[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bug#310652: d-i leaks an entire (reduced) copy of libc during anna run



This is a problem for the GTK fe too, since from my past experience i've notice it requires the full libc: the gtk frontend cannot be started until the full libc has been installed.

Attilio

Joey Hess wrote:
Package: lowmem
Severity: normal
Tags: d-i

This is a stange package to file the bug on, since the real bug is in
d-i, but mostly people doing lowmem will care about this, and a lowmem
hack may be the best/only way to deal with the problem.

d-i boots an initrd containing a reduced libc.so, that's about 640k big.
During the anna run this is replaced by a fullsize libc. Then d-i goes
on and all is well.. except the original libc.so is kept in memory by
all the programs that were started up running it (init, cdebconf, etc).

I've verified this by checking the /proc/nnn/maps files. Just look for
the "(deleted)". During a netinst install, the only such lines by the
end of the anna run are for libc.so.6, ld-linux.so, libdl.so.2. I
imagine it might be worse if any other libraries are out of sync between
the initrd and udebs and get upgraded though.

I can't think of a clean way to fix this, but a lowmem hack that forces
d-i to restart after anna might be worth coding up, to reclaim this >
640k of memory.




Reply to: