block: Introduce the UFQ I/O scheduler

Introduce IOSCHED_UFQ, a blk-mq elevator ("ufq: User-programmable
Flexible Queueing") whose policy is supplied by an eBPF program via
struct_ops (insert, dispatch, merge, finish, etc.).

When no eBPF program is attached, the UFQ I/O scheduler uses a simple,
per-ctx queueing policy (similar to none). After an eBPF program is
attached, the user-defined scheduling policy replaces UFQ’s built-in
queueing policy, while per-ctx queues remain available as a fallback
mechanism.

Signed-off-by: Kaitao Cheng <[email protected]>

Author: Kaitao Cheng

block/Kconfig.iosched

@@ -44,4 +44,12 @@ config BFQ_CGROUP_DEBUG
 	Enable some debugging help. Currently it exports additional stat
 	files in a cgroup which can be useful for debugging.
 
+config IOSCHED_UFQ
+	tristate "UFQ I/O scheduler"
+	default y
+	help
+	The UFQ I/O scheduler is a programmable I/O scheduler. When
+	enabled, an out-of-kernel I/O scheduler based on eBPF can be
+	designed to interact with it, leveraging its customizable
+	hooks to redefine I/O scheduling policies.
 endmenu

block/Makefile

@@ -24,6 +24,7 @@ obj-$(CONFIG_MQ_IOSCHED_DEADLINE)	+= mq-deadline.o
 obj-$(CONFIG_MQ_IOSCHED_KYBER)	+= kyber-iosched.o
 bfq-y				:= bfq-iosched.o bfq-wf2q.o bfq-cgroup.o
 obj-$(CONFIG_IOSCHED_BFQ)	+= bfq.o
+obj-$(CONFIG_IOSCHED_UFQ)	+= ufq-iosched.o ufq-bpfops.o ufq-kfunc.o
 
 obj-$(CONFIG_BLK_DEV_INTEGRITY) += bio-integrity.o blk-integrity.o t10-pi.o \
 				   bio-integrity-auto.o bio-integrity-fs.o

block/blk-merge.c

@@ -774,8 +774,8 @@ u8 bio_seg_gap(struct request_queue *q, struct bio *prev, struct bio *next,
  * For non-mq, this has to be called with the request spinlock acquired.
  * For mq with scheduling, the appropriate queue wide lock should be held.
  */
-static struct request *attempt_merge(struct request_queue *q,
-				     struct request *req, struct request *next)
+static struct request *attempt_merge(struct request_queue *q, struct request *req,
+				     struct request *next, bool nohash)
 {
 	if (!rq_mergeable(req) || !rq_mergeable(next))
 		return NULL;
@@ -842,7 +842,7 @@ static struct request *attempt_merge(struct request_queue *q,
 
 	req->__data_len += blk_rq_bytes(next);
 
-	if (!blk_discard_mergable(req))
+	if (!nohash && !blk_discard_mergable(req))
 		elv_merge_requests(q, req, next);
 
 	blk_crypto_rq_put_keyslot(next);
@@ -868,7 +868,7 @@ static struct request *attempt_back_merge(struct request_queue *q,
 	struct request *next = elv_latter_request(q, rq);
 
 	if (next)
-		return attempt_merge(q, rq, next);
+		return attempt_merge(q, rq, next, false);
 
 	return NULL;
 }
@@ -879,11 +879,17 @@ static struct request *attempt_front_merge(struct request_queue *q,
 	struct request *prev = elv_former_request(q, rq);
 
 	if (prev)
-		return attempt_merge(q, prev, rq);
+		return attempt_merge(q, prev, rq, false);
 
 	return NULL;
 }
 
+struct request *bpf_attempt_merge(struct request_queue *q, struct request *rq,
+				  struct request *next)
+{
+	return attempt_merge(q, rq, next, true);
+}
+
 /*
  * Try to merge 'next' into 'rq'. Return true if the merge happened, false
  * otherwise. The caller is responsible for freeing 'next' if the merge
@@ -892,7 +898,7 @@ static struct request *attempt_front_merge(struct request_queue *q,
 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
 			   struct request *next)
 {
-	return attempt_merge(q, rq, next);
+	return attempt_merge(q, rq, next, false);
 }
 
 bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
@@ -1169,3 +1175,34 @@ bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
 	}
 }
 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);
+
+bool blk_mq_sched_merge_fn(struct request_queue *q, struct bio *bio,
+		unsigned int nr_segs, struct request **merged_request,
+		struct request *rq, enum elv_merge type, void (*fn)
+		(struct request_queue *, struct request *, enum elv_merge))
+{
+	switch (type) {
+	case ELEVATOR_BACK_MERGE:
+		if (!blk_mq_sched_allow_merge(q, rq, bio))
+			return false;
+		if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
+			return false;
+		*merged_request = attempt_back_merge(q, rq);
+		if (!*merged_request)
+			fn(q, rq, ELEVATOR_BACK_MERGE);
+		return true;
+	case ELEVATOR_FRONT_MERGE:
+		if (!blk_mq_sched_allow_merge(q, rq, bio))
+			return false;
+		if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK)
+			return false;
+		*merged_request = attempt_front_merge(q, rq);
+		if (!*merged_request)
+			fn(q, rq, ELEVATOR_FRONT_MERGE);
+		return true;
+	case ELEVATOR_DISCARD_MERGE:
+		return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK;
+	default:
+		return false;
+	}
+}

block/blk-mq-sched.h

@@ -7,6 +7,10 @@
 
 #define MAX_SCHED_RQ (16 * BLKDEV_DEFAULT_RQ)
 
+bool blk_mq_sched_merge_fn(struct request_queue *q, struct bio *bio,
+		unsigned int nr_segs, struct request **merged_request,
+		struct request *rq, enum elv_merge type, void (*fn)
+		(struct request_queue *, struct request *, enum elv_merge));
 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
 		unsigned int nr_segs, struct request **merged_request);
 bool blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio,

block/blk-mq.c

@@ -796,7 +796,7 @@ static void blk_mq_finish_request(struct request *rq)
 	}
 }
 
-static void __blk_mq_free_request(struct request *rq)
+void __blk_mq_free_request(struct request *rq)
 {
 	struct request_queue *q = rq->q;
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
@@ -1844,6 +1844,12 @@ static bool dispatch_rq_from_ctx(struct sbitmap *sb, unsigned int bitnr,
 		if (list_empty(&ctx->rq_lists[type]))
 			sbitmap_clear_bit(sb, bitnr);
 	}
+
+	if (dispatch_data->rq) {
+		dispatch_data->rq->rq_flags |= RQF_STARTED;
+		if (hctx->queue->last_merge == dispatch_data->rq)
+			hctx->queue->last_merge = NULL;
+	}
 	spin_unlock(&ctx->lock);
 
 	return !dispatch_data->rq;

block/blk-mq.h

@@ -56,7 +56,7 @@ void blk_mq_flush_busy_ctxs(struct blk_mq_hw_ctx *hctx, struct list_head *list);
 struct request *blk_mq_dequeue_from_ctx(struct blk_mq_hw_ctx *hctx,
 					struct blk_mq_ctx *start);
 void blk_mq_put_rq_ref(struct request *rq);
-
+void __blk_mq_free_request(struct request *rq);
 /*
  * Internal helpers for allocating/freeing the request map
  */

block/blk.h

@@ -444,6 +444,8 @@ static inline unsigned get_max_segment_size(const struct queue_limits *lim,
 
 int ll_back_merge_fn(struct request *req, struct bio *bio,
 		unsigned int nr_segs);
+struct request *bpf_attempt_merge(struct request_queue *q, struct request *rq,
+				  struct request *next);
 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
 				struct request *next);
 unsigned int blk_recalc_rq_segments(struct request *rq);

