1 ======================= 1 =======================
2 Direct Access for files 2 Direct Access for files
3 ======================= 3 =======================
4 4
5 Motivation 5 Motivation
6 ---------- 6 ----------
7 7
8 The page cache is usually used to buffer reads 8 The page cache is usually used to buffer reads and writes to files.
9 It is also used to provide the pages which are 9 It is also used to provide the pages which are mapped into userspace
10 by a call to mmap. 10 by a call to mmap.
11 11
12 For block devices that are memory-like, the pa 12 For block devices that are memory-like, the page cache pages would be
13 unnecessary copies of the original storage. T 13 unnecessary copies of the original storage. The `DAX` code removes the
14 extra copy by performing reads and writes dire 14 extra copy by performing reads and writes directly to the storage device.
15 For file mappings, the storage device is mappe 15 For file mappings, the storage device is mapped directly into userspace.
16 16
17 17
18 Usage 18 Usage
19 ----- 19 -----
20 20
21 If you have a block device which supports `DAX 21 If you have a block device which supports `DAX`, you can make a filesystem
22 on it as usual. The `DAX` code currently only 22 on it as usual. The `DAX` code currently only supports files with a block
23 size equal to your kernel's `PAGE_SIZE`, so yo 23 size equal to your kernel's `PAGE_SIZE`, so you may need to specify a block
24 size when creating the filesystem. 24 size when creating the filesystem.
25 25
26 Currently 5 filesystems support `DAX`: ext2, e 26 Currently 5 filesystems support `DAX`: ext2, ext4, xfs, virtiofs and erofs.
27 Enabling `DAX` on them is different. 27 Enabling `DAX` on them is different.
28 28
29 Enabling DAX on ext2 and erofs 29 Enabling DAX on ext2 and erofs
30 ------------------------------ 30 ------------------------------
31 31
32 When mounting the filesystem, use the ``-o dax 32 When mounting the filesystem, use the ``-o dax`` option on the command line or
33 add 'dax' to the options in ``/etc/fstab``. T 33 add 'dax' to the options in ``/etc/fstab``. This works to enable `DAX` on all files
34 within the filesystem. It is equivalent to th 34 within the filesystem. It is equivalent to the ``-o dax=always`` behavior below.
35 35
36 36
37 Enabling DAX on xfs and ext4 37 Enabling DAX on xfs and ext4
38 ---------------------------- 38 ----------------------------
39 39
40 Summary 40 Summary
41 ------- 41 -------
42 42
43 1. There exists an in-kernel file access mode 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 44 the statx flag `STATX_ATTR_DAX`. See the manpage for statx(2) for details
45 about this access mode. 45 about this access mode.
46 46
47 2. There exists a persistent flag `FS_XFLAG_D 47 2. There exists a persistent flag `FS_XFLAG_DAX` that can be applied to regular
48 files and directories. This advisory flag 48 files and directories. This advisory flag can be set or cleared at any
49 time, but doing so does not immediately af 49 time, but doing so does not immediately affect the `S_DAX` state.
50 50
51 3. If the persistent `FS_XFLAG_DAX` flag is s 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 subd 52 be inherited by all regular files and subdirectories that are subsequently
53 created in this directory. Files and subdi 53 created in this directory. Files and subdirectories that exist at the time
54 this flag is set or cleared on the parent 54 this flag is set or cleared on the parent directory are not modified by
55 this modification of the parent directory. 55 this modification of the parent directory.
56 56
57 4. There exist dax mount options which can ov 57 4. There exist dax mount options which can override `FS_XFLAG_DAX` in the
58 setting of the `S_DAX` flag. Given underl 58 setting of the `S_DAX` flag. Given underlying storage which supports `DAX` the
59 following hold: 59 following hold:
60 60
61 ``-o dax=inode`` means "follow `FS_XFLAG_ 61 ``-o dax=inode`` means "follow `FS_XFLAG_DAX`" and is the default.
62 62
63 ``-o dax=never`` means "never set `S_DAX` 63 ``-o dax=never`` means "never set `S_DAX`, ignore `FS_XFLAG_DAX`."
64 64
65 ``-o dax=always`` means "always set `S_DAX 65 ``-o dax=always`` means "always set `S_DAX` ignore `FS_XFLAG_DAX`."
