| Message ID | 20241213-work-mount-rbtree-lockless-v3-0-6e3cdaf9b280@kernel.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | fs: lockless mntns lookup | expand |
On Fri, 2024-12-13 at 00:03 +0100, Christian Brauner wrote: > Hey, > > This now also includes selftests for iterating mount namespaces both > backwards and forwards. > > Currently we take the read lock when looking for a mount namespace to > list mounts in. We can make this lockless. The simple search case can > just use a sequence counter to detect concurrent changes to the rbtree. > > For walking the list of mount namespaces sequentially via nsfs we keep a > separate rcu list as rb_prev() and rb_next() aren't usable safely with > rcu. > > Since creating mount namespaces is a relatively rare event compared with > querying mounts in a foreign mount namespace this is worth it. Once > libmount and systemd pick up this mechanism to list mounts in foreign > mount namespaces this will be used very frequently. > > Thanks! > Christian > > --- > Changes in v3: > - Add selftests. > - Put list_head into a union with the wait_queue_head_t for poll instead > of the mnt_ns_tree_node which would've risked breaking rbtree > traversal. > - Handle insertion into the mount namespace list correctly by making use > of the rbtree position information after the mount namespace has been > added to it. > - Improve the documentation for the new list_bidir_{del,prev}_rcu(). > - Link to v2: https://lore.kernel.org/r/20241212-work-mount-rbtree-lockless-v2-0-4fe6cef02534@kernel.org > > Changes in v2: > - Remove mnt_ns_find_it_at() by switching to rb_find_rcu(). > - Add separate list to lookup sequential mount namespaces. > - Link to v1: https://lore.kernel.org/r/20241210-work-mount-rbtree-lockless-v1-0-338366b9bbe4@kernel.org > > --- > Christian Brauner (10): > mount: remove inlude/nospec.h include > fs: add mount namespace to rbtree late > fs: lockless mntns rbtree lookup > rculist: add list_bidir_{del,prev}_rcu() > fs: lockless mntns lookup for nsfs > fs: simplify rwlock to spinlock > seltests: move nsfs into filesystems subfolder > selftests: add tests for mntns iteration > selftests: remove unneeded include > samples: add test-list-all-mounts > > fs/mount.h | 18 +- > fs/namespace.c | 163 ++++++++------ > fs/nsfs.c | 5 +- > include/linux/rculist.h | 47 +++++ > samples/vfs/.gitignore | 1 + > samples/vfs/Makefile | 2 +- > samples/vfs/test-list-all-mounts.c | 235 +++++++++++++++++++++ > .../selftests/{ => filesystems}/nsfs/.gitignore | 1 + > .../selftests/{ => filesystems}/nsfs/Makefile | 4 +- > .../selftests/{ => filesystems}/nsfs/config | 0 > .../selftests/filesystems/nsfs/iterate_mntns.c | 149 +++++++++++++ > .../selftests/{ => filesystems}/nsfs/owner.c | 0 > .../selftests/{ => filesystems}/nsfs/pidns.c | 0 > tools/testing/selftests/pidfd/pidfd.h | 1 - > 14 files changed, 546 insertions(+), 80 deletions(-) > --- > base-commit: 40384c840ea1944d7c5a392e8975ed088ecf0b37 > change-id: 20241207-work-mount-rbtree-lockless-7d4071b74f18 > Reviewed-by: Jeff Layton <jlayton@kernel.org>
Hey, This now also includes selftests for iterating mount namespaces both backwards and forwards. Currently we take the read lock when looking for a mount namespace to list mounts in. We can make this lockless. The simple search case can just use a sequence counter to detect concurrent changes to the rbtree. For walking the list of mount namespaces sequentially via nsfs we keep a separate rcu list as rb_prev() and rb_next() aren't usable safely with rcu. Since creating mount namespaces is a relatively rare event compared with querying mounts in a foreign mount namespace this is worth it. Once libmount and systemd pick up this mechanism to list mounts in foreign mount namespaces this will be used very frequently. Thanks! Christian --- Changes in v3: - Add selftests. - Put list_head into a union with the wait_queue_head_t for poll instead of the mnt_ns_tree_node which would've risked breaking rbtree traversal. - Handle insertion into the mount namespace list correctly by making use of the rbtree position information after the mount namespace has been added to it. - Improve the documentation for the new list_bidir_{del,prev}_rcu(). - Link to v2: https://lore.kernel.org/r/20241212-work-mount-rbtree-lockless-v2-0-4fe6cef02534@kernel.org Changes in v2: - Remove mnt_ns_find_it_at() by switching to rb_find_rcu(). - Add separate list to lookup sequential mount namespaces. - Link to v1: https://lore.kernel.org/r/20241210-work-mount-rbtree-lockless-v1-0-338366b9bbe4@kernel.org --- Christian Brauner (10): mount: remove inlude/nospec.h include fs: add mount namespace to rbtree late fs: lockless mntns rbtree lookup rculist: add list_bidir_{del,prev}_rcu() fs: lockless mntns lookup for nsfs fs: simplify rwlock to spinlock seltests: move nsfs into filesystems subfolder selftests: add tests for mntns iteration selftests: remove unneeded include samples: add test-list-all-mounts fs/mount.h | 18 +- fs/namespace.c | 163 ++++++++------ fs/nsfs.c | 5 +- include/linux/rculist.h | 47 +++++ samples/vfs/.gitignore | 1 + samples/vfs/Makefile | 2 +- samples/vfs/test-list-all-mounts.c | 235 +++++++++++++++++++++ .../selftests/{ => filesystems}/nsfs/.gitignore | 1 + .../selftests/{ => filesystems}/nsfs/Makefile | 4 +- .../selftests/{ => filesystems}/nsfs/config | 0 .../selftests/filesystems/nsfs/iterate_mntns.c | 149 +++++++++++++ .../selftests/{ => filesystems}/nsfs/owner.c | 0 .../selftests/{ => filesystems}/nsfs/pidns.c | 0 tools/testing/selftests/pidfd/pidfd.h | 1 - 14 files changed, 546 insertions(+), 80 deletions(-) --- base-commit: 40384c840ea1944d7c5a392e8975ed088ecf0b37 change-id: 20241207-work-mount-rbtree-lockless-7d4071b74f18