There are some races with IOlock and page unlock due to bad
assumption.
So make things more explicit with a flag to say if we own
the lock, or the 'Writeback' flag.
This allows us to remove the extra flag to iounlock_block
which is really nice because it always confused me.
A significant but subtle part of the locking involved the
fact that one a page has been read, it will not be read
again, so when readpage calls lafs_iocheck_block it is not
possible that it will unlock incorrectly.
It could conceivably unwriteback incorrectly, but next patch
should fix that.
projects / LaFS.git / commit