rust: block: introduce kernel::block::mq module

Add initial abstractions for working with blk-mq.

This patch is a maintained, refactored subset of code originally published
by Wedson Almeida Filho <[email protected]> [1].

[1] https://github.com/wedsonaf/linux/tree/f2cfd2fe0e2ca4e90994f96afe268bbd4382a891/rust/kernel/blk/mq.rs

Cc: Wedson Almeida Filho <[email protected]>
Signed-off-by: Andreas Hindborg <[email protected]>
Reviewed-by: Benno Lossin <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jens Axboe <[email protected]>

Author: metaspace

rust/bindings/bindings_helper.h

@@ -7,6 +7,8 @@
  */
 
 #include <kunit/test.h>
+#include <linux/blk_types.h>
+#include <linux/blk-mq.h>
 #include <linux/errname.h>
 #include <linux/ethtool.h>
 #include <linux/jiffies.h>
@@ -20,6 +22,7 @@
 
 /* `bindgen` gets confused at certain things. */
 const size_t RUST_CONST_HELPER_ARCH_SLAB_MINALIGN = ARCH_SLAB_MINALIGN;
+const size_t RUST_CONST_HELPER_PAGE_SIZE = PAGE_SIZE;
 const gfp_t RUST_CONST_HELPER_GFP_ATOMIC = GFP_ATOMIC;
 const gfp_t RUST_CONST_HELPER_GFP_KERNEL = GFP_KERNEL;
 const gfp_t RUST_CONST_HELPER_GFP_KERNEL_ACCOUNT = GFP_KERNEL_ACCOUNT;

rust/helpers.c

@@ -186,3 +186,19 @@ static_assert(
 	__alignof__(size_t) == __alignof__(uintptr_t),
 	"Rust code expects C `size_t` to match Rust `usize`"
 );
+
+// This will soon be moved to a separate file, so no need to merge with above.
+#include <linux/blk-mq.h>
+#include <linux/blkdev.h>
+
+void *rust_helper_blk_mq_rq_to_pdu(struct request *rq)
+{
+	return blk_mq_rq_to_pdu(rq);
+}
+EXPORT_SYMBOL_GPL(rust_helper_blk_mq_rq_to_pdu);
+
+struct request *rust_helper_blk_mq_rq_from_pdu(void *pdu)
+{
+	return blk_mq_rq_from_pdu(pdu);
+}
+EXPORT_SYMBOL_GPL(rust_helper_blk_mq_rq_from_pdu);

rust/kernel/block.rs

@@ -0,0 +1,5 @@
+// SPDX-License-Identifier: GPL-2.0
+
+//! Types for working with the block layer.
+
+pub mod mq;

