'apt-get update' segfaults with 03d386a on qemu-user about qemu-m68k HOT 6 CLOSED

This is resolved by reverting the following commits:

a56d4d73b557bdc0171b116886e210827e0ca3cb
8a9587cbd88fdc31cab4e8793f6b15014a3c83d2
211c7bc91d3f2aafbd1cc51f931fcb7e88437b93
5772073d88375164286614e05242a6a0c43bceb6

which are all the commits related to adding netlink support. Although this is already known, I just document it here for others that might run into the same bug.

Adrian

from qemu-m68k.

As of this comment, the relevant commits to revert are now:

65e65b2
7c9a671
f0a0acf
b72a43c

from qemu-m68k.

As I rebase this branch regularly, the commit IDs change, so the title of these commits are:

linux-user: add partial SOL_NETLINK level in setsockopt()/getsockopt()
linux-user: add netlink audit
linux-user: support netlink protocol NETLINK_KOBJECT_UEVENT
linux-user: add rtnetlink(7) support

I plan to do a 680x0-v2.6.0 which should never be rebased (based on qemu-2.6.0). I will remove the netlink stuff from this branch.

from qemu-m68k.

I've found the bugs.

I've a fix for SEGFAULT:
linux-user: fd_trans_host_to_target_data() must process only received data
https://patchwork.ozlabs.org/patch/636628/
linux-user: don't swap NLMSG_DATA() fields
https://patchwork.ozlabs.org/patch/636626/

And I know why "apt-get update" hangs: we byte-swap the data in the guest memory and this corrupts the sequence number (libc/check_pf.c checks the sequence number of the sendto() with the ones of the recvmsg() and the one of the sendto() is corrupted by our byte-swapping).

from qemu-m68k.

Should be fixed in 680x0-master-dev now. Please check.

from qemu-m68k.

I think it's safe to close this one. I haven't seen any issues with apt-get update for a while now.

from qemu-m68k.