Re: [gentoo-dev] call for testers: udev predictable network interface names
On Thu, Jan 17, 2013 at 9:51 AM, Ian Stakenvicius a...@gentoo.org wrote: On 16/01/13 09:55 PM, Rich Freeman wrote: SUBSYSTEM==tty, DRIVERS==pl2303, KERNELS==4-1:1.0, KERNEL==ttyUSB*, SYMLINK=mythser/rca1 I'm not sure if rules are additive - if these symlinks would show up in addition to whatever other ones are created by other rules... I should look this up before making an authoritative response but I believe that SYMLINK= would mean no, it's not additive. If you changed that to SYMLINK+= then it would be additive. That worked. Looks like /dev/serial/by-path would accomplish what I ended up doing. The by-id directory only lists one of my two serial devices. I suspect this is because the devices are completely identical, aside from being plugged into two different ports. Rich
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Fri, Jan 18, 2013 at 10:07:42AM -0500, Ian Stakenvicius wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 18/01/13 09:54 AM, William Hubbs wrote: On Fri, Jan 18, 2013 at 08:33:13AM -0500, Ian Stakenvicius wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 18/01/13 07:24 AM, viv...@gmail.com wrote: Since for servers predictable names are useful and for desktop (which usually have only one ethernet that never change) Is it possible to set desktop profiles to still use ethX, and base profile to use new naming scheme? For wireless situation may be different, many of them are external, could wireless be managed differently? In short, no. At least, not unless the functionality that is currently a configure-time thing is changed into a build-time/install-time thing controlled via a use flag. Actually,this is how I set you up by dropping the file in /etc/udev/rules.d/80-net-name-slot.rules. Nothing changes on your system unless you remove this file and do not have 70-persistent-net.rules. William ..right, but default behaviour can't be changed automatically depending on what profile you're on, as vivo requested, since profiles don't control configuration (just use flags) Right, and we have a policy against using use flags to control the installation of configuration files. vivo, what is your concern here exactly? William pgpnT__EsTSSb.pgp Description: PGP signature
Re: [gentoo-dev] call for testers: udev predictable network interface names
2013/1/19 William Hubbs willi...@gentoo.org On Fri, Jan 18, 2013 at 10:07:42AM -0500, Ian Stakenvicius wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 18/01/13 09:54 AM, William Hubbs wrote: On Fri, Jan 18, 2013 at 08:33:13AM -0500, Ian Stakenvicius wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 18/01/13 07:24 AM, viv...@gmail.com wrote: Since for servers predictable names are useful and for desktop (which usually have only one ethernet that never change) Is it possible to set desktop profiles to still use ethX, and base profile to use new naming scheme? For wireless situation may be different, many of them are external, could wireless be managed differently? In short, no. At least, not unless the functionality that is currently a configure-time thing is changed into a build-time/install-time thing controlled via a use flag. Actually,this is how I set you up by dropping the file in /etc/udev/rules.d/80-net-name-slot.rules. Nothing changes on your system unless you remove this file and do not have 70-persistent-net.rules. William ..right, but default behaviour can't be changed automatically depending on what profile you're on, as vivo requested, since profiles don't control configuration (just use flags) Right, and we have a policy against using use flags to control the installation of configuration files. vivo, what is your concern here exactly? William My concern was to make simple desktop users happy while leaving the servers safe. The answers given in the previous emails are satisfying, since they cover exhaustively what is in place and what could be (or not) done. Thanks, Francesco
Re: [gentoo-dev] call for testers: udev predictable network interface names
Il 09/01/2013 23:13, William Hubbs ha scritto: All, as you probably know by now, udev-197 has hit the tree. This new version implements a new feature called predictable network interface names [1], which I have currently turned off for live systems, because it will require migration on the part of the user. When you upgrade to this new version of udev, you will find a file /etc/udev/rules.d/80-net-name-slot.rules on your system. It currently has comments explaining what is happening. As long as this file is in place, this feature is not activated. That is why there is not a news item. If you do nothing, nothing changes. What I would like to do is find some people who are willing to migrate and report any issues they find. I would like this to be the default for everyone at some point, so I want to document the migration process and find out if there are any bugs in tools because they expect the eth* names. Thoughts? William [1] http://www.freedesktop.org/wiki/Software/systemd/PredictableNetworkInterfaceNames Since for servers predictable names are useful and for desktop (which usually have only one ethernet that never change) Is it possible to set desktop profiles to still use ethX, and base profile to use new naming scheme? For wireless situation may be different, many of them are external, could wireless be managed differently?
