wip: ublk zoned

include/ublk_cmd.h

@@ -17,6 +17,8 @@
 #define	UBLK_CMD_STOP_DEV	0x07
 #define	UBLK_CMD_SET_PARAMS	0x08
 #define	UBLK_CMD_GET_PARAMS	0x09
+#define	UBLK_CMD_REPORT_ZONES_FETCH 10
+#define	UBLK_CMD_REPORT_ZONES_COMMIT_FETCH 11
 
 /*
  * IO commands, issued by ublk server, and handled by ublk driver.
@@ -74,6 +76,11 @@
  */
 #define UBLK_F_NEED_GET_DATA (1UL << 2)
 
+/*
+ * Enable zoned device support
+ */
+#define UBLK_F_ZONED (1ULL << 3)
+
 /* device state */
 #define UBLK_S_DEV_DEAD	0
 #define UBLK_S_DEV_LIVE	1
@@ -123,6 +130,12 @@ struct ublksrv_ctrl_dev_info {
 #define		UBLK_IO_OP_DISCARD	3
 #define		UBLK_IO_OP_WRITE_SAME	4
 #define		UBLK_IO_OP_WRITE_ZEROES	5
+#define		UBLK_IO_OP_ZONE_OPEN		10
+#define		UBLK_IO_OP_ZONE_CLOSE		11
+#define		UBLK_IO_OP_ZONE_FINISH		12
+#define		UBLK_IO_OP_ZONE_APPEND		13
+#define		UBLK_IO_OP_ZONE_RESET		15
+/* TODO: Add reset_all command */
 
 #define		UBLK_IO_F_FAILFAST_DEV		(1U << 8)
 #define		UBLK_IO_F_FAILFAST_TRANSPORT	(1U << 9)
@@ -208,6 +221,12 @@ struct ublk_param_discard {
 	__u16	reserved0;
 };
 
+struct ublk_param_zoned {
+	__u64 max_open_zones;
+	__u64 max_active_zones;
+	__u64 max_append_size;
+};
+
 struct ublk_params {
 	/*
 	 * Total length of parameters, userspace has to set 'len' for both
@@ -218,10 +237,12 @@ struct ublk_params {
 	__u32	len;
 #define UBLK_PARAM_TYPE_BASIC           (1 << 0)
 #define UBLK_PARAM_TYPE_DISCARD         (1 << 1)
+#define UBLK_PARAM_TYPE_ZONED           (1 << 2)
 	__u32	types;			/* types of parameter included */
 
 	struct ublk_param_basic		basic;
 	struct ublk_param_discard	discard;
+	struct ublk_param_zoned zoned;
 };
 
 #endif

include/ublksrv.h

@@ -24,6 +24,7 @@
 #include <sys/uio.h>
 #include <sys/eventfd.h>
 #include <linux/fs.h>
+#include <linux/blkzoned.h>
 #include <unistd.h>
 #include <string.h>
 #include <pthread.h>
@@ -114,6 +115,7 @@ struct ublksrv_ctrl_dev {
 
 	cpu_set_t *queues_cpuset;
 
+	uint64_t zone_size_sectors;
 	unsigned long reserved[4];
 };
 
@@ -294,6 +296,9 @@ struct ublksrv_tgt_type {
 
 	void *(*alloc_io_buf)(struct ublksrv_queue *q, int tag, int size);
 	void (*free_io_buf)(struct ublksrv_queue *q, void *buf, int tag);
+	void (*report_zone)(struct ublksrv_dev *dev,
+			uint64_t offset_sectors,
+			struct blk_zone *zone_info);
 
 	int  type;
 	unsigned ublk_flags;	//flags required for ublk driver
@@ -367,6 +372,16 @@ extern struct ublksrv_dev *ublksrv_dev_init(const struct ublksrv_ctrl_dev *
 		ctrl_dev);
 extern void ublksrv_dev_deinit(struct ublksrv_dev *dev);
 
+extern int ublksrv_ctrl_report_zones_commit_and_fetch(
+		const struct ublksrv_ctrl_dev *ctrl_dev,
+		struct io_uring *ring,
+		struct blk_zone *zone_info,
+		uint64_t *cqe_data);
+extern int ublksrv_ctrl_report_zones_fetch(
+		const struct ublksrv_ctrl_dev *ctrl_dev,
+		struct io_uring *ring,
+		uint64_t *cqe_data);
+
 /* target json has to include the following key/value */
 #define UBLKSRV_TGT_NAME_MAX_LEN 32
 struct ublksrv_tgt_base_json {

lib/ublksrv_cmd.c

@@ -1,6 +1,8 @@
 // SPDX-License-Identifier: MIT or LGPL-2.1-only
 
 #include <config.h>
+#include <stdint.h>
+#include <linux/blkzoned.h>
 
 #include "ublksrv_priv.h"
 
@@ -51,39 +53,51 @@ static inline void ublksrv_ctrl_init_cmd(struct ublksrv_ctrl_dev *dev,
 			dev->dev_info.dev_id, data->cmd_op, cmd);
 }
 
