Re: [PATCH blktests 2/2] loop/001: verify all partitions are removed

[Omar Sandoval <osandov@osandov.com>]

On Fri, Mar 15, 2019 at 10:00:27AM +0800, Dongli Zhang wrote:
> 
> 
> On 3/15/19 1:55 AM, Omar Sandoval wrote:
> > On Thu, Mar 14, 2019 at 07:45:17PM +0800, Dongli Zhang wrote:
> >> loop/001 does not test whether all partitions are removed successfully
> >> during loop device partition scanning. As a result, the regression
> >> introduced by 0da03cab87e6 ("loop: Fix deadlock when calling
> >> blkdev_reread_part()") can not be detected.
> >>
> >> The regression will generate below message in dmesg:
> >>
> >> [  464.414043] __loop_clr_fd: partition scan of loop0 failed (rc=-22)
> >>
> >> and leave orphan partitions like below:
> >>
> >> - /dev/loop0p1
> >> - /sys/block/loop0/loop0p1
> >>
> >> This patch verifies all partitions are removed by checking if there is
> >> /sys/block/loopX/loopXpY left. The expected number of partitions left is 0.
> > 
> > Thanks for the test! A couple of comments below.
> > 
> >> Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com>
> >> ---
> >>  tests/loop/001     | 5 +++++
> >>  tests/loop/001.out | 1 +
> >>  2 files changed, 6 insertions(+)
> >>
> >> diff --git a/tests/loop/001 b/tests/loop/001
> >> index 47f760a..a0326b7 100755
> >> --- a/tests/loop/001
> >> +++ b/tests/loop/001
> >> @@ -4,6 +4,9 @@
> >>  #
> >>  # Test loop device partition scanning. Regression test for commit e02898b42380
> >>  # ("loop: fix LO_FLAGS_PARTSCAN hang").
> >> +#
> >> +# Test loop device paritition scanning. Regression test for commit 758a58d0bc67
> >> +# ("loop: set GENHD_FL_NO_PART_SCAN after blkdev_reread_part()").
> > 
> > These can just be combined to
> > 
> > # Test loop device partition scanning. Regression test for commits e02898b42380
> > # ("loop: fix LO_FLAGS_PARTSCAN hang") and 758a58d0bc67 ("loop: set
> > # GENHD_FL_NO_PART_SCAN after blkdev_reread_part()").
> > 
> >>  . tests/loop/rc
> >>  
> >> @@ -24,9 +27,11 @@ test() {
> >>  		mkpart primary 50% 100%
> >>  
> >>  	loop_device="$(losetup -P -f --show "$TMPDIR/img")"
> >> +	loop_name=${loop_device:5}
> >>  	lsblk -ln "$loop_device" | wc -l
> >>  
> >>  	losetup -d "$loop_device"
> >> +	ls /sys/block/$loop_name | grep loop | wc -l
> > 
> > We can just repeat the same `lsblk -ln "$loop_device" | wc -l` from
> > earlier, right? That's a bit cleaner than the hardcoded string slicing
> > and ls.
> 
> Seems 'lsblk' does not work here.
> 
> step1: truncate -s 100M /tmp/tmp.raw
> step2: parted /tmp/tmp.raw --script mklabel msdos \
>        mkpart primary 0% 50% mkpart primary 50% 100%
> step3: losetup -P -f --show /tmp/tmp.raw
> 
> Now we are able to see two loop partitions from 'lsblk'
> 
> # lsblk -ln /dev/loop0
> loop0     7:0    0  100M  0 loop
> loop0p1 259:0    0   50M  0 loop
> loop0p2 259:1    0   50M  0 loop
> 
> 
> step4: # losetup -d /dev/loop0
> 
> There is below syslog as  partscan is failed.
> 
> [  261.181049] __loop_clr_fd: partition scan of loop0 failed (rc=-22)
> 
> 
> There are 2 partitions left:
> 
> # ls /dev | grep loop0
> loop0
> loop0p1
> loop0p2
> 
> # ls /sys/block/loop0 | grep loop
> loop0p1
> loop0p2
> 
> 
> However, 'lsblk -ln' does not report the orphan paritions:
> 
> # lsblk -ln
> sr0   11:0    1 1024M  0 rom
> sda    8:0    0   20G  0 disk
> sda2   8:2    0    1K  0 part
> sda5   8:5    0  4.1G  0 part [SWAP]
> sda1   8:1    0 15.9G  0 part /
> 
> 
> Therefore, we would not be able to use 'lsblk' here.

I see. I think we should check both lsblk and sysfs here. How about
something like
https://github.com/osandov/blktests/commit/6c1237cd358008024ece90bd915a67c23add8a2a?