Question on release_one_tty

Question on release_one_tty

[Cyrill Gorcunov]

Hi guys, could you please explain me the sequence

static void release_one_tty(struct work_struct *work)
{
	struct tty_struct *tty =
		container_of(work, struct tty_struct, hangup_work);
	struct tty_driver *driver = tty->driver;

	if (tty->ops->cleanup)
		tty->ops->cleanup(tty);

	tty->magic = 0;
-->	tty_driver_kref_put(driver);
-->	module_put(driver->owner);

why tty_driver_kref_put is called before module_put? As far as I understand
tty_driver_kref_put may call the destruct_tty_driver which eventually does

static void destruct_tty_driver(struct kref *kref)
{
	struct tty_driver *driver = container_of(kref, struct tty_driver, kref);
	...
	kfree(driver->cdevs);
	kfree(driver->ports);
	kfree(driver->termios);
	kfree(driver->ttys);
-->	kfree(driver);
}

so that the module_put(driver->owner) would access freed memory. Should not we
call the reverse module_put and then tty_driver_kref_put, or I miss something
obvious?

	Cyrill

Re: Question on release_one_tty

[Pavel Emelyanov]

On 08/07/2014 12:25 PM, Cyrill Gorcunov wrote:
> Hi guys, could you please explain me the sequence
> 
> static void release_one_tty(struct work_struct *work)
> {
> 	struct tty_struct *tty =
> 		container_of(work, struct tty_struct, hangup_work);
> 	struct tty_driver *driver = tty->driver;
> 
> 	if (tty->ops->cleanup)
> 		tty->ops->cleanup(tty);
> 
> 	tty->magic = 0;
> -->	tty_driver_kref_put(driver);
> -->	module_put(driver->owner);
> 
> why tty_driver_kref_put is called before module_put? As far as I understand
> tty_driver_kref_put may call the destruct_tty_driver which eventually does
> 
> static void destruct_tty_driver(struct kref *kref)
> {
> 	struct tty_driver *driver = container_of(kref, struct tty_driver, kref);
> 	...
> 	kfree(driver->cdevs);
> 	kfree(driver->ports);
> 	kfree(driver->termios);
> 	kfree(driver->ttys);
> -->	kfree(driver);
> }
> 
> so that the module_put(driver->owner) would access freed memory. Should not we
> call the reverse module_put and then tty_driver_kref_put, or I miss something
> obvious?

If you put the module it can be unloaded at any time killing the code that would
be potentially required by kref_put.

> 	Cyrill
> .
> 

Re: Question on release_one_tty

[Cyrill Gorcunov]

On Thu, Aug 07, 2014 at 12:28:58PM +0400, Pavel Emelyanov wrote:
> On 08/07/2014 12:25 PM, Cyrill Gorcunov wrote:
> > Hi guys, could you please explain me the sequence
> > 
> > static void release_one_tty(struct work_struct *work)
> > {
> > 	struct tty_struct *tty =
> > 		container_of(work, struct tty_struct, hangup_work);
> > 	struct tty_driver *driver = tty->driver;
> > 
> > 	if (tty->ops->cleanup)
> > 		tty->ops->cleanup(tty);
> > 
> > 	tty->magic = 0;
> > -->	tty_driver_kref_put(driver);
> > -->	module_put(driver->owner);
> > 
> > why tty_driver_kref_put is called before module_put? As far as I understand
> > tty_driver_kref_put may call the destruct_tty_driver which eventually does
> > 
> > static void destruct_tty_driver(struct kref *kref)
> > {
> > 	struct tty_driver *driver = container_of(kref, struct tty_driver, kref);
> > 	...
> > 	kfree(driver->cdevs);
> > 	kfree(driver->ports);
> > 	kfree(driver->termios);
> > 	kfree(driver->ttys);
> > -->	kfree(driver);
> > }
> > 
> > so that the module_put(driver->owner) would access freed memory. Should not we
> > call the reverse module_put and then tty_driver_kref_put, or I miss something
> > obvious?
> 
> If you put the module it can be unloaded at any time killing the code that would
> be potentially required by kref_put.

So how this code supposed to work then? I mean tty_driver_kref_put must never call
for destruct_tty_driver, otherwise we're accessing freed memory.

