Handle large read/write/sendfile to File; return actual bytes written #60321
+57
−15
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
xal-0
added
io
The return value for the `jl_fs_write/read/sendfile` functions is an `int`, even through they all take `size_t` length arguments, resulting in potentially truncated counts for large file operations. Instead, on success (including partial read/write), we should return the `uv_fs_t.result` field. I have added a test for this but have left it disabled because it allocates a 5 GiB Vector to read into. One consequence of the truncated return values is that we previously mangled large files while copying them with sendfile. If the return value is truncated to a random negative value, it manifests as JuliaLang#30723 or JuliaLang#39868. If it is truncated to a positive value, sendfile will create a very large destination file with thousands of repeated sections of the source file, causing JuliaLang#56537. The bug will only manifest if LLVM decides to zero-extend the return value. This also changes `jl_fs_write` to return the actual number of bytes written, rather than the number of bytes it was asked to write. If a write returns EAGAIN (usually because the file is O_NONBLOCK/O_NODELAY), we never see the actual number of bytes written.
xal-0
force-pushed
the
raw-file-io-retval
branch
from
633307e to
0dae450
Compare
We use raw Base.Filesystem.File IO on PTYs in the REPL precompile script. Due to JuliaLang#24440, spawning a subprocess using the REPL's shell mode clears the O_NONBLOCK flag set on the PTY, resulting in deadlocks when libuv later reads or writes to it from Julia code. I haven't identified what caused this to become more of a problem recently, but I don't think some extra compilation latency when starting shell mode is worth it if it's causing mystery timeouts in CI.
Member
Author
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The return value for the
jl_fs_write/read/sendfilefunctions is anint, even through they all takesize_tlength arguments, resulting in potentially truncated counts for large file operations. Instead, on success (including partial read/write), we should return theuv_fs_t.resultfield.I have added a test for this but have left it disabled because it allocates a 5 GiB Vector to read into.
One consequence of the truncated return values is that we previously mangled large files while copying them with sendfile. If the return value is truncated to a random negative value, it manifests as #30723 or #39868. If it is truncated to a positive value, sendfile will create a very large destination file with thousands of repeated sections of the source file, causing #56537. The bug will only manifest if LLVM decides to zero-extend the return value.
This also changes
jl_fs_writeto return the actual number of bytes written, rather than the number of bytes it was asked to write. If a write returns EAGAIN (usually because the file is O_NONBLOCK/O_NODELAY), we never see the actual number of bytes written.