Move raw format implementation into a structured directory layout:
raw_disk.rs -> formats/raw/mod.rs (RawDisk)
raw_sync.rs -> formats/raw/worker/sync.rs (RawSync)
raw_async.rs -> formats/raw/worker/async_uring.rs (RawAsync)
raw_async_aio.rs -> formats/raw/worker/async_aio.rs (RawAio)
raw_async_io_tests.rs -> formats/raw/worker/tests.rs
Update imports in fixed_vhd_sync.rs and fixed_vhd_async.rs to use
the new paths. Re-export formats::raw as raw_disk in lib.rs to
preserve the external API.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Move async_io.rs, fcntl.rs, and request.rs into block/src/io/. These
files are generic I/O infrastructure shared by all formats rather
than format specific code.
The io/ directory name clashes with std::io in lib.rs, so the module
is declared as io_impl via #[path] and the submodules are re-exported
at the crate root to keep existing import paths working.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
FixedVhdDisk is the format level DiskFile wrapper for VHD images.
Rename it to VhdDisk to match the <Format>Disk convention used by
RawDisk, QcowDisk, and VhdxDisk. The Fixed prefix remains on the
workers (FixedVhdSync, FixedVhdAsync) since those are specific to
the fixed subformat.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Align with the <Format>Disk wrapper naming convention. VHDx
has a single on disk format so no variant suffix is needed.
The async backend will be added to VhdxDisk.
No functional change.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
The VHDX synchronous async I/O backend still converted owned
AsyncIoOperation targets back into raw iovec slices before calling the
VHDX read and write helpers. That kept pointer dereferences in the owned
path and allowed a backend mistake to violate the safety boundary.
Handle owned VHDX reads and writes through AsyncIoOperation copy helpers
instead. The VHDX file operations still run synchronously, but data is
copied through operation-owned buffers or guest-memory targets without
reconstructing Rust slices from raw iovec pointers.
Assisted-by: Codex:GPT-5
Signed-off-by: Dylan Reid <dgreid@fb.com>
Similar to the other functions fixed in this series, qcow has helpers
that dereference whatever pointers are passed but are labeled safe.
Use the newly added ops helpers to, instead, provide a safe interface
and implementation.
Assisted-by: Codex:GPT-5.5
Signed-off-by: Dylan Reid <dgreid@fb.com>
Drop the borrowed iovec AsyncIo entry points now that all callers use
owned operations. Rename the transitional owned batch and completion
methods to the final trait names and remove the borrowed submission
helpers from the queue wrappers.
This removes a bunch of known safety foot-guns so future-us don't
accidentally use them.
Assisted-by: Codex:GPT-5.5
Signed-off-by: Dylan Reid <dgreid@fb.com>
Switch virtio-blk request construction and the users to the owned
AsyncIo data path added in the series. Read bounce buffers now return
through AsyncIoCompletion before being copied back to guest memory.
This makes the main virtio async block I/O path use retained request
memory. qcow still has raw-iovec fallback paths at this point; those
are removed in follow-up commits.
Leave the legacy borrowed iovec trait methods in place for a follow-up
cleanup commit to minimize single-commit churn.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Similar to the previous commits, use UringDataIo for qcow async. Again,
the legacy interfaces are kept(at the expense of some temporary code).
The temporary code is unsound, like the existing code, but will be
removed soon.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Start using AioDataIo from RawFileAsyncAio. This adds a temporary
submit_borrowed_operation to enable preserving the unsafe iovec api
until we can remove it in the forthcoming commits.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Add the new AsyncIo apis to fixed_vhd_async. Later commits update
callers to use them and remove their unsound counterparts that take
iovecs.
While doing this, make the owned path validate offset plus length
instead of only the starting offset, so requests that extend past the
VHD size are rejected.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Route RawFileAsync data I/O through UringDataIo's owned-operation
retention path while keeping borrowed submissions available for the
legacy AsyncIo calls during the transition.
This mostly moves code around, temporarily moving uring handling from
RawFileAsync to the UringDataIo, including the unsafe iovec access.
This enables the UringIo to be added to RawFileAsync incrementally.
Later commits will remove the unsafe paths when the callers are updated
to use the new functions.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Implement the new AsyncIo members for qcow_sync while keeping the
legacy borrowed iovec methods in place.
The code before and after this commit is equally unsound. This
intermediate state is not a safety regression and allows for a
bisectable transition to the fully sound code at the end of the
series.
This temporary state breaks out the iovec accesses to helpers used
from both the old and new code and updates the safety comments to
reflect reality.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Refactor VhdxSync around shared iovec helpers which are marked unsafe.
Use these to implement safe wrappers for the new AsyncIo trait.
Leave the existing, unsound read/write vectored calls in place until
all callers are converted to the new interface later in this series.
