Age | Commit message (Collapse) | Author |
|
|
|
|
|
https://rust-lang.github.io/rust-clippy/master/index.html#assertions_on_result_states
|
|
Signed-off-by: Costin-Robert Sin <sin.costinrobert@gmail.com>
|
|
The existing AIO implementation has some problems:
1) The in_progress field is checked at runtime, not compile time.
2) The mutable field is checked at runtime, not compile time.
3) A downstream lio_listio user must store extra state to track whether
the whole operation is partially, completely, or not at all
submitted.
4) Nix does heap allocation itself, rather than allowing the caller to
choose it. This can result in double (or triple, or quadruple)
boxing.
5) There's no easy way to use lio_listio to submit multiple operations with
a single syscall, but poll each individually.
6) The lio_listio usage is far from transparent and zero-cost.
7) No aio_readv or aio_writev support.
8) priority has type c_int; should be i32
9) aio_return should return a usize instead of an isize, since it only
uses negative values to indicate errors, which Rust represents via
the Result type.
This rewrite solves several problems:
1) Unsolved. I don't think it can be solved without something like
C++'s guaranteed type elision. It might require changing the
signature of Future::poll too.
2) Solved.
3) Solved, by the new in_progress method and by removing the complicated
lio_listio resubmit code.
4) Solved.
5) Solved.
6) Solved, by removing the lio_listo resubmit code. It can be
reimplemented downstream if necessary. Or even in Nix, but it
doesn't fit Nix's theme of zero-cost abstractions.
7) Solved.
8) Solved.
9) Solved.
The rewrite includes functions that don't work on FreeBSD, so add CI
testing for FreeBSD 14 too.
By default only enable tests that will pass on FreeBSD 12.3. But run a
CI job on FreeBSD 14 and set a flag that will enable such tests.
|
|
parking_lot provides synchronization primitives which aren't
poisoned on panic. This makes it easier to determine which tests
are failing, as a test failure no longer causes all subsequent tests
using that mutex to fail.
|
|
And this time, start running Clippy in CI
|
|
Now that Nix's weird error types are eliminated, there's no reason not
to simply use Errno as the Error type.
|
|
For many of Nix's consumers it be convenient to easily convert a Nix
error into a std::io::Error. That's currently not possible because of
the InvalidPath, InvalidUtf8, and UnsupportedOperation types that have
no equivalent in std::io::Error.
However, very few of Nix's public APIs actually return those unusual
errors. So a more useful API would be for Nix's standard error type to
implement Into<std::io::Error>.
This commit makes Error a simple NewType around Errno. For most
functions it's a drop-in replacement. There are only three exceptions:
* clearenv now returns a bespoke error type. It was the only Nix
function whose error couldn't be cleanly mapped onto an Errno.
* sys::signal::signal now returns Error(Errno::ENOTSUP) instead of
Error::UnsupportedOperation when the user passes an incompatible
argument to `handler`.
* When a NixPath exceeds PATH_MAX, it will now return
Error(Errno::ENAMETOOLONG) instead of Error::InvalidPath.
In the latter two cases there is now some abiguity about whether the
error code was generated by Nix or by the OS. But I think the ambiguity
is worth it for the sake of being able to implement Into<io::Error>.
This commit also introduces Error::Sys() as a migration aid. Previously
that as an enum variant. Now it's a function, but it will work in many
of the same contexts as the original.
Fixes #1155
|
|
|
|
* libc::aiocb must not be moved while the kernel has a pointer to it.
This change enforces that requirement by using std::pin.
* Split LioCbBuilder out of LioCb. struct LioCb relied on the
(incorrect) assumption that a Vec's elements have a stable location in
memory. That's not true; they can be moved during Vec::push. The
solution is to use a Vec in the new Builder struct, but finalize it to
a boxed slice (which doesn't support push) before allowing it to be
submitted to the kernel.
* Eliminate owned buffer types. mio-aio no longer uses owned buffers
with nix::aio. There's little need for it in the world of
async/await. I'm not aware of any other consumers. This
substantially simplifies the code.
|
|
On Cirrus-CI, this test frequently fails with EINVAL. The error goes
away if I add a line of debugging, so it's probably a timing issue. But
I can't debug it myself.
Issue #1361
|
|
On OSX, this test has begun to fail in CI on OSX. Presumably it's because
aio_suspend was getting interrupted by a signal.
|
|
|
|
I suspect that the segfault is due to a stack overflow on musl's signal
stack, but I can't reproduce the failure locally.
Fixes #1169
|
|
`assert_eq!` gives more debug info when the test fails by default than
`assert!`. This should help make debugging easier.
|
|
On Travis (and only on Travis) this test crashes. It hits an internal
assertion within glibc. It happens reliably with rustc 1.37.0. Ignore
the test until Travis updates its images; then we'll try again.
