From: Jan Kara <jack@suse.cz>
To: Christoph Hellwig <hch@lst.de>
Cc: Christian Brauner <brauner@kernel.org>,
Alexander Viro <viro@zeniv.linux.org.uk>,
"Darrick J. Wong" <djwong@kernel.org>, Jan Kara <jack@suse.cz>,
Jens Axboe <axboe@kernel.dk>, Avi Kivity <avi@scylladb.com>,
Damien Le Moal <dlemoal@kernel.org>,
Naohiro Aota <naohiro.aota@wdc.com>,
Johannes Thumshirn <jth@kernel.org>,
linux-xfs@vger.kernel.org, linux-fsdevel@vger.kernel.org,
io-uring@vger.kernel.org
Subject: Re: [PATCH 3/5] iomap: rework REQ_FUA selection
Date: Mon, 24 Nov 2025 12:05:14 +0100 [thread overview]
Message-ID: <saep4t5rujrszxabjiou5x5y5o26thwdyu3vr5wf6uvn6zxpf4@ybkwqxwl55d2> (raw)
In-Reply-To: <20251113170633.1453259-4-hch@lst.de>
On Thu 13-11-25 18:06:28, Christoph Hellwig wrote:
> The way how iomap_dio_can_use_fua and the caller is structured is
> a bit confusing, as the main guarding condition is hidden in the
> helper, and the secondary conditions are split between caller and
> callee.
>
> Refactor the code, so that iomap_dio_bio_iter itself tracks if a write
> might need metadata updates based on the iomap type and flags, and
> then have a condition based on that to use the FUA flag.
>
> Note that this also moves the REQ_OP_WRITE assignment to the end of
> the branch to improve readability a bit.
>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
Looks good. Feel free to add:
Reviewed-by: Jan Kara <jack@suse.cz>
Honza
> ---
> fs/iomap/direct-io.c | 78 +++++++++++++++++++++++++++-----------------
> 1 file changed, 48 insertions(+), 30 deletions(-)
>
> diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
> index 765ab6dd6637..fb2d83f640ef 100644
> --- a/fs/iomap/direct-io.c
> +++ b/fs/iomap/direct-io.c
> @@ -287,23 +287,6 @@ static int iomap_dio_zero(const struct iomap_iter *iter, struct iomap_dio *dio,
> return 0;
> }
>
> -/*
> - * Use a FUA write if we need datasync semantics and this is a pure data I/O
> - * that doesn't require any metadata updates (including after I/O completion
> - * such as unwritten extent conversion) and the underlying device either
> - * doesn't have a volatile write cache or supports FUA.
> - * This allows us to avoid cache flushes on I/O completion.
> - */
> -static inline bool iomap_dio_can_use_fua(const struct iomap *iomap,
> - struct iomap_dio *dio)
> -{
> - if (iomap->flags & (IOMAP_F_SHARED | IOMAP_F_DIRTY))
> - return false;
> - if (!(dio->flags & IOMAP_DIO_WRITE_THROUGH))
> - return false;
> - return !bdev_write_cache(iomap->bdev) || bdev_fua(iomap->bdev);
> -}
> -
> static int iomap_dio_bio_iter(struct iomap_iter *iter, struct iomap_dio *dio)
> {
> const struct iomap *iomap = &iter->iomap;
> @@ -332,7 +315,24 @@ static int iomap_dio_bio_iter(struct iomap_iter *iter, struct iomap_dio *dio)
> return -EINVAL;
>
> if (dio->flags & IOMAP_DIO_WRITE) {
> - bio_opf |= REQ_OP_WRITE;
> + bool need_completion_work = true;
> +
> + switch (iomap->type) {
> + case IOMAP_MAPPED:
> + /*
> + * Directly mapped I/O does not inherently need to do
> + * work at I/O completion time. But there are various
> + * cases below where this will get set again.