66 66
67 ``-o dax`` is a legacy option which i 67 ``-o dax`` is a legacy option which is an alias for ``dax=always``.
68 68
69 .. warning:: 69 .. warning::
70 70
71 The option ``-o dax`` may be removed in 71 The option ``-o dax`` may be removed in the future so ``-o dax=always`` is
72 the preferred method for specifying this 72 the preferred method for specifying this behavior.
73 73
74 .. note:: 74 .. note::
75 75
76 Modifications to and the inheritance beh 76 Modifications to and the inheritance behavior of `FS_XFLAG_DAX` remain
77 the same even when the filesystem is mou 77 the same even when the filesystem is mounted with a dax option. However,
78 in-core inode state (`S_DAX`) will be ov 78 in-core inode state (`S_DAX`) will be overridden until the filesystem is
79 remounted with dax=inode and the inode i 79 remounted with dax=inode and the inode is evicted from kernel memory.
80 80
81 5. The `S_DAX` policy can be changed via: 81 5. The `S_DAX` policy can be changed via:
82 82
83 a) Setting the parent directory `FS_XFLAG_ 83 a) Setting the parent directory `FS_XFLAG_DAX` as needed before files are
84 created 84 created
85 85
86 b) Setting the appropriate dax="foo" mount 86 b) Setting the appropriate dax="foo" mount option
87 87
88 c) Changing the `FS_XFLAG_DAX` flag on exi 88 c) Changing the `FS_XFLAG_DAX` flag on existing regular files and
89 directories. This has runtime constrai 89 directories. This has runtime constraints and limitations that are
90 described in 6) below. 90 described in 6) below.
91 91
92 6. When changing the `S_DAX` policy via toggl 92 6. When changing the `S_DAX` policy via toggling the persistent `FS_XFLAG_DAX`
93 flag, the change to existing regular files 93 flag, the change to existing regular files won't take effect until the
94 files are closed by all processes. 94 files are closed by all processes.
95 95
96 96
97 Details 97 Details
98 ------- 98 -------
99 99
100 There are 2 per-file dax flags. One is a pers 100 There are 2 per-file dax flags. One is a persistent inode setting (`FS_XFLAG_DAX`)
101 and the other is a volatile flag indicating th 101 and the other is a volatile flag indicating the active state of the feature
102 (`S_DAX`). 102 (`S_DAX`).
103 103
104 `FS_XFLAG_DAX` is preserved within the filesys 104 `FS_XFLAG_DAX` is preserved within the filesystem. This persistent config
105 setting can be set, cleared and/or queried usi 105 setting can be set, cleared and/or queried using the `FS_IOC_FS`[`GS`]`ETXATTR` ioctl
106 (see ioctl_xfs_fsgetxattr(2)) or an utility su 106 (see ioctl_xfs_fsgetxattr(2)) or an utility such as 'xfs_io'.
107 107
108 New files and directories automatically inheri 108 New files and directories automatically inherit `FS_XFLAG_DAX` from
109 their parent directory **when created**. Ther 109 their parent directory **when created**. Therefore, setting `FS_XFLAG_DAX` at
110 directory creation time can be used to set a d 110 directory creation time can be used to set a default behavior for an entire
111 sub-tree. 111 sub-tree.
