00:04:56 does /etc/sysctl.conf.local works just like /etc/sysctl.conf ? 00:15:16 so I've managed to narrow things down regarding these DRM fault errors when using picom. it's picom's vsync stuff that triggers it 00:20:19 hernan, It's supposed to work just the same. The system ships with a template /etc/sysctl.conf so you can use the .local file to avoid modifying it though I modify it here. 00:28:29 rwp: thanks 01:00:09 so I'm wanting to try out 15-CURRENT to see if the drm-61-kmod drivers have this same issue, but I can't seem to jump up to 15-CURRENT through pkgbase, running env ABI=freebsd:15:amd64 pkg-static upgrade -r pkgbase 01:00:36 gives "an error occured while fetching package" and "repository pkgbase has no meta file, using default settings" 01:00:55 eventually just spitting out "unable to update repository pkgbase" and erroring out 01:01:49 gives a warning to consider running pkg bootstrap -f, but that tries to install pkg through the pkgbase repo which fails 01:06:09 it seems like the mirror is giving a 404 error on GET /freebsd:15:amd64/base_weekly/meta.conf 01:10:00 weird, just manually using curl on meta.conf gives a 404 on my laptop, but on my (Linux) desktop I get the file 01:12:27 oh, it's case-sensitive 01:15:32 now it's proceeding. I've heard about CURRENT having a bunch of debug options enabled. is that true for the binaries in pkgbase, or just for the traditional builds? 01:17:32 either way, should be sufficient for testing this 01:23:31 also, if I decide to start tweaking the settings for i915, which sysctl namespace do I use? I'm seeing i915's tunables at both hw.i915kms.* and compat.linuxkpi.i915_* 01:46:01 Woot, first server of the day made the migration from Debian to FreeBSD. Made me think that it'd be cool if the installer could configure bridges out of the box, and also if it could throw and enable in cpu-microcode. 01:46:10 throw in and enable* 01:49:15 so at least first impression of drm-61-kmod is that it doesn't throw the fault errors, though this could just be a case like 510 on 14-STABLE where it seems fine at first and then suddenly it starts to throw them 01:50:05 I don't really want to run CURRENT though 01:50:54 tm512: STABLE is a development branch, remember. 02:05:55 mason: I'm aware, I dunno what point you're trying to make 02:20:33 so nevermind, fault errors with drm-61-kmod as well 02:25:42 I have no idea why these are so inconsistent. ends up getting my hopes up when they don't happen for a couple boots 02:52:16 apparently has nothing to do with vblank scheduling, because disabling frame pacing in picom has no effect on it 03:00:28 and evidently it's not even related to vsync? earlier the errors were going away when I disabled vsync in picom 03:25:06 forcing picom to use DRI2 rather than DRI3 seems to fix the issue, hmm 03:35:30 kind of a garbage solution though since that makes picom take considerably more CPU which basically has this CPU running constantly around 55-60C even without other load 03:45:33 >STABLE is a development branch 03:45:39 yes the terminology is a bit... confusing 03:49:00 I still don't really understand the point of bringing it up. I'm quite familiar with running STABLE 03:49:26 does picom have some advantage vs xorg? 03:49:33 i've never heard of it 03:50:44 it's a compositor for X11, it's not something mutually exclusive with it 03:51:02 it's the only way I've found to reliably fix screen tearing on X11 04:33:19 different topic. i'm struggling to understand how port building works 04:33:39 the Makefiles define a variable WWW to be a github repo. But I can't find a reference to that variable anywhere else 04:39:33 this is in the lsof port. i'm guess MASTER_SITES is just set to ${WWW} somewhere I can't see. 04:41:00 johnjaye: WWW is an informational-only thing 04:41:24 very mysterious. the lsof port has no mention of where to actually get the source code 04:41:27 it used to be in the pkg-descr, but it got moved into a make var to make it easier for some tooling to get at for display or something like that 04:41:42 or rather it has WWW but then doesn't use it. 04:41:46 for lsof it's here: https://cgit.freebsd.org/ports/tree/sysutils/lsof/Makefile#n21 04:41:47 Title: Makefile « lsof « sysutils - ports - FreeBSD ports tree 04:41:52 USE_GITHUB=yes, GH_ACCOUNT=lsof-org 