| Age | Commit message (Collapse) | Author | Lines |
|---|
|
|
|
|
|
provide thread name to OS for Solarish systems
Fixes #62302
Passes a Linux bootstrap build. python x.py test src/tools/tidy happy.
I tested this with a small test binary that spawns a few threads, and verified
that:
- on an illumos system lacking the libc function, the binary runs but no OS-level
thread names are set
- on an illumos system with the feature, the binary runs, and the thread names are
visible and correct under tools like MDB, pstack, core dump, etc.
|
|
|
|
|
|
|
|
Fix few Clippy warnings
|
|
|
|
|
|
|
|
filedesc: don't use ioctl(FIOCLEX) on Linux
All `ioctl(2)`s will fail on `O_PATH` file descriptors on Linux (because
they use `&empty_fops` as a security measure against `O_PATH` descriptors
affecting the backing file).
As a result, `File::try_clone()` and various other methods would always
fail with `-EBADF` on `O_PATH` file descriptors. The solution is to simply
use `F_SETFD` (as is used on other unices) which works on `O_PATH`
descriptors because it operates through the `fnctl(2)` layer and not
through `ioctl(2)`s.
Since this code is usually only used in strange error paths (a broken or
ancient kernel), the extra overhead of one syscall shouldn't cause any
dramas. Most other systems programming languages also use the fnctl(2)
so this brings us in line with them.
Fixes: rust-lang/rust#62314
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
|
|
All ioctl(2)s will fail on O_PATH file descriptors on Linux (because
they use &empty_fops as a security measure against O_PATH descriptors
affecting the backing file).
As a result, File::try_clone() and various other methods would always
fail with -EBADF on O_PATH file descriptors. The solution is to simply
use F_SETFD (as is used on other unices) which works on O_PATH
descriptors because it operates through the fnctl(2) layer and not
through ioctl(2)s.
Since this code is usually only used in strange error paths (a broken or
ancient kernel), the extra overhead of one syscall shouldn't cause any
dramas. Most other systems programming languages also use the fnctl(2)
so this brings us in line with them.
Fixes: rust-lang/rust#62314
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
|
|
request at least ptr-size alignment from posix_memalign
Fixes https://github.com/rust-lang/rust/issues/62251
|
|
|
|
|
|
|
|
|
|
|
|
For `libc` types that will be initialized in FFI calls, we can just use
`MaybeUninit` and then pass around raw pointers.
For `sun_path_offset()`, which really wants `offset_of`, all callers
have a real `sockaddr_un` available, so we can use that reference.
|
|
macos tlv workaround
fixes: #60141
Includes:
* remove dead code: `requires_move_before_drop`. This hasn't been needed for a while now (oops I should have removed it in #57655)
* redox had a copy of `fast::Key` (not sure why?). That has been removed.
* Perform a `read_volatile` on OSX to reduce `tlv_get_addr` calls per `__getit` from (4-2 depending on context) to 1.
`tlv_get_addr` is relatively expensive (~1.5ns on my machine).
Previously, in contexts where `__getit` was inlined, 4 calls to `tlv_get_addr` were performed per lookup. For some reason when `__getit` is not inlined this is reduced to 2x - and performance improves to match.
After this PR, I have only ever seen 1x call to `tlv_get_addr` per `__getit`, and macos now benefits from situations where `__getit` is inlined.
I'm not sure if the `read_volatile(&&__KEY)` trick is working around an LLVM bug, or a rustc bug, or neither.
r? @alexcrichton
|
|
Print PermissionExt::mode() in octal in Documentation Examples
Printing the file permission mode on unix systems in decimal feels unintuitive. Printing it in octal gives the expected form of e.g. `664`.
|
|
|
|
This commit removes all in-tree support for generating backtraces in
favor of depending on the `backtrace` crate on crates.io. This resolves
a very longstanding piece of duplication where the standard library has
long contained the ability to generate a backtrace on panics, but the
code was later extracted and duplicated on crates.io with the
`backtrace` crate. Since that fork each implementation has seen various
improvements one way or another, but typically `backtrace`-the-crate has
lagged behind libstd in one way or another.
