This commit adds the necessary definitions for target specs and such as well as
the necessary support in libstd to compile basic `aarch64-pc-windows-msvc`
binaries. The target is not currently built on CI, but it can be built locally
with:
./configure --target=aarch64-pc-windows-msvc --set rust.lld
./x.py build src/libstd --target aarch64-pc-windows-msvc
Currently this fails to build `libtest` due to a linker bug (seemingly in LLD?)
which hasn't been investigate yet. Otherwise though with libstd you can build a
hello world program (linked with LLD). I've not tried to execute it yet, but it
at least links!
Full support for this target is still a long road ahead, but this is hopefully a
good stepping stone to get started.
Points of note about this target are:
* Currently defaults to `panic=abort` as support is still landing in LLVM for
SEH on AArch64.
* Currently defaults to LLD as a linker as I was able to get farther with it
than I was with `link.exe`
PR #47252 switched stack inspection functions of dbghelp.dll
to their newer alternatives that also capture inlined context.
Unfortunately, said new alternatives are not present in older
dbghelp.dll versions.
In particular Windows 7 at the time of writing has dbghelp.dll
version 6.1.7601 from 2010, that lacks StackWalkEx and friends.
Fixes#50138
This commit updates the OS random number generator on Windows to match the
upstream implementation in the `rand` crate. First proposed in
rust-lang-nursery/rand#111 this implementation uses a "private" API of
`RtlGenRandom`. Despite the [documentation][dox] indicating this is a private
function its widespread use in Chromium and Firefox as well as [comments] from
Microsoft internally indicates that it's highly unlikely to break.
Another motivation for switching this is to also attempt to make progress
on #44911. It may be the case that this function succeeds while the previous
implementation may fail in "weird" scenarios.
[dox]: https://msdn.microsoft.com/en-us/library/windows/desktop/aa387694(v=vs.85).aspx
[comments]: https://github.com/rust-lang-nursery/rand/issues/111#issuecomment-316140155
Windows io::Error: also format NTSTATUS error codes
`NTSTATUS` errors may be encoded as `HRESULT`, see [[MS-ERREF]](https://msdn.microsoft.com/en-us/library/cc231198.aspx). These error codes can still be formatted using `FormatMessageW` but require some different parameters to be passed in.
I wasn't sure if this needed a test and if so, how to test it. Presumably we wouldn't want to make our tests dependent on localization-dependent strings returned from `FormatMessageW`.
Users that get an `err: NTSTATUS` will need to do `io::Error::from_raw_os_error(err|0x1000_0000)` (the equivalent of [`HRESULT_FROM_NT`](https://msdn.microsoft.com/en-us/library/ms693780(VS.85).aspx))
Gecko recently had a bug reported [1] with a deadlock in the Rust TLS
implementation for Windows. TLS destructors are implemented in a sort of ad-hoc
fashion on Windows as it doesn't natively support destructors for TLS keys. To
work around this the runtime manages a list of TLS destructors and registers a
hook to get run whenever a thread exits. When a thread exits it takes a look at
the list and runs all destructors.
Unfortunately it turns out that there's a lock which is held when our "at thread
exit" callback is run. The callback then attempts to acquire a lock protecting
the list of TLS destructors. Elsewhere in the codebase while we hold a lock over
the TLS destructors we try to acquire the same lock held first before our
special callback is run. And as a result, deadlock!
This commit sidesteps the issue with a few small refactorings:
* Removed support for destroying a TLS key on Windows. We don't actually ever
exercise this as a public-facing API, and it's only used during `lazy_init`
during racy situations. To handle that we just synchronize `lazy_init`
globally on Windows so we never have to call `destroy`.
* With no need to support removal the global synchronized `Vec` was tranformed
to a lock-free linked list. With the removal of locks this means that
iteration no long requires a lock and as such we won't run into the deadlock
problem mentioned above.
Note that it's still a general problem that you have to be extra super careful
in TLS destructors. For example no code which runs a TLS destructor on Windows
can call back into the Windows API to do a dynamic library lookup. Unfortunately
I don't know of a great way around that, but this at least fixes the immediate
problem that Gecko was seeing which is that with "well behaved" destructors the
system would still deadlock!
