/* -*- auto-fill -*- */
Overview of the Virtual File System
Richard Gooch <rgooch@atnf.csiro.au>
5-JUL-1999
Conventions used in this document <section>
=================================
Each section in this document will have the string "<section>" at the
right-hand side of the section title. Each subsection will have
"<subsection>" at the right-hand side. These strings are meant to make
it easier to search through the document.
NOTE that the master copy of this document is available online at:
http://www.atnf.csiro.au/~rgooch/linux/docs/vfs.txt
What is it? <section>
===========
The Virtual File System (otherwise known as the Virtual Filesystem
Switch) is the software layer in the kernel that provides the
filesystem interface to userspace programs. It also provides an
abstraction within the kernel which allows different filesystem
implementations to co-exist.
A Quick Look At How It Works <section>
============================
In this section I'll briefly describe how things work, before
launching into the details. I'll start with describing what happens
when user programs open and manipulate files, and then look from the
other view which is how a filesystem is supported and subsequently
mounted.
Opening a File <subsection>
--------------
The VFS implements the open(2), stat(2), chmod(2) and similar system
calls. The pathname argument is used by the VFS to search through the
directory entry cache (dentry cache or "dcache"). This provides a very
fast look-up mechanism to translate a pathname (filename) into a
specific dentry.
An individual dentry usually has a pointer to an inode. Inodes are the
things that live on disc drives, and can be regular files (you know:
those things that you write data into), directories, FIFOs and other
beasts. Dentries live in RAM and are never saved to disc: they exist
only for performance. Inodes live on disc and are copied into memory
when required. Later any changes are written back to disc. The inode
that lives in RAM is a VFS inode, and it is this which the dentry
points to. A single inode can be pointed to by multiple dentries
(think about hardlinks).
The dcache is meant to be a view into your entire filespace. Unlike
Linus, most of us losers can't fit enough dentries into RAM to cover
all of our filespace, so the dcache has bits missing. In order to
resolve your pathname into a dentry, the VFS may have to resort to
creating dentries along the way, and then loading the inode. This is
done by looking up the inode.
To look up an inode (usually read from disc) requires that the VFS
calls the lookup() method of the parent directory inode. This method
is installed by the specific filesystem implementation that the inode
lives in. There will be more on this later.
Once the VFS has the required dentry (and hence the inode), we can do
all those boring things like open(2) the file, or stat(2) it to peek
at the inode data. The stat(2) operation is fairly simple: once the
VFS has the dentry, it peeks at the inode data and passes some of it
back to userspace.
Opening a file requires another operation: allocation of a file
structure (this is the kernel-side implementation of file
descriptors). The freshly allocated file structure is initialised with
a pointer to the dentry and a set of file operation member functions.
These are taken from the inode data. The open() file method is then
called so the specific filesystem implementation can do it's work. You
can see that this is another switch performed by the VFS.
The file structure is placed into the file descriptor table for the
process.
Reading, writing and closing files (and other assorted VFS operations)
is done by using the userspace file descriptor to grab the appropriate
file structure, and then calling the required file structure method
function to do whatever is required.
For as long as the file is open, it keeps the dentry "open" (in use),
which in turn means that the VFS inode is still in use.
All VFS system calls (i.e. open(2), stat(2), read(2), write(2),
chmod(2) and so on) are called from a process context. You should
assume that these calls are made without any kernel locks being
held. This means that the processes may be executing the same piece of
filesystem or driver code at the same time, on different
processors. You should ensure that access to shared resources is
protected by appropriate locks.
Registering and Mounting a Filesystem <subsection>
-------------------------------------
If you want to support a new kind of filesystem in the kernel, all you
need to do is call register_filesystem(). You pass a structure
describing the filesystem implementation (struct file_system_type)
which is then added to an internal table of supported filesystems. You
can do:
% cat /proc/filesystems
to see what filesystems are currently available on your system.
