Re: [PATCH 2/2] fix up new filp allocators

[Dave Hansen <haveblue@us.ibm.com>]

On Fri, 2007-11-09 at 16:26 -0500, Trond Myklebust wrote:
> >  #include <linux/sunrpc/svc.h>
> >  #include <linux/nfsd/nfsd.h>
> >  #include <linux/nfsd/cache.h>
> > +#include <linux/file.h>
> >  #include <linux/mount.h>
> >  #include <linux/workqueue.h>
> >  #include <linux/smp_lock.h>
> > @@ -1303,7 +1304,7 @@ static inline void
> >  nfs4_file_downgrade(struct file *filp, unsigned int share_access)
> >  {
> >  	if (share_access & NFS4_SHARE_ACCESS_WRITE) {
> > -		put_write_access(filp->f_path.dentry->d_inode);
> > +		drop_file_write_access(filp);
> >  		filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
> >  	}
> >  }
> 
> Hmm... The NFS server may also try to 'upgrade' an open file request to
> a read/write request whenever the client issues a new OPEN request for
> WRITE using the same open_owner.

Can you point me to some code?  I'll try and go fix it up.

> > diff -puN fs/pipe.c~fix-idmapd-bug fs/pipe.c
> > --- linux-2.6.git/fs/pipe.c~fix-idmapd-bug	2007-11-09 12:13:15.000000000 -0800
> > +++ linux-2.6.git-dave/fs/pipe.c	2007-11-09 12:13:15.000000000 -0800
> > @@ -959,13 +959,10 @@ struct file *create_write_pipe(void)
> >  	struct dentry *dentry;
> >  	struct qstr name = { .name = "" };
> >  
> > -	f = get_empty_filp();
> > -	if (!f)
> > -		return ERR_PTR(-ENFILE);
> >  	err = -ENFILE;
> >  	inode = get_pipe_inode();
> >  	if (!inode)
> > -		goto err_file;
> > +		goto err;
> >  
> >  	err = -ENOMEM;
> >  	dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &name);
> > @@ -980,13 +977,14 @@ struct file *create_write_pipe(void)
> >  	 */
> >  	dentry->d_flags &= ~DCACHE_UNHASHED;
> >  	d_instantiate(dentry, inode);
> > -	f->f_path.mnt = mntget(pipe_mnt);
> > -	f->f_path.dentry = dentry;
> > +
> > +	f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_pipe_fops);
> > +	err = -ENFILE;
> > +	if (!f)
> > +		goto err_inode;
> >  	f->f_mapping = inode->i_mapping;
> >  
> >  	f->f_flags = O_WRONLY;
> > -	f->f_op = &write_pipe_fops;
> > -	f->f_mode = FMODE_WRITE;
> >  	f->f_version = 0;
> >  
> >  	return f;
> > @@ -994,8 +992,7 @@ struct file *create_write_pipe(void)
> >   err_inode:
> >  	free_pipe_info(inode);
> >  	iput(inode);
> > - err_file:
> > -	put_filp(f);
> > + err:
> >  	return ERR_PTR(err);
> >  }
> 
> Does RPC idmapd open a regular pipe too? I thought it only used
> rpc_pipefs?

No.  I the the actual bug triggered by idmapd was from an eventpoll filp
that had been acquired through anon_inode_getfd().  I just noticed the
pipe code here could be another potential bug of the same type.

-- Dave

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/