04:42:08 the project defaults to the port name, so you'll probably find it at github.com/lsof-org/lsof 04:42:35 meaning USE_GITHUB is used elsewhere to get the source from the WWW var? 04:42:52 no, WWW isn't used by anything in the process 04:43:05 USE_GITHUB says "it's a github thing, setup master sites and whatnot accordingly" 04:44:28 ok. i can't find that var in /usr/share/mk or in the source tree of bmake anywhere 04:44:42 ports has their own framework built atop bmake 04:44:50 look in Mk/ in the ports repo 04:45:50 wow 04:45:58 this is a correct response :-) 04:46:16 studying bsd ports is like reading a detective novel 04:46:28 every answer leads to another level of question 04:46:48 just don't read bsd.port.mk too hard 04:47:01 i am in fact reading it right now 04:47:18 or rather i spent the last hour reading it carefully 04:47:31 there's some really fun shit in there where a port can include bsd.port.pre.mk and bsd.port.post.mk separately, and parts of bsd.port.mk will be excluded based on which one you've included in which position 04:53:14 i see 04:53:48 i'm also trying to wrap my head around makefile being so complicated 04:54:03 it has a bunch of weird pseudo-shell like features like variables and for each loops 04:55:44 it's also called Mk not mk.... 04:56:38 i guess now i know why there is an entire manual on this 05:00:12 weird, so sometimes, FreeBSD reports the bluetooth chip as "Realtek Bluetooth Radio", but then other times it reports it as a "Realtek 802.11ac" something or other 05:06:27 also, picom seems fine on drm-515-kmod with Xorg using the intel DDX, even with DRI3 enabled. I seem to recall the exact same issue happening regardless of whether I used intel or modesetting, though, both produced fault errors 05:06:37 all of these inconsistencies with this laptop are really getting to me 05:07:44 tm512, Which laptop is it? 05:09:42 a first-gen ThinkPad E14 05:09:58 tm512, Thanks 05:17:58 tm512, Both bluetooth & Wifi most likely are on the same card. One or the other component is likely being identified first 05:19:52 parv: they're on the same card, but they're separate devices. wifi is connected over PCIe, bt over USB 05:41:52 another thing I'm dealing with are these "acpi_ec0: EcCommand: no response to 0x84" messages, seemingly mostly when my system is under load 06:02:00 tm512: The obvious reason to note that it's a development branch is that you should expect inconsistency and failed experiments from time to time, as compared with a RELEASE branch which is intended for production by dint of an expectation of reliability. Nothing too mysterious. 06:05:16 feels kinda dismissive tbh, also these packages are the exact same I'd be using on RELEASE 06:05:22 same exact drm drivers 06:05:25 same version of picom 06:06:08 and from what I understand, STABLE is an "officially supported" branch anyway, unlike CURRENT 06:44:31 mason: also, even if the issues I'm having, particularly with the pipe errors, aren't present in 14-RELEASE (kind of a big if since the drivers aren't in base), it's still important to catch regressions 06:45:37 hrm. so i thought to build a port you go into the port directory and type 'make install' or 'make' 06:46:23 but if it's using variables like USE_GITHUB or SR or what not from the ../Mk dir how is it finding those 06:50:26 for these fault errors I have no idea where the blame lays. I dunno if it's i915kms, which ultimately gets fault errors, or if it's mesa's iris driver, or if it's the Xorg DDX drivers 06:51:16 I don't think it's ultimately picom's fault because it doesn't seem to be doing anything all too weird 09:54:41 is there a way to change the label of an existing fat filesystem without reformatting with newfs_msdos(8) again? 09:58:17 if I'd like PGSQL support in libreoffice do I have any other choice than recompiling the whole stuff ...? 10:42:13 kenrap: dosfslabel/fatlabel, though that seems to be a Linux utility, hrm 10:43:01 ports has it (from dosfstools) but it's for the linuxulator 10:46:17 emulators/mtools can do that (with mlabel) 10:47:37 mage: if there's no subpackage/flavour with postgres support, then yes. or else you could open a bugzilla PR and ask for pgsql to be enabled by default 10:51:24 kenrap: https://www.gnu.org/software/mtools/manual/html_node/mlabel.html 10:51:25 Title: Mtools 4.0.20: mlabel 11:03:45 hi all 11:53:21 damn, GNU to rescue then, thanks lw 11:58:31 *to the rescue 11:59:18 Time for me to pass out, zzz... 12:13:12 Hello guys, when i try e.g. zpool create example /dev/da2 and zpool responds: "cannot use '/dev/da2': must be a block device or regular file 12:13:25 what would i have to do different? 