When a request is made to mount a block device onto a directory in
your filespace the VFS will call the appropriate method for the
specific filesystem. The dentry for the mount point will then be
updated to point to the root inode for the new filesystem.
It's now time to look at things in more detail.
struct file_system_type <section>
=======================
This describes the filesystem. As of kernel 2.1.99, the following
members are defined:
struct file_system_type {
const char *name;
int fs_flags;
struct super_block *(*read_super) (struct super_block *, void *, int);
struct file_system_type * next;
};
name: the name of the filesystem type, such as "ext2", "iso9660",
"msdos" and so on
fs_flags: various flags (i.e. FS_REQUIRES_DEV, FS_NO_DCACHE, etc.)
read_super: the method to call when a new instance of this
filesystem should be mounted
next: for internal VFS use: you should initialise this to NULL
The read_super() method has the following arguments:
struct super_block *sb: the superblock structure. This is partially
initialised by the VFS and the rest must be initialised by the
read_super() method
void *data: arbitrary mount options, usually comes as an ASCII
string
int silent: whether or not to be silent on error
The read_super() method must determine if the block device specified
in the superblock contains a filesystem of the type the method
supports. On success the method returns the superblock pointer, on
failure it returns NULL.
The most interesting member of the superblock structure that the
read_super() method fills in is the "s_op" field. This is a pointer to
a "struct super_operations" which describes the next level of the
filesystem implementation.
struct super_operations <section>
=======================
This describes how the VFS can manipulate the superblock of your
filesystem. As of kernel 2.1.99, the following members are defined:
struct super_operations {
void (*read_inode) (struct inode *);
int (*write_inode) (struct inode *, int);
void (*put_inode) (struct inode *);
void (*drop_inode) (struct inode *);
void (*delete_inode) (struct inode *);
int (*notify_change) (struct dentry *, struct iattr *);
void (*put_super) (struct super_block *);
void (*write_super) (struct super_block *);
int (*statfs) (struct super_block *, struct statfs *, int);
int (*remount_fs) (struct super_block *, int *, char *);
void (*clear_inode) (struct inode *);
};
All methods are called without any locks being held, unless otherwise
noted. This means that most methods can block safely. All methods are
only called from a process context (i.e. not from an interrupt handler
or bottom half).
read_inode: this method is called to read a specific inode from the
mounted filesystem. The "i_ino" member in the "struct inode"
will be initialised by the VFS to indicate which inode to
read. Other members are filled in by this method
write_inode: this method is called when the VFS needs to write an
inode to disc. The second parameter indicates whether the write
should be synchronous or not, not all filesystems check this flag.
put_inode: called when the VFS inode is removed from the inode
cache. This method is optional
drop_inode: called when the last access to the inode is dropped,
with the inode_lock spinlock held.
This method should be either NULL (normal unix filesystem
semantics) or "generic_delete_inode" (for filesystems that do not
want to cache inodes - causing "delete_inode" to always be
called regardless of the value of i_nlink)
The "generic_delete_inode()" behaviour is equivalent to the
old practice of using "force_delete" in the put_inode() case,
but does not have the races that the "force_delete()" approach
had.
delete_inode: called when the VFS wants to delete an inode
notify_change: called when VFS inode attributes are changed. If this
is NULL the VFS falls back to the write_inode() method. This
is called with the kernel lock held
put_super: called when the VFS wishes to free the superblock
(i.e. unmount). This is called with the superblock lock held
write_super: called when the VFS superblock needs to be written to
disc. This method is optional
statfs: called when the VFS needs to get filesystem statistics. This
is called with the kernel lock held
remount_fs: called when the filesystem is remounted. This is called
with the kernel lock held
clear_inode: called then the VFS clears the inode. Optional
The read_inode() method is responsible for filling in the "i_op"
field. This is a pointer to a "struct inode_operations" which
describes the methods that can be performed on individual inodes.
