tmpfs: use a seperate lock to protect all_nodes

Previously, tmpfs used diskfs_node_refcnt_lock to serialize access to
the all_nodes and some other related global state related to memory
consumption.

Use a separate lock to protect all_nodes, and atomic operations to
access the state related to memory consumption. Adjust the reference
counting accordingly.  Every node in the all_nodes carries a light
reference.  When we are asked to give up that light reference, we
reacquire our lock momentarily to check whether someone else
reacquired a reference through the all_nodes.

* tmpfs/tmpfs.h (num_files, tmpfs_space_used): Use atomic operations
for these variables.
(adjust_used): Use atomic operations.
(get_used): New convenience function to atomically retrieve
tmpfs_space_used.
* tmpfs/node.c (all_nodes_lock): New lock.
(diskfs_alloc_node): Use a separate lock to protect all_nodes.
Adjust the reference counting accordingly.
(diskfs_free_node): Likewise.
(diskfs_cached_lookup):Likewise.
(diskfs_node_iterate): Likewise.
(diskfs_node_norefs): Do not remove the node from all_nodes.  This
actually looks like a mistake, I do not know why they did that here as
well as in diskfs_free_node.
(diskfs_try_dropping_softrefs): Check whether someone reacquired a
reference, and if so hold on to our light reference.
(diskfs_grow): Use atomic operations.
* tmpfs/tmpfs.c (diskfs_set_statfs): Likewise.

0 files changed, 0 insertions, 0 deletions