Re: [PATCH v2 2/5] ntfs: report attribute errors to fsnotify

[liubaolin]

在 2026/9/16 15:38, Hyunchul Lee 写道:
> Hi Baolin,
>
> >   folio_err_out:
> > @@ -3720,7 +3737,8 @@ static int ntfs_attr_update_meta(struct attr_record *a, struct ntfs_inode *ni,
> >   static int __ntfs_attr_update_mapping_pairs(struct ntfs_inode *ni,
> >   					    s64 from_vcn,
> >   					    struct ntfs_inode *locked_ni,
> > -					    bool defer_attrlist)
> > +					    bool defer_attrlist,
> > +					    bool *error_reported)
> >   {
> >   	struct ntfs_attr_search_ctx *ctx;
> >   	struct ntfs_inode *base_ni;
> > @@ -4086,8 +4104,11 @@ static int __ntfs_attr_update_mapping_pairs(struct ntfs_inode *ni,
> >   	return 0;
> >   
> >   put_err_out:
> > -	if (ctx)
> > +	if (ctx) {
> > +		if (error_reported && ctx->error_reported)
> > +			*error_reported = true;
> >   		ntfs_attr_put_search_ctx(ctx);
> > +	}
> >   	return err;
> >   }
>
> __ntfs_attr_update_mapping_pairs() does:
>
>    ntfs_attr_put_search_ctx(ctx);
>    ctx = NULL
>
> Therefore any later mapping or other errors can therefore lose report
> state and duplicate reports. And The error_reported boolean only tell
> us that some lower-level path has already emitted an error. Consequently
> a report can incorrectly suppress a separate report for a later rollback
> or resotre failure, even though that failure represents an independant
> filesystem integrity problem.
>
> Instead of carrying error_reported, would it be simpler to report each
> error at its detection point and let fanofify merge repeated error
> events? or would there be any alternatives?
Hi Hyunchul,
  Thanks for the feedback. I agree that reporting errors at their 
detection points and relying on fanotify's event merging is simpler. 
I'll update the series accordingly and send v3.

Thanks,
Baolin.