On Sun, Aug 28, 2022 at 12:50:03PM +0800, Ming Lei wrote: > ublk document is missed when merging ublk driver, so add it now. > Better say "Add documentation for ublk subsystem. It was supposed to be documented when merging the driver, but missing at that time." > diff --git a/Documentation/block/ublk.rst b/Documentation/block/ublk.rst > new file mode 100644 > index 000000000000..9e8f7ba518a3 > --- /dev/null > +++ b/Documentation/block/ublk.rst > @@ -0,0 +1,203 @@ > +.. SPDX-License-Identifier: GPL-2.0 > + > +========================================== > +Userspace block device driver(ublk driver) > +========================================== > + > +Overview > +======== > + > +ublk is one generic framework for implementing block device logic from > +userspace. It is very helpful to move virtual block drivers into userspace, > +such as loop, nbd and similar block drivers. It can help to implement new > +virtual block device, such as ublk-qcow2, and there was several attempts > +of implementing qcow2 driver in kernel. > + > +ublk block device(``/dev/ublkb*``) is added by ublk driver. Any IO request > +submitted to ublk device will be forwarded to ublk's userspace part( > +ublksrv [1]), and after the IO is handled by ublksrv, the result is > +committed back to ublk driver, then ublk IO request can be completed. With > +this way, any specific IO handling logic is totally done inside ublksrv, > +and ublk driver doe _not_ handle any device specific IO logic, such as > +loop's IO handling, NBD's IO communication, or qcow2's IO mapping, ... > + > +/dev/ublkbN is driven by blk-mq request based driver, each request is > +assigned by one queue wide unique tag. ublksrv assigns unique tag to each > +IO too, which is 1:1 mapped with IO of /dev/ublkb*. > + > +Both the IO request forward and IO handling result committing are done via > +io_uring passthrough command, that is why ublk is also one io_uring based > +block driver. It has been observed that io_uring passthrough command can get > +better IOPS than block IO. So ublk is one high performance implementation > +of userspace block device. Not only IO request communication is done by > +io_uring, but also the preferred IO handling in ublksrv is io_uring based > +approach too. > + > +ublk provides control interface to set/get ublk block device parameters, and > +the interface is extendable and kabi compatible, so basically any ublk request > +queue's parameter or ublk generic feature parameters can be set/get via this > +extendable interface. So ublk is generic userspace block device framework, such > +as, it is easy to setup one ublk device with specified block parameters from > +userspace. > + > +How to use ublk > +=============== > + > +After building ublksrv[1], ublk block device(``/dev/ublkb*``) can be added > +and deleted by the utility, then existed block IO applications can talk with > +it. > + > +See usage details in README[2] of ublksrv, for example of ublk-loop: > + > +- add ublk device: > + ublk add -t loop -f ublk-loop.img > + > +- use it: > + mkfs.xfs /dev/ublkb0 > + mount /dev/ublkb0 /mnt > + .... # all IOs are handled by io_uring!!! > + umount /mnt > + > +- get ublk dev info: > + ublk list > + > +- delete ublk device > + ublk del -a > + ublk del -n $ublk_dev_id > + > +Design > +====== > + > +Control plane > +------------- > + > +ublk driver provides global misc device node(``/dev/ublk-control``) for > +managing and controlling ublk devices with help of several control commands: > + > +- UBLK_CMD_ADD_DEV > + Add one ublk char device(``/dev/ublkc*``) which is talked with ublksrv wrt. > + IO command communication. Basic device info is sent together with this > + command, see UAPI structure of ublksrv_ctrl_dev_info, such as nr_hw_queues, > + queue_depth, and max IO request buffer size, which info is negotiated with > + ublk driver and sent back to ublksrv. After this command is completed, the > + basic device info can't be changed any more. > + > +- UBLK_CMD_SET_PARAMS / UBLK_CMD_GET_PARAMS > + Set or get ublk device's parameters, which can be generic feature related, > + or request queue limit related, but can't be IO logic specific, cause