xfstests and linux nfs testing

xfstests and linux nfs testing

[Steve French]

Was running into a strange error with xfstests over nfs - must