GitLab

->atomic_open() is another monster (the other is ->rename() of course).
It operates look-up, create, and open in a single method.
Strictly speaking the behaviour is not atomic from the branch fs's point
of view, while the atomicity is kept in aufs's. This is a second-best
approach. For details, refer to the design document in previous commit.

A simple list 'si_aopen' can be put into aufs inode object, but I don't
think it a good idea because the number of inodes is much larger than
the number of super_block. If I put it into inode, it can be an
un-interesting memory pressure I am afraid.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

This new function au_aopen_or_create() tries calling branch fs's
->atomic_open() first. If it is not set, call vfs_create() instead.
By putting this behaviour into aufs's add_simple(), many codes can be
shared.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Extend do_open_dir(), au_do_open_nondir() and au_do_open() to receive an
additional parameter h_file, which is an opened file object by branch
fs's ->atomic_open(). By this design/commit, aufs doesn't have to
duplicate many codes into a new aufs_atomic_open() (in later commit),
and can simply share them.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Following the design in another commit, aufs calls branch fs's
->atomic_open() if exits. Ideally it would be better to call
VFS:do_last, lookup_open() or atomic_open, but it is very hard for
aufs. This implementation is far from the best.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement several f_op functions for non-dir.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

For details, read the document in this commit.
I don't like this approach, but there is no other way currently. But it
seems that UnionMount is trying add siblings of f_dentry and d_inode for
linux-4.0 or later. It may become another light for aufs too.

The finfo object which has ever mmapped is excluded from
refreshing (based upon fi_mmapped). Otherwise we may corrupt the process
memory space.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

By default, aufs doesn't copy-up the file in open(2).
The file write operation is one of the trigger of the copy-up.
Although I understand that O_RDWR or O_WRONLY should trigger the
copy-up, it is not a good idea for the case of open(O_RDWR) +
mmap(MAP_PRIVATE). In this case, all changes are not written-back to the
file on disk, and the copy-up is meaningless entirely.
In other words, aufs postpone the copy-up as possible.

This design also applies to the file operation after branch management.
Some of the opened file need to be refreshed after add/del/mod
branches. Eg. detect the revealed same named one, open it, close the old
one internally while the virtual file is kept opened.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement f_op->open() for non-directory.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Support for XATTR and ACL including several branch attributes to ignore
the copy error around XATTR and ACL.

NFS always sets MS_POSIXACL regardless its mount option 'noacl.'
When MS_POSIXACL is set, generic_permission() calls check_acl() (via
acl_permission_check()) and gets -EOPNOTSUPP because the NFS branch is
mounted as 'noacl.'
In aufs, h_permission() should not call generic_permission() in this
case.
The similar thing happens in coping-up XATTR. vfs_getxattr_alloc()
returns -EOPNOTSUPP.

See also the document in this commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

An enhancement for udba=none if possible.
The condition is same to the 'no ->d_revalidate()' patch series.
Refresh i_op in all cached inodes at the remount-time.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

An enhancement for udba=none if possible.
The condition is same to the 'no ->d_revalidate()' patch series.
In this mode, i_op->getattr() is less important, and the default VFS
handler is enough.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement i_op->get/setattr().
setattr() is another trigger of the copy-up. The file may or may not be
opened. And some of sub-functions are commonly used to XATTR operations
in later commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement i_op->link().
As aufs supports 'pseudo-link', aufs_link() can make it without
copying-up. In the case of aufs_link() has to copy-up, the name of the
target file is used as-is, and it is pseudo-linked. In other words,
calling link(2) after the copy-up is unnecessary.

See also struct.txt in previous commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement i_op->rename().
This is a big monster and I don't like it.

In this version, the copy-up always happen.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement i_op->rename().
This is a big monster and I don't like it.

In order to call d_move() in aufs lock section, FS_RENAME_DOES_D_MOVE is
set to fstype.f_flags.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement i_op->rmdir() with supporting logical deletion by whiteout
including all children.

As struct.txt in previous commit described, the target dir is renamed to
a whiteout-ed temporary unique name in rmdir(2), and then removed
asynchronously by the system global workqueue.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement i_op->unlink() with supporting logical deletion by whiteout.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Here are entry adding inode operations, i_op->create(), symlink(),
mkdir(), mknod(), and tmpfile().
Obviously they return EOPNOTSUPP when the target branch fs doesn't
support the operation.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement dir.i_op->get_link().
I am afraid there may exist the case skipping updating the inode atime.
It means the aufs inode atime may be reverted by the real fs inode atime
silently.  But I don't think it a big problem.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement dir.i_op->lookup() and ->permission().
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Optimize out ->d_revalidate() if possible.
If the refreshing failed, then ->d_revalidate() remains. In this case,
aufs has two types of dentries. One has ->d_revalidate, the other
doesn't. I am afraid it will confuse me someday.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement d_op->d_revalidate().
This is another core part of UDBA (cf. lookup.txt in another commit).
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement basic sb_op->show_options(), statfs() and sync_fs() simply.
- show_options() doesn't print the default values (AuOpt_Def).
- statfs() will have an option to summarize the numbers from branches
  (in later commit).
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Call the functions in previous commits to refresh all.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Maintain the internal array including corresponding XINO file and sysfs
entries.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

