Most of the udev database from initramfs should be deleted before
starting udev in the real root. udevadm: info --cleanup-db deletes
all database entries in /run/udev. Events that processed IMPORT{db},
or mark devices explicitely as persistent, will be excluded.
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.
Warning: noted but unhandled ioctl 0x641e with no size/direction hints
Syscall param socketcall.setsockopt(optval) points to uninitialised byte(s)
at 0x6A30BAA: setsockopt (syscall-template.S:82)
by 0x4E36E1C: udev_monitor_filter_update (libudev-monitor.c:347)
by 0x4E36ED8: udev_monitor_enable_receiving (libudev-monitor.c:373)
by 0x8A9E68E: I830ScreenInit (intel_driver.c:838)
by 0x42EE9E: AddScreen (dispatch.c:3890)
by 0x471561: InitOutput (xf86Init.c:738)
by 0x422C18: main (main.c:205)
Location 0x7ff0000c2 is 2 bytes inside local var "filter"
declared at libudev-monitor.c:256, in frame #1 of thread 1
Uninitialised value was created by a stack allocation
at 0x4E30860: ??? (in /lib64/libudev.so.0.10.0)
Thanks to Zdenek Kabelac.
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.
When I tried to boot a system with 256 disks x 4 paths with
device-mapper, udevadm trigger (--type=devices) that was called
from start_udev ended up dumping a core due to a segmentation
fault.
In udev_enumerate_get_list_entry(), if it finds the devices that
should be delayed, it calls syspath_add(). If realloc() in
syspath_add() allocates the required memory at a different memory
address, referring prev->len afterward causes the segmentation
fault.
Signed-off-by: Kei Tokunaga <tokunaga.keiich@jp.fujitsu.com>
We need to preserve the database of network interfaces while we
rename them. Use the kernel's numbers wherever possible, instead
of the device names.
Fix wrong database filenames which contain a '/', translated
from '!' in the kernel name.
Fix segfault for kobject pathes where the subsystem can not be
determined from sysfs.
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.
Harald Hoyer discovered some incorrect behavior while debugging
problems with network interface renaming:
Udev events might be queued for devices which are renamed. A new
device registered the same time may claime the old name and create
a database entry for it. The current rename logic would move over
this databse entry to the wrong device.
CC udev/udevd.o
In file included from udev/udev.h:27,
from udev/udevd.c:47:
./libudev/libudev-private.h: In function ‘udev_selinux_setfscreateconat’:
./libudev/libudev-private.h:230: warning: declaration of ‘dirfd’ shadows a global declaration
/usr/include/dirent.h:224: warning: shadowed declaration is here
Signed-off-by: Yin Kangkai <kangkai.yin@intel.com>
Signed-off-by: Martin Pitt <martin.pitt@ubuntu.com>
Kay Sievers
Thomas Egerer
Kei Tokunaga
Yin Kangkai
Marco d'Itri