1 .. SPDX-License-Identifier: GPL-2.0 1 .. SPDX-License-Identifier: GPL-2.0
2 2
3 =========================================== 3 ===========================================
4 Userspace block device driver (ublk driver) 4 Userspace block device driver (ublk driver)
5 =========================================== 5 ===========================================
6 6
7 Overview 7 Overview
8 ======== 8 ========
9 9
10 ublk is a generic framework for implementing b 10 ublk is a generic framework for implementing block device logic from userspace.
11 The motivation behind it is that moving virtua 11 The motivation behind it is that moving virtual block drivers into userspace,
12 such as loop, nbd and similar can be very help 12 such as loop, nbd and similar can be very helpful. It can help to implement
13 new virtual block device such as ublk-qcow2 (t 13 new virtual block device such as ublk-qcow2 (there are several attempts of
14 implementing qcow2 driver in kernel). 14 implementing qcow2 driver in kernel).
15 15
16 Userspace block devices are attractive because 16 Userspace block devices are attractive because:
17 17
18 - They can be written many programming languag 18 - They can be written many programming languages.
19 - They can use libraries that are not availabl 19 - They can use libraries that are not available in the kernel.
20 - They can be debugged with tools familiar to 20 - They can be debugged with tools familiar to application developers.
21 - Crashes do not kernel panic the machine. 21 - Crashes do not kernel panic the machine.
22 - Bugs are likely to have a lower security imp 22 - Bugs are likely to have a lower security impact than bugs in kernel
23 code. 23 code.
24 - They can be installed and updated independen 24 - They can be installed and updated independently of the kernel.
25 - They can be used to simulate block device ea 25 - They can be used to simulate block device easily with user specified
26 parameters/setting for test/debug purpose 26 parameters/setting for test/debug purpose
27 27
28 ublk block device (``/dev/ublkb*``) is added b 28 ublk block device (``/dev/ublkb*``) is added by ublk driver. Any IO request
29 on the device will be forwarded to ublk usersp 29 on the device will be forwarded to ublk userspace program. For convenience,
30 in this document, ``ublk server`` refers to ge 30 in this document, ``ublk server`` refers to generic ublk userspace
31 program. ``ublksrv`` [#userspace]_ is one of s 31 program. ``ublksrv`` [#userspace]_ is one of such implementation. It
32 provides ``libublksrv`` [#userspace_lib]_ libr 32 provides ``libublksrv`` [#userspace_lib]_ library for developing specific
33 user block device conveniently, while also gen 33 user block device conveniently, while also generic type block device is
34 included, such as loop and null. Richard W.M. 34 included, such as loop and null. Richard W.M. Jones wrote userspace nbd device
35 ``nbdublk`` [#userspace_nbdublk]_ based on `` 35 ``nbdublk`` [#userspace_nbdublk]_ based on ``libublksrv`` [#userspace_lib]_.
36 36
37 After the IO is handled by userspace, the resu 37 After the IO is handled by userspace, the result is committed back to the
38 driver, thus completing the request cycle. Thi 38 driver, thus completing the request cycle. This way, any specific IO handling
39 logic is totally done by userspace, such as lo 39 logic is totally done by userspace, such as loop's IO handling, NBD's IO
40 communication, or qcow2's IO mapping. 40 communication, or qcow2's IO mapping.
41 41
42 ``/dev/ublkb*`` is driven by blk-mq request-ba 42 ``/dev/ublkb*`` is driven by blk-mq request-based driver. Each request is
43 assigned by one queue wide unique tag. ublk se 43 assigned by one queue wide unique tag. ublk server assigns unique tag to each
44 IO too, which is 1:1 mapped with IO of ``/dev/ 44 IO too, which is 1:1 mapped with IO of ``/dev/ublkb*``.
45 45
46 Both the IO request forward and IO handling re 46 Both the IO request forward and IO handling result committing are done via
47 ``io_uring`` passthrough command; that is why 47 ``io_uring`` passthrough command; that is why ublk is also one io_uring based
48 block driver. It has been observed that using 48 block driver. It has been observed that using io_uring passthrough command can
49 give better IOPS than block IO; which is why u 49 give better IOPS than block IO; which is why ublk is one of high performance
50 implementation of userspace block device: not 50 implementation of userspace block device: not only IO request communication is
51 done by io_uring, but also the preferred IO ha 51 done by io_uring, but also the preferred IO handling in ublk server is io_uring
52 based approach too. 52 based approach too.
