Am Freitag, den 15.12.2006, 11:41 -0600 schrieb Bruno Wolff III: > I am currently using the write barrier support in ext3 and software raid > (raid 1) and was wondering if I added dm-crypt into the mix, if I would > still get the same guaranties for write barriers? For pre-2.6.19 this definitely works, as dm-crypt passes the requests through unmodified (except for changing the data, not the ordering in any way). But with 2.6.19 writes are passed through a workqueue. The more I think about this, the more I get to the conclusion that on SMP systems the ordering could get changed and barrier requests arrive in the wrong order. With multiple CPUs there are multiple worker threads, which could run encryption for write requests in parallel and thus cause the requests arrive at underlying device in a different order. This should by addressed.
Attachment:
signature.asc
Description: Dies ist ein digital signierter Nachrichtenteil
