[PATCH v4] io_uring: limit inflight IO

[PATCH v4] io_uring: limit inflight IO

[Jens Axboe]

With unbounded request times, we can potentially have a lot of IO
inflight. As we provide no real backpressure unless
IORING_SETUP_CQ_NODROP is set, and even there there's quite some delay
between overflows and backpressure being applied, let's put some safety
in place to avoid going way overboard.

This limits the maximum number of inflight IO for any given io_ring_ctx
to twice the CQ ring size. This is a losely managed limit, we only check
for every SQ ring size number of events. That should be good enough to
achieve our goal, which is to prevent massively deep queues. If these
are async requests, they would just be waiting for an execution slot
anyway.

We return -EBUSY if we can't queue anymore IO. The caller should reap
some completions and retry the operation after that. Note that this is
a "should never hit this" kind of condition, as driving the depth into
CQ overflow situations is unreliable.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

---

Changes since v3:
- Fixup masking for when to check (Pavel)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 472bbc11688c..a2548a6dd195 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -745,7 +745,7 @@ static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx,
 	struct io_kiocb *req;
 
 	if (!percpu_ref_tryget(&ctx->refs))
-		return NULL;
+		return ERR_PTR(-ENXIO);
 
 	if (!state) {
 		req = kmem_cache_alloc(req_cachep, gfp);
@@ -791,7 +791,7 @@ static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx,
 	if (req)
 		goto got_it;
 	percpu_ref_put(&ctx->refs);
-	return NULL;
+	return ERR_PTR(-EBUSY);
 }
 
 static void io_free_req_many(struct io_ring_ctx *ctx, void **reqs, int *nr)
@@ -2997,6 +2997,31 @@ static bool io_get_sqring(struct io_ring_ctx *ctx, struct sqe_submit *s)
 	return false;
 }
 
+static bool io_sq_over_limit(struct io_ring_ctx *ctx, unsigned to_submit)
+{
+	unsigned inflight;
+
+	if (!list_empty(&ctx->cq_overflow_list)) {
+		io_cqring_overflow_flush(ctx, false);
+		return true;
+	}
+
+	/*
+	 * This doesn't need to be super precise, so only check every once
+	 * in a while.
+	 */
+	if ((ctx->cached_sq_head & ~ctx->sq_mask) !=
+	    ((ctx->cached_sq_head + to_submit) & ~ctx->sq_mask))
+		return false;
+
+	/*
+	 * Limit us to 2x the CQ ring size
+	 */
+	inflight = ctx->cached_sq_head -
+		  (ctx->cached_cq_tail + atomic_read(&ctx->cached_cq_overflow));
+	return inflight > 2 * ctx->cq_entries;
+}
+
 static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
 			  struct file *ring_file, int ring_fd,
 			  struct mm_struct **mm, bool async)
@@ -3007,10 +3032,8 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
 	int i, submitted = 0;
 	bool mm_fault = false;
 
-	if (!list_empty(&ctx->cq_overflow_list)) {
-		io_cqring_overflow_flush(ctx, false);
+	if (unlikely(io_sq_over_limit(ctx, nr)))
 		return -EBUSY;
-	}
 
 	if (nr > IO_PLUG_THRESHOLD) {
 		io_submit_state_start(&state, ctx, nr);
@@ -3022,9 +3045,9 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
 		unsigned int sqe_flags;
 
 		req = io_get_req(ctx, statep);
-		if (unlikely(!req)) {
+		if (unlikely(IS_ERR(req))) {
 			if (!submitted)
-				submitted = -EAGAIN;
+				submitted = PTR_ERR(req);
 			break;
 		}
 		if (!io_get_sqring(ctx, &req->submit)) {
@@ -3045,8 +3068,10 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
 		if (link && (sqe_flags & IOSQE_IO_DRAIN)) {
 			if (!shadow_req) {
 				shadow_req = io_get_req(ctx, NULL);
-				if (unlikely(!shadow_req))
+				if (unlikely(IS_ERR(shadow_req))) {
+					shadow_req = NULL;
 					goto out;
+				}
 				shadow_req->flags |= (REQ_F_IO_DRAIN | REQ_F_SHADOW_DRAIN);
 				refcount_dec(&shadow_req->refs);
 			}

-- 
Jens Axboe

Re: [PATCH v4] io_uring: limit inflight IO

[Pavel Begunkov]

[-- Attachment #1.1: Type: text/plain, Size: 4653 bytes --]

