cygwin: fix off-by-one in utsname.sysname length#5099
Open
phdye wants to merge 2 commits into
Open
Conversation
Cygwin's <sys/utsname.h> defines _UTSNAME_LENGTH as 65, and all
six fields of struct utsname are 65 bytes:
struct utsname {
char sysname [65];
char nodename [65];
char release [65];
char version [65];
char machine [65];
char domainname[65];
};
Verified by C-side struct probe on a Cygwin x86_64 host:
sizeof(struct utsname) = 390 (= 6 * 65)
offsetof(sysname) = 0
offsetof(nodename) = 65
offsetof(release) = 130
offsetof(version) = 195
offsetof(machine) = 260
The libc crate declares sysname as 66 bytes, which shifts every
other field one byte forward in Rust's view of the struct.
Reading the fields with CStr::from_ptr therefore drops the first
character of nodename, release, version, and machine. sysname
itself is unaffected because field 0 starts at offset 0 either
way.
Empirically, on a host where `uname -a` reports
CYGWIN_NT-10.0-26200 xps-ne 3.6.7-1.x86_64 ...
a Rust caller of libc::uname() sees:
sysname = "CYGWIN_NT-10.0-26200" (correct)
nodename = "ps-ne" (lost 'x')
release = ".6.7-1.x86_64" (lost '3')
machine = "86_64" (lost 'x')
Downstream consequences include silently mangled hostnames in
Rust-based logging/telemetry on Cygwin, broken `uname -m` output
in `uutils/coreutils`, and malformed wheel platform tags
produced by `maturin` (cygwin_86_64 instead of cygwin_x86_64)
which Cygwin Python's `pip` then refuses with
"is not a supported wheel on this platform."
Fix: declare sysname at the correct length of 65 bytes, matching
the other five fields. After the fix, all five public fields sit
at the correct C offsets and CStr reads return the full strings.
Adds a runtime test that calls libc::uname() and asserts:
1. nodename matches gethostname() — guards against the field
shifting back to a misaligned offset.
2. every variable-length field reads as a non-empty string.
The C-side struct layout cross-validation in libc-test/ctest
already catches a regression of the [c_char; 65] length at
compile time. This test is a runtime safety net in case
utsname is ever moved to ctest's skip list for Cygwin.
rustbot
added
S-waiting-on-review
stable-nominated
3 tasks
JohnTitor
reviewed
May 12, 2026
Member
There was a problem hiding this comment.
Could you drop this test? We don't have concrete testing structure for this type and I'd figure out it before actually adding anything.
Contributor
There was a problem hiding this comment.
On that note @phdye were you able to run libc-test on this target? It should have caught things like this
tgross35
reviewed
Jun 18, 2026
Comment on lines
449
to
456
| pub struct utsname { | ||
| pub sysname: [c_char; 66], | ||
| pub sysname: [c_char; 65], | ||
| pub nodename: [c_char; 65], | ||
| pub release: [c_char; 65], | ||
| pub version: [c_char; 65], | ||
| pub machine: [c_char; 65], | ||
| pub domainname: [c_char; 65], | ||
| } |
Contributor
There was a problem hiding this comment.
Honestly you can just define a nonpublic _SYS_UTSNAME_H and use that for all the fields, to match the source.
This part of the change seems fine but cc maintainer @Berrysoft in case
Contributor
|
rust-lang does not have a policy against AI use but please handwrite all PR descriptions and commit messages. The amount of text is overwhelming and not necessary for a simple length fix, all you need to do is say what changed and link the source. |
Contributor
|
😢That's... I have already found that the length of |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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.
Description
libcdeclares the Cygwinutsnamestruct'ssysnamefield as[c_char; 66], but Cygwin's<sys/utsname.h>declares all six fields as_UTSNAME_LENGTH= 65 bytes. The over-declaration shifts every subsequent field one byte forward in Rust's view of the struct, soCStr::from_ptrreadsnodename,release,version,machine, anddomainnamestarting at byte 1 instead of byte 0 — silently dropping the first character of each.This PR changes a single character (
66→65) and adds a runtime regression test.Empirical effect on a Cygwin x86_64 host where
uname -areportsCYGWIN_NT-10.0-26200 xps-ne 3.6.7-1.x86_64 ...:CYGWIN_NT-10.0-26200CYGWIN_NT-10.0-26200ps-nexps-ne.6.7-1.x86_643.6.7-1.x86_6486_64x86_64sysnamereads correctly because field 0 starts at offset 0 either way; subsequent fields each begin one byte too late.Downstream this propagates to
platform-info,uutils/coreutils'suname -m, andmaturin's wheel-tag formatter — which produces*-cygwin_86_64.whl(instead ofcygwin_x86_64) that Cygwin Python'spipthen refuses withis not a supported wheel on this platform.Sources
Cygwin's
<sys/utsname.h>in the newlib-cygwin tree:https://sourceware.org/git/?p=newlib-cygwin.git;a=blob;f=winsup/cygwin/include/sys/utsname.h
Checklist
libc-test/semverhave been updated — N/A; correcting layout of an existing struct field, not adding/removing symbols.*LASTor*MAXare includedcargo test --target x86_64-pc-cygwinpasses onlibc-0.2with this fix; passes onmainwith this fix plus the cpuset_t typo fix from companion PR cygwin: fix cpuset_t typo in CPU_ZERO #5098. The added regression test verifiesuname.nodename == gethostname()and that all variable-length fields read non-empty.Note: this PR is independent but
maindoes not currently build onx86_64-pc-cygwindue to a separatecpuset_ttypo, addressed in companion PR #5098. The fix here was verified end-to-end on both thelibc-0.2branch (which builds cleanly) and onmainwith both fixes stacked.@rustbot label +stable-nominated