*** Description of problem:

In dosfstools-4.2 a new option appeared, called "--mbr". Under circumstances that matter much to libguestfs, namely when the disk being formatted is (a) non-removable and (b) unpartitioned (that is, a whole disk, not a partition), the "--mbr" option defaults to "yes". (For the sake of the reproducer below, we'll use an explicit "--mbr=yes".) When this option is in effect, "mkfs.fat" overlays the first sector of the medium (the FAT boot sector) with a bogus MBR partition table. The table has one entry, and it describes a partition that spans the entire disk, covering the partition table itself, resident in the first sector of the medium. This is supposed to placate Windows (otherwise Windows does not recognize the FAT filesystem on a non-removable device), and to work with Linux utilities as well (the kernel itself, and blkid, and so on). However, in case the chosen FAT variant is FAT12, parted rejects the partition table with the error message "Invalid partition table - recursive partition".

*** Version-Release number of selected component (if applicable):

- parted-3.4-6.el9.x86_64
  (it should reproduce with upstream parted too, as of commit b20227adf575)

- dosfstools-4.2-3.el9.x86_64

*** How reproducible:

Always.

*** Steps to Reproduce:

(1) rm -f image
    fallocate -l 1M image
    mkfs.fat --mbr=yes image
    parted -m -s -- image unit b print 

(2) rm -f image
    fallocate -l 255M image
    mkfs.fat -F 12 --mbr=yes image
    parted -m -s -- image unit b print 

(In both cases, FAT12 is used. In the first case, "mkfs.fat" picks FAT12 automatically, based on the size of the image. In the second case, we force FAT12.)

*** Actual results:

parted prints the following:

(1)

(stderr) Error: Invalid partition table - recursive partition on /root/image.
(stdout) BYT;
(stdout) /root/image:1048576B:file:512:512:unknown::;

(2)

(stderr) Error: Invalid partition table - recursive partition on /root/image.
(stdout) BYT;
(stdout) /root/image:267386880B:file:512:512:unknown::;

and parted exits with status 1. No partition information is listed.

*** Expected results:

In case (2), the disk image (255MB) is large enough for us to force either FAT16 or FAT32 (passing "-F 16" and "-F 32" to "mkfs.fat", respectively). In those cases, the partition output is [almost] as expected:

(2 / FAT16)

(stdout) BYT;
(stdout) /root/image:267386880B:file:512:512:loop::;
(stdout) 1:0B:267386879B:267386880B:fat16::;

(2 / FAT32)

(stdout) BYT;
(stdout) /root/image:267386880B:file:512:512:loop::;
(stdout) 1:0B:267386879B:267386880B:fat32::;

(And exit status 0 in both sub-cases.)

The same kind of output is expected with FAT12 as well.

Note: where I wrote "[almost] as expected" above, I alluded to the *separate bug* that "parted" prints the partition table *type* as "loop", instead of "mbr". I'm going to file a separate BZ about that.