Re: Re: md:When opened /proc/mdstat, increase the refcount of

["majianpeng" <majianpeng@gmail.com>]

my patch is:
---
 fs/proc/inode.c |   13 +++++++++++--
 1 files changed, 11 insertions(+), 2 deletions(-)

diff --git a/fs/proc/inode.c b/fs/proc/inode.c
index 84fd323..78dbb03 100644
--- a/fs/proc/inode.c
+++ b/fs/proc/inode.c
@@ -317,6 +317,7 @@ static int proc_reg_open(struct inode *inode, struct file *file)
 	int (*open)(struct inode *, struct file *);
 	int (*release)(struct inode *, struct file *);
 	struct pde_opener *pdeo;
+	const struct file_operations *fops;
 
 	/*
 	 * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
@@ -339,7 +340,12 @@ static int proc_reg_open(struct inode *inode, struct file *file)
 		return -ENOENT;
 	}
 	pde->pde_users++;
-	open = pde->proc_fops->open;
+	/*
+	 *If proc_fops defined in module and used .owner = THIS_MODULE,so open must
+	 *get a reference of module.
+	 */
+	fops = fops_get(pde->proc_fops);
+	open = fops->open;
 	release = pde->proc_fops->release;
 	spin_unlock(&pde->pde_unload_lock);
 
@@ -354,8 +360,10 @@ static int proc_reg_open(struct inode *inode, struct file *file)
 		/* Strictly for "too late" ->release in proc_reg_release(). */
 		pdeo->release = release;
 		list_add(&pdeo->lh, &pde->pde_openers);
-	} else
+	} else	{
+		fops_put(pde->proc_fops);
 		kfree(pdeo);
+	}
 	__pde_users_dec(pde);
 	spin_unlock(&pde->pde_unload_lock);
 	return rv;
@@ -402,6 +410,7 @@ static int proc_reg_release(struct inode *inode, struct file *file)
 	}
 	pde->pde_users++;
 	release = pde->proc_fops->release;
+	fops_put(pde->proc_fops);
 	if (pdeo) {
 		list_del(&pdeo->lh);
 		kfree(pdeo);

I tested ok,but Al Viro said deadlock, I did not tested.

------------------				 
majianpeng
2012-02-27

-------------------------------------------------------------
发件人：NeilBrown
发送日期：2012-02-26 19:01:32
收件人：majianpeng
抄送：linux-raid
主题：Re: md:When opened /proc/mdstat, increase the refcount of

On Sun, 26 Feb 2012 16:33:25 +0800 "majianpeng" <majianpeng@gmail.com> wrote:

> I wrote a letter to maintainer of procfs. He's answer is follow:
> >No.  Read remove_proc_entry(), especially the part under ->pde_unload_lock.
> >The whole damn point of that stuff is that opened file on procfs does *not*
> >pin the module down; IO in progress does, but that's it.
> >You can't deadlock rmmod foobar </proc/crap/foobar; with your patch we'll
> >be back to a pile of deadlocks in there.
> >IOW, NAK.

Hmmm... that sounds like Al Viro.

I see the point of that code now.  And your patch would be open to the same
problem wouldn't it. i.e.
    rmmod md_mod < /proc/mdstat

would deadlock??

So we still need to find a fix that actually works correctly.

That probably means moving the call to 'poll_wait' into procfs code, and using
a wait_queue_head which is managed by procfs.

So maybe:
 - put a wait_queue_head in 'struct proc_dir_entry'
 - have proc_reg_poll call poll_wait passing that wait_queue_head
 - write a function e.g. proc_poll_wake which calls wake_up on that
   wait_queue_head
 - have md.c save the return value from proc_create, and call the above
   function instead of calling "wake_up(&md_event_waiters)"

That should work.

NeilBrown