Re: [gentoo-dev] call for testers: udev predictable network interface names
-BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 18/01/13 07:24 AM, viv...@gmail.com wrote: Since for servers predictable names are useful and for desktop (which usually have only one ethernet that never change) Is it possible to set desktop profiles to still use ethX, and base profile to use new naming scheme? For wireless situation may be different, many of them are external, could wireless be managed differently? In short, no. At least, not unless the functionality that is currently a configure-time thing is changed into a build-time/install-time thing controlled via a use flag. -BEGIN PGP SIGNATURE- Version: GnuPG v2.0.19 (GNU/Linux) iF4EAREIAAYFAlD5TxkACgkQ2ugaI38ACPCQHAD7BEIoXLuskCfv/TllbCDaW94u 84t/PufZ03LJLjqzWlAA/Azuvil7oLWAzTxSDuHT+oheJsPvf4tBFmQUojSf+WIj =FOCB -END PGP SIGNATURE-
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Fri, Jan 18, 2013 at 08:33:13AM -0500, Ian Stakenvicius wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 18/01/13 07:24 AM, viv...@gmail.com wrote: Since for servers predictable names are useful and for desktop (which usually have only one ethernet that never change) Is it possible to set desktop profiles to still use ethX, and base profile to use new naming scheme? For wireless situation may be different, many of them are external, could wireless be managed differently? In short, no. At least, not unless the functionality that is currently a configure-time thing is changed into a build-time/install-time thing controlled via a use flag. Actually,this is how I set you up by dropping the file in /etc/udev/rules.d/80-net-name-slot.rules. Nothing changes on your system unless you remove this file and do not have 70-persistent-net.rules. William pgpLqFtcVV6CJ.pgp Description: PGP signature
Re: [gentoo-dev] call for testers: udev predictable network interface names
-BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 18/01/13 09:54 AM, William Hubbs wrote: On Fri, Jan 18, 2013 at 08:33:13AM -0500, Ian Stakenvicius wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 18/01/13 07:24 AM, viv...@gmail.com wrote: Since for servers predictable names are useful and for desktop (which usually have only one ethernet that never change) Is it possible to set desktop profiles to still use ethX, and base profile to use new naming scheme? For wireless situation may be different, many of them are external, could wireless be managed differently? In short, no. At least, not unless the functionality that is currently a configure-time thing is changed into a build-time/install-time thing controlled via a use flag. Actually,this is how I set you up by dropping the file in /etc/udev/rules.d/80-net-name-slot.rules. Nothing changes on your system unless you remove this file and do not have 70-persistent-net.rules. William ..right, but default behaviour can't be changed automatically depending on what profile you're on, as vivo requested, since profiles don't control configuration (just use flags) -BEGIN PGP SIGNATURE- Version: GnuPG v2.0.19 (GNU/Linux) iF4EAREIAAYFAlD5ZT4ACgkQ2ugaI38ACPCECQD6A78Wgm30Tx0RIfgblZhAu4d2 /2NFMtZng4JQlgmbCc8BAJZgPOgH3fxhSl+pRBpWFkZu/v5kwqxs+h+9ooBJZ5nG =MhsO -END PGP SIGNATURE-
Re: [gentoo-dev] call for testers: udev predictable network interface names
On 17/01/13 04:49, Greg KH wrote: On Wed, Jan 16, 2013 at 06:36:59AM -0500, Rich Freeman wrote: On Tue, Jan 15, 2013 at 10:42 PM, Peter Stuge pe...@stuge.se wrote: Rich Freeman wrote: Not that anybody is taking requests, but it would be really handy if serial ports were deterministically labeled. Does /dev/serial/* solve the problem? I don't see this directory at all on my system. Do you have a usb-serial device plugged in? You need a serial device for it to show up, and you need to be using udev. greg k-h Right, I have 3G Huawei USB modem attached and I see: $ ls /dev/serial/* /dev/serial/by-id: usb-Huawei_Technologies_HUAWEI_Mobile-if00-port0 usb-Huawei_Technologies_HUAWEI_Mobile-if03-port0 usb-Huawei_Technologies_HUAWEI_Mobile-if04-port0 /dev/serial/by-path: pci-:00:1d.0-usb-0:1.2:1.0-port0 pci-:00:1d.0-usb-0:1.2:1.4-port0 pci-:00:1d.0-usb-0:1.2:1.3-port0
Re: [gentoo-dev] call for testers: udev predictable network interface names
-BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 16/01/13 09:55 PM, Rich Freeman wrote: SUBSYSTEM==tty, DRIVERS==pl2303, KERNELS==4-1:1.0, KERNEL==ttyUSB*, SYMLINK=mythser/rca1 I'm not sure if rules are additive - if these symlinks would show up in addition to whatever other ones are created by other rules... I should look this up before making an authoritative response but I believe that SYMLINK= would mean no, it's not additive. If you changed that to SYMLINK+= then it would be additive. -BEGIN PGP SIGNATURE- Version: GnuPG v2.0.19 (GNU/Linux) iF4EAREIAAYFAlD4D98ACgkQ2ugaI38ACPBvJQD/dFlhO8q9voNAMedF1TBIyEK8 /IXoXUjuWMxwaBrDlSwA/i8wB6BfkWyVopPDboikcl1K37hFrEhE3npaLbIhrtbX =HA4k -END PGP SIGNATURE-