Issue #1099
|
|
|
|
It helps deal with errors like EAGAIN, which can result in a subset of
an LioCb's operations being queued. The test is only enabled on
FreeBSD, because it requires intimate knowledge of AIO system limits.
|
|
Supporting the bytes crate was unnecessarily specific. This change
replaces from_bytes and from_bytes_mut with from_boxed_slice and
from_boxed_mut_slice, which can work with anything that implements
Borrow<[u8]> and BorrowMut<[u8]>, respectively.
|
|
The new LioCb structure allows us to control the exact arguments passed
to lio_listio, guaranteeing that each call gets a unique storage
location for the list argument. This prevents clients from misusing
lio_listio in a way that causes events to get dropped from a kqueue
Fixes #870
|
|
|
|
|
|
It's not actually safe to read into an `Rc<[u8]>`. It only worked
because of a coincidental `unsafe` block. Replace that type with
`BytesMut` from the bytes crate. For consistency's sake, use `Bytes`
for writing too, and completely remove methods relating to `Rc<[u8]>`.
Note that the `AioCb` will actually own the `BytesMut` object. The
caller must call `into_buffer` to get it back once the I/O is complete.
Fixes #788
|
|
|
|
Reads a little bit easier
|
|
Makes it more clear what's being cloned
|
|
|
|
It's unclear why these were static in the first place.
|
|
As of Rust 1.17 'static lifetimes are implied when
declaring consts.
|
|
Several tests make the assumption that all data is written, which
is not guaranteed with write(), so use write_all() instead.
|
|
The libc_bitflags! macro was replaced with a non-recursive one supporting
only public structs. I could not figure out how to make the old macro work
with the upgrade, so I reworked part of the bitflags! macro directly to suit
our needs, much as the original recursive macro was made. There are no uses
of this macro for non-public structs, so this is not a problem for internal code.
|
|
773: Add more accessors for AioCb r=asomers a=asomers
|
|
This fixes the following warning during run of cargo test
warning: variable does not need to be mutable
--> test/sys/test_aio.rs:16:13
|
16 | fn poll_aio(mut aiocb: &mut AioCb) -> Result<()> {
| ^^^^^^^^^
|
= note: #[warn(unused_mut)] on by default
|
|
|
|
Previously, the `AioCb`'s `in_progress` field would erroneously be set
to `true`, even if the syscall had an error
Fixes #714
|
|
Printing a warning message to stderr isn't really appropriate, because
there's no way to guarantee that stderr is even valid. Nor is
aio_suspend necessarily an appropriate action to take.
|
|
These are assumed to be QEMU issues, as they also fail on mips.
|
|
Note that this is now only available for Linux as support is missing in libc
for Android (see rust-lang/libc#671).
As part of this work the SIGUSR2 signal mutex was altered to be a general
signal mutex. This is because all signal handling is shared across all threads
in the Rust test harness, so if you alter one signal, depending on whether it's
additive or may overwrite the mask for other signals, it could break the other
ones. Instead of putting this on the user, just broaden the scope of the mutex
so that any altering of signal handling needs to use it.
|
|
They have four problems:
* The chdir tests change the process's cwd, which is global. Protect them
all with a mutex.
* The wait tests will reap any subprocess, and several tests create
subprocesses. Protect them all with a mutex so only one
subprocess-creating test will run at a time.
* When a multithreaded test forks, the child process can sometimes block in
the stack unwinding code. It blocks on a mutex that was held by a
different thread in the parent, but that thread doesn't exist in the
child, so a deadlock results. Fix this by immediately calling
std::process:exit in the child processes.
* My previous attempt at thread safety in the aio tests didn't work, because
anonymous MutexGuards drop immediately. Fix this by naming the
SIGUSR2_MTX MutexGuards.
Fixes #251
|
|
Seems that pretty much all aio tests fail on x64 musl builds.
|
|
|
|
|
|
The existing AioCb constructors work for simple programs where
everything is stored on the stack. But in more complicated programs the
borrow checker can't prove that a buffer will outlive the AioCb that
references it. Fix this problem by introducting
AioCb::from_boxed_slice, which takes a reference-counted buffer.
Fixes #575
|
|
Adds a mutex to protect access to SIGUSR2 signal handlers by the AIO
tests.
Fixes #578
|
|
Also, fix style bug in AIO tests
|
|
If an AioCb has any in-kernel state, AioCb.drop will print a warning and
wait for it to complete.
|
|
Prevent immutable buffers from being used with aio_read or lio_listio
with LIO_READ. AioCb.from_slice no longer needs to be unsafe.
|
|
|
|
|