struct inode_operations <section>
=======================
This describes how the VFS can manipulate an inode in your
filesystem. As of kernel 2.1.99, the following members are defined:
struct inode_operations {
struct file_operations * default_file_ops;
int (*create) (struct inode *,struct dentry *,int);
int (*lookup) (struct inode *,struct dentry *);
int (*link) (struct dentry *,struct inode *,struct dentry *);
int (*unlink) (struct inode *,struct dentry *);
int (*symlink) (struct inode *,struct dentry *,const char *);
int (*mkdir) (struct inode *,struct dentry *,int);
int (*rmdir) (struct inode *,struct dentry *);
int (*mknod) (struct inode *,struct dentry *,int,dev_t);
int (*rename) (struct inode *, struct dentry *,
struct inode *, struct dentry *);
int (*readlink) (struct dentry *, char *,int);
struct dentry * (*follow_link) (struct dentry *, struct dentry *);
int (*readpage) (struct file *, struct page *);
int (*writepage) (struct page *page, struct writeback_control *wbc);
int (*bmap) (struct inode *,int);
void (*truncate) (struct inode *);
int (*permission) (struct inode *, int);
int (*smap) (struct inode *,int);
int (*updatepage) (struct file *, struct page *, const char *,
unsigned long, unsigned int, int);
int (*revalidate) (struct dentry *);
};
Again, all methods are called without any locks being held, unless
otherwise noted.
default_file_ops: this is a pointer to a "struct file_operations"
which describes how to open and then manipulate open files
create: called by the open(2) and creat(2) system calls. Only
required if you want to support regular files. The dentry you
get should not have an inode (i.e. it should be a negative
dentry). Here you will probably call d_instantiate() with the
dentry and the newly created inode
lookup: called when the VFS needs to look up an inode in a parent
directory. The name to look for is found in the dentry. This
method must call d_add() to insert the found inode into the
dentry. The "i_count" field in the inode structure should be
incremented. If the named inode does not exist a NULL inode
should be inserted into the dentry (this is called a negative
dentry). Returning an error code from this routine must only
be done on a real error, otherwise creating inodes with system
calls like create(2), mknod(2), mkdir(2) and so on will fail.
If you wish to overload the dentry methods then you should
initialise the "d_dop" field in the dentry; this is a pointer
to a struct "dentry_operations".
This method is called with the directory inode semaphore held
link: called by the link(2) system call. Only required if you want
to support hard links. You will probably need to call
d_instantiate() just as you would in the create() method
unlink: called by the unlink(2) system call. Only required if you
want to support deleting inodes
symlink: called by the symlink(2) system call. Only required if you
want to support symlinks. You will probably need to call
d_instantiate() just as you would in the create() method
mkdir: called by the mkdir(2) system call. Only required if you want
to support creating subdirectories. You will probably need to
call d_instantiate() just as you would in the create() method
rmdir: called by the rmdir(2) system call. Only required if you want
to support deleting subdirectories
mknod: called by the mknod(2) system call to create a device (char,
block) inode or a named pipe (FIFO) or socket. Only required
if you want to support creating these types of inodes. You
will probably need to call d_instantiate() just as you would
in the create() method
readlink: called by the readlink(2) system call. Only required if
you want to support reading symbolic links
follow_link: called by the VFS to follow a symbolic link to the
inode it points to. Only required if you want to support
symbolic links
struct file_operations <section>
======================
This describes how the VFS can manipulate an open file. As of kernel
2.1.99, the following members are defined:
struct file_operations {
loff_t (*llseek) (struct file *, loff_t, int);
ssize_t (*read) (struct file *, char *, size_t, loff_t *);
ssize_t (*write) (struct file *, const char *, size_t, loff_t *);
int (*readdir) (struct file *, void *, filldir_t);
unsigned int (*poll) (struct file *, struct poll_table_struct *);
int (*ioctl) (struct inode *, struct file *, unsigned int, unsigned long);
int (*mmap) (struct file *, struct vm_area_struct *);
int (*open) (struct inode *, struct file *);
int (*release) (struct inode *, struct file *);
int (*fsync) (struct file *, struct dentry *);
int (*fasync) (struct file *, int);
int (*check_media_change) (kdev_t dev);
int (*revalidate) (kdev_t dev);
int (*lock) (struct file *, int, struct file_lock *);
};
Again, all methods are called without any locks being held, unless
otherwise noted.