[1]: https://bugzilla.mozilla.org/show_bug.cgi?id=1358151
Add peek APIs to std::net
Adds "peek" APIs to `std::net` sockets, including:
- `UdpSocket.peek()`
- `UdpSocket.peek_from()`
- `TcpStream.peek()`
These methods enable socket reads without side-effects. That is, repeated calls to `peek()` return identical data. This is accomplished by providing the POSIX flag `MSG_PEEK` to the underlying socket read operations.
This also moves the current implementation of `recv_from` out of the platform-independent `sys_common` and into respective `sys/windows` and `sys/unix` implementations. This allows for more platform-dependent implementations where necessary.
Fixes#38980
Support unprivileged symlink creation in Windows
Symlink creation on Windows has in the past basically required admin; it’s being opened up a bit in the Creators Update, so that at least people who have put their computers into Developer Mode will be able to create symlinks without special privileges. (It’s unclear from what Microsoft has said whether Developer Mode will be required in the final Creators Update release, but sadly I expect it still will be, so this *still* won’t be as helpful as I’d like.)
Because of compatibility concerns, they’ve hidden this new functionality behind a new flag in the CreateSymbolicLink dwFlags: `SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE`. So we add this flag in order to join the party.
Sources:
- https://blogs.windows.com/buildingapps/2016/12/02/symlinks-windows-10/ is the official announcement (search for CreateSymbolicLink)
- https://news.ycombinator.com/item?id=13096354 on why the new flag.
These methods enable socket reads without side-effects. That is,
repeated calls to peek() return identical data. This is accomplished
by providing the POSIX flag MSG_PEEK to the underlying socket read
operations.
This also moves the current implementation of recv_from out of the
platform-independent sys_common and into respective sys/windows and
sys/unix implementations. This allows for more platform-dependent
implementations.
This is done by adding a function that can return a filename
to pass to backtrace_create_state. The filename is obtained in
a safe way by first getting the filename, locking the file so it can't
be moved, and then getting the filename again and making sure it's the same.
See: https://github.com/rust-lang/rust/pull/37359#issuecomment-260123399
Issue: #33985
std: Add a nonblocking `Child::try_wait` method
This commit adds a new method to the `Child` type in the `std::process` module
called `try_wait`. This method is the same as `wait` except that it will not
block the calling thread and instead only attempt to collect the exit status. On
Unix this means that we call `waitpid` with the `WNOHANG` flag and on Windows it
just means that we pass a 0 timeout to `WaitForSingleObject`.
Currently it's possible to build this method out of tree, but it's unfortunately
tricky to do so. Specifically on Unix you essentially lose ownership of the pid
for the process once a call to `waitpid` has succeeded. Although `Child` tracks
this state internally to be resilient to multiple calls to `wait` or a `kill`
after a successful wait, if the child is waited on externally then the state
inside of `Child` is not updated. This means that external implementations of
this method must be extra careful to essentially not use a `Child`'s methods
after a call to `waitpid` has succeeded (even in a nonblocking fashion).
By adding this functionality to the standard library it should help canonicalize
these external implementations and ensure they can continue to robustly reuse
the `Child` type from the standard library without worrying about pid ownership.
Symlink creation on Windows has in the past basically required admin;
it’s being opened up a bit in the Creators Update, so that at least
people who have put their computers into Developer Mode will be able to
create symlinks without special privileges. (Microsoft are being
cautious about it all; the Developer Mode requirement makes it so that
it this won’t be as helpful as I’d like, but it’s still an improvement
over requiring admin.)
Because of compatibility concerns, they’ve hidden this new functionality
behind a new flag in the CreateSymbolicLink dwFlags:
SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE. So we add this flag in
order to join the party.
Older Windows doesn’t like this new flag, though, so if we encounter
ERROR_INVALID_PARAMETER we try again without the new flag.
Sources:
- https://blogs.windows.com/buildingapps/2016/12/02/symlinks-windows-10/
is the official announcement (search for CreateSymbolicLink)
- https://news.ycombinator.com/item?id=13096354 on why the new flag.
- https://twitter.com/richturn_ms/status/818167548269051905 confirming
that Developer Mode will still be required.
This commit adds a new method to the `Child` type in the `std::process` module
called `try_wait`. This method is the same as `wait` except that it will not
block the calling thread and instead only attempt to collect the exit status. On
Unix this means that we call `waitpid` with the `WNOHANG` flag and on Windows it
just means that we pass a 0 timeout to `WaitForSingleObject`.
