This is a follow-up of [0].
# Advantages
- This saves dozens of unneeded clone()s across the whole code base
- Makes it much easier to reason about how parameters are used
(often we passed owned Arc/Rc versions without actually needing
ownership)
# Exceptions
For certain code paths, the alternatives would require awkward or overly
complex code, and in some cases the functions are the logical owners of
the values they take. In those cases, I've added
#[allow(clippy::needless_pass_by_value)].
This does not mean that one should not improve this in the future.
[0] 6a86c157af
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
On-behalf-of: SAP philipp.schuster@sap.com
This helps to uncover expensive and needless clones in the code base.
For example, I prevented extensive clones in the snapshot path where
(nested) BTreeMap's have been cloned over and over again. Further,
the lint helps devs to much better reason about the ownership of
parameters.
All of these changes have been done manually with the necessary
caution. A few structs that are cheap to clone are now `copy` so that
this lint won't trigger for them.
I didn't enable the lint so far as it is a massive rabbit hole and
needs much more fixes. Nevertheless, it is very useful.
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
On-behalf-of: SAP philipp.schuster@sap.com
The granularity has significant implications in typical cloud
deployments with network storage. The Linux kernel will sync advisory
locks to network file systems, but these backends may have different
policies and handle locks differently. For example, Netapp speaks a NFS
API but will treat advisory OFD locks for the whole file as mandatory
locks, whereas byte-range locks for the whole file will remain
advisory [0].
As it is a valid use case to prevent multiple CHV instances from
accessing the same disk but disk management software (e.g., Cinder in
OpenStack) should be able to snapshot disks while VMs are running, we
need special control over the lock granularity. Therefore, it is a valid
use case to lock the whole byte range of a disk image without
technically locking the whole file - to get the best of both worlds.
This also brings CHVs behavior in line with QEMU [1].
Whole-file locks remain a valid use case and could be supported later.
This patch only provides the necessary groundwork; making it
configurable is out of scope for now.
[0] https://kb.netapp.com/on-prem/ontap/da/NAS/NAS-KBs/How_is_Mandatory_Locking_supported_for_NFSv4_on_ONTAP_9
[1] <qemu>/util/osdep.c::qemu_lock_fcntl()
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
On-behalf-of: SAP philipp.schuster@sap.com
Consuming `&Arc<T>` as argument is almost always an antipattern as it
hides whether the callee is going to take over (shared) ownership
(by .clone()) or not. Instead, it is better to consume `&dyn T` or
`Arc<dyn T>` to be more explicit. This commit cleans up the code.
The change is very mechanic and was very easy to implement across the
code base.
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
On-behalf-of: SAP philipp.schuster@sap.com
Fix clippy warning `uninlined_format_args` reported by rustc rustc
1.89.0 (29483883e 2025-08-04).
```console
warning: variables can be used directly in the `format!` string
--> block/src/lib.rs:649:17
|
649 | info!("{} failed to create io_uring instance: {}", error_msg, e);
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#uninlined_format_args
= note: `#[warn(clippy::uninlined_format_args)]` on by default
help: change this to
|
649 - info!("{} failed to create io_uring instance: {}", error_msg, e);
649 + info!("{error_msg} failed to create io_uring instance: {e}");
|
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Cache and batch IO requests after parsing all
items in the queue, improving performance—especially
for small block sizes—by reducing per-request overhead.
Introduced two methods in the AsyncIo trait for batch
submission, with implementation in the raw disk backend.
This method should be called during/after parsing all block IO requests
in the available queue. If the batch submission is not enabled, by
default it does the old way of submitting requests.
Signed-off-by: Bo Chen <bchen@crusoe.ai>
Signed-off-by: Muminul Islam <muislam@microsoft.com>
Instead of returning boolean return an struct of completion status
so that it can be cached for batch submission.
Signed-off-by: Bo Chen <bchen@crusoe.ai>
Signed-off-by: Muminul Islam <muislam@microsoft.com>
This patch changes the read-only check using acked features bit, which
will help to check more features.
Signed-off-by: Songqian Li <sionli@tencent.com>
Instead of exiting on IO errors, report the errors to the guest with
VIRTIO_BLK_S_IOERR. For example, the guest kernel will log something
similar to this if the nbd behind /dev/vdc is unexpectedly disconnected:
[ 166.033957] I/O error, dev vdc, sector 264 op 0x1:(WRITE) flags 0x9800 phys_seg 1 prio class 2
[ 166.035083] Aborting journal on device vdc-8.
[ 166.037307] Buffer I/O error on dev vdc, logical block 9, lost sync page write
[ 166.038471] JBD2: I/O error when updating journal superblock for vdc-8.
[...]
[ 174.234470] EXT4-fs (vdc): I/O error while writing superblock
In case the rootfs is not located on the affected block device, this
will not crash the guest.
Fixes: #6995
Signed-off-by: Gauthier Jolly <contact@gjolly.fr>
This streamlines the code base to follow best practices for
error handling in Rust: Each error struct implements
std::error::Error (most due via thiserror::Error derive macro)
and sets its source accordingly.