Re: [gentoo-dev] call for testers: udev predictable network interface names
Tobias Klausmann wrote: It has been rather nifty that if I walk up to a random machine with exactly one NIC (that I've been asked to examine/fix), I _know_ that there will be eth0 and only that. Only as long as that system hasn't seen *another* NIC first, if it has persistent interface name udev rules. //Peter
Re: [gentoo-dev] call for testers: udev predictable network interface names
Hi! On Thu, 17 Jan 2013, Peter Stuge wrote: Tobias Klausmann wrote: It has been rather nifty that if I walk up to a random machine with exactly one NIC (that I've been asked to examine/fix), I _know_ that there will be eth0 and only that. Only as long as that system hasn't seen *another* NIC first, if it has persistent interface name udev rules. I was talking about strictly kernel order vs. predictable-net. Persistent-net has VM-related downsides as pointed out in the udev page about the whole thing. Regards, Tobias
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Thu, Jan 17, 2013 at 07:44:39PM +0100, Tobias Klausmann wrote: Hi! On Thu, 17 Jan 2013, Peter Stuge wrote: Tobias Klausmann wrote: It has been rather nifty that if I walk up to a random machine with exactly one NIC (that I've been asked to examine/fix), I _know_ that there will be eth0 and only that. Only as long as that system hasn't seen *another* NIC first, if it has persistent interface name udev rules. I was talking about strictly kernel order vs. predictable-net. Persistent-net has VM-related downsides as pointed out in the udev page about the whole thing. The problem is the kernel names are not dependable. If you have one network card right now, sure, it will be eth0. But, suppose you buy another network card and plug it into the system. Now you have no way to know that eth0 will refer to the card you think it does. With the predictable names, on my system for example, I know that enp1s5 will always refer to the same nic, even if I put a new one in the box. William pgpv8MI42FXdP.pgp Description: PGP signature
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Tue, Jan 15, 2013 at 10:42 PM, Peter Stuge pe...@stuge.se wrote: Rich Freeman wrote: Not that anybody is taking requests, but it would be really handy if serial ports were deterministically labeled. Does /dev/serial/* solve the problem? I don't see this directory at all on my system. Rich
Re: [gentoo-dev] call for testers: udev predictable network interface names
Hi! On Tue, 15 Jan 2013, Greg KH wrote: So anyone who relies on network names right now to be deterministic, and you have more than one network device in your system, should seriously reconsider how they are naming their devices, as it will not work if you only rely on the kernel. You might have gotten lucky for the past 5 years, but you never know what could happen if you reboot today. Seriously, I've seen it happen all the time. It has been rather nifty that if I walk up to a random machine with exactly one NIC (that I've been asked to examine/fix), I _know_ that there will be eth0 and only that. OTOH, maybe it's a good idea to make admins do ip link sh and ip addr sh every time they examine a new computer -- it goes a long way to root out wrong assumptions in that field. Regards, Tobias PS: Do not use ifconfig. Ever. Except if there's no iproute. And then you should only use ifconfig to enable downloading of iproute :)
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Wed, Jan 16, 2013 at 10:19 AM, Tobias Klausmann klaus...@gentoo.org wrote: Hi! On Tue, 15 Jan 2013, Greg KH wrote: So anyone who relies on network names right now to be deterministic, and you have more than one network device in your system, should seriously reconsider how they are naming their devices, as it will not work if you only rely on the kernel. You might have gotten lucky for the past 5 years, but you never know what could happen if you reboot today. Seriously, I've seen it happen all the time. It has been rather nifty that if I walk up to a random machine with exactly one NIC (that I've been asked to examine/fix), I _know_ that there will be eth0 and only that. OTOH, maybe it's a good idea to make admins do ip link sh and ip addr sh every time they examine a new computer -- it goes a long way to root out wrong assumptions in that field. Regards, Tobias PS: Do not use ifconfig. Ever. Except if there's no iproute. And then you should only use ifconfig to enable downloading of iproute :) I would actually like to see iproute2 added to the system set.
