
|
|
Fix a stupid logic bug, I introduced with the udev.c simplification. We
want to look at class and block devices only.
|
|
On Tue, Sep 07, 2004 at 12:46:43PM +0200, Kay Sievers wrote:
> On Mon, 2004-09-06 at 20:18 -0700, Tom Rini wrote:
> > I noticed somewhat recently that my enet devices weren't being renamed
> > on boot anymore. I don't quite know when this got broken (or rather, if
> > it was supposed to be working. I swear it worked for me once..), but
> > the following seems to do it.
>
> I think it never worked in the udevstart case. It worked only with the
> hotplug-event-udev, I expect.
>
> > In udev_scan_class(), look for not just
> > %s/%s/dev (which everything with a dev node has), but %s/%s/dev* (both
> > of my enet devices, sis900 & 3c59x only have device) and if that
> > exists, pass this along to udev.
>
> Yeah, network devices don't have a devnode and therefore no "dev", but
> they are all in /sys/class/net/. We may just test if we are there
> instead of the "device" match.
How about something like this. It adds all the net devices without
looking at the attributes and keeps the remaining logic like it is.
It also removes certain levels of indirection and much simplifies the
udevstart process. We surely don't need to open and close the udevdb
for every node while iterating over the list. (We are about 5% faster on
my box)
It's not well tested, so it would be nice if someone can have a look
at it, before a broken udevstart renders any system unbootable.
|
|
Move setting UDEV_NO_SLEEP into main(). I thought about moving
udev_init_config() around, but it still must be invoked in both udev and
udevstart cases, and before udev_hotplug() is called. An alternative
would be to have main() do:
if (is_udevstart) {
... current ...
} else {
udev_init_config();
return udev_hotplug();
}
And move setting UDEV_NO_SLEEP into udev_start(). I can redo it that
way, if you prefer.
Signed-off-by: Tom Rini <trini@kernel.crashing.org>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>
|
|
Hi,
The following patch makes udev/udevstart be a common binary. First,
doing this grows udev by a total of 1.8kB (ppc32, stripped) whereas
udevstart by itself is 6.4kB. I know you mentioned being able to
replace udevstart with a script, but at 1.8kB I don't think it'll be
easy to beat this with size there. Next, the following are by-eye
timings of before, after, and with devfs on a slow, but still usable
embedded platform (config stripped down to more-or-less bare for
ramdisk):
-- Embedded Planet RPX LITE, 64Mhz MPC 823e --
devfs : 15.333s, 15.253s, 14.988s (15.191s avg)
udev-pristine : 18.675s, 18.079s, 18.418s (18.390s avg)
udev-multi : 14.587s, 14.747s, 14.868s (14.734s avg)
The patch ends up being rather large to add this, as in doing so I ended
up making all refs (that I hit..) to devpath/subsystem be marked as
'const'.
Signed-off-by: Tom Rini <trini@kernel.crashing.org>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>
|