Currently it's possible to build this method out of tree, but it's unfortunately
tricky to do so. Specifically on Unix you essentially lose ownership of the pid
for the process once a call to `waitpid` has succeeded. Although `Child` tracks
this state internally to be resilient to multiple calls to `wait` or a `kill`
after a successful wait, if the child is waited on externally then the state
inside of `Child` is not updated. This means that external implementations of
this method must be extra careful to essentially not use a `Child`'s methods
after a call to `waitpid` has succeeded (even in a nonblocking fashion).
By adding this functionality to the standard library it should help canonicalize
these external implementations and ensure they can continue to robustly reuse
the `Child` type from the standard library without worrying about pid ownership.
This commit fixes a mistake introduced in #31618 where overlapped handles were
leaked to child processes on Windows. On Windows once a handle is in overlapped
mode it should always have I/O executed with an instance of `OVERLAPPED`. Most
child processes, however, are not prepared to have their stdio handles in
overlapped mode as they don't use `OVERLAPPED` on reads/writes to the handle.
Now we haven't had any odd behavior in Rust up to this point, and the original
bug was introduced almost a year ago. I believe this is because it turns out
that if you *don't* pass an `OVERLAPPED` then the system will [supply one for
you][link]. In this case everything will go awry if you concurrently operate on
the handle. In Rust, however, the stdio handles are always locked, and there's
no way to not use them unlocked in libstd. Due to that change we've always had
synchronized access to these handles, which means that Rust programs typically
"just work".
Conversely, though, this commit fixes the test case included, which exhibits
behavior that other programs Rust spawns may attempt to execute. Namely, the
stdio handles may be concurrently used and having them in overlapped mode wreaks
havoc.
[link]: https://blogs.msdn.microsoft.com/oldnewthing/20121012-00/?p=6343Closes#38811
Ctrl-Z returns from Stdin.read() when reading from the console on Windows
Fixes#19914.
Fixes read(), read_to_string(), read_to_end(), etc.
r? @alexcrichton
On unix like systems, the underlying file corresponding to any given path may
change at any time. This function makes it possible to set the permissions of
the a file corresponding to a `File` object even if its path changes.
This commit is an implementation of [RFC 1513] which allows applications to
alter the behavior of panics at compile time. A new compiler flag, `-C panic`,
is added and accepts the values `unwind` or `panic`, with the default being
`unwind`. This model affects how code is generated for the local crate, skipping
generation of landing pads with `-C panic=abort`.
[RFC 1513]: https://github.com/rust-lang/rfcs/blob/master/text/1513-less-unwinding.md
Panic implementations are then provided by crates tagged with
`#![panic_runtime]` and lazily required by crates with
`#![needs_panic_runtime]`. The panic strategy (`-C panic` value) of the panic
runtime must match the final product, and if the panic strategy is not `abort`
then the entire DAG must have the same panic strategy.
With the `-C panic=abort` strategy, users can expect a stable method to disable
generation of landing pads, improving optimization in niche scenarios,
decreasing compile time, and decreasing output binary size. With the `-C
panic=unwind` strategy users can expect the existing ability to isolate failure
in Rust code from the outside world.
Organizationally, this commit dismantles the `sys_common::unwind` module in
favor of some bits moving part of it to `libpanic_unwind` and the rest into the
`panicking` module in libstd. The custom panic runtime support is pretty similar
to the custom allocator support with the only major difference being how the
panic runtime is injected (takes the `-C panic` flag into account).
Removes all unstable and deprecated APIs prior to the 1.8 release. All APIs that
are deprecated in the 1.8 release are sticking around for the rest of this
cycle.
Some notable changes are:
* The `dynamic_lib` module was moved into `rustc_back` as the compiler still
relies on a few bits and pieces.
* The `DebugTuple` formatter now special-cases an empty struct name with only
one field to append a trailing comma.
Semantically there's actually no reason for us to spawn threads as part of the
call to `wait_with_output`, and that's generally an incredibly heavyweight
operation for just reading a few bytes (especially when stderr probably rarely
has bytes!). An equivalent operation in terms of what's implemented today would
be to just drain both pipes of all contents and then call `wait` on the child
process itself.