12:21:39 saohh: normally you just use 'da2', not '/dev/da2', but i'd still expect that to work. what does 'ls -l /dev/da2' say? 12:22:02 hello thank you for helping, let me check ;) 12:22:43 da2 does exist in the folder /dev if thats the question xD 12:23:07 saohh: does it work if you just do 'zpool create example da2'? 12:23:15 crw-r----- 1 root operator 0x8e Feb 20 11:41 /dev/da2 12:23:20 this is what it replies 12:23:55 with "just" da2 also same answer (block device or regular file) 12:24:45 are you root? 12:25:01 no but i use sudo to have permissions to run the command 12:25:24 can you paste the entire command you're running? 12:26:13 just as written (simple to test / identify the issue): zpool create example da2 12:26:35 that's not the exact command though since you said you used sudo? 12:28:05 yes i did -.- shame on me somewhere running through handbook etc. i saw so many different ways i lost the sudo in front of the commands -.- 12:28:23 erasing the example pool is sudo zpool delete example ? 12:28:39 bcs. i would like to try a raidz2 pool with 4 disks 12:28:44 (sudo) zpool destroy example 12:28:49 da2, 3, 4 and 5 ;) 12:28:51 ok mom 12:29:17 do you use ZFS yourself? 12:29:55 maybe you can give me a recommendation, i have 6 disks completely for my homeserver and i would like to use ZFS 12:30:16 question is what is the best way of doing that raidz2? bcs. of 2 disks can fail? 12:31:17 yes, i use zfs on my desktop (2 disks in separate pools) and on a fileserver with 8 disks in a raidz2. if you have four disks, assuming they're reasonably large (>2TB) i would go with raidz2, but if you don't mind an increased risk of disk failure on rebuild (e.g. if you have backups or the data or not important) then a mirror-stripe will give better performance 12:31:58 wait, do you have 4 disks or 6 disks? you said both 12:33:19 well xD right now i have 2 SSDs as a mirror for the OS (FreeBSD) and 4 DELL SAS 1.8TB Harddisks but i wait for 2 more chassies to end up having 6 for the data pool (possibly then raidz2) 12:34:30 if i create that raidz2 now with 4 disks can i expand it later by 2 more? 12:35:12 sorry, i need to do a work thing, i'll be back in ~10 mins if no one else answers 12:35:26 sure thank you anyhow you helped me already 12:38:08 saohh: yes, you can add additional devices to a raidz pool, see 'man zpool-attach'. i think (?) this feature is new in freebsd 14.0. i've never tried it so i can't say how well it works 12:38:54 well thats awesome, once i have the additional two caddies with me, i will put the two disks in the server and let you know 12:39:01 we can find out together i run 14.0 ;) 12:39:09 be very careful when doing this because if you add the disks in the wrong way (e.g. adding them as top-level vdevs instead of adding them to the raidz) it's annoying to remove them 12:39:30 (although that's not as bad as it used to be where if you accidentally added a disk you could never remove it) 12:39:55 i will get back here once i have them as i mentioned ;) so we can figure out, the offical way and if it works, maybe interesting for others facing that too 12:40:12 and a knowledgeable guy like you could then confirm that to them directly here ;) 14:12:34 I am back and i received the additional drives 14:20:56 lw are you there? if yes, we could try together adding the two disks now to the raidz2 pool ;) 14:40:11 saohh: i am around but i've never done this before so i might not be the best person to help ... 14:40:43 i tried it zpool attach poolname device_to_add 14:41:01 but it also wants a "new device" as argument, whatever that will be? 14:41:13 attach [-fsw] [-o property=value] 14:42:10 as i have no data on it, its just an experiment, to let you know if it works, if it doesnt it doesnt, i delete the pool and build a new one with all 6 disks 14:42:11 saohh: you need to specify the raidz vdev to attach to, so say you had a pool like this: https://www.le-fay.org/tmp/30d/mR9nDX.txt - the raidz vdev here is "raidz2-0" so you want "zpool attach data raidz2-0 " 14:42:53 (i see this pool has suffered the native encryption bug again, i should get around to rebuilding it from scratch without encryption...) 