Fix fsync absence when fsync=N parameter specified
#1799
Conversation
Please squash the two commits into one, we should not have fix-up commits in the history. Once you do, I'll take a look again. |
RomanSofyin
force-pushed
the
r.sofin/fsync-fixup
branch
from
b1c0bb5
to
2b11473
Compare
|
Hi @axboe, can you look at this again? |
|
You still have the should_fsync() change in there. If you want to have it take a const td, then please do a prep patch with just that. Like I said initially, it's not related to your fix at all outside of you adding a new caller that needs it done. That's by definition a prep patch, should not be included with a fix. |
|
I created PR #1805 which I guess we merge first, then I update the current one in case there will be a conflict. |
|
Since the change depends on the other one, please just have both commits in a single pr. |
Make `should_fsync()` safer by using a pointer to a constant thread_data structure instance. Signed-off-by: Roman Sofin roma.sofin@gmail.com
When `fsync=N` parameter specified and the final IO block issued happens to be Nth an fsync should follow it is missed. This update checks if it needs to loop again in `do_io` to issue that final fsync when all the IO work is done. Signed-off-by: Roman Sofin roma.sofin@gmail.com
RomanSofyin
force-pushed
the
r.sofin/fsync-fixup
branch
from
2b11473
to
d3396c6
Compare
|
Should be fine now. If so, I'll close #1805. |
|
@axboe does this look fine? |
When
fsync=Nparameter specified and the final IO block issued happens to be Nth the fsync should follow it is missed. This update checks if it needs to loop again indo_ioto issue that final fsync when all the IO work is done.Signed-off-by: Roman Sofin roma.sofin@gmail.com