On Fri, 2008-09-12 at 13:10 +0100, Hugh Dickins wrote: > So long as the I/O schedulers guarantee that a WRITE bio submitted > to an area already covered by a DISCARD_NOBARRIER bio cannot pass that > DISCARD_NOBARRIER - ... > That seems a reasonable guarantee to me, and perhaps it's trivially > obvious to those who know their I/O schedulers; but I don't, so I'd > like to hear such assurance given. No, that's the point. the I/O schedulers _don't_ give you that guarantee at all. They can treat DISCARD_NOBARRIER just like a write. That's all it is, really -- a special kind of WRITE request without any data. But -- and this came as a bit of a shock to me -- they don't guarantee that writes don't cross writes on their queue. If you issue two WRITE requests to the same sector, you have to make sure for _yourself_ that there is some kind of barrier between them to keep them in the right order. Does swap do that, when a page on the disk is deallocated and then used for something else? -- David Woodhouse Open Source Technology Centre David.Woodhouse@xxxxxxxxx Intel Corporation -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