14:43:37 ah alright mom 14:44:09 let me point out again that i've never actually done this and am just reading the manpage and telling you what i think it says :-) 14:44:19 if there's no data on the pool i guess it doesn't matter if it breaks though 14:44:29 exactly the point 14:44:35 right now nothing is on it ;) 14:45:07 lw, can you read a manpage and tell me what to do? 14:45:25 CrtxReavr: depends, what are you trying to do? have you tried "rm -rf /"? 14:45:37 lw, six times! 14:45:43 funny xP 14:45:47 wow you must have found a bug in rm if that didn't fix the problem 14:45:49 Tried of re-installing BTW. .. FreeBSD sucks! 14:46:07 "rm" = "remove malfunction" btw, it normally fixes all your problem 14:46:28 That's the "read mail real fast" utility, right? 14:46:44 cannot attach da6 to raidz2-0: can only attach to mirrors and top-level disks 14:46:49 hmm 14:47:06 saohh: i'm on 15.0, let me see what the 14.0 manpage says about this, one sec 14:47:46 saohh: ok, seems like 14.0 does not have this feature :-( sorry, i really thought it was in 14 already (ref: https://man.freebsd.org/cgi/man.cgi?query=zpool-attach&apropos=0&sektion=0&manpath=FreeBSD+14.0-RELEASE+and+Ports&arch=default&format=html) 14:47:48 Title: zpool-attach 14:48:32 saohh: so you'd have to destroy the pool and recreate it from scratch with the new disks 14:48:51 but when i do zpool --help it shows attach as a possible option oO 14:48:54 seems like it's not in 14-STABLE either so it won't be in 14.1, that's a shame 14:49:10 saohh: attach has existed forever, but it could only attach devices to a mirror, not a raidz. attaching to a raidz is the new functionality in 15.0 14:49:27 oh holy moly alright 14:49:51 as someone wanting to build a homeserver while learning BSD its better to stick to the stable right? 14:50:07 stick with -RELEASE if you're new, e.g. 14.0-pX 14:50:34 Probably the security branch, which is release+security patches. 14:50:35 yeah i use the 14 release p5 ;) 14:50:38 14-STABLE is okay for experienced users, i would never recommend 15-CURRENT unless you feel comfortable fixing bugs in the source code :-) 14:50:45 Mostly security patches. . . sometimes functionality patches. 14:51:08 "STABLE" is a development branch. . . they try to behave with it, but shit happens. 14:51:28 it is a development branch but at least most of the changes have been tested for a week or two in current, so it's a bit less likely to break 14:51:29 * CrtxReavr is a 13.2p9 luddite. 14:51:54 ok so destroy it is then 14:52:35 CrtxReavr: i think you've been using freebsd a while... remember when we didn't have patch releases and you had to run -stable if you wanted security fixes? :-) 14:52:51 I do. 14:52:55 after that changed i guess they got more adventerous with merging changes into current 14:53:01 And it wasn't bad, really. 14:53:14 Things were fixed quickly. 14:53:55 * CrtxReavr cut his teeth on 3.0 FWIW. 15:00:47 CrtxReavr: my primary memory from those days is sshd having a new security vulnerability every other week 15:00:57 at one point it seemed like telnetd might be more secure 15:03:24 I rememeber a telnetd RCE as root exploit. 15:03:54 It was fun. . . the PoC code even had a curses "progress bar" - it was all very hollywood. 15:03:58 a few systems had the LD_PRELOAD issue, and there was the Solaris -l-froot issue... did freebsd have one? 15:04:26 It was. . . I think in the early 4.x days. 15:05:03 Back then, so many people ran both telnet & ssh - with telnet as a sort of backup. 15:06:02 lw, that solaris -froot thing is present in un-patched Solaris 10. . . which still runs in a scary number of places. 15:06:31 how old are you guys? 15:06:43 i had expected over 40 i would be the grandfather here xD 15:07:18 * CrtxReavr is just south of dirt. 15:07:29 54 if it matters. 15:08:06 CrtxReavr: anyone still running that version must not have upgraded for ~20 years... which, yeah, that's probably believable 15:08:20 saohh: i'm 20 and i've been 20 for 20 years 15:09:11 If we're counting that, way, I've been twelve for 32 years. 