However, this _does_ bring up another issue that I have been
wondering about: would it be better to make the inode semaphore not
a semaphore but a read-write lock?
I say yes, and if I recall correctly Thomas Schoebel-Theur was
recommending such a thing in the early revisions of the vfs rewrites.
We don't have any blocking read-write locks, but they shouldn't be
hard to implement similarly to the rw-spinlocks.
Subsystems of Alexey's new routing code would benefit tremendously
from such a beast as well.
Later,
David "Sparc" Miller
davem@caip.rutgers.edu