Several hash table implementations in the networking were
remotely exploitable.  Remote attackers could launch attacks
whereby, using carefully choosen forged source addresses, make
every routing cache entry get hashed into the same hash chain.

Netfilter's IP conntrack module and the TCP syn-queue implementation
had identical vulnerabilities and have been fixed too.

The choosen solution to the problem involved using Bob's Jenkins
hash along with a randomly choosen input.  For the ipv4 routing
cache we take things one step further and periodically choose a
new random secret.  By default this happens every 10 minutes, but
this is configurable by the user via sysctl knobs.

into nuts.ninka.net:/home/davem/src/BK/net-2.5

With this one and 100 dummy interfaces everything now is working
flawlessly, case closed :-)

This cures the problem reported in lkml when there are lots
of net devices, tested with 50 dummy interfaces, there is
still one glitch when there are more info produced than
one page, where one entry is missed, I'm working on this
one now.

Ah, dev_seq_show needs only seq_puts, not the more expensive
seq_printf.

Thanks to Randy Dunlap for spotting this one, now to study report
about /proc/net/dev breakage with many interfaces.

into nuts.ninka.net:/home/davem/src/BK/net-2.5

into kernel.bkbits.net:/home/davem/net-2.5

Try to avoid calling "pfn_to_page()" on invalid pfn's.  It used to be
legal, but the CONFIG_DISCONTIGMEM people want us to try to avoid it,
since they do magic stuff in their "pfn_to_page" translations.

Just make x86-64/amd64 compile again.  Only architecture specific
changes.

And a workaround for the Opteron prefetch bug.

Also remove the obsolete LVM1 ioctl emulation code.

Brown paperbag time. I forgot to take the modrm byte in account
with the prefetch patch replacement.  With 3.2 it worked because
it used the right registers in my configuration.

But gcc 2.96 uses a different register in __dpath and the prefetch becomes
4 bytes with modrm and the original nop needs to be as long as that too.

Leftovers from the mem_map_reserve-removal patch.

into home.transmeta.com:/home/torvalds/v2.5/linux

Changes:

  - Convert to a static highlevel handle for all drivers. Gets rid of
    a kmalloc for each driver and consolidates the highlevel handle and
    highlevel ops. Reduces points of failure as well.
  - Move host number allocation to hosts.c, giving all drivers access to
    it.
  - Implemented S800 changes for core and sbp2. Thanks to TI and LaCie
    for hardware to test this. 1394b support coming soon.
  - Convert nodemgr to using a similar device classification hack as
    usb. The old class_num hack I had in place was broken do to its
    removal. This is a much cleaner solution.

into kroah.com:/home/linux/linux/BK/gregkh-2.5

Missed parts of the ioctl32 compatibility patch: a few ioctls from
PA-RISC, and missing <linux/compat_ioctl.h> file.

Here are two vicam.c patches to clean up and restore copyright notices
for 2.4.21-rc1 and 2.5.68-bk9. At some point in time Pavel and my
copyrights were remove inadvertenly from the code.

drivers/input/misc/uinput.c::uinput_alloc_device(): reduce stack size
from 0x480 to 0x24;

reduce stack usage in sw_connect() from 0x490 to 0x98 on P4 SMP (gcc 3.2);

 - kill dups in ide.c and ide-disk.c
 - read_24() is exported so rename it to ide_read_24()
 - add it ide.h

They are unused, not needed and identical to generic ones so kill 'em:
    task_read_24(),
    task_try_to_flush_leftover_data(),
    taskfile_dump_status(),
    taskfile_error().

There's an (fortunately unused) devfs in misc_register currently,
when the name of the miscdevice contains a slash the name is used
as devfs name instead of misc/<name>.  Kill if as we have .devfs_name
for this kind of stuff now.

ioctl32 cleanups are pretty neccessary (we have 6+ copies of 600+
lines tables, all getting slightly out of sync, not speaking about
surrounding code produced by cut-and-paste).

There's three drivers in the tree that workaround the suboptimal
devfs name choice of the misc device layer (/dev/misc/<foo>) using
devfs_mk_symlink.    Switch them to set miscdev.devfs_name instead
to get the right name from the very beginning.

into home.transmeta.com:/home/torvalds/v2.5/linux

into samba.org:/home/paulus/kernel/for-linus-ppc

into home.transmeta.com:/home/torvalds/v2.5/linux

- Remove all uses of unnamed structs and unions in the driver to make
  old and newer gcc versions happy. Makes it a bit uglier IMO but at
  least people will stop hassling me

into cantab.net:/home/aia21/ntfs-2.5

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

From: William Lee Irwin III <wli@holomorphy.com>, Robert Love

Add some spinlock protection around the oom-killer state.