rust/kernel/block/mq.rs

@@ -0,0 +1,98 @@
+// SPDX-License-Identifier: GPL-2.0
+
+//! This module provides types for implementing block drivers that interface the
+//! blk-mq subsystem.
+//!
+//! To implement a block device driver, a Rust module must do the following:
+//!
+//! - Implement [`Operations`] for a type `T`.
+//! - Create a [`TagSet<T>`].
+//! - Create a [`GenDisk<T>`], via the [`GenDiskBuilder`].
+//! - Add the disk to the system by calling [`GenDiskBuilder::build`] passing in
+//!   the `TagSet` reference.
+//!
+//! The types available in this module that have direct C counterparts are:
+//!
+//! - The [`TagSet`] type that abstracts the C type `struct tag_set`.
+//! - The [`GenDisk`] type that abstracts the C type `struct gendisk`.
+//! - The [`Request`] type that abstracts the C type `struct request`.
+//!
+//! The kernel will interface with the block device driver by calling the method
+//! implementations of the `Operations` trait.
+//!
+//! IO requests are passed to the driver as [`kernel::types::ARef<Request>`]
+//! instances. The `Request` type is a wrapper around the C `struct request`.
+//! The driver must mark end of processing by calling one of the
+//! `Request::end`, methods. Failure to do so can lead to deadlock or timeout
+//! errors. Please note that the C function `blk_mq_start_request` is implicitly
+//! called when the request is queued with the driver.
+//!
+//! The `TagSet` is responsible for creating and maintaining a mapping between
+//! `Request`s and integer ids as well as carrying a pointer to the vtable
+//! generated by `Operations`. This mapping is useful for associating
+//! completions from hardware with the correct `Request` instance. The `TagSet`
+//! determines the maximum queue depth by setting the number of `Request`
+//! instances available to the driver, and it determines the number of queues to
+//! instantiate for the driver. If possible, a driver should allocate one queue
+//! per core, to keep queue data local to a core.
+//!
+//! One `TagSet` instance can be shared between multiple `GenDisk` instances.
+//! This can be useful when implementing drivers where one piece of hardware
+//! with one set of IO resources are represented to the user as multiple disks.
+//!
+//! One significant difference between block device drivers implemented with
+//! these Rust abstractions and drivers implemented in C, is that the Rust
+//! drivers have to own a reference count on the `Request` type when the IO is
+//! in flight. This is to ensure that the C `struct request` instances backing
+//! the Rust `Request` instances are live while the Rust driver holds a
+//! reference to the `Request`. In addition, the conversion of an integer tag to
+//! a `Request` via the `TagSet` would not be sound without this bookkeeping.
+//!
+//! [`GenDisk`]: gen_disk::GenDisk
+//! [`GenDisk<T>`]: gen_disk::GenDisk
+//! [`GenDiskBuilder`]: gen_disk::GenDiskBuilder
+//! [`GenDiskBuilder::build`]: gen_disk::GenDiskBuilder::build
+//!
+//! # Example
+//!
+//! ```rust
+//! use kernel::{
+//!     alloc::flags,
+//!     block::mq::*,
+//!     new_mutex,
+//!     prelude::*,
+//!     sync::{Arc, Mutex},
+//!     types::{ARef, ForeignOwnable},
+//! };
+//!
+//! struct MyBlkDevice;
+//!
+//! #[vtable]
+//! impl Operations for MyBlkDevice {
+//!
+//!     fn queue_rq(rq: ARef<Request<Self>>, _is_last: bool) -> Result {
+//!         Request::end_ok(rq);
+//!         Ok(())
+//!     }
+//!
+//!     fn commit_rqs() {}
+//! }
+//!
+//! let tagset: Arc<TagSet<MyBlkDevice>> =
+//!     Arc::pin_init(TagSet::new(1, 256, 1), flags::GFP_KERNEL)?;
+//! let mut disk = gen_disk::GenDiskBuilder::new()
+//!     .capacity_sectors(4096)
+//!     .build(format_args!("myblk"), tagset)?;
+//!
+//! # Ok::<(), kernel::error::Error>(())
+//! ```
+
+pub mod gen_disk;
+mod operations;
+mod raw_writer;
+mod request;
+mod tag_set;
+
+pub use operations::Operations;
+pub use request::Request;
+pub use tag_set::TagSet;

