Re: [PATCH v8] ntfs: convert a named $DATA stream safely when making room in a record

[Wahid Khan <wahidzk0091@gmail.com>]

Hi Namjae,

-EAGAIN must not leave ntfs_resident_attr_resize(). The callers return
it as is, so a normal blocking write() or setxattr() would get EAGAIN
for a "would block" that never was, and index.c already uses -EAGAIN
internally as its "redo the insert" code (ntfs_ib_insert() ->
resplit), so a stray one from ntfs_attr_truncate() could be taken for
that.

-ENOSPC is what this function already returns when the attribute does
not fit and cannot be made non-resident, so callers already pass it up
as a normal error.
It is not exactly true here, it is a short lived "busy" and not a full
record, but there is no way to wait for folio 0 at that point without
the deadlock (the reader holds it and waits for mrec_lock), and no
errno fits better. I can use -EBUSY or add a comment saying so if you
prefer.

Thanks,
Wahid