53 53
54 ublk provides control interface to set/get ubl 54 ublk provides control interface to set/get ublk block device parameters.
55 The interface is extendable and kabi compatibl 55 The interface is extendable and kabi compatible: basically any ublk request
56 queue's parameter or ublk generic feature para 56 queue's parameter or ublk generic feature parameters can be set/get via the
57 interface. Thus, ublk is generic userspace blo 57 interface. Thus, ublk is generic userspace block device framework.
58 For example, it is easy to setup a ublk device 58 For example, it is easy to setup a ublk device with specified block
59 parameters from userspace. 59 parameters from userspace.
60 60
61 Using ublk 61 Using ublk
62 ========== 62 ==========
63 63
64 ublk requires userspace ublk server to handle 64 ublk requires userspace ublk server to handle real block device logic.
65 65
66 Below is example of using ``ublksrv`` to provi 66 Below is example of using ``ublksrv`` to provide ublk-based loop device.
67 67
68 - add a device:: 68 - add a device::
69 69
70 ublk add -t loop -f ublk-loop.img 70 ublk add -t loop -f ublk-loop.img
71 71
72 - format with xfs, then use it:: 72 - format with xfs, then use it::
73 73
74 mkfs.xfs /dev/ublkb0 74 mkfs.xfs /dev/ublkb0
75 mount /dev/ublkb0 /mnt 75 mount /dev/ublkb0 /mnt
76 # do anything. all IOs are handled by io_ 76 # do anything. all IOs are handled by io_uring
77 ... 77 ...
78 umount /mnt 78 umount /mnt
79 79
80 - list the devices with their info:: 80 - list the devices with their info::
81 81
82 ublk list 82 ublk list
83 83
84 - delete the device:: 84 - delete the device::
85 85
86 ublk del -a 86 ublk del -a
87 ublk del -n $ublk_dev_id 87 ublk del -n $ublk_dev_id
88 88
89 See usage details in README of ``ublksrv`` [#u 89 See usage details in README of ``ublksrv`` [#userspace_readme]_.
90 90
91 Design 91 Design
92 ====== 92 ======
93 93
94 Control plane 94 Control plane
95 ------------- 95 -------------
96 96
97 ublk driver provides global misc device node ( 97 ublk driver provides global misc device node (``/dev/ublk-control``) for
98 managing and controlling ublk devices with hel 98 managing and controlling ublk devices with help of several control commands:
99 99
100 - ``UBLK_CMD_ADD_DEV`` 100 - ``UBLK_CMD_ADD_DEV``
101 101
102 Add a ublk char device (``/dev/ublkc*``) whi 102 Add a ublk char device (``/dev/ublkc*``) which is talked with ublk server
103 WRT IO command communication. Basic device i 103 WRT IO command communication. Basic device info is sent together with this
104 command. It sets UAPI structure of ``ublksrv 104 command. It sets UAPI structure of ``ublksrv_ctrl_dev_info``,
105 such as ``nr_hw_queues``, ``queue_depth``, a 105 such as ``nr_hw_queues``, ``queue_depth``, and max IO request buffer size,
106 for which the info is negotiated with the dr 106 for which the info is negotiated with the driver and sent back to the server.
