fs: sysfs: do not remove files if group is null
PATCH 1/1 fs: sysfs: do not remove files if group is null
The current implementation allows the remove_files() API to be called without checking if the grp->name is null. Ensure that the group name is valid prior to removing files.
This patch fixes a race condition where device_del() will cleanup sysfs entries prior to device managed sysfs entries. This results in a NULL group->name and a system error during device cleanup.
To reproduce the issue, simply create a new child device in a platform driver of your choice. Add a sysfs file group using devm API. On driver exist ensure to unregister your child device. Do not call devm_device_remove_group() and leave it up to the implementation to automatically clean up the files. Here is where you will see a kernel error complaining that the files have already been removed.
--- fs/sysfs/group.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
fs/sysfs/group.c +3 -2
@@ -286,9 +286,10 @@ void sysfs_remove_group(struct kobject *kobj,kernfs_get (kn); } -remove_files (kn, grp); -if (grp->name) +if (grp->name) { +remove_files (kn, grp);kernfs_remove (kn); + }kernfs_put (kn); }
Signature · 1 line
2.25.1
Greg KH
maintainerQuoted 11 lines · Daniel Steger
The current implementation allows the
remove_files()API to be called without checking if thegrp->nameis null. Ensure that the group name is valid prior to removing files.This patch fixes a race condition where
device_del()will cleanup sysfs entries prior to device managed sysfs entries. This results in a NULLgroup->nameand a system error during device cleanup.To reproduce the issue, simply create a new child device in a platform driver of your choice. Add a sysfs file group using devm API.
What driver is doing this today?
Quoted 1 line
On driver exist ensure to unregister your child device.
What child device? Why is a platform driver creating a child device at all?
Quoted 15 lines
Do not call
devm_device_remove_group()and leave it up to the implementation to automatically clean up the files. Here is where you will see a kernel error complaining that the files have already been removed.Signed-off-by Daniel Steger <[email protected]>--- fs/sysfs/group.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)fs/sysfs/group.c +0
-0
fs/sysfs/group.c void sysfs_remove_group(struct kobject *kobj, · line 290
Quoted 7 lines
fs/sysfs/group.c +2
-2@@ -286,9 +286,10 @@ void sysfs_remove_group(struct kobject *kobj,kernfs_get (kn); } -remove_files (kn, grp); -if (grp->name) +if (grp->name) { +remove_files (kn, grp);
What about groups without names? Will then now not be removed properly? Why does the name matter here?
thanks,
greg k-h