Re: Question on release_one_tty

[Pavel Emelyanov]

On 08/07/2014 12:34 PM, Cyrill Gorcunov wrote:
> On Thu, Aug 07, 2014 at 12:28:58PM +0400, Pavel Emelyanov wrote:
>> On 08/07/2014 12:25 PM, Cyrill Gorcunov wrote:
>>> Hi guys, could you please explain me the sequence
>>>
>>> static void release_one_tty(struct work_struct *work)
>>> {
>>> 	struct tty_struct *tty =
>>> 		container_of(work, struct tty_struct, hangup_work);
>>> 	struct tty_driver *driver = tty->driver;
>>>
>>> 	if (tty->ops->cleanup)
>>> 		tty->ops->cleanup(tty);
>>>
>>> 	tty->magic = 0;
>>> -->	tty_driver_kref_put(driver);
>>> -->	module_put(driver->owner);
>>>
>>> why tty_driver_kref_put is called before module_put? As far as I understand
>>> tty_driver_kref_put may call the destruct_tty_driver which eventually does
>>>
>>> static void destruct_tty_driver(struct kref *kref)
>>> {
>>> 	struct tty_driver *driver = container_of(kref, struct tty_driver, kref);
>>> 	...
>>> 	kfree(driver->cdevs);
>>> 	kfree(driver->ports);
>>> 	kfree(driver->termios);
>>> 	kfree(driver->ttys);
>>> -->	kfree(driver);
>>> }
>>>
>>> so that the module_put(driver->owner) would access freed memory. Should not we
>>> call the reverse module_put and then tty_driver_kref_put, or I miss something
>>> obvious?
>>
>> If you put the module it can be unloaded at any time killing the code that would
>> be potentially required by kref_put.
> 
> So how this code supposed to work then? I mean tty_driver_kref_put must never call
> for destruct_tty_driver, otherwise we're accessing freed memory.

mod = driver->owner;
tty_driver_kref_put(driver);
module_put(mod);

Check the upstream whether the same issue exists there.

Re: Question on release_one_tty

[Cyrill Gorcunov]

On Thu, Aug 07, 2014 at 01:05:28PM +0400, Pavel Emelyanov wrote:
> > 
> > So how this code supposed to work then? I mean tty_driver_kref_put must never call
> > for destruct_tty_driver, otherwise we're accessing freed memory.
> 
> mod = driver->owner;
> tty_driver_kref_put(driver);
> module_put(mod);
> 
> Check the upstream whether the same issue exists there.

Same in tty.git

static void release_one_tty(struct work_struct *work)
{
	struct tty_struct *tty =
		container_of(work, struct tty_struct, hangup_work);
	struct tty_driver *driver = tty->driver;

	if (tty->ops->cleanup)
		tty->ops->cleanup(tty);

	tty->magic = 0;
	tty_driver_kref_put(driver);
	module_put(driver->owner);

	spin_lock(&tty_files_lock);
	list_del_init(&tty->tty_files);
	spin_unlock(&tty_files_lock);

	put_pid(tty->pgrp);
	put_pid(tty->session);
	free_tty_struct(tty);
}

so I think yes, the fix is needed. I'll prepare the patch
and test/post for review.

[PATCH] tty: Fix potential use after free in release_one_tty

[Cyrill Gorcunov]

On Thu, Aug 07, 2014 at 01:18:12PM +0400, Cyrill Gorcunov wrote:
> > mod = driver->owner;
> > tty_driver_kref_put(driver);
> > module_put(mod);
> > 
> > Check the upstream whether the same issue exists there.
> 
> Same in tty.git
> 
> static void release_one_tty(struct work_struct *work)
> {
---
From: Cyrill Gorcunov <gorcunov@gmail.com>
Subject: tty: Fix potential use after free in release_one_tty

In case if we're releasing the last tty reference the following
call sequence is possible

tty_driver_kref_put
  destruct_tty_driver
    kfree(driver);