llseek: called when the VFS needs to move the file position index
read: called by read(2) and related system calls
write: called by write(2) and related system calls
readdir: called when the VFS needs to read the directory contents
poll: called by the VFS when a process wants to check if there is
activity on this file and (optionally) go to sleep until there
is activity. Called by the select(2) and poll(2) system calls
ioctl: called by the ioctl(2) system call
mmap: called by the mmap(2) system call
open: called by the VFS when an inode should be opened. When the VFS
opens a file, it creates a new "struct file" and initialises
the "f_op" file operations member with the "default_file_ops"
field in the inode structure. It then calls the open method
for the newly allocated file structure. You might think that
the open method really belongs in "struct inode_operations",
and you may be right. I think it's done the way it is because
it makes filesystems simpler to implement. The open() method
is a good place to initialise the "private_data" member in the
file structure if you want to point to a device structure
release: called when the last reference to an open file is closed
fsync: called by the fsync(2) system call
fasync: called by the fcntl(2) system call when asynchronous
(non-blocking) mode is enabled for a file
Note that the file operations are implemented by the specific
filesystem in which the inode resides. When opening a device node
(character or block special) most filesystems will call special
support routines in the VFS which will locate the required device
driver information. These support routines replace the filesystem file
operations with those for the device driver, and then proceed to call
the new open() method for the file. This is how opening a device file
in the filesystem eventually ends up calling the device driver open()
method. Note the devfs (the Device FileSystem) has a more direct path
from device node to device driver (this is an unofficial kernel
patch).
Directory Entry Cache (dcache) <section>
------------------------------
struct dentry_operations
========================
This describes how a filesystem can overload the standard dentry
operations. Dentries and the dcache are the domain of the VFS and the
individual filesystem implementations. Device drivers have no business
here. These methods may be set to NULL, as they are either optional or
the VFS uses a default. As of kernel 2.1.99, the following members are
defined:
struct dentry_operations {
int (*d_revalidate)(struct dentry *);
int (*d_hash) (struct dentry *, struct qstr *);
int (*d_compare) (struct dentry *, struct qstr *, struct qstr *);
void (*d_delete)(struct dentry *);
void (*d_release)(struct dentry *);
void (*d_iput)(struct dentry *, struct inode *);
};
d_revalidate: called when the VFS needs to revalidate a dentry. This
is called whenever a name look-up finds a dentry in the
dcache. Most filesystems leave this as NULL, because all their
dentries in the dcache are valid
d_hash: called when the VFS adds a dentry to the hash table
d_compare: called when a dentry should be compared with another
d_delete: called when the last reference to a dentry is
deleted. This means no-one is using the dentry, however it is
still valid and in the dcache
d_release: called when a dentry is really deallocated
d_iput: called when a dentry loses its inode (just prior to its
being deallocated). The default when this is NULL is that the
VFS calls iput(). If you define this method, you must call
iput() yourself
Each dentry has a pointer to its parent dentry, as well as a hash list
of child dentries. Child dentries are basically like files in a
directory.
Directory Entry Cache APIs
--------------------------
There are a number of functions defined which permit a filesystem to
manipulate dentries:
dget: open a new handle for an existing dentry (this just increments
the usage count)
dput: close a handle for a dentry (decrements the usage count). If
the usage count drops to 0, the "d_delete" method is called
and the dentry is placed on the unused list if the dentry is
still in its parents hash list. Putting the dentry on the
unused list just means that if the system needs some RAM, it
goes through the unused list of dentries and deallocates them.
