On 2020/9/3 1:45, Darrick J. Wong wrote:
On Wed, Sep 02, 2020 at 10:38:28AM -0700, Ira Weiny wrote:
On Wed, Sep 02, 2020 at 10:03:26AM -0700, Darrick J. Wong wrote:
On Wed, Sep 02, 2020 at 01:11:00PM +0800, Xiao Yang wrote:
Hi Darrick,
It is reasonable for your concern to add a check in VFS, but checking all
defined xflags is too rough in VFS if one filesystem only supports few
xflags. :-)
I was advocating for two levels of flags checks: one in the VFS for
undefined flags, and a second check in each filesystem for whichever
flag it wants to recognize. I was not implying that the VFS checks
would be sufficient on their own.
I've not really followed this thread completely but wouldn't this proposed
check in the VFS layer be redundant because the set of flags the filesystem
accepts should always be a strict subset of the VFS flags?
Hi,
I also think this check in the VFS is redundant. :-)
Yes. It's 100% CYA. I wouldn't be that bent out of shape if the vfs
part never happens, but as we already have a vfs argument checker
function in addition to the per-fs validation I don't see why we would
leave a gap... ;)
After looking at vfs_ioc_fssetxattr_check(), why do we need to move the
check of extent
size hint to vfs? It seems a xfs-specific flag, right?
btw:
It is fine to move DAX and project id to vfs because they are supported
by more than one
filesystem(e.g. ext4 and xfs).
Best Regards,
Xiao Yang
--D
Ira
.
