**** BEGIN LOGGING AT Fri Oct 26 02:59:59 2018 Oct 26 03:20:59 hi. build an image from git, now have a strange issue. when connecting over ssh, dropbear spikes to 100% cpu usage; connection never gets established. what could be causing this? Oct 26 03:21:31 luci works fine. i have set up rudimentary shell access via netcat for the time being Oct 26 03:22:01 platform is mt7621 (dir-860l) Oct 26 03:22:42 built from nbd's branch, if that matters Oct 26 04:33:42 paulius: ssh keygen? Oct 26 04:34:48 mangix: normally a fresh system generates the keys in a few seconds upon first login? Oct 26 04:35:08 shouldnt take forever Oct 26 04:36:08 "a few seconds" might very well be half a minute or more Oct 26 04:37:49 entropy during early boot is a scarce ressource to begin with - and some targets are much worse in that regards (e.g. mvebu) than others (e.g. ar71xx/ ipq40xx/ ipq806x), which do mix in (bad-) hwrng from the noise received from the wireless cards Oct 26 04:39:16 with all previous builds, and builds by others, it's just a few seconds on this machine Oct 26 04:39:26 mt7621 isn't that slow Oct 26 04:39:49 it doesn't really have much to do with processing power Oct 26 04:40:12 so how come a snapshot image works fine Oct 26 04:40:19 for example Oct 26 04:41:15 i'm not talking about a minute or two either, i'm talking hours Oct 26 04:41:54 unlikely for the ssh keys, but not impossible either for larger amounts of entropy (e.g. ssl certs for VPN uses) Oct 26 04:41:55 untill I rebooted it at least Oct 26 04:42:51 what has dropbear to do with ssl certs and vpns? Oct 26 04:45:54 i'm just trying to SSH into a freshy built openwrt box, and dropbear just hangs. if I kill the dropbear process, another one spawns with 100% cpu usage, even if I'm not trying to connect Oct 26 04:46:29 I have said that gathering entropy can take a long time, a sufficient amount of entropy is needed when reading from /dev/randrom (which is blocking) - and this in turn is needed for generating ssh host keys (dropbear), a relatively small amount, or ssl CAs, certs, etc. - where the amount can be a lot bigger (e.g. 4k RSA keys) Oct 26 04:47:24 /can/, doesn't need to - depending on the platform, star alignments, sun spots, etc. Oct 26 04:49:20 if I download a build off of openwrt.org and install it on the same box, the key generation takes like 2 seconds. if I flash my own build on the exact same box in the exact same environment, it takes hours. Oct 26 04:49:27 how is that? Oct 26 04:50:26 doesn't make sense Oct 26 04:51:29 obviously something in my config is causing the issue. either that or a bug somewhere in the latest commits Oct 26 04:54:58 on the client side, ssh -v -v -v shows that it hangs on this: debug1: expecting SSH2_MSG_KEX_ECDH_REPLY Oct 26 05:01:16 pkgadd: google says people were experiencing this issue on gcc7 as opposed to gcc6. I used gcc8 Oct 26 05:15:41 https://forum.openwrt.org/t/solved-master-mt7621-with-kernel-4-14-dropbear-doesnt-play-ball/11916/11 Oct 26 05:15:45 MTU issue? Oct 26 05:26:30 that last commit link doesn't work for me. Oct 26 05:26:53 I have lots of experience with MTU issues, so I might be able to help if I get more information Oct 26 05:27:03 I also have MT7621 available to me. Oct 26 05:27:10 (device with MT7621) Oct 26 05:28:26 paulius: if you can tcpdump at both ends of the session and send me the PCAP files I might be able to tell you if it's MTU issue Oct 26 05:34:00 paulius: if you can build an image for RB750G3 instead of DIR-860l I can test it for you. It's in the same MT7621 menuconfig as your 860l Oct 26 05:35:19 SwedeMike: sure, I'm kae a build for you Oct 26 05:35:36 make* Oct 26 05:44:30 SwedeMike: https://ufile.io/x3wq9 the zip password