Re: Review D38047 ... and then there was one....

From: Marek Zarychta <zarychtam_at_plan-b.pwste.edu.pl>
Date: Mon, 07 Oct 2024 04:07:59 UTC
W dniu 7.10.2024 o 00:36, Mark Millard pisze:
> Marek Zarychta <zarychtam_at_plan-b.pwste.edu.pl> wrote on
> Date: Sun, 06 Oct 2024 20:13:58 UTC :
>
>> W dniu 6.10.2024 o 22:04, David Cross pisze:
>>> Here’s the thing. The current implementation of nscd DOESN’T WORK at all. There is a symbol that nscd exports that libc is supposed to use as a flag to bypass lookups for nscd itself. But that symbol isn’t exported right.
>>>
>>> You will need to recompile libc and nscd. (I just do a buildworld to make sure i get everything as there are makefile changes related to the aforementioned symbol changes.
>> Yes, without world installed this patched nscd won't even start:
>>
>> [host] /usr/src# service nscd start
>> Starting nscd.
>> limits: setrlimit pipebuf: Invalid argument
>> /etc/rc.d/nscd: WARNING: failed to start nscd
> . . .
>
> This note is only about the "limits: setrlimit pipebuf:
> Invalid argument" notice.
>
> The main [so: 15] pipebuf related commits were done during
> 2024-Sep-20 UTC. If one has a kernel that predates those but
> a world for which limits now tries to use the new pipebuf
> material, the result is messages like that:
>
> limits: setrlimit pipebuf: Invalid argument
>
> (or related such messages).
>
>
> For reference for main [so: 15]:
>
> Fri, 20 Sep 2024
>      . . .
>      • git: 3458bbd39778 - main - kernel: add RLIMIT_PIPEBUF Konstantin Belousov
>      • git: 54a8d1fbbf65 - main - getrlimit(2): document RLIMIT_PIPEBUF Konstantin Belousov
>      • git: a4c04958f526 - main - libutil: support RLIMIT_PIPEBUF Konstantin Belousov
>      • git: 5d92f20c7d31 - main - bin/sh: support RLIMIT_PIPEBUF Konstantin Belousov
>      • git: f54f41403d14 - main - usr.bin/limits: support RLIMIT_PIPEBUF Konstantin Belousov
>      • git: b029e29e0d8b - main - login.conf: add a placeholder for the pipebuf limit Konstantin Belousov
>      • git: 80133d678ecb - main - procstat: support RLIMIT_PIPEBUF Konstantin Belousov
>      • git: 8ae779832c6f - main - privs: add PRIV_PIPEBUF Konstantin Belousov
>      • git: 7672cbef2c1e - main - pipes: reserve configured percentage of buffers zone to superuser Konstantin Belousov
>      . . .
>      • git: d6074f73af5c - main - pipe: use pipe subsystem KVA counter instead of pipe_map size Konstantin Belousov
>      • git: 40769168a5ee - main - pipespace_new(): decrease uidinfo pipebuf usage if reservation check failed Konstantin Belousov
>      . . .
>      • git: a52b30ff98cd - main - sys_pipe: consistently use cr_ruidinfo for accounting of pipebuf Konstantin Belousov
>      • git: af96ccc6a508 - main - uifree(9): report non-zero values for all shared resources Konstantin Belousov
>      • git: 2c1963d46335 - main - procfs rlimit: handle pipebuf Konstantin Belousov
>      • git: c84d8db0ab3d - main - procfs: ensure that RLIMIT_IDENT is properly updated when a limit is added Konstantin Belousov
>
> The combination of an older kernel and a newer world will not be
> nicely behaved when any non-kernel code from the above ends up
> involved.
>
>
> stable/14 has now also had the commits:
>
> Sat, 05 Oct 2024
>      • git: 1508dce2502d - stable/14 - procfs: ensure that RLIMIT_IDENT is properly updated when a limit is added Konstantin Belousov
>      . . .
>      • git: b7eecc86c3bd - stable/14 - kernel: add RLIMIT_PIPEBUF Konstantin Belousov
>      • git: d20f0dae2f97 - stable/14 - getrlimit(2): document RLIMIT_PIPEBUF Konstantin Belousov
>      • git: a03f7c040ce7 - stable/14 - libutil: support RLIMIT_PIPEBUF Konstantin Belousov
>      • git: d5ed8778bf3b - stable/14 - bin/sh: support RLIMIT_PIPEBUF Konstantin Belousov
>      • git: 25902860b270 - stable/14 - usr.bin/limits: support RLIMIT_PIPEBUF Konstantin Belousov
>      • git: 524b9810de6a - stable/14 - login.conf: add a placeholder for the pipebuf limit Konstantin Belousov
>      • git: 6600090e678e - stable/14 - procstat: support RLIMIT_PIPEBUF Konstantin Belousov
>      • git: fd9babb1b85f - stable/14 - privs: add PRIV_PIPEBUF Konstantin Belousov
>      • git: d532d9926ee7 - stable/14 - pipes: reserve configured percentage of buffers zone to superuser Konstantin Belousov
>      • git: 6536b979b856 - stable/14 - pipe: use pipe subsystem KVA counter instead of pipe_map size Konstantin Belousov
>      • git: a8c663bb4261 - stable/14 - pipespace_new(): decrease uidinfo pipebuf usage if reservation check failed Konstantin Belousov
>      • git: c15b2e046e8c - stable/14 - sys_pipe: consistently use cr_ruidinfo for accounting of pipebuf Konstantin Belousov
>      . . .
>      • git: fc9070bf1d16 - stable/14 - procfs rlimit: handle pipebuf Konstantin Belousov
>      . . .
>
> Again, the combination of an older kernel and a newer world will not be
> nicely behaved.
>
> ===
> Mark Millard
> marklmi at yahoo.com
>
>
>
Thank you for this deep investigation. Anyway, after the kernel and 
world update, I could test the patch, but there was no significant 
progress, at least in lookup timing (as I pointed out before).

-- 
Marek Zarychta