[PATCH v3 bpf-next 1/4] kernfs: remove iattr_mutex

[Jan Kiszka]

On 02.07.25 14:17, Christian Brauner wrote:
> On Wed, Jul 02, 2025 at 11:47:58AM +0100, André Draszik wrote:
>> Hi,
>>
>> On Sun, 2025-06-22 at 23:38 -0700, Song Liu wrote:
>>> From: Christian Brauner <brauner at kernel.org>
>>>
>>> All allocations of struct kernfs_iattrs are serialized through a global
>>> mutex. Simply do a racy allocation and let the first one win. I bet most
>>> callers are under inode->i_rwsem anyway and it wouldn't be needed but
>>> let's not require that.
>>>
>>> Signed-off-by: Christian Brauner <brauner at kernel.org>
>>> Acked-by: Greg Kroah-Hartman <gregkh at linuxfoundation.org>
>>> Acked-by: Tejun Heo <tj at kernel.org>
>>> Signed-off-by: Song Liu <song at kernel.org>
>>
>> On next-20250701, ls -lA gives errors on /sys:
>>
>> $ ls -lA /sys/
>> ls: /sys/: No data available
>> ls: /sys/kernel: No data available
>> ls: /sys/power: No data available
>> ls: /sys/class: No data available
>> ls: /sys/devices: No data available
>> ls: /sys/dev: No data available
>> ls: /sys/hypervisor: No data available
>> ls: /sys/fs: No data available
>> ls: /sys/bus: No data available
>> ls: /sys/firmware: No data available
>> ls: /sys/block: No data available
>> ls: /sys/module: No data available
>> total 0
>> drwxr-xr-x   2 root root 0 Jan  1  1970 block
>> drwxr-xr-x  52 root root 0 Jan  1  1970 bus
>> drwxr-xr-x  88 root root 0 Jan  1  1970 class
>> drwxr-xr-x   4 root root 0 Jan  1  1970 dev
>> drwxr-xr-x  11 root root 0 Jan  1  1970 devices
>> drwxr-xr-x   3 root root 0 Jan  1  1970 firmware
>> drwxr-xr-x  10 root root 0 Jan  1  1970 fs
>> drwxr-xr-x   2 root root 0 Jul  2 09:43 hypervisor
>> drwxr-xr-x  14 root root 0 Jan  1  1970 kernel
>> drwxr-xr-x 251 root root 0 Jan  1  1970 module
>> drwxr-xr-x   3 root root 0 Jul  2 09:43 power
>>
>>
>> and my bisect is pointing to this commit. Simply reverting it also fixes
>> the errors.
>>
>>
>> Do you have any suggestions?
> 
> Yes, apparently the xattr selftest don't cover sysfs/kernfs. The issue
> is that the commit changed listxattr() to skip allocation of the xattr
> header and instead just returned ENODATA. We should just allocate like
> before tested just now:
> 
> user1 at localhost:~$ sudo ls -al /sys/kernel/
> total 0
> drwxr-xr-x  17 root root    0 Jul  2 13:41 .
> dr-xr-xr-x  12 root root    0 Jul  2 13:41 ..
> -r--r--r--   1 root root 4096 Jul  2 13:41 address_bits
> drwxr-xr-x   3 root root    0 Jul  2 13:41 boot_params
> drwxr-xr-x   2 root root    0 Jul  2 13:41 btf
> drwxr-xr-x   2 root root    0 Jul  2 13:41 cgroup
> drwxr-xr-x   2 root root    0 Jul  2 13:41 config
> -r--r--r--   1 root root 4096 Jul  2 13:41 cpu_byteorder
> -r--r--r--   1 root root 4096 Jul  2 13:41 crash_elfcorehdr_size
> drwx------  34 root root    0 Jul  2 13:41 debug
> -r--r--r--   1 root root 4096 Jul  2 13:41 fscaps
> -r--r--r--   1 root root 4096 Jul  2 13:41 hardlockup_count
> drwxr-xr-x   2 root root    0 Jul  2 13:41 iommu_groups
> drwxr-xr-x 344 root root    0 Jul  2 13:41 irq
> -r--r--r--   1 root root 4096 Jul  2 13:41 kexec_crash_loaded
> -rw-r--r--   1 root root 4096 Jul  2 13:41 kexec_crash_size
> -r--r--r--   1 root root 4096 Jul  2 13:41 kexec_loaded
> drwxr-xr-x   9 root root    0 Jul  2 13:41 mm
> -r--r--r--   1 root root   84 Jul  2 13:41 notes
> -r--r--r--   1 root root 4096 Jul  2 13:41 oops_count
> -rw-r--r--   1 root root 4096 Jul  2 13:41 profiling
> -rw-r--r--   1 root root 4096 Jul  2 13:41 rcu_expedited
> -rw-r--r--   1 root root 4096 Jul  2 13:41 rcu_normal
> -r--r--r--   1 root root 4096 Jul  2 13:41 rcu_stall_count
> drwxr-xr-x   2 root root    0 Jul  2 13:41 reboot
> drwxr-xr-x   2 root root    0 Jul  2 13:41 sched_ext
> drwxr-xr-x   4 root root    0 Jul  2 13:41 security
> drwxr-xr-x 190 root root    0 Jul  2 13:41 slab
> -r--r--r--   1 root root 4096 Jul  2 13:41 softlockup_count
> drwxr-xr-x   2 root root    0 Jul  2 13:41 software_nodes
> drwxr-xr-x   4 root root    0 Jul  2 13:41 sunrpc
> drwxr-xr-x   6 root root    0 Jul  2 13:41 tracing
> -r--r--r--   1 root root 4096 Jul  2 13:41 uevent_seqnum
> -r--r--r--   1 root root 4096 Jul  2 13:41 vmcoreinfo
> -r--r--r--   1 root root 4096 Jul  2 13:41 warn_count
> 
> I'm folding:
> 
> diff --git a/fs/kernfs/inode.c b/fs/kernfs/inode.c
> index 3c293a5a21b1..457f91c412d4 100644
> --- a/fs/kernfs/inode.c
> +++ b/fs/kernfs/inode.c
> @@ -142,9 +142,9 @@ ssize_t kernfs_iop_listxattr(struct dentry *dentry, char *buf, size_t size)
>         struct kernfs_node *kn = kernfs_dentry_node(dentry);
>         struct kernfs_iattrs *attrs;
> 
> -       attrs = kernfs_iattrs_noalloc(kn);
> +       attrs = kernfs_iattrs(kn);
>         if (!attrs)
> -               return -ENODATA;
> +               return -ENOMEM;
> 
>         return simple_xattr_list(d_inode(dentry), &attrs->xattrs, buf, size);
>  }
> 
> which brings it back to the old behavior.
> 

...but it looks like v3 was merged as-is in the end, without this fixup.
Is there some separate patch in the pipeline, or was this forgotten?

> I'm also adding a selftest for this behavior. Patch appended.

Jan

-- 
Siemens AG, Foundational Technologies
Linux Expert Center