If the dentry has already been unhashed and the usage count
drops to 0, in this case the dentry is deallocated after the
"d_delete" method is called
d_drop: this unhashes a dentry from its parents hash list. A
subsequent call to dput() will dellocate the dentry if its
usage count drops to 0
d_delete: delete a dentry. If there are no other open references to
the dentry then the dentry is turned into a negative dentry
(the d_iput() method is called). If there are other
references, then d_drop() is called instead
d_add: add a dentry to its parents hash list and then calls
d_instantiate()
d_instantiate: add a dentry to the alias hash list for the inode and
updates the "d_inode" member. The "i_count" member in the
inode structure should be set/incremented. If the inode
pointer is NULL, the dentry is called a "negative
dentry". This function is commonly called when an inode is
created for an existing negative dentry
d_lookup: look up a dentry given its parent and path name component
It looks up the child of that given name from the dcache
hash table. If it is found, the reference count is incremented
and the dentry is returned. The caller must use d_put()
to free the dentry when it finishes using it.
RCU-based dcache locking model
------------------------------
On many workloads, the most common operation on dcache is
to look up a dentry, given a parent dentry and the name
of the child. Typically, for every open(), stat() etc.,
the dentry corresponding to the pathname will be looked
up by walking the tree starting with the first component
of the pathname and using that dentry along with the next
component to look up the next level and so on. Since it
is a frequent operation for workloads like multiuser
environments and webservers, it is important to optimize
this path.
Prior to 2.5.10, dcache_lock was acquired in d_lookup and thus
in every component during path look-up. Since 2.5.10 onwards,
fastwalk algorithm changed this by holding the dcache_lock
at the beginning and walking as many cached path component
dentries as possible. This signficantly decreases the number
of acquisition of dcache_lock. However it also increases the
lock hold time signficantly and affects performance in large
SMP machines. Since 2.5.62 kernel, dcache has been using
a new locking model that uses RCU to make dcache look-up
lock-free.
The current dcache locking model is not very different from the existing
dcache locking model. Prior to 2.5.62 kernel, dcache_lock
protected the hash chain, d_child, d_alias, d_lru lists as well
as d_inode and several other things like mount look-up. RCU-based
changes affect only the way the hash chain is protected. For everything
else the dcache_lock must be taken for both traversing as well as
updating. The hash chain updations too take the dcache_lock.
The significant change is the way d_lookup traverses the hash chain,
it doesn't acquire the dcache_lock for this and rely on RCU to
ensure that the dentry has not been *freed*.
Dcache locking details
----------------------
For many multi-user workloads, open() and stat() on files are
very frequently occurring operations. Both involve walking
of path names to find the dentry corresponding to the
concerned file. In 2.4 kernel, dcache_lock was held
during look-up of each path component. Contention and
cacheline bouncing of this global lock caused significant
scalability problems. With the introduction of RCU
in linux kernel, this was worked around by making
the look-up of path components during path walking lock-free.
Safe lock-free look-up of dcache hash table
===========================================
Dcache is a complex data structure with the hash table entries
also linked together in other lists. In 2.4 kernel, dcache_lock
protected all the lists. We applied RCU only on hash chain
walking. The rest of the lists are still protected by dcache_lock.
Some of the important changes are :
1. The deletion from hash chain is done using hlist_del_rcu() macro which
doesn't initialize next pointer of the deleted dentry and this
allows us to walk safely lock-free while a deletion is happening.
2. Insertion of a dentry into the hash table is done using
hlist_add_head_rcu() which take care of ordering the writes -
the writes to the dentry must be visible before the dentry
is inserted. This works in conjuction with hlist_for_each_rcu()
while walking the hash chain. The only requirement is that
all initialization to the dentry must be done before hlist_add_head_rcu()
since we don't have dcache_lock protection while traversing
the hash chain. This isn't different from the existing code.
