[PATCH -next] proc: fix task_struct infoleak

[PATCH -next] proc: fix task_struct infoleak

[Vasiliy Kulikov]

proc_pid_permission() doesn't put task_struct on every /proc/$pid/
access.  A demo from Hugh Dickins:

while :; do ps; grep KernelStack /proc/meminfo; sleep 1; done

Reported-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Vasiliy Kulikov <segoon@openwall.com>
---
 This is a patch against a hidepid patchset from -mm.

 fs/proc/base.c |    9 +++++++--
 1 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/fs/proc/base.c b/fs/proc/base.c
index 8caf5cb..0e5c577 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -646,9 +646,14 @@ static bool has_pid_permissions(struct pid_namespace *pid,
 static int proc_pid_permission(struct inode *inode, int mask)
 {
 	struct pid_namespace *pid = inode->i_sb->s_fs_info;
-	struct task_struct *task = get_proc_task(inode);
+	struct task_struct *task;
+	int has_perms;
+
+	task = get_proc_task(inode);
+	has_perms = has_pid_permissions(pid, task, 1);
+	put_task_struct(task);
 
-	if (!has_pid_permissions(pid, task, 1)) {
+	if (!has_perms) {
 		if (pid->hide_pid == 2) {
 			/*
 			 * Let's make getdents(), stat(), and open()
-- 
1.7.0.4

Re: [PATCH -next] proc: fix task_struct infoleak

[Al Viro]

On Sun, Dec 11, 2011 at 10:28:21PM +0400, Vasiliy Kulikov wrote:
> proc_pid_permission() doesn't put task_struct on every /proc/$pid/
> access.  A demo from Hugh Dickins:
> 
> while :; do ps; grep KernelStack /proc/meminfo; sleep 1; done
> 
> Reported-by: Hugh Dickins <hughd@google.com>
> Signed-off-by: Vasiliy Kulikov <segoon@openwall.com>
> ---
>  This is a patch against a hidepid patchset from -mm.

Choose saner commit summary, please.  It's not information leak, for pity sake
- it's a plain and simple memory leak...

Speaking of which, I've a couple of memory leaks in mainline procfs and
mqueue; Alexey, are you OK with that sucker going directly to Linus or
would you rather push it yourself?  See the patch below...

diff --git a/fs/proc/root.c b/fs/proc/root.c
index 9a8a2b7..03102d9 100644
--- a/fs/proc/root.c
+++ b/fs/proc/root.c
@@ -91,20 +91,18 @@ static struct file_system_type proc_fs_type = {
 
 void __init proc_root_init(void)
 {
-	struct vfsmount *mnt;
 	int err;
 
 	proc_init_inodecache();
 	err = register_filesystem(&proc_fs_type);
 	if (err)
 		return;
-	mnt = kern_mount_data(&proc_fs_type, &init_pid_ns);
-	if (IS_ERR(mnt)) {
+	err = pid_ns_prepare_proc(&init_pid_ns);
+	if (err) {
 		unregister_filesystem(&proc_fs_type);
 		return;
 	}
 
-	init_pid_ns.proc_mnt = mnt;
 	proc_symlink("mounts", NULL, "self/mounts");
 
 	proc_net_init();
@@ -209,5 +207,5 @@ int pid_ns_prepare_proc(struct pid_namespace *ns)
 
 void pid_ns_release_proc(struct pid_namespace *ns)
 {
-	mntput(ns->proc_mnt);
+	kern_unmount(ns->proc_mnt);
 }

Re: [PATCH -next] proc: fix task_struct infoleak

[Hugh Dickins]

On Sun, 11 Dec 2011, Vasiliy Kulikov wrote:

> proc_pid_permission() doesn't put task_struct on every /proc/$pid/
> access.  A demo from Hugh Dickins:
> 
> while :; do ps; grep KernelStack /proc/meminfo; sleep 1; done
> 
> Reported-by: Hugh Dickins <hughd@google.com>
> Signed-off-by: Vasiliy Kulikov <segoon@openwall.com>

Thank you, yes, that fixes it for me:
Tested-by: Hugh Dickins <hughd@google.com>
but hopefully akpm will just fold this into the original patch.

Hmm, tiny niggle, but I notice that has_pid_permissions() is
declared, reasonably, as bool; so wouldn't it be better for
has_perms to be declared as bool instead of int?  Hardly
worth bothering about, but if you've a mind to respin with
a corrected Subject as Al rightly suggests.

> ---
>  This is a patch against a hidepid patchset from -mm.
> 
>  fs/proc/base.c |    9 +++++++--
>  1 files changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 8caf5cb..0e5c577 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -646,9 +646,14 @@ static bool has_pid_permissions(struct pid_namespace *pid,
>  static int proc_pid_permission(struct inode *inode, int mask)
>  {
>  	struct pid_namespace *pid = inode->i_sb->s_fs_info;
> -	struct task_struct *task = get_proc_task(inode);
> +	struct task_struct *task;
> +	int has_perms;
> +
> +	task = get_proc_task(inode);
> +	has_perms = has_pid_permissions(pid, task, 1);
> +	put_task_struct(task);
>  
> -	if (!has_pid_permissions(pid, task, 1)) {
> +	if (!has_perms) {
>  		if (pid->hide_pid == 2) {
>  			/*
>  			 * Let's make getdents(), stat(), and open()
> -- 
> 1.7.0.4

Re: [PATCH -next] proc: fix task_struct infoleak

[Vasiliy Kulikov]

On Sun, Dec 11, 2011 at 20:59 +0000, Al Viro wrote:
> On Sun, Dec 11, 2011 at 10:28:21PM +0400, Vasiliy Kulikov wrote:
> > proc_pid_permission() doesn't put task_struct on every /proc/$pid/
> > access.  A demo from Hugh Dickins:
> > 
> > while :; do ps; grep KernelStack /proc/meminfo; sleep 1; done
> > 
> > Reported-by: Hugh Dickins <hughd@google.com>
> > Signed-off-by: Vasiliy Kulikov <segoon@openwall.com>
> > ---
> >  This is a patch against a hidepid patchset from -mm.
> 
> Choose saner commit summary, please.  It's not information leak, for pity sake
> - it's a plain and simple memory leak...

Oops!  s/infoleak/memleak/ surely, bad habit...

Thank you, I'll resend the patch.


-- 
Vasiliy Kulikov
http://www.openwall.com - bringing security into open computing environments