false physical blocksize detected

[mraerino]

I have a cdrom in a VM that is emulated from an iso on the host system.

in the VM, fdisk reports these block sizes:

$ sudo fdisk -l /dev/sr0 
Disk /dev/sr0: 4 MiB, 4194304 bytes, 2048 sectors
Disk model: QEMU DVD-ROM    
Units: sectors of 1 * 2048 = 2048 bytes
Sector size (logical/physical): 2048 bytes / 2048 bytes
I/O size (minimum/optimal): 2048 bytes / 2048 bytes

while diskfs is detecting a different physical block size:

disk, _ := diskfs.OpenWithMode("/dev/sr0", diskfs.ReadOnly)
fs, _ := disk.GetFilesystem(0)

DEBU[0000] checking device: /dev/sr0                    
DEBU[0000] initDisk(): start                            
DEBU[0000] initDisk(): block device                     
DEBU[0000] initDisk(): logical block size 2048, physical block size 4096 
DEBU[0000] trying fat32                                 
DEBU[0000] fat32 failed: blocksize for FAT32 must be either 512 bytes or 0, not 2048 
DEBU[0000] trying iso9660 with physical block size 4096 

what could cause this? is there something i can tune so this can't happen?

---

related to #103

[deitch]

The sector sizes are determined here. If you dig in, you can see that it determines the device type (correctly, a block device, when you give it /dev/sr0), and then it calls getSectorSizes (I linked to the Unix one, since, as far as I can tell, you are running on a Unix-like system):

func getSectorSizes(f *os.File) (int64, int64, error) {
	fd := f.Fd()
	logicalSectorSize, err := unix.IoctlGetInt(int(fd), blksszGet)
	if err != nil {
		return 0, 0, fmt.Errorf("Unable to get device logical sector size: %v", err)
	}
	physicalSectorSize, err := unix.IoctlGetInt(int(fd), blkbszGet)
	if err != nil {
		return 0, 0, fmt.Errorf("Unable to get device physical sector size: %v", err)
	}
	return int64(logicalSectorSize), int64(physicalSectorSize), nil
}

I find it interesting that fdisk returned a different value than the ioctl calls, as I think that fdisk does those ioctl calls. It is worth looking into their source code.

The source for fdisk listing (fdisk -l) is here

[deitch]

Yup, code starts here:

/*
 * Get physical block device size. The BLKPBSZGET is supported since Linux
 * 2.6.32. For old kernels is probably the best to assume that physical sector
 * size is the same as logical sector size.
 *
 * Example:
 *
 * rc = blkdev_get_physector_size(fd, &physec);
 * if (rc || physec == 0) {
 *	rc = blkdev_get_sector_size(fd, &physec);
 *	if (rc)
 *		physec = DEFAULT_SECTOR_SIZE;
 * }
 */
#ifdef BLKPBSZGET
int blkdev_get_physector_size(int fd, int *sector_size)
{
	if (ioctl(fd, BLKPBSZGET, &sector_size) >= 0)
		return 0;
	return -1;
}
#else
int blkdev_get_physector_size(int fd __attribute__((__unused__)), int *sector_size)
{
	*sector_size = DEFAULT_SECTOR_SIZE;
	return 0;
}
#endif

And the definitions here:

/* block device topology ioctls, introduced in 2.6.32 (commit ac481c20) */
# ifndef BLKIOMIN
#  define BLKIOMIN   _IO(0x12,120)
#  define BLKIOOPT   _IO(0x12,121)
#  define BLKALIGNOFF _IO(0x12,122)
#  define BLKPBSZGET _IO(0x12,123)
# endif

I wonder if it is being misset on different platforms, because we have it fixed here:

	blkbszGet                        = 0x80081270

It is possible, especially in a virtualized environment (I saw you are running in qemu).

Hmm, oh hold on. Even with a fixed number, I am seeing it as we set it to the value of BLKBSZGET and not BLKPBSZGET, which are different values.

Let's start with a branch that sets it to the "right" fixed value for now, then let's see about getting it dynamically determined.

[deitch]

@mraerino can you try against this branch and see what you get? https://github.com/diskfs/go-diskfs/tree/pbsize

[mraerino]

works! #107 (review)