Re: [gentoo-dev] call for testers: udev predictable network interface names
Il 16/01/2013 17:25, Mike Gilbert ha scritto: On Wed, Jan 16, 2013 at 10:19 AM, Tobias Klausmann klaus...@gentoo.org wrote: Hi! On Tue, 15 Jan 2013, Greg KH wrote: So anyone who relies on network names right now to be deterministic, and you have more than one network device in your system, should seriously reconsider how they are naming their devices, as it will not work if you only rely on the kernel. You might have gotten lucky for the past 5 years, but you never know what could happen if you reboot today. Seriously, I've seen it happen all the time. It has been rather nifty that if I walk up to a random machine with exactly one NIC (that I've been asked to examine/fix), I _know_ that there will be eth0 and only that. OTOH, maybe it's a good idea to make admins do ip link sh and ip addr sh every time they examine a new computer -- it goes a long way to root out wrong assumptions in that field. Regards, Tobias PS: Do not use ifconfig. Ever. Except if there's no iproute. And then you should only use ifconfig to enable downloading of iproute :) I would actually like to see iproute2 added to the system set. additionally (or indipendently) I would like to see it in bin instead of sbin
Re: [gentoo-dev] call for testers: udev predictable network interface names
On 01/16/2013 05:25 PM, Mike Gilbert wrote: It has been rather nifty that if I walk up to a random machine with exactly one NIC (that I've been asked to examine/fix), I _know_ that there will be eth0 and only that. ++ I would actually like to see iproute2 added to the system set. ++ -- Michael Weber Gentoo Developer web: https://xmw.de/ mailto: Michael Weber x...@gentoo.org
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Wed, Jan 16, 2013 at 06:36:59AM -0500, Rich Freeman wrote: On Tue, Jan 15, 2013 at 10:42 PM, Peter Stuge pe...@stuge.se wrote: Rich Freeman wrote: Not that anybody is taking requests, but it would be really handy if serial ports were deterministically labeled. Does /dev/serial/* solve the problem? I don't see this directory at all on my system. Do you have a usb-serial device plugged in? You need a serial device for it to show up, and you need to be using udev. greg k-h
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Wed, Jan 16, 2013 at 9:49 PM, Greg KH gre...@gentoo.org wrote: On Wed, Jan 16, 2013 at 06:36:59AM -0500, Rich Freeman wrote: On Tue, Jan 15, 2013 at 10:42 PM, Peter Stuge pe...@stuge.se wrote: Rich Freeman wrote: Not that anybody is taking requests, but it would be really handy if serial ports were deterministically labeled. Does /dev/serial/* solve the problem? I don't see this directory at all on my system. Do you have a usb-serial device plugged in? You need a serial device for it to show up, and you need to be using udev. Yes, I have two plugged in and they're working fine. However, perhaps my custom rules are preventing them from showing up: SUBSYSTEM==tty, DRIVERS==pl2303, KERNELS==4-1:1.0, KERNEL==ttyUSB*, SYMLINK=mythser/rca1 SUBSYSTEM==tty, DRIVERS==pl2303, KERNELS==3-3:1.0, KERNEL==ttyUSB*, SYMLINK=mythser/rca2 I'm not sure if rules are additive - if these symlinks would show up in addition to whatever other ones are created by other rules, or if these would be exclusive. I hard-coded them to specific physical USB ports so that they would be persistent. If I plug them in elsewhere they still get ttyUSBn devices, but no symlinks. Rich
Re: [gentoo-dev] call for testers: udev predictable network interface names
-BEGIN PGP SIGNED MESSAGE- Hash: SHA256 Hi all, I respect both sides of the discussion, because: a) I once set up an old P3-700 with 5+1 eth cards in 6 different networks as (bridging)router and truly benefited from the ability to change a broken NIC - which happened quite often due scrap-metal hardware - without ending up with martian packages, dhcp service on the wrong places. But that was 1 incident in 10 years. b) I use multi-nic servers, some with onboard and extention NICs c) I tend to move my setups (esp. my laptop) around between different hardware (nearly identical thinkpad R61/X61), and I _share_ my installation with other/new users by cloning my disc (well rsync), lets call this stageN installation. d) I abuse an old multiport GBit card as GBit switch in my desktop, besides an onboard one. e) Some distro/driver constellations (archlinux?) tend to name their wireless lan eth*. This resulted in one decision per setup, whether or not to set /etc/conf.d/udev's persistent_net_disable=yes persistent_cd_disable=yes Either to avoid random names due hardware replacement (a) or changed module loading order (b, inside debian initrd) or to just use kernel names (eth0, wlan0) because no other cards present (c) or the NIC drivers compiled into the kernel (d). e) never happened to me. It always bugged me to fix/reboot systems which needlessly end