O_DIRECT support for ext3.

It works OK in all journalling modes.

Updates to the file metadata and inode are journalled as usual.

If the system crashes during an appending O_DIRECT write then journal
recovery will truncate the written-to file back to the length which it
had on entry to that write.

If the system crashes during a file overwrite to existing blocks then
the file contents will be an unknown mixture of old and new.

If the system crashes during a file overwrite which instantiates new
blocks in the middle of the file then there is a possibility of
uninitialised disk blocks being present in the file post-recovery.

mpage_writepages() does a lock_page() on pages to be written back, even
when it is being used for page reclaim writeback.

This is normally OK, because the page is unlocked quickly - pages are
unlocked during writeback and nobody should be performing __GFP_FS
allocations inside lock_page().

But it has introduced a ranking problem in ext3:

generic_file_write
->lock_page
  ->ext3_prepare_write
    ->journal_start	(waits for a commit)

versus

ext3_create()
->journal_start()
  ->ext3_new_inode(GFP_KERNEL)
    ->page reclaim
      ->mpage_writepages
        ->lock_page	(locks up, transaction is held open)

Maybe sometime, I'll have to turn mpage_writepages' lock_page into a
trylock if the caller is PF_MEMALLOC.  But for now, let's make ext3's
inside-transaction allocations use GFP_NOFS.  There is only one of them.

This is a performance and correctness fix against the writeback paths.

The writeback code has competing requirements.  Sometimes it is used
for "memory cleansing": kupdate, bdflush, writer throttling, page
allocator writeback, etc.  And sometimes this same code is used for
data integrity pruposes: fsync, msync, fdatasync, sync, umount, various
other kernel-internal uses.

The problem is: how to handle a dirty buffer or page which is currently
under writeback.

For memory cleansing, we just want to skip that buffer/page and go onto
the next one.  But for sync, we must wait on the old writeback and then
start new writeback.

mpage_writepages() is current correct for cleansing, but incorrect for
sync.  block_write_full_page() is currently correct for sync, but
inefficient for cleansing.

The fix is fairly simple.

- In mpage_writepages(), don't skip the page is it's a sync
operation.

- In block_write_full_page(), skip the buffer if it is a sync
operation.  And return -EAGAIN to tell the caller that the writeout
didn't work out.  The caller must then set the page dirty again and
move it onto mapping->dirty_pages.

This is an extension of the writepage API: writepage can now return
EAGAIN.  There are only three callers, and they have been updated.

fail_writepage() and ext3_writepage() were actually doing this by
hand.  They have been changed to return -EAGAIN.  NTFS will want to
be able to return -EAGAIN from its writepage as well.

- A sticky question is: how to tell the writeout code which mode it
is operating in?  Cleansing or sync?

It's such a tiny code change that I didn't have the heart to go and
propagate a `mode' argument down every instance of writepages() and
writepage() in the kernel.  So I passed it in via current->flags.

Incidentally, the occurrence of a locked-and-dirty buffer in
block_write_full_page() is fairly rare: normally the collision avoidance
happens at the address_space level, via PageWriteback.  But some
mappings (blockdevs, ext3 files, etc) have their dirty buffers written
out via submit_bh().  It is these buffers which can stall
block_write_full_page().

This wart will be pretty intrusive to fix.  ext3 needs to become fully
page-based (ugh.  It's a block-based journalling filesystem, and pages
are unnatural).  blockdev mappings are still written out by buffers
because that's how filesystems use them.  Putting _all_ metadata
(indirects, inodes, superblocks, etc) into standalone address_spaces
would fix that up.

- filemap_fdatawrite() sets PF_SYNC.  So filemap_fdatawrite() is the
kernel function which will start writeback against a mapping for
"data integrity" purposes, whereas the unexported, internal-only
do_writepages() is the writeback function which is used for memory
cleansing.  This difference is the reason why I didn't consolidate
those functions ages ago...

- Lots of code paths had a bogus extra call to filemap_fdatawait(),
which I previously added in a moment of weak-headedness.  They have
all been removed.

A reworked version of the batched page freeing and lock amortisation
for VMA teardown.

It walks the existing 507-page list in the mmu_gather_t in 16-page
chunks, drops their refcounts in 16-page chunks, and de-LRUs and
frees any resulting zero-count pages in up-to-16 page chunks.

Clean up put_page() and page_cache_release().  It's pretty simple now:

#define page_cache_get(page)           get_page(page)
#define page_cache_release(page)       put_page(page)

it was only being used in invalidate_inode_pages(), and from there,
pagevec_release() does the same thing.

As suggested by Daniel - it's a bug to run put_page_testzero
against a zero-ref page.

please apply this patch (Robert ACK-ed it). While there is a preemptible
kernel entry already, i think listing this at the scheduler entry is
justfied, preemption has a number of scheduler interactions.

this is an updated version of the LDT fixes. It fixes the following kinds
of problems:

 - fix a possible gcc optimization causing a race causing the loading of a
   corrupt LDT descriptor upon context switch. [this fix got simplified
   over previous versions.]

 - remove an unconditional OOM printk, and there's no need to set ->size
   in the OOM path.

 - fix preemption bugs, load_LDT()/clear_LDT() was not preemption-safe,
   when it was used outside of spinlocks.

the context-switch race is the following. 'LDT modification' is the
following operation: the seg->ldt pointer is modified, then seg->size is
modified. In theory gcc is free to reschedule the two modifications, and
first modify ->size, then ->ldt. Thus if this modification is not
synchronized with context-switches, another thread might see a temporary
state of the new ->size [which was increased], but still the old pointer.
Ie.:

	CPU0				CPU1

	pc->size = newsize;
					load_LDT(); // (oldptr, newsize)
	pc->ldt = newptr;

the corrupt LDT is loaded until the SMP cross-call is sent, leaving the
window open for many usecs.

the fix is to put a wmb() after ->ldt modifications. [this is also in
preparation of not-write-ordered SMP x86 designs.]

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

some mainboards (Andrew Morton's Dell) report that even everything
is okay with AUX. Also remove a check for very old AMI i8042's, which
could generate false positives on modern buggy mainboards.

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

jfs_get_blocks should return up to the number of blocks in the
extent rather than limiting itself to one block, as the initial,
trivial implementation did.  This greatly reduces the overhead of
O_DIRECT reads and writes.

Submitted by Badari Pulavarty (pbadari@us.ibm.com)

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

Submitted by Steve Best.