is SwedeMike Oct 26 05:48:59 paulius: ok, I'll get back to you in an hour or so with the results, have some things that need to be done first Oct 26 05:49:28 great, thanks! Oct 26 06:10:28 paulius: r8373-bc3d47cd12 works for me Oct 26 06:17:12 paulius: what commit of nbd's Oct 26 06:17:29 and what .config and files overlay (if any) Oct 26 06:18:01 paulius: I'm seeing the same problem you did, and it's not MTU problem from what I can tell. Oct 26 06:18:33 paulius: I see 5 dropbear instances all running 20% CPU each when it's doing that KEX Oct 26 06:18:42 (in top) Oct 26 06:19:45 oh, those were just the old ones that were stuck Oct 26 06:20:07 I killed them and tried again, now I have one dropbear using 25% CPU Oct 26 06:20:08 that's 80% of the CPU that you'll never get back again Oct 26 06:20:27 oh, no, i misread Oct 26 06:20:43 running strace on the dropbear that uses 25% CPU yields nothing Oct 26 06:21:43 hm, this is a quad core machine. Does top work differently here, is 25% one core busy-waiting? Oct 26 06:22:11 that's not what I expected from top Oct 26 06:22:19 dual core, actually with some kind of hyperthreading-ish thing Oct 26 06:22:32 russell--: yes, /proc/cpuinfo reports 5 cores. Oct 26 06:22:33 4 Oct 26 06:23:00 core 0 and core 1 Oct 26 06:24:02 on processor 0,1,2,3 Oct 26 06:24:33 grep core /proc/cpuinfo Oct 26 06:26:24 russell--: ah, nice, wasn't aware that it reported that nowadays. Oct 26 06:46:46 * russell-- wonders if CONFIG_GPIO_WATCHDOG is responsible? Oct 26 06:47:24 just comparing openwrt master (as of recently) to nbd/master Oct 26 09:14:31 in my testing of MT7621 IPv6 routing on RB750GR3 it seems this is not properly hw accelerated. I am getting conflicting reports from others who are testing other MT7621 devices whether IPv6 routing is accelerated or not. Anyone have any insights? When I test on my UBNT ER-X (MT7621) using their OS, IPv6 routing is accelerated. Oct 26 09:50:03 https://git.openwrt.org/?p=openwrt/staging/ldir.git;a=commitdiff;h=eaba9e2b288abae1f3e079153b8bdd0a8c9854de Sane/Insane ? Oct 26 10:00:56 ldir: I would say ACK, but then I notice /usr/share/... for .conf files and it makes me shiver Oct 26 10:01:11 how about /etc/dnsmasq.d include dir and stuff everything in there Oct 26 10:01:28 and can drop a bunch of crap from the init script maybe Oct 26 10:32:10 SwedeMike: sorry, was away. any idea what's causing it? Oct 26 10:33:24 I shall have a think Oct 26 10:35:38 paulius: sorry, no. I didn't see it on my image I compiled on ubuntu 16.04 either. Oct 26 10:40:32 russell--: from Thu Oct 25 18:49:19 2018 +0200 Oct 26 10:42:14 russell--: so this one I guess https://git.openwrt.org/?p=openwrt/staging/nbd.git;a=commit;h=a6d02a757273439133eef70a658c49059c18018a Oct 26 10:42:33 SwedeMike: which GCC/binutils version? Oct 26 10:44:27 paulius: Ubuntu comes with gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.10). The build system downloaded gcc-7.3.0.tar.xz so I presume that's what's being used for cross-compiling? Oct 26 10:45:01 that's what's currently default in menuconfig, yes Oct 26 10:45:12 binutils in ubuntu is 2.26.1-1ubuntu1~16.04.7 Oct 26 10:45:40 build system used binutils-2.31.1.tar.xz Oct 26 10:45:44 i think only the one used in cross-compiling matters Oct 26 10:46:43 okay. guess I'll try recompiling with gcc 7.3. which 'make' cleaning method should I use? Oct 26 10:47:02 so that ccache gets cleared Oct 26 10:47:16 dirclean? Oct 26 10:47:47 that should retain .config, right? Oct 26 10:54:59 bloody sysupgrade failed again bah Oct 26 10:57:56 karlp: happened to me as well yesterday Oct 26 11:25:05 on an unrelated note, the ntfs-3g package seems to be broken Oct 26 11:37:53 hates whatsapp - so confusing. How