up with eth1/wlan1 because some stupid pre-persistent_net_disable did not recognize beeing run on an entirely different hardware. So can we just watch out for the disable=yes setting and migrate it during udev's pkg_install phases __and__ post an big fat warning (elog, news item) on the wall? I assume most linux users do not operate servers/multi-nic/multi-networking setups, do not clone their setups to other hardware. Given that, these user will almost only see the 'my nics changed names and i cannot connect to the internet' errors due some moronic or unavoidable change in initrd/module loading. That might be the driving force behind udev persistence in the first place. I'd be glad if I we respect setups w/ custom-built kernels, w/o initrds, roots capable of choosing network-name-persistence iff needed, users adoring the possibility of just dd(1)'ing installations to new hardware without reinstalling or entering an new product code. rant=1; And I'd like to avoid dozends of conversations like Yeah, your setup/firewall/rouing/... command no longer works, eth0 is no enp0xx2_at_home_lid_open or was it _bluetooth_turned_off. Didn't you read the post on some derps mailing list. with haunted people not knowing better than asking me about their problems. Not to mention all online documentation/forum posts referring to eth0. rant=0; Keep up the good work! Michael - -- Michael Weber Gentoo Developer web: https://xmw.de/ mailto: Michael Weber x...@gentoo.org -BEGIN PGP SIGNATURE- Version: GnuPG v2.0.19 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/ iF4EAREIAAYFAlD1FmAACgkQknrdDGLu8JA68wD/Vuw8mL7O0T398QR7OetqDoLN pQ7kJz9nveemDxw7o9MBAJSsyQ/DWIKLsqudXjlXhTPQEd0Od6vDBEL6IeFtXCjc =AfSI -END PGP SIGNATURE-
Re: [gentoo-dev] call for testers: udev predictable network interface names
-BEGIN PGP SIGNED MESSAGE- Hash: SHA256 Hi, This can have serious security implications [1] For whom? The often cited end user not running any network service, not even sshd? Without firewalls, routing or dhcp_d_? Some avahi-discovery woodoo stuff unaware of network topology at all? Maybe the M$/Windows mechanism asking the user to classify an newly discovered network as (and shutting down network communication until done so) isn't the worst solution at all. (Well, that would need an dbus like service to pop up this box *hihi*) [Generally speaking] Linux developed from an highly specialized group of users to an broad spectrum from I have control, leave my unique setup alone to I have no idea what I'm doing/I'm unwilling to read/Lets sudo random search results kinda users. Not all are enlightened. Good part is the media coverage, money invested/wasted/... Hard part is to find an compromise for all users. So lets provide something that works w/o interaction or master knowledge and not annoys the crap out of users - for all users. [about NIC names] Changing the netdev names way from eth*/wlan*/wwan*/ results in a hell of obsolete documentation. Opt-out urges users into either adapt their setups or disable the rules. This LAN/WLAN eth0/eth1 mess could be fixed by assuring Wi-Fi NICs being called wlan*, and running WPA stuff just there. The upcoming UMTS/broadband interfaces are called wwan*? *duck* Last point - as long as identification of LAN networks isn't handled properly, the consistency of NIC names it the lesser security concern for users carring around their laptops. Enough! Michael [1] http://www.freedesktop.org/wiki/Software/systemd/PredictableNetworkInterfaceNames On 01/09/2013 11:13 PM, William Hubbs wrote: All, as you probably know by now, udev-197 has hit the tree. This new version implements a new feature called predictable network interface names [1], which I have currently turned off for live systems, because it will require migration on the part of the user. When you upgrade to this new version of udev, you will find a file /etc/udev/rules.d/80-net-name-slot.rules on your system. It currently has comments explaining what is happening. As long as this file is in place, this feature is not activated. That is why there is not a news item. If you do nothing, nothing changes. What I would like to do is find some people who are willing to migrate and report any issues they find. I would like this to be the default for everyone at some point, so I want to document the migration process and find out if there are any bugs in tools because they expect the eth* names. Thoughts? William [1] http://www.freedesktop.org/wiki/Software/systemd/PredictableNetworkInterfaceNames - -- Michael Weber Gentoo Developer web: https://xmw.de/ mailto: Michael Weber x...@gentoo.org -BEGIN PGP SIGNATURE- Version: GnuPG v2.0.19 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/ iF4EAREIAAYFAlD1HmkACgkQknrdDGLu8JDLRQD+P0pO8z0WHnELVYOgQrEQi0wm Xp1kG1pQhYTCN271T6EBAJvRSacaBE7hdIaTCRH7VUoeugWdktQaXE935kqhFCNV =BWkO -END PGP SIGNATURE-
Re: [gentoo-dev] call for testers: udev predictable network interface names
-BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 15/01/13 04:16 AM, Michael Weber wrote: Hi, This can have serious security implications [1] For whom? I think the idea there is that a user expects eth0 and eth1 to stay the same, writes iptables rules on a per-interface basis to control what they want, then update the kernel or make some other change (upgraded udev, maybe? :D) which swaps them around and poof, the rules they thought were correct don't end up protecting them they way they assumed it would... Not saying this is necessarily valid, just saying how I interpreted their meaning of serious security implications. [about NIC names] ... Opt-out urges users into either adapt their setups or disable the rules. Unless i'm mistaken (and i haven't done any sort of comprehensive search so I could be), I believe the majority of package rollouts for systemd-udev is going to provide an opt-in rather than an opt-out. I understand the general point here, that systemd-udev upstream perhaps should also be defaulting to an opt-in, but there isn't a whole lot of benefit in making that point on the gentoo ML.. :) -BEGIN PGP SIGNATURE- Version: GnuPG v2.0.19 (GNU/Linux) iF4EAREIAAYFAlD1YKMACgkQ2ugaI38ACPA8OgEAtK1Y3vHB3oBQyAdmZHYFZcBW 4g9ry2YFts41Zu1wuXcA/REe9lunWnLQ9w4uZNxvFnZ0LqEK9lMrOP0pJEr3UHAq =06X2 -END PGP SIGNATURE-
Re: [gentoo-dev] call for testers: udev predictable network interface names
-BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 15/01/13 03:42 AM, Michael Weber wrote: e) Some distro/driver constellations (archlinux?) tend to name their wireless lan eth*. [...] e) never happened to me. It has for me, but not for a *LONG* time -- iirc it was prior to 2.6.16 and I think it was with an (externally compiled) ipw2100 driver. Neither of which are supported now in general, and certainly not with a current udev. I think (e) has been sorted out long ago in the kernel. -BEGIN PGP SIGNATURE- Version: GnuPG v2.0.19 (GNU/Linux) iF4EAREIAAYFAlD1YbYACgkQ2ugaI38ACPC1vQEArVONIEOlLPrvd4PV7NnXszOg AOTxveWpT5drCAV681sA/1WuQwKaqnvfoZReEedNk6Uthedp8dSSIVyvsYaEj0Ud =0Hnr -END PGP SIGNATURE-
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Tue, Jan 15, 2013 at 08:58:59AM -0500, Ian Stakenvicius wrote: On 15/01/13 04:16 AM, Michael Weber wrote: Hi, This can have serious security implications [1] For whom? I think the idea there is that a user expects eth0 and eth1 to stay the same, writes iptables rules on a per-interface basis to control what they want, then update the kernel or make some other change (upgraded udev, maybe? :D) which swaps them around and poof, the rules they thought were correct don't end up protecting them they way they assumed it would... Not saying this is necessarily valid, just saying how I interpreted their meaning of serious security implications. Yes, that is true. And it's not udev that could rename the interface (hint, it wouldn't), it's the kernel, it _never_ guarantees the same interface name every time you boot. You might just be getting lucky, but really, PCI busses can be enumerated in different ways, USB devices can come and go and initialize sometimes slower one boot from another, and lots of other things can happen. So anyone who relies on network names right now to be deterministic, and you have more than one network device in your system, should seriously reconsider how they are naming their devices, as it will not work if you only rely on the kernel. You might have gotten lucky for the past 5 years, but you never know what could happen if you reboot today. Seriously, I've seen it happen all the time. Hope this helps explain things a bit better. greg k-h
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Tue, Jan 15, 2013 at 1:58 PM, Greg KH gre...@gentoo.org wrote: And it's not udev that could rename the interface (hint, it wouldn't), it's the kernel, it _never_ guarantees the same interface name every time you boot. You might just be getting lucky, but really, PCI busses can be enumerated in different ways, USB devices can come and go and initialize sometimes slower one boot from another, and lots of other things can happen. Not that anybody is taking requests, but it would be really handy if serial ports were deterministically labeled. I ended up having to hack my udev rules to hard-code a symlink a USB serial device to a specific hardware USB port. It has broken once or twice over the years, but has otherwise been reliable. Otherwise, if you have more than one USB serial interface there is no way to know which one will end up with what minor number, which is a bummer if they aren't hooked up to the same thing. Rich
Re: [gentoo-dev] call for testers: udev predictable network interface names
Rich Freeman wrote: Not that anybody is taking requests, but it would be really handy if serial ports were deterministically labeled. Does /dev/serial/* solve the problem? //Peter
