cachepc-linux

dax.rst (11483B)

=======================
Direct Access for files
=======================

Motivation
----------

The page cache is usually used to buffer reads and writes to files.
It is also used to provide the pages which are mapped into userspace
by a call to mmap.

For block devices that are memory-like, the page cache pages would be
unnecessary copies of the original storage.  The `DAX` code removes the
extra copy by performing reads and writes directly to the storage device.
For file mappings, the storage device is mapped directly into userspace.


Usage
-----

If you have a block device which supports `DAX`, you can make a filesystem
on it as usual.  The `DAX` code currently only supports files with a block
size equal to your kernel's `PAGE_SIZE`, so you may need to specify a block
size when creating the filesystem.

Currently 5 filesystems support `DAX`: ext2, ext4, xfs, virtiofs and erofs.
Enabling `DAX` on them is different.

Enabling DAX on ext2 and erofs
------------------------------

When mounting the filesystem, use the ``-o dax`` option on the command line or
add 'dax' to the options in ``/etc/fstab``.  This works to enable `DAX` on all files
within the filesystem.  It is equivalent to the ``-o dax=always`` behavior below.


Enabling DAX on xfs and ext4
----------------------------

Summary
-------

 1. There exists an in-kernel file access mode flag `S_DAX` that corresponds to
    the statx flag `STATX_ATTR_DAX`.  See the manpage for statx(2) for details
    about this access mode.

 2. There exists a persistent flag `FS_XFLAG_DAX` that can be applied to regular
    files and directories. This advisory flag can be set or cleared at any
    time, but doing so does not immediately affect the `S_DAX` state.

 3. If the persistent `FS_XFLAG_DAX` flag is set on a directory, this flag will
    be inherited by all regular files and subdirectories that are subsequently
    created in this directory. Files and subdirectories that exist at the time
    this flag is set or cleared on the parent directory are not modified by
    this modification of the parent directory.

 4. There exist dax mount options which can override `FS_XFLAG_DAX` in the
    setting of the `S_DAX` flag.  Given underlying storage which supports `DAX` the
    following hold:

    ``-o dax=inode``  means "follow `FS_XFLAG_DAX`" and is the default.

    ``-o dax=never``  means "never set `S_DAX`, ignore `FS_XFLAG_DAX`."

    ``-o dax=always`` means "always set `S_DAX` ignore `FS_XFLAG_DAX`."

    ``-o dax``      is a legacy option which is an alias for ``dax=always``.

    .. warning::

      The option ``-o dax`` may be removed in the future so ``-o dax=always`` is
      the preferred method for specifying this behavior.

    .. note::

      Modifications to and the inheritance behavior of `FS_XFLAG_DAX` remain
      the same even when the filesystem is mounted with a dax option.  However,
      in-core inode state (`S_DAX`) will be overridden until the filesystem is
      remounted with dax=inode and the inode is evicted from kernel memory.

 5. The `S_DAX` policy can be changed via:

    a) Setting the parent directory `FS_XFLAG_DAX` as needed before files are
       created

    b) Setting the appropriate dax="foo" mount option

    c) Changing the `FS_XFLAG_DAX` flag on existing regular files and
       directories.  This has runtime constraints and limitations that are
       described in 6) below.

 6. When changing the `S_DAX` policy via toggling the persistent `FS_XFLAG_DAX`
    flag, the change to existing regular files won't take effect until the
    files are closed by all processes.


Details
-------

There are 2 per-file dax flags.  One is a persistent inode setting (`FS_XFLAG_DAX`)
and the other is a volatile flag indicating the active state of the feature
(`S_DAX`).

`FS_XFLAG_DAX` is preserved within the filesystem.  This persistent config
setting can be set, cleared and/or queried using the `FS_IOC_FS`[`GS`]`ETXATTR` ioctl
(see ioctl_xfs_fsgetxattr(2)) or an utility such as 'xfs_io'.

New files and directories automatically inherit `FS_XFLAG_DAX` from
their parent directory **when created**.  Therefore, setting `FS_XFLAG_DAX` at
directory creation time can be used to set a default behavior for an entire
sub-tree.

To clarify inheritance, here are 3 examples:

Example A:

.. code-block:: shell

  mkdir -p a/b/c
  xfs_io -c 'chattr +x' a
  mkdir a/b/c/d
  mkdir a/e

  ------[outcome]------

  dax: a,e
  no dax: b,c,d

Example B:

.. code-block:: shell

  mkdir a
  xfs_io -c 'chattr +x' a
  mkdir -p a/b/c/d

  ------[outcome]------

  dax: a,b,c,d
  no dax:

Example C:

.. code-block:: shell

  mkdir -p a/b/c
  xfs_io -c 'chattr +x' c
  mkdir a/b/c/d

  ------[outcome]------

  dax: c,d
  no dax: a,b

The current enabled state (`S_DAX`) is set when a file inode is instantiated in
memory by the kernel.  It is set based on the underlying media support, the
value of `FS_XFLAG_DAX` and the filesystem's dax mount option.

statx can be used to query `S_DAX`.

.. note::

  That only regular files will ever have `S_DAX` set and therefore statx
  will never indicate that `S_DAX` is set on directories.

Setting the `FS_XFLAG_DAX` flag (specifically or through inheritance) occurs even
if the underlying media does not support dax and/or the filesystem is
overridden with a mount option.