107 When this command is completed, the basic de 107 When this command is completed, the basic device info is immutable.
108 108
109 - ``UBLK_CMD_SET_PARAMS`` / ``UBLK_CMD_GET_PAR 109 - ``UBLK_CMD_SET_PARAMS`` / ``UBLK_CMD_GET_PARAMS``
110 110
111 Set or get parameters of the device, which c 111 Set or get parameters of the device, which can be either generic feature
112 related, or request queue limit related, but 112 related, or request queue limit related, but can't be IO logic specific,
113 because the driver does not handle any IO lo 113 because the driver does not handle any IO logic. This command has to be
114 sent before sending ``UBLK_CMD_START_DEV``. 114 sent before sending ``UBLK_CMD_START_DEV``.
115 115
116 - ``UBLK_CMD_START_DEV`` 116 - ``UBLK_CMD_START_DEV``
117 117
118 After the server prepares userspace resource 118 After the server prepares userspace resources (such as creating per-queue
119 pthread & io_uring for handling ublk IO), th 119 pthread & io_uring for handling ublk IO), this command is sent to the
120 driver for allocating & exposing ``/dev/ublk 120 driver for allocating & exposing ``/dev/ublkb*``. Parameters set via
121 ``UBLK_CMD_SET_PARAMS`` are applied for crea 121 ``UBLK_CMD_SET_PARAMS`` are applied for creating the device.
122 122
123 - ``UBLK_CMD_STOP_DEV`` 123 - ``UBLK_CMD_STOP_DEV``
124 124
125 Halt IO on ``/dev/ublkb*`` and remove the de 125 Halt IO on ``/dev/ublkb*`` and remove the device. When this command returns,
126 ublk server will release resources (such as 126 ublk server will release resources (such as destroying per-queue pthread &
127 io_uring). 127 io_uring).
128 128
129 - ``UBLK_CMD_DEL_DEV`` 129 - ``UBLK_CMD_DEL_DEV``
130 130
131 Remove ``/dev/ublkc*``. When this command re 131 Remove ``/dev/ublkc*``. When this command returns, the allocated ublk device
132 number can be reused. 132 number can be reused.
133 133
134 - ``UBLK_CMD_GET_QUEUE_AFFINITY`` 134 - ``UBLK_CMD_GET_QUEUE_AFFINITY``
135 135
136 When ``/dev/ublkc`` is added, the driver cre 136 When ``/dev/ublkc`` is added, the driver creates block layer tagset, so
137 that each queue's affinity info is available 137 that each queue's affinity info is available. The server sends
138 ``UBLK_CMD_GET_QUEUE_AFFINITY`` to retrieve 138 ``UBLK_CMD_GET_QUEUE_AFFINITY`` to retrieve queue affinity info. It can
139 set up the per-queue context efficiently, su 139 set up the per-queue context efficiently, such as bind affine CPUs with IO
140 pthread and try to allocate buffers in IO th 140 pthread and try to allocate buffers in IO thread context.
141 141
142 - ``UBLK_CMD_GET_DEV_INFO`` 142 - ``UBLK_CMD_GET_DEV_INFO``
143 143
144 For retrieving device info via ``ublksrv_ctr 144 For retrieving device info via ``ublksrv_ctrl_dev_info``. It is the server's
145 responsibility to save IO target specific in 145 responsibility to save IO target specific info in userspace.