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Wed, Jan 09, 2013 at 11:33:42PM -0500, Rich Freeman wrote: On Wed, Jan 9, 2013 at 11:21 PM, Daniel Campbell dlcampb...@gmx.com wrote: So long as users retain the choice of keeping eth* or wlan*, no complaints from me. I (and others) came to Gentoo to get away from systemd, and this smells of a systemd-ism. Will eudev be pursuing this as well? Keep in mind that this is a udev announcement, not a eudev announcement. Udev is generally going to follow upstream, so if avoiding systemd is your main goal in life you probably will want to stick with eudev, which might or might not adopt this feature. For the record, I have no plans of forcing systemd on anyone. I still maintain OpenRC and plan to continue doing so. As described on the wiki, it is very simple to turn this feature off either by adding your own persistent rules in /etc/udev/rules.d or by overriding the 80-net-slot-name.rules file by putting a file in /etc/udev/rules.d with that name. So, how is this a systemd-ism? a lot of software has defaults that you can reconfigure. William pgpNpURpjZgz8.pgp Description: PGP signature
Re: [gentoo-dev] call for testers: udev predictable network interface names
-BEGIN PGP SIGNED MESSAGE- Hash: SHA256 On 09/01/13 11:21 PM, Daniel Campbell wrote: [1] http://www.freedesktop.org/wiki/Software/systemd/PredictableNetworkInterfaceNames So long as users retain the choice of keeping eth* or wlan*, no complaints from me. I (and others) came to Gentoo to get away from systemd, and this smells of a systemd-ism. Will eudev be pursuing this as well? The eudev team hasn't discussed it yet, but it's on the agenda for our next meeting. I believe that we will be implementing the functionality (probably by default) in the eudev software package, but we will not be enabling this by default in the eudev ebuilds (at least, not any time soon). Also of note, though, is that the eudev package (and ebuild) will still have available (although not by default) the old legacy persistent-net functionality. I am planning to update the rules generator for this to use the same attributes as the new method though, which should be theoretically more reliable than the old attributes. - --- Finally, something that wasn't mentioned yet -- if a user has / still uses a 75-persistent-net.rules from old udev's, or any custom iface-naming rules, then the new 80-net-name-slot.rules will also not take effect on any of the interfaces that were named in these other rules. -BEGIN PGP SIGNATURE- Version: GnuPG v2.0.19 (GNU/Linux) iF4EAREIAAYFAlDuydMACgkQ2ugaI38ACPDaLwD+JCn43am7AkSkz4/7d/IisXAp U9wm1hD2hqjAe2RjAQUBAKGwBTRAcDDx5od26ip99svgnWu6TQw2DKSICWq8BGQd =T9fH -END PGP SIGNATURE-
[gentoo-dev] call for testers: udev predictable network interface names
All, as you probably know by now, udev-197 has hit the tree. This new version implements a new feature called predictable network interface names [1], which I have currently turned off for live systems, because it will require migration on the part of the user. When you upgrade to this new version of udev, you will find a file /etc/udev/rules.d/80-net-name-slot.rules on your system. It currently has comments explaining what is happening. As long as this file is in place, this feature is not activated. That is why there is not a news item. If you do nothing, nothing changes. What I would like to do is find some people who are willing to migrate and report any issues they find. I would like this to be the default for everyone at some point, so I want to document the migration process and find out if there are any bugs in tools because they expect the eth* names. Thoughts? William [1] http://www.freedesktop.org/wiki/Software/systemd/PredictableNetworkInterfaceNames pgpulsp4FiePK.pgp Description: PGP signature
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Wed, 9 Jan 2013 16:13:10 -0600 William Hubbs willi...@gentoo.org wrote: http://www.freedesktop.org/wiki/Software/systemd/PredictableNetworkInterfaceNames This seems like a good thing for some systems. Will there be a news item when 197 (or greater) goes stable informing people that the option is available and if they want to use it they can do so? In my (ordinary user) opinion, there should be. Chris
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Wed, Jan 09, 2013 at 02:59:10PM -0800, Christopher Head wrote: On Wed, 9 Jan 2013 16:13:10 -0600 William Hubbs willi...@gentoo.org wrote: http://www.freedesktop.org/wiki/Software/systemd/PredictableNetworkInterfaceNames This seems like a good thing for some systems. Will there be a news item when 197 (or greater) goes stable informing people that the option is available and if they want to use it they can do so? In my (ordinary user) opinion, there should be. There will definitely be a newsitem before this hits stable. Once we figure out what the migration will involve and if there are any bugs we need to worry about, I want to discuss making this the default setup before we go stable. There is a way for users to opt out if we default this to on, but I think the new naming scheme has advantages over the traditional eth* wlan* etc names. I did the migration myself on my box this afternoon in just a minute or two; it wasn't painful at all. William pgphcw4Aq3cCV.pgp Description: PGP signature