ublk > + driver does not handle any IO logic. This command has to be sent before > + sending UBLK_CMD_START_DEV. > + > +- UBLK_CMD_START_DEV > + After ublksrv prepares userspace resource such as, creating per-queue > + pthread & io_ruing for handling ublk IO, this command is set for ublk > + driver to allocate & expose /dev/ublkb*. Parameters set via > + UBLK_CMD_SET_PARAMS are applied for creating /dev/ublkb*. > + > +- UBLK_CMD_STOP_DEV > + Quiesce IO on /dev/ublkb* and delete the disk. After this command returns, > + ublksrv can release resource, such as destroy per-queue pthread & io_uring > + for handling io command. > + > +- UBLK_CMD_DEL_DEV > + Delete /dev/ublkc*. After this command returns, the allocated ublk device > + number can be reused. > + > +- UBLK_CMD_GET_QUEUE_AFFINITY > + After /dev/ublkc is added, ublk driver creates block layer tagset, so each > + queue's affinity info is available, ublksrv sends UBLK_CMD_GET_QUEUE_AFFINITY > + to retrieve queue affinity info, so ublksrv can setup the per-queue context > + efficiently, such as bind affine CPUs with IO pthread, and try to allocate > + buffers in IO thread context. > + > +- UBLK_CMD_GET_DEV_INFO > + For retrieve device info of ublksrv_ctrl_dev_info. And it is ublksrv's > + responsibility to save IO target specific info in userspace. > + > +Data plane > +---------- > + > +ublksrv needs to create per-queue IO pthread & io_uring for handling IO > +command (io_uring passthrough command), and the per-queue IO pthread > +focuses on IO handling and shouldn't handle any control & management > +task. > + > +ublksrv's IO is assigned by one unique tag, which is 1:1 mapping with IO > +request of /dev/ublkb*. > + > +UAPI structure of ublksrv_io_desc is defined for describing each IO from > +ublk driver. One fixed mmaped area(array) on /dev/ublkc* is provided for > +exporting IO info to ublksrv, such as IO offset, length, OP/flags and > +buffer address. Each ublksrv_io_desc instance can be indexed via queue id > +and IO tag directly. > + > +Following IO commands are communicated via io_uring passthrough command, > +and each command is only for forwarding ublk IO and committing IO result > +with specified IO tag in the command data: > + > +- UBLK_IO_FETCH_REQ > + Sent from ublksrv IO pthread for fetching future coming IO request > + issued to /dev/ublkb*. This command is just sent once from ublksrv IO > + pthread for ublk driver to setup IO forward environment. > + > +- UBLK_IO_COMMIT_AND_FETCH_REQ > + After one IO request is issued to /dev/ublkb*, ublk driver stores this > + IO's ublksrv_io_desc to the specified mapped area, then the previous > + received IO command of this IO tag, either UBLK_IO_FETCH_REQ or > + UBLK_IO_COMMIT_AND_FETCH_REQ, is completed, so ulksrv gets the IO > + notification via io_uring. > + > + After ublksrv handles this IO, this IO's result is committed back to ublk > + driver by sending UBLK_IO_COMMIT_AND_FETCH_REQ back. Once ublkdrv received > + this command, it parses the IO result and complete the IO request to > + /dev/ublkb*. Meantime setup environment for fetching future IO request > + with this IO tag. So UBLK_IO_COMMIT_AND_FETCH_REQ is reused for both > + fetching request and committing back IO result. > + > +- UBLK_IO_NEED_GET_DATA > + ublksrv pre-allocates IO buffer for each IO at default, any new project > + should use this IO buffer to communicate with ublk driver. But existed > + project may not work or be changed to in this way, so add this command > + to provide chance for userspace to use its existed buffer for handling > + IO. > + > +- data copy between ublkserv IO buffer and ublk block IO request > + ublk driver needs to copy ublk block IO request pages into ublksrv buffer > + (pages) first for WRITE before notifying ublksrv of the coming IO, so > + ublksrv can hanldle WRITE request. > + > + After ublksrv handles READ request and sends UBLK_IO_COMMIT_AND_FETCH_REQ > + to ublksrv, ublkdrv needs to copy read ublksrv buffer(pages) to the ublk > + IO request pages. > + > +Future development > +================== > + > +Container-ware ublk deivice > +--------------------------- > + > +ublk