In addition VHDX code assume it's safe to create slices to GuestMemory
via the AsyncAdaptor in existing code and explicitly after this
change. This is technically unsound as it can easily create multiple
mut refs. At least this is 'llvm update breaks the code' UB, not guest
exploitable UB...
Signed-off-by: Dylan Reid <dgreid@fb.com>
Implement the safe AsyncIo interface the fixed VHD synchronous wrapper
and delegate the actual I/O through RawFileSync. This maintains the
existing interfaces until the callers are converted later in the series.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Implement the owned AsyncIo path for RawFileSync while keeping the
legacy borrowed iovec methods available until all callers can be
converted.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Add owned data-operation, completion, and batch methods to AsyncIo.
These will be used as safe alternatives to the existing, comically
unsafe, but marked safe interface.
Over the course of the following commits, users are converted to the
new interface and after all users of the unsound interfaces are
removed, they are removed.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Similar to uring io added in the parent commit. These async ops deal
with buffer ownership across aio calls.
This will be used in the (increasingly rare) case of io_uring not being
available or desirable.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Add the shared helpers and UringDataIo queue that drive uring async I/O
operations. UringDataIo is the key component responsible for keeping the
memory pointers active while async operations are ongoing. It uses the
async core added in previous commits.
Later commits will change the block backends to use this instead of the
lower level abstractions directly.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Add `AsyncIoOperation` and `AsyncIoCompletion` as the owned request and
completion types that will be used to ensure buffers for async io
outlive the operations that use them. Later commits will update the
`AsyncIo` trait to expose apis using only these instead of raw iovecs.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Add `OwnedIoBuffer` to be used for host owned buffers. These are buffers
backed with either a `Vec` or an aligned allocation and will be used for
bounce buffers. This is host owned memory that can later be copied to
guest memory.
Later commits will use this to ensure backing memory outlives async
operations in a centralized, verified way.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Introduce GuestMemoryTarget to own the combination of an Arc to
GuestMemory and a set of ranges/iovecs. This will be used in the
following commits to replace the iovec pointers that are passed to the
backend operations unsafely.
Assisted-by: Codex:GPT-5
Assisted-by: Claude:Opus-4.7
Signed-off-by: Dylan Reid <dgreid@fb.com>
QCOW2 compressed L2 entries encode their extent layout in bits that
overlap with the flags used by standard L2 entries. In particular,
bit 0 can be part of the compressed entry layout, so it must not be
interpreted as ZERO_FLAG until the entry has first been ruled out as
compressed.
Keep compressed deallocation ahead of zero-flag handling in both the
shared QcowMetadata path and the legacy QcowFile path. This ensures
WRITE_ZEROES deallocates compressed clusters instead of treating a
compressed entry with bit 0 set as an existing logical-zero marker.
Add regression coverage that forces bit 0 on a compressed L2 entry
and verifies WRITE_ZEROES still clears the entry through the
compressed-cluster path.
Assisted-by: Codex:GPT-5
Signed-off-by: Ian Klemm <hi@ianklemm.de>
QCOW2 empty L2 entries in an overlay mean that reads fall
through to the backing file. Reusing the punch_hole path for
WRITE_ZEROES therefore turns a full-cluster zero operation on an
unallocated overlay cluster into backing data exposure.
Keep discard/punch_hole behavior unchanged, but let WRITE_ZEROES
request a logical-zero marker when the image has a backing file.
ZERO_FLAG entries now read as zeros in both the legacy QcowFile
path and the shared runtime metadata path. Partial writes after
such entries seed new clusters from zeros instead of backing data.
Treat ZERO_FLAG entries as logical holes for SEEK_HOLE/SEEK_DATA.
Empty overlay entries with a backing file still report data because
the data exists in the backing file.
Avoid cluster-sized userspace zero buffers when materializing
zero-flagged clusters by zeroing the allocated host range directly.
This keeps recycled clusters safe without making partial writes
allocate large zero-filled Vecs.
Add regression coverage for legacy QcowFile, QcowSync, direct I/O,
QcowAsync/io_uring overlay paths, and a large-cluster partial-write
case.
Assisted-by: Codex:GPT-5
Signed-off-by: Ian Klemm <hi@ianklemm.de>
QCOW2 backing file paths stored in image headers may be
relative. These paths should be interpreted relative to the image
that references them, but the block backend opened them relative to
the process working directory.
Resolve the current image path inside parse_qcow() from the open
file descriptor and use its parent directory for relative backing
paths. Recursive backing chains work the same way because each layer
is parsed from its own file descriptor.
Signed-off-by: Rowen-Ye <rowenye1@gmail.com>
Some block devices (ZFS volume) may require BLKDISCARD and BLKZEROOUT
ioctls for discard and write_zeroes operations respectively.
