Re: [PATCH v4] io_uring/register: add IORING_REGISTER_CLONE_FILES opcode

Harshal Chavan <[email protected]> writes:

> 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.
>
> Signed-off-by: Harshal Chavan <[email protected]>

Hello,

Do you have the liburing side and test cases?

A few comments inline.

> ---
>  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(+)
>
> diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
> index 909fb7aea638..0727602ce12f 100644
> --- a/include/uapi/linux/io_uring.h
> +++ b/include/uapi/linux/io_uring.h
> @@ -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

> +	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;
> diff --git a/io_uring/register.c b/io_uring/register.c
> index dce5e2f9cf77..bbc8c506ea2d 100644
> --- a/io_uring/register.c
> +++ b/io_uring/register.c
> @@ -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)
> diff --git a/io_uring/rsrc.c b/io_uring/rsrc.c
> index 650303626be6..a598e5af4c0a 100644
> --- a/io_uring/rsrc.c
> +++ b/io_uring/rsrc.c
> @@ -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)
> +{
> +	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?

> +
> +	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;
> +
> +	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);

> +	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;
> +	}
> +
> +	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.

-- 
Gabriel Krisman Bertazi