driver doesn't handle any IO logic, and its function is well defined > +so far, and very limited userspace interfaces are needed, and each one is > +well defined too, then it is very likely to make ublk device one > +container-ware block device in future, as Stefan Hajnoczi suggested[3], by > +removing ADMIN privilege. > + > +Zero copy > +--------- > + > +Wrt. zero copy support, it is one generic requirement for nbd, fuse or > +similar drivers, one problem Xiaoguang mentioned is that pages mapped to > +userspace can't be remapped any more in kernel with existed mm interfaces, > +and it can be involved when submitting direct IO to /dev/ublkb*. Also > +Xiaoguang reported that big request may benefit from zero copy a lot, > +such as >= 256KB IO. > + At what thread on lore the problem is mentioned? > + > +References > +========== > + > +[1] https://github.com/ming1/ubdsrv > + > +[2] https://github.com/ming1/ubdsrv/blob/master/README > + > +[3] https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/ The documentation can be improved: - Its slug should be added to table of contents (index.rst) - Use footnote syntax for external references - The wordings are weird, almost to the point that I can't comprehend it (barely understand the meaning). Here's the improv: ---- >8 ---- diff --git a/Documentation/block/index.rst b/Documentation/block/index.rst index 68f115f2b1c6bf..c4c73db748a81f 100644 --- a/Documentation/block/index.rst +++ b/Documentation/block/index.rst @@ -23,3 +23,4 @@ Block stat switching-sched writeback_cache_control + ublk diff --git a/Documentation/block/ublk.rst b/Documentation/block/ublk.rst index 9e8f7ba518a3bb..7262c8b198074b 100644 --- a/Documentation/block/ublk.rst +++ b/Documentation/block/ublk.rst @@ -14,56 +14,63 @@ virtual block device, such as ublk-qcow2, and there was several attempts of implementing qcow2 driver in kernel. ublk block device(``/dev/ublkb*``) is added by ublk driver. Any IO request -submitted to ublk device will be forwarded to ublk's userspace part( -ublksrv [1]), and after the IO is handled by ublksrv, the result is -committed back to ublk driver, then ublk IO request can be completed. With -this way, any specific IO handling logic is totally done inside ublksrv, -and ublk driver doe _not_ handle any device specific IO logic, such as -loop's IO handling, NBD's IO communication, or qcow2's IO mapping, ... +on the device will be forwarded to the userspace program ``ublksrv`` +[#userspace]_. +After the IO is handled by userspace, the result is committed back to the +driver, thus completing the request cycle. This way, any specific IO handling +logic is totally done by userspace, such as +loop's IO handling, NBD's IO communication, or qcow2's IO mapping. -/dev/ublkbN is driven by blk-mq request based driver, each request is +``/dev/ublkbN`` is driven by blk-mq request-based driver. Each request is assigned by one queue wide unique tag. ublksrv assigns unique tag to each -IO too, which is 1:1 mapped with IO of /dev/ublkb*. +IO too, which is 1:1 mapped with IO of ``/dev/ublkb*``. Both the IO request forward and IO handling result committing are done via -io_uring passthrough command, that is why ublk is also one io_uring based -block driver. It has been observed that io_uring passthrough command can get -better IOPS than block IO. So ublk is one high performance implementation -of userspace block device. Not only IO request communication is done by -io_uring, but also the preferred IO handling in ublksrv is io_uring based -approach too. +``io_uring`` passthrough command; that is why ublk is also one io_uring based +block driver. It has been observed that using io_uring passthrough command can +give better IOPS than block IO; which is why ublk is one of high performance +implementation of userspace block device: not only IO request communication is +done by io_uring, but also the preferred IO handling in ublksrv is io_uring +based approach too. -ublk provides control interface to set/get ublk block device parameters, and -the interface is extendable and kabi compatible, so basically any ublk request +ublk provides control interface to set/get ublk