1=======================
2Direct Access for files
3=======================
4
5Motivation
6----------
7
8The page cache is usually used to buffer reads and writes to files.
9It is also used to provide the pages which are mapped into userspace
10by a call to mmap.
11
12For block devices that are memory-like, the page cache pages would be
13unnecessary copies of the original storage.  The `DAX` code removes the
14extra copy by performing reads and writes directly to the storage device.
15For file mappings, the storage device is mapped directly into userspace.
16
17
18Usage
19-----
20
21If you have a block device which supports `DAX`, you can make a filesystem
22on it as usual.  The `DAX` code currently only supports files with a block
23size equal to your kernel's `PAGE_SIZE`, so you may need to specify a block
24size when creating the filesystem.
25
26Currently 5 filesystems support `DAX`: ext2, ext4, xfs, virtiofs and erofs.
27Enabling `DAX` on them is different.
28
29Enabling DAX on ext2 and erofs
30------------------------------
31
32When mounting the filesystem, use the ``-o dax`` option on the command line or
33add 'dax' to the options in ``/etc/fstab``.  This works to enable `DAX` on all files
34within the filesystem.  It is equivalent to the ``-o dax=always`` behavior below.
35
36
37Enabling DAX on xfs and ext4
38----------------------------
39
40Summary
41-------
42
43 1. There exists an in-kernel file access mode flag `S_DAX` that corresponds to
44    the statx flag `STATX_ATTR_DAX`.  See the manpage for statx(2) for details
45    about this access mode.
46
47 2. There exists a persistent flag `FS_XFLAG_DAX` that can be applied to regular
48    files and directories. This advisory flag can be set or cleared at any
49    time, but doing so does not immediately affect the `S_DAX` state.
50
51 3. If the persistent `FS_XFLAG_DAX` flag is set on a directory, this flag will
52    be inherited by all regular files and subdirectories that are subsequently
53    created in this directory. Files and subdirectories that exist at the time
54    this flag is set or cleared on the parent directory are not modified by
55    this modification of the parent directory.
56
57 4. There exist dax mount options which can override `FS_XFLAG_DAX` in the
58    setting of the `S_DAX` flag.  Given underlying storage which supports `DAX` the
59    following hold:
60
61    ``-o dax=inode``  means "follow `FS_XFLAG_DAX`" and is the default.
62
63    ``-o dax=never``  means "never set `S_DAX`, ignore `FS_XFLAG_DAX`."
64
65    ``-o dax=always`` means "always set `S_DAX` ignore `FS_XFLAG_DAX`."
66
67    ``-o dax``      is a legacy option which is an alias for ``dax=always``.
68
69    .. warning::
70
71      The option ``-o dax`` may be removed in the future so ``-o dax=always`` is
72      the preferred method for specifying this behavior.
73
74    .. note::
75
76      Modifications to and the inheritance behavior of `FS_XFLAG_DAX` remain
77      the same even when the filesystem is mounted with a dax option.  However,
78      in-core inode state (`S_DAX`) will be overridden until the filesystem is
79      remounted with dax=inode and the inode is evicted from kernel memory.
80
81 5. The `S_DAX` policy can be changed via:
82
83    a) Setting the parent directory `FS_XFLAG_DAX` as needed before files are
84       created
85
86    b) Setting the appropriate dax="foo" mount option
87
88    c) Changing the `FS_XFLAG_DAX` flag on existing regular files and
89       directories.  This has runtime constraints and limitations that are
90       described in 6) below.
91
92 6. When changing the `S_DAX` policy via toggling the persistent `FS_XFLAG_DAX`
93    flag, the change to existing regular files won't take effect until the
94    files are closed by all processes.
95
96
97Details
98-------
99
100There are 2 per-file dax flags.  One is a persistent inode setting (`FS_XFLAG_DAX`)
101and the other is a volatile flag indicating the active state of the feature
102(`S_DAX`).
