From: Jaegeuk Kim via Linux-f2fs-devel <linux-f2fs-devel@lists.sourceforge.net>
To: Nanzhe Zhao <nzzhao@126.com>
Cc: linux-kernel@vger.kernel.org, linux-f2fs-devel@lists.sourceforge.net
Subject: Re: [f2fs-dev] [PATCH 1/2 v4] f2fs: support large folio for immutable non-compressed case
Date: Sun, 4 Jan 2026 03:20:42 +0000 [thread overview]
Message-ID: <aVncihATY3KXN59M@google.com> (raw)
In-Reply-To: <16d92acc.17ac.19b837e5d80.Coremail.nzzhao@126.com>
On 01/03, Nanzhe Zhao wrote:
> Dear Kim:
> Thanks for your quick reply!
>
> I applied the two bug fixes on my local branch and found that
> I still couldn't pass my test of generating and reading a heavily
> fragmented file.
>
> The root cause is that current code will treat hole blocks as mapped
> blocks as well and mistakenly increment read_pages_pending, resulting
> task hung in readahead.
>
> Inside f2fs_map_blocks():
>
> /* DIO READ and hole case, should not map the blocks. */
> if (!(flag == F2FS_GET_BLOCK_DIO && is_hole && !map->m_may_create))
> map->m_flags |= F2FS_MAP_MAPPED;
>
> it will have map->m_flags marked with F2FS_MAP_MAPPED in non-DIO and
> no blocks creation context for NULL_ADDR and NEW_ADDR, except for
> holes mapped to an unallocated dnode.
>
> Personally, I think a better fix is to add a helper function
> f2fs_block_needs_zeroing(). The condition could be: return true if the
> current blkaddr is NULL_ADDR or NEW_ADDR.
>
> Then we can reverse the order of the checks under the got_it: label:
> first `if (f2fs_block_needs_zeroing()) ...`, and then `else if
> (map->m_flags & F2FS_MAP_MAPPED)`, while keeping all the logic inside
> those statements unchanged.
>
> For the parameters of f2fs_block_needs_zeroing(), I think we can pass
> `struct f2fs_map_blocks` directly, because it already contains all the
> information we need. Also, if we later want to support batching
> contiguous physical block mappings and bio additions inside the loop,
> this signature should be more extensible.
>
> If you think this approach makes sense, I can send a patch to fix all
> three bugs. Thank you.
I think that's feasbile. Could you please post a patch to discuss further?
Thanks,
>
> Best regards,
> Nanzhe Zhao
_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
WARNING: multiple messages have this Message-ID (diff)
From: Jaegeuk Kim <jaegeuk@kernel.org>
To: Nanzhe Zhao <nzzhao@126.com>
Cc: Chao Yu <chao@kernel.org>,
linux-kernel@vger.kernel.org,
linux-f2fs-devel@lists.sourceforge.net
Subject: Re: [f2fs-dev] [PATCH 1/2 v4] f2fs: support large folio for immutable non-compressed case
Date: Sun, 4 Jan 2026 03:20:42 +0000 [thread overview]
Message-ID: <aVncihATY3KXN59M@google.com> (raw)
In-Reply-To: <16d92acc.17ac.19b837e5d80.Coremail.nzzhao@126.com>
On 01/03, Nanzhe Zhao wrote:
> Dear Kim:
> Thanks for your quick reply!
>
> I applied the two bug fixes on my local branch and found that
> I still couldn't pass my test of generating and reading a heavily
> fragmented file.
>
> The root cause is that current code will treat hole blocks as mapped
> blocks as well and mistakenly increment read_pages_pending, resulting
> task hung in readahead.
>
> Inside f2fs_map_blocks():
>
> /* DIO READ and hole case, should not map the blocks. */
> if (!(flag == F2FS_GET_BLOCK_DIO && is_hole && !map->m_may_create))
> map->m_flags |= F2FS_MAP_MAPPED;
>
> it will have map->m_flags marked with F2FS_MAP_MAPPED in non-DIO and
> no blocks creation context for NULL_ADDR and NEW_ADDR, except for
> holes mapped to an unallocated dnode.
>
> Personally, I think a better fix is to add a helper function
> f2fs_block_needs_zeroing(). The condition could be: return true if the
> current blkaddr is NULL_ADDR or NEW_ADDR.
