Description of problem: Loader no longer shows the mac address during "Networking Device" selection. It worked with yesterdays rawhide but is missing with today. I do not know which component to attribute this change to yet. Booting into rawhide-20090520 to perform a network install.. ┌─────────┤ Networking Device ├──────────┐ │ │ │ You have multiple network devices on │ │ this system. Which would you like to │ │ install through? │ │ │ │ eth0 - │ │ eth1 - │ │ eth2 - │ │ │ │ ┌────┐ ┌──────────┐ ┌──────┐ │ │ │ OK │ │ Identify │ │ Back │ │ │ └────┘ └──────────┘ └──────┘ │ │ │ │ │ └────────────────────────────────────────┘ Booting into rawhide-20090519 to perform a network install.. +----------------+ Networking Device +----------------+ | | | You have multiple network devices on this system. | | Which would you like to install through? | | | | eth0 - Ethernet device eth0 - 00:11:25:7e:28:64 | | eth1 - Ethernet device eth1 - 00:0d:60:b9:6b:c8 | | eth2 - Ethernet device eth2 - 00:11:25:7e:28:65 | | | | +----+ +----------+ +------+ | | | OK | | Identify | | Back | | | +----+ +----------+ +------+ | | | | | +-----------------------------------------------------+ I've experienced this on x86_64 and ppc so far
I can reproduce this at will in rawhide. I replace loader on the initrd with a locally built one (no code changes) and it works.
Uh oh, that smacks of environmental differences. Was your local built loader done in a minimal chroot, or just on your regular install file system? I'm wondering if we need some tool brought along that we aren't getting in the chroot?
Switching to RPM_OPT_FLAGS for CFLAGS makes it reproducible. Patch posted, although it's really not obvious to me why this would fail now.
http://git.fedorahosted.org/git/?p=anaconda.git;a=commitdiff;h=e8a00c3a91072e9d00b877f70266735375b9bdbf
Bill Nottingham, There's an off-by-one error at your link: http://git.fedorahosted.org/git/?p=anaconda.git;a=commitdiff;h=e8a00c3a91072e9d00b877f70266735375b9bdbf Filling the buffer with 64 nulls, then reading in (up to) 64 bytes, could overwrite all the nulls. This code needs to either: 1) use the return value of read() instead of strlen() 2) make the buffer (and memset()) one byte bigger (65 bytes) or 3) try reading only 63 bytes. Take care, - Jason
In practice, the sysfs files are never that long. But defensive programming, etc., yes.
No longer observing this failure as of anaconda-11.5.0.56. This will be fixed in F11. Thanks!
*** Bug 517123 has been marked as a duplicate of this bug. ***