[PATCH 2/2] tty: Fix lockless tty buffer race

[PATCH 2/2] tty: Fix lockless tty buffer race

[Peter Hurley]

Commit 6a20dbd6caa2358716136144bf524331d70b1e03,
"tty: Fix race condition between __tty_buffer_request_room and flush_to_ldisc"
correctly identifies an unsafe race condition between
__tty_buffer_request_room() and flush_to_ldisc(), where the consumer
flush_to_ldisc() prematurely advances the head before consuming the
last of the data committed. For example:

           CPU 0                     |            CPU 1
__tty_buffer_request_room            | flush_to_ldisc
  ...                                |   ...
                                     |   count = head->commit - head->read
  n = tty_buffer_alloc()             |
  b->commit = b->used                |
  b->next = n                        |
                                     |   if (!count)                /* T */
                                     |     if (head->next == NULL)  /* F */
                                     |     buf->head = head->next

In this case, buf->head has been advanced but head->commit may have
been updated with a new value.

Instead of reintroducing an unnecessary lock, fix the race locklessly.
Read the commit-next pair in the reverse order of writing, which guarantees
the commit value read is the latest value written if the head is
advancing.

Reported-by: Manfred Schlaegl <manfred.schlaegl@gmx.at>
Cc: <stable@vger.kernel.org> # 3.12.x+
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
---
 drivers/tty/tty_buffer.c | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

diff --git a/drivers/tty/tty_buffer.c b/drivers/tty/tty_buffer.c
index 8ebd9f8..cf78d19 100644
--- a/drivers/tty/tty_buffer.c
+++ b/drivers/tty/tty_buffer.c
@@ -258,7 +258,11 @@ static int __tty_buffer_request_room(struct tty_port *port, size_t size,
 			n->flags = flags;
 			buf->tail = n;
 			b->commit = b->used;
-			smp_mb();
+			/* paired w/ barrier in flush_to_ldisc(); ensures the
+			 * latest commit value can be read before the head is
+			 * advanced to the next buffer
+			 */
+			smp_wmb();
 			b->next = n;
 		} else if (change)
 			size = 0;
@@ -444,17 +448,24 @@ static void flush_to_ldisc(struct work_struct *work)
 
 	while (1) {
 		struct tty_buffer *head = buf->head;
+		struct tty_buffer *next;
 		int count;
 
 		/* Ldisc or user is trying to gain exclusive access */
 		if (atomic_read(&buf->priority))
 			break;
 
+		next = head->next;
+		/* paired w/ barrier in __tty_buffer_request_room();
+		 * ensures commit value read is not stale if the head
+		 * is advancing to the next buffer
+		 */
+		smp_rmb();
 		count = head->commit - head->read;
 		if (!count) {
-			if (head->next == NULL)
+			if (next == NULL)
 				break;
-			buf->head = head->next;
+			buf->head = next;
 			tty_buffer_free(port, head);
 			continue;
 		}
-- 
1.8.1.2

Re: [PATCH 2/2] tty: Fix lockless tty buffer race

[Peter Hurley]

On 05/02/2014 10:56 AM, Peter Hurley wrote:
> Commit 6a20dbd6caa2358716136144bf524331d70b1e03,
> "tty: Fix race condition between __tty_buffer_request_room and flush_to_ldisc"
> correctly identifies an unsafe race condition between
> __tty_buffer_request_room() and flush_to_ldisc(), where the consumer
> flush_to_ldisc() prematurely advances the head before consuming the
> last of the data committed. For example:
>
>             CPU 0                     |            CPU 1
> __tty_buffer_request_room            | flush_to_ldisc
>    ...                                |   ...
>                                       |   count = head->commit - head->read
>    n = tty_buffer_alloc()             |
>    b->commit = b->used                |
>    b->next = n                        |
>                                       |   if (!count)                /* T */
>                                       |     if (head->next == NULL)  /* F */
>                                       |     buf->head = head->next
>
> In this case, buf->head has been advanced but head->commit may have
> been updated with a new value.
>
> Instead of reintroducing an unnecessary lock, fix the race locklessly.
> Read the commit-next pair in the reverse order of writing, which guarantees
> the commit value read is the latest value written if the head is
> advancing.
>
> Reported-by: Manfred Schlaegl <manfred.schlaegl@gmx.at>
> Cc: <stable@vger.kernel.org> # 3.12.x+

The patch submitted by Manfred notes the commits which introduced the
race [1], but attributes those commits to the 3.11 cycle. Those commits
were merged in the 3.12 cycle.

Regards,
Peter Hurley

[1] commits e8437d7ecbc50198705331449367d401ebb3181f,
"tty: Make driver-side flip buffers lockless", and
e9975fdec0138f1b2a85b9624e41660abd9865d4,
"tty: Ensure single-threaded flip buffer consumer with mutex"

Re: [PATCH 2/2] tty: Fix lockless tty buffer race

[Manfred Schlaegl]

On 2014-05-02 17:05, Peter Hurley wrote:
> On 05/02/2014 10:56 AM, Peter Hurley wrote:
>> Commit 6a20dbd6caa2358716136144bf524331d70b1e03,
>> "tty: Fix race condition between __tty_buffer_request_room and flush_to_ldisc"
>> correctly identifies an unsafe race condition between
>> __tty_buffer_request_room() and flush_to_ldisc(), where the consumer
>> flush_to_ldisc() prematurely advances the head before consuming the
>> last of the data committed. For example:
>>
>>             CPU 0                     |            CPU 1
>> __tty_buffer_request_room            | flush_to_ldisc
>>    ...                                |   ...
>>                                       |   count = head->commit - head->read
>>    n = tty_buffer_alloc()             |
>>    b->commit = b->used                |
>>    b->next = n                        |
>>                                       |   if (!count)                /* T */
>>                                       |     if (head->next == NULL)  /* F */
>>                                       |     buf->head = head->next
>>
>> In this case, buf->head has been advanced but head->commit may have
>> been updated with a new value.
>>
>> Instead of reintroducing an unnecessary lock, fix the race locklessly.
>> Read the commit-next pair in the reverse order of writing, which guarantees
>> the commit value read is the latest value written if the head is
>> advancing.

This is a fine solution! I'll verify this against my previous experimental setup
(3.12.x and 3.12.x-rt25), but I dont't expect any problems.

>>
>> Reported-by: Manfred Schlaegl <manfred.schlaegl@gmx.at>
>> Cc: <stable@vger.kernel.org> # 3.12.x+
> 
> The patch submitted by Manfred notes the commits which introduced the
> race [1], but attributes those commits to the 3.11 cycle. Those commits
> were merged in the 3.12 cycle.

You are right. I'm sorry for this.


Regars,
Manfred