From: Alex Lyashkov (shadow_at_psoft.net)
Date: Sun 25 Jan 2004 - 04:25:17 GMT
On Saturday 24 January 2004 00:58, Herbert Poetzl wrote:
> On Sat, Jan 24, 2004 at 12:42:12AM +0200, Alex Lyashkov wrote:
> > On Saturday 24 January 2004 00:25, Herbert Poetzl wrote:
> > > Hi Community!
> > >
> > > another quite big step in vserver evolution awaits
> > > your testing/approval ;)
> > >
> > > I had to get 1.3.6 out, before I start attacking the
> > > network and virtualization issues, and this seemed
> > > like a good time, so here it is:
> > >
> > > http://www.13thfloor.at/vserver/d_release/v1.3.6/
> > >
> > > what was done?
> > >
> > > * heavy restructuring (revolution part II)
> > > * memory accounting and resource limits
> > > * xfs iunlink support was added
> > > * finalized the next generation procfs
> > > * inode attribute/xid syscall command
> > > * uts and context name syscall commands
> > >
> > > you can download a all-in-one patche for 2.4.25-pre7
> > > or the tar archive of the broken out patches.
> > > (2.4.25 patches will be available when the kernel
> > > is released ;)
> > >
> > > it should work with older tools, but to utilize
> > > the latest and greatest features, enricos util-vserver
> > > 0.27.199 (or later) package is required ...
> > >
> > > if you want to aid in development, please test it
> > > (for best results, on production like scenarios)
> > > and provide some feedback ...
> > >
> > > enjoy,
> > > Herbert
> >
> > At linux-vserver has bug with selected source address for outgoning
> > packets. It bug exist because not divided routed cache info per context.
> > routing informations about sending packets be stored in routing cache
> > (net/ipv4/route.c; /proc/net/rt_cache) and it`s info be used before any
> > call ip_route_output.
> > its can be tested simple:
>
> # ifconfig eth0 10.0.0.2 netmask 255.255.255.0
>
> > 1) assign to vps address from network other then host system.
>
> # ifconfig eth0:1 192.168.0.2 netmask 255.255.255.0
>
> > 2) send any packet from vps to any host in that network.
>
> # chbind --ip 192.168.0.2 ping -c 3 192.168.0.1
>
> > 3) try to send packet from host system to selected host.
> > packet been have vps source address.
>
> # ping -c 1 10.0.0.1
you try ping different addess - it create new routes in cache.
try ping -c 3 192.168.0.1 from host.
routes for this addess be cached in rt_cashe.
>
> 23:55:21.328698 52:54:0:12:34:56 ff:ff:ff:ff:ff:ff 0806 60: arp who-has
> 10.0.0.1 tell 10.0.0.2 23:55:21.328787 0:ff:e0:f7:ab:81 52:54:0:12:34:56
> 0806 42: arp reply 10.0.0.1 is-at 0:ff:e0:f7:ab:81 23:55:21.333256
> 52:54:0:12:34:56 0:ff:e0:f7:ab:81 0800 98: 10.0.0.2 > 10.0.0.1: icmp: echo
> request (DF) (ttl 64, id 0, len 84) 23:55:21.333398 0:ff:e0:f7:ab:81
> 52:54:0:12:34:56 0800 98: 10.0.0.1 > 10.0.0.2: icmp: echo reply (ttl 64, id
> 5575, len 84)
>
> # uname -a
> Linux (none) 2.4.25-pre7-vs1.3.6 #1 SMP Fri Jan 23 21:33:03 CET 2004 i686
> unknown
>
> sorry, can't reproduce ...
>
> > Simple fix for it - add context tag at rt_key structure. more correctly -
> > full separate it`s cache.
>
> patches against vs1.3.6 and vs1.24 are always welcome ;)
>
for me be difficulty work with 2 brances.
you can backport it`s part from my last release - it use idea about using tag
for separate routing caches.
last snapshot use full separated caches.
-- With best regards, Alex _______________________________________________ Vserver mailing list Vserver_at_list.linux-vserver.org http://list.linux-vserver.org/mailman/listinfo/vserver