Re: [External] [PATCH v13 08/10] fuse: update inode size/mtime after passthrough write

[Jens Axboe <axboe@kernel.dk>]

On 9/25/23 4:43 AM, Amir Goldstein wrote:
> Jens,
> 
> Are there any IOCB flags that overlayfs (or backing_aio) need
> to set or clear, besides IOCB_DIO_CALLER_COMP, that
> would prevent calling completion from interrupt context?

There are a few flags that may get set (like WAITQ or ALLOC_CACHE), but
I think that should all work ok as-is as the former is just state in
that iocb and that is persistent (and only for the read path), and
ALLOC_CACHE should just be propagated.

> Or is the proper way to deal with this is to defer completion
> to workqueue in the common backing_aio helpers that
> I am re-factoring from overlayfs?

No, deferring to a workqueue would defeat the purpose of the flag, which
is telling you that the caller will ensure that the end_io callback will
happen from task context and need not be deferred to a workqueue. I can
take a peek at how to wire it up properly for overlayfs, have some
travel coming up in a few days.

> IIUC, that could also help overlayfs support
> IOCB_DIO_CALLER_COMP?
> 
> Is my understanding correct?

If you peek at fs.h and find the CALLER_COMP references, it'll tell you
a bit about how it works. This is new with the 6.6-rc kernel, there's a
series of patches from me that went in through the iomap tree that
hooked that up. Those commits have an explanation as well.

-- 
Jens Axboe