rust/kernel/block/mq/gen_disk.rs

@@ -0,0 +1,215 @@
+// SPDX-License-Identifier: GPL-2.0
+
+//! Generic disk abstraction.
+//!
+//! C header: [`include/linux/blkdev.h`](srctree/include/linux/blkdev.h)
+//! C header: [`include/linux/blk_mq.h`](srctree/include/linux/blk_mq.h)
+
+use crate::block::mq::{raw_writer::RawWriter, Operations, TagSet};
+use crate::error;
+use crate::{bindings, error::from_err_ptr, error::Result, sync::Arc};
+use core::fmt::{self, Write};
+
+/// A builder for [`GenDisk`].
+///
+/// Use this struct to configure and add new [`GenDisk`] to the VFS.
+pub struct GenDiskBuilder {
+    rotational: bool,
+    logical_block_size: u32,
+    physical_block_size: u32,
+    capacity_sectors: u64,
+}
+
+impl Default for GenDiskBuilder {
+    fn default() -> Self {
+        Self {
+            rotational: false,
+            logical_block_size: bindings::PAGE_SIZE as u32,
+            physical_block_size: bindings::PAGE_SIZE as u32,
+            capacity_sectors: 0,
+        }
+    }
+}
+
+impl GenDiskBuilder {
+    /// Create a new instance.
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    /// Set the rotational media attribute for the device to be built.
+    pub fn rotational(mut self, rotational: bool) -> Self {
+        self.rotational = rotational;
+        self
+    }
+
+    /// Validate block size by verifying that it is between 512 and `PAGE_SIZE`,
+    /// and that it is a power of two.
+    fn validate_block_size(size: u32) -> Result<()> {
+        if !(512..=bindings::PAGE_SIZE as u32).contains(&size) || !size.is_power_of_two() {
+            Err(error::code::EINVAL)
+        } else {
+            Ok(())
+        }
+    }
+
+    /// Set the logical block size of the device to be built.
+    ///
+    /// This method will check that block size is a power of two and between 512
+    /// and 4096. If not, an error is returned and the block size is not set.
+    ///
+    /// This is the smallest unit the storage device can address. It is
+    /// typically 4096 bytes.
+    pub fn logical_block_size(mut self, block_size: u32) -> Result<Self> {
+        Self::validate_block_size(block_size)?;
+        self.logical_block_size = block_size;
+        Ok(self)
+    }
+
+    /// Set the physical block size of the device to be built.
+    ///
+    /// This method will check that block size is a power of two and between 512
+    /// and 4096. If not, an error is returned and the block size is not set.
+    ///
+    /// This is the smallest unit a physical storage device can write
+    /// atomically. It is usually the same as the logical block size but may be
+    /// bigger. One example is SATA drives with 4096 byte physical block size
+    /// that expose a 512 byte logical block size to the operating system.
+    pub fn physical_block_size(mut self, block_size: u32) -> Result<Self> {
+        Self::validate_block_size(block_size)?;
+        self.physical_block_size = block_size;
+        Ok(self)
+    }
+
+    /// Set the capacity of the device to be built, in sectors (512 bytes).
+    pub fn capacity_sectors(mut self, capacity: u64) -> Self {
+        self.capacity_sectors = capacity;
+        self
+    }
+
+    /// Build a new `GenDisk` and add it to the VFS.
+    pub fn build<T: Operations>(
+        self,
+        name: fmt::Arguments<'_>,
+        tagset: Arc<TagSet<T>>,
+    ) -> Result<GenDisk<T>> {
+        let lock_class_key = crate::sync::LockClassKey::new();
+
+        // SAFETY: `tagset.raw_tag_set()` points to a valid and initialized tag set
+        let gendisk = from_err_ptr(unsafe {
+            bindings::__blk_mq_alloc_disk(
+                tagset.raw_tag_set(),
+                core::ptr::null_mut(), // TODO: We can pass queue limits right here
+                core::ptr::null_mut(),
+                lock_class_key.as_ptr(),
+            )
+        })?;
+
+        const TABLE: bindings::block_device_operations = bindings::block_device_operations {
+            submit_bio: None,
+            open: None,
+            release: None,
+            ioctl: None,
+            compat_ioctl: None,
+            check_events: None,
