[PATCH] blk-mq: Avoid race condition with uninitialized requests

[PATCH] blk-mq: Avoid race condition with uninitialized requests

[David Hildenbrand]

This patch should fix the bug reported in https://lkml.org/lkml/2014/9/11/249.

Test is still pending.

David Hildenbrand (1):
  blk-mq: Avoid race condition with uninitialized requests

 block/blk-mq.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

-- 
1.8.5.5

[PATCH] blk-mq: Avoid race condition with uninitialized requests

[David Hildenbrand]

This patch should fix the bug reported in https://lkml.org/lkml/2014/9/11/249.

We have to initialize at least the atomic_flags and the cmd_flags when
allocating storage for the requests.

Otherwise blk_mq_timeout_check() might dereference uninitialized pointers when
racing with the creation of a request.

Also move the reset of cmd_flags for the initializing code to the point where a
request is freed. So we will never end up with pending flush request indicators
that might trigger dereferences of invalid pointers in blk_mq_timeout_check().

Cc: stable@vger.kernel.org
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
---
 block/blk-mq.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 383ea0c..eed6340 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -203,7 +203,6 @@ __blk_mq_alloc_request(struct blk_mq_alloc_data *data, int rw)
 	if (tag != BLK_MQ_TAG_FAIL) {
 		rq = data->hctx->tags->rqs[tag];
 
-		rq->cmd_flags = 0;
 		if (blk_mq_tag_busy(data->hctx)) {
 			rq->cmd_flags = REQ_MQ_INFLIGHT;
 			atomic_inc(&data->hctx->nr_active);
@@ -258,6 +257,7 @@ static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
 
 	if (rq->cmd_flags & REQ_MQ_INFLIGHT)
 		atomic_dec(&hctx->nr_active);
+	rq->cmd_flags = 0;
 
 	clear_bit(REQ_ATOM_STARTED, &rq->atomic_flags);
 	blk_mq_put_tag(hctx, tag, &ctx->last_tag);
@@ -1404,6 +1404,8 @@ static struct blk_mq_tags *blk_mq_init_rq_map(struct blk_mq_tag_set *set,
 		left -= to_do * rq_size;
 		for (j = 0; j < to_do; j++) {
 			tags->rqs[i] = p;
+			tags->rqs[i]->atomic_flags = 0;
+			tags->rqs[i]->cmd_flags = 0;
 			if (set->ops->init_request) {
 				if (set->ops->init_request(set->driver_data,
 						tags->rqs[i], hctx_idx, i,
-- 
1.8.5.5

Re: [PATCH] blk-mq: Avoid race condition with uninitialized requests

[Christian Borntraeger]

On 09/18/2014 11:04 AM, David Hildenbrand wrote:
> This patch should fix the bug reported in https://lkml.org/lkml/2014/9/11/249.
> 
> We have to initialize at least the atomic_flags and the cmd_flags when
> allocating storage for the requests.
> 
> Otherwise blk_mq_timeout_check() might dereference uninitialized pointers when
> racing with the creation of a request.
> 
> Also move the reset of cmd_flags for the initializing code to the point where a
> request is freed. So we will never end up with pending flush request indicators
> that might trigger dereferences of invalid pointers in blk_mq_timeout_check().
> 
> Cc: stable@vger.kernel.org
> Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>

So far our testers were not able to reproduce the problem with this patch applied, we will continue testing over weekend. I will give an ACK/NACK on monday then.

Thanks


> ---
>  block/blk-mq.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 383ea0c..eed6340 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -203,7 +203,6 @@ __blk_mq_alloc_request(struct blk_mq_alloc_data *data, int rw)
>  	if (tag != BLK_MQ_TAG_FAIL) {
>  		rq = data->hctx->tags->rqs[tag];
> 
> -		rq->cmd_flags = 0;
>  		if (blk_mq_tag_busy(data->hctx)) {
>  			rq->cmd_flags = REQ_MQ_INFLIGHT;
>  			atomic_inc(&data->hctx->nr_active);
> @@ -258,6 +257,7 @@ static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
> 
>  	if (rq->cmd_flags & REQ_MQ_INFLIGHT)
>  		atomic_dec(&hctx->nr_active);
> +	rq->cmd_flags = 0;
> 
>  	clear_bit(REQ_ATOM_STARTED, &rq->atomic_flags);
>  	blk_mq_put_tag(hctx, tag, &ctx->last_tag);
> @@ -1404,6 +1404,8 @@ static struct blk_mq_tags *blk_mq_init_rq_map(struct blk_mq_tag_set *set,
>  		left -= to_do * rq_size;
>  		for (j = 0; j < to_do; j++) {
>  			tags->rqs[i] = p;
> +			tags->rqs[i]->atomic_flags = 0;
> +			tags->rqs[i]->cmd_flags = 0;
>  			if (set->ops->init_request) {
>  				if (set->ops->init_request(set->driver_data,
>  						tags->rqs[i], hctx_idx, i,
> 