As a part of branch-management, aufs maintains all cached inodes,
dentries, and opened files in remounting.
This commits handles the opened files by counting the number of them,
generating an array of their pointers. I don't like such array
approach, but I don't have another idea.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

As a part of branch-management, aufs maintains all cached inodes,
dentries, and opened files in remounting.
This commits handles the cached dentries by calling the VFS internal
function d_walk().
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

As a part of branch-management, aufs maintains all cached inodes,
dentries, and opened files in remounting.
This commits handles the cached inodes by counting the number of cached
inodes, generating an array of their pointers. I don't like such array
approach, but I don't have another idea.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

It is hard to implement readdir(3) for aufs virtual directory.
It considers the every whiteout in a single direcotry, as well as the
(first) opaque marker (diropq).
This implementation consumes memory a lot, and I'd suggest you to try
RDU (readdir in userspace) in later commit.

See also the document in previous commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

This commit is just to prepare for the succeeding commit, and split to
suppress the size of a single commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Expand finfo to support for a directory.
For readdir(3), see VDIR and RDU in later commits.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

The structure is very similar to iinfo and dinfo (in previous commits).
This commit is for non-dir files. For a directory, see later commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

As a part of looking-up, construct a virtual inode.
After branch-management (add/del branches), the inode has to be
refreshed to represent a revealed file.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Now aufs becomes mountable with very few features.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

In order to prevent firing the notify event from aufs itself, hnotify
feature is suspend/resume-able. They are combined with mutex lock/unlock
for the parent dir.

See also previous commits.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

The feature is constructed by two layers. One is generic interface, and
the other is exact implementation. This is rather historical. Originally
aufs implemented this feature based upon 'inotify.' Later 'fsnotify'
made 'inotify' obsolete. During the transition period, these two layers
were introduced to support both of 'inotify' and 'fsnotify.' Currently
only 'fsnotify' is supported, but the layers are kept for the future
use.

This feature is compiled only when CONFIG_AUFS_HNOTIFY is enabled.
See also the document in previous commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

This is the hardest test to support UDBA (users' direct branch access).
It uses 'fsnotify' internally.  Detecting UDBA, decrements the
generation of the cached aufs objects.  In the next access to the file,
aufs detects the generation is obsoleted and tries refreshing it.
Eventually aufs cache will be updated to latest status.

The fsnotify is set on the cached dirs on the non-RR branches.
The RR (real readonly) branches will never be modified and it is
unnecessary to set fsnotify for them.

This commit is for the declarations mainly, and the body parts will be
in succeeding commits.

This feature is compiled only when CONFIG_AUFS_HNOTIFY is enabled.
See also the document in this commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

The main part is in previous commit.
This commit handles the generation of aufs objects, to make sure the
inode in the file handle is still valid.
In order not to confuse NFSD, the various operation returns ESTALE for
NFSD where it used to return EBUSY.

See also the document in this commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Implement exporting via NFS.
The file handle is rather large (40 bytes at most + the file handle on a
branch).
The non-virtual filesystems can use an anonymous (disconnected) dentry
as long as the inode is identified, but aufs needs a dentry with dinfo
which is usually constructed.  So aufs has to find or generate the
normal dentry from the file handle in decoding.  Eg. in aufs, there
should never be the anonymous dentry.

In decoding the file handle, if both of the dentry and the inode which
are corresponding the file handle are still in cache, then they are
returned immediately.  Otherwise aufs has to find the cached parent dir
from the file handle.  If the parent dir is not cached either, the aufs
tries these steps.
- decode the branch fs's file handle and get the parent dir
- generate the path of the parent dir on the branch
- convert the branch path to aufs's path
- lookup the inode number under the aufs' path
The last one is the slowest case.

exportfs_decode_fh() (actually reconnect_path()) acquires mutex, and
this behaviour violates the locking order between aufs si_rwsem.  This
is not a problem since internal exportfs_decode_fh() is called for the
branch fs.
Simply use lockdep_off/on to silence the lockdep message.

See also the document in later commit.
This is compiled only when CONFIG_AUFS_EXPORT is enabled.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

As a result of branch management, the virtual inode may point a
different real inode from it used to. And aufs has to maintain its
address_space_operations, since its definition may affect the
behaviour.
I know some people (including grsec-patch) doesn't like a non-const
address_space_operations, but in order to keep the consistency of the
behaviour, the correct address_space_operations is important.

See also the document in this commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>