The goal here is to remove this duplication of a fairly critical piece
of code and ensure that there's only one source of truth for generating
backtraces between the standard library and the crate on crates.io.
Recently I've been working to bring the `backtrace` crate on crates.io
up to speed with the support in the standard library which includes:
* Support for `StackWalkEx` on MSVC to recover inline frames with
debuginfo.
* Using `libbacktrace` by default on MinGW targets.
* Supporting `libbacktrace` on OSX as an option.
* Ensuring all the requisite support in `backtrace`-the-crate compiles
with `#![no_std]`.
* Updating the `libbacktrace` implementation in `backtrace`-the-crate to
initialize the global state with the correct filename where necessary.
After reviewing the code in libstd the `backtrace` crate should be at
exact feature parity with libstd today. The backtraces generated should
have the same symbols and same number of frames in general, and there's
not known divergence from libstd currently.
Note that one major difference between libstd's backtrace support and
the `backtrace` crate is that on OSX the crates.io crate enables the
`coresymbolication` feature by default. This feature, however, uses
private internal APIs that aren't published for OSX. While they provide
more accurate backtraces this isn't appropriate for libstd distributed
as a binary, so libstd's dependency on the `backtrace` crate explicitly
disables this feature and forces OSX to use `libbacktrace` as a
symbolication strategy.
The long-term goal of this refactoring is to eventually move us towards
a world where we can drop `libbacktrace` entirely and simply use Gimli
and the surrounding crates for backtrace support. That's still aways off
but hopefully will much more easily enabled by having the source of
truth for backtraces live in crates.io!
Procedurally if we go forward with this I'd like to transfer the
`backtrace-rs` crate to the rust-lang GitHub organization as well, but I
figured I'd hold off on that until we get closer to merging.
|
|
DoubleEndedIterators."
This reverts commit 3e86cf36b5114f201868bf459934fe346a76a2d4.
|
|
convert custom try macro to `?`
resolves #60580
r? @frewsxcv
|
|
Fall back to `/dev/urandom` on `EPERM` for `getrandom`
This can happen because of seccomp or some VMs.
Fixes #52609.
|
|
This reverts commit d252f3b77f3b7d4cd59620588f9d026633c05816.
|
|
|
|
|
|
Remove bitrig support from rust
Resolves #60743
using `find` and `rg` I delete every occurence of "bitrig" in the sources, expect for the llvm submodule (is this correct?).
There's also this file https://github.com/rust-lang/rls/blob/5b8e99bb61958ca8abcb7c5eda70521726be1065/rls-analysis/test_data/rust-analysis/libstd-af9bacceee784405.json which contains a bitrig string in it. What to do with that?
|
|
fix Miri
This reverts https://github.com/rust-lang/rust/pull/60156, which turned out to be a dead end (see https://github.com/rust-lang/rust/pull/60469).
r? @oli-obk
|
|
Add implementations of last in terms of next_back on a bunch of DoubleEndedIterators
Provided a `DoubleEndedIterator` has finite length, `Iterator::last` is equivalent to `DoubleEndedIterator::next_back`. But searching forwards through the iterator when it's unnecessary is obviously not good for performance. I ran into this on one of the collection iterators.
I tried adding appropriate overloads for a bunch of the iterator adapters like filter, map, etc, but I ran into a lot of type inference failures after doing so.
The other interesting case is what to do with `Repeat`. Do we consider it part of the contract that `Iterator::last` will loop forever on it? The docs do say that the iterator will be evaluated until it returns None. This is also relevant for the adapters, it's trivially easy to observe whether a `Map` adapter invoked its closure a zillion times or just once for the last element.
|
|
This reverts commit 54aefc6a2d076b74921a8d78c5d8c68c13bfa4a7.
|
|
|
|
|
|
resolves #60580
|
|
Make `std::fs::copy` attempt to create copy-on-write clones of files on MacOS
The behaviour of MacOS now matches Linux which uses `copy_file_range` to perform CoW file copies where available and supported by the underlying filesystem.
|
|
|
|
use SecRandomCopyBytes on macOS in Miri
This is a hack to fix https://github.com/rust-lang/miri/issues/686: on macOS, rustc will open `/dev/urandom` to initialize a `HashMap`. That's quite hard to emulate properly in Miri without a full-blown implementation of file descriptors. However, Miri needs an implementation of `SecRandomCopyBytes` anyway to support [getrandom](https://crates.io/crates/getrandom), so using it here should work just as well.