block device parameters. +The interface is extendable and kabi compatible: basically any ublk request queue's parameter or ublk generic feature parameters can be set/get via this -extendable interface. So ublk is generic userspace block device framework, such -as, it is easy to setup one ublk device with specified block parameters from -userspace. +extendable interface. Thus ublk is generic userspace block device framework. +For example, it is easy to setup one ublk device with specified block +parameters from userspace. -How to use ublk -=============== +Using ublk +========== -After building ublksrv[1], ublk block device(``/dev/ublkb*``) can be added -and deleted by the utility, then existed block IO applications can talk with -it. +ublk requires ublksrv [#userspace]_ to be installed. The userspace program manages ublk +block devices (``/dev/ublkb*``), while block IO applications can talk with +them. -See usage details in README[2] of ublksrv, for example of ublk-loop: +See usage details in README of ublksrv [#userspace_readme]_. -- add ublk device: - ublk add -t loop -f ublk-loop.img +Below is example of using ublk as loop device. -- use it: - mkfs.xfs /dev/ublkb0 - mount /dev/ublkb0 /mnt - .... # all IOs are handled by io_uring!!! - umount /mnt +- add ublk device:: -- get ublk dev info: - ublk list + ublk add -t loop -f ublk-loop.img -- delete ublk device - ublk del -a - ublk del -n $ublk_dev_id +- format with xfs, then use it:: + + mkfs.xfs /dev/ublkb0 + mount /dev/ublkb0 /mnt + # do anything. all IOs are handled by io_uring + ... + umount /mnt + +- get ublk dev info:: + + ublk list + +- delete ublk device:: + + ublk del -a + ublk del -n $ublk_dev_id Design ====== @@ -74,130 +81,143 @@ Control plane ublk driver provides global misc device node(``/dev/ublk-control``) for managing and controlling ublk devices with help of several control commands: -- UBLK_CMD_ADD_DEV - Add one ublk char device(``/dev/ublkc*``) which is talked with ublksrv wrt. - IO command communication. Basic device info is sent together with this - command, see UAPI structure of ublksrv_ctrl_dev_info, such as nr_hw_queues, - queue_depth, and max IO request buffer size, which info is negotiated with - ublk driver and sent back to ublksrv. After this command is completed, the - basic device info can't be changed any more. +- ``UBLK_CMD_ADD_DEV`` -- UBLK_CMD_SET_PARAMS / UBLK_CMD_GET_PARAMS - Set or get ublk device's parameters, which can be generic feature related, - or request queue limit related, but can't be IO logic specific, cause ublk - driver does not handle any IO logic. This command has to be sent before - sending UBLK_CMD_START_DEV. + Add one ublk char device(``/dev/ublkc*``) which is talked with ublksrv + WRT IO command communication. Basic device info is sent together with this + command. It sets UAPI structure of ``ublksrv_ctrl_dev_info``, + such as ``nr_hw_queues``, ``queue_depth``, and max IO request buffer size, + for which the info is negotiated with ublk driver and sent back to ublksrv. + After this command is completed, the basic device info is immutable. -- UBLK_CMD_START_DEV - After ublksrv prepares userspace resource such as, creating per-queue - pthread & io_ruing for handling ublk IO, this command is set for ublk - driver to allocate & expose /dev/ublkb*. Parameters set via - UBLK_CMD_SET_PARAMS are applied for creating /dev/ublkb*. +- ``UBLK_CMD_SET_PARAMS`` / ``UBLK_CMD_GET_PARAMS`` -- UBLK_CMD_STOP_DEV - Quiesce IO on /dev/ublkb* and delete the disk. After this command returns, - ublksrv can release resource, such as destroy per-queue pthread & io_uring - for handling io command. + Set or get ublk device's parameters, which can be either generic feature + related, or request queue limit related, but can't be IO logic specific, + because ublk driver does not handle any IO logic. This command has to be + sent before sending ``UBLK_CMD_START_DEV``. -- UBLK_CMD_DEL_DEV - Delete /dev/ublkc*. After this command returns, the allocated ublk device +- ``UBLK_CMD_START_DEV`` + + After ublksrv prepares userspace resources (such as creating per-queue + pthread & io_uring for handling ublk IO), this