block: fix a device invalidation regression

Historically we only set the capacity to zero for devices that support
partitions (independ of actually having partitions created).  Doing that
is rather inconsistent, but changing it broke legacy udisks polling for
legacy ide-cdrom devices.  Use the crude a crude check for devices that
either are non-removable or partitionable to get the sane behavior for
most device while not breaking userspace for this particular setup.

Fixes: a1548b6744 ("block: move rescan_partitions to fs/block_dev.c")
Reported-by: He Zhe <zhe.he@windriver.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: He Zhe <zhe.he@windriver.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This commit is contained in:
Christoph Hellwig 2020-03-18 09:12:06 +01:00 committed by Jens Axboe
parent 11bde98600
commit d981cb5b9f

View File

@ -1520,10 +1520,22 @@ rescan:
if (ret)
return ret;
if (invalidate)
set_capacity(disk, 0);
else if (disk->fops->revalidate_disk)
disk->fops->revalidate_disk(disk);
/*
* Historically we only set the capacity to zero for devices that
* support partitions (independ of actually having partitions created).
* Doing that is rather inconsistent, but changing it broke legacy
* udisks polling for legacy ide-cdrom devices. Use the crude check
* below to get the sane behavior for most device while not breaking
* userspace for this particular setup.
*/
if (invalidate) {
if (disk_part_scan_enabled(disk) ||
!(disk->flags & GENHD_FL_REMOVABLE))
set_capacity(disk, 0);
} else {
if (disk->fops->revalidate_disk)
disk->fops->revalidate_disk(disk);
}
check_disk_size_change(disk, bdev, !invalidate);