Re: [PATCH v3 0/11] vfs: add inode iteration methods
From: Andreas Gruenbacher
Date: Fri Nov 20 2026 - 10:01:22 EST
On Fri, Nov 20, 2026 at 4:13 AM Huang Shijie <huangsj@xxxxxxxx> wrote:
> This series adds two superblock inode iteration helpers,
> super_iter_inodes() and super_iter_inodes_locked(), and converts the
> open-coded sb->s_inodes walks in the VFS to them. The design is based on
> Dave Chinner's patch set:
> "vfs: improving inode cache iteration scalability"
> https://lore.kernel.org/all/20241002014017.3801899-1-david@xxxxxxxxxxxxx/
>
> About the two helpers:
>
> int super_iter_inodes(struct super_block *sb, ino_iter_fn iter_fn,
> ino_iter_eligible_fn eligible, void *priv,
> int flags);
>
> static inline int super_iter_inodes_locked(struct super_block *sb,
> ino_iter_fn iter_fn,
> void *priv);
>
> - super_iter_inodes() runs @iter_fn with a referenced inode and with
> sb->s_inode_list_lock __dropped__, so the callback is free to block and
> to do whatever it wants with the inode except drop the reference the
> iterator holds. The iterator releases that one reference per inode
> itself, so the callback cannot leave a dangling inode behind.
> It is exported because GFS2 can be built as a module.
>
> - super_iter_inodes_locked() keeps sb->s_inode_list_lock held for the
> whole walk and takes no inode references at all. Its callback
> therefore only has an existence guarantee: it must not block and
> must do its own inode state checks and its own reference handling.
> It is internal to the VFS, being declared in fs/internal.h.
>
> - The callback is separate from the caller's data, so a caller keeps
> its state in its own structure instead of filling in a shared
> context:
>
> typedef int (*ino_iter_fn)(struct inode *inode, void *priv);
>
> - %INO_ITER_F_REFERENCED restricts the safe iterator to inodes that
> still have a reference. It preserves the "only handle referenced
> inodes" rule of landlock's sb-delete walk.
>
> - %INO_ITER_F_NOREF does the opposite: only inodes whose reference
> count has already reached zero are handed out. %INO_ITER_F_NOIGET
> suppresses the reference. Neither keeps the inode alive, so the
> caller must make it unclaimable itself.
>
> Patch 1 adds super_iter_inodes().
> Patch 2 adds super_iter_inodes_locked().
> Patch 3-5 convert the callers which cannot use referenced inodes
> (dquot removal, nr_blockdev_pages() and the
> generic_shutdown_super() poisoning loop).
> Patch 6-11 convert the remaining callers (inode eviction, sync_bdevs,
> drop_pagecache_sb, add_dquot_ref, landlock's hook_sb_delete
> and gfs2_evict_inodes()).
>
>
> BTW:
> This patch set depends a small patch:
> "fs/drop_caches: skip filesystems without fs writeback"
> https://lkml.org/lkml/2026/9/23/675
That link doesn't work for me. When referring to other postings, can
you please use lore links?
There also is no posting of a patch with that title from that day on
https://lore.kernel.org/linux-fsdevel/. You may be referring to
https://lore.kernel.org/linux-fsdevel/20260920072811.2064247-2-huangsj@xxxxxxxx/,
but that patch queue no longer applies. So please post patch queues in
a state that applies and optionally, make a git branch available.
(Also, enabling "git config --global format.useAutoBase whenAble" is a
possibility, but it wouldn't have helped in this case.
> v2-->v3:
> -the callers' state no longer goes through a shared struct, the
> callback now takes the caller's own @priv.
>
> -super_iter_inodes() takes @eligible directly, so a caller can reject
> an inode before the reference is paid for.
>
> -rename super_iter_inodes_unsafe() to super_iter_inodes_locked() and
> drop the rcu_read_lock(), which the list lock already covers.
>
> -drop @ctx->prev; the resume point is gone with the unsafe iterator.
>
> -evict_inodes() no longer goes through the locked iterator.
> it now uses the unlocked iterator with INO_ITER_F_NOREF | INO_ITER_F_NOIGET.
>
> -each caller gets its own patch.
>
> v2: https://lore.kernel.org/all/20261113115702.586594-1-huangsj@xxxxxxxx/
> v1: https://lore.kernel.org/all/20261015093854.2082600-1-huangsj@xxxxxxxx/
>
> Huang Shijie (11):
> vfs: add super_iter_inodes()
> vfs: add super_iter_inodes_locked()
> fs/quota: convert remove_dquot_ref() to super_iter_inodes_locked()
> block: convert nr_blockdev_pages() to super_iter_inodes_locked()
> fs: convert generic_shutdown_super() poison loop
> fs/inode: convert evict_inodes() to super_iter_inodes()
> block: convert sync_bdevs() to super_iter_inodes()
> fs/drop_caches: convert drop_pagecache_sb() to super_iter_inodes()
> fs/quota: convert add_dquot_ref() to super_iter_inodes()
> landlock: convert hook_sb_delete() to super_iter_inodes()
> gfs2: convert gfs2_evict_inodes() to super_iter_inodes()
>
> block/bdev.c | 101 ++++++++++--------------
> fs/drop_caches.c | 45 ++++-------
> fs/gfs2/ops_fstype.c | 29 ++-----
> fs/inode.c | 61 ++++++---------
> fs/internal.h | 31 ++++++++
> fs/quota/dquot.c | 139 ++++++++++++++++-----------------
> fs/super.c | 106 ++++++++++++++++++++++---
> include/linux/fs/super_types.h | 29 +++++++
> security/landlock/fs.c | 138 ++++++++++++--------------------
> 9 files changed, 370 insertions(+), 309 deletions(-)
>
> --
> 2.53.0
Thanks,
Andreas