Rob Norris
Rob Norris
Right, I think I understand it. Its annoying but I think shows up a deficiency. BIOs have two limits: number of segments, and max size. On the real devices I've...
Main change in last push is to track number of bytes used when scanning the ABD up front. If adding the current page would blow the BIO byte limit, we...
I also reordered the commits to put the ABD page stuff first, so I could collapse the original version with the updated one, since it was just a mess. I'm...
Bummer. Still, lots further. I'll check it out soon!
> Bummer. Still, lots further. I'll check it out soon! Got it, just a counting bug.
Last push reorganises the commits to keep the individual diffs cleaner. The intermediate stages are mostly unuseful on their own (though they compile and work), but they show a logical...
Alright, _almost_ there. The EL7 crash logs showed two issues. The VERIFY trip was because of a miscount, because it turns out you can't trust `bi_vcnt` as "number of segments...
Pretty sure I figured it all out. There was a real infinite loop, I screwed up handling zero space left in the BIO. The rest was `checkpoint_removal` apparently requiring a...
EL7 failures are a total system hang during `zpool_reopen/zpool_reopen_003_pos`. I did get a kernel crash dump on a local run: ``` Jan 12 18:34:36 centos7 kernel: ------------[ cut here ]------------...
Short answer is that I'm corrupting memory somewhere in the BIO cleanup stage. I don't know if its a 3.x thing or a EL7 thing, but something is different in...