[PATCH 1/3] blk-mq: Return true if request was completed

[PATCH 1/3] blk-mq: Return true if request was completed

[Keith Busch]

A driver may have internal state to cleanup if we're pretending a request
timeout occured. Return 'false' if the command wasn't actually completed
due to the error injection, and true otherwise.

Signed-off-by: Keith Busch <keith.busch@intel.com>
---
 block/blk-mq.c         | 5 +++--
 include/linux/blk-mq.h | 2 +-
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 3f91c6e5b17a..f91951800a64 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -630,11 +630,12 @@ static void hctx_lock(struct blk_mq_hw_ctx *hctx, int *srcu_idx)
  *	Ends all I/O on a request. It does not handle partial completions.
  *	The actual completion happens out-of-order, through a IPI handler.
  **/
-void blk_mq_complete_request(struct request *rq)
+bool blk_mq_complete_request(struct request *rq)
 {
 	if (unlikely(blk_should_fake_timeout(rq->q)))
-		return;
+		return false;
 	__blk_mq_complete_request(rq);
+	return true;
 }
 EXPORT_SYMBOL(blk_mq_complete_request);
 
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 2286dc12c6bc..dec6ef385492 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -264,7 +264,7 @@ void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
 				bool kick_requeue_list);
 void blk_mq_kick_requeue_list(struct request_queue *q);
 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
-void blk_mq_complete_request(struct request *rq);
+bool blk_mq_complete_request(struct request *rq);
 bool blk_mq_bio_list_merge(struct request_queue *q, struct list_head *list,
 			   struct bio *bio);
 bool blk_mq_queue_stopped(struct request_queue *q);
-- 
2.14.4

[PATCH 2/3] scsi: Do not rely on blk-mq for double completions

[Keith Busch]

The scsi timeout error handling had been directly updating the request
state to prevent a natural completion and error handling from completing
the same request twice. Fix this layering violation by having scsi
control the fate of its commands with scsi owned flags rather than
use blk-mq's.

Signed-off-by: Keith Busch <keith.busch@intel.com>
---
 drivers/scsi/scsi_error.c | 17 +++--------------
 drivers/scsi/scsi_lib.c   |  6 +++++-
 include/scsi/scsi_cmnd.h  |  5 ++++-
 3 files changed, 12 insertions(+), 16 deletions(-)

diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index c736d61b1648..f89e829a1c51 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -199,6 +199,9 @@ scsi_abort_command(struct scsi_cmnd *scmd)
 		return FAILED;
 	}
 
+	if (test_and_set_bit(__SCMD_COMPLETE, &scmd->flags))
+		return SUCCESS;
+
 	spin_lock_irqsave(shost->host_lock, flags);
 	if (shost->eh_deadline != -1 && !shost->last_reset)
 		shost->last_reset = jiffies;
@@ -296,20 +299,6 @@ enum blk_eh_timer_return scsi_times_out(struct request *req)
 		rtn = host->hostt->eh_timed_out(scmd);
 
 	if (rtn == BLK_EH_DONE) {
-		/*
-		 * For blk-mq, we must set the request state to complete now
-		 * before sending the request to the scsi error handler. This
-		 * will prevent a use-after-free in the event the LLD manages
-		 * to complete the request before the error handler finishes
-		 * processing this timed out request.
-		 *
-		 * If the request was already completed, then the LLD beat the
-		 * time out handler from transferring the request to the scsi
-		 * error handler. In that case we can return immediately as no
-		 * further action is required.
-		 */
-		if (req->q->mq_ops && !blk_mq_mark_complete(req))
-			return rtn;
 		if (scsi_abort_command(scmd) != SUCCESS) {
 			set_host_byte(scmd, DID_TIME_OUT);
 			scsi_eh_scmd_add(scmd);
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index c7fccbb8f554..1e74137f1073 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -2044,8 +2044,11 @@ static int scsi_mq_prep_fn(struct request *req)
 
 static void scsi_mq_done(struct scsi_cmnd *cmd)
 {
+	if (test_and_set_bit(__SCMD_COMPLETE, &cmd->flags))
+		return;
 	trace_scsi_dispatch_cmd_done(cmd);
-	blk_mq_complete_request(cmd->request);
+	if (unlikely(!blk_mq_complete_request(cmd->request)))
+		clear_bit(__SCMD_COMPLETE, &cmd->flags);
 }
 
