io_uring/zcrx: notify user when out of buffers

There are currently no easy ways for the user to know if zcrx is out of
buffers and page pool fails to allocate. Add uapi for zcrx to communicate
it back.

It's implemented as a separate CQE, which for now is posted to the creator
ctx. To use it, on registration the user space needs to pass an instance
of struct zcrx_notification_desc, which tells the kernel the user_data
for resulting CQEs and which event types are expected / allowed.

When an allowed event happens, zcrx will post a CQE containing the
specified user_data, and lower bits of cqe->res will be set to the event
mask. Before the kernel could post another notification of the given
type, the user needs to acknowledge that it processed the previous one
by issuing IORING_REGISTER_ZCRX_CTRL with ZCRX_CTRL_ARM_NOTIFICATION.

The only notification type the patch implements is
ZCRX_NOTIF_NO_BUFFERS, but we'll need more of them in the future.

Co-developed-by: Vishwanath Seshagiri <vishs@meta.com>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Vishwanath Seshagiri <vishs@meta.com>
Link: https://patch.msgid.link/35cd307a03a43583838a2e151fc641c69abd786f.1779189667.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

4 files changed, 92 insertions, 4 deletions

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index fb6ed52bae61..02c02e14f392 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -160,7 +160,7 @@ static void io_poison_cached_req(struct io_kiocb *req)
 	req->apoll = IO_URING_PTR_POISON;
 }
 
-static void io_poison_req(struct io_kiocb *req)
+void io_poison_req(struct io_kiocb *req)
 {
 	io_poison_cached_req(req);
 	req->async_data = IO_URING_PTR_POISON;
diff --git a/io_uring/io_uring.h b/io_uring/io_uring.h
index 1b657b714373..cb736b815422 100644
--- a/io_uring/io_uring.h
+++ b/io_uring/io_uring.h
@@ -213,6 +213,7 @@ bool __io_alloc_req_refill(struct io_ring_ctx *ctx);
 
 void io_activate_pollwq(struct io_ring_ctx *ctx);
 void io_restriction_clone(struct io_restriction *dst, struct io_restriction *src);
+void io_poison_req(struct io_kiocb *req);
 
 static inline void io_lockdep_assert_cq_locked(struct io_ring_ctx *ctx)
 {
diff --git a/io_uring/zcrx.c b/io_uring/zcrx.c
index 1e194de7d6d2..2bca9a58dcf6 100644
--- a/io_uring/zcrx.c
+++ b/io_uring/zcrx.c
@@ -766,6 +766,8 @@ static int import_zcrx(struct io_ring_ctx *ctx,
 		return -EINVAL;
 	if (reg->if_rxq || reg->rq_entries || reg->area_ptr || reg->region_ptr)
 		return -EINVAL;
+	if (reg->notif_desc)
+		return -EINVAL;
 	if (reg->flags & ~ZCRX_REG_IMPORT)
 		return -EINVAL;
 
@@ -854,6 +856,7 @@ netdev_put_unlock:
 int io_register_zcrx(struct io_ring_ctx *ctx,
 		     struct io_uring_zcrx_ifq_reg __user *arg)
 {
+	struct zcrx_notification_desc notif;
 	struct io_uring_zcrx_area_reg area;
 	struct io_uring_zcrx_ifq_reg reg;
 	struct io_uring_region_desc rd;
@@ -897,10 +900,22 @@ int io_register_zcrx(struct io_ring_ctx *ctx,
 	if (copy_from_user(&area, u64_to_user_ptr(reg.area_ptr), sizeof(area)))
 		return -EFAULT;
 
+	memset(&notif, 0, sizeof(notif));
+	if (reg.notif_desc && copy_from_user(&notif, u64_to_user_ptr(reg.notif_desc),
+					     sizeof(notif)))
+		return -EFAULT;
+	if (notif.type_mask & ~ZCRX_NOTIF_TYPE_MASK)
+		return -EINVAL;
+	if (notif.__resv1 || !mem_is_zero(&notif.__resv2, sizeof(notif.__resv2)))
+		return -EINVAL;
+
 	ifq = io_zcrx_ifq_alloc(ctx);
 	if (!ifq)
 		return -ENOMEM;
 
+	ifq->notif_data = notif.user_data;
+	ifq->allowed_notif_mask = notif.type_mask;
+
 	if (ctx->user) {
 		get_uid(ctx->user);
 		ifq->user = ctx->user;
@@ -952,7 +967,8 @@ int io_register_zcrx(struct io_ring_ctx *ctx,
 		goto err;
 	}
 
-	zcrx_set_ring_ctx(ifq, ctx);
+	if (notif.type_mask)
+		zcrx_set_ring_ctx(ifq, ctx);
 	return 0;
 err:
 	scoped_guard(mutex, &ctx->mmap_lock)
@@ -1125,6 +1141,48 @@ static unsigned io_zcrx_refill_slow(struct page_pool *pp, struct io_zcrx_ifq *if
 	return allocated;
 }
 