This allows future work that nicely prints the error chains,
for example.
So far, the convention is that each error prints its
sub error as part of its Display::fmt() impl.
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
On-behalf-of: SAP philipp.schuster@sap.com
# What
This commit introduces file-based advisory locking for the files backing
up the block devices by using the fcntl() syscall with OFD locks. The
per-open-file-descriptor (OFD) locks are more robust than traditional
POSIX locks (F_SETLK) as they are not tied to process IDs and avoid
common issues in multithreaded or multi-fd scenarios [1]. Therefore,
we don't use `std::fs::File::try_lock()`, which is backed by F_SETLKW.
The locking mechanism is aware of the `readonly` property and allows
`n` readers or `1` writer (exclusive mode).
As the locks are advisory, multiple cloud-hypervisor processes can
prevent themselves from writing to the same file. However, this is not
a system-wide file-system level locking mechanism preventing to open()
a file.
The introduced new locking mechanism does not cover vhost-user devices.
# Why
To prevent misconfiguration and improve safety, it is good practice to
protect disk image files with a locking mechanism. Experience and common
best practices suggest that advisory locks are preferable over mandatory
locks due to better compatibility and fewer pitfalls (in fs space).
The introduced functionality is aligned with the approach taken by
QEMU [0], and is also recommended in [1].
# Implementation Details
We need to ensure that not only normal operation keeps working but also
state save/resume and live-migration. Especially for live migration,
it is crucial that the sender VMM releases the locks when the VM stops
so the receiver VMM can acquire them right after that.
Therefore, the locking and releasing happen directly on the block
device struct. The device manager knows all block devices and can
forward requests to these types.
Last but not least, this commit uses on explicit lock acquiring
but implicit lock releasing (FD close). It only explicitly releases
the locks where this integrates more smoothly into the existing
code.
# Testing
I tested
- normal operation
- state save/resume,
- device hot plugging,
- and live-migration
with read/shared and write/exclusive locks.
One can use the `fcntl-tool` to test if locks are actually acquired
or released [2].
# Links
[0] 825b96dbce/util/osdep.c (L266)
[1] https://apenwarr.ca/log/20101213
[2] https://crates.io/crates/fcntl-tool
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
On-behalf-of: SAP philipp.schuster@sap.com
Rust has a new way of constructing other error and clippy complains if
we are still using the older way to construct error message. Thus,
migrate to the new approach suggested by the clippy.
Warning from beta compiler:
error: this can be `std::io::Error::other(_)`
--> block/src/vhdx/mod.rs:142:17
|
| / std::io::Error::new(
| | std::io::ErrorKind::Other,
| | format!("Failed to update VHDx header: {e}"),
| | )
| |_________________^
|
= help: for further information visit
https://rust-lang.github.io/rust-clippy/master/index.html#io_other_error
help: use `std::io::Error::other`
std::io::Error::other(
format!("Failed to update VHDx header: {e}"),
Signed-off-by: Jinank Jain <jinankjain@microsoft.com>
This allows the guest to put in more than one segment per request. It
can improve the throughput of the system.
Introduce a new check to make sure the queue size configured by the user
is large enough to hold at least one segment.
Signed-off-by: Wei Liu <liuwe@microsoft.com>
Replace `map_or()` on false condition with `is_some_and` to provide
better readability, as suggestted by v1.84.0-beta.1 `cargo clippy`.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
By introducing `imports_granularity="Module"` format strategy,
effectively groups imports from the same module into one line or block,
improving maintainability and readability.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Historically the Cloud Hypervisor coding style has been to ensure that
all imports are ordered and placed in a single group. Unfortunately
cargo fmt has no support for ensuring that all imports are in a single
group so if whitespace lines were added as part of the import statements
then they would only be odered correctly in the group.
By adopting "group_imports="StdExternalCrate" we can enforce a style
where imports are placed in at most three groups for std, external
crates and the crate itself. Choosing a style enforceable by the tooling
reduces the reviewer burden.
Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
When the rate limit was reached it was possible for the notification to
the guest to be lost since the logic to handle the notification was
tightly coupled with processing the queue. The notification would
eventually be triggered when the rate limit pool was refilled but this
could add significant latency.
Address this by refactoring the code to separate processing queue and
signalling - the processing of the queue is suspended when the rate
limit is reached but the signalling will still be attempted if needed
(i.e. VIRTIO_F_EVENT_IDX is still considered.)
Signed-off-by: wuxinyue <wuxinyue.wxy@antgroup.com>
Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
Support event idx feature for virtio-blk device.
This feature could improve disk IO performance by suppressing
notifications from guest to host and interrupts from host to
guest, which has been already supported in virtio-net and
vhost-user devices.
To achieve this, virtqueue's event-idx-related API is
leveraged for avail_event field update and needs_notification
check.
Fixes: #6580
Signed-off-by: wuxinyue <wuxinyue.wxy@antgroup.com>
As per VirtIO spec 1.2 section 5.2.6, the `status` field is a byte, not
u32. cloud-hypervisor writes an `u32` to guest memory, which
accidentally zeros out the following 3 bytes, and may corrupt guest OS
internal state.
