Handle ERELOOKUP from VOP_FSYNC()

[kib]

Summary

We need to repeat the operation if the vnode was relocked.

Reported by: markj

[markj]

Thanks, this fixes the panic I saw.

sys/vm/vm_object.c

1238

Why isn't it necessary to relock the vnode here like we do in md? And if we can simply call VOP_FSYNC() in a loop like this, why not handle the retry entirely in the VOP?

[kib]

sys/vm/vm_object.c

1238

Generally it is useful to unlock the vnode since it allows SU thread and bufdaemon to clean more dependencies that perhaps caused the relock during ffs_fsyncvnode().

I added the relock there.

[kib]

Relock for vm_object_sync()

[markj]

sys/vm/vm_object.c

1244

[kib]

Avoid duplicating VOP_FSYNC() call.
Move comment to the more logical place. Make it stylish.

[markj]

sys/vm/vm_object.c

1249