Linux Kernel Architecture

Chapter 8: The Virtual Filesystem

The actual work is delegated todo_umount.

❑ If a superblock-specificumount_beginfunctionisdefined,itisinvoked.Thisallows,forinstance,

network filesystems to terminate communication with the remote partner before an unmount is

forced.

❑ umount_treeis invoked if the mounted filesystem is no longer needed (this is indicated

by the usage counter)orifMNT_DETACHwas specified to force a filesystem unmount. The

real work is delegated toumount_treeandrelease_mounts. Essentially, the first one is

responsible for decrementing the counterd_mounted, while the latter one uses the data stored

inmnt_mountpointandmnt_parentto restore the original state before the new filesystem

was mounted. The structures of the previously mounted filesystem are also removed from the

kernel lists.

Automatic Expiration

The kernel also provides some infrastructure to allow automatic expiration of mounts. When a mount is

not used by any process or the kernel itself, it will be automatically removed from the VFS mount tree if

automatic expiration is used. Currently the NFS and AFS network filesystems use this offer. Allvfsmount

instances of submounts that are supposed to expire automatically must be collected on a linked list that

usesvfsmount->mnt_expireto chain the elements together.

It then suffices to periodically applymark_mounts_for_expiryon the list. The function scans through

all entries. A mount is unused if its usage count is 1, that is, if it is only referenced by the parent mount.

When such an unused mount is found,mnt_expiry_markis set. Whenmark_mounts_for_expiryfinds

an unused entry withmnt_expiry_markset on the next list traversal, the mount is removed from the

namespace.

Note that themntputis responsible to clearmnt_expiry_mark. This ensures that a mount that has already

been on the expiration list but became used again is not immediately expired when it becomes once more

unused. The code flow is as follows:

1. An unused mount is marked for expiry bymark_mounts_for_expiry.


2. After this, the mount comes into use again, so itsmnt_countis increased. This prevents
   mark_mounts_for_expiryfrom removing the mount from the namespace despite the expi-
   ration mark still being set.


3. When the usage count is decreased withmntput, the function will also ensure that the
   expiration mark is removed. Themark_mounts_for_expirycircle can thus commence as
   usual.

Pseudo-Filesystems

Filesystems do not necessarily require an underlying block medium. They can either use memory as

backing store, as is the case for ramfs and tmpfs, or can require no backing store at all, as procfs and

sysfs do — their contents are generated synthetically from information contained in the kernel’s data

structures. While filesystems of this type are already quite distinct from the traditional concepts, it is still

possible to take a further step forward. How? All filesystems, be they virtual or not, have one common

property: They are visible to userspace in the form of files and directories. However, this property is not

sacrosanct.Pseudo-filesystems are filesystems that cannot be mounted and are thus never directly visible

in userland.