15:09:23 s/\,// 15:09:39 that sounds lovely xD healthy attitude i suppose 15:10:02 you guys use freebsd professionally or say more likely unix(es)? or just home usage? 15:10:47 both, although we're mostly Windows at work (our entire business is built around supporting a specific niche Windows application) most of the support infrastructure runs freebsd, websites etc 15:11:10 interesting as i plan to look into webserver 15:11:17 Nginx or Apache? 15:11:27 freenginx 15:11:46 both, but i'm thinking about switching to Apache exclusively as nginx has basically no advantage for us and has fewer features 15:12:05 +1 to apache 15:12:45 less weird staff, better documented, more predictable 15:12:54 I don't currently use FreeBSD professionally, but I've certainly deployed it for many services at scale and places all around the world. 15:12:56 isn't Apache way too slower than nginx? 15:13:22 isnt it supposed to be better with higher load (Nginx bcs. of its design)? propably anyhow at numbers a normal privat person using it at home will never matter 15:13:23 I didn't notice that 15:13:27 tercaL: it's not "way slower". it's less scalable in some workloads (but much less so with the event mpm), but that only matters if you're serving very high volumes of traffic 15:14:15 like, if i was serving static files or proxy traffic at 10Gbps... i'd probably lean towards nginx... but there's nothing like our workload. and if you're running PHP or something, PHP itself is going to be the bottleneck 15:14:55 in nginx you pass the php over a socket no? 15:15:09 agree, and if used php - use fpm instead of mod_php 15:15:10 i found that concept interesting fpm php or so no? 15:15:10 it's not like in 2002 when Apache required a separate worker process for every incoming connection 15:15:17 saohh: you do it that way in Apache as well, with php-fpm 15:15:39 i mean, you can still use mod_php in Apache if you really need to but that's basically just worse in every way, so don't do that 15:16:30 aha so basicly apache took over this idea from NGinx and now its equally scalable, with the lack of being as quick as NGinx for static content? 15:16:42 Web servers are something I've never run at scale. . . just one-off deployments, for relatively low traffic. 15:16:54 I am kind of an Apache luddite though. 15:16:59 i wouldn't say it came "from nginx", the idea was well known long before nginx existed. it just took Apache a while to adopt it because it's a large and complicated piece of software 15:17:04 and for that you use CrtxReavr? Apache? 15:17:28 Look back two lines. 15:17:47 fwiw, i used to work at an extremely large and well known website based on a PHP app with Apache (and this was back in the Apache 1.3 days, so not event mpm) and Apache was *never* the bottleneck. 15:17:53 oh yeah you wrote that while i wrote the question xD 15:18:19 not to say that there's never any reason to use nginx, but i think people wrongly dismiss Apache based on 20-year-old information without properly considering their requirements. Apache can do a lot of stuff nginx can't 15:18:22 is both equally complex when it comes to running multiple sites? 15:18:31 or is one easier then the other in that regard? 15:18:51 on FreeBSD that is ;) 15:19:16 that's just down to what you prefer, both of them can do virtual hosting and the total lines of configuration isn't that different between them 15:19:23 try both, see what you like 15:19:32 lw, that race-condition thing I brought up yesterday with named & dhcp6c. . . never did get that sorted in a clean way: https://bpa.st/5EPQ 15:19:33 Title: View paste 5EPQ 15:19:54 CrtxReavr: what was the actual problem? 15:20:19 I'm think I'm facing doing something ugly like adding a 'sleep 10' to the beginning of the named startup script. 15:20:28 you want named to depend on DHCP? why? 15:21:02 lw, when named starts, it bitches that it can't create its v6 interfaces. . . . but if I restart named, it's fine. 15:21:32 i've never used named on a dhcp system, but it should recognise new addresses when dhcp adds them and reconfigure itself 15:21:36 relying dhcp6c to pull DHCPv6-PD and configure an external and internal interface. 15:21:47 (maybe this doesn't work as well with ipv6?) 