Hello, I have been debugging weird failures when using encrypted DVDs (bko#194965 for interested) but in the end it all boils down to the fact that __blkdev_get() updates i_size of bdev inode only of the first open. This seems as a sensible thing to do but there is some weird behavior resulting out of this for devices with removable media: 1) If someone has the device (such as /dev/sr0) open while inserting the media, bdev size will not get updated. This results in the media being accessible but the device size is wrong resulting in weird and hard to debug failures. 2) This is especially annoying when pktcdvd is in the game as pktcdvd device holds corresponding sr device permanently open. Upon some inspection this seems to be an issue with how check_disk_change() (called from sr_block_open()) interacts with __blkdev_get(). If partition scan is enabled, check_disk_change() will call flush_disk() which sets bdev->bd_invalidated. And __blkdev_get() seeing bd_invalidated will call rescan_partitions() which ends up updating bdev size through check_disk_size_change(). But without partitioning none of this happens and the disk size remains stale. Now it seems strange that partitioned and unpartitioned devices behave differently. So I'd be inclined to just unify the behavior and use bd_invalidated for unpartitioned devices as well. Does anyone see a problem with that? Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