to add a contact. Oct 26 11:40:01 it just uses system contacts? Oct 26 11:40:21 it's just a really shit interface. Oct 26 11:40:31 * karlp loves it, it's why I started using it. Oct 26 11:40:49 so much better than snapshit Oct 26 11:44:38 ok, so I've been invited to a group chat. Other paricipants appear as phone numbers, or in some cases as a number & a name, I'd like to add the number & name to my contacts so it knows who this person is 'properly'. How? The help is useless. Oct 26 11:44:59 Thing is, I managed to do it yesterday, but today I can't. Oct 26 11:46:29 hrm, never tried that, only been in group chats with people I already knew. Oct 26 11:46:44 someone did say to me that they didn't know who "some number" was later though Oct 26 13:06:57 Is anyone going to OpenWRT Summit in Lisbon/Portugal next week ? Oct 26 13:08:44 SwedeMike: pkgadd: my dropbear problem went away after changing -O3 back to -Os in the target build options. GCC bug? Oct 26 13:12:40 used GCC8. Oct 26 13:17:20 stintel: the reality is that those files are already in those locations. Oct 26 13:18:17 falk0n: yes. Oct 26 13:18:39 any idea why Kconfig has problems with dependencies when selecting all package + all kmod? Oct 26 13:19:14 my build fails because kmod-ath depends on iw, but iw is only selected as module, not installed. kmod-ath=y, iw=m Oct 26 13:39:53 ldir: yes, but that doesn't make it less wrong :) Oct 26 13:46:31 falk0n: couple of us, yes Oct 26 13:58:39 stintel: I'd argue that mostly static, almost never changed resources files belong to /usr/share not /etc Oct 26 13:59:59 so, I'd like to be able to make sure radio X is always called 'wiphy0', not phy7, etc. Oct 26 14:01:03 I was thinking I could add a section in /etc/config/wireless to map PCI bus to a name, and then use that in ipq806x/base-files/etc/hotplug.d/ieee80211/10_fix_wifi_mac to set the phy name Oct 26 14:01:17 any suggestions for improvement? Oct 26 14:02:18 since the mac80211 script machinery already identifies the phy to match the configuration, why not rename it directly in the default script? Oct 26 14:02:41 phy name changes on driver reload Oct 26 14:03:04 and what script are you talking about? Oct 26 14:03:28 /lib/netifd/wireless/mac80211.sh Oct 26 14:03:55 in drv_mac80211_setup() it calls find_phy() which populates $phy Oct 26 14:05:39 you could do an `iw phy "$phy" set name "$1"; phy="$1"` there Oct 26 14:05:46 ok, so in that setup() method I could rename it to wiphyX Oct 26 14:06:06 now, how to let user specify what to name it to? Oct 26 14:06:07 yeah, or the uci name ($1) Oct 26 14:06:14 which by default is radio0, radio1 etc. Oct 26 14:06:28 but if you change it in /etc/config/wireless, it'll get reflected in the phy name Oct 26 14:06:52 I'm nervous about changing 'radio1' to wiphy1, and for my app, I need wiphyX Oct 26 14:07:12 so, maybe a phyname=wiphyX in the radio section? Oct 26 14:07:33 yep, that'd work too Oct 26 14:08:01 can you suggest uci command to get/set this? Oct 26 14:08:01 still feels redundant with the radioX identifier though Oct 26 14:08:57 yeah, but there might be some code somewhere that would puke on a name change...and if it is a new option, then at least it would be easy to not configure it and have old behaviour Oct 26 14:09:37 could tweak it later to use radioX by default if my experiment works out OK Oct 26 14:11:08 I'd try this: http://sprunge.us/DiWV3E Oct 26 14:11:34 then uci set wireless.