[lvm-devel] [LVM2 PATCH] libdm: Fail to add tree node when requested major/minor is used.

[Dave Wysochanski]

On Wed, 2008-12-17 at 18:03 +0100, Milan Broz wrote:
> Fail to add tree node when requested major/minor is used.
> 
> (_info_by_dev() is just moved for definition before use here)
> 
> Fixes https://bugzilla.redhat.com/show_bug.cgi?id=204992
> 
> Signed-off-by: Milan Broz <mbroz at redhat.com>
> ---
>  libdm/libdm-deptree.c |   64 ++++++++++++++++++++++++++++----------------------
>  1 file changed, 36 insertions(+), 28 deletions(-)
> 
> Index: LVM2/libdm/libdm-deptree.c
> ===================================================================
> --- LVM2.orig/libdm/libdm-deptree.c	2008-12-17 16:28:33.000000000 +0100
> +++ LVM2/libdm/libdm-deptree.c	2008-12-17 17:39:51.000000000 +0100
> @@ -544,6 +544,35 @@ static int _node_clear_table(struct dm_t
>  	return r;
>  }
>  
> +static int _info_by_dev(uint32_t major, uint32_t minor, int with_open_count,
> +			struct dm_info *info)
> +{
> +	struct dm_task *dmt;
> +	int r;
> +
> +	if (!(dmt = dm_task_create(DM_DEVICE_INFO))) {
> +		log_error("_info_by_dev: dm_task creation failed");
> +		return 0;
> +	}
> +
> +	if (!dm_task_set_major(dmt, major) || !dm_task_set_minor(dmt, minor)) {
> +		log_error("_info_by_dev: Failed to set device number");
> +		dm_task_destroy(dmt);
> +		return 0;
> +	}
> +
> +	if (!with_open_count && !dm_task_no_open_count(dmt))
> +		log_error("Failed to disable open_count");
> +
> +	if ((r = dm_task_run(dmt)))
> +		r = dm_task_get_info(dmt, info);
> +
> +	dm_task_destroy(dmt);
> +
> +	return r;
> +}
> +
> +
>  struct dm_tree_node *dm_tree_add_new_dev(struct dm_tree *dtree,
>  					    const char *name,
>  					    const char *uuid,
> @@ -559,6 +588,13 @@ struct dm_tree_node *dm_tree_add_new_dev
>  
>  	/* Do we need to add node to tree? */
>  	if (!(dnode = dm_tree_find_node_by_uuid(dtree, uuid))) {
> +
> +		if (major && minor && _info_by_dev(major, minor, 0, &info) && info.exists) {
> +			log_error("Requested major:minor %i:%i number is already used.",
> +				  major, minor);
> +			return NULL;
> +		}
> +


Looks to me like this condition rarely/never is true because major is
always 253 for the devices created but the _info_by_dev() call uses the
major value on the cmdline.  When we create the device the creation
often still fails without explaining the real reason:
# tools/lvm lvcreate -n linear -My --major 200 --minor 75 -L 50M vgtest
  Rounding up size to full physical extent 52.00 MB
  device-mapper: create ioctl failed: No such device or address
  Aborting. Failed to activate new LV to wipe the start of it.
# tools/lvm lvcreate -n linear -My --major 253 --minor 75 -L 50M vgtest
  Rounding up size to full physical extent 52.00 MB
  Requested major:minor 253:75 number is already used.
  Aborting. Failed to activate new LV to wipe the start of it.


Should we be ignoring --major on the lvcreate cmdline, overriding it, or
is there another bug?