[RFC PATCH] dm: add deviceless messages

[RFC PATCH] dm: add deviceless messages

[Mikulas Patocka]

The VDO code needs to perform some calculations before creating a device. 
The calculations should be performed in the kernel, so that the same logic 
is not duplicated in the kernel and in the userspace.

However, currently it is not possible to call a device mapper target 
when no device exists.

This commit solves the problem by adding "deviceless" messages. If the 
message starts with '%', the name is a name of a target, not a name of a 
device. The message will be directed to the target by calling the 
"deviceless_message" method. The target can perform the requested 
calculations in this method and return the result back to userspace.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>

---
 drivers/md/dm-ioctl.c         |   57 +++++++++++++++++++++++++++++++++---------
 include/linux/device-mapper.h |    4 ++
 2 files changed, 49 insertions(+), 12 deletions(-)

Index: linux-2.6/drivers/md/dm-ioctl.c
===================================================================
--- linux-2.6.orig/drivers/md/dm-ioctl.c	2026-05-06 22:15:36.000000000 +0200
+++ linux-2.6/drivers/md/dm-ioctl.c	2026-05-11 12:53:37.000000000 +0200
@@ -1743,6 +1743,33 @@ static int table_status(struct file *fil
 	return 0;
 }
 
+static int message_for_module(const char *target_name, unsigned int argc, char **argv,
+			      char *result, unsigned int maxlen)
+{
+	struct target_type *tt;
+	int r;
+
+	if (**argv != '%')
+		return 2;
+
+	tt = dm_get_target_type(target_name);
+	if (!tt) {
+		DMERR("unknown target name %s", target_name);
+		return -EINVAL;
+	}
+
+	if (!tt->deviceless_message) {
+		DMERR("target %s doesn't support deviceless messages", target_name);
+		r = -EINVAL;
+	} else {
+		r = tt->deviceless_message(argc, argv, result, maxlen);
+	}
+
+	dm_put_target_type(tt);
+
+	return r;
+}
+
 /*
  * Process device-mapper dependent messages.  Messages prefixed with '@'
  * are processed by the DM core.  All others are delivered to the target.
@@ -1788,10 +1815,6 @@ static int target_message(struct file *f
 	char *result = get_result_buffer(param, param_size, &maxlen);
 	int srcu_idx;
 
-	md = find_device(param);
-	if (!md)
-		return -ENXIO;
-
 	if (tmsg < (struct dm_target_msg *) param->data ||
 	    invalid_str(tmsg->message, (void *) param + param_size)) {
 		DMERR("Invalid target message parameters.");
@@ -1811,9 +1834,19 @@ static int target_message(struct file *f
 		goto out_argv;
 	}
 
-	r = message_for_md(md, argc, argv, result, maxlen);
+	r = message_for_module(param->name, argc, argv, result, maxlen);
 	if (r <= 1)
+		goto out_deviceless;
+
+	md = find_device(param);
+	if (!md) {
+		r = -ENXIO;
 		goto out_argv;
+	}
+
+	r = message_for_md(md, argc, argv, result, maxlen);
+	if (r <= 1)
+		goto out_md;
 
 	table = dm_get_live_table(md, &srcu_idx);
 	if (!table) {
@@ -1838,14 +1871,13 @@ static int target_message(struct file *f
 		r = -EINVAL;
 	}
 
- out_table:
+out_table:
 	dm_put_live_table(md, srcu_idx);
- out_argv:
-	kfree(argv);
- out:
+out_md:
 	if (r >= 0)
 		__dev_status(md, param);
-
+	dm_put(md);
+out_deviceless:
 	if (r == 1) {
 		param->flags |= DM_DATA_OUT_FLAG;
 		if (dm_message_test_buffer_overflow(result, maxlen))
@@ -1854,8 +1886,9 @@ static int target_message(struct file *f
 			param->data_size = param->data_start + strlen(result) + 1;
 		r = 0;
 	}
-
-	dm_put(md);
+out_argv:
+	kfree(argv);
+out:
 	return r;
 }
 