103
104`FS_XFLAG_DAX` is preserved within the filesystem.  This persistent config
105setting can be set, cleared and/or queried using the `FS_IOC_FS`[`GS`]`ETXATTR` ioctl
106(see ioctl_xfs_fsgetxattr(2)) or an utility such as 'xfs_io'.
107
108New files and directories automatically inherit `FS_XFLAG_DAX` from
109their parent directory **when created**.  Therefore, setting `FS_XFLAG_DAX` at
110directory creation time can be used to set a default behavior for an entire
111sub-tree.
112
113To clarify inheritance, here are 3 examples:
114
115Example A:
116
117.. code-block:: shell
118
119  mkdir -p a/b/c
120  xfs_io -c 'chattr +x' a
121  mkdir a/b/c/d
122  mkdir a/e
123
124  ------[outcome]------
125
126  dax: a,e
127  no dax: b,c,d
128
129Example B:
130
131.. code-block:: shell
132
133  mkdir a
134  xfs_io -c 'chattr +x' a
135  mkdir -p a/b/c/d
136
137  ------[outcome]------
138
139  dax: a,b,c,d
140  no dax:
141
142Example C:
143
144.. code-block:: shell
145
146  mkdir -p a/b/c
147  xfs_io -c 'chattr +x' c
148  mkdir a/b/c/d
149
150  ------[outcome]------
151
152  dax: c,d
153  no dax: a,b
154
155The current enabled state (`S_DAX`) is set when a file inode is instantiated in
156memory by the kernel.  It is set based on the underlying media support, the
157value of `FS_XFLAG_DAX` and the filesystem's dax mount option.
158
159statx can be used to query `S_DAX`.
160
161.. note::
162
163  That only regular files will ever have `S_DAX` set and therefore statx
164  will never indicate that `S_DAX` is set on directories.
165
166Setting the `FS_XFLAG_DAX` flag (specifically or through inheritance) occurs even
167if the underlying media does not support dax and/or the filesystem is
168overridden with a mount option.
169
170
171Enabling DAX on virtiofs
172----------------------------
173The semantic of DAX on virtiofs is basically equal to that on ext4 and xfs,
174except that when '-o dax=inode' is specified, virtiofs client derives the hint
175whether DAX shall be enabled or not from virtiofs server through FUSE protocol,
176rather than the persistent `FS_XFLAG_DAX` flag. That is, whether DAX shall be
177enabled or not is completely determined by virtiofs server, while virtiofs
178server itself may deploy various algorithm making this decision, e.g. depending
179on the persistent `FS_XFLAG_DAX` flag on the host.
180
181It is still supported to set or clear persistent `FS_XFLAG_DAX` flag inside
182guest, but it is not guaranteed that DAX will be enabled or disabled for
183corresponding file then. Users inside guest still need to call statx(2) and
184check the statx flag `STATX_ATTR_DAX` to see if DAX is enabled for this file.
185
186
187Implementation Tips for Block Driver Writers
188--------------------------------------------
189
190To support `DAX` in your block driver, implement the 'direct_access'
191block device operation.  It is used to translate the sector number
192(expressed in units of 512-byte sectors) to a page frame number (pfn)
193that identifies the physical page for the memory.  It also returns a
194kernel virtual address that can be used to access the memory.
195
196The direct_access method takes a 'size' parameter that indicates the
197number of bytes being requested.  The function should return the number
198of bytes that can be contiguously accessed at that offset.  It may also
199return a negative errno if an error occurs.
200
201In order to support this method, the storage must be byte-accessible by
202the CPU at all times.  If your device uses paging techniques to expose
203a large amount of memory through a smaller window, then you cannot
204implement direct_access.  Equally, if your device can occasionally
205stall the CPU for an extended period, you should also not attempt to
206implement direct_access.