15:22:28 or are you using listen-on{} in named.conf? i guess that would not work like this 15:23:21 CrtxReavr: fwiw this might not work because just because dhcp6c has started, doesn't mean it acquired and assigned the address 15:23:43 this feels like a fragile configuration in general, like what if your isp's dhcp server is slow one time? ... i'd rather fix the root cause i think 15:24:04 listen-on { any; }; 15:24:04 listen-on-v6 { any; }; 15:24:07 Yep. 15:24:41 Hello. Quick question on the default install and config with FreeBSD14. /etc/ssh/sshd_config has the following commented: "#PermitRootLogin no". I am attempting to harden my system, but with the default, i am not able to log in as root. Is there another setting elsewhere that is closing the ability to log in as root? It s 15:24:49 Someone suggested using netwait, but its v6 support seems. . . poor. 15:24:54 ooi, why do you want named to listen on a DHCP-PD prefix? does your ISP guarantee the prefix will never change? 15:25:16 RoyalYork: what's in sshd_config, commented out, is the default settings. so the default setting is "PermitRootLogin no". 15:25:19 RoyalYork, U dont have to 15:25:32 use "su" to elevate to root 15:25:35 lw, no. . . but it will be mostly stable - and that's okay, I can deal with it changing in an automated fashion. 15:25:41 create user in wheel group 15:25:57 su is bush league - use sudo. 15:26:04 su would not work for me too you can not login as root over remote, thats what i had 15:26:10 Okay, so if I leave it uncommented, then users will *not* be able to login as root directly? Good. Because I thought I had to uncomment the line myself for it to take affect 15:26:10 nerozero is right. and then, do: su - 15:26:11 CrtxReavr: all the cool kids use doas now, apparently 15:26:23 yeah i used my IDrac and as Root installed SUDO, added my user done ;) 15:26:36 u can do basically everything with build in su 15:26:38 * CrtxReavr still uses screen instead of tmux, so I've given up on being cool. 15:26:49 RoyalYork: if you leave it *commented* the default is not to permit root login. if you uncomment it, nothing changes as you are explicitly specifying the default. to allow root login, you would have to uncomment it and change it to say "PermitRootLogin yes" (or "PermitRootLogin without-password") 15:27:05 * CrtxReavr litterally has three decades of muscle memory for screen. 15:27:37 Thank you, lw. Is there some master config with the those defaults enabled and uncommented? 15:27:53 is this the case under FreeBSD now too (doas vs sudo)? 15:27:54 o_O 15:28:09 RoyalYork: no, the defaults are built into the sshd binary. but /etc/ssh/sshd_config lists all the default values, so that's the file you're looking for 15:28:16 Thanks 15:28:49 like if you're looking for something similar to /etc/rc.conf vs /etc/defaults/rc.conf, sshd doesn't have that 15:32:01 Ya that was what I thinking of. Thanks for clarifying 15:54:32 Oh, weird. 14.0, just installed on my T480, and the after-install shell doesn't have the network configured. 15:55:06 dhclient to the rescue, but I don't think I've seen it unconfigured before. 15:56:04 You do have to pick that option during the install, unless you intend to set it up yourself. 15:56:27 CrtxReavr: Where during install? I didn't do it last night and the network remained configured. Same install stick. 15:56:30 i was going to say, maybe an obvious question, but did you actually pick the option to configure networking in the installer 15:56:34 RoyalYork: you can also check `sshd -T` for the effective settings (which of course is not 100% safe, but a good start, if you assume the system is un-tampered) 15:56:36 I did. 15:57:33 ridcully_, thank you 15:58:00 shows everything that is set 15:58:01 :D 16:00:33 shows `permitrootlogin no` for me - so maybe _is_ set on your machine? 16:10:14 that's the deault value 16:22:01 Oof. And, the installed T480 doesn't boot. Haven't dug into why yet. Just used the installer, nothing custom. 16:26:43 Hm. It popped in a fallback bootloader properly. 16:29:41 I don't see the NVMe disk in the boot order priority list, oddly. 16:46:32 this isnot freebsd related but with freebsd nginx/1.24.0 package i experince this weird thing that /robots.txt is not matched, request goes to proxy pass. here is my conf: https://paste24.net/p/egXK6a 16:46:33 Title: Troubleshoot IT easily - paste service 16:46:55 can someone verify perhaps? 