Skip to contentSkip to conversation

io_uring/register: add IORING_REGISTER_CLONE_FILES opcode

PATCH Superseded Harshal Chavan 2 participants 3 messages Jun 19 – Jun 23 io_uring

v4 of 7 Superseded Latest is v7 →

lore
Share & tools
lore
#0 lore raw

PATCH v4 1/1 io_uring/register: add IORING_REGISTER_CLONE_FILES opcode

Currently, if an application wants to duplicate registered file descriptors from one io_uring instance to another, it must manually unregister and re-register them, incurring unnecessary overhead.

Add IORING_REGISTER_CLONE_FILES to allow direct cloning of the file table from a source ring to a destination ring. This implementation strictly mirrors the io_clone_buffers UAPI, supporting partial offsets and the IORING_REGISTER_DST_REPLACE flag.

To ensure lock synchronization safety, destination nodes are strictly allocated as new, private io_rsrc_nodes rather than sharing references across rings.

---
v4:
- Updated Signed-off-by to use real name and moved above the scissors line (Greg KH).
v3:
- Rewrote the cloning loop to allocate private destination nodes via io_rsrc_node_alloc to fix non-atomic ref lock synchronization (Jens).
- Maintained partial offset/copy support to mirror io_clone_buffers UAPI (Jens).
- Gated the replacement free check on ctx->file_table.data.nr (Gabriel).
- Prevented self-cloning by checking ctx == src_ctx (Gabriel).
- Removed submitter_task check to allow cross-thread pooling setups (Gabriel).
v2:
- Dropped unrelated whitespace formatting changes from v1

---
 include/uapi/linux/io_uring.h |  12 +++
 io_uring/register.c           |   6 ++
 io_uring/rsrc.c               | 149 ++++++++++++++++++++++++++++++++++
 io_uring/rsrc.h               |   1 +
 4 files changed, 168 insertions(+)
include/uapi/linux/io_uring.h +12 -0
index 909fb7aea638..0727602ce12f 100644
@@ -723,6 +723,9 @@ enum io_uring_register_op {
 	/* register bpf filtering programs */
 	IORING_REGISTER_BPF_FILTER		= 37,
 
+	/* clone file descriptors from another ring*/
+	IORING_REGISTER_CLONE_FILES		= 38,
+
 	/* this goes last */
 	IORING_REGISTER_LAST,
 
@@ -854,6 +857,15 @@ struct io_uring_clone_buffers {
 	__u32	pad[3];
 };
 
+struct io_uring_clone_files {
+	__u32 src_fd;
+	__u32 flags;
+	__u32 src_off;
+	__u32 dst_off;
+	__u32 nr;
+	__u32 pad[3];
+};
+
 struct io_uring_buf {
 	__u64	addr;
 	__u32	len;
io_uring/register.c +6 -0
index dce5e2f9cf77..bbc8c506ea2d 100644
@@ -924,6 +924,12 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
 			break;
 		ret = io_register_clone_buffers(ctx, arg);
 		break;
+	case IORING_REGISTER_CLONE_FILES:
+		ret = -EINVAL;
+		if (!arg || nr_args != 1)
+			break;
+		ret = io_register_clone_files(ctx, arg);
+		break;
 	case IORING_REGISTER_ZCRX_IFQ:
 		ret = -EINVAL;
 		if (!arg || nr_args != 1)
io_uring/rsrc.c +149 -0

155 lines not on this page · raw message

io_uring/rsrc.h +1 -0
index 44e3386f7c1c..32f5c47c46af 100644
@@ -75,6 +75,7 @@ int io_prep_reg_iovec(struct io_kiocb *req, struct iou_vec *iv,
 			const struct iovec __user *uvec, size_t uvec_segs);
 
 int io_register_clone_buffers(struct io_ring_ctx *ctx, void __user *arg);
+int io_register_clone_files(struct io_ring_ctx *ctx, void __user *arg);
 int io_sqe_buffers_unregister(struct io_ring_ctx *ctx);
 int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg,
 			    unsigned int nr_args, u64 __user *tags);
Signature · 1 line
2.54.0
#1 lore raw
Quoted 9 lines · Harshal Chavan

Currently, if an application wants to duplicate registered file descriptors from one io_uring instance to another, it must manually unregister and re-register them, incurring unnecessary overhead.