146 146
147 - ``UBLK_CMD_GET_DEV_INFO2`` <<
148 Same purpose with ``UBLK_CMD_GET_DEV_INFO``, <<
149 provide path of the char device of ``/dev/ub <<
150 permission check, and this command is added <<
151 ublk device, and introduced with ``UBLK_F_UN <<
152 Only the user owning the requested device ca <<
153 <<
154 How to deal with userspace/kernel compatibil <<
155 <<
156 1) if kernel is capable of handling ``UBLK_F <<
157 <<
158 If ublk server supports ``UBLK_F_UNPRIVILE <<
159 <<
160 ublk server should send ``UBLK_CMD_GET_DEV <<
161 unprivileged application needs to query de <<
162 when the application has no idea if ``UBLK <<
163 given the capability info is stateless, an <<
164 retrieve it via ``UBLK_CMD_GET_DEV_INFO2`` <<
165 <<
166 If ublk server doesn't support ``UBLK_F_UN <<
167 <<
168 ``UBLK_CMD_GET_DEV_INFO`` is always sent t <<
169 UBLK_F_UNPRIVILEGED_DEV isn't available fo <<
170 <<
171 2) if kernel isn't capable of handling ``UBL <<
172 <<
173 If ublk server supports ``UBLK_F_UNPRIVILE <<
174 <<
175 ``UBLK_CMD_GET_DEV_INFO2`` is tried first, <<
176 ``UBLK_CMD_GET_DEV_INFO`` needs to be retr <<
177 ``UBLK_F_UNPRIVILEGED_DEV`` can't be set <<
178 <<
179 If ublk server doesn't support ``UBLK_F_UN <<
180 <<
181 ``UBLK_CMD_GET_DEV_INFO`` is always sent t <<
182 ``UBLK_F_UNPRIVILEGED_DEV`` isn't availabl <<
183 <<
184 - ``UBLK_CMD_START_USER_RECOVERY`` <<
185 <<
186 This command is valid if ``UBLK_F_USER_RECOV <<
187 command is accepted after the old process ha <<
188 and ``/dev/ublkc*`` is released. User should <<
189 a new process which re-opens ``/dev/ublkc*`` <<
190 ublk device is ready for the new process. <<
191 <<
192 - ``UBLK_CMD_END_USER_RECOVERY`` <<
193 <<
194 This command is valid if ``UBLK_F_USER_RECOV <<
195 command is accepted after ublk device is qui <<
196 opened ``/dev/ublkc*`` and get all ublk queu <<
197 returns, ublk device is unquiesced and new I <<
198 new process. <<
199 <<
200 - user recovery feature description <<
201 <<
202 Two new features are added for user recovery <<
203 ``UBLK_F_USER_RECOVERY_REISSUE``. <<
204 <<
205 With ``UBLK_F_USER_RECOVERY`` set, after one <<
206 handler) is dying, ublk does not delete ``/d <<
207 recovery stage and ublk device ID is kept. I <<
208 responsibility to recover the device context <<
209 Requests which have not been issued to users <<
210 which have been issued to userspace are abor <<
211 <<
212 With ``UBLK_F_USER_RECOVERY_REISSUE`` set, a <<
213 server's io handler) is dying, contrary to ` <<
214 requests which have been issued to userspace <<
215 re-issued to the new process after handling <<
216 ``UBLK_F_USER_RECOVERY_REISSUE`` is designed <<
217 double-write since the driver may issue the <<
218 might be useful to a read-only FS or a VM ba <<
219 <<
220 Unprivileged ublk device is supported by passi <<
221 Once the flag is set, all control commands can <<
222 user. Except for command of ``UBLK_CMD_ADD_DEV <<
223 the specified char device(``/dev/ublkc*``) is <<
224 commands by ublk driver, for doing that, path <<
225 be provided in these commands' payload from ub <<
226 ublk device becomes container-ware, and device <<
227 can be controlled/accessed just inside this co <<
228 <<
229 Data plane 147 Data plane
230 ---------- 148 ----------
231 149
232 ublk server needs to create per-queue IO pthre 150 ublk server needs to create per-queue IO pthread & io_uring for handling IO
233 commands via io_uring passthrough. The per-que 151 commands via io_uring passthrough. The per-queue IO pthread
234 focuses on IO handling and shouldn't handle an 152 focuses on IO handling and shouldn't handle any control & management
235 tasks. 153 tasks.
236 154
237 The's IO is assigned by a unique tag, which is 155 The's IO is assigned by a unique tag, which is 1:1 mapping with IO
238 request of ``/dev/ublkb*``. 156 request of ``/dev/ublkb*``.