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Wed, 9 Jan 2013 18:13:21 -0600 William Hubbs willi...@gentoo.org wrote: There is a way for users to opt out if we default this to on, but I think the new naming scheme has advantages over the traditional eth* wlan* etc names. I think it should be taken with a grain of salt. The page mentions how it lets you replace a failed NIC without losing its name. But given a simple computer with just one NIC, if the NIC fails and is replaced (perhaps by a different type of NIC in a different slot, or perhaps an onboard NIC disabled in the BIOS and replaced by an add-in), the name could change, while the kernel’s automatically assigned name will not: eth0 (this also applies to a computer with one Ethernet NIC and one wifi NIC: eth0 and wlan0). That fact was never mentioned on the wiki page, even though it applies to a heck of a lot of systems. Perhaps something to include when the Gentoo docs are put together, as part of the balance of reasons to choose one way or the other? Chris
Re: [gentoo-dev] call for testers: udev predictable network interface names
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 01/09/2013 04:13 PM, William Hubbs wrote: All, as you probably know by now, udev-197 has hit the tree. This new version implements a new feature called predictable network interface names [1], which I have currently turned off for live systems, because it will require migration on the part of the user. When you upgrade to this new version of udev, you will find a file /etc/udev/rules.d/80-net-name-slot.rules on your system. It currently has comments explaining what is happening. As long as this file is in place, this feature is not activated. That is why there is not a news item. If you do nothing, nothing changes. What I would like to do is find some people who are willing to migrate and report any issues they find. I would like this to be the default for everyone at some point, so I want to document the migration process and find out if there are any bugs in tools because they expect the eth* names. Thoughts? William [1] http://www.freedesktop.org/wiki/Software/systemd/PredictableNetworkInterfaceNames So long as users retain the choice of keeping eth* or wlan*, no complaints from me. I (and others) came to Gentoo to get away from systemd, and this smells of a systemd-ism. Will eudev be pursuing this as well? -BEGIN PGP SIGNATURE- Version: GnuPG v2.0.19 (GNU/Linux) Comment: Using GnuPG with undefined - http://www.enigmail.net/ iQEcBAEBAgAGBQJQ7kHbAAoJEJUrb08JgYgHJ20H/20G6Pkq+hIB1546UKR/Kti+ VmxaFdi5msWjtor6xzzBaVsaWjdp1ovCHqL1EeuIDg6JUIpeQ2HiAlj9OqaP 9Kg1xATiTw8TKOiGF4r6J1ysfDgFI/K/5CCsMr1Eea6+8m6EUI+yOR5K5xSXZbkR 9Pti3JIrE6t3EkY1EdWguPOGRiiSshjbessNbIzWe/SM/92aDbylQp0ut4DjXn7F XZyPQ+mCzU2tNWVq8HYuqg6xoO1izk6huYWc9jjxwGXfdewxPN6ebng7uDRhIQSK QR5dSLpoLEkrC5aZqmtuz2v5zqxgJWz30uNZl6JG8dCrdAntB9JmYwFTVkpekHY= =6v04 -END PGP SIGNATURE-
Re: [gentoo-dev] call for testers: udev predictable network interface names
On Wed, Jan 9, 2013 at 11:21 PM, Daniel Campbell dlcampb...@gmx.com wrote: So long as users retain the choice of keeping eth* or wlan*, no complaints from me. I (and others) came to Gentoo to get away from systemd, and this smells of a systemd-ism. Will eudev be pursuing this as well? Keep in mind that this is a udev announcement, not a eudev announcement. Udev is generally going to follow upstream, so if avoiding systemd is your main goal in life you probably will want to stick with eudev, which might or might not adopt this feature. You might want to take discussion of eudev planned features to its dedicated list. Rich
Re: [gentoo-dev] call for testers: udev predictable network interface names
On 01/09/2013 10:33 PM, Rich Freeman wrote: On Wed, Jan 9, 2013 at 11:21 PM, Daniel Campbell dlcampb...@gmx.com wrote: So long as users retain the choice of keeping eth* or wlan*, no complaints from me. I (and others) came to Gentoo to get away from systemd, and this smells of a systemd-ism. Will eudev be pursuing this as well? Keep in mind that this is a udev announcement, not a eudev announcement. Udev is generally going to follow upstream, so if avoiding systemd is your main goal in life you probably will want to stick with eudev, which might or might not adopt this feature. You might want to take discussion of eudev planned features to its dedicated list. Rich My apologies. It wasn't my intent to derail the discussion with my simple yes/no question.