Index: linux-2.6/include/linux/device-mapper.h
===================================================================
--- linux-2.6.orig/include/linux/device-mapper.h	2026-04-27 19:06:34.000000000 +0200
+++ linux-2.6/include/linux/device-mapper.h	2026-05-11 12:45:45.000000000 +0200
@@ -93,6 +93,9 @@ typedef void (*dm_status_fn) (struct dm_
 typedef int (*dm_message_fn) (struct dm_target *ti, unsigned int argc, char **argv,
 			      char *result, unsigned int maxlen);
 
+typedef int (*dm_deviceless_message_fn) (unsigned int argc, char **argv,
+			      char *result, unsigned int maxlen);
+
 /*
  * Called with *forward == true. If it remains true, the ioctl should be
  * forwarded to bdev. If it is reset to false, the target already fully handled
@@ -214,6 +217,7 @@ struct target_type {
 	dm_resume_fn resume;
 	dm_status_fn status;
 	dm_message_fn message;
+	dm_deviceless_message_fn deviceless_message;
 	dm_prepare_ioctl_fn prepare_ioctl;
 	dm_report_zones_fn report_zones;
 	dm_busy_fn busy;

Re: [RFC PATCH] dm: add deviceless messages

[Mikulas Patocka]

On Tue, 12 May 2026, Mikulas Patocka wrote:

> The VDO code needs to perform some calculations before creating a device. 
> The calculations should be performed in the kernel, so that the same logic 
> is not duplicated in the kernel and in the userspace.
> 
> However, currently it is not possible to call a device mapper target 
> when no device exists.
> 
> This commit solves the problem by adding "deviceless" messages. If the 
> message starts with '%', the name is a name of a target, not a name of a 
> device. The message will be directed to the target by calling the 
> "deviceless_message" method. The target can perform the requested 
> calculations in this method and return the result back to userspace.
> 
> Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>

This is example code how to use deviceless messages. You can call it with 
"dmsetup message linear 0 %Hello"

Mikulas

---
 drivers/md/dm-linear.c |   10 ++++++++++
 1 file changed, 10 insertions(+)

Index: linux-2.6/drivers/md/dm-linear.c
===================================================================
--- linux-2.6.orig/drivers/md/dm-linear.c	2026-05-11 13:00:08.000000000 +0200
+++ linux-2.6/drivers/md/dm-linear.c	2026-05-11 13:00:14.000000000 +0200
@@ -199,6 +199,15 @@ static size_t linear_dax_recovery_write(
 #define linear_dax_recovery_write NULL
 #endif
 
+static int linear_deviceless_message(unsigned int argc, char **argv, char *result, unsigned int maxlen)
+{
+	if (argc == 1 && !strcasecmp(argv[0], "%Hello")) {
+		snprintf(result, maxlen, "World!");
+		return 1;
+	}
+	return -EINVAL;
+}
+
 static struct target_type linear_target = {
 	.name   = "linear",
 	.version = {1, 5, 0},
@@ -211,6 +220,7 @@ static struct target_type linear_target
 	.dtr    = linear_dtr,
 	.map    = linear_map,
 	.status = linear_status,
+	.deviceless_message = linear_deviceless_message,
 	.prepare_ioctl = linear_prepare_ioctl,
 	.iterate_devices = linear_iterate_devices,
 	.direct_access = linear_dax_direct_access,

Re: [RFC PATCH] dm: add deviceless messages

[Benjamin Marzinski]

On Tue, May 12, 2026 at 03:36:01PM +0200, Mikulas Patocka wrote:
> The VDO code needs to perform some calculations before creating a device. 
> The calculations should be performed in the kernel, so that the same logic 
> is not duplicated in the kernel and in the userspace.
> 
> However, currently it is not possible to call a device mapper target 
> when no device exists.
> 
> This commit solves the problem by adding "deviceless" messages. If the 
> message starts with '%', the name is a name of a target, not a name of a 
> device. The message will be directed to the target by calling the 
> "deviceless_message" method. The target can perform the requested 
> calculations in this method and return the result back to userspace.
> 
> Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>