+            unlock_native_capacity: None,
+            getgeo: None,
+            set_read_only: None,
+            swap_slot_free_notify: None,
+            report_zones: None,
+            devnode: None,
+            alternative_gpt_sector: None,
+            get_unique_id: None,
+            // TODO: Set to THIS_MODULE. Waiting for const_refs_to_static feature to
+            // be merged (unstable in rustc 1.78 which is staged for linux 6.10)
+            // https://github.com/rust-lang/rust/issues/119618
+            owner: core::ptr::null_mut(),
+            pr_ops: core::ptr::null_mut(),
+            free_disk: None,
+            poll_bio: None,
+        };
+
+        // SAFETY: `gendisk` is a valid pointer as we initialized it above
+        unsafe { (*gendisk).fops = &TABLE };
+
+        let mut raw_writer = RawWriter::from_array(
+            // SAFETY: `gendisk` points to a valid and initialized instance. We
+            // have exclusive access, since the disk is not added to the VFS
+            // yet.
+            unsafe { &mut (*gendisk).disk_name },
+        )?;
+        raw_writer.write_fmt(name)?;
+        raw_writer.write_char('\0')?;
+
+        // SAFETY: `gendisk` points to a valid and initialized instance of
+        // `struct gendisk`. We have exclusive access, so we cannot race.
+        unsafe {
+            bindings::blk_queue_logical_block_size((*gendisk).queue, self.logical_block_size)
+        };
+
+        // SAFETY: `gendisk` points to a valid and initialized instance of
+        // `struct gendisk`. We have exclusive access, so we cannot race.
+        unsafe {
+            bindings::blk_queue_physical_block_size((*gendisk).queue, self.physical_block_size)
+        };
+
+        // SAFETY: `gendisk` points to a valid and initialized instance of
+        // `struct gendisk`. `set_capacity` takes a lock to synchronize this
+        // operation, so we will not race.
+        unsafe { bindings::set_capacity(gendisk, self.capacity_sectors) };
+
+        if !self.rotational {
+            // SAFETY: `gendisk` points to a valid and initialized instance of
+            // `struct gendisk`. This operation uses a relaxed atomic bit flip
+            // operation, so there is no race on this field.
+            unsafe { bindings::blk_queue_flag_set(bindings::QUEUE_FLAG_NONROT, (*gendisk).queue) };
+        } else {
+            // SAFETY: `gendisk` points to a valid and initialized instance of
+            // `struct gendisk`. This operation uses a relaxed atomic bit flip
+            // operation, so there is no race on this field.
+            unsafe {
+                bindings::blk_queue_flag_clear(bindings::QUEUE_FLAG_NONROT, (*gendisk).queue)
+            };
+        }
+
+        crate::error::to_result(
+            // SAFETY: `gendisk` points to a valid and initialized instance of
+            // `struct gendisk`.
+            unsafe {
+                bindings::device_add_disk(core::ptr::null_mut(), gendisk, core::ptr::null_mut())
+            },
+        )?;
+
+        // INVARIANT: `gendisk` was initialized above.
+        // INVARIANT: `gendisk` was added to the VFS via `device_add_disk` above.
+        Ok(GenDisk {
+            _tagset: tagset,
+            gendisk,
+        })
+    }
+}
+
+/// A generic block device.
+///
+/// # Invariants
+///
+///  - `gendisk` must always point to an initialized and valid `struct gendisk`.
+///  - `gendisk` was added to the VFS through a call to
+///     `bindings::device_add_disk`.
+pub struct GenDisk<T: Operations> {
+    _tagset: Arc<TagSet<T>>,
+    gendisk: *mut bindings::gendisk,
+}
+
+// SAFETY: `GenDisk` is an owned pointer to a `struct gendisk` and an `Arc` to a
+// `TagSet` It is safe to send this to other threads as long as T is Send.
+unsafe impl<T: Operations + Send> Send for GenDisk<T> {}
+
+impl<T: Operations> Drop for GenDisk<T> {
+    fn drop(&mut self) {
+        // SAFETY: By type invariant, `self.gendisk` points to a valid and
+        // initialized instance of `struct gendisk`, and it was previously added
+        // to the VFS.
+        unsafe { bindings::del_gendisk(self.gendisk) };
+    }
+}