>
> Then we can reverse the order of the checks under the got_it: label:
> first `if (f2fs_block_needs_zeroing()) ...`, and then `else if
> (map->m_flags & F2FS_MAP_MAPPED)`, while keeping all the logic inside
> those statements unchanged.
>
> For the parameters of f2fs_block_needs_zeroing(), I think we can pass
> `struct f2fs_map_blocks` directly, because it already contains all the
> information we need. Also, if we later want to support batching
> contiguous physical block mappings and bio additions inside the loop,
> this signature should be more extensible.
>
> If you think this approach makes sense, I can send a patch to fix all
> three bugs. Thank you.
I think that's feasbile. Could you please post a patch to discuss further?
Thanks,
>
> Best regards,
> Nanzhe Zhao
next prev parent reply other threads:[~2026-01-04 3:21 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-11-20 23:54 [f2fs-dev] [PATCH 1/2] f2fs: support large folio for immutable non-compressed case Jaegeuk Kim via Linux-f2fs-devel
2025-11-20 23:54 ` Jaegeuk Kim
2025-11-20 23:54 ` [f2fs-dev] [PATCH 2/2] f2fs: add a tracepoint to see large folio read submission Jaegeuk Kim via Linux-f2fs-devel
2025-11-20 23:54 ` Jaegeuk Kim
2025-11-21 10:23 ` [f2fs-dev] " Chao Yu via Linux-f2fs-devel
2025-11-21 10:23 ` Chao Yu
2025-11-21 10:20 ` [f2fs-dev] [PATCH 1/2] f2fs: support large folio for immutable non-compressed case Chao Yu via Linux-f2fs-devel
2025-11-21 10:20 ` Chao Yu
2025-11-22 1:17 ` Jaegeuk Kim via Linux-f2fs-devel
2025-11-22 1:17 ` Jaegeuk Kim
2025-11-25 1:38 ` Chao Yu via Linux-f2fs-devel
2025-11-25 1:38 ` Chao Yu
2025-12-01 19:31 ` [f2fs-dev] [PATCH 1/2 v2] " Jaegeuk Kim via Linux-f2fs-devel
2025-12-01 19:31 ` Jaegeuk Kim
2025-12-01 21:37 ` Chao Yu via Linux-f2fs-devel
2025-12-01 21:37 ` Chao Yu
2025-12-01 22:30 ` [f2fs-dev] [PATCH 1/2 v3] " Jaegeuk Kim via Linux-f2fs-devel
2025-12-01 22:30 ` Jaegeuk Kim
2025-12-01 22:37 ` Chao Yu via Linux-f2fs-devel
2025-12-01 22:37 ` Chao Yu
2025-12-02 2:38 ` [f2fs-dev] [PATCH 1/2 v4] " Jaegeuk Kim via Linux-f2fs-devel
2025-12-02 2:38 ` Jaegeuk Kim
2025-12-02 18:07 ` Chao Yu via Linux-f2fs-devel
2025-12-02 18:07 ` Chao Yu
2025-12-09 8:32 ` Chao Yu via Linux-f2fs-devel
2025-12-09 8:32 ` Chao Yu
2025-12-09 18:38 ` Jaegeuk Kim via Linux-f2fs-devel
2025-12-09 18:38 ` Jaegeuk Kim
2026-01-01 11:20 ` Nanzhe Zhao
2026-01-01 11:20 ` Nanzhe Zhao
2026-01-02 6:23 ` Jaegeuk Kim via Linux-f2fs-devel
2026-01-02 6:23 ` Jaegeuk Kim
2026-01-03 10:54 ` Nanzhe Zhao
2026-01-03 10:54 ` Nanzhe Zhao
2026-01-04 3:20 ` Jaegeuk Kim via Linux-f2fs-devel [this message]
2026-01-04 3:20 ` Jaegeuk Kim
2025-11-22 1:18 ` [f2fs-dev] [PATCH 1/2 v2] " Jaegeuk Kim via Linux-f2fs-devel
2025-11-22 1:18 ` Jaegeuk Kim
2025-12-16 19:20 ` [f2fs-dev] [PATCH 1/2] " patchwork-bot+f2fs--- via Linux-f2fs-devel
2025-12-16 19:20 ` patchwork-bot+f2fs
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=aVncihATY3KXN59M@google.com \
--to=linux-f2fs-devel@lists.sourceforge.net \
--cc=jaegeuk@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=nzzhao@126.com \
/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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.