16:47:31 with stable nginx version and freebsd 13.2-RELEASE-p9 17:47:12 ivaat: There is #nginx 17:53:45 i was there 17:54:30 guy who tested has ver nginx/1.25.2 17:54:45 anyways i tested angie and in there same conf works 17:55:15 what can happen if a device is unplugged/disconnected without being unmounted on the supposition that no read/write is going on? 17:55:15 i tought to let know of maintainer since im not going to open bug in nginx own tracker 17:56:31 if no read and write then issue command sync before unplugging 17:57:02 web forums paint a bleak picture, but i cannot logically imagine that it can do a harm to the file system 17:57:52 patanga: data can be in the kernel buffer cache and not yet written to the disk, so you could lose some amount of data that was written since the last sync 17:58:36 depending on the filesystem, this could cause metadata corruption if the metadata changes hadn't been fully written (this should not happen on modern journaled filesystems, but could on FAT or older UFS) 17:59:01 lw: ivaat: ok, i understand that. my question is: can there be done any harm to the file system, apart from losing data still in the cache? 17:59:19 patanga: yes, if the cache in the cache is filesystem metadata. 17:59:27 s/cache in the/data in the 17:59:40 sync is not writing cache to disk? 17:59:48 lw: oh, metadata. huh, i dont know if i ever changed the metadata of a fs 18:00:09 this is why fsck exists, to give metadata errors caused by unclean dismounting of a filesystem (usually on a crash, but disconnecting a mount drive is semantically the same as a crash in many ways) 18:02:06 patanga: every time you write to a file, create a file, delete a file, you change the fs metadata 18:02:31 the only exception there is on many filesystems, overwriting existing data of a file in-place (i.e., not changing the file size) might not require a metadata update 18:03:04 even reading a file can change the fs metadata if atime is enabled 18:04:47 lw: seems that i have to study file systems and metadata. however what you say seems to confirm that unmounting is important 18:05:28 patanga: it is, yes, in the sense that there's no reason not to do it. in practice, many/most of these errors can be fixed by fsck, but that only guarantees that the filesystem is internally consistent, not that all data could be recovered 18:13:09 lw: thanks for the elaborate explanation 18:21:50 patanga, Additionally one of the kernel thread processes is the syncer thread which looks for modified cached pages (dirty pages) that are candidates for writing to disk. If a page is modified/dirty and IIRC aged more than 30 seconds then the syncer thread will queue those pages to be written to disk. 18:22:17 Meaning that if one writes a bunch of data and then let's the system idle for a while then in the background all of the modified pages will be written to disk. 18:23:02 At some point all queued work will be done and finished. At that point a power failure or a detach of the storage device won't have much outstanding and it is unlikely that any data would be lost or any serious corruption. fsck can fix everything. 18:23:36 But if the storage device is detached before all of these dirty blocks have been written then data loss is almost assured. And depending upon the file system the corruption might be large. 18:25:26 Gracefully unmounting will sync the data explicitly and wait for the pages to be written before detaching. Then it will mark the mount as being cleanly unmounted. If not marked as cleanly unmounted then the mount point will remain marked as dirty and will want an fsck before mounting again. 18:26:28 That description is a little fuzzy being off the top of the head. But the concepts are there. One should always unmount explicitly and allow it to sync queued data and mark it as cleanly unmounted. 18:39:49 Next excitement, for some reason saying nvd0p1:/efi/freebsd/loader.efi failed. gpt/efiboot0:/efi/freebsd/loader.efi worked. 18:40:10 (saying this to efibootmgr(8) that is) 18:42:20 It's a terrible thing but of my systems with UEFI firmware most of mine are buggy in various ways. Only my one laptop booting CoreBoot seems to work correctly for UEFI. The others all have problems. 18:43:28 This one, despite presenting a couple funny tricks on install, appears to be booting successfully now that I've manually popped in a boot variable. 