The nat core calls skb_ip_make_writable() with a length too short.
Found by nfsim.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

UDP and TCP refer to potentially stale pointers after calling
skb_ip_make_writable(), and UDP calls it with the wrong len argument.
All found by nfsim.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Here is a patch that just makes it use get_free_pages to test the TLB
theory.  Another obvious improvement would be to not use list_heads for
the hash table buckets - a single pointer would likely suffice and it
would cut the hash table in half, saving cache, TLB and memory.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

The multiport match doesn't support ranges of ports, so a new match
called "mport" was written.  Now we have versioning of matches and
targets, we can simply put this extension in multiport revision 1.

Also, removes gratuitous checking in match: we basically trust
iptables userspace these days.
Signed-off-by: Pablo Neira Ayuso <pablo@eurodev.net>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> (modified)
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Anders Fugmann <afu@fugmann.dhs.org> wrote a patch to add bitops to
ipt_MARK.  I made a version based on the revision patch.

Bart De Schuymer <bdschuym@pandora.be> provided the idea of overriding
the target type.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This adds a new getsockopt to iptables, which allows userspace to
query the revision number of extensions.  iptables 1.3.0 (to be
released soon) already has support for this.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Pablo Neira <pablo@eurodev.net> came up with the idea of stealing the
last byte in the (already nul-terminated) name of a target/match to
create a version.

That patch motivated me to clean up the match and target searching in
ip_tables, which makes this patch more trivial.  I also renamed
"version" to "revision" since the word "version" was used for iptables
itself.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

We now have a convenient kmod macro try_then_request_module(). 

1) Split our current "generic" function into separate
   find_target_lock(), find_table_lock() and find_match_lock()
   functions explicitly.

2) Have those functions do try_module_get() for us, and fix up the one
   caller who didn't do that anyway.

3) Have the caller use try_then_request_module().

4) Remove __ipt_mutex_up() and __ipt_find_target_lock() which weren't
   used (even in patch-o-matic AFAICT).

This cleanup takes us closer to using standard list macros everywhere,
and makes the version patch simpler.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

We currently choose a "random" IP address to NAT to, where we have a
range.  Martin Josefsson pointed out that he uses the SAME target in
iptables because changing IP addresses breaks Internet banking sites
(among others) which assume the customer will be coming from a
consistent IP address.

In fact, we spend a fair bit of effort trying to balance the number of
connections we NAT to each IP address.  We can come pretty damn close
just hashing the source and destination IP addresses, and it has the
consistency property which is so desirable, as well as being faster.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Now we no longer do extra mangling, warn in that case.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

On NF_IP_LOCAL_OUT, when destination NAT changes the destination
interface, we also change the source address, so the packet is the
same as if it were generated to go that way in the first place.  This
is not strictly necessary, I believe.

This patch rips that code out to see what breaks.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

On NF_IP_LOCAL_OUT or NF_IP_PRE_ROUTING, if destination NAT is not
sufficient to create a unique tuple, we try changing the source port
as well.  However, this is also not strictly necessary: if the tuple
is not unique, we will also try to change the source on the
NF_IP_POST_ROUTING hook.

When we finally confirm the connection, if the tuple is still not
unique the packet will be dropped (this is required anyway as we could
race: the conntrack is not placed in the hash until the packet is
about to leave the box anyway).

In fact, we only need best effort *everywhere*.

This patch rips that code out to see what breaks.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

ip_conntrack_alter_reply checks that the reply isn't already taken,
but there's little point, since there's *still* a race after it is
called (which we handle at confirm time anyway).
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

The NAT code has the concept of multiple ranges: you can say "map this
connection onto IP 192.168.1.2 - 192.168.1.4, 192.168.1.7 ports
1024-65535, and 192.168.1.10".  I implemented this because we could.

But it's not actually *used* by many (any?) people, and you can
approximate this by a random match (from patch-o-matic) if you really
want to.  It adds complexity to the code.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Writing the nfsim testcase for the ECN target revealed a hole in the
rule checking: when checking whether the rule specified TCP, you need
to check it isn't inverted.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

nfsim now generates non-linear packets: when run under valgrind it
finds linear assumptions very nicely.  This is the second thing I
tried, and it found a real bug.

In this case, checking the UDP checksum is correct, but overkill, and
like the ICMP header handling, does not handle non-linear packets.  Remove UDP
checksum, fix ICMP error.

Signed-off-by: Rusty Russell
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

OK, somehow I managed to botch this one.  It happens to work fine, but I
should have been more careful with forward porting this 1+ year old patch.
The exec-time calc should go in bprm_apply_creds, not bprm_free_security.