112 112
113 To clarify inheritance, here are 3 examples: 113 To clarify inheritance, here are 3 examples:
114 114
115 Example A: 115 Example A:
116 116
117 .. code-block:: shell 117 .. code-block:: shell
118 118
119 mkdir -p a/b/c 119 mkdir -p a/b/c
120 xfs_io -c 'chattr +x' a 120 xfs_io -c 'chattr +x' a
121 mkdir a/b/c/d 121 mkdir a/b/c/d
122 mkdir a/e 122 mkdir a/e
123 123
124 ------[outcome]------ 124 ------[outcome]------
125 125
126 dax: a,e 126 dax: a,e
127 no dax: b,c,d 127 no dax: b,c,d
128 128
129 Example B: 129 Example B:
130 130
131 .. code-block:: shell 131 .. code-block:: shell
132 132
133 mkdir a 133 mkdir a
134 xfs_io -c 'chattr +x' a 134 xfs_io -c 'chattr +x' a
135 mkdir -p a/b/c/d 135 mkdir -p a/b/c/d
136 136
137 ------[outcome]------ 137 ------[outcome]------
138 138
139 dax: a,b,c,d 139 dax: a,b,c,d
140 no dax: 140 no dax:
141 141
142 Example C: 142 Example C:
143 143
144 .. code-block:: shell 144 .. code-block:: shell
145 145
146 mkdir -p a/b/c 146 mkdir -p a/b/c
147 xfs_io -c 'chattr +x' c 147 xfs_io -c 'chattr +x' c
148 mkdir a/b/c/d 148 mkdir a/b/c/d
149 149
150 ------[outcome]------ 150 ------[outcome]------
151 151
152 dax: c,d 152 dax: c,d
153 no dax: a,b 153 no dax: a,b
154 154
155 The current enabled state (`S_DAX`) is set whe 155 The current enabled state (`S_DAX`) is set when a file inode is instantiated in
156 memory by the kernel. It is set based on the 156 memory by the kernel. It is set based on the underlying media support, the
157 value of `FS_XFLAG_DAX` and the filesystem's d 157 value of `FS_XFLAG_DAX` and the filesystem's dax mount option.
158 158
159 statx can be used to query `S_DAX`. 159 statx can be used to query `S_DAX`.
160 160
161 .. note:: 161 .. note::
162 162
163 That only regular files will ever have `S_DA 163 That only regular files will ever have `S_DAX` set and therefore statx
164 will never indicate that `S_DAX` is set on d 164 will never indicate that `S_DAX` is set on directories.
165 165
166 Setting the `FS_XFLAG_DAX` flag (specifically 166 Setting the `FS_XFLAG_DAX` flag (specifically or through inheritance) occurs even
167 if the underlying media does not support dax a 167 if the underlying media does not support dax and/or the filesystem is
168 overridden with a mount option. 168 overridden with a mount option.
169 169
170 170
171 Enabling DAX on virtiofs 171 Enabling DAX on virtiofs
172 ---------------------------- 172 ----------------------------
173 The semantic of DAX on virtiofs is basically e 173 The semantic of DAX on virtiofs is basically equal to that on ext4 and xfs,
174 except that when '-o dax=inode' is specified, 174 except that when '-o dax=inode' is specified, virtiofs client derives the hint
175 whether DAX shall be enabled or not from virti 175 whether DAX shall be enabled or not from virtiofs server through FUSE protocol,
176 rather than the persistent `FS_XFLAG_DAX` flag 176 rather than the persistent `FS_XFLAG_DAX` flag. That is, whether DAX shall be
177 enabled or not is completely determined by vir 177 enabled or not is completely determined by virtiofs server, while virtiofs
178 server itself may deploy various algorithm mak 178 server itself may deploy various algorithm making this decision, e.g. depending
179 on the persistent `FS_XFLAG_DAX` flag on the h 179 on the persistent `FS_XFLAG_DAX` flag on the host.
180 180
181 It is still supported to set or clear persiste 181 It is still supported to set or clear persistent `FS_XFLAG_DAX` flag inside
182 guest, but it is not guaranteed that DAX will 182 guest, but it is not guaranteed that DAX will be enabled or disabled for
183 corresponding file then. Users inside guest st 183 corresponding file then. Users inside guest still need to call statx(2) and
184 check the statx flag `STATX_ATTR_DAX` to see i 184 check the statx flag `STATX_ATTR_DAX` to see if DAX is enabled for this file.
185 185
186 186
187 Implementation Tips for Block Driver Writers 187 Implementation Tips for Block Driver Writers
188 -------------------------------------------- 188 --------------------------------------------
189 189
190 To support `DAX` in your block driver, impleme 190 To support `DAX` in your block driver, implement the 'direct_access'
191 block device operation. It is used to transla 191 block device operation. It is used to translate the sector number
192 (expressed in units of 512-byte sectors) to a 192 (expressed in units of 512-byte sectors) to a page frame number (pfn)
193 that identifies the physical page for the memo 193 that identifies the physical page for the memory. It also returns a
194 kernel virtual address that can be used to acc 194 kernel virtual address that can be used to access the memory.