Reviewed-by: Benjamin Marzinski <bmarzins@redhat.com>

Re: [RFC PATCH] dm: add deviceless messages

[Matthew Sakai]

On 5/12/26 9:36 AM, Mikulas Patocka wrote:
> The VDO code needs to perform some calculations before creating a device.
> The calculations should be performed in the kernel, so that the same logic
> is not duplicated in the kernel and in the userspace.
> 
> However, currently it is not possible to call a device mapper target
> when no device exists.
> 
> This commit solves the problem by adding "deviceless" messages. If the
> message starts with '%', the name is a name of a target, not a name of a
> device. The message will be directed to the target by calling the
> "deviceless_message" method. The target can perform the requested
> calculations in this method and return the result back to userspace.
> 
> Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>

I do wish that we could clean up the message interface. The sector 
number argument is completely meaningless for a deviceless message, so 
in theory we could distinguish this case by omitting the argument, e.g. 
"dmsetup message <target_type> <command>" and not require the special 
'%' notation, either. But that looks like it would require a lot more 
rearrangement of the message handling code.

I agree this does what we need, so...

Reviewed-by: Matthew Sakai <msakai@redhat.com>

> ---
>   drivers/md/dm-ioctl.c         |   57 +++++++++++++++++++++++++++++++++---------
>   include/linux/device-mapper.h |    4 ++
>   2 files changed, 49 insertions(+), 12 deletions(-)
> 
> Index: linux-2.6/drivers/md/dm-ioctl.c
> ===================================================================
> --- linux-2.6.orig/drivers/md/dm-ioctl.c	2026-05-06 22:15:36.000000000 +0200
> +++ linux-2.6/drivers/md/dm-ioctl.c	2026-05-11 12:53:37.000000000 +0200
> @@ -1743,6 +1743,33 @@ static int table_status(struct file *fil
>   	return 0;
>   }
>   
> +static int message_for_module(const char *target_name, unsigned int argc, char **argv,
> +			      char *result, unsigned int maxlen)
> +{
> +	struct target_type *tt;
> +	int r;
> +
> +	if (**argv != '%')
> +		return 2;
> +
> +	tt = dm_get_target_type(target_name);
> +	if (!tt) {
> +		DMERR("unknown target name %s", target_name);
> +		return -EINVAL;
> +	}
> +
> +	if (!tt->deviceless_message) {
> +		DMERR("target %s doesn't support deviceless messages", target_name);
> +		r = -EINVAL;
> +	} else {
> +		r = tt->deviceless_message(argc, argv, result, maxlen);
> +	}
> +
> +	dm_put_target_type(tt);
> +
> +	return r;
> +}
> +
>   /*
>    * Process device-mapper dependent messages.  Messages prefixed with '@'
>    * are processed by the DM core.  All others are delivered to the target.
> @@ -1788,10 +1815,6 @@ static int target_message(struct file *f
>   	char *result = get_result_buffer(param, param_size, &maxlen);
>   	int srcu_idx;
>   
> -	md = find_device(param);
> -	if (!md)
> -		return -ENXIO;
> -
>   	if (tmsg < (struct dm_target_msg *) param->data ||
>   	    invalid_str(tmsg->message, (void *) param + param_size)) {
>   		DMERR("Invalid target message parameters.");
> @@ -1811,9 +1834,19 @@ static int target_message(struct file *f
>   		goto out_argv;
>   	}
>   
> -	r = message_for_md(md, argc, argv, result, maxlen);
> +	r = message_for_module(param->name, argc, argv, result, maxlen);
>   	if (r <= 1)
> +		goto out_deviceless;
> +
> +	md = find_device(param);
> +	if (!md) {
> +		r = -ENXIO;
>   		goto out_argv;
> +	}
> +
> +	r = message_for_md(md, argc, argv, result, maxlen);
> +	if (r <= 1)
> +		goto out_md;
>   
>   	table = dm_get_live_table(md, &srcu_idx);
>   	if (!table) {
> @@ -1838,14 +1871,13 @@ static int target_message(struct file *f
>   		r = -EINVAL;
>   	}
>   
> - out_table:
> +out_table:
>   	dm_put_live_table(md, srcu_idx);
> - out_argv:
> -	kfree(argv);
> - out:
> +out_md:
>   	if (r >= 0)
>   		__dev_status(md, param);
> -
> +	dm_put(md);
> +out_deviceless:
>   	if (r == 1) {
>   		param->flags |= DM_DATA_OUT_FLAG;
>   		if (dm_message_test_buffer_overflow(result, maxlen))
> @@ -1854,8 +1886,9 @@ static int target_message(struct file *f
>   			param->data_size = param->data_start + strlen(result) + 1;
>   		r = 0;
>   	}
> -
> -	dm_put(md);
> +out_argv:
> +	kfree(argv);
> +out:
>   	return r;
>   }
>   
> Index: linux-2.6/include/linux/device-mapper.h
> ===================================================================
> --- linux-2.6.orig/include/linux/device-mapper.h	2026-04-27 19:06:34.000000000 +0200
> +++ linux-2.6/include/linux/device-mapper.h	2026-05-11 12:45:45.000000000 +0200
> @@ -93,6 +93,9 @@ typedef void (*dm_status_fn) (struct dm_
>   typedef int (*dm_message_fn) (struct dm_target *ti, unsigned int argc, char **argv,
>   			      char *result, unsigned int maxlen);
>   
> +typedef int (*dm_deviceless_message_fn) (unsigned int argc, char **argv,
> +			      char *result, unsigned int maxlen);
> +
>   /*
>    * Called with *forward == true. If it remains true, the ioctl should be
>    * forwarded to bdev. If it is reset to false, the target already fully handled
> @@ -214,6 +217,7 @@ struct target_type {
>   	dm_resume_fn resume;
>   	dm_status_fn status;
>   	dm_message_fn message;
> +	dm_deviceless_message_fn deviceless_message;
>   	dm_prepare_ioctl_fn prepare_ioctl;
>   	dm_report_zones_fn report_zones;
>   	dm_busy_fn busy;
> 

