fuse: optimize writepages search

fuse: optimize writepages search

Re-work fi->writepages, replacing list with rb-tree.  This improves
performance because kernel fuse iterates through fi->writepages for each
writeback page and typical number of entries is about 800 (for 100MB of
fuse writeback).

Before patch:

10240+0 records in
10240+0 records out
10737418240 bytes (11 GB) copied, 41.3473 s, 260 MB/s

 2  1      0 57445400  40416 6323676    0    0    33 374743 8633 19210  1  8 88  3  0

  29.86%  [kernel]               [k] _raw_spin_lock
  26.62%  [fuse]                 [k] fuse_page_is_writeback

After patch:

10240+0 records in
10240+0 records out
10737418240 bytes (11 GB) copied, 21.4954 s, 500 MB/s

 2  9      0 53676040  31744 10265984    0    0    64 854790 10956 48387  1  6 88  6  0

  23.55%  [kernel]             [k] copy_user_enhanced_fast_string
   9.87%  [kernel]             [k] __memcpy
   3.10%  [kernel]             [k] _raw_spin_lock

Signed-off-by: Maxim Patlasov <mpatlasov@virtuozzo.com>
Signed-off-by: Vasily Averin <vvs@virtuozzo.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>