fuse: support writable mmap

Quoting Linus (3 years ago, FUSE inclusion discussions):

  "User-space filesystems are hard to get right. I'd claim that they
   are almost impossible, unless you limit them somehow (shared
   writable mappings are the nastiest part - if you don't have those,
   you can reasonably limit your problems by limiting the number of
   dirty pages you accept through normal "write()" calls)."

Instead of attempting the impossible, I've just waited for the dirty page
accounting infrastructure to materialize (thanks to Peter Zijlstra and
others).  This nicely solved the biggest problem: limiting the number of pages
used for write caching.

Some small details remained, however, which this largish patch attempts to
address.  It provides a page writeback implementation for fuse, which is
completely safe against VM related deadlocks.  Performance may not be very
good for certain usage patterns, but generally it should be acceptable.

It has been tested extensively with fsx-linux and bash-shared-mapping.

Fuse page writeback design
--------------------------

fuse_writepage() allocates a new temporary page with GFP_NOFS|__GFP_HIGHMEM.
It copies the contents of the original page, and queues a WRITE request to the
userspace filesystem using this temp page.

The writeback is finished instantly from the MM's point of view: the page is
removed from the radix trees, and the PageDirty and PageWriteback flags are
cleared.

For the duration of the actual write, the NR_WRITEBACK_TEMP counter is
incremented.  The per-bdi writeback count is not decremented until the actual
write completes.

On dirtying the page, fuse waits for a previous write to finish before
proceeding.  This makes sure, there can only be one temporary page used at a
time for one cached page.

This approach is wasteful in both memory and CPU bandwidth, so why is this
complication needed?

The basic problem is that there can be no guarantee about the time in which
the userspace filesystem will complete a write.  It may be buggy or even
malicious, and fail to complete WRITE requests.  We don't want unrelated parts
of the system to grind to a halt in such cases.

Also a filesystem may need additional resources (particularly memory) to
complete a WRITE request.  There's a great danger of a deadlock if that
allocation may wait for the writepage to finish.

Currently there are several cases where the kernel can block on page
writeback:

  - allocation order is larger than PAGE_ALLOC_COSTLY_ORDER
  - page migration
  - throttle_vm_writeout (through NR_WRITEBACK)
  - sync(2)

Of course in some cases (fsync, msync) we explicitly want to allow blocking.
So for these cases new code has to be added to fuse, since the VM is not
tracking writeback pages for us any more.

As an extra safetly measure, the maximum dirty ratio allocated to a single
fuse filesystem is set to 1% by default.  This way one (or several) buggy or
malicious fuse filesystems cannot slow down the rest of the system by hogging
dirty memory.

With appropriate privileges, this limit can be raised through
'/sys/class/bdi/<bdi>/max_ratio'.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>