On 2025-04-10 15:13, Nelson H. F. Beebe wrote:
I made this change against the original test file:
% diff dfly-openat-bug.c dfly-openat-bug-3.c
8c8
< int dirfd = open ("d/", O_SEARCH | O_DIRECTORY);
---
> int dirfd = open ("d/", O_RDONLY | O_DIRECTORY);
11c11
< int fd = openat (dirfd, "f", O_EXCL | O_CREAT | O_WRONLY, 0660);
---
That diff is chopped off; what did line 11 get changed to?
Here is the test run:
% ls -ldR d
dr-xr-xr-x 1 beebe wheel 0 Apr 10 16:06 d
% gcc dfly-openat-bug-3.c && ./a.out
d/f: Permission denied
The final message now appears to be accurate, and expected due to the
directory permissions.
Yes, evidently we didn't fully reproduce the symptoms you reported,
symptoms that I quoted in <https://bugs.gnu.org/77640#17>. The relevant
line there is:
openat(0x4,0x49f822,0xa01,0x180) ERR#17 'File exists'
0x180 == 0660 and I assume 0xa01 equals (O_EXCL | O_CREAT | O_WRONLY).
If so, I don't understand why our attempt to reproduce the bug failed,
unless it has something to do with a change to line 11 above.
I'm still unclear about why gzip would need (or want) to be able to
write into a no-write-access directory in the failing test.
Explanation?
The test is making sure that gzip outputs a reasonable diagnosic when
asked to write into an unwriteable directory. Unfortunately this isn't
happening on DragonflyBSD, evidently due to a kernel bug there.