block/ufq-bpfops.c

@@ -0,0 +1,213 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2026 KylinSoft Corporation.
+ * Copyright (c) 2026 Kaitao Cheng <[email protected]>
+ */
+#include <linux/init.h>
+#include <linux/types.h>
+#include <linux/bpf_verifier.h>
+#include <linux/bpf.h>
+#include <linux/btf.h>
+#include <linux/btf_ids.h>
+#include <linux/string.h>
+#include "ufq-iosched.h"
+
+struct ufq_iosched_ops ufq_ops;
+
+static const struct bpf_func_proto *
+bpf_ufq_get_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
+{
+	return bpf_base_func_proto(func_id, prog);
+}
+
+static bool bpf_ufq_is_valid_access(int off, int size,
+				    enum bpf_access_type type,
+				    const struct bpf_prog *prog,
+				    struct bpf_insn_access_aux *info)
+{
+	if (type != BPF_READ)
+		return false;
+	if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
+		return false;
+	if (off % size != 0)
+		return false;
+
+	/*
+	 * merge_req's third argument is int *type.  btf_ctx_access() treats
+	 * pointers that are not "pointer to struct" as scalars (no reg_type),
+	 * so loading the pointer from ctx leaves a SCALAR and *type stores
+	 * fail verification.  Model it as a read/write buffer of merge_type.
+	 */
+	if (off == 16 && size == sizeof(__u64) &&
+	    prog->aux->attach_func_name &&
+	    !strcmp(prog->aux->attach_func_name, "merge_req")) {
+		if (!btf_ctx_access(off, size, type, prog, info))
+			return false;
+		info->reg_type = PTR_TO_BUF;
+		return true;
+	}
+
+	return btf_ctx_access(off, size, type, prog, info);
+}
+
+static const struct bpf_verifier_ops bpf_ufq_verifier_ops = {
+	.get_func_proto = bpf_ufq_get_func_proto,
+	.is_valid_access = bpf_ufq_is_valid_access,
+};
+
+static int bpf_ufq_init_member(const struct btf_type *t,
+			       const struct btf_member *member,
+			       void *kdata, const void *udata)
+{
+	const struct ufq_iosched_ops *uops = udata;
+	struct ufq_iosched_ops *ops = kdata;
+	u32 moff = __btf_member_bit_offset(t, member) / 8;
+	int ret;
+
+	switch (moff) {
+	case offsetof(struct ufq_iosched_ops, name):
+		ret = bpf_obj_name_cpy(ops->name, uops->name,
+				       sizeof(ops->name));
+		if (ret < 0)
+			return ret;
+		if (ret == 0)
+			return -EINVAL;
+		return 1;
+	/* other var adding .... */
+	}
+
+	return 0;
+}
+
+static int bpf_ufq_check_member(const struct btf_type *t,
+				const struct btf_member *member,
+				const struct bpf_prog *prog)
+{
+	return 0;
+}
+
+static int bpf_ufq_enable(struct ufq_iosched_ops *ops)
+{
+	ufq_ops = *ops;
+	return 0;
+}
+
+static void bpf_ufq_disable(struct ufq_iosched_ops *ops)
+{
+	memset(&ufq_ops, 0, sizeof(ufq_ops));
+}
+
+static int bpf_ufq_reg(void *kdata, struct bpf_link *link)
+{
+	return bpf_ufq_enable(kdata);
+}
+
+static void bpf_ufq_unreg(void *kdata, struct bpf_link *link)
+{
+	bpf_ufq_disable(kdata);
+}
+
+static int bpf_ufq_init(struct btf *btf)
+{
+	return 0;
+}
+
+static int bpf_ufq_update(void *kdata, void *old_kdata, struct bpf_link *link)
+{
+	/*
+	 * UFQ does not support live-updating an already-attached BPF scheduler:
+	 * partial failure during callback setup (e.g. init_sched) would be hard
+	 * to reason about, and update can race with unregister/teardown.
+	 */
+	return -EOPNOTSUPP;
+}
+
+static int bpf_ufq_validate(void *kdata)
+{
+	return 0;
+}
+
+static int init_sched_stub(struct request_queue *q)
+{
+	return -EPERM;
+}
+
+static int exit_sched_stub(struct request_queue *q)
+{
+	return -EPERM;
+}
+
+static int insert_req_stub(struct request_queue *q, struct request *rq,
+			   blk_insert_t flags)
+{
+	return 0;
+}
+
+static struct request *dispatch_req_stub(struct request_queue *q)
+{
+	return NULL;
+}
+
+static bool has_req_stub(struct request_queue *q, int rqs_count)
+{
+	return rqs_count > 0;
+}
+
+static void finish_req_stub(struct request *rq)
+{
+}
+
+static struct request *former_req_stub(struct request_queue *q, struct request *rq)
+{
+	return NULL;
+}
+
+static struct request *next_req_stub(struct request_queue *q, struct request *rq)
+{
+	return NULL;
+}
+
+static struct request *merge_req_stub(struct request_queue *q, struct request *rq,
+				      int *type)
+{
+	*type = ELEVATOR_NO_MERGE;
+	return NULL;
+}
+
+static void req_merged_stub(struct request_queue *q, struct request *rq,
+			    int type)
+{
+}
+
+static struct ufq_iosched_ops __bpf_ops_ufq_ops = {
+	.init_sched	= init_sched_stub,
+	.exit_sched	= exit_sched_stub,
+	.insert_req	= insert_req_stub,
+	.dispatch_req	= dispatch_req_stub,
+	.has_req	= has_req_stub,
+	.former_req	= former_req_stub,
+	.next_req	= next_req_stub,
+	.merge_req	= merge_req_stub,
+	.req_merged	= req_merged_stub,
+	.finish_req	= finish_req_stub,
+};
+
+static struct bpf_struct_ops bpf_iosched_ufq_ops = {
+	.verifier_ops = &bpf_ufq_verifier_ops,
+	.reg = bpf_ufq_reg,
+	.unreg = bpf_ufq_unreg,
+	.check_member = bpf_ufq_check_member,
+	.init_member = bpf_ufq_init_member,
+	.init = bpf_ufq_init,
+	.update = bpf_ufq_update,
+	.validate = bpf_ufq_validate,
+	.name = "ufq_iosched_ops",
+	.owner = THIS_MODULE,
+	.cfi_stubs = &__bpf_ops_ufq_ops
+};
+
+int bpf_ufq_ops_init(void)
+{
+	return register_bpf_struct_ops(&bpf_iosched_ufq_ops, ufq_iosched_ops);
+}
+