On 10/11/2019 00:01, Jens Axboe wrote:
> With unbounded request times, we can potentially have a lot of IO
> inflight. As we provide no real backpressure unless
> IORING_SETUP_CQ_NODROP is set, and even there there's quite some delay
> between overflows and backpressure being applied, let's put some safety
> in place to avoid going way overboard.
> 
> This limits the maximum number of inflight IO for any given io_ring_ctx
> to twice the CQ ring size. This is a losely managed limit, we only check
> for every SQ ring size number of events. That should be good enough to
> achieve our goal, which is to prevent massively deep queues. If these
> are async requests, they would just be waiting for an execution slot
> anyway.
> 
> We return -EBUSY if we can't queue anymore IO. The caller should reap
> some completions and retry the operation after that. Note that this is
> a "should never hit this" kind of condition, as driving the depth into
> CQ overflow situations is unreliable.
> 
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> 
> ---
> 
> Changes since v3:
> - Fixup masking for when to check (Pavel)
> 
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 472bbc11688c..a2548a6dd195 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -745,7 +745,7 @@ static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx,
>  	struct io_kiocb *req;
>  
>  	if (!percpu_ref_tryget(&ctx->refs))
> -		return NULL;
> +		return ERR_PTR(-ENXIO);
>  
>  	if (!state) {
>  		req = kmem_cache_alloc(req_cachep, gfp);
> @@ -791,7 +791,7 @@ static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx,
>  	if (req)
>  		goto got_it;
>  	percpu_ref_put(&ctx->refs);
> -	return NULL;
> +	return ERR_PTR(-EBUSY);
>  }
>  
>  static void io_free_req_many(struct io_ring_ctx *ctx, void **reqs, int *nr)
> @@ -2997,6 +2997,31 @@ static bool io_get_sqring(struct io_ring_ctx *ctx, struct sqe_submit *s)
>  	return false;
>  }
>  
> +static bool io_sq_over_limit(struct io_ring_ctx *ctx, unsigned to_submit)
> +{
> +	unsigned inflight;
> +
> +	if (!list_empty(&ctx->cq_overflow_list)) {
> +		io_cqring_overflow_flush(ctx, false);
> +		return true;
> +	}
> +
> +	/*
> +	 * This doesn't need to be super precise, so only check every once
> +	 * in a while.
> +	 */
> +	if ((ctx->cached_sq_head & ~ctx->sq_mask) !=
> +	    ((ctx->cached_sq_head + to_submit) & ~ctx->sq_mask))
> +		return false;

Still not quite agree, with "!=" it will check almost every time.
We want to continue with the checks only when we overrun sq_entries
(i.e. changed higher bits).


There is a simplified trace: let
1) mask == 3, ~mask == 4
2) head == 0,
3) submitting by one

1: (0 & 4) != (1 & 4): false // do checks
2: (1 & 4) != (2 & 4): false // do checks
3: (2 & 4) != (3 & 4): false // do checks
4: (3 & 4) != (4 & 4): true // skip, return true


The rest looks good,
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>


> +
> +	/*
> +	 * Limit us to 2x the CQ ring size
> +	 */
> +	inflight = ctx->cached_sq_head -
> +		  (ctx->cached_cq_tail + atomic_read(&ctx->cached_cq_overflow));
> +	return inflight > 2 * ctx->cq_entries;
> +}
> +
>  static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
>  			  struct file *ring_file, int ring_fd,
>  			  struct mm_struct **mm, bool async)
> @@ -3007,10 +3032,8 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
>  	int i, submitted = 0;
>  	bool mm_fault = false;
>  
> -	if (!list_empty(&ctx->cq_overflow_list)) {
> -		io_cqring_overflow_flush(ctx, false);
> +	if (unlikely(io_sq_over_limit(ctx, nr)))
>  		return -EBUSY;
> -	}
>  
>  	if (nr > IO_PLUG_THRESHOLD) {
>  		io_submit_state_start(&state, ctx, nr);
> @@ -3022,9 +3045,9 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
>  		unsigned int sqe_flags;
>  
>  		req = io_get_req(ctx, statep);
> -		if (unlikely(!req)) {
> +		if (unlikely(IS_ERR(req))) {
>  			if (!submitted)
> -				submitted = -EAGAIN;
> +				submitted = PTR_ERR(req);
>  			break;
>  		}
>  		if (!io_get_sqring(ctx, &req->submit)) {
> @@ -3045,8 +3068,10 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
>  		if (link && (sqe_flags & IOSQE_IO_DRAIN)) {
>  			if (!shadow_req) {
>  				shadow_req = io_get_req(ctx, NULL);
> -				if (unlikely(!shadow_req))
> +				if (unlikely(IS_ERR(shadow_req))) {
> +					shadow_req = NULL;
>  					goto out;
> +				}
>  				shadow_req->flags |= (REQ_F_IO_DRAIN | REQ_F_SHADOW_DRAIN);
>  				refcount_dec(&shadow_req->refs);
>  			}
> 

-- 
Pavel Begunkov


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH v4] io_uring: limit inflight IO

[Jens Axboe]

On 11/10/19 6:36 AM, Pavel Begunkov wrote:
>> @@ -2997,6 +2997,31 @@ static bool io_get_sqring(struct io_ring_ctx *ctx, struct sqe_submit *s)
>>   	return false;
>>   }
>>   
>> +static bool io_sq_over_limit(struct io_ring_ctx *ctx, unsigned to_submit)
>> +{
>> +	unsigned inflight;
>> +
>> +	if (!list_empty(&ctx->cq_overflow_list)) {
>> +		io_cqring_overflow_flush(ctx, false);
>> +		return true;
>> +	}
>> +
>> +	/*
>> +	 * This doesn't need to be super precise, so only check every once
>> +	 * in a while.
>> +	 */
>> +	if ((ctx->cached_sq_head & ~ctx->sq_mask) !=
>> +	    ((ctx->cached_sq_head + to_submit) & ~ctx->sq_mask))
>> +		return false;
> 
> Still not quite agree, with "!=" it will check almost every time.
> We want to continue with the checks only when we overrun sq_entries
> (i.e. changed higher bits).
> 
> 
> There is a simplified trace: let
> 1) mask == 3, ~mask == 4
> 2) head == 0,
> 3) submitting by one
> 
> 1: (0 & 4) != (1 & 4): false // do checks
> 2: (1 & 4) != (2 & 4): false // do checks
> 3: (2 & 4) != (3 & 4): false // do checks
> 4: (3 & 4) != (4 & 4): true // skip, return true

Huh yes, it's reversed, isn't it. Should be:

	if ((ctx->cached_sq_head & ~ctx->sq_mask) ==
	    ((ctx->cached_sq_head + to_submit) & ~ctx->sq_mask))
		return false;

> The rest looks good,
> Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>

Thanks, I'll make the above change too.

-- 
Jens Axboe