195 195
196 The direct_access method takes a 'size' parame 196 The direct_access method takes a 'size' parameter that indicates the
197 number of bytes being requested. The function 197 number of bytes being requested. The function should return the number
198 of bytes that can be contiguously accessed at 198 of bytes that can be contiguously accessed at that offset. It may also
199 return a negative errno if an error occurs. 199 return a negative errno if an error occurs.
200 200
201 In order to support this method, the storage m 201 In order to support this method, the storage must be byte-accessible by
202 the CPU at all times. If your device uses pag 202 the CPU at all times. If your device uses paging techniques to expose
203 a large amount of memory through a smaller win 203 a large amount of memory through a smaller window, then you cannot
204 implement direct_access. Equally, if your dev 204 implement direct_access. Equally, if your device can occasionally
205 stall the CPU for an extended period, you shou 205 stall the CPU for an extended period, you should also not attempt to
206 implement direct_access. 206 implement direct_access.
207 207
208 These block devices may be used for inspiratio 208 These block devices may be used for inspiration:
209 - brd: RAM backed block device driver 209 - brd: RAM backed block device driver
210 - dcssblk: s390 dcss block device driver 210 - dcssblk: s390 dcss block device driver
211 - pmem: NVDIMM persistent memory driver 211 - pmem: NVDIMM persistent memory driver
212 212
213 213
214 Implementation Tips for Filesystem Writers 214 Implementation Tips for Filesystem Writers
215 ------------------------------------------ 215 ------------------------------------------
216 216
217 Filesystem support consists of: 217 Filesystem support consists of:
218 218
219 * Adding support to mark inodes as being `DAX` 219 * Adding support to mark inodes as being `DAX` by setting the `S_DAX` flag in
220 i_flags 220 i_flags
221 * Implementing ->read_iter and ->write_iter op 221 * Implementing ->read_iter and ->write_iter operations which use
222 :c:func:`dax_iomap_rw()` when inode has `S_D 222 :c:func:`dax_iomap_rw()` when inode has `S_DAX` flag set
223 * Implementing an mmap file operation for `DAX 223 * Implementing an mmap file operation for `DAX` files which sets the
224 `VM_MIXEDMAP` and `VM_HUGEPAGE` flags on 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_ 225 include handlers for fault, pmd_fault, page_mkwrite, pfn_mkwrite. These
226 handlers should probably call :c:func:`dax_i 226 handlers should probably call :c:func:`dax_iomap_fault()` passing the
227 appropriate fault size and iomap operations. 227 appropriate fault size and iomap operations.
228 * Calling :c:func:`iomap_zero_range()` passing 228 * Calling :c:func:`iomap_zero_range()` passing appropriate iomap operations
229 instead of :c:func:`block_truncate_page()` f 229 instead of :c:func:`block_truncate_page()` for `DAX` files
230 * Ensuring that there is sufficient locking be 230 * Ensuring that there is sufficient locking between reads, writes,
231 truncates and page faults 231 truncates and page faults
232 232
233 The iomap handlers for allocating blocks must 233 The iomap handlers for allocating blocks must make sure that allocated blocks
234 are zeroed out and converted to written extent 234 are zeroed out and converted to written extents before being returned to avoid
235 exposure of uninitialized data through mmap. 235 exposure of uninitialized data through mmap.
236 236
237 These filesystems may be used for inspiration: 237 These filesystems may be used for inspiration:
238 238
239 .. seealso:: 239 .. seealso::
240 240
241 ext2: see Documentation/filesystems/ext2.rst 241 ext2: see Documentation/filesystems/ext2.rst
242 242
243 .. seealso:: 243 .. seealso::
244 244
245 xfs: see Documentation/admin-guide/xfs.rst 245 xfs: see Documentation/admin-guide/xfs.rst
246 246
247 .. seealso:: 247 .. seealso::
248 248
249 ext4: see Documentation/filesystems/ext4/ 249 ext4: see Documentation/filesystems/ext4/
250 250
251 251
252 Handling Media Errors 252 Handling Media Errors
253 --------------------- 253 ---------------------
254 254
255 The libnvdimm subsystem stores a record of kno 255 The libnvdimm subsystem stores a record of known media error locations for
256 each pmem block device (in gendisk->badblocks) 256 each pmem block device (in gendisk->badblocks). If we fault at such location,
257 or one with a latent error not yet discovered, 257 or one with a latent error not yet discovered, the application can expect
258 to receive a `SIGBUS`. Libnvdimm also allows c 258 to receive a `SIGBUS`. Libnvdimm also allows clearing of these errors by simply
259 writing the affected sectors (through the pmem 259 writing the affected sectors (through the pmem driver, and if the underlying