Signed-off-by: Changyuan Lyu <changyuanl@google.com>
With the nightly toolchain (2024-02-18) cargo check will flag up
redundant imports either because they are pulled in by the prelude on
earlier match.
Remove those redundant imports.
Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
Currently the only way to set the affinity for virtio block threads is
to boot the VM, search for the tid of each of the virtio block threads,
then set the affinity manually. This commit adds an option to pin virtio
block queues to specific host cpus (similar to pinning vcpus to host
cpus). A queue_affinity option has been added to the disk flag in
the cli to specify a mapping of queue indices to host cpus.
Signed-off-by: acarp <acarp@crusoeenergy.com>
Add a 'rate_limit_groups' field to VmConfig that defines a set of
named RateLimiterGroups.
When the 'rate_limit_group' field of DiskConfig is defined, all
virtio-blk queues will be rate-limited by a shared RateLimiterGroup.
The lifecycle of all RateLimiterGroups is tied to the Vm.
A RateLimiterGroup may exist even if no Disks are configured to use
the RateLimiterGroup. Disks may be hot-added or hot-removed from the
RateLimiterGroup.
When the 'rate_limiter' field of DiskConfig is defined, we construct
an anonymous RateLimiterGroup whose lifecycle is tied to the Disk.
This is primarily done for api backwards compatability. Importantly,
the behavior is not the same! This implementation rate_limits the
aggregate bandwidth / iops of an individual disk rather than the
bandwidth / iops of an individual queue of a disk.
When neither the 'rate_limit_group' or the 'rate_limiter' fields of
DiskConfig is defined, the Disk is not rate-limited.
Signed-off-by: Thomas Barrett <tbarrett@crusoeenergy.com>
The cumulative average formula [1] requires to use signed integers
for proper calculations, while calculated result (e.g. cumulative
average) is always positive. This patch reflects the above requirements
in our code.
[1] https://en.wikipedia.org/wiki/Moving_average#Cumulative_averageFixes: #5745
Signed-off-by: Bo Chen <chen.bo@intel.com>
There is a "LATENCY_SCALE" being used for calculating cumulative average
latency, so it should also be used for the latency of the first op.
See: #5712
Signed-off-by: Bo Chen <chen.bo@intel.com>
Logically until we have handled the first operation the latency is
infinite; this logic was applied to the minimum latency originally but
this patch extends that logic to the maximum and average latency.
To prevent the initial average latency being skewed by the inclusion of
infinity the average value is initally seeded with the first measured
latency.
Fixes: #5704
Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
This commit merges crates `qcow`, `vhdx` and `block_util` into the
crate `block`, which can allow `qcow` to use functions from `block_util`
without introducing a circular crate dependency.
This commit is based on crosvm implementation:
f2eecc4152
Signed-off-by: Yu Li <liyu.yukiteru@bytedance.com>
This change is important to do a proper resource cleanup. We decided
to do this repetitive approach as VirtioCommon can't implement Drop
without major changes to the corresponding code. Also, devices such as
Net can't easily use the epoll_threads-abstraction from VirtioCommon as
it has multiple threads with different semantics.
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
Add new latency counters for virtio-block device, including
minimal latency, maximal latency, and average latency for block
read and write.
The average latency is calculated based on cumulative average.
Signed-off-by: Yong He <alexyonghe@tencent.com>
Rather than aggregate the completion list into an intermediate vector
instead adjust the API to provide one completion item at a time.
With DHAT this shows the number of heap allocations has decreased.
Before:
dhat: Total: 623,852 bytes in 8,157 blocks
After:
dhat: Total: 380,444 bytes in 3,469 blocks
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
During analysis of the asynchrous block I/O handling it was observed
that the majority of the time the completion events occur in the same
order as submissions. Further the maximum number of inflight requests
during the boot time is much lower than the size of the queue.
Through the use of a double ended queue (VecDequeue) with a reasonable
pre-allocation capacity we can have O(1) allocation free addition of
items to the list of inflight requests and mostly O(1) matching of
completed requests to submissions.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
There is duplicated code when handlin queue events in handle_event()
refactor and introduce a new helper function.
Signed-off-by: Hao Xu <howeyxu@tencent.com>
The information about the identifier related to a Snapshot is only
relevant from the BTreeMap perspective, which is why we can get rid of
the duplicated identifier in every Snapshot structure.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Following the new restore design, it is not appropriate to set every
virtio device threads into a paused state after they've been started.
This is why we remove the line of code pausing the devices only after
they've been restored, and replace it with a small patch in every virtio
device implementation. When a virtio device is created as part of a
restored VM, the associated "paused" boolean is set to true. This
ensures the corresponding thread will be directly parked when being
started, avoiding the thread to be in a different state than the one it
was on the source VM during the snapshot.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
TEST=Boot `--disk readonly=on` along with a guest that tries to write
(unmodified hypervisor-fw) and observe that the virtio device thread no
longer panics.
Fixes: #4888
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
