[MTASCsft PATCH 37/??] MT-, AS- and AC-Safety docs: manual/users.texi

[Joseph S. Myers]

On Fri, 31 Jan 2014, Carlos O'Donell wrote:

> In glibc this is implemented via setXid and has to iterate over all
> threads making them call signal handlers to call setXid themselves
> because the kernel does not provide an atomic way to change all of
> the gid's at the same time for the "process."

Really I'd suggest working with the kernel people to get new syscalls for 
atomically changing ids across an entire process, as suggested in bug 
13347 comment 5, to fix the problems of partial failure as well as not 
being fully safe.  Don't assume that if the kernel people refused in 2004 
(as said in the libc-hacker message you found) they'd refuse now - all the 
security issues associated with set*id failing when programs didn't test 
for it are more recent than then, so the issues are likely to be 
considered more significant now than in 2004.

(More generally, it might be good to have a list of cases where we think 
there are deficiencies in the kernel/userspace interface that cause 
problems for implementing something in glibc, along with proposed fixes 
and whether and when those have been proposed for kernel inclusion.  I 
doubt this is the only one.)

-- 
Joseph S. Myers
joseph@codesourcery.com