239 157
240 UAPI structure of ``ublksrv_io_desc`` is defin 158 UAPI structure of ``ublksrv_io_desc`` is defined for describing each IO from
241 the driver. A fixed mmapped area (array) on `` !! 159 the driver. A fixed mmaped area (array) on ``/dev/ublkc*`` is provided for
242 exporting IO info to the server; such as IO of 160 exporting IO info to the server; such as IO offset, length, OP/flags and
243 buffer address. Each ``ublksrv_io_desc`` insta 161 buffer address. Each ``ublksrv_io_desc`` instance can be indexed via queue id
244 and IO tag directly. 162 and IO tag directly.
245 163
246 The following IO commands are communicated via 164 The following IO commands are communicated via io_uring passthrough command,
247 and each command is only for forwarding the IO 165 and each command is only for forwarding the IO and committing the result
248 with specified IO tag in the command data: 166 with specified IO tag in the command data:
249 167
250 - ``UBLK_IO_FETCH_REQ`` 168 - ``UBLK_IO_FETCH_REQ``
251 169
252 Sent from the server IO pthread for fetching 170 Sent from the server IO pthread for fetching future incoming IO requests
253 destined to ``/dev/ublkb*``. This command is 171 destined to ``/dev/ublkb*``. This command is sent only once from the server
254 IO pthread for ublk driver to setup IO forwa 172 IO pthread for ublk driver to setup IO forward environment.
255 173
256 - ``UBLK_IO_COMMIT_AND_FETCH_REQ`` 174 - ``UBLK_IO_COMMIT_AND_FETCH_REQ``
257 175
258 When an IO request is destined to ``/dev/ubl 176 When an IO request is destined to ``/dev/ublkb*``, the driver stores
259 the IO's ``ublksrv_io_desc`` to the specifie 177 the IO's ``ublksrv_io_desc`` to the specified mapped area; then the
260 previous received IO command of this IO tag 178 previous received IO command of this IO tag (either ``UBLK_IO_FETCH_REQ``
261 or ``UBLK_IO_COMMIT_AND_FETCH_REQ)`` is comp 179 or ``UBLK_IO_COMMIT_AND_FETCH_REQ)`` is completed, so the server gets
262 the IO notification via io_uring. 180 the IO notification via io_uring.
263 181
264 After the server handles the IO, its result 182 After the server handles the IO, its result is committed back to the
265 driver by sending ``UBLK_IO_COMMIT_AND_FETCH 183 driver by sending ``UBLK_IO_COMMIT_AND_FETCH_REQ`` back. Once ublkdrv
266 received this command, it parses the result 184 received this command, it parses the result and complete the request to
267 ``/dev/ublkb*``. In the meantime setup envir 185 ``/dev/ublkb*``. In the meantime setup environment for fetching future
268 requests with the same IO tag. That is, ``UB 186 requests with the same IO tag. That is, ``UBLK_IO_COMMIT_AND_FETCH_REQ``
269 is reused for both fetching request and comm 187 is reused for both fetching request and committing back IO result.
270 188
271 - ``UBLK_IO_NEED_GET_DATA`` 189 - ``UBLK_IO_NEED_GET_DATA``
272 190
273 With ``UBLK_F_NEED_GET_DATA`` enabled, the W 191 With ``UBLK_F_NEED_GET_DATA`` enabled, the WRITE request will be firstly
274 issued to ublk server without data copy. The 192 issued to ublk server without data copy. Then, IO backend of ublk server
275 receives the request and it can allocate dat 193 receives the request and it can allocate data buffer and embed its addr
276 inside this new io command. After the kernel 194 inside this new io command. After the kernel driver gets the command,
277 data copy is done from request pages to this 195 data copy is done from request pages to this backend's buffer. Finally,
278 backend receives the request again with data 196 backend receives the request again with data to be written and it can
279 truly handle the request. 197 truly handle the request.
