Re: [PATCH 21/34] kernfs, sysfs, cgroup, intel_rdt: Support fs_context [ver #12]

[Andrei Vagin <avagin@gmail.com>]

On Sun, Nov 18, 2018 at 08:23:42PM -0800, Andrei Vagin wrote:
> On Fri, Sep 21, 2018 at 05:33:01PM +0100, David Howells wrote:
> > @@ -1993,57 +2009,53 @@ int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask, int ref_flags)
> >  	return ret;
> >  }
> >  
> > -struct dentry *cgroup_do_mount(struct file_system_type *fs_type, int flags,
> > -			       struct cgroup_root *root, unsigned long magic,
> > -			       struct cgroup_namespace *ns)
> > +int cgroup_do_get_tree(struct fs_context *fc)
> >  {
> > -	struct dentry *dentry;
> > -	bool new_sb;
> > +	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
> > +	int ret;
> >  
> > -	dentry = kernfs_mount(fs_type, flags, root->kf_root, magic, &new_sb);
> > +	ctx->kfc.root = ctx->root->kf_root;
> > +
> > +	ret = kernfs_get_tree(fc);
> > +	if (ret < 0)
> > +		goto out_cgrp;
> >  
> >  	/*
> >  	 * In non-init cgroup namespace, instead of root cgroup's dentry,
> >  	 * we return the dentry corresponding to the cgroupns->root_cgrp.
> >  	 */
> > -	if (!IS_ERR(dentry) && ns != &init_cgroup_ns) {
> > +	if (ctx->ns != &init_cgroup_ns) {
> >  		struct dentry *nsdentry;
> >  		struct cgroup *cgrp;
> >  
> >  		mutex_lock(&cgroup_mutex);
> >  		spin_lock_irq(&css_set_lock);
> >  
> > -		cgrp = cset_cgroup_from_root(ns->root_cset, root);
> > +		cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root);
> >  
> >  		spin_unlock_irq(&css_set_lock);
> >  		mutex_unlock(&cgroup_mutex);
> >  
> > -		nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb);
> > -		dput(dentry);
> > -		dentry = nsdentry;
> > +		nsdentry = kernfs_node_dentry(cgrp->kn, fc->root->d_sb);
> > +		if (IS_ERR(nsdentry))
> > +			return PTR_ERR(nsdentry);
> > +		dput(fc->root);
> > +		fc->root = nsdentry;
> >  	}
> >  
> > -	if (IS_ERR(dentry) || !new_sb)
> > -		cgroup_put(&root->cgrp);
> 
> I don't see where this cgroup_put() has been moved.

David, have you looked at this problem? It isn't fixed in linux-next
yet.

https://travis-ci.org/avagin/linux/jobs/463960763

Thanks,
Andrei

> 
> With this patch, the next script works only once, on the second attempt
> it hangs up on mounting a cgroup file system.
> 
> This is the only suspicious place in this patch what I have found.
> 
> [root@fc24 ~]# cat fs-vs-cg 
> d=$(mktemp -d /tmp/cg.XXXXXX)
> mkdir $d/a
> mkdir $d/b
> mount -t cgroup -o none,name=xxxx xxx $d/a
> mount -t cgroup -o none,name=xxxx xxx $d/b
> umount $d/a
> umount $d/b
> 
> [root@fc24 ~]# unshare -m --propagation private bash -x fs-vs-cg
> ++ mktemp -d /tmp/cg.XXXXXX
> + d=/tmp/cg.yUfagS
> + mkdir /tmp/cg.yUfagS/a
> + mkdir /tmp/cg.yUfagS/b
> + mount -t cgroup -o none,name=xxxx xxx /tmp/cg.yUfagS/a
> + mount -t cgroup -o none,name=xxxx xxx /tmp/cg.yUfagS/b
> + umount /tmp/cg.yUfagS/a
> + umount /tmp/cg.yUfagS/b
> [root@fc24 ~]# unshare -m --propagation private bash -x fs-vs-cg
> ++ mktemp -d /tmp/cg.XXXXXX
> + d=/tmp/cg.ippWUn
> + mkdir /tmp/cg.ippWUn/a
> + mkdir /tmp/cg.ippWUn/b
> + mount -t cgroup -o none,name=xxxx xxx /tmp/cg.ippWUn/a
> ^Z
> [1]+  Stopped                 unshare -m --propagation private bash -x fs-vs-cg
> 
> [root@fc24 ~]# ps
>   PID TTY          TIME CMD
>   556 pts/0    00:00:00 bash
>   591 pts/0    00:00:00 bash
>   595 pts/0    00:00:00 mount
>   596 pts/0    00:00:00 ps
> 
> [root@fc24 ~]# bg
> [1]+ unshare -m --propagation private bash -x fs-vs-cg &
> 
> [root@fc24 ~]# cat /proc/595/stack 
> [<0>] msleep+0x38/0x40
> [<0>] cgroup1_get_tree+0x4e1/0x72c
> [<0>] vfs_get_tree+0x5e/0x140
> [<0>] do_mount+0x326/0xc70
> [<0>] ksys_mount+0xba/0xd0
> [<0>] __x64_sys_mount+0x21/0x30
> [<0>] do_syscall_64+0x60/0x210
> [<0>] entry_SYSCALL_64_after_hwframe+0x49/0xbe
> [<0>] 0xffffffffffffffff
>