@wifi-device[0].phyname=foobar0 Oct 26 14:13:29 the [0] means first wifi-device entry in the wireless file? Oct 26 14:13:33 yes Oct 26 14:14:58 I guess that is OK, people can just not re-order that file unless they really want different behaviour Oct 26 14:17:00 hmm, that doesn't work, maybe because I have (and want) the radios disabled so that OpenWRT stuff doesn't manage them Oct 26 14:17:23 but I would like it to rename the phys :) Oct 26 14:17:57 then a hotplug script might be the better choice Oct 26 14:18:06 you could store the mac->name map somewhere in /etc Oct 26 14:19:00 and in the hotplug script compare /sys/class/ieee80211/$phyname/macaddress against that /etc file to pull out the desired name Oct 26 14:19:31 alternatively, if you want to keep /etc/config/wireless for that, you'd have to copy that find_phy() code Oct 26 14:19:52 since the openwrt wifi config can reference phys by sysfs path, macaddr or phy name Oct 26 14:22:00 given a phyX, do you know how to create the 'path' setting in the wireless config ? soc/1b700000.pci/pci0001:00/0001:00:00.0/0001:01:00.0 Oct 26 14:22:21 if so, maybe that is the proper thing to match on, since it is more immutable than MAC address I'd guess? Oct 26 14:24:33 hi Oct 26 14:29:49 Ok, I'm going to commit the permissions tweak, and we can argue about file locations later :-) Oct 26 14:34:43 my sell programming is weak...any idea why this doesn't return just the PCI bus ID w/out the PCI_SLOT_NAME= prefix? Oct 26 14:34:53 DEVPATH=/class/ieee80211/phy7 Oct 26 14:34:59 LOT=`grep PCI_SLOT_NAME /sys${DEVPATH}/device/uevent` Oct 26 14:35:07 PCI=${SLOT##*/PCI_SLOT_NAME=} Oct 26 14:35:18 er, SLOT=`grep... Oct 26 14:38:58 lynxis: wigyori I'm trying to go, I work in Lisbon so I'm at "home", and if by any chance I'm not able to go to the summit at least know the guys/girls that also like and work on openwrt Oct 26 14:39:43 So at least to buy a few beers and have fun Oct 26 15:56:00 falk0n: would be nice if you can join us Oct 26 15:56:04 jow, any idea what I am doing wrong here? https://pastebin.com/WAXxG6VS And more importantly, how can I see debugging messages and errors in this code (ie, not in dmesg, so somewhere else?) Oct 26 16:34:46 doh, just need to echo to a /tmp file of course.... Oct 26 17:00:40 lynxis: I got my ticket, it is official I'm able to go so Oct 26 17:24:30 ok, this doesn't work on busybox tail, and I cannot find any 'real' tail offered. Maybe it is in some other package? I'm trying: Oct 26 17:24:44 tail --lines=+11 foo.txt Oct 26 17:26:05 you want last 10, starting from line 11? Oct 26 17:26:27 I want all lines after line 10 Oct 26 17:26:47 tail -h says "tail -n +10" Oct 26 17:27:39 so just plain "tail +10 foo.tx"? Oct 26 17:31:25 I gues so, I originally wrote that code in 2007, so maybe things were different then :) Oct 26 17:34:42 some older fedora needs -n +10, so I'll stick with that Oct 26 17:44:00 wow, 'top' is insane on openwrt! Oct 26 17:44:42 busybox not openwrt Oct 26 17:44:50 sure Oct 26 17:45:10 either way, interesting thing to waste instructions on :) Oct 26 18:23:47 I crashed my system by creating too many vdevs, and file system went read-only, and now on reboot it stays like that: Oct 26 18:23:51 [ 31.942522] UBIFS error (ubi0:1 pid 570): ubifs_iget: failed to read inode 1623, error -2 Oct 26 18:23:51 [ 31.942588] UBIFS error (ubi0:1 pid 570): ubifs_lookup: dead directory entry 'dhcpd', error -2 Oct 26 18:23:51 [ 31.949813] UBIFS warning (ubi0:1 pid 570): ubifs_ro_mode.part.0: switched to read-only mode, error -2 Oct 26 19:57:38 greearb_: I think you b0rked it Oct 26 19:58:11 well yeah, but.... Oct 26 19:58:23 sysupgrade a new image made it work again Oct 26 20:17:14 how can i have realtime connection attempts log to console (which i already have) Oct 26 20:17:26 wifi auth attempts Oct 26 20:23:45 or where is the system log located Oct 26 20:48:11 nm. logread -f | grep hostapd Oct 26 20:51:47 logread -f -e hostapd Oct 26 20:52:49 even better