 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
@@ -2104,6 +2107,7 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
 			goto out_dec_host_busy;
 		req->rq_flags |= RQF_DONTPREP;
 	} else {
+		cmd->flags &= ~SCMD_COMPLETE;
 		blk_mq_start_request(req);
 	}
 
diff --git a/include/scsi/scsi_cmnd.h b/include/scsi/scsi_cmnd.h
index c891ada3c5c2..acef13c628d3 100644
--- a/include/scsi/scsi_cmnd.h
+++ b/include/scsi/scsi_cmnd.h
@@ -58,6 +58,9 @@ struct scsi_pointer {
 #define SCMD_TAGGED		(1 << 0)
 #define SCMD_UNCHECKED_ISA_DMA	(1 << 1)
 #define SCMD_INITIALIZED	(1 << 2)
+
+#define __SCMD_COMPLETE		3
+#define SCMD_COMPLETE		(1 << __SCMD_COMPLETE)
 /* flags preserved across unprep / reprep */
 #define SCMD_PRESERVED_FLAGS	(SCMD_UNCHECKED_ISA_DMA | SCMD_INITIALIZED)
 
@@ -144,7 +147,7 @@ struct scsi_cmnd {
 					 * to be at an address < 16Mb). */
 
 	int result;		/* Status code from lower level driver */
-	int flags;		/* Command flags */
+	unsigned long flags;	/* Command flags */
 
 	unsigned char tag;	/* SCSI-II queued command tag */
 };
-- 
2.14.4

[PATCH 3/3] blk-mq: Simplify request completion state

[Keith Busch]

There are no more users relying on blk-mq request states to prevent
double completions, so replace the relatively expensive cmpxchg operation
with WRITE_ONCE.

Signed-off-by: Keith Busch <keith.busch@intel.com>
---
 block/blk-mq.c         |  4 +---
 include/linux/blk-mq.h | 14 --------------
 2 files changed, 1 insertion(+), 17 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index f91951800a64..9d569e74cbe3 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -565,9 +565,7 @@ static void __blk_mq_complete_request(struct request *rq)
 	bool shared = false;
 	int cpu;
 
-	if (!blk_mq_mark_complete(rq))
-		return;
-
+	WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
 	/*
 	 * Most of single queue controllers, there is only one irq vector
 	 * for handling IO completion, and the only irq's affinity is set
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index dec6ef385492..ec0a2688a373 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -293,20 +293,6 @@ void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
 
 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
 
-/**
- * blk_mq_mark_complete() - Set request state to complete
- * @rq: request to set to complete state
- *
- * Returns true if request state was successfully set to complete. If
- * successful, the caller is responsibile for seeing this request is ended, as
- * blk_mq_complete_request will not work again.
- */
-static inline bool blk_mq_mark_complete(struct request *rq)
-{
-	return cmpxchg(&rq->state, MQ_RQ_IN_FLIGHT, MQ_RQ_COMPLETE) ==
-			MQ_RQ_IN_FLIGHT;
-}
-
 /*
  * Driver command data is immediately after the request. So subtract request
  * size to get back to the original request, add request size to get the PDU.
-- 
2.14.4

Re: [PATCH 2/3] scsi: Do not rely on blk-mq for double completions

[Bart Van Assche]