Re: [RFC PATCH] dm: add deviceless messages

[Mikulas Patocka]

On Tue, 12 May 2026, Matthew Sakai wrote:

> 
> 
> On 5/12/26 9:36 AM, Mikulas Patocka wrote:
> > The VDO code needs to perform some calculations before creating a device.
> > The calculations should be performed in the kernel, so that the same logic
> > is not duplicated in the kernel and in the userspace.
> > 
> > However, currently it is not possible to call a device mapper target
> > when no device exists.
> > 
> > This commit solves the problem by adding "deviceless" messages. If the
> > message starts with '%', the name is a name of a target, not a name of a
> > device. The message will be directed to the target by calling the
> > "deviceless_message" method. The target can perform the requested
> > calculations in this method and return the result back to userspace.
> > 
> > Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
> 
> I do wish that we could clean up the message interface. The sector number
> argument is completely meaningless for a deviceless message, so in theory we
> could distinguish this case by omitting the argument, e.g. "dmsetup message
> <target_type> <command>" and not require the special '%' notation, either. But
> that looks like it would require a lot more rearrangement of the message
> handling code.

I wanted to design it in such a way that we don't have to update the 
libdevmapper library or the dmsetup tool to send deviceless messages - 
that's why there is the unused sector number.

In the past, dm_stats was glueued to messages in a similar way (the 
dm_stats messages start with '@'), the sector number is also unused in 
this case.

Perhaps we could introduce a new ioctl DM_DEVICELESS_MSG_CMD and send 
deviceless messages with it, but that would need coordination with the 
libdevmapper+dmsetup maintainers. I think it would create maintenance 
burden (you couldn't use deviceless messages unless you update 
libdevmapper, you couldn't update libdevmapper unless you update libc, you 
couldn't update libc unless you update the whole distribution ... etc.).

> I agree this does what we need, so...
> 
> Reviewed-by: Matthew Sakai <msakai@redhat.com>

When you make a patch that needs deviceless messages, let me know and I 
will commit it. I will not commit deviceless messages now, because I want 
to have someone using it when committing it.

Mikulas