Hi Samuel, Just a quick ping on this V6 since it’s been three weeks and more than a month since v5.
It would be great to get your feedback on the core architecture when you have a moment so we can figure out next steps. Thanks, Milos On Wed, Apr 1, 2026, 10:55 AM Milos Nikic <[email protected]> wrote: > Hi Samuel, > > Makes perfect sense. Thanks for getting back to me! > No rush on the review I know this is very large. > Let me know if some form of breaking up would help. (for instance jbd2 > headers have not changed at all since v1, diskfs transaction wrapping is > separate and self contained...these can all be viewed separately alongside > the journal in the ext2fs). > On the flips side i can also generate just the v5-v6 diff, if that would > help! Just let me know > > Here is a v6. As mentioned, it builds cleanly with no compiler warnings on > both 32-bit and 64-bit, and works smoothly on both architectures. > The edge-case deadlock on a full journal has been fixed, and I've added > comments explaining the logic there. > > As a quick reassurance regarding the buildd machines: the journal is > strictly opt-in. If this code is merged as-is, zero journaling paths are > actually exercised by default. (the weird assert blowup in v5 got fixed up > as well) > > To initialize and run it, one must explicitly execute: > sudo tune2fs -j /dev/nbd0 > (and then restart the Hurd). > > If the journal is enabled and does happen to trigger any weird edge cases > on the buildd machines down the road, please just send me the logs or any > available info about what failed etc. I will gladly replicate and fix them > locally. > > In an emergency, if a build ever gets blocked due to the journal, one can > always instantly disable the journal and unblock the machine with: > sudo tune2fs -O ^has_journal /dev/nbd0 > > Best, > Milos > > > On Wed, Apr 1, 2026 at 12:22 AM Samuel Thibault <[email protected]> > wrote: > >> Milos Nikic, le mar. 31 mars 2026 21:17:10 -0700, a ecrit: >> > Should I send a v6 now with those fixes included, or hold onto them >> until >> > you've had a chance to look at the core v5 architecture? >> >> Yes, you can send a v6, I haven't had the time to have a look at v5, so >> better look at both at the same time. >> >> Thanks for having taken the time to stress-test, that'll avoid me >> finding these on buildd machines and not have time to investigate. >> >> Samuel >> >