Thanks to Stephen for spotting my mistake.
Signed-off-by: Chris Wright <chrisw@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

into ppc970.osdl.org:/home/torvalds/v2.6/linux

into fys.uio.no:/home/linux/bitkeeper/nfsclient-2.6

	Don't label all retries as "server seeing garbage".
	Report correct error for buffer overflows.
	Fix incorrect buffer overflow test that was masking
		AUTH_REJECTEDCRED, AUTH_REJECTEDVERF,... errors.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

 Description:
 Fix an NFS client bug introduced in 2.6.9-rc1.  The "df" command was
 reporting the size of NFS file systems incorrectly.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

     cache.
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 The behavior enabled by the "noac" mount option should be precisely
 equivalent to setting acreg{min,max} or acdir{min,max} to zero via mount
 options.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 Other parts of the NFS client invoke nfs_wb_all() when they want to flush dirty
 cache pages.  The direct path needs to do that, too.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 The initial implementation of NFS direct reads was entirely synchronous.
 The direct read logic issued one NFS READ operation at a time, and waited
 for the server's reply before issuing the next one.  For large direct
 read requests, this is unnecessarily slow.

 This patch changes the NFS direct read path to dispatch NFS READ operations
 for a single direct read request in parallel and wait for them once.  The
 direct read path is still synchronous in nature, but because the NFS READ
 operations are going in parallel, the completion wait should be much shorter.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 Reduce stack utilization in the NFS direct read path by using a
 dynamically allocated nfs_read_data structure instead of allocating one
 on the stack.  This reduces stack utilization of nfs_direct_read_seg
 from over 900 bytes to less than 100 bytes.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 Reduce stack utilization in the NFS direct write path by using a
 dynamically allocated nfs_write_data structure instead of allocating one
 on the stack.  This reduces stack utilization of nfs_direct_write_seg
 from over 900 bytes to less than 100 bytes.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 Immediately return control to the application if a short NFS write is
 detected in the NFS client's direct write path.  This is better behavior
 than what the direct write path does today, which could result in data
 appearing at the wrong offset in the file.

 Eventually this code path should retry short writes at least once before
 giving up.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 Replace a C macro with sizeof().
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 Description:
 Ethereal and other tools display RPC XIDs in host order.  This patch
 changes the RPC trace messages that display XIDs to print them in host
 order so they can be easily matched to XIDs that appear in Ethereal.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 Improve trace debugging messages for NFSv2/3 GETATTR and SETATTR
 procedures.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 Recently a patch set was accepted to allow the Linux NFS client to handle
 short writes by retrying the unwritten portion of the request.  The only
 case that now results in an error is when the server makes no progress;
 that is, writes zero bytes.

 This patch changes the kernel log warning that is generated in that case
 to reflect the error condition more accurately.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 sillyrenames when renaming to an existing file.
Signed-off-by: Greg Banks <gnb@melbourne.sgi.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

      know there are no pending timers.
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

      inodes may be marked as stale in one instance (causing the dentry
      to be dropped) then re-enabled in the next instance.
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

Although it usually means that someone has deleted a file on the server,
the ESTALE error may also indicate that the sysadmin has used exportfs to
deny our client access to the server. Most NFS implementations therefore
consider it a non-permanent condition, and allow inodes to "recover" when
the sysadmin re-enables access.
If, however, you want to work with broken servers, like unfsd, that reuse
filehandles for new files after the original file gets deleted, then
"recovery" is impossible, since it may be that the filehandle now points
to a different file. Note that this is broken server behaviour that may
happen even without us ever seeing the ESTALE error.
In order to minimize (but we can never eliminate entirely) this race
condition on unfsd servers, Linux has traditionally made ESTALE a
permanent condition on all filehandles except the root filehandle.

The problem is that if we apply this strict staleness criterion to
directories (particularly so for he current directory), then all
processes will need to re-walk the path starting from the mount point,
in order to recover from the sysadmin intervention case. As this is not
usual on other *NIX implementations, and may in any case be undermined by
caching rules etc, this is being seen as a usability problem.

This patch makes ESTALE a non-permanent condition on directories, but
preserves the current behaviour for non-directories.
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

 Replace the kmalloc() and kfree() calls in this path with appropriate
 invocations of nfs_writedata_alloc() and nfs_writedata_free().  This
 makes nfs_writepage_sync match all the other write paths in fs/nfs/write.c.

 Category: Maintainability, performance
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>

as a void * when we know perfectly well what it is.  And we're passing
some arguments that we don't actually use.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>