Bug #3964
closed
krbd: 32-bit, kernel 3.2.0 system can't do O_DIRECT writes to mapped rbd image with snapshot
Added by Dan Mick over 11 years ago.
Updated almost 11 years ago.
Description
fghaas reported, I reproduced on a precise 32-bit system:
create an image, map, writes work fine, even with dd oflag=direct
make a snapshot, dd oflag=direct fails
- Subject changed from 32-bit, kernel 3.2.0 system can't do O_DIRECT writes to mapped rbd image with snapshot to krbd: 32-bit, kernel 3.2.0 system can't do O_DIRECT writes to mapped rbd image with snapshot
rbd create image -s 100
rbd map image
dd if=/dev/zero of=/dev/rbd0 oflag=direct count=10
rbd snap create image@s1
dd if=/dev/zero of=/dev/rbd0 oflag=direct count=10 [<< fails]
rbd snap rm image@s1
dd if=/dev/zero of=/dev/rbd0 oflag=direct count=10
This works fine on current testing 3.6.0-00210-g8cc17ca
added test to suite, commit:c782d2ac531cbb7650968e62f0b24e6136a64359
So is this then a request to port whatever it was that
fixed the problem back to 3.2?
If so, how do we prioritize this?
And Sage, your (Dan's) test script is nice; does it ever
get run (i.e., nightly)?
I don't know if Sage tested 32-bit, or if it matters, and no, that script was just a reproduction scenario; as far as I know we just found out about this last night. I'll check with Josh if this doesn't sound familiar to you, Alex, but, I don't know what our path forward is; I suspect we'd prefer people to upgrade kernels, but, I dunno.
Josh thinks 32-bitness probably doesn't matter, and remembers problems with snapshots that were fixed long ago; I guess it comes down to "do we backport anything to 3.2", and I suspect we'd really rather not...
...and to answer your other question Alex, there's now a workunit test Sage just added
in c782d2ac531cbb7650968e62f0b24e6136a64359
- Priority changed from Normal to Low
- Status changed from New to Won't Fix
Also available in: Atom
PDF