All the higher level database operations need to be locked for the whole
duration of the operation, so nobody should be using the lower-level
tasks or raw read/write functions. We can also remove the locks from
them.
The database can now safely be used by multiple servers.
Must be locked for the whole operation due to the find free/add
read-modify-write.
Preserves the previous behavior of sending back an ID of 0 in case of
allocation failure, for better or worse.
There is (currently) only one storage area that is used by all servers,
so it needs to be managed by its own class shared among them.
The occupied mask is also moved as it reflects the storage contents and
so can't be stored by each server.
Put documentation with each bitmask and use the object directly. Node ID
range checks can be removed as the bitmask itself checks and we don't
expect to trip them.
Substantially cleans up the code.
The StorageManager read_block/write_block methods only return failure if
an out of bounds access is performed. Assert statically that this does
not happen.
Also remove the now-impossible failed to add node state.
Thomas Watson
bugobliterator
Andrew Tridgell