command is set for ublk + driver to allocate & expose ``/dev/ublkb*``. Parameters set via + ``UBLK_CMD_SET_PARAMS`` are applied for creating the device. + +- ``UBLK_CMD_STOP_DEV`` + + Halt IO on ``/dev/ublkb*`` and remove the device. When this command returns, + ublksrv will release resources (such as destroying per-queue pthread & + io_uring). + +- ``UBLK_CMD_DEL_DEV`` + + Remove ``/dev/ublkc*``. When this command returns, the allocated ublk device number can be reused. -- UBLK_CMD_GET_QUEUE_AFFINITY - After /dev/ublkc is added, ublk driver creates block layer tagset, so each - queue's affinity info is available, ublksrv sends UBLK_CMD_GET_QUEUE_AFFINITY - to retrieve queue affinity info, so ublksrv can setup the per-queue context - efficiently, such as bind affine CPUs with IO pthread, and try to allocate +- ``UBLK_CMD_GET_QUEUE_AFFINITY`` + + When ``/dev/ublkc`` is added, ublk driver creates block layer tagset, so + that each + queue's affinity info is available. ublksrv sends + ``UBLK_CMD_GET_QUEUE_AFFINITY`` + to retrieve queue affinity info. It can setup the per-queue context + efficiently, such as bind affine CPUs with IO pthread and try to allocate buffers in IO thread context. -- UBLK_CMD_GET_DEV_INFO - For retrieve device info of ublksrv_ctrl_dev_info. And it is ublksrv's +- ``UBLK_CMD_GET_DEV_INFO`` + + For retrieving device info via ``ublksrv_ctrl_dev_info``. It is ublksrv's responsibility to save IO target specific info in userspace. Data plane ---------- ublksrv needs to create per-queue IO pthread & io_uring for handling IO -command (io_uring passthrough command), and the per-queue IO pthread +commands via io_uring passthrough. The per-queue IO pthread focuses on IO handling and shouldn't handle any control & management -task. +tasks. -ublksrv's IO is assigned by one unique tag, which is 1:1 mapping with IO -request of /dev/ublkb*. +ublksrv's IO is assigned by a unique tag, which is 1:1 mapping with IO +request of ``/dev/ublkb*``. -UAPI structure of ublksrv_io_desc is defined for describing each IO from -ublk driver. One fixed mmaped area(array) on /dev/ublkc* is provided for -exporting IO info to ublksrv, such as IO offset, length, OP/flags and -buffer address. Each ublksrv_io_desc instance can be indexed via queue id +UAPI structure of ``ublksrv_io_desc`` is defined for describing each IO from +ublk driver. A fixed mmaped area (array) on ``/dev/ublkc*`` is provided for +exporting IO info to ublksrv; such as IO offset, length, OP/flags and +buffer address. Each ``ublksrv_io_desc`` instance can be indexed via queue id and IO tag directly. -Following IO commands are communicated via io_uring passthrough command, +The following IO commands are communicated via io_uring passthrough command, and each command is only for forwarding ublk IO and committing IO result with specified IO tag in the command data: -- UBLK_IO_FETCH_REQ - Sent from ublksrv IO pthread for fetching future coming IO request - issued to /dev/ublkb*. This command is just sent once from ublksrv IO +- ``UBLK_IO_FETCH_REQ`` + + Sent from ublksrv IO pthread for fetching future incoming IO requests + destined to ``/dev/ublkb*``. This command is sent only once from ublksrv IO pthread for ublk driver to setup IO forward environment. -- UBLK_IO_COMMIT_AND_FETCH_REQ - After one IO request is issued to /dev/ublkb*, ublk driver stores this - IO's ublksrv_io_desc to the specified mapped area, then the previous - received IO command of this IO tag, either UBLK_IO_FETCH_REQ or - UBLK_IO_COMMIT_AND_FETCH_REQ, is completed, so ulksrv gets the IO +- ``UBLK_IO_COMMIT_AND_FETCH_REQ`` + + When an IO request is destined to ``/dev/ublkb*``, ublk driver stores + the IO's ``ublksrv_io_desc`` to the specified mapped area; then the + previous received IO command of this IO tag (either UBLK_IO_FETCH_REQ or + UBLK_IO_COMMIT_AND_FETCH_REQ) is completed, so ublksrv gets the IO notification via io_uring. - After ublksrv handles this IO, this IO's result is committed back to ublk - driver by sending UBLK_IO_COMMIT_AND_FETCH_REQ back. Once ublkdrv received - this command, it parses the IO result and complete the IO request to - /dev/ublkb*. Meantime setup environment for fetching future IO request - with this IO tag. So UBLK_IO_COMMIT_AND_FETCH_REQ is reused for both - fetching request and committing back IO result. + After ublksrv handles the IO, its result is committed back to ublk + driver by sending ``UBLK_IO_COMMIT_AND_FETCH_REQ`` back. Once ublkdrv + received this command, it parses the result and complete the request to + ``/dev/ublkb*``. In the meantime setup environment for fetching future + requests with the same IO tag. That is, ``UBLK_IO_COMMIT_AND_FETCH_REQ`` + is reused for both fetching request and committing back IO result. -- UBLK_IO_NEED_GET_DATA - ublksrv pre-allocates IO buffer for each IO at default, any new project - should use this IO buffer to communicate with ublk driver. But existed - project may not work or be changed to in this way, so add this command - to provide chance for userspace to use its existed buffer for handling - IO. +- ``UBLK_IO_NEED_GET_DATA`` + + ublksrv pre-allocates IO buffer for each IO by default. Any new projects + should use this buffer to communicate with ublk driver. However, + existing + projects may break or not able to consume the new buffer interface; that's + why this command is added for backwards compatibility so that existing + projects can still consume existing buffers. - data copy between ublkserv IO buffer and ublk block IO request - ublk driver needs to copy ublk block IO request pages into ublksrv buffer - (pages) first for WRITE before notifying ublksrv of the coming IO, so - ublksrv can hanldle WRITE request. - After ublksrv handles READ request and sends UBLK_IO_COMMIT_AND_FETCH_REQ - to ublksrv, ublkdrv needs to copy read ublksrv buffer(pages) to the ublk - IO request pages. + ublk driver needs to copy the block IO request pages into ublksrv buffer + (pages) first for WRITE before notifying ublksrv of the coming IO, so + that ublksrv can hanldle WRITE request. + + When ublksrv handles READ request and sends ``UBLK_IO_COMMIT_AND_FETCH_REQ`` + to ublksrv, ublkdrv needs to copy read ublksrv buffer (pages) to the IO + request pages. Future development ================== -Container-ware ublk deivice ---------------------------- +Container-aware ublk deivice +---------------------------- -ublk driver doesn't handle any IO logic, and its function is well defined -so far, and very limited userspace interfaces are needed, and each one is -well defined too, then it is very likely to make ublk device one -container-ware block device in future, as Stefan Hajnoczi suggested[3], by -removing ADMIN privilege. +ublk driver doesn't handle any IO logic. Its function is well defined +for now, and very limited userspace interfaces are needed, which is also +well defined too. It is possible to make ublk devices container-aware block +devices in future as Stefan Hajnoczi suggested [#stefan]_, by removing +ADMIN privilege. Zero copy --------- -Wrt. zero copy support, it is one generic requirement for nbd, fuse or -similar drivers, one problem Xiaoguang mentioned is that pages mapped to -userspace can't be remapped any more in kernel with existed mm interfaces, -and it can be involved when submitting direct IO to /dev/ublkb*. Also -Xiaoguang reported that big request may benefit from zero copy a lot, -such as >= 256KB IO. +Wrt. zero copy support, which is a generic requirement for nbd, fuse or +similar drivers, a problem Xiaoguang mentioned is that pages mapped to +userspace can't be remapped any more in kernel with existing mm interfaces. +This can occurs when destining direct IO to ``/dev/ublkb*``. Also +he reported that big requests (>= 256 KB IO) may benefit a lot from zero copy. References ========== -[1] https://github.com/ming1/ubdsrv +.. [#userspace] https://github.com/ming1/ubdsrv -[2] https://github.com/ming1/ubdsrv/blob/master/README +.. [#userspace_readme] https://github.com/ming1/ubdsrv/blob/master/README -[3] https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/ +.. [#stefan] https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/ Thanks. -- An old man doll... just what I always wanted! - Clara
Attachment:
signature.asc
Description: PGP signature