18:44:01 I should write a patch though. This is a pretty common laptop, especially since they're now coming off corporate leases. 18:44:13 I should write up a wiki page noting how to install, also. 18:45:24 My imperfect-by-a-lot understanding is that UEFI requires both the on disk files and the on motherboard non-volatile EFI VARS to be in sync. Which means that one can't just move a disk from one system to another and boot it because the new system won't have a matching EFI VAR to configure booting that disk. 18:45:45 Which is different from Legacy BIOS which would just boot the disk if it was moved into the system. 18:46:06 rwp: Typically you can use the fallback naming and that'll work. 18:46:34 Boot variables sometimes don't work at all - not common lately, but it's been the case - in which case the fallback naming is required. This is probably why FreeBSD opts for it. 18:46:54 I have a handful of HP z210 workstations and exactly one of them seems to work okay. But the rest have this UEFI bug that resets the efi vars after the boot. Meaning that after changing the efibootmgr settings it will boot once and once exactly and the second boot afterward will fail. 18:47:36 The notion is that lacking a boot variable to direct what runs, the fallback file (\EFI\BOOT\BOOTX64.EFI) is tried 18:48:15 Even your z210s ought to work with this. Of course, my ThinkPad T480 appears not to, which would make it the very first system I've seen fail this particular way. 18:49:19 This is mostly a problem if trying to multi-boot. That's basically impossible on this HP firmware. Requiring the fallback. Which means only one system can boot. Requiring a multiboot helper like rEFInd to boot. 18:49:28 Yep. 18:49:40 GRUB ends up being able to do this as well. 18:49:56 rwp: https://wiki.freebsd.org/MasonLoringBliss/BootingFreeBSDfromUEFIGRUB 18:49:57 Title: MasonLoringBliss/BootingFreeBSDfromUEFIGRUB - FreeBSD Wiki 18:54:05 On my HP z210 firmware even trying to use the FreeBSD default installation with it as the single booting OS would not boot. This was 12 and 13. I haven't tried 14 on the problematic machines. 18:54:22 Instead I went ahead and installed them using Legacy BIOS booting. Which of course works and is well tested. 18:54:22 rwp: I wonder if you hit the same thing I just did. 18:54:32 Ah, yeah, that's probably simpler. 18:54:40 But at the time I was trying to exercise UEFI since that is the new path that all machines will have at some point. 18:55:07 I'll make a note to write up install notes. 18:57:35 UEFI only requires this for regular booting storage. But it allows us to boot removable storage without corresponding efi vars. I have this fuzzy memory reading that someone suggested that we could configure internal storage to be removable and then on that system it could boot internal disks as if they were removable without corresponding efi vars. I have been wanting to try that. 18:58:22 Because if that worked then it would make UEFI more robust for booting systems. Then one could migrate disks from say a failed motherboard into another system and then boot the system without needing to boot rescue media and configuring the efi vars. 18:59:22 Life and time is what keeps everything from happening all at once. I haven't had time to work on it more. The systems are running now. Can't disrupt the running systems. I will need a new system to try this on at some point. 19:17:52 rwp: Ah, removable storage, yeah. I believe the UEFI firmware looks for the fallback naming on whatever disk you select, in non-legacy boot mode where you choose a disk. 20:43:44 so i see pkg lists a lot of different types of emacs but i only see one emacs in ports tree 20:43:59 does that mean the different versions are made from having some parameters or config files or something like this? 20:49:40 flavours, but there are only three full, canna and nox 21:04:33 rwp: thanks for your explanations! what i still somehow fail to grasp is "if the storage device is detached before all of these dirty blocks have been written ... the corruption might be large"; how could other data than the one still not written to the device be affected? will its metadata interfere with the metadata of the data written before safely?