+static void zcrx_notif_tw(struct io_tw_req tw_req, io_tw_token_t tw)
+{
+	struct io_kiocb *req = tw_req.req;
+	struct io_ring_ctx *ctx = req->ctx;
+
+	io_post_aux_cqe(ctx, req->cqe.user_data, req->cqe.res, 0);
+	percpu_ref_put(&ctx->refs);
+	io_poison_req(req);
+	kmem_cache_free(req_cachep, req);
+}
+
+static void zcrx_send_notif(struct io_zcrx_ifq *ifq, unsigned type)
+{
+	gfp_t gfp = GFP_ATOMIC | __GFP_NOWARN | __GFP_ZERO;
+	u32 type_mask = 1 << type;
+	struct io_kiocb *req;
+
+	if (!(type_mask & ifq->allowed_notif_mask))
+		return;
+
+	guard(spinlock_bh)(&ifq->ctx_lock);
+	if (!ifq->master_ctx)
+		return;
+	if (type_mask & ifq->fired_notifs)
+		return;
+
+	req = kmem_cache_alloc(req_cachep, gfp);
+	if (unlikely(!req))
+		return;
+
+	ifq->fired_notifs |= type_mask;
+
+	req->opcode = IORING_OP_NOP;
+	req->cqe.user_data = ifq->notif_data;
+	req->cqe.res = type;
+	req->ctx = ifq->master_ctx;
+	percpu_ref_get(&req->ctx->refs);
+	req->tctx = NULL;
+	req->io_task_work.func = zcrx_notif_tw;
+	io_req_task_work_add(req);
+}
+
 static netmem_ref io_pp_zc_alloc_netmems(struct page_pool *pp, gfp_t gfp)
 {
 	struct io_zcrx_ifq *ifq = io_pp_to_ifq(pp);
@@ -1141,8 +1199,10 @@ static netmem_ref io_pp_zc_alloc_netmems(struct page_pool *pp, gfp_t gfp)
 		goto out_return;
 
 	allocated = io_zcrx_refill_slow(pp, ifq, netmems, to_alloc);
-	if (!allocated)
+	if (!allocated) {
+		zcrx_send_notif(ifq, ZCRX_NOTIF_NO_BUFFERS);
 		return 0;
+	}
 out_return:
 	zcrx_sync_for_device(pp, ifq, netmems, allocated);
 	allocated--;
@@ -1291,12 +1351,32 @@ static int zcrx_flush_rq(struct io_ring_ctx *ctx, struct io_zcrx_ifq *zcrx,
 	return 0;
 }
 
+static int zcrx_arm_notif(struct io_ring_ctx *ctx, struct io_zcrx_ifq *zcrx,
+			  struct zcrx_ctrl *ctrl)
+{
+	const struct zcrx_ctrl_arm_notif *an = &ctrl->zc_arm_notif;
+	unsigned type_mask;
+
+	if (an->notif_type >= __ZCRX_NOTIF_TYPE_LAST)
+		return -EINVAL;
+	if (!mem_is_zero(&an->__resv, sizeof(an->__resv)))
+		return -EINVAL;
+
+	guard(spinlock_bh)(&zcrx->ctx_lock);
+	type_mask = 1U << an->notif_type;
+	if (type_mask & ~zcrx->fired_notifs)
+		return -EINVAL;
+	zcrx->fired_notifs &= ~type_mask;
+	return 0;
+}
+
 int io_zcrx_ctrl(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
 {
 	struct zcrx_ctrl ctrl;
 	struct io_zcrx_ifq *zcrx;
 
 	BUILD_BUG_ON(sizeof(ctrl.zc_export) != sizeof(ctrl.zc_flush));
+	BUILD_BUG_ON(sizeof(ctrl.zc_export) != sizeof(ctrl.zc_arm_notif));
 
 	if (nr_args)
 		return -EINVAL;
@@ -1314,6 +1394,8 @@ int io_zcrx_ctrl(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
 		return zcrx_flush_rq(ctx, zcrx, &ctrl);
 	case ZCRX_CTRL_EXPORT:
 		return zcrx_export(ctx, zcrx, &ctrl, arg);
+	case ZCRX_CTRL_ARM_NOTIFICATION:
+		return zcrx_arm_notif(ctx, zcrx, &ctrl);
 	}
 
 	return -EOPNOTSUPP;
diff --git a/io_uring/zcrx.h b/io_uring/zcrx.h
index 76389a5dd50f..e8b7717d6adf 100644
--- a/io_uring/zcrx.h
+++ b/io_uring/zcrx.h
@@ -9,7 +9,9 @@
 #include <net/net_trackers.h>
 
 #define ZCRX_SUPPORTED_REG_FLAGS	(ZCRX_REG_IMPORT | ZCRX_REG_NODEV)
-#define ZCRX_FEATURES			(ZCRX_FEATURE_RX_PAGE_SIZE)
+#define ZCRX_FEATURES			(ZCRX_FEATURE_RX_PAGE_SIZE |\
+					 ZCRX_FEATURE_NOTIFICATION)
+#define ZCRX_NOTIF_TYPE_MASK		(1U << ZCRX_NOTIF_NO_BUFFERS)
 
 struct io_zcrx_mem {
 	unsigned long			size;
@@ -75,6 +77,9 @@ struct io_zcrx_ifq {
 
 	spinlock_t			ctx_lock;
 	struct io_ring_ctx		*master_ctx;
+	u32				allowed_notif_mask;
+	u32				fired_notifs;
+	u64				notif_data;
 };
 
 #if defined(CONFIG_IO_URING_ZCRX)