I did simulation of that DM_BUFFER_FULL_FLAG (I modified the kernel code
to return this spuriously) and found out that there are no userspace
problems with it.
The userspace code detects this flag in the function "dm_task_run",
doubles the buffer size and jumps back to "repeat_ioctl:" label. A
function "_do_dm_ioctl" is called from there, this function constructs
"struct dm_ioctl *dmi" from scratch, so there is no possibility that
returned content of the structure from the previous call leaks into the
Do you want still to modify the kernel to accept multiple device
designators if they refer to the same device? For handling
DM_BUFFER_FULL_FLAG, it is not needed.
BTW. there is an unrelated feature of dmestup --- if you specify multiple
device designators, it ignores the others --- for example, the command:
"dmsetup info vg1-thinp_pool -u
LVM-L95tc8BLZ7tHTFc8uIZOdiR6usFzyHTOiAA3JkKgjH1DNw63Wy zXHx9RbVftrgox -j
254 -m 5"
actually passes only "vg1-thinp_pool" to the kernel driver, it ignores the
uuid and major/minor numbers. Do you think it should be fixed?
dm-devel mailing list