Commit e79c6a4fc923 ("net: make net namespace sysctls belong to container's owner")
added default values for i_uid/i_gid.
These however are only used when ctl_table_root->set_ownership is not
implemented.
But the callbacks themselves could fail to compute i_uid/i_gid and they
all need to have the same fallback logic for this case.
This is unnecessary code duplication and prone to errors.
For example net_ctl_set_ownership() missed the fallback.
Instead always initialize i_uid/i_gid inside the sysfs core so
set_ownership() can safely skip setting them.
Fixes: e79c6a4fc923 ("net: make net namespace sysctls belong to container's owner")
Cc: [email protected]
Signed-off-by: Thomas Weißschuh <[email protected]>
---
Changes in v2:
- Move the fallback logic to the sysctl core
- Link to v1: https://lore.kernel.org/r/[email protected]
---
fs/proc/proc_sysctl.c | 6 ++----
1 file changed, 2 insertions(+), 4 deletions(-)
diff --git a/fs/proc/proc_sysctl.c b/fs/proc/proc_sysctl.c
index 37cde0efee57..9e34ab9c21e4 100644
--- a/fs/proc/proc_sysctl.c
+++ b/fs/proc/proc_sysctl.c
@@ -479,12 +479,10 @@ static struct inode *proc_sys_make_inode(struct super_block *sb,
make_empty_dir_inode(inode);
}
+ inode->i_uid = GLOBAL_ROOT_UID;
+ inode->i_gid = GLOBAL_ROOT_GID;
if (root->set_ownership)
root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
- else {
- inode->i_uid = GLOBAL_ROOT_UID;
- inode->i_gid = GLOBAL_ROOT_GID;
- }
return inode;
}
---
base-commit: ff9c18e435b042596c9d48badac7488e3fa76a55
change-id: 20240315-sysctl-net-ownership-bc4e17eaeea6
Best regards,
--
Thomas Weißschuh <[email protected]>
From: "Thomas Weißschuh" <[email protected]>
Date: Fri, 22 Mar 2024 20:51:11 +0100
> Commit e79c6a4fc923 ("net: make net namespace sysctls belong to container's owner")
> added default values for i_uid/i_gid.
The commit that added the default is 5ec27ec735ba ("fs/proc/proc_sysctl.c:
fix the default values of i_uid/i_gid on /proc/sys inodes.")
> These however are only used when ctl_table_root->set_ownership is not
> implemented.
> But the callbacks themselves could fail to compute i_uid/i_gid and they
> all need to have the same fallback logic for this case.
>
> This is unnecessary code duplication and prone to errors.
> For example net_ctl_set_ownership() missed the fallback.
>
> Instead always initialize i_uid/i_gid inside the sysfs core so
> set_ownership() can safely skip setting them.
>
> Fixes: e79c6a4fc923 ("net: make net namespace sysctls belong to container's owner")
> Cc: [email protected]
> Signed-off-by: Thomas Weißschuh <[email protected]>
> ---
> Changes in v2:
> - Move the fallback logic to the sysctl core
> - Link to v1: https://lore.kernel.org/r/[email protected]
> ---
> fs/proc/proc_sysctl.c | 6 ++----
> 1 file changed, 2 insertions(+), 4 deletions(-)
>
> diff --git a/fs/proc/proc_sysctl.c b/fs/proc/proc_sysctl.c
> index 37cde0efee57..9e34ab9c21e4 100644
> --- a/fs/proc/proc_sysctl.c
> +++ b/fs/proc/proc_sysctl.c
> @@ -479,12 +479,10 @@ static struct inode *proc_sys_make_inode(struct super_block *sb,
> make_empty_dir_inode(inode);
> }
>
> + inode->i_uid = GLOBAL_ROOT_UID;
> + inode->i_gid = GLOBAL_ROOT_GID;
> if (root->set_ownership)
> root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
> - else {
> - inode->i_uid = GLOBAL_ROOT_UID;
> - inode->i_gid = GLOBAL_ROOT_GID;
> - }
>
> return inode;
> }
>
> ---
> base-commit: ff9c18e435b042596c9d48badac7488e3fa76a55
> change-id: 20240315-sysctl-net-ownership-bc4e17eaeea6
>
> Best regards,
> --
> Thomas Weißschuh <[email protected]>
On Fri, Mar 22, 2024 at 08:51:11PM +0100, Thomas Wei?schuh wrote:
> Commit e79c6a4fc923 ("net: make net namespace sysctls belong to container's owner")
> added default values for i_uid/i_gid.
> These however are only used when ctl_table_root->set_ownership is not
> implemented.
> But the callbacks themselves could fail to compute i_uid/i_gid and they
> all need to have the same fallback logic for this case.
All this text is related to commit e79c6a4fc923; which is not the one at
fault. I think mentioning e79c6a4fc923 here is incorrect. Please mention
5ec27ec735ba0 ("fs/proc/proc_sysctl.c: fix the default values of
i_uid/i_gid on /proc/sys inodes") and how it missed adjusting
net_ctl_set_ownership when it was committed.
>
> This is unnecessary code duplication and prone to errors.
> For example net_ctl_set_ownership() missed the fallback.
This paragraph can be dropped
>
> Instead always initialize i_uid/i_gid inside the sysfs core so
> set_ownership() can safely skip setting them.
>
> Fixes: e79c6a4fc923 ("net: make net namespace sysctls belong to container's owner")
Why did you leave this? Do you disagree with my comment in
https://lore.kernel.org/all/[email protected]/?
> Cc: [email protected]
> Signed-off-by: Thomas Wei?schuh <[email protected]>
> ---
> Changes in v2:
> - Move the fallback logic to the sysctl core
> - Link to v1: https://lore.kernel.org/r/[email protected]
> ---
> fs/proc/proc_sysctl.c | 6 ++----
> 1 file changed, 2 insertions(+), 4 deletions(-)
>
> diff --git a/fs/proc/proc_sysctl.c b/fs/proc/proc_sysctl.c
> index 37cde0efee57..9e34ab9c21e4 100644
> --- a/fs/proc/proc_sysctl.c
> +++ b/fs/proc/proc_sysctl.c
> @@ -479,12 +479,10 @@ static struct inode *proc_sys_make_inode(struct super_block *sb,
> make_empty_dir_inode(inode);
> }
>
> + inode->i_uid = GLOBAL_ROOT_UID;
> + inode->i_gid = GLOBAL_ROOT_GID;
> if (root->set_ownership)
> root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
> - else {
> - inode->i_uid = GLOBAL_ROOT_UID;
> - inode->i_gid = GLOBAL_ROOT_GID;
> - }
>
> return inode;
> }
>
> ---
> base-commit: ff9c18e435b042596c9d48badac7488e3fa76a55
Now that v6.9-rc1 is out, I would rebase your next version there.
The change looks good, but the commit message still needs work.
--
Joel Granados