On Wed, 2018-11-14 at 09:26 -0700, Keith Busch wrote:
> The scsi timeout error handling had been directly updating the request
> state to prevent a natural completion and error handling from completing
> the same request twice. Fix this layering violation by having scsi
> control the fate of its commands with scsi owned flags rather than
> use blk-mq's.
> 
> Signed-off-by: Keith Busch <keith.busch@intel.com>
> ---
>  drivers/scsi/scsi_error.c | 17 +++--------------
>  drivers/scsi/scsi_lib.c   |  6 +++++-
>  include/scsi/scsi_cmnd.h  |  5 ++++-
>  3 files changed, 12 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
> index c736d61b1648..f89e829a1c51 100644
> --- a/drivers/scsi/scsi_error.c
> +++ b/drivers/scsi/scsi_error.c
> @@ -199,6 +199,9 @@ scsi_abort_command(struct scsi_cmnd *scmd)
>  		return FAILED;
>  	}
>  
> +	if (test_and_set_bit(__SCMD_COMPLETE, &scmd->flags))
> +		return SUCCESS;
> +
>  	spin_lock_irqsave(shost->host_lock, flags);
>  	if (shost->eh_deadline != -1 && !shost->last_reset)
>  		shost->last_reset = jiffies;
> @@ -296,20 +299,6 @@ enum blk_eh_timer_return scsi_times_out(struct request
> *req)
>  		rtn = host->hostt->eh_timed_out(scmd);
>  
>  	if (rtn == BLK_EH_DONE) {
> -		/*
> -		 * For blk-mq, we must set the request state to complete
> now
> -		 * before sending the request to the scsi error handler.
> This
> -		 * will prevent a use-after-free in the event the LLD
> manages
> -		 * to complete the request before the error handler
> finishes
> -		 * processing this timed out request.
> -		 *
> -		 * If the request was already completed, then the LLD beat
> the
> -		 * time out handler from transferring the request to the
> scsi
> -		 * error handler. In that case we can return immediately
> as no
> -		 * further action is required.
> -		 */
> -		if (req->q->mq_ops && !blk_mq_mark_complete(req))
> -			return rtn;
>  		if (scsi_abort_command(scmd) != SUCCESS) {
>  			set_host_byte(scmd, DID_TIME_OUT);
>  			scsi_eh_scmd_add(scmd);
> diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
> index c7fccbb8f554..1e74137f1073 100644
> --- a/drivers/scsi/scsi_lib.c
> +++ b/drivers/scsi/scsi_lib.c
> @@ -2044,8 +2044,11 @@ static int scsi_mq_prep_fn(struct request *req)
>  
>  static void scsi_mq_done(struct scsi_cmnd *cmd)
>  {
> +	if (test_and_set_bit(__SCMD_COMPLETE, &cmd->flags))
> +		return;
>  	trace_scsi_dispatch_cmd_done(cmd);
> -	blk_mq_complete_request(cmd->request);
> +	if (unlikely(!blk_mq_complete_request(cmd->request)))
> +		clear_bit(__SCMD_COMPLETE, &cmd->flags);
>  }
>  
>  static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
> @@ -2104,6 +2107,7 @@ static blk_status_t scsi_queue_rq(struct
> blk_mq_hw_ctx *hctx,
>  			goto out_dec_host_busy;
>  		req->rq_flags |= RQF_DONTPREP;
>  	} else {
> +		cmd->flags &= ~SCMD_COMPLETE;
>  		blk_mq_start_request(req);
>  	}


Hi Keith,

Please Cc Martin Petersen and the scsi mailing list for SCSI patches.

Regarding this patch: I think this patch introduces a subtle but severe bug
in the SCSI core, namely that if an abort is processed concurrently with
request completion with "fake timeout" enabled that the abort is ignored.

Bart.

Re: [PATCH 2/3] scsi: Do not rely on blk-mq for double completions

[Keith Busch]

On Wed, Nov 14, 2018 at 09:51:36AM -0800, Bart Van Assche wrote:
> Regarding this patch: I think this patch introduces a subtle but severe bug
> in the SCSI core, namely that if an abort is processed concurrently with
> request completion with "fake timeout" enabled that the abort is ignored.

That requires the following occur concurrently:

  1. A real completion
  2. A real timeout
  3. A fake timeout

That can't happen on a production kernel, and highly improbable
on the fake one. We can still fix it by having scsi timeout return
BLK_EH_RESET_TIMER in this case. I didn't like adding code just to work
around error injection, but there isn't a good alternative at the moment.

Re: [PATCH 2/3] scsi: Do not rely on blk-mq for double completions

[Keith Busch]

On Wed, Nov 14, 2018 at 11:00:18AM -0700, Keith Busch wrote:
> On Wed, Nov 14, 2018 at 09:51:36AM -0800, Bart Van Assche wrote:
> > Regarding this patch: I think this patch introduces a subtle but severe bug
> > in the SCSI core, namely that if an abort is processed concurrently with
> > request completion with "fake timeout" enabled that the abort is ignored.
> 
> That requires the following occur concurrently:
> 
>   1. A real completion
>   2. A real timeout
>   3. A fake timeout
> 
> That can't happen on a production kernel, and highly improbable
> on the fake one. We can still fix it by having scsi timeout return
> BLK_EH_RESET_TIMER in this case. I didn't like adding code just to work
> around error injection, but there isn't a good alternative at the moment.

So do this instead:

--8<---
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index ff372b335ced..d343024e732a 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -199,9 +199,6 @@ scsi_abort_command(struct scsi_cmnd *scmd)
 		return FAILED;
 	}
 
-	if (test_and_set_bit(__SCMD_COMPLETE, &scmd->flags))
-		return SUCCESS;
-
 	spin_lock_irqsave(shost->host_lock, flags);
 	if (shost->eh_deadline != -1 && !shost->last_reset)
 		shost->last_reset = jiffies;
@@ -299,6 +296,8 @@ enum blk_eh_timer_return scsi_times_out(struct request *req)
 		rtn = host->hostt->eh_timed_out(scmd);
 
 	if (rtn == BLK_EH_DONE) {
+		if (test_and_set_bit(__SCMD_COMPLETE, &scmd->flags))
+			return BLK_EH_RESET_TIMER;
 		if (scsi_abort_command(scmd) != SUCCESS) {
 			set_host_byte(scmd, DID_TIME_OUT);
 			scsi_eh_scmd_add(scmd);
-->8---

Re: [PATCH 2/3] scsi: Do not rely on blk-mq for double completions

[Keith Busch]

Sorry everyone, this was the previous verision. Please ignore, I'm
sending out the updated one now.