207
208These block devices may be used for inspiration:
209- brd: RAM backed block device driver
210- dcssblk: s390 dcss block device driver
211- pmem: NVDIMM persistent memory driver
212
213
214Implementation Tips for Filesystem Writers
215------------------------------------------
216
217Filesystem support consists of:
218
219* Adding support to mark inodes as being `DAX` by setting the `S_DAX` flag in
220  i_flags
221* Implementing ->read_iter and ->write_iter operations which use
222  :c:func:`dax_iomap_rw()` when inode has `S_DAX` flag set
223* Implementing an mmap file operation for `DAX` files which sets the
224  `VM_MIXEDMAP` and `VM_HUGEPAGE` flags on the `VMA`, and setting the vm_ops to
225  include handlers for fault, pmd_fault, page_mkwrite, pfn_mkwrite. These
226  handlers should probably call :c:func:`dax_iomap_fault()` passing the
227  appropriate fault size and iomap operations.
228* Calling :c:func:`iomap_zero_range()` passing appropriate iomap operations
229  instead of :c:func:`block_truncate_page()` for `DAX` files
230* Ensuring that there is sufficient locking between reads, writes,
231  truncates and page faults
232
233The iomap handlers for allocating blocks must make sure that allocated blocks
234are zeroed out and converted to written extents before being returned to avoid
235exposure of uninitialized data through mmap.
236
237These filesystems may be used for inspiration:
238
239.. seealso::
240
241  ext2: see Documentation/filesystems/ext2.rst
242
243.. seealso::
244
245  xfs:  see Documentation/admin-guide/xfs.rst
246
247.. seealso::
248
249  ext4: see Documentation/filesystems/ext4/
250
251
252Handling Media Errors
253---------------------
254
255The libnvdimm subsystem stores a record of known media error locations for
256each pmem block device (in gendisk->badblocks). If we fault at such location,
257or one with a latent error not yet discovered, the application can expect
258to receive a `SIGBUS`. Libnvdimm also allows clearing of these errors by simply
259writing the affected sectors (through the pmem driver, and if the underlying
260NVDIMM supports the clear_poison DSM defined by ACPI).
261
262Since `DAX` IO normally doesn't go through the ``driver/bio`` path, applications or
263sysadmins have an option to restore the lost data from a prior ``backup/inbuilt``
264redundancy in the following ways:
265
2661. Delete the affected file, and restore from a backup (sysadmin route):
267   This will free the filesystem blocks that were being used by the file,
268   and the next time they're allocated, they will be zeroed first, which
269   happens through the driver, and will clear bad sectors.
270
2712. Truncate or hole-punch the part of the file that has a bad-block (at least
272   an entire aligned sector has to be hole-punched, but not necessarily an
273   entire filesystem block).
274
275These are the two basic paths that allow `DAX` filesystems to continue operating
276in the presence of media errors. More robust error recovery mechanisms can be
277built on top of this in the future, for example, involving redundancy/mirroring
278provided at the block layer through DM, or additionally, at the filesystem
279level. These would have to rely on the above two tenets, that error clearing
280can happen either by sending an IO through the driver, or zeroing (also through
281the driver).
282
283
284Shortcomings
285------------
286
287Even if the kernel or its modules are stored on a filesystem that supports
288`DAX` on a block device that supports `DAX`, they will still be copied into RAM.
289
290The DAX code does not work correctly on architectures which have virtually
291mapped caches such as ARM, MIPS and SPARC.
292
293Calling :c:func:`get_user_pages()` on a range of user memory that has been
294mmapped from a `DAX` file will fail when there are no 'struct page' to describe
295those pages.  This problem has been addressed in some device drivers
296by adding optional struct page support for pages under the control of
297the driver (see `CONFIG_NVDIMM_PFN` in ``drivers/nvdimm`` for an example of
298how to do this). In the non struct page cases `O_DIRECT` reads/writes to
299those memory ranges from a non-`DAX` file will fail
300
301
302.. note::
303
304  `O_DIRECT` reads/writes _of a `DAX` file do work, it is the memory that
305  is being accessed that is key here).  Other things that will not work in
306  the non struct page case include RDMA, :c:func:`sendfile()` and
307  :c:func:`splice()`.
308