3. The dentry looked up without holding dcache_lock by cannot be
returned for walking if it is unhashed. It then may have a NULL
d_inode or other bogosity since RCU doesn't protect the other
fields in the dentry. We therefore use a flag DCACHE_UNHASHED to
indicate unhashed dentries and use this in conjunction with a
per-dentry lock (d_lock). Once looked up without the dcache_lock,
we acquire the per-dentry lock (d_lock) and check if the
dentry is unhashed. If so, the look-up is failed. If not, the
reference count of the dentry is increased and the dentry is returned.
4. Once a dentry is looked up, it must be ensured during the path
walk for that component it doesn't go away. In pre-2.5.10 code,
this was done holding a reference to the dentry. dcache_rcu does
the same. In some sense, dcache_rcu path walking looks like
the pre-2.5.10 version.
5. All dentry hash chain updations must take the dcache_lock as well as
the per-dentry lock in that order. dput() does this to ensure
that a dentry that has just been looked up in another CPU
doesn't get deleted before dget() can be done on it.
6. There are several ways to do reference counting of RCU protected
objects. One such example is in ipv4 route cache where
deferred freeing (using call_rcu()) is done as soon as
the reference count goes to zero. This cannot be done in
the case of dentries because tearing down of dentries
require blocking (dentry_iput()) which isn't supported from
RCU callbacks. Instead, tearing down of dentries happen
synchronously in dput(), but actual freeing happens later
when RCU grace period is over. This allows safe lock-free
walking of the hash chains, but a matched dentry may have
been partially torn down. The checking of DCACHE_UNHASHED
flag with d_lock held detects such dentries and prevents
them from being returned from look-up.
Maintaining POSIX rename semantics
==================================
Since look-up of dentries is lock-free, it can race against
a concurrent rename operation. For example, during rename
of file A to B, look-up of either A or B must succeed.
So, if look-up of B happens after A has been removed from the
hash chain but not added to the new hash chain, it may fail.
Also, a comparison while the name is being written concurrently
by a rename may result in false positive matches violating
rename semantics. Issues related to race with rename are
handled as described below :
1. Look-up can be done in two ways - d_lookup() which is safe
from simultaneous renames and __d_lookup() which is not.
If __d_lookup() fails, it must be followed up by a d_lookup()
to correctly determine whether a dentry is in the hash table
or not. d_lookup() protects look-ups using a sequence
lock (rename_lock).
2. The name associated with a dentry (d_name) may be changed if
a rename is allowed to happen simultaneously. To avoid memcmp()
in __d_lookup() go out of bounds due to a rename and false
positive comparison, the name comparison is done while holding the
per-dentry lock. This prevents concurrent renames during this
operation.
3. Hash table walking during look-up may move to a different bucket as
the current dentry is moved to a different bucket due to rename.
But we use hlists in dcache hash table and they are null-terminated.
So, even if a dentry moves to a different bucket, hash chain
walk will terminate. [with a list_head list, it may not since
termination is when the list_head in the original bucket is reached].
Since we redo the d_parent check and compare name while holding
d_lock, lock-free look-up will not race against d_move().
4. There can be a theoritical race when a dentry keeps coming back
to original bucket due to double moves. Due to this look-up may
consider that it has never moved and can end up in a infinite loop.
But this is not any worse that theoritical livelocks we already
have in the kernel.
Important guidelines for filesystem developers related to dcache_rcu
====================================================================
1. Existing dcache interfaces (pre-2.5.62) exported to filesystem
don't change. Only dcache internal implementation changes. However
filesystems *must not* delete from the dentry hash chains directly
using the list macros like allowed earlier. They must use dcache
APIs like d_drop() or __d_drop() depending on the situation.
2. d_flags is now protected by a per-dentry lock (d_lock). All
access to d_flags must be protected by it.
3. For a hashed dentry, checking of d_count needs to be protected
by d_lock.
Papers and other documentation on dcache locking
================================================
1. Scaling dcache with RCU (http://linuxjournal.com/article.php?sid=7124).
2. http://lse.sourceforge.net/locking/dcache/dcache.html
