Re: Regression on linux-next (next-20250120)

[Al Viro <viro@zeniv.linux.org.uk>]

On Tue, Jan 28, 2025 at 04:00:58PM +0000, Borah, Chaitanya Kumar wrote:

> Unfortunately this change does not help us. I think it is the methods member that causes the problem. So the following change solves the problem for us.
> 
> 
> --- a/fs/debugfs/file.c
> +++ b/fs/debugfs/file.c
> @@ -102,6 +102,8 @@ static int __debugfs_file_get(struct dentry *dentry, enum dbgfs_get_mode mode)
>                 if (!fsd)
>                         return -ENOMEM;
> 
> +               fsd->methods = 0;
> +
>                 if (mode == DBGFS_GET_SHORT) {
>                         const struct debugfs_short_fops *ops;
>                         ops = fsd->short_fops = DEBUGFS_I(inode)->short_fops;

D'OH.

Both are needed, actually.  Slightly longer term I would rather
split full_proxy_{read,write,lseek}() into short and full variant,
getting rid of the "check which pointer is non-NULL" and killed
the two remaining users of debugfs_real_fops() outside of
fs/debugfs/file.c; then we could union these ->..._fops pointers,
but until then they need to be initialized.

And yes, ->methods obviously needs to be initialized.

Al, bloody embarrassed ;-/