[PATCH 2/3] scsi: Do not rely on blk-mq for double completions

[Keith Busch]

The scsi timeout error handling had been directly updating the request
state to prevent a natural completion and error handling from completing
the same request twice. Fix this layering violation by having scsi
control the fate of its commands with scsi owned flags rather than
use blk-mq's.

Signed-off-by: Keith Busch <keith.busch@intel.com>
---
 drivers/scsi/scsi_error.c | 17 +++--------------
 drivers/scsi/scsi_lib.c   |  6 +++++-
 include/scsi/scsi_cmnd.h  |  5 ++++-
 3 files changed, 12 insertions(+), 16 deletions(-)

diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index c736d61b1648..f89e829a1c51 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -199,6 +199,9 @@ scsi_abort_command(struct scsi_cmnd *scmd)
 		return FAILED;
 	}
 
+	if (test_and_set_bit(__SCMD_COMPLETE, &scmd->flags))
+		return SUCCESS;
+
 	spin_lock_irqsave(shost->host_lock, flags);
 	if (shost->eh_deadline != -1 && !shost->last_reset)
 		shost->last_reset = jiffies;
@@ -296,20 +299,6 @@ enum blk_eh_timer_return scsi_times_out(struct request *req)
 		rtn = host->hostt->eh_timed_out(scmd);
 
 	if (rtn == BLK_EH_DONE) {
-		/*
-		 * For blk-mq, we must set the request state to complete now
-		 * before sending the request to the scsi error handler. This
-		 * will prevent a use-after-free in the event the LLD manages
-		 * to complete the request before the error handler finishes
-		 * processing this timed out request.
-		 *
-		 * If the request was already completed, then the LLD beat the
-		 * time out handler from transferring the request to the scsi
-		 * error handler. In that case we can return immediately as no
-		 * further action is required.
-		 */
-		if (req->q->mq_ops && !blk_mq_mark_complete(req))
-			return rtn;
 		if (scsi_abort_command(scmd) != SUCCESS) {
 			set_host_byte(scmd, DID_TIME_OUT);
 			scsi_eh_scmd_add(scmd);
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index c7fccbb8f554..1e74137f1073 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -2044,8 +2044,11 @@ static int scsi_mq_prep_fn(struct request *req)
 
 static void scsi_mq_done(struct scsi_cmnd *cmd)
 {
+	if (test_and_set_bit(__SCMD_COMPLETE, &cmd->flags))
+		return;
 	trace_scsi_dispatch_cmd_done(cmd);
-	blk_mq_complete_request(cmd->request);
+	if (unlikely(!blk_mq_complete_request(cmd->request)))
+		clear_bit(__SCMD_COMPLETE, &cmd->flags);
 }
 
 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
@@ -2104,6 +2107,7 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
 			goto out_dec_host_busy;
 		req->rq_flags |= RQF_DONTPREP;
 	} else {
+		cmd->flags &= ~SCMD_COMPLETE;
 		blk_mq_start_request(req);
 	}
 
diff --git a/include/scsi/scsi_cmnd.h b/include/scsi/scsi_cmnd.h
index c891ada3c5c2..acef13c628d3 100644
--- a/include/scsi/scsi_cmnd.h
+++ b/include/scsi/scsi_cmnd.h
@@ -58,6 +58,9 @@ struct scsi_pointer {
 #define SCMD_TAGGED		(1 << 0)
 #define SCMD_UNCHECKED_ISA_DMA	(1 << 1)
 #define SCMD_INITIALIZED	(1 << 2)
+
+#define __SCMD_COMPLETE		3
+#define SCMD_COMPLETE		(1 << __SCMD_COMPLETE)
 /* flags preserved across unprep / reprep */
 #define SCMD_PRESERVED_FLAGS	(SCMD_UNCHECKED_ISA_DMA | SCMD_INITIALIZED)
 
@@ -144,7 +147,7 @@ struct scsi_cmnd {
 					 * to be at an address < 16Mb). */
 
 	int result;		/* Status code from lower level driver */
-	int flags;		/* Command flags */
+	unsigned long flags;	/* Command flags */
 
 	unsigned char tag;	/* SCSI-II queued command tag */
 };
-- 
2.14.4