[RFC v3 0/1] fs/namespace: defer RCU sync for MNT_DETACH umount

From: Lucas Karpinski
Date: Wed Jun 26 2024 - 16:16:13 EST


Hi all,

Attached is v3 of the umount optimization. Please take a look at v1 for
the original introduction to the problem. Al made it clear in the
RFC v1 that if a filesystem is shut down by umount(2), that the
shut down needs to be completed before the return from the syscall. In
the RFC v2, it was pointed out that call_rcu can block within an
interrupt context. This RFC v3 addresses both that and removes
unneccesary code by taking advantage of mntput for the cleanup of lazy
umounts.

Lucas

v3:
- Removed unneeded code for lazy umount case.
- Don't block within interrupt context.
v2:
- Only defer releasing umount'ed filesystems for lazy umounts
v1:
https://lore.kernel.org/all/20230119205521.497401-1-echanude@xxxxxxxxxx/

Lucas Karpinski (1):
fs/namespace: remove RCU sync for MNT_DETACH umount

fs/namespace.c | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)

--
2.45.2