Enabling DAX on virtiofs
----------------------------
The semantic of DAX on virtiofs is basically equal to that on ext4 and xfs,
except that when '-o dax=inode' is specified, virtiofs client derives the hint
whether DAX shall be enabled or not from virtiofs server through FUSE protocol,
rather than the persistent `FS_XFLAG_DAX` flag. That is, whether DAX shall be
enabled or not is completely determined by virtiofs server, while virtiofs
server itself may deploy various algorithm making this decision, e.g. depending
on the persistent `FS_XFLAG_DAX` flag on the host.

It is still supported to set or clear persistent `FS_XFLAG_DAX` flag inside
guest, but it is not guaranteed that DAX will be enabled or disabled for
corresponding file then. Users inside guest still need to call statx(2) and
check the statx flag `STATX_ATTR_DAX` to see if DAX is enabled for this file.


Implementation Tips for Block Driver Writers
--------------------------------------------

To support `DAX` in your block driver, implement the 'direct_access'
block device operation.  It is used to translate the sector number
(expressed in units of 512-byte sectors) to a page frame number (pfn)
that identifies the physical page for the memory.  It also returns a
kernel virtual address that can be used to access the memory.

The direct_access method takes a 'size' parameter that indicates the
number of bytes being requested.  The function should return the number
of bytes that can be contiguously accessed at that offset.  It may also
return a negative errno if an error occurs.

In order to support this method, the storage must be byte-accessible by
the CPU at all times.  If your device uses paging techniques to expose
a large amount of memory through a smaller window, then you cannot
implement direct_access.  Equally, if your device can occasionally
stall the CPU for an extended period, you should also not attempt to
implement direct_access.

These block devices may be used for inspiration:
- brd: RAM backed block device driver
- dcssblk: s390 dcss block device driver
- pmem: NVDIMM persistent memory driver


Implementation Tips for Filesystem Writers
------------------------------------------

Filesystem support consists of:

* Adding support to mark inodes as being `DAX` by setting the `S_DAX` flag in
  i_flags
* Implementing ->read_iter and ->write_iter operations which use
  :c:func:`dax_iomap_rw()` when inode has `S_DAX` flag set
* Implementing an mmap file operation for `DAX` files which sets the
  `VM_MIXEDMAP` and `VM_HUGEPAGE` flags on the `VMA`, and setting the vm_ops to
  include handlers for fault, pmd_fault, page_mkwrite, pfn_mkwrite. These
  handlers should probably call :c:func:`dax_iomap_fault()` passing the
  appropriate fault size and iomap operations.
* Calling :c:func:`iomap_zero_range()` passing appropriate iomap operations
  instead of :c:func:`block_truncate_page()` for `DAX` files
* Ensuring that there is sufficient locking between reads, writes,
  truncates and page faults

The iomap handlers for allocating blocks must make sure that allocated blocks
are zeroed out and converted to written extents before being returned to avoid
exposure of uninitialized data through mmap.

These filesystems may be used for inspiration:

.. seealso::

  ext2: see Documentation/filesystems/ext2.rst

.. seealso::

  xfs:  see Documentation/admin-guide/xfs.rst

.. seealso::

  ext4: see Documentation/filesystems/ext4/


Handling Media Errors
---------------------

The libnvdimm subsystem stores a record of known media error locations for
each pmem block device (in gendisk->badblocks). If we fault at such location,
or one with a latent error not yet discovered, the application can expect
to receive a `SIGBUS`. Libnvdimm also allows clearing of these errors by simply
writing the affected sectors (through the pmem driver, and if the underlying
NVDIMM supports the clear_poison DSM defined by ACPI).

Since `DAX` IO normally doesn't go through the ``driver/bio`` path, applications or
sysadmins have an option to restore the lost data from a prior ``backup/inbuilt``
redundancy in the following ways:

1. Delete the affected file, and restore from a backup (sysadmin route):
   This will free the filesystem blocks that were being used by the file,
   and the next time they're allocated, they will be zeroed first, which
   happens through the driver, and will clear bad sectors.

2. Truncate or hole-punch the part of the file that has a bad-block (at least
   an entire aligned sector has to be hole-punched, but not necessarily an
   entire filesystem block).

These are the two basic paths that allow `DAX` filesystems to continue operating
in the presence of media errors. More robust error recovery mechanisms can be
built on top of this in the future, for example, involving redundancy/mirroring
provided at the block layer through DM, or additionally, at the filesystem
level. These would have to rely on the above two tenets, that error clearing
can happen either by sending an IO through the driver, or zeroing (also through
the driver).


Shortcomings
------------

Even if the kernel or its modules are stored on a filesystem that supports
`DAX` on a block device that supports `DAX`, they will still be copied into RAM.

The DAX code does not work correctly on architectures which have virtually
mapped caches such as ARM, MIPS and SPARC.

Calling :c:func:`get_user_pages()` on a range of user memory that has been
mmaped from a `DAX` file will fail when there are no 'struct page' to describe
those pages.  This problem has been addressed in some device drivers
by adding optional struct page support for pages under the control of
the driver (see `CONFIG_NVDIMM_PFN` in ``drivers/nvdimm`` for an example of
how to do this). In the non struct page cases `O_DIRECT` reads/writes to
those memory ranges from a non-`DAX` file will fail 


.. note::

  `O_DIRECT` reads/writes _of a `DAX` file do work, it is the memory that
  is being accessed that is key here).  Other things that will not work in
  the non struct page case include RDMA, :c:func:`sendfile()` and
  :c:func:`splice()`.