Add IORING_REGISTER_CLONE_FILES to allow direct cloning of the file table from a source ring to a destination ring. This implementation strictly mirrors the io_clone_buffers UAPI, supporting partial offsets and the IORING_REGISTER_DST_REPLACE flag.

Quoted 5 lines

To ensure lock synchronization safety, destination nodes are strictly allocated as new, private io_rsrc_nodes rather than sharing references across rings.

Hello,

Do you have the liburing side and test cases?

A few comments inline.

Quoted 12 lines
---
 include/uapi/linux/io_uring.h |  12 +++
 io_uring/register.c           |   6 ++
 io_uring/rsrc.c               | 149 ++++++++++++++++++++++++++++++++++
 io_uring/rsrc.h               |   1 +
 4 files changed, 168 insertions(+)
include/uapi/linux/io_uring.h +0 -0
index 909fb7aea638..0727602ce12f 100644

include/uapi/linux/io_uring.h enum io_uring_register_op { · line 726

Quoted 4 lines
include/uapi/linux/io_uring.h +1 -0
@@ -723,6 +723,9 @@ enum io_uring_register_op {
 	/* register bpf filtering programs */
 	IORING_REGISTER_BPF_FILTER		= 37,
 
+	/* clone file descriptors from another ring*/

^ spacing

Quoted 71 lines
include/uapi/linux/io_uring.h +11 -0
+	IORING_REGISTER_CLONE_FILES		= 38,
+
 	/* this goes last */
 	IORING_REGISTER_LAST,
 
@@ -854,6 +857,15 @@ struct io_uring_clone_buffers {
 	__u32	pad[3];
 };
 
+struct io_uring_clone_files {
+	__u32 src_fd;
+	__u32 flags;
+	__u32 src_off;
+	__u32 dst_off;
+	__u32 nr;
+	__u32 pad[3];
+};
+
 struct io_uring_buf {
 	__u64	addr;
 	__u32	len;
io_uring/register.c +6 -0
index dce5e2f9cf77..bbc8c506ea2d 100644
@@ -924,6 +924,12 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
 			break;
 		ret = io_register_clone_buffers(ctx, arg);
 		break;
+	case IORING_REGISTER_CLONE_FILES:
+		ret = -EINVAL;
+		if (!arg || nr_args != 1)
+			break;
+		ret = io_register_clone_files(ctx, arg);
+		break;
 	case IORING_REGISTER_ZCRX_IFQ:
 		ret = -EINVAL;
 		if (!arg || nr_args != 1)
io_uring/rsrc.c +25 -0
index 650303626be6..a598e5af4c0a 100644
@@ -1303,6 +1303,155 @@ int io_register_clone_buffers(struct io_ring_ctx *ctx, void __user *arg)
 	return ret;
 }
 
+static int io_clone_file_node(struct io_ring_ctx *ctx,
+			      struct io_rsrc_node *src_node,
+			      int dst_index,
+			      struct io_file_table *new_table)
+{
+	struct io_rsrc_node *dst_node;
+	struct file *file;
+
+	dst_node = io_rsrc_node_alloc(ctx, IORING_RSRC_FILE);
+	if (!dst_node)
+		return -ENOMEM;
+
+	file = io_slot_file(src_node);
+	get_file(file);
+	io_fixed_file_set(dst_node, file);
+
+	new_table->data.nodes[dst_index] = dst_node;
+	io_file_bitmap_set(new_table, dst_index);
+
+	return 0;
+}
+
+static int io_clone_files(struct io_ring_ctx *ctx, struct io_ring_ctx *src_ctx,
+			  struct io_uring_clone_files *arg)
+{

io_uring/rsrc.c int io_register_clone_buffers(struct io_ring_ctx *ctx, void __user *arg) · line 1340

