From: Alex Lyashkov (shadow_at_psoft.net)
Date: Tue 27 May 2003 - 17:15:21 BST
Hello All
sorry for unanswered mail. I test RSS limit, and after testers answer - i
publish snapshort.
>
> Suggestions for readjustments:
>
> - context creation/destruction should be independant
> of the actual processes (within the context)
>
> what I mean: create a context, set its properties
> (limits, rootpath, capabilities, ...), THEN fork
> some process into this context ...
hm.. no problem. add one function to new_s_context syscall.
it function must call alloc_s_context for created context with default
assigneds. Examle this - read inode from fs.
>
> - vroot (chroot path), capabilites and limits should
> become implicit features of a context
hm.. i think if add check contexts access in namei.c:vfs_permission (as check
in Herbert quotas) it not need add hack "dir 000" for disable escape from
VPS.
>
> - chroot jail (and the permissions)
in work.
> - vserver root path (location)
yes.
> - quota files & support
yes.
> - disk limits (virtual on shared partition)
yes. but if vps have only one partition. if need more one partions in vps have
2 way:
1) map all partitions to one virtual. but we can add per partions limits.
2) add list for maping vps:partion:super_block to
vps:virtual_partion:super_block.
all ways need change little code - functions
s_context_super/s_context_get_super.
> - capability issues (venter vs. native)
hm. if task entered to context - all lmits for this task switched to context
limits. or you speak about other ?
> - process handling (freeze/stop/kill a vs)
on snapshot. CPU QoS beetwen context in process.
> - accounting/statistics across vs reboot
statistics save from kernel and restored on start.
> - networking (interfaces, iptables, ...)
devices, iptables - yes.
network packet scheduler - not.
network QoS beetwen context in process.
> - scheduling
scheduling per context ? as "Fair CPU scheduler" in ASPCompleate ?
I work on this way.
-- With best regards, Alex