Re: sysfs: duplicate filename '74' can not be created

* Re: sysfs: duplicate filename '74' can not be created
@ 2008-03-04 16:09 Alexey Dobriyan
  2008-03-04 16:33 ` Dhaval Giani
  2008-03-07  7:24 ` Holger Kiehl
  0 siblings, 2 replies; 15+ messages in thread
From: Alexey Dobriyan @ 2008-03-04 16:09 UTC (permalink / raw)
  To: Holger Kiehl; +Cc: linux-kernel, dhaval, vatsa, mingo

[Cc'ing /sys/kernel/uids guys]

> Mar  3 13:50:01 helena kernel: sysfs: duplicate filename '74' can not be created
> Mar  3 13:50:01 helena kernel: WARNING: at fs/sysfs/dir.c:424 sysfs_add_one()
> Mar  3 13:50:01 helena kernel: Pid: 21911, comm: sshd Not tainted 2.6.24.3 #1

I see only one suspicious place: KOBJ_ADD is sent even if sysfs file
creation failed and kobject is happily removed. But this almost
certainly separate bug.



You can try this debugging patch. Post several first consecutive messages.
[crosses fingers]

--- a/kernel/user.c
+++ b/kernel/user.c
@@ -179,8 +179,11 @@ static int user_kobject_create(struct user_struct *up)
 		goto done;
 
 	error = sysfs_create_file(kobj, &up->user_attr.attr);
-	if (error)
+	if (error) {
+		printk("%s: %p->uid = %d\n", __func__, up, up->uid);
+		WARN_ON(1);
 		kobject_del(kobj);
+	}
 
 	kobject_uevent(kobj, KOBJ_ADD);
 
@@ -363,6 +366,7 @@ struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid)
 		}
 
 		if (sched_create_user(new) < 0) {
+			WARN_ON(1);
 			key_put(new->uid_keyring);
 			key_put(new->session_keyring);
 			kmem_cache_free(uid_cachep, new);
@@ -371,6 +375,7 @@ struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid)
 		}
 
 		if (user_kobject_create(new)) {
+			WARN_ON(1);
 			sched_destroy_user(new);
 			key_put(new->uid_keyring);
 			key_put(new->session_keyring);
@@ -386,6 +391,10 @@ struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid)
 		spin_lock_irq(&uidhash_lock);
 		up = uid_hash_find(uid, hashent);
 		if (up) {
+#ifdef CONFIG_FAIR_USER_SCHED
+			printk("%s: %p->uid = %d\n", __func__, up, up->uid);
+			WARN_ON(1);
+#endif
 			/* This case is not possible when CONFIG_FAIR_USER_SCHED
 			 * is defined, since we serialize alloc_uid() using
 			 * uids_mutex. Hence no need to call

diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
index 7af90fc..3479c55 100644
--- a/kernel/user_namespace.c
+++ b/kernel/user_namespace.c
@@ -32,6 +32,8 @@ static struct user_namespace *clone_user_ns(struct user_namespace *old_ns)
 	struct user_struct *new_user;
 	int n;
 
+	printk("%s: ENTER\n", __func__);
+
 	ns = kmalloc(sizeof(struct user_namespace), GFP_KERNEL);
 	if (!ns)
 		return ERR_PTR(-ENOMEM);


^ permalink raw reply related	[flat|nested] 15+ messages in thread