Quoted 10 lines
io_uring/rsrc.c +10 -0
+	struct io_file_table new_file_table;
+	unsigned int dst_nr = ctx->file_table.data.nr;
+	unsigned int src_nr = src_ctx->file_table.data.nr;
+	unsigned int new_nr, i;
+
+	lockdep_assert_held(&ctx->uring_lock);
+	lockdep_assert_held(&src_ctx->uring_lock);
+
+	if (ctx->user != src_ctx->user || ctx->mm_account != src_ctx->mm_account)
+		return -EINVAL;

I don't think it makes sense to check ->user here. But is mm_account necessary either? How could you get the src_ctx from another process?

Quoted 79 lines
io_uring/rsrc.c +79 -0
+
+	if (dst_nr && !(arg->flags & IORING_REGISTER_DST_REPLACE))
+		return -EBUSY;
+
+	if (!src_nr)
+		return -ENXIO;
+
+	if (!arg->nr)
+		arg->nr = src_nr;
+	else if (arg->nr > src_nr)
+		return -EINVAL;
+
+	if (check_add_overflow(arg->src_off, arg->nr, &i) || i > src_nr)
+		return -EINVAL;
+	if (check_add_overflow(arg->dst_off, arg->nr, &i))
+		return -EINVAL;
+
+	new_nr = max(dst_nr, arg->dst_off + arg->nr);
+	if (new_nr > IORING_MAX_FIXED_FILES)
+		return -EINVAL;
+
+	memset(&new_file_table, 0, sizeof(new_file_table));
+	if (!io_alloc_file_tables(ctx, &new_file_table, new_nr))
+		return -ENOMEM;
+
+	/* Copy original nodes from before the cloned range */
+	for (i = 0; i < min(arg->dst_off, dst_nr); i++) {
+		struct io_rsrc_node *src_node = io_rsrc_node_lookup(&ctx->file_table.data, i);
+
+		if (!src_node)
+			continue;
+		if (io_clone_file_node(ctx, src_node, i, &new_file_table))
+			goto out;
+	}
+
+	/* Copy the actual cloned range from the source ring */
+	for (i = 0; i < arg->nr; i++) {
+		struct io_rsrc_node *src_node = io_rsrc_node_lookup(&src_ctx->file_table.data,
+				arg->src_off + i);
+
+		if (!src_node)
+			continue;
+		if (io_clone_file_node(ctx, src_node, arg->dst_off + i, &new_file_table))
+			goto out;
+	}
+
+	/* Copy original nodes from after the cloned range */
+	for (i = arg->dst_off + arg->nr; i < dst_nr; i++) {
+		struct io_rsrc_node *src_node = io_rsrc_node_lookup(&ctx->file_table.data, i);
+
+		if (!src_node)
+			continue;
+		if (io_clone_file_node(ctx, src_node, i, &new_file_table))
+			goto out;
+	}
+
+	/* free the old file table if there is any data present */
+	if (dst_nr)
+		io_free_file_tables(ctx, &ctx->file_table);
+
+	WARN_ON_ONCE(ctx->file_table.data.nr);
+	ctx->file_table = new_file_table;
+	io_file_table_set_alloc_range(ctx, 0, ctx->file_table.data.nr);
+	return 0;
+
+out:
+	/* Error Path: Safely destroy whatever we partially built */
+	io_free_file_tables(ctx, &new_file_table);
+	return -ENOMEM;
+}
+
+int io_register_clone_files(struct io_ring_ctx *ctx, void __user *arg)
+{
+	struct io_uring_clone_files clone_arg;
+	struct io_ring_ctx *src_ctx;
+	bool registered_src;
+	struct file *file;
+	int ret;
+

io_uring/rsrc.c int io_register_clone_buffers(struct io_ring_ctx *ctx, void __user *arg) · line 1429

