Do not label any files in the udev runtime directory, but only nodes,
links and directories below /dev.
In case the runtime directory falls back to /dev/.udev, label this
directory once at udevd startup, but never anything below it.
We should bind the udev socket from systemd, so we are sure
that the abstract namespace socket is always bound by a root
process and there is never a window during an update where
an untrusted process can steal our socket.
Also split the udev.service file, so that the daemon can be
updated/restarted without triggering any coldplug events.
Syslog wants to distinguish the sorce of messages. We should
indicate that this is a userspace message (LOG_DAEMON) and not
a kernel message (LOG_KERNEL).
Renaming network devices might delay events for the other device, which has
the same devpath in the meantime as the original event. Causing a delay until
the timout of the event is reached.
Look at the ifindex/devnum of the devices to check if they are really
the same devices.
For example, not all systems have PROC_KCORE enabled. Avoid a broken symbolic
link in those cases.
Signed-off-by: Yin Kangkai <kangkai.yin@intel.com>
Signed-off-by: Martin Pitt <martin.pitt@ubuntu.com>
We special-handle events with a TIMEOUT= set, so they don't get queued
or wait for parent events to finish, to make sure we can handle them
as fast as possible.
With this change we first try to find an idle worker process before
forking a new one.
It handles only RUN but not IMPORT and PROGRAM. There is no sane way
to suppress program execution. Most important programs run with IMPORT
these days. Also events can no longer suppressed with the libudev
netlink messages, so UDEV_RUN does nothing useful and is just
inconsistent.
There is no way to ignore an event these days. Libudev events can
not be suppressed. It only prevents RUN keys from being executed,
which results in an inconsistent behavior in current setups.
This should also address:
On Tue, Nov 3, 2009 at 16:21, Marco d'Itri <md@linux.it> wrote:
> udev_rules_new() in udev/udev-rules.c unconditionally creates the
> directory.
> This is a problem because the function is called also by e.g. udevadm
> test, and creating /dev/.udev/ when it does not exist is an unacceptable
> side effect which will break everything else that checks for its
> existence to know if udev is running.
On Wed, Oct 7, 2009 at 21:46, Alan Jenkins <sourcejedi.lkml@googlemail.com> wrote:
> Udev would have avoided the race prior to
>
> 82c785e "udevd: remove check for dev_t, DEVPATH_OLD takes care of that"
>
> (the "check" removed here used to serialize events based on the device
> major:minor number).
On Wed, Oct 7, 2009 at 22:31, Michael Guntsche <mike@it-loops.com> wrote:
> add /module/8250_pnp (module)
> remove /devices/platform/serial8250/tty/ttyS0 (tty)
> add /devices/pnp0/00:05/tty/ttyS0 (tty)
On Wed, Sep 23, 2009 at 23:11, Matthias Schwarzott <zzam@gentoo.org> wrote:
> It is about ioctl failures on amd64:
> http://bugs.gentoo.org/show_bug.cgi?id=286041
>
> A bad parameter type to an ioctl() call causes udev-146 to generate "error
> getting buffer for inotify" messages in syslog. The offending code is
> roughly:
>
> ssize_t nbytes, pos;
> // ...
> ioctl(fd, FIONREAD, &nbytes);
>
> where ssize_t is 64 bits on amd64, but the kernel code for FIONREAD (at least
> through gentoo-sources-2.6.31) uses type int:
>
> p = (void __user *) arg;
> switch (cmd) {
> case FIONREAD:
> // ...
> ret = put_user(send_len, (int __user *) p);
>
> so the upper 32 bits of "nbytes" are left uninitialized, and the subsequent
> malloc(nbytes) fails unless those 32 bits happen to be zero (or the system has
> a LOT of memory).
External programs triggered by events (via RUN=) will inherit udev's
signal mask, which is set to block all but SIGALRM. For most utilities,
this is OK, but if we start daemons from RUN=, we run into trouble
(especially as SIGCHLD is blocked).
This change saves the original sigmask when udev starts, and restores it
just before we exec() the external command.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
It did not work for the last couple of releases.
If RUN{record_failed}+="..." is given, a non-zero execution will mark
the event as failed. Recorded failed events can be re-triggered with:
udevadm trigger --type=failed
The failed tracking _might_ be useful for things which might not be
ready to be executed at early bootup, but a bit later when the needed
dependencies are available. In many cases though, it indicates that
something is used in a way it should not.
We may need to handle SIGCHLD before the queued worker message. The last
reference, from the SIGCHLD or the worker message will clean up the worker
context. In case we receive an unexpected SIGCHLD with an error, we let
the event fail and clean up the worker context.
Persistent network rules write out new rules files. When rules change,
we need to kill all workers to update the in-memory copy of the rules.
We need to make sure, that a worker finshes its work for all device
messages it has accepted, before it exits after a SIGTERM from the main
process.
