[Posted September 16, 2026 by corbet]
The Register looks forward to the upcoming Fedora 45 release.
The biggest surprise is that Linux's legacy in-kernel console – the text-mode interface normally hidden beneath the GUI – has been replaced with a software-controlled alternative. The replacement is kmscon, a userspace terminal emulator that has been in development for more than a decade.
to post comments
target user ?
Posted Sep 16, 2026 19:40 UTC (Wed) by ballombe (subscriber, #9523) [Link] (19 responses)
target user ?
Posted Sep 16, 2026 19:43 UTC (Wed) by cyperpunks (subscriber, #39406) [Link] (4 responses)
target user ?
Posted Sep 16, 2026 21:27 UTC (Wed) by barryascott (subscriber, #80640) [Link]
target user ?
Posted Sep 17, 2026 4:11 UTC (Thu) by wtarreau (subscriber, #51152) [Link] (2 responses)
target user ?
Posted Sep 17, 2026 5:17 UTC (Thu) by pbonzini (subscriber, #60935) [Link]
Unicode support, better keyboard configuration, speed, and "no one wants to maintain fbcon/fbdev in the kernel" are the main things it brings.
target user ?
Posted Sep 21, 2026 6:40 UTC (Mon) by kaesaecracker (subscriber, #126447) [Link]
Target user here. I've been using the user space terminal on my desktops for a while because it renders my nerd font glyphs and can be customized.
target user ?
Posted Sep 16, 2026 19:53 UTC (Wed) by zhalas (subscriber, #114013) [Link] (11 responses)
target user ?
Posted Sep 16, 2026 20:42 UTC (Wed) by ballombe (subscriber, #9523) [Link] (9 responses)
There is no way to know that a particular dynamically allocated /dev/pts is linked to some linux console and need to be handled specially.
target user ?
Posted Sep 17, 2026 1:09 UTC (Thu) by zhalas (subscriber, #114013) [Link] (8 responses)
target user ?
Posted Sep 17, 2026 7:14 UTC (Thu) by taladar (subscriber, #68407) [Link] (7 responses)
Distinguishing local authentication from remote authentication is one thing that comes to mind.
target user ?
Posted Sep 17, 2026 7:45 UTC (Thu) by josh (subscriber, #17465) [Link] (6 responses)
PAM knows what's trying to authenticate; if SSH is authenticating then it's remote, and if login or gdm is authenticating then it's local. I don't think there's value in detecting whether something is a "local TTY".
target user ?
Posted Sep 17, 2026 21:40 UTC (Thu) by turistu (guest, #164830) [Link] (5 responses)
Not all Linux systems have to use PAM, yet. Fortunately.
target user ?
Posted Sep 17, 2026 23:24 UTC (Thu) by AdamW (subscriber, #48457) [Link]
Fedora does, though.
target user ?
Posted Sep 18, 2026 10:16 UTC (Fri) by josh (subscriber, #17465) [Link] (3 responses)
As someone who has no particular reason not to, what exactly is the problem with PAM?
pam
Posted Sep 20, 2026 6:37 UTC (Sun) by donald.buczek (subscriber, #112892) [Link] (1 responses)
As someone who has no particular reason not to, what exactly is the problem with PAM?
Nobody answered yet, so I send a few points from the top of my head:
- Plugins: I co-maintain multi-user systems that can't be rebooted, on a self-maintained distribution, so in-process .so plugins hurt. PAM's pam_unix also pulls in NSS, which has the same design.
- Upgrades are fragile with plugins: After a rolling upgrade, long-running daemons (sshd, httpd, cron) hold old-ABI code in memory while newly loaded modules use the new internal API. Crashes (segfaults) are among the milder outcomes.
- You sometimes have unresolvable diamond dependencies with glibc->PAM->nss modules. Dependency hell.
- Documentation: The API documentation is bad. Basic concepts like "session" are not defined but vaguely assumed. "key=value" strings without clear semantics are passed back and forth between components.
- Security: Security-relevant code doesn't run in a separate process. Unprivileged usage requires a setuid helper (unix_chkpwd).
pam
Posted Sep 20, 2026 9:39 UTC (Sun) by bluca (subscriber, #118303) [Link]
target user ?
