linux-kernel - Re: [PATCH] KVM: release anon file in failure path of vm creation

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Thu, 14 Jul 2016 17:46:47 +0100
From:	Al Viro <viro@...IV.linux.org.uk>
To:	Paolo Bonzini <pbonzini@...hat.com>
Cc:	Liu Shuo <shuo.a.liu@...el.com>, linux-kernel@...r.kernel.org,
	kvm@...r.kernel.org, Zhang Yanmin <yanmin_zhang@...ux.intel.com>,
	He Bo <bo.he@...el.com>, Liu Shuo <shuox.liu@...il.com>,
	Radim Krčmář <rkrcmar@...hat.com>
Subject: Re: [PATCH] KVM: release anon file in failure path of vm creation

On Tue, Jul 12, 2016 at 12:24:39PM +0200, Paolo Bonzini wrote:

> On 12/07/2016 11:38, Liu Shuo wrote:
> > The failure of create debugfs of VM will return directly without release
> > the anon file. It will leak memory and file descriptors, even through
> > be not serious.
> > 
> > Signed-off-by: Liu Shuo <shuo.a.liu@...el.com>

This is broken.

> > @@ -3067,6 +3068,7 @@ static int kvm_dev_ioctl_create_vm(unsigned long type)
> >  
> >  	if (kvm_create_vm_debugfs(kvm, r) < 0) {
> >  		kvm_put_kvm(kvm);
> > +		sys_close(r);

You have no warranty whatsoever that descriptor table has not been changed
by that point.  You should *NEVER* use sys_close() on failure exit paths
like that.  Moreover, this kvm_put_kvm() becomes a double-put, since
closing the damn file will drop that reference to kvm.

> >  		return -ENOMEM;
> >  	}
> >  
> > 
> 
> Thanks, applied to kvm/master.

Please, revert.  anon_inode_getfd() should be used only when there's no
possible failures past its call.  What you need in this case (due to fucking
ugly API - decriptor number used as part of debugfs pathname) is something
like this (against mainline):

[kvm] don't use anon_inode_getfd() before possible failures

Once anon_inode_getfd() has succeeded, it's impossible to undo in a clean
way and no, sys_close() is not usable in such cases.  Use anon_inode_getfile()
and get_unused_fd_flags() to get struct file and descriptor and do *not*
install the file into descriptor table until after the last possible failure
exit.

Signed-off-by: Al Viro <viro@...iv.linux.org.uk>
---
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 48bd520..7f82c6c 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -3048,6 +3048,7 @@ static int kvm_dev_ioctl_create_vm(unsigned long type)
 {
 	int r;
 	struct kvm *kvm;
+	struct file *file;
 
 	kvm = kvm_create_vm(type);
 	if (IS_ERR(kvm))
@@ -3059,17 +3060,25 @@ static int kvm_dev_ioctl_create_vm(unsigned long type)
 		return r;
 	}
 #endif
-	r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
+	r = get_unused_fd_flags(O_CLOEXEC);
 	if (r < 0) {
 		kvm_put_kvm(kvm);
 		return r;
 	}
+	file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
+	if (IS_ERR(file)) {
+		put_unused_fd(r);
+		kvm_put_kvm(kvm);
+		return PTR_ERR(file);
+	}
 
 	if (kvm_create_vm_debugfs(kvm, r) < 0) {
-		kvm_put_kvm(kvm);
+		put_unused_fd(r);
+		fput(file);
 		return -ENOMEM;
 	}
 
+	fd_install(r, file);
 	return r;
 }