fork: normalize copy_process() error return before ERR_PTR().

copy_process() returns ERR_PTR(retval) from its error path, so retval
must be a negative errno. If retval is zero or positive, ERR_PTR(retval)
produces a non-error pointer that is not caught by IS_ERR() in
kernel_clone().

A BPF_MODIFY_RETURN program attached to security_task_alloc() can return
a positive value. copy_process() treats the non-zero return as a failure
and then returns ERR_PTR(1). kernel_clone() does not treat that as an
error and later dereferences the pointer, causing a kernel crash.

Normalize unexpected non-negative values before returning ERR_PTR() from
copy_process(). This keeps the fix local to the fork error path and does
not change BPF_MODIFY_RETURN verifier behavior.

The issue has been reported and discussed upstream, but the verifier-side
fix attempt has not been accepted. Carry this targeted fix in deepin-kernel
to prevent the reported denial-of-service.

Link: https://lore.kernel.org/bpf/973a1b7b-8ee7-407a-890a-11455d9cc5bf@std.uestc.edu.cn/
Link: https://lore.kernel.org/all/20260411163556.8567-1-yangfeng59949@163.com/
Reported-by: Quan Sun <2022090917019@std.uestc.edu.cn>
Reported-by: Yinhao Hu <dddddd@hust.edu.cn>
Reported-by: Kaiyan Mei <M202472210@hust.edu.cn>
Signed-off-by: hushijia <hushijia1@uniontech.com>

Author: husjdodoanthing

kernel/fork.c

@@ -2512,6 +2512,13 @@ __latent_entropy struct task_struct *copy_process(
 	spin_lock_irq(&current->sighand->siglock);
 	hlist_del_init(&delayed.node);
 	spin_unlock_irq(&current->sighand->siglock);
+	/*
+	 * The error path returns ERR_PTR(retval), which requires a negative
+	 * errno. Normalize unexpected values to avoid returning non-error
+	 * pointers to callers.
+	 */
+	if (unlikely(retval >= 0))
+		retval = -EINVAL;
 	return ERR_PTR(retval);
 }