Posted Sep 21, 2026 15:01 UTC (Mon) by turistu (guest, #164830) [Link]
First of all, it does not do what you think it does:
PAM knows what's trying to authenticate; if SSH is authenticating then it's remote, and if login or gdm is authenticating then it's local.
Any program can pretend to pam that it's login or ssh or lightdm or whatever. It's just an argument to pam_start().
So pam does not actually know anything.
NB: on Debian/Fedora/etc, the program does not need any privileges for this -- any user can authenticate as themselves via pam; the default pam modules are able to achieve that by calling a "helper" setuid program behind the scenes.
target user ?
Posted Sep 17, 2026 21:12 UTC (Thu) by turistu (guest, #164830) [Link]
Currently, there's no way to know which /dev/pts/n is actually the controlling terminal of a process in Linux, or which one is paired to a master pty file descriptor.
The kernel can only tell you its major:minor number, which can exist as different devices in different devpts mounts.
It's quite easy to make appear for silly sysadmins and stupid tools that some process is running on e.g. /dev/pts/0, and completely hide from view which process is actually controlling its master pty peer.
And there are other pretty nasty bugs related to devpts. First of all being its existence in the first place -- how about replacing it with something better before getting rid of virtual terminals?
target user ?
Posted Sep 16, 2026 20:29 UTC (Wed) by chder (subscriber, #96621) [Link] (1 responses)
target user ?
Posted Sep 16, 2026 20:52 UTC (Wed) by nim-nim (subscriber, #34454) [Link]
Braille display and text to speech accessibility support
Posted Sep 16, 2026 22:50 UTC (Wed) by jasonjgw (subscriber, #52080) [Link]
There was discussion on the BRLTTY mailing list a while ago of developing a protocol to enable braille access to user-space consoles (without any graphical desktop running, obviously). The question now is whether a solution will be implemented before any plan to deprecate and, perhaps, ultimately remove the kernel's console support goes ahead. Likewise for users who rely on text to speech to access the console. The Linux console has been non-visually accessible via text to speech and braille devices for decades, so any removal of this support would disadvantage users and administrators who depend on this functionality.
kmscon is great
Posted Sep 17, 2026 1:53 UTC (Thu) by dankamongmen (subscriber, #35141) [Link] (1 responses)
i looked at it when i was doing notcurses, and it was a fantastic little piece of software. it was unmaintained at the time, though. i'm guessing fedora has put some love behind it? really good to see this out there.
kmscon is great
Posted Sep 17, 2026 2:54 UTC (Thu) by rahulsundaram (subscriber, #21946) [Link]
Yes, kmscon was unmaintained, looking at the repo (linked in the blurb), it looks like activity has picked up and the developer who proposed this change (linked earlier in one of the comments) is doing the work to iron out the rough edges as part of the change to make it the default. I would expect other distros to follow given the advantages here.
Adios serial console?
Posted Sep 17, 2026 14:44 UTC (Thu) by NightMonkey (subscriber, #23051) [Link] (3 responses)
Perhaps this is obvious, but is support for serial console access then going away in the kernel? Cheers.
Adios serial console?
Posted Sep 17, 2026 15:06 UTC (Thu) by hmh (subscriber, #3838) [Link]
Adios serial console?
Posted Sep 17, 2026 16:56 UTC (Thu) by intelfx (subscriber, #130118) [Link]
No, serial consoles (tty) and virtual terminals/virtual consoles (vt/vc/vtcon) layers are two separate things. This is strictly about the latter; nobody is proposing to get rid of ttys (which are, at the very least, still required for any kind of terminal emulation, including kmscon the vtcon replacement). :-)
Adios serial console?
Posted Sep 18, 2026 9:13 UTC (Fri) by LtWorf (subscriber, #124958) [Link]
How do you debug your virtual machines then?
VT switching/locking support
Posted Sep 21, 2026 4:45 UTC (Mon) by cypherpunks2 (guest, #152408) [Link]
Does this support standard VT switching and locking? Commands like VT_LOCKSWITCH that are used by screen locking utilities like physlock?
Instagram Engineering
Shopify Engineering