Quoted 10 lines
io_uring/rsrc.c +10 -0
+	if (copy_from_user(&clone_arg, arg, sizeof(clone_arg)))
+		return -EFAULT;
+	if (clone_arg.flags &
+	    ~(IORING_REGISTER_SRC_REGISTERED | IORING_REGISTER_DST_REPLACE))
+		return -EINVAL;
+
+	if (memchr_inv(clone_arg.pad, 0, sizeof(clone_arg.pad)))
+		return -EINVAL;
+
+	registered_src = (clone_arg.flags & IORING_REGISTER_SRC_REGISTERED) != 0;

This is better written as

registered_src = !!(clone_arg.flags & IORING_REGISTER_SRC_REGISTERED);

Quoted 9 lines
io_uring/rsrc.c +9 -0
+	file = io_uring_ctx_get_file(clone_arg.src_fd, registered_src);
+	if (IS_ERR(file))
+		return PTR_ERR(file);
+
+	src_ctx = file->private_data;
+	/* Same ring clone is not allowed */
+	if (src_ctx == ctx) {
+		ret = -EINVAL;
+		goto out;

io_uring/rsrc.c int io_register_clone_buffers(struct io_ring_ctx *ctx, void __user *arg) · line 1448

Quoted 10 lines
io_uring/rsrc.c +10 -0
+	}
+
+	mutex_unlock(&ctx->uring_lock);
+	lock_two_rings(ctx, src_ctx);
+
+	ret = io_clone_files(ctx, src_ctx, &clone_arg);
+
+out:
+	if (src_ctx != ctx)
+		mutex_unlock(&src_ctx->uring_lock);

Make the mutex_unlock unconditionally above the out label. It is never locked in the error context.

Signature · 1 line
Gabriel Krisman Bertazi
#2 lore raw
Quoted 5 lines · Gabriel Krisman Bertazi @ 2026-06-22 20:04 UTC writes:

Hello,

Do you have the liburing side and test cases?

A few comments inline.

Hello,
Yes I will update the liburing side with helper function and add appropriate test cases.

Quoted 2 lines

include/uapi/linux/io_uring.h enum io_uring_register_op { · line 726

Quoted 1 line
include/uapi/linux/io_uring.h +1 -0
+	/* clone file descriptors from another ring*/

^ spacing

Fixed in v5

Quoted 5 lines

Review

Quoted 2 lines
(quoted diff) +2 -0
+	if (ctx->user != src_ctx->user || ctx->mm_account != src_ctx->mm_account)
+		return -EINVAL;

I don't think it makes sense to check ->user here. But is mm_account necessary either? How could you get the src_ctx from another process?

Yes, Keeping this check would unnecessarily break valid use case like Root user passing FDs to guest user

Removed it completely in v5, thanks for catching this!.

Quoted 5 lines

io_uring/rsrc.c int io_register_clone_buffers(struct io_ring_ctx *ctx, void __user *arg) · line 1429

Quoted 1 line
io_uring/rsrc.c +1 -0
+	registered_src = (clone_arg.flags & IORING_REGISTER_SRC_REGISTERED) != 0;

This is better written as

registered_src = !!(clone_arg.flags & IORING_REGISTER_SRC_REGISTERED);

Understood, updated this in v5

Quoted 6 lines

io_uring/rsrc.c int io_register_clone_buffers(struct io_ring_ctx *ctx, void __user *arg) · line 1448

Quoted 3 lines
io_uring/rsrc.c +3 -0
+out:
+	if (src_ctx != ctx)
+		mutex_unlock(&src_ctx->uring_lock);

Make the mutex_unlock unconditionally above the out label. It is never locked in the error context.

Yes, moved the unlock statement before out without any conditions.

Thank you for the review.

Regards,
Harshal Chavan