> + */
> + need_completion_work = false;
> + break;
> + case IOMAP_UNWRITTEN:
> + dio->flags |= IOMAP_DIO_UNWRITTEN;
> + need_zeroout = true;
> + break;
> + default:
> + break;
> + }
>
> if (iomap->flags & IOMAP_F_ATOMIC_BIO) {
> /*
> @@ -345,22 +345,40 @@ static int iomap_dio_bio_iter(struct iomap_iter *iter, struct iomap_dio *dio)
> bio_opf |= REQ_ATOMIC;
> }
>
> - if (iomap->type == IOMAP_UNWRITTEN) {
> - dio->flags |= IOMAP_DIO_UNWRITTEN;
> - need_zeroout = true;
> - }
> -
> - if (iomap->flags & IOMAP_F_SHARED)
> + if (iomap->flags & IOMAP_F_SHARED) {
> + /*
> + * Unsharing of needs to update metadata at I/O
> + * completion time.
> + */
> + need_completion_work = true;
> dio->flags |= IOMAP_DIO_COW;
> + }
>
> - if (iomap->flags & IOMAP_F_NEW)
> + if (iomap->flags & IOMAP_F_NEW) {
> + /*
> + * Newly allocated blocks might need recording in
> + * metadata at I/O completion time.
> + */
> + need_completion_work = true;
> need_zeroout = true;
> - else if (iomap->type == IOMAP_MAPPED &&
> - iomap_dio_can_use_fua(iomap, dio))
> - bio_opf |= REQ_FUA;
> + }
>
> - if (!(bio_opf & REQ_FUA))
> - dio->flags &= ~IOMAP_DIO_WRITE_THROUGH;
> + /*
> + * Use a FUA write if we need datasync semantics and this is a
> + * pure overwrite that doesn't require any metadata updates.
> + *
> + * This allows us to avoid cache flushes on I/O completion.
> + */
> + if (dio->flags & IOMAP_DIO_WRITE_THROUGH) {
> + if (!need_completion_work &&
> + !(iomap->flags & IOMAP_F_DIRTY) &&
> + (!bdev_write_cache(iomap->bdev) ||
> + bdev_fua(iomap->bdev)))
> + bio_opf |= REQ_FUA;
> + else
> + dio->flags &= ~IOMAP_DIO_WRITE_THROUGH;
> + }
> + bio_opf |= REQ_OP_WRITE;
> } else {
> bio_opf |= REQ_OP_READ;
> }
> --
> 2.47.3
>
--
Jan Kara <jack@suse.com>
SUSE Labs, CR
next prev parent reply other threads:[~2025-11-24 11:05 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-11-13 17:06 enable iomap dio write completions from interrupt context v2 Christoph Hellwig
2025-11-13 17:06 ` [PATCH 1/5] fs, iomap: remove IOCB_DIO_CALLER_COMP Christoph Hellwig
2025-11-13 17:09 ` Jens Axboe
2025-11-13 17:06 ` [PATCH 2/5] iomap: always run error completions in user context Christoph Hellwig
2025-11-13 17:06 ` [PATCH 3/5] iomap: rework REQ_FUA selection Christoph Hellwig
2025-11-24 11:05 ` Jan Kara [this message]
2025-11-13 17:06 ` [PATCH 4/5] iomap: support write completions from interrupt context Christoph Hellwig
2025-11-24 11:11 ` Jan Kara
2025-11-13 17:06 ` [PATCH 5/5] iomap: invert the polarity of IOMAP_DIO_INLINE_COMP Christoph Hellwig
2025-11-14 11:47 ` enable iomap dio write completions from interrupt context v2 Christian Brauner
-- strict thread matches above, loose matches on Subject: below --
2025-11-12 7:21 enable iomap dio write completions from interrupt context Christoph Hellwig
2025-11-12 7:21 ` [PATCH 3/5] iomap: rework REQ_FUA selection Christoph Hellwig
2025-11-12 20:07 ` Jan Kara
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=saep4t5rujrszxabjiou5x5y5o26thwdyu3vr5wf6uvn6zxpf4@ybkwqxwl55d2 \
--to=jack@suse.cz \
--cc=avi@scylladb.com \
--cc=axboe@kernel.dk \
--cc=brauner@kernel.org \
--cc=djwong@kernel.org \
--cc=dlemoal@kernel.org \
--cc=hch@lst.de \
--cc=io-uring@vger.kernel.org \
--cc=jth@kernel.org \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-xfs@vger.kernel.org \
--cc=naohiro.aota@wdc.com \
--cc=viro@zeniv.linux.org.uk \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox