fuse-resolve: Attempt fd-migration in resolver, if migration

was never attempted.

Since fd is always associated with an inode, we can create an
fd only after resolver resolves an inode. So, there is a possibility
that graph-switch can happen after resolver kicks in, but before
it can complete, thereby resulting in the newly created fd not
migrated to new graph. So, instead of migrating fds only during
graph-switch, we give a second chance during fd-resolution.

As an example, consider following sequence of events during a
create call:
1. create wants to resolve parent inode, hence it starts
   resolution for parent
2. graph-switch happens (it can happen since fuse-request reader
   thread returns after winding lookup calls)
3. fd-migration of all the fds which are currently in fdtable is
   attempted (Note that the fd corresponding to current create call
   is not yet created and added to fd-table, hence it will not be
   migrated as part of graph switch)
4. resolution of parent triggered as part of create, completes
5. fd is created in fuse_create_resume and this fd is not migrated
   to new graph
6. Any future fops on this fd will fail with EBADF errors
   (create call itself will succeed)

Change-Id: Iae06ecfaca24eaacb2e166ffefbbbb57446332ba
BUG: 804592
Signed-off-by: Raghavendra G <raghavendra@gluster.com>
Reviewed-on: http://review.gluster.com/3181
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>

0 files changed, 0 insertions, 0 deletions