GEOM weirdness with SD flash reader?

[Nate Lawson]

Dmitry Morozovsky wrote:
> On Sun, 1 Jul 2007, Nate Lawson wrote:
> 
> NL> Simply trying to mount a FAT filesystem in an external USB SD flash
> NL> reader.  The reader works fine, no errors in dmesg.
> NL> 
> NL> deenlo# fdisk da2
> NL> ******* Working on device /dev/da2 *******
> NL> parameters extracted from in-core disklabel are:
> NL> cylinders=982 heads=64 sectors/track=32 (2048 blks/cyl)
> NL> 
> NL> parameters to be used for BIOS calculations are:
> NL> cylinders=982 heads=64 sectors/track=32 (2048 blks/cyl)
> NL> 
> NL> Media sector size is 512
> NL> Warning: BIOS sector numbering starts with sector 1
> NL> Information from DOS bootblock is:
> NL> The data for partition 1 is:
> NL> sysid 6 (0x06),(Primary 'big' DOS (>= 32MB))
> NL>     start 243, size 2011917 (982 Meg), flag 0
> NL>         beg: cyl 0/ head 3/ sector 55;
> NL>         end: cyl 998/ head 3/ sector 3
> NL> The data for partition 2 is:
> NL> <UNUSED>
> NL> The data for partition 3 is:
> NL> <UNUSED>
> NL> The data for partition 4 is:
> NL> <UNUSED>
> NL> deenlo# mount -t msdosfs /dev/da2s1 /mnt/dos
> NL> mount_msdosfs: /dev/da2s1: No such file or directory
> NL> deenlo# mount -t msdosfs /dev/da2 /mnt/dos
> NL> mount_msdosfs: /dev/da2: Invalid argument
> NL> deenlo# mount -t msdosfs /dev/da2s1 /mnt/dos
> NL> [success]
> NL> 
> NL> So I'm guessing GEOM had to decide to create /dev/da2s1 after I tried to
> NL> mount /dev/da2 directly?  Nothing else would create the da2s1 device.
> NL> Once the device was created with make_dev(), repeated unmounts/mounts
> NL> work fine.
> 
> Do you insert your card after reader got initialized? I have internal reader in 
> my work machine, and have to use
> 
> dd if=/dev/null of=/dev/daN count=0
> 
> to [re]initialize GEOM providers on newly inserted card...

Yes, you're exactly right.  However, shouldn't a read from a device
trigger a re-taste if the media is removable?  So the fdisk of da2
should have caused it to re-read the MBR and create /dev/da2s1.  I think
maybe there should be a flag set for devices with removable media that
causes the device to be re-tasted on every mount request.

Other ideas how to fix this?

-- 
Nate