2.4.15pre1aa1

Andrea Arcangeli (andrea@suse.de)
Fri, 9 Nov 2001 19:59:00 +0100


URL:

ftp://ftp.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.15pre1aa1.bz2
ftp://ftp.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.15pre1aa1/

Only in 2.4.14pre7aa2: 00_alpha-cc-1
Only in 2.4.14pre7aa2: 00_alpha-cc-1-freq-dev-1
Only in 2.4.14pre7aa2: 00_increase-logbuffer-2

In mainline.

Only in 2.4.14pre7aa2: 00_netconsole-2.4.10-C2-1.bz2
Only in 2.4.15pre1aa1: 00_netconsole-2.4.10-C2-2.bz2

Rediffed.

Only in 2.4.14pre7aa2: 10_vm-12
Only in 2.4.15pre1aa1: 10_vm-13

Latest vm updates. Most important if we take a swapin on
an exclusive swap cache that is getting swapped out (so
locked) we don't need to lock_page or to do_wp_page, we
can takeover the swapcache despite it's locked, if it's
exclusive. This is possible because we can learn if it's exlcusive
without the need of taking the page lock thanks to latest Linus's saner
locking recent changes. So this update still delivers non blocking minor
swapin faults, _but_ without wasteful cows.

Also merged anon pages into the lru, now that the race I spotted in
release_page_cache is fixed in 2.4.15pre1 (fixed by Linus, thanks!).

Only in 2.4.14pre7aa2: 50_uml-patch-2.4.13-3.bz2
Only in 2.4.15pre1aa1: 50_uml-patch-2.4.13-5.bz2

Latest Jeff's update.

Only in 2.4.14pre7aa2: 60_atomic-lookup-4
Only in 2.4.15pre1aa1: 60_atomic-lookup-5

Rediffed.

Only in 2.4.14pre7aa2: 60_tux-2.4.13-ac5-A5.bz2
Only in 2.4.15pre1aa1: 60_tux-2.4.13-ac5-B0.bz2

Latest Ingo's update (didn't checked if this fixes the trivial
kernel-wise hostname DoS yet).

Tux is obviously potentially less safe than any userspace webserver [in
this case it DoS the while machine, it's not enough to restart the
webserver like you could do if the same overflow bug was in an
userspace webserver], and of course security wise there's no chroot
that can help if the kernel overflows.

The advantage of Tux is that it can reach higher performance by saving
the cost of syscalls, tlb flushing, 4M global pagetables, and by using
internal kernel functionalities without being slowed down by user
protection API and by being able to use _all_ kernel internal
functionalities even if an user API is missing (like for cpus_allowed
and zerocopy parsing in the dma buffer etc...). But it's by no means
"safer" than its userspace counterparts, period. Or more precisely: if
it is safer it's not because it's in kernel but it's because Ingo wrote
it :).

Andrea
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/