This will only have an effect when libstd is compiled specifically for Miri, but that will generally be the case when people use `cargo miri`.
This is clearly a hack, so I am opening this to start a discussion about whether we are okay with such a hack or not.
Cc @oli-obk
|
|
This can happen because of seccomp or some VMs.
Fixes #52609.
|
|
This renames `std::io::IoVec` to `std::io::IoSlice` and
`std::io::IoVecMut` to `std::io::IoSliceMut`, and stabilizes
`std::io::IoSlice`, `std::io::IoSliceMut`,
`std::io::Read::read_vectored`, and `std::io::Write::write_vectored`.
Closes #58452
|
|
|
|
DoubleEndedIterators.
r?Manishearth
|
|
sync_all should flush all metadata in macos/ios, so it should call
fcntl with the F_FULLFSYNC flag as sync_data does.
Fixes #55920
|
|
|
|
std: Add `{read,write}_vectored` for more types
This commit implements the `{read,write}_vectored` methods on more types
in the standard library, namely:
* `std::fs::File`
* `std::process::ChildStd{in,out,err}`
* `std::io::Std{in,out,err}`
* `std::io::Std{in,out,err}Lock`
* `std::io::Std{in,out,err}Raw`
Where supported the OS implementations hook up to native support,
otherwise it falls back to the already-defaulted implementation.
|
|
This commit implements the `{read,write}_vectored` methods on more types
in the standard library, namely:
* `std::fs::File`
* `std::process::ChildStd{in,out,err}`
* `std::io::Std{in,out,err}`
* `std::io::Std{in,out,err}Lock`
* `std::io::Std{in,out,err}Raw`
Where supported the OS implementations hook up to native support,
otherwise it falls back to the already-defaulted implementation.
|
|
|
|
std: Avoid usage of `Once` in `Instant`
This commit removes usage of `Once` from the internal implementation of
time utilities on OSX and Windows. It turns out that we accidentally hit
a deadlock today (#59020) via events that look like:
* A thread invokes `park_timeout`
* Internally, only on OSX, `park_timeout` calls `Instant::elapsed`
* Inside of `Instant::elapsed` on OSX we enter a `Once` to initialize
global timer data
* Inside of `Once`, it attempts to `park`
This means on the same stack frame, when there's contention, we're
calling `park` from inside `park_timeout`, causing a deadlock!
The solution implemented in this commit was to remove usage of `Once`
and instead just do a small dance with atomics. There's no real need we
need to guarantee that the global information is only learned once, only
that it's only *stored* once. This implementation may have multiple
threads invoke `mach_timebase_info`, but only one will store the global
information which will amortize the cost for all other threads.
A similar fix has been applied to windows to be uniform across our
implementations, but looking at the code on Windows no deadlock was
possible. This is purely just a consistency update for Windows and in
theory a slightly leaner implementation.
Closes #59020
|
|
This commit removes usage of `Once` from the internal implementation of
time utilities on OSX and Windows. It turns out that we accidentally hit
a deadlock today (#59020) via events that look like:
* A thread invokes `park_timeout`
* Internally, only on OSX, `park_timeout` calls `Instant::elapsed`
* Inside of `Instant::elapsed` on OSX we enter a `Once` to initialize
global timer data
* Inside of `Once`, it attempts to `park`
This means on the same stack frame, when there's contention, we're
calling `park` from inside `park_timeout`, causing a deadlock!
The solution implemented in this commit was to remove usage of `Once`
and instead just do a small dance with atomics. There's no real need we
need to guarantee that the global information is only learned once, only
that it's only *stored* once. This implementation may have multiple
threads invoke `mach_timebase_info`, but only one will store the global
information which will amortize the cost for all other threads.
A similar fix has been applied to windows to be uniform across our
implementations, but looking at the code on Windows no deadlock was
possible. This is purely just a consistency update for Windows and in
theory a slightly leaner implementation.
Closes #59020
|