260 NVDIMM supports the clear_poison DSM defined b 260 NVDIMM supports the clear_poison DSM defined by ACPI).
261 261
262 Since `DAX` IO normally doesn't go through the 262 Since `DAX` IO normally doesn't go through the ``driver/bio`` path, applications or
263 sysadmins have an option to restore the lost d 263 sysadmins have an option to restore the lost data from a prior ``backup/inbuilt``
264 redundancy in the following ways: 264 redundancy in the following ways:
265 265
266 1. Delete the affected file, and restore from 266 1. Delete the affected file, and restore from a backup (sysadmin route):
267 This will free the filesystem blocks that w 267 This will free the filesystem blocks that were being used by the file,
268 and the next time they're allocated, they w 268 and the next time they're allocated, they will be zeroed first, which
269 happens through the driver, and will clear 269 happens through the driver, and will clear bad sectors.
270 270
271 2. Truncate or hole-punch the part of the file 271 2. 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-pun 272 an entire aligned sector has to be hole-punched, but not necessarily an
273 entire filesystem block). 273 entire filesystem block).
274 274
275 These are the two basic paths that allow `DAX` 275 These are the two basic paths that allow `DAX` filesystems to continue operating
276 in the presence of media errors. More robust e 276 in the presence of media errors. More robust error recovery mechanisms can be
277 built on top of this in the future, for exampl 277 built on top of this in the future, for example, involving redundancy/mirroring
278 provided at the block layer through DM, or add 278 provided at the block layer through DM, or additionally, at the filesystem
279 level. These would have to rely on the above t 279 level. These would have to rely on the above two tenets, that error clearing
280 can happen either by sending an IO through the 280 can happen either by sending an IO through the driver, or zeroing (also through
281 the driver). 281 the driver).
282 282
283 283
284 Shortcomings 284 Shortcomings
285 ------------ 285 ------------
286 286
287 Even if the kernel or its modules are stored o 287 Even if the kernel or its modules are stored on a filesystem that supports
288 `DAX` on a block device that supports `DAX`, t 288 `DAX` on a block device that supports `DAX`, they will still be copied into RAM.
289 289
290 The DAX code does not work correctly on archit 290 The DAX code does not work correctly on architectures which have virtually
291 mapped caches such as ARM, MIPS and SPARC. 291 mapped caches such as ARM, MIPS and SPARC.
292 292
293 Calling :c:func:`get_user_pages()` on a range 293 Calling :c:func:`get_user_pages()` on a range of user memory that has been
294 mmapped from a `DAX` file will fail when there 294 mmapped from a `DAX` file will fail when there are no 'struct page' to describe
295 those pages. This problem has been addressed 295 those pages. This problem has been addressed in some device drivers
296 by adding optional struct page support for pag 296 by adding optional struct page support for pages under the control of
297 the driver (see `CONFIG_NVDIMM_PFN` in ``drive 297 the driver (see `CONFIG_NVDIMM_PFN` in ``drivers/nvdimm`` for an example of
298 how to do this). In the non struct page cases 298 how to do this). In the non struct page cases `O_DIRECT` reads/writes to
299 those memory ranges from a non-`DAX` file will 299 those memory ranges from a non-`DAX` file will fail
300 300
301 301
302 .. note:: 302 .. note::
303 303
304 `O_DIRECT` reads/writes _of a `DAX` file do 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 305 is being accessed that is key here). Other things that will not work in
306 the non struct page case include RDMA, :c:fu 306 the non struct page case include RDMA, :c:func:`sendfile()` and
307 :c:func:`splice()`. 307 :c:func:`splice()`.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.