There is no good way to probe whether fallocate is supported on a block
device. Arguably, punch_hole and write_zeroes are rare. Instead of
having a complex scheme for the IO uring backend, we force it to always
use ioctls. The code can be changed if the synchronized ioctls become a
performance issue.
Changes:
- Detect block devices at construction time
- Use BLKDISCARD ioctl for punch_hole (discard) on block devices
- Use BLKZEROOUT ioctl for write_zeroes on block devices
- Add BLKDISCARD/BLKZEROOUT to VirtioBlock seccomp whitelist
- Keep fallocate() path for regular files (no behavior change)
- Consolidate some helper functions to the new sparse module
Signed-off-by: Wei Liu <liuwe@microsoft.com>
probe_sparse_support() and DiskTopology::is_block_device() each carry
their own copy of the same fstat()+S_IFMT dance to ask "is this fd a
block device?". Hoist a single pub helper
pub(crate) fn is_block_device(fd: RawFd) -> bool
into block::lib and route both call sites through it. Drop the
MaybeUninit gymnastics in favour of mem::zeroed() since libc::stat is
POD.
Drop DiskTopology::is_block_device since it is now just a one line
wrapper around the new helper function.
Pure refactor in preparation for the BLKDISCARD/BLKZEROOUT support,
which needs the same probe in three more backends.
Signed-off-by: Wei Liu <liuwe@microsoft.com>
submit_batch_requests pushed each BatchRequest into the io_uring SQ in
turn and used `?` to bail on the first push failure.
Leaving the initial SQEs visible to the kernel — but submitter.submit()
was never called, and every other call site in this file gates submit()
behind a preceding sq.push() that now also fails on the full ring.
This could allow a guest to DoS it's own queue or worse if the buffer is
freed early.
Signed-off-by: Dylan Reid <dgreid@fb.com>
The bounce buffer for an unaligned descriptor was allocated in
execute_async and leaked on error paths, even though, for the sync case
the kernel already had a pointer to the buffer.
Clean this up by moving ownership of the buffer to the AlignedOperation
type. To make it actually safe, stop stashing a guest memory pointer for
the duration of the op. Instead, save the guest address and pass guest
memory back to the complete function.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Keep the underlying io_uring submission queue push error in raw async
I/O paths instead of replacing it with a generic full-queue message.
Signed-off-by: Keith Adler <kadler@cloudflare.com>
Request::execute and Request::execute_async checked each data descriptor
against `disk_nsectors` using the request's fixed start sector. With
sector = disk_nsectors-1 and N descriptors of 512 bytes each, every
descriptor passed (top = disk_nsectors) but the vectored I/O
collectively read/wrote N*512 bytes starting at the last sector — N-1
sectors past EOF.
For the io_uring/aio raw backends this lets the guest extend the host
disk image beyond its provisioned size, exhausting the host filesystem.
For fixed-VHD images (footer at end of file) the same chain overwrites
the footer with guest-controlled bytes, corrupting the disk image.
Replace the per-descriptor check with a chain-wide check_data_bounds().
Pre-validating the entire request before beginning the operation avoids
having to unroll a partial submit.
Signed-off-by: Dylan Reid <dgreid@fb.com>
Verify that a freshly created sparse QCOW2 image reports a
physical size smaller than its logical size.
Assisted-by: Claude:Opus-4.6
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Verify that try_clone preserves the backend dispatch for both
sync and io_uring backends.
Assisted-by: Claude:Opus-4.6
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Verify that the sync backend disables batch requests and the
io_uring backend enables them.
Assisted-by: Claude:Opus-4.6
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Delete QcowDiskSync and QcowDiskAsync wrapper structs along with
their DiskFile trait impls. Only the AsyncIo worker structs
QcowSync and QcowAsync remain. Reduce module visibility of
qcow_sync and qcow_async to pub(crate).
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Replace QcowDiskSync and QcowDiskAsync constructors in the
qcow_sync and qcow_async test modules with QcowDisk::new,
passing use_io_uring=false and use_io_uring=true respectively.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Update open_qcow2 to construct QcowDisk instead of choosing
between QcowDiskAsync and QcowDiskSync. The backend decision
is now made inside QcowDisk::create_async_io.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Introduce QcowDisk, a unified DiskFile implementation for QCOW2
disk images that handles backend selection at runtime via a
use_io_uring flag, matching the pattern used by FixedVhdDisk.
The wrapper delegates to QcowSync or QcowAsync based on the flag
and includes a compile time guard that returns an error when
io_uring is requested but the feature is not enabled.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
The physical size of a fixed VHD is the data region plus the 512
byte footer. Verify it differs from the logical size.
Assisted-by: Claude:Opus-4.6
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Fixed VHDs do not support resize. Verify the error is returned.
Assisted-by: Claude:Opus-4.6
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Verify that a cloned disk produces the same async I/O backend
as the original for both sync and io_uring paths.
Assisted-by: Claude:Opus-4.6
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
