On 2017-06-13 00:57, Corinna Vinschen wrote:> On Jun 12 13:08, Joe Lowe wrote:
>> On 2017-06-12 03:27, Corinna Vinschen wrote:
>>> On Jun 9 15:44, Joe Lowe wrote:
>>>> - else
>>>> + else if (res == -1)
>>>> {
>>>> - /* Volume moint point or unrecognized reparse point type.
>>>> + /* Volume moint point or unhandled reparse point.
>>>> Make sure the open handle is not used in later stat
>>>> calls.
>>>> The handle has been opened with the
>>>> FILE_OPEN_REPARSE_POINT
>>>> flag, so it's a handle to the reparse point, not a
>>>> handle
>>>> - to the volumes root dir. */
>>>> + to the reparse point target. */
>>>> pflags &= ~PC_KEEP_HANDLE;
>>>> - /* Volume mount point: The filesystem information for the top
>>>> - level directory should be for the volume top level directory,
>>>> - rather than for the reparse point itself. So we fetch the
>>>> - filesystem information again, but with a NULL handle.
>>>> - This does what we want because fs_info::update opens the
>>>> - handle without FILE_OPEN_REPARSE_POINT. */
>>>> - if (res == -1)
>>>> - fs.update (&upath, NULL);
>>>> + /* The filesystem information should be for the target of
>>>> + the reparse point rather than for the reparse point itself.
>>>> + So we fetch the filesystem information again, but with a
>>>> + NULL handle. This does what we want because fs_info::update
>>>> + opens the handle without FILE_OPEN_REPARSE_POINT. */
>>>> + fs.update (&upath, NULL);
>>>> + }
>>>> + else
>>>> + {
>>>> + /* Unknown reparse point type: HSM, dedup, compression, ...
>>>> + Treat as normal directory. */
>>>> }
>>>
>>> Nothing against reordering the code, but this drops removing
>>> PC_KEEP_HANDLE from pflags if res == 0, i.e., for unknown reparse
>>> points.
>>
>> Changing the handling for unknown (3rd party) reparse tags is
>> not the primary point of the patch. But the fallback handling
>> for reparse tags should be to do what most win32 apps would do,
>> treat as a normal file/directory and open without the
>> FILE_OPEN_REPARSE_TAG flag.
>>
...
> I don't get what you're arguing for. Removing the PC_KEEP_HANDLE
> bit will close the (wrong) search handle from path_conv, so that
> subsequent stat calls will stat the target of the reparse point.
> By keeping it in case of unknown reparse points, subsequent stats
> will return the wrong info (reparse point rather than target).
OK. Next pass should be a smaller diff, and I will double check
things. Please check what I end up with in reworked patch and let
me know then if there still is a concern.
Joe L.