Re: [PATCH] blk-mq: Avoid race condition with uninitialized requests

[Christian Borntraeger]

On 09/18/2014 11:04 AM, David Hildenbrand wrote:
> This patch should fix the bug reported in https://lkml.org/lkml/2014/9/11/249.
> 
> We have to initialize at least the atomic_flags and the cmd_flags when
> allocating storage for the requests.
> 
> Otherwise blk_mq_timeout_check() might dereference uninitialized pointers when
> racing with the creation of a request.
> 
> Also move the reset of cmd_flags for the initializing code to the point where a
> request is freed. So we will never end up with pending flush request indicators
> that might trigger dereferences of invalid pointers in blk_mq_timeout_check().
> 
> Cc: stable@vger.kernel.org
> Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>

Can you please add
Reported-by: Paulo De Rezende Pinatti <ppinatti@linux.vnet.ibm.com>
Tested-by: Paulo De Rezende Pinatti <ppinatti@linux.vnet.ibm.com>

as Paulo did the testing work?

We thing this patch is fine and should go upstream.



Christian


> ---
>  block/blk-mq.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 383ea0c..eed6340 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -203,7 +203,6 @@ __blk_mq_alloc_request(struct blk_mq_alloc_data *data, int rw)
>  	if (tag != BLK_MQ_TAG_FAIL) {
>  		rq = data->hctx->tags->rqs[tag];
> 
> -		rq->cmd_flags = 0;
>  		if (blk_mq_tag_busy(data->hctx)) {
>  			rq->cmd_flags = REQ_MQ_INFLIGHT;
>  			atomic_inc(&data->hctx->nr_active);
> @@ -258,6 +257,7 @@ static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
> 
>  	if (rq->cmd_flags & REQ_MQ_INFLIGHT)
>  		atomic_dec(&hctx->nr_active);
> +	rq->cmd_flags = 0;
> 
>  	clear_bit(REQ_ATOM_STARTED, &rq->atomic_flags);
>  	blk_mq_put_tag(hctx, tag, &ctx->last_tag);
> @@ -1404,6 +1404,8 @@ static struct blk_mq_tags *blk_mq_init_rq_map(struct blk_mq_tag_set *set,
>  		left -= to_do * rq_size;
>  		for (j = 0; j < to_do; j++) {
>  			tags->rqs[i] = p;
> +			tags->rqs[i]->atomic_flags = 0;
> +			tags->rqs[i]->cmd_flags = 0;
>  			if (set->ops->init_request) {
>  				if (set->ops->init_request(set->driver_data,
>  						tags->rqs[i], hctx_idx, i,
> 

Re: [PATCH] blk-mq: Avoid race condition with uninitialized requests

[Jens Axboe]

On 2014-09-22 08:15, Christian Borntraeger wrote:
> On 09/18/2014 11:04 AM, David Hildenbrand wrote:
>> This patch should fix the bug reported in https://lkml.org/lkml/2014/9/11/249.
>>
>> We have to initialize at least the atomic_flags and the cmd_flags when
>> allocating storage for the requests.
>>
>> Otherwise blk_mq_timeout_check() might dereference uninitialized pointers when
>> racing with the creation of a request.
>>
>> Also move the reset of cmd_flags for the initializing code to the point where a
>> request is freed. So we will never end up with pending flush request indicators
>> that might trigger dereferences of invalid pointers in blk_mq_timeout_check().
>>
>> Cc: stable@vger.kernel.org
>> Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
>
> Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
>
> Can you please add
> Reported-by: Paulo De Rezende Pinatti <ppinatti@linux.vnet.ibm.com>
> Tested-by: Paulo De Rezende Pinatti <ppinatti@linux.vnet.ibm.com>
>
> as Paulo did the testing work?
>
> We thing this patch is fine and should go upstream.

I might have to pick'n rebase the series, in which case I'll add it. But 
I already queued it up last week, so if I don't, then I can't easily add 
it. I wish the git notes wasn't such a horrible and unusable hack, so we 
had a chance to annotate commits without having to rewrite history...

-- 
Jens Axboe