-static int __ublksrv_ctrl_cmd(struct ublksrv_ctrl_dev *dev,
-		struct ublksrv_ctrl_cmd_data *data)
+static int __ublksrv_ctrl_cmd_with_ring(struct ublksrv_ctrl_dev *dev,
+		struct io_uring *ring, struct ublksrv_ctrl_cmd_data *data,
+		uint64_t *cqe_data)
 {
 	struct io_uring_sqe *sqe;
 	struct io_uring_cqe *cqe;
 	int ret = -EINVAL;
 
-	sqe = io_uring_get_sqe(&dev->ring);
+	sqe = io_uring_get_sqe(ring);
 	if (!sqe) {
 		fprintf(stderr, "can't get sqe ret %d\n", ret);
 		return ret;
 	}
 
 	ublksrv_ctrl_init_cmd(dev, sqe, data);
 
-	ret = io_uring_submit(&dev->ring);
+	ret = io_uring_submit(ring);
 	if (ret < 0) {
 		fprintf(stderr, "uring submit ret %d\n", ret);
 		return ret;
 	}
 
-	ret = io_uring_wait_cqe(&dev->ring, &cqe);
+	ret = io_uring_wait_cqe(ring, &cqe);
 	if (ret < 0) {
 		fprintf(stderr, "wait cqe: %s\n", strerror(-ret));
 		return ret;
 	}
-	io_uring_cqe_seen(&dev->ring, cqe);
+
+	if (cqe_data) {
+		*cqe_data = cqe->big_cqe[0];
+	}
+
+	io_uring_cqe_seen(ring, cqe);
 
 	ublksrv_printf(stdout, "dev %d, ctrl cqe res %d, user_data %llx\n",
 			dev->dev_info.dev_id, cqe->res, cqe->user_data);
 	return cqe->res;
 }
 
+static int __ublksrv_ctrl_cmd(struct ublksrv_ctrl_dev *dev,
+		struct ublksrv_ctrl_cmd_data *data)
+{
+	return __ublksrv_ctrl_cmd_with_ring(dev, &dev->ring, data, NULL);
+}
+
 void ublksrv_ctrl_deinit(struct ublksrv_ctrl_dev *dev)
 {
 	close(dev->ring.ring_fd);
@@ -160,6 +174,33 @@ int ublksrv_ctrl_get_affinity(struct ublksrv_ctrl_dev *ctrl_dev)
 	return 0;
 }
 
+int ublksrv_ctrl_report_zones_fetch(const struct ublksrv_ctrl_dev *ctrl_dev,
+		struct io_uring *ring,
+		uint64_t *cqe_data) {
+
+	struct ublksrv_ctrl_cmd_data data = {
+		.cmd_op = UBLK_CMD_REPORT_ZONES_FETCH,
+	};
+
+	return __ublksrv_ctrl_cmd_with_ring(ctrl_dev, ring, &data, cqe_data);
+}
+
+int ublksrv_ctrl_report_zones_commit_and_fetch(
+		const struct ublksrv_ctrl_dev *ctrl_dev,
+		struct io_uring *ring,
+		struct blk_zone *zone_info,
+		uint64_t *cqe_data) {
+
+	struct ublksrv_ctrl_cmd_data data = {
+		.cmd_op = UBLK_CMD_REPORT_ZONES_COMMIT_FETCH,
+		.flags = CTRL_CMD_HAS_BUF,
+		.addr = (__u64)zone_info,
+		.len = sizeof(struct blk_zone),
+	};
+
+	return __ublksrv_ctrl_cmd_with_ring(ctrl_dev, ring, &data, cqe_data);
+}
+
 /*
  * Start the ublksrv device:
  *

lib/ublksrv_json.cpp

@@ -97,8 +97,14 @@ NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE_WITH_DEFAULT(struct ublk_param_discard,
 	max_write_zeroes_sectors,
 	max_discard_segments,
 	reserved0)
+
+NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE_WITH_DEFAULT(struct ublk_param_zoned,
+	max_open_zones,
+	max_active_zones,
+	max_append_size)
+
 NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE_WITH_DEFAULT(struct ublk_params,
-	len, types, basic, discard)
+	len, types, basic, discard, zoned)
 
 int ublksrv_json_write_params(const struct ublk_params *p,
 		char *jbuf, int len)

tgt_loop.cpp

@@ -2,7 +2,9 @@
 
 #include <config.h>
 
+#include <linux/blkzoned.h>
 #include <poll.h>
+#include <stdint.h>
 #include <sys/epoll.h>
 #include "ublksrv_aio.h"
 #include "ublksrv_tgt.h"
@@ -37,6 +39,8 @@ static bool backing_supports_discard(char *name)
 	return false;
 }
 
+extern struct ublksrv_tgt_type loop_ctrl_tgt_type;
+
 static int loop_init_tgt(struct ublksrv_dev *dev, int type, int argc, char
 		*argv[])
 {
@@ -98,6 +102,30 @@ static int loop_init_tgt(struct ublksrv_dev *dev, int type, int argc, char
 		return -2;
 	}
 