On Unix we can implement this through some convenient use of the `select`
function, whereas on Windows we can make use of overlapped I/O. Note that on
Windows this requires us to use named pipes instead of anonymous pipes, but
they're semantically the same under the hood.
Also back out keepalive support for TCP since the API is perhaps not
actually what we want. You can't read the interval on Windows, and
we should probably separate the functionality of turning keepalive on
and overriding the interval.
Fix `read_link` to also be able to read the target of junctions on Windows.
Also the path returned should not include a NT namespace, and there were
some problems with permissions.
Have all Cargo-built crates pass `--cfg cargobuild` and then add appropriate
`#[cfg]` definitions to all crates to avoid linking anything if this is passed.
This should help allow libstd to compile with both the makefiles and with Cargo.
Tracking issue: #30014
This implements the RFC and makes a few other changes.
I have added a few extra tests, and made the Windows and
Unix code as similar as possible.
Part of the RFC mentions the unstable OpenOptionsExt trait
on Windows (see #27720). I have added a few extra methods
to future-proof it for CreateFile2.
This is a standard "clean out libstd" commit which removes all 1.5-and-before
deprecated functionality as it's now all been deprecated for at least one entire
cycle.
This commit is an implementation of [RFC 1288][rfc] which adds two new unstable
types to the `std::time` module. The `Instant` type is used to represent
measurements of a monotonically increasing clock suitable for measuring time
withing a process for operations such as benchmarks or just the elapsed time to
do something. An `Instant` favors panicking when bugs are found as the bugs are
programmer errors rather than typical errors that can be encountered.
[rfc]: https://github.com/rust-lang/rfcs/pull/1288
The `SystemTime` type is used to represent a system timestamp and is not
monotonic. Very few guarantees are provided about this measurement of the system
clock, but a fixed point in time (`UNIX_EPOCH`) is provided to learn about the
relative distance from this point for any particular time stamp.
This PR takes the same implementation strategy as the `time` crate on crates.io,
namely:
| Platform | Instant | SystemTime |
|------------|--------------------------|--------------------------|
| Windows | QueryPerformanceCounter | GetSystemTimeAsFileTime |
| OSX | mach_absolute_time | gettimeofday |
| Unix | CLOCK_MONOTONIC | CLOCK_REALTIME |
These implementations can perhaps be refined over time, but they currently
satisfy the requirements of the `Instant` and `SystemTime` types while also
being portable across implementations and revisions of each platform.
* Delete `sys::unix::{c, sync}` as these are now all folded into libc itself
* Update all references to use `libc` as a result.
* Update all references to the new flat namespace.
* Moves all windows bindings into sys::c
As discovered in #29298, `env::set_var("", "")` will panic, but it turns out
that it *also* deadlocks on Unix systems. This happens because if a panic
happens while holding the environment lock, we then go try to read
RUST_BACKTRACE, grabbing the environment lock, causing a deadlock.
Specifically, the changes made here are:
* The environment lock is pushed into `std::sys` instead of `std::env`. This
also only puts it in the Unix implementation, not Windows where the functions
are already threadsafe.
* The `std::sys` implementation now returns `io::Result` so panics are
explicitly at the `std::env` level.
Note: for now, this change only affects `-windows-gnu` builds.
So why was this `libgcc` dylib dependency needed in the first place?
The stack unwinder needs to know about locations of unwind tables of all the modules loaded in the current process. The easiest portable way of achieving this is to have each module register itself with the unwinder when loaded into the process. All modules compiled by GCC do this by calling the __register_frame_info() in their startup code (that's `crtbegin.o` and `crtend.o`, which are automatically linked into any gcc output).
Another important piece is that there should be only one copy of the unwinder (and thus unwind tables registry) in the process. This pretty much means that the unwinder must be in a shared library (unless everything is statically linked).
Now, Rust compiler tries very hard to make sure that any given Rust crate appears in the final output just once. So if we link the unwinder statically to one of Rust's crates, everything should be fine.
Unfortunately, GCC startup objects are built under assumption that `libgcc` is the one true place for the unwind info registry, so I couldn't find any better way than to replace them. So out go `crtbegin`/`crtend`, in come `rsbegin`/`rsend`!
A side benefit of this change is that rustc is now more in control of the command line that goes to the linker, so we could stop using `gcc` as the linker driver and just invoke `ld` directly.