where @driver is used in next module_put call, which leads to

 | [ 285.964007] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
 | [ 285.964007] Workqueue: events release_one_tty
 | [ 285.964007] task: ffff8800cc7ea5f0 ti: ffff8800cb800000 task.ti: ffff8800cb800000
 | [ 285.964007] RIP: 0010:[<ffffffff810aeaf5>] [<ffffffff810aeaf5>] module_put+0x24/0xf4
 | [ 285.964007] RSP: 0018:ffff8800cb801d48 EFLAGS: 00010213
 | [ 285.964007] RAX: ffff8800cb801fd8 RBX: ffff8800ca3429d0 RCX: ffff8800cb1db400
 | [ 285.964007] RDX: 0000000000000000 RSI: ffffffff817349c1 RDI: 0000000000000001
 | [ 285.964007] RBP: ffff8800cb801d60 R08: ffff8800cd632b40 R09: 0000000000000000
 | [ 285.964007] R10: 00000000ffffffff R11: ffff88011f40a000 R12: 6b6b6b6b6b6b6b6b
 | [ 285.964007] R13: ffff8800ca342520 R14: 0000000000000000 R15: ffff88011f5d8200
 | [ 285.964007] FS: 0000000000000000(0000) GS:ffff88011f400000(0000) knlGS:0000000000000000
 | [ 285.964007] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b
 | [ 285.964007] CR2: 00007faf5229d090 CR3: 0000000001c0b000 CR4: 00000000000006f0
 | [ 285.964007] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 | [ 285.964007] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
 | [ 285.964007] Stack:
 | [ 285.964007] ffff8800ca3429d0 ffff8800ca342a30 ffff8800ca342520 ffff8800cb801d88
 | [ 285.964007] ffffffff8146554a ffff8800cc77cc78 ffff8800ca3429d0 ffff88011f5d3800
 | [ 285.964007] ffff8800cb801e08 ffffffff810683c1 ffffffff810682ff 0000000000000046
 | [ 285.964007] Call Trace:
 | [ 285.964007] [<ffffffff8146554a>] release_one_tty+0x54/0xa3
 | [ 285.964007] [<ffffffff810683c1>] process_one_work+0x223/0x404
 | [ 285.964007] [<ffffffff810682ff>] ? process_one_work+0x161/0x404
 | [ 285.964007] [<ffffffff81068971>] worker_thread+0x136/0x205
 | [ 285.964007] [<ffffffff8106883b>] ? rescuer_thread+0x26a/0x26a
 | [ 285.964007] [<ffffffff8106e5bf>] kthread+0xa2/0xaa
 | [ 285.964007] [<ffffffff810a4586>] ? trace_hardirqs_on_caller+0x16/0x1eb
 | [ 285.964007] [<ffffffff8106e51d>] ? __kthread_parkme+0x65/0x65
 | [ 285.964007] [<ffffffff8173f59c>] ret_from_fork+0x7c/0xb0
 | [ 285.964007] [<ffffffff8106e51d>] ? __kthread_parkme+0x65/0x65
 | [ 285.964007] Code: 09 00 5b 41 5c 5d c3 0f 1f 44 00 00 55 48 85 ff 48 89 e5 41 55 41 54 49 89 fc 53 0f 84 d3 00
 | 00 00 bf 01 00 00 00 e8 d0 a1 fc ff <49> 8b 84 24 50 02 00 00 65 48 ff 40 08 4c 8b 6d 08 0f 1f 44 00

so simply keep a local reference to the module owner and
use it later.

(Note I hit the bug on custom kernel but I think vanilla one suffers
 the same problem).

CC: Pavel Emelyanov <xemul@parallels.com>
CC: Jiri Slaby <jslaby@suse.cz>
CC: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
---
 drivers/tty/tty_io.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Index: linux-2.6.git/drivers/tty/tty_io.c
===================================================================
--- linux-2.6.git.orig/drivers/tty/tty_io.c
+++ linux-2.6.git/drivers/tty/tty_io.c
@@ -1559,13 +1559,14 @@ static void release_one_tty(struct work_
 	struct tty_struct *tty =
 		container_of(work, struct tty_struct, hangup_work);
 	struct tty_driver *driver = tty->driver;
+	struct module *owner = driver->owner;
 
 	if (tty->ops->cleanup)
 		tty->ops->cleanup(tty);
 
 	tty->magic = 0;
 	tty_driver_kref_put(driver);
-	module_put(driver->owner);
+	module_put(owner);
 
 	spin_lock(&tty_files_lock);
 	list_del_init(&tty->tty_files);

Re: [PATCH] tty: Fix potential use after free in release_one_tty

[Greg Kroah-Hartman]

