fs: copy: use copy_file_range on Linux
Linux 4.5 introduced a new system call [copy_file_range](http://man7.org/linux/man-pages/man2/copy_file_range.2.html) to copy data from one file to another.
This PR uses the new system call (if available). This has several advantages:
1. No need to constantly copy data from userspace to kernel space, if the buffer is small or the file is large
2. On some filesystems, like BTRFS, the kernel can leverage internal fs mechanisms for huge performance gains
3. Filesystems on the network dont need to copy data between the host and the client machine (they have to in the current read/write implementation)
I have created a small library that also implements the new system call for some huge performance gains here: https://github.com/nicokoch/fastcopy
Benchmark results are in the README
Previously, every `open64` was accompanied by a `ioctl(…, FIOCLEX)`,
because some old Linux version would ignore the `O_CLOEXEC` flag we pass
to the `open64` function.
Now, we check whether the `CLOEXEC` flag is set on the first file we
open – if it is, we won't do extra syscalls for every opened file. If it
is not set, we fall back to the old behavior of unconditionally calling
`ioctl(…, FIOCLEX)` on newly opened files.
On old Linuxes, this amounts to one extra syscall per process, namely
the `fcntl(…, F_GETFD)` call to check the `CLOEXEC` flag.
On new Linuxes, this reduces the number of syscalls per opened file by
one, except for the first file, where it does the same number of
syscalls as before (`fcntl(…, F_GETFD)` to check the flag instead of
`ioctl(…, FIOCLEX)` to set it).
Most users would expect set_permissions(Metadata.permissions()) to be
non-destructive. While we can't guarantee this, we can at least pass
the needed info to chmod.
Also update the PermissionsExt documentation to disambiguate what it
contains, and to refer to the underlying value as `st_mode` rather than
its type `mode_t`.
Closes#44147
This commit adds the needed modifications to compile the std crate
for the L4 Runtime environment (L4Re).
A target for the L4Re was introduced in commit:
c151220a84
In many aspects implementations for linux also apply for the L4Re
microkernel.
Two uncommon characteristics had to be resolved:
* L4Re has no network funktionality
* L4Re has a maximum stacksize of 1Mb for threads
Co-authored-by: Sebastian Humenda <sebastian.humenda@tu-dresden.de>
The readdir_r function is deprecated on newer Posix systems because of
various problems, and not implemented at all for Fuchsia. There are
already implementations using both, and this patch switches Fuchsia
over to the readdir-based one.
Fixes#40021 for Fuchsia, but that issue also contains discussion of
what should happen for other Posix systems.
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.
The DirEntryExt::ino() implementation was omitted from the first
iteration of this patch, because a dependency needed to be
configured. The fix is straightforward enough.
It is good practice to implement Debug for public types, and
indicating what directory you're reading seems useful.
Signed-off-by: David Henningsson <diwic@ubuntu.com>
These functions allow to read from and write to a file in one atomic
action from multiple threads, avoiding the race between the seek and the
read.
The functions are named `{read,write}_at` on non-Windows (which don't
change the file cursor), and `seek_{read,write}` on Windows (which
change the file cursor).
std: fix `readdir` errors for solaris
A `NULL` from `readdir` could be the end of stream or an error. The only
way to know is to check `errno`, so it must be set to a known value first,
like a 0 that POSIX will never use.
This currently only matters for solaris targets, as the other unix platforms
are using `readdir_r` with a direct error return indication. However, this is
getting deprecated (#34668) so they should all eventually switch to `readdir`.
This PR adds `set_errno`, uses it to clear the value before calling `readdir`,
then checks it again after to see the reason for a `NULL`. A few other small
fixes are included just to get solaris compiling at all.
I couldn't get cross-compilation completely going, so I don't have a good way
to test this beyond a smoke-test cargo build of std. I'd appreciate input from
someone more familiar with solaris -- cc @nbaksalyar?
A `NULL` from `readdir` could be the end of stream or an error. The
only way to know is to check `errno`, so it must be set to a known value
first, like a 0 that POSIX will never use.
This patch adds `set_errno`, uses it to clear the value before calling
`readdir`, then checks it again after to see the reason for a `NULL`.
The Gecko folks currently use Android API level 9 for their builds, so they're
requesting that we move back our minimum supported API level from 18 to 9. Turns
out, ABI-wise at least, there's not that many changes we need to take care of.
The `ftruncate64` API appeared in android-12 and the `log2` and `log2f` APIs
appeared in android-18. We can have a simple shim for `ftruncate64` which falls
back on `ftruncate` and the `log2` function can be approximated with just
`ln(f) / ln(2)`.
This should at least get the standard library building on API level 9, although
the tests aren't quite happening there just yet. As we seem to be growing a
number of Android compatibility shims, they're now centralized in a common
`sys::android` module.
Automated conversion using the untry tool [1] and the following command:
```
$ find -name '*.rs' -type f | xargs untry
```
at the root of the Rust repo.
[1]: https://github.com/japaric/untry
This pushes the implementation detail of proxying `read_to_end` through to
`read_to_end_uninitialized` all the way down to the `FileDesc` and `Handle`
implementations on Unix/Windows. This way intermediate layers will also be able
to take advantage of this optimized implementation.
This commit also adds the optimized implementation for `ChildStdout` and
`ChildStderr`.
Squashed 10 commits:
1) The main cause of the problem is that libstd/os/mod.rs treats emscripten targets as an alias of linux targets, whereas liblibc treats emscripten targets as musl-compliant, so it gets a slightly different struct stat64 defined.
This commit adds conditional compilation checks to use the correct timestamp format on fs metadata functions in the case of compiling to emscripten targets.
2) Update previous commit to comply with rust formatting standards.
Removed tab characters, remove trailing whitespaces.
3) Move emscripten changes into their own file under libstd/os/emscripten
Put libstd/os/linux/fs back to the way it was.
4) Cannot use stat.st_ctim on emscripten to get created time.
5) Remove compile-time conditionals for target_env = musl, it looks like musl builds compile fine already.
6) Undone some formatting changes that are no longer needed,
Removed some more target_env="musl" compilation checks that I missed from my previous commit.
7) upgrade to liblibc e19309c, it fixes the differences in the musl stat and stat64 definitions.
8) Undo the compile-time checks to check for emscripten (or musl targets) in the FileAttr struct.
No longer needed after updating liblibc to e19309c.
9) Change the MetadataExt implementation of emscripten fs.rs module to match the changes in new liblibc.
10) remove a stray return statement, should have been removed in the previous commit.
Android should use 64-bit LFS symbols for `lseek` and `ftruncate`, lest
those offset parameters suffer a lossy cast down to a 32-bit `off_t`.
Unlike GNU/Linux, Android's `stat`, `dirent`, and related functions are
always 64-bit LFS compatible, and `open` already implies `O_LARGEFILE`,
so all those don't need to follow Linux. It might be nice to unify them
anyway, but those other LFS symbols aren't present in API 18 bionic.
r? @alexcrichton
Some struct members have a slighty different name on NetBSD. This has been
fixed in the libc crate, but not in libstd.
This also removes `st_spare` from MetadataExt, since it is private field
reserved for future use.
Android should use 64-bit LFS symbols for `lseek` and `ftruncate`, lest
those offset parameters suffer a lossy cast down to a 32-bit `off_t`.
Unlike GNU/Linux, Android's `stat`, `dirent`, and related functions are
always 64-bit LFS compatible, and `open` already implies `O_LARGEFILE`,
so all those don't need to follow Linux. It might be nice to unify them
anyway, but those other LFS symbols aren't present in API 18 bionic.
r? @alexcrichton
