2023-04-12 03:15:32

by 李扬韬

Subject: [PATCH] erofs: don't access kobject object members directly

It is better not to directly access the internal members of the
kobject object, especially kobject_init_and_add() may failure.
BTW remove unnecessary kobject_del(), kobject_put() actually covers
kobject removal automatically, which is single stage removal.

Signed-off-by: Yangtao Li <[email protected]>
---
fs/erofs/internal.h | 1 +
fs/erofs/sysfs.c | 4 ++--
2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
index f675050af2bb..f364b1e9b35b 100644
--- a/fs/erofs/internal.h
+++ b/fs/erofs/internal.h
@@ -177,6 +177,7 @@ struct erofs_sb_info {
/* sysfs support */
struct kobject s_kobj; /* /sys/fs/erofs/<devname> */
struct completion s_kobj_unregister;
+ bool sysfs_registered;

/* fscache support */
struct fscache_volume *volume;
diff --git a/fs/erofs/sysfs.c b/fs/erofs/sysfs.c
index 435e515c0792..c38018d3c442 100644
--- a/fs/erofs/sysfs.c
+++ b/fs/erofs/sysfs.c
@@ -228,6 +228,7 @@ int erofs_register_sysfs(struct super_block *sb)
kfree(str);
if (err)
goto put_sb_kobj;
+ sbi->sysfs_registered = true;
return 0;

put_sb_kobj:
@@ -240,8 +241,7 @@ void erofs_unregister_sysfs(struct super_block *sb)
{
struct erofs_sb_info *sbi = EROFS_SB(sb);

- if (sbi->s_kobj.state_in_sysfs) {
- kobject_del(&sbi->s_kobj);
+ if (sbi->sysfs_registered) {
kobject_put(&sbi->s_kobj);
wait_for_completion(&sbi->s_kobj_unregister);
}
--
2.35.1

2023-04-12 03:34:00

by Gao Xiang

Subject: Re: [PATCH] erofs: don't access kobject object members directly

(+ Greg, Huang Jianan, Dongliang Mu)

On 2023/4/12 11:11, Yangtao Li wrote:
> It is better not to directly access the internal members of the
> kobject object, especially kobject_init_and_add() may failure.
> BTW remove unnecessary kobject_del(), kobject_put() actually covers
> kobject removal automatically, which is single stage removal.

Please at least Cc the proper people --- and write
the previous background to the commit message *again*.

I'd like to know the preferred way to do this since
the previous discussion is unfinished.

Also if you grep `state_in_sysfs`, actually there are
some similar in-tree usage, also device_is_registered() is
implemented like this:

static inline int device_is_registered(struct device *dev)
{
return dev->kobj.state_in_sysfs;
}

Again, I'm not against this patch (I even prefer this one),
but you should Cc the proper people and finish the previous
topic first.

Thanks,
Gao Xiang

>
> Signed-off-by: Yangtao Li <[email protected]>
> ---
> fs/erofs/internal.h | 1 +
> fs/erofs/sysfs.c | 4 ++--
> 2 files changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
> index f675050af2bb..f364b1e9b35b 100644
> --- a/fs/erofs/internal.h
> +++ b/fs/erofs/internal.h
> @@ -177,6 +177,7 @@ struct erofs_sb_info {
> /* sysfs support */
> struct kobject s_kobj; /* /sys/fs/erofs/<devname> */
> struct completion s_kobj_unregister;
> + bool sysfs_registered;
>
> /* fscache support */
> struct fscache_volume *volume;
> diff --git a/fs/erofs/sysfs.c b/fs/erofs/sysfs.c
> index 435e515c0792..c38018d3c442 100644
> --- a/fs/erofs/sysfs.c
> +++ b/fs/erofs/sysfs.c
> @@ -228,6 +228,7 @@ int erofs_register_sysfs(struct super_block *sb)
> kfree(str);
> if (err)
> goto put_sb_kobj;
> + sbi->sysfs_registered = true;
> return 0;
>
> put_sb_kobj:
> @@ -240,8 +241,7 @@ void erofs_unregister_sysfs(struct super_block *sb)
> {
> struct erofs_sb_info *sbi = EROFS_SB(sb);
>
> - if (sbi->s_kobj.state_in_sysfs) {
> - kobject_del(&sbi->s_kobj);
> + if (sbi->sysfs_registered) {
> kobject_put(&sbi->s_kobj);
> wait_for_completion(&sbi->s_kobj_unregister);
> }