+	if (dev->ctrl_dev->dev_info.flags & UBLK_F_ZONED) {
+		uint32_t zone_size;
+		if (ioctl(fd, BLKGETZONESZ, &zone_size)) {
+			syslog(LOG_ERR, "%s: BLKGETSONESZ ioctl failed for %s\n",
+				__func__, file);
+			return -1;
+		}
+
+		if (zone_size == 0) {
+			syslog(LOG_ERR, "%s: target %s is not zoned\n",
+				__func__, file);
+			return -1;
+		}
+
+		p.basic.chunk_sectors = zone_size;
+		p.types |= UBLK_PARAM_TYPE_ZONED;
+		/* TODO: lift these values from loop target */
+		p.zoned.max_open_zones = 14;
+		p.zoned.max_active_zones = 14;
+		// TODO: Dropping const qualifier
+		((struct ublksrv_ctrl_dev *)dev->ctrl_dev)->zone_size_sectors =
+			zone_size;
+	}
+
 	if (fstat(fd, &st) < 0)
 		return -2;
 
@@ -166,6 +194,10 @@ static int loop_init_tgt(struct ublksrv_dev *dev, int type, int argc, char
 			jbuf = ublksrv_tgt_realloc_json_buf(dev, &jbuf_size);
 	} while (ret < 0);
 
+	// TODO: Dropping const qualifier
+	((struct ublksrv_ctrl_dev *)dev->ctrl_dev)->tgt_ops =
+		&loop_ctrl_tgt_type;
+
 	return 0;
 }
 
@@ -256,6 +288,37 @@ static co_io_job __loop_handle_io_async(struct ublksrv_queue *q,
 	struct io_uring_cqe *cqe;
 	int ret;
 
+	const struct ublksrv_io_desc *iod = ublksrv_get_iod(q, tag);
+	unsigned ublk_op = ublksrv_get_op(iod);
+	unsigned long ioctl_op = 0;
+
+	switch (ublk_op) {
+	case UBLK_IO_OP_ZONE_OPEN:
+		ioctl_op = BLKOPENZONE;
+		break;
+	case UBLK_IO_OP_ZONE_CLOSE:
+		ioctl_op = BLKCLOSEZONE;
+		break;
+	case UBLK_IO_OP_ZONE_FINISH:
+		ioctl_op = BLKFINISHZONE;
+		break;
+	case UBLK_IO_OP_ZONE_RESET:
+		ioctl_op = BLKRESETZONE;
+		break;
+	}
+
+	if (ioctl_op) {
+		struct blk_zone_range range = {
+			.sector = iod->start_sector,
+			.nr_sectors = q->dev->ctrl_dev->zone_size_sectors,
+		};
+
+		ret = ioctl(q->dev->tgt.fds[1], ioctl_op, &range);
+
+		ublksrv_complete_io(q, tag, ret);
+		co_return;
+	}
+
 	io->queued_tgt_io = 0;
  again:
 	ret = loop_queue_tgt_io(q, io, tag);
@@ -310,6 +373,43 @@ static void loop_deinit_tgt(struct ublksrv_dev *dev)
 	free(dev->tgt.tgt_data);
 }
 
+static void loop_report_zone(struct ublksrv_dev *dev, uint64_t offset_sectors,
+	struct blk_zone *zone_info)
+{
+	struct {
+		struct blk_zone_report report;
+		struct blk_zone zone;
+	} request = {
+		.report = {
+			.sector = offset_sectors,
+			.nr_zones = 1,
+		},
+	};
+
+	if (ioctl(dev->tgt.fds[1], BLKREPORTZONE, &request)) {
+		syslog(LOG_ERR, "%s: BLKREPORTZONE failed\n", __func__);
+		// TODO: Report error
+		return;
+	}
+
+	if (request.report.nr_zones == 0) {
+		/* Reporting zero length zone to indicate end */
+		memset(zone_info, 0, sizeof(*zone_info));
+		return;
+	}
+
+	if (request.report.nr_zones > 1) {
+		syslog(LOG_ERR,
+			"%s: BLKREPORTZONE reported wrong number of "
+			"zones: %d\n",
+			__func__, request.report.nr_zones);
+		// TODO: Report error
+		return;
+	}
+
+	memcpy(zone_info, &request.zone, sizeof(*zone_info));
+}
+
 struct ublksrv_tgt_type  loop_tgt_type = {
 	.handle_io_async = loop_handle_io_async,
 	.tgt_io_done = loop_tgt_io_done,
@@ -320,6 +420,10 @@ struct ublksrv_tgt_type  loop_tgt_type = {
 	.name	=  "loop",
 };
 
+struct ublksrv_tgt_type loop_ctrl_tgt_type = {
+	.report_zone = loop_report_zone,
+};
+
 static void tgt_loop_init() __attribute__((constructor));
 
 static void tgt_loop_init(void)