Horton

Re: Checking pgwin32_is_junction() errors

Thomas Munro <thomas.munro@gmail.com>

View thread View on postgr.es
From: Thomas Munro <thomas.munro@gmail.com>
To: Andrew Dunstan <andrew@dunslane.net>
Cc: Michael Paquier <michael@paquier.xyz>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2022-08-06T01:02:41Z
Lists: pgsql-hackers 
On Fri, Aug 5, 2022 at 9:17 PM Thomas Munro <thomas.munro@gmail.com> wrote:
> Hmm, POSIX says st_link should contain the length of a symlink's
> target path, so I suppose we should probably set that even though we
> never consult it.  Here's a version that does that.  I also removed
> the rest of the now redundant #ifdef S_ISLNK conditions.

Pushed.

Hmm, this stuff could *really* use a little test framework that's run
by check-world, that exercises these various replacement operations.
But I also suspect that problems in this area are likely to be due to
concurrency.  It's hard to make a simple test that simulates the case
where a file is unlinked between system calls within stat() and hits
the STATUS_DELETE_PENDING case.  That check is code I cargo-culted in
this patch.  So much of the stuff we've had in the tree relating to
that area has been wrong in the past...

Commits

  1. Fix readlink() for non-PostgreSQL junction points on Windows.

  2. Fix stat() for recursive junction points on Windows.

  3. aix: fix misreading of condition in 8f12a4e7add

  4. Make unlink() work for junction points on Windows.

  5. Replace pgwin32_is_junction() with lstat().

  6. Provide lstat() for Windows.