On Thu, Aug 07, 2014 at 01:34:28PM +0400, Cyrill Gorcunov wrote:
> On Thu, Aug 07, 2014 at 01:18:12PM +0400, Cyrill Gorcunov wrote:
> > > mod = driver->owner;
> > > tty_driver_kref_put(driver);
> > > module_put(mod);
> > > 
> > > Check the upstream whether the same issue exists there.
> > 
> > Same in tty.git
> > 
> > static void release_one_tty(struct work_struct *work)
> > {
> ---
> From: Cyrill Gorcunov <gorcunov@gmail.com>
> Subject: tty: Fix potential use after free in release_one_tty
> 
> In case if we're releasing the last tty reference the following
> call sequence is possible
> 
> tty_driver_kref_put
>   destruct_tty_driver
>     kfree(driver);
> 
> where @driver is used in next module_put call, which leads to
> 
>  | [ 285.964007] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
>  | [ 285.964007] Workqueue: events release_one_tty
>  | [ 285.964007] task: ffff8800cc7ea5f0 ti: ffff8800cb800000 task.ti: ffff8800cb800000
>  | [ 285.964007] RIP: 0010:[<ffffffff810aeaf5>] [<ffffffff810aeaf5>] module_put+0x24/0xf4
>  | [ 285.964007] RSP: 0018:ffff8800cb801d48 EFLAGS: 00010213
>  | [ 285.964007] RAX: ffff8800cb801fd8 RBX: ffff8800ca3429d0 RCX: ffff8800cb1db400
>  | [ 285.964007] RDX: 0000000000000000 RSI: ffffffff817349c1 RDI: 0000000000000001
>  | [ 285.964007] RBP: ffff8800cb801d60 R08: ffff8800cd632b40 R09: 0000000000000000
>  | [ 285.964007] R10: 00000000ffffffff R11: ffff88011f40a000 R12: 6b6b6b6b6b6b6b6b
>  | [ 285.964007] R13: ffff8800ca342520 R14: 0000000000000000 R15: ffff88011f5d8200
>  | [ 285.964007] FS: 0000000000000000(0000) GS:ffff88011f400000(0000) knlGS:0000000000000000
>  | [ 285.964007] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b
>  | [ 285.964007] CR2: 00007faf5229d090 CR3: 0000000001c0b000 CR4: 00000000000006f0
>  | [ 285.964007] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>  | [ 285.964007] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
>  | [ 285.964007] Stack:
>  | [ 285.964007] ffff8800ca3429d0 ffff8800ca342a30 ffff8800ca342520 ffff8800cb801d88
>  | [ 285.964007] ffffffff8146554a ffff8800cc77cc78 ffff8800ca3429d0 ffff88011f5d3800
>  | [ 285.964007] ffff8800cb801e08 ffffffff810683c1 ffffffff810682ff 0000000000000046
>  | [ 285.964007] Call Trace:
>  | [ 285.964007] [<ffffffff8146554a>] release_one_tty+0x54/0xa3
>  | [ 285.964007] [<ffffffff810683c1>] process_one_work+0x223/0x404
>  | [ 285.964007] [<ffffffff810682ff>] ? process_one_work+0x161/0x404
>  | [ 285.964007] [<ffffffff81068971>] worker_thread+0x136/0x205
>  | [ 285.964007] [<ffffffff8106883b>] ? rescuer_thread+0x26a/0x26a
>  | [ 285.964007] [<ffffffff8106e5bf>] kthread+0xa2/0xaa
>  | [ 285.964007] [<ffffffff810a4586>] ? trace_hardirqs_on_caller+0x16/0x1eb
>  | [ 285.964007] [<ffffffff8106e51d>] ? __kthread_parkme+0x65/0x65
>  | [ 285.964007] [<ffffffff8173f59c>] ret_from_fork+0x7c/0xb0
>  | [ 285.964007] [<ffffffff8106e51d>] ? __kthread_parkme+0x65/0x65
>  | [ 285.964007] Code: 09 00 5b 41 5c 5d c3 0f 1f 44 00 00 55 48 85 ff 48 89 e5 41 55 41 54 49 89 fc 53 0f 84 d3 00
>  | 00 00 bf 01 00 00 00 e8 d0 a1 fc ff <49> 8b 84 24 50 02 00 00 65 48 ff 40 08 4c 8b 6d 08 0f 1f 44 00
> 
> so simply keep a local reference to the module owner and
> use it later.
> 
> (Note I hit the bug on custom kernel but I think vanilla one suffers
>  the same problem).
> 
> CC: Pavel Emelyanov <xemul@parallels.com>
> CC: Jiri Slaby <jslaby@suse.cz>
> CC: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
> ---
>  drivers/tty/tty_io.c |    3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> Index: linux-2.6.git/drivers/tty/tty_io.c
> ===================================================================
> --- linux-2.6.git.orig/drivers/tty/tty_io.c
> +++ linux-2.6.git/drivers/tty/tty_io.c
> @@ -1559,13 +1559,14 @@ static void release_one_tty(struct work_
>  	struct tty_struct *tty =
>  		container_of(work, struct tty_struct, hangup_work);
>  	struct tty_driver *driver = tty->driver;
> +	struct module *owner = driver->owner;
>  
>  	if (tty->ops->cleanup)
>  		tty->ops->cleanup(tty);
>  
>  	tty->magic = 0;
>  	tty_driver_kref_put(driver);
> -	module_put(driver->owner);
> +	module_put(owner);
>  
>  	spin_lock(&tty_files_lock);
>  	list_del_init(&tty->tty_files);

Looks good, thanks, can you resend it in a format I can apply it in (one
that doesn't require me to edit it by hand)?  Then I'll be glad to queue
it up after 3.17-rc1 is out.

thanks,

greg k-h