Oct 26 21:02:09 TIL -e Oct 26 21:15:01 what? Oct 26 21:15:45 ahhhhh :-) Oct 26 21:42:14 bugger, after 5 months uptime, my /tmp is full on my work station since mate-term is storing huge amounts of backlog and evidently has no way to clean them w/out killing the process :( Oct 26 22:00:16 anyone ever see a company making OEM cases for something like the Netgear r7800? Oct 26 22:18:38 bah, my laptop is not getting the static DHCP lease configured in /etc/config/dhcp most of the time, and now apparently my new printer also has that problem Oct 26 22:19:24 multiple DHCP servers on the network? Oct 26 22:20:35 nope Oct 26 22:20:45 the lease is written in the dnsmasq leases file with the wrong IP Oct 26 22:21:13 all of a sudden it just says "address not available" Oct 26 22:21:19 * stintel reads code Oct 26 22:21:50 it thinks it had assigned the address already.... Oct 26 22:22:26 I've see that on cisco dhcp services too Oct 26 22:22:38 /* Check for renewal of a lease which is outside the allowed range. */ Oct 26 22:23:55 and while googling I read that staticly configured lease has to be within the dynamic range Oct 26 22:24:26 within work's PXE boot env... really annoying when setting up more than the DHCP range count of servers... even with small batches... Oct 26 22:24:32 option start 100 and option limit 100, that means 192.168.0.100 - 192.168.0.200 Oct 26 22:24:39 and I have configured .47 as static lease Oct 26 22:25:02 but how the heck does it work then for days/weeks and then suddenly complains about it Oct 26 22:25:19 dnsmasq sucks hard than an industrial vacuum cleaner Oct 26 22:25:32 it assigns it... but you can never renew Oct 26 22:26:03 initial request... you have no IP... and it will just look it up in the static table Oct 26 22:26:30 next you try to renew an IP outside the range Oct 26 22:28:09 it renewed fine for weeks Oct 26 22:28:33 maybe it should check MAC before looking at the IP one is trying to renew... but I'm not certain that's what the protocol specifies... Oct 26 22:30:44 http://lists.thekelleys.org.uk/pipermail/dnsmasq-discuss/2018q1/012098.html Oct 26 22:30:54 looks like this might be causing problems Oct 26 22:37:47 hmm no, it's not enabled by default Oct 26 22:40:05 stintel: I'm still on dnsmasq 2.80rc1-2, but all my ~270 static leases are outside the range for the dynamic pool (172.20.0.0/14, option start '10', option limit '230' - static leases like 172.21.1.0 ...) Oct 26 22:40:51 and that is working - and has been working for years Oct 26 22:40:54 pkgadd: yeah, it always worked for me like that, only my laptop has been not getting the static lease IP for quite some time already (don't really remember) Oct 26 22:41:02 and now I notice my new printer does it too Oct 26 22:41:27 it's really not acceptable :/ Oct 26 22:41:36 yep Oct 26 22:45:06 and apparently I cannot reboot the printer via the webinterface :/ Oct 26 22:46:39 ok, I can toggle the switchport to make it do a new DHCP release Oct 26 22:47:06 after stopping dnsmasq, rm'ing the bad lease from the leases file and restarting dnsmasq it gets the correct IP again Oct 26 22:47:08 but for how long :/ Oct 27 00:48:37 Is there an easy way to understand ethernet numbering? I have a Zyxel NBG6817, and plan to set it up as a strictly internal router (no firewall, no significant distinction between the "WAN" ethernet and the "LAN" ethernet ports) Oct 27 00:49:57 why do I have eth0 and eth0.2, and eth1 and eth1.1? Oct 27 00:50:27 Why is the IP address on the .1 or .2 subinterface, and not just on eth0 or eth1? Oct 27 00:51:22 there are two CPU ports connected to the same switch Oct 27 00:52:15 eth1.1 used for LAN, eth0.2 for WAN **** ENDING LOGGING AT Sat Oct 27 03:00:00 2018