Re: [RFC][PATCH] inotify 0.13.1

[Robert Love]

OK.

I am working on redoing the inotify locking wrt inodes.

Most of the time that we grab inode->i_lock, we don't need it.  We do
need it when allocating or freeing inode->inotify_data.  We also need it
when first obtaining the inotify_data reference count (see below).

So we need an inotify_data reference count to track its lifetime.  We
already have ->watcher_count, which functions pretty well as this ref
count, but we need to formalize it, make it atomic_t, and always grab it
under ->i_lock.  Freeing the structure is then a consequence of the ref
count dropping to zero via atomic_dec_and_test().

Then we can add inotify_data->lock to protect the contents of
inotify_data from concurrent access and not use i_lock.

Anyone dereferencing inotify_data (or setting/freeing it for the first
time) needs to hold a ref count.  This is an existing bug.

We can hopefully end up only taking inode->i_lock when grabbing this ref
count, and no other time.  We won't have to nest it with dev->lock any
more.

That is the plan, anyhow.  Patch later tonight or tomorrow, with luck.

	Robert Love


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/