Re: Kernel bug in dcache.h:266; 2.5.64, EIP at sysfs_remove_dir

[Patrick Mochel]

On Sun, 9 Mar 2003, Patrick Mochel wrote:

> 
> On Sun, 9 Mar 2003, Martin J. Bligh wrote:
> 
> > Look for akpm's latest -mm tree release notes - the patch is embedded
> > in there.
> 
> Actually, sysfs is officially fscked in 2.5.64. I'm looking into it, so 
> more reports of it crashing are not necessary. :) I apologize for the 
> inconveniences this has caused.

I was able to reproduce the Oops with a USB device on multiple insert/
removals. The patch below fixes the Oops for me. Could people who have 
seen the Oops try it out and let me know if it helps them? 

[ Unfortunately, I can't test some of the exact failure scenarios, as I 
don't use ppp, and my one system with PCMCIA has decided that it doesn't 
want to let me (physically) insert cards anymore.. ]

Thanks,

	-pat

===== fs/sysfs/dir.c 1.4 vs edited =====
--- 1.4/fs/sysfs/dir.c	Sat Mar  8 23:42:32 2003
+++ edited/fs/sysfs/dir.c	Sun Mar  9 14:25:26 2003
@@ -98,7 +98,6 @@
 			 * Unlink and unhash.
 			 */
 			spin_unlock(&dcache_lock);
-			d_delete(d);
 			simple_unlink(dentry->d_inode,d);
 			dput(d);
 			spin_lock(&dcache_lock);
===== fs/sysfs/inode.c 1.83 vs edited =====
--- 1.83/fs/sysfs/inode.c	Mon Mar  3 17:11:29 2003
+++ edited/fs/sysfs/inode.c	Sun Mar  9 14:25:45 2003
@@ -93,19 +93,14 @@
 		/* make sure dentry is really there */
 		if (victim->d_inode && 
 		    (victim->d_parent->d_inode == dir->d_inode)) {
-			simple_unlink(dir->d_inode,victim);
-			d_delete(victim);
-
 			pr_debug("sysfs: Removing %s (%d)\n", victim->d_name.name,
 				 atomic_read(&victim->d_count));
-			/*
-			 * Drop reference from initial sysfs_get_dentry().
-			 */
-			dput(victim);
+
+			simple_unlink(dir->d_inode,victim);
+
 		}
-		
-		/**
-		 * Drop the reference acquired from sysfs_get_dentry() above.
+		/*
+		 * Drop reference from sysfs_get_dentry() above.
 		 */
 		dput(victim);
 	}