280 198
281 ``UBLK_IO_NEED_GET_DATA`` adds one additiona 199 ``UBLK_IO_NEED_GET_DATA`` adds one additional round-trip and one
282 io_uring_enter() syscall. Any user thinks th 200 io_uring_enter() syscall. Any user thinks that it may lower performance
283 should not enable UBLK_F_NEED_GET_DATA. ublk 201 should not enable UBLK_F_NEED_GET_DATA. ublk server pre-allocates IO
284 buffer for each IO by default. Any new proje 202 buffer for each IO by default. Any new project should try to use this
285 buffer to communicate with ublk driver. Howe 203 buffer to communicate with ublk driver. However, existing project may
286 break or not able to consume the new buffer 204 break or not able to consume the new buffer interface; that's why this
287 command is added for backwards compatibility 205 command is added for backwards compatibility so that existing projects
288 can still consume existing buffers. 206 can still consume existing buffers.
289 207
290 - data copy between ublk server IO buffer and 208 - data copy between ublk server IO buffer and ublk block IO request
291 209
292 The driver needs to copy the block IO reques 210 The driver needs to copy the block IO request pages into the server buffer
293 (pages) first for WRITE before notifying the 211 (pages) first for WRITE before notifying the server of the coming IO, so
294 that the server can handle WRITE request. 212 that the server can handle WRITE request.
295 213
296 When the server handles READ request and sen 214 When the server handles READ request and sends
297 ``UBLK_IO_COMMIT_AND_FETCH_REQ`` to the serv 215 ``UBLK_IO_COMMIT_AND_FETCH_REQ`` to the server, ublkdrv needs to copy
298 the server buffer (pages) read to the IO req 216 the server buffer (pages) read to the IO request pages.
299 217
300 Future development 218 Future development
301 ================== 219 ==================
>> 220
>> 221 Container-aware ublk deivice
>> 222 ----------------------------
>> 223
>> 224 ublk driver doesn't handle any IO logic. Its function is well defined
>> 225 for now and very limited userspace interfaces are needed, which is also
>> 226 well defined too. It is possible to make ublk devices container-aware block
>> 227 devices in future as Stefan Hajnoczi suggested [#stefan]_, by removing
>> 228 ADMIN privilege.
302 229
303 Zero copy 230 Zero copy
304 --------- 231 ---------
305 232
306 Zero copy is a generic requirement for nbd, fu 233 Zero copy is a generic requirement for nbd, fuse or similar drivers. A
307 problem [#xiaoguang]_ Xiaoguang mentioned is t 234 problem [#xiaoguang]_ Xiaoguang mentioned is that pages mapped to userspace
308 can't be remapped any more in kernel with exis 235 can't be remapped any more in kernel with existing mm interfaces. This can
309 occurs when destining direct IO to ``/dev/ublk 236 occurs when destining direct IO to ``/dev/ublkb*``. Also, he reported that
310 big requests (IO size >= 256 KB) may benefit a 237 big requests (IO size >= 256 KB) may benefit a lot from zero copy.
311 238
312 239
313 References 240 References
314 ========== 241 ==========
315 242
316 .. [#userspace] https://github.com/ming1/ubdsr 243 .. [#userspace] https://github.com/ming1/ubdsrv
317 244
318 .. [#userspace_lib] https://github.com/ming1/u 245 .. [#userspace_lib] https://github.com/ming1/ubdsrv/tree/master/lib
319 246
320 .. [#userspace_nbdublk] https://gitlab.com/rwm 247 .. [#userspace_nbdublk] https://gitlab.com/rwmjones/libnbd/-/tree/nbdublk
321 248
322 .. [#userspace_readme] https://github.com/ming 249 .. [#userspace_readme] https://github.com/ming1/ubdsrv/blob/master/README
323 250
324 .. [#stefan] https://lore.kernel.org/linux-blo 251 .. [#stefan] https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
325 252
326 .. [#xiaoguang] https://lore.kernel.org/linux- 253 .. [#xiaoguang] https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.