On Sun, Jan 26, 2025 at 04:54:55PM +0100, Alexander Gordeev wrote:
> > > Since the version next-20250120 [2], we are seeing the following regression
> >
> > Ugh... To narrow the things down, could you see if replacing
> > fsd = kmalloc(sizeof(*fsd), GFP_KERNEL);
> > with
> > fsd = kzalloc(sizeof(*fsd), GFP_KERNEL);
> > in fs/debugfs/file.c:__debugfs_file_get() affects the test?
>
> This change fixes lots of the below failures in our CI. FWIW:
>
> Tested-by: Alexander Gordeev <agordeev@xxxxxxxxxxxxx>
The real fix follows:
[PATCH] Fix the missing initializations in __debugfs_file_get()
both method table pointers in debugfs_fsdata need to be initialized,
obviously...
Fixes: 41a0ecc0997c "debugfs: get rid of dynamically allocation proxy_ops"
Fucked-up-by: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
Signed-off-by: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
---
diff --git a/fs/debugfs/file.c b/fs/debugfs/file.c
index e33cc77699cd..212cd8128e1f 100644
--- a/fs/debugfs/file.c
+++ b/fs/debugfs/file.c
@@ -111,6 +111,7 @@ static int __debugfs_file_get(struct dentry *dentry, enum dbgfs_get_mode mode)
fsd->methods |= HAS_READ;
if (ops->write)
fsd->methods |= HAS_WRITE;
+ fsd->real_fops = NULL;
} else {
const struct file_operations *ops;
ops = fsd->real_fops = DEBUGFS_I(inode)->real_fops;
@@ -124,6 +125,7 @@ static int __debugfs_file_get(struct dentry *dentry, enum dbgfs_get_mode mode)
fsd->methods |= HAS_IOCTL;
if (ops->poll)
fsd->methods |= HAS_POLL;
+ fsd->short_fops = NULL;
}
refcount_set(&fsd->active_users, 1);
init_completion(&fsd->active_users_drained);
