Regarding Filesystem Handler Table Functions

[Bharath Suri]

On Thu, May 13, 2010 at 11:26 PM, Chris Johns <chrisj at rtems.org> wrote:
> On 13/05/10 11:53 PM, Bharath Suri wrote:
>>
>> I am right now facing some difficulty wrt Filesystem Handler functions.
>>
>> Following is the setup.
>>
>> /imfs exists, /tmp/test exists
>>
>> A filesystem is mounted at /imfs, with the
>> rtems_filesystem_operations_table argument mostly same as IMFS_ops,
>> except that the evalformake_h = NULL and rename_h = NULL.
>>
>> /imfs/test is created.
>>
>> When node info ( rtems_filesystem_location_info_t) of /tmp/test is
>> seen, mt_entry is that of the root file system and ops is the IMFS_ops
>> (OK)
>> When node info ( rtems_filesystem_location_info_t) of /imfs/test is
>> seen, mt_entry is that of the new file system mounted, but ops is
>> IMFS_ops, instead of the modified one.
>>
>> I am assuming that the ops should be the modified one, which was
>> provided during the mount.
>>
>
> It should be if the mount is successful. Is it ? Are you checking for errors
> ?
>

Errors are being checked for and the mount was successful in this case.

>> Mount operation was performed in the below manner:
>>
>> rtems_filesystem_operations_table  IMFS_ops_no_evalformake = IMFS_ops;
>> IMFS_ops_no_evalformake.evalformake_h = NULL;
>> mount(&mt_entry,&IMFS_ops_no_evalformake,
>> RTEMS_FILESYSTEM_READ_WRITE, NULL, "/imfs");
>>
>> Please let me know if the assumption is wrong or about any mistakes in
>> the procedure I am following.
>
> I think you are being detected in 'Is_node_fs_root' as being a root fs and
> attempting to mount it twice. Check the mount code in libcsupport.

The mount was successful, and Is_node_fs_root returned a FALSE.

Following are some observations with respect to the ops of the
temp_mt_entry (which later becomes the mt_entry)

The location info of the mount point is obtained and in this case if
would be of the /imfs. The ops for this would be IMFS_ops. This is
assigned to temp_mt_entry->ops.

Later, a call to

fs_ops->fsmount_me_h( temp_mt_entry ), fs_ops being the ops as sent to
mount function
(in this case, fs_ops->evalformake_h = NULL)

fs_ops->fsmount_me_h = IMFS_initialize which will further call
IMFS_initialize_support (with ops = IMFS_ops)
So, now the temp_mt_entry->mt_fs_root.ops is set to IMFS_ops (NOK)

In the end, the temp_mt_entry is copied to mt_entry, and this has the
ops as IMFS_ops

IMO, the implementation is in a way that the ops assigned to mt_entry
is decided by the (requested_ops)->fsmount_me_h
i.e, if (requested_ops)->fsmount_me_h = IMFS_Initialize, ops assigned
to mt_entry is IMFS_ops.
Same is the case with dosfs
if (requested_ops)->fsmount_me_h = msdos_Initialize, ops assigned to
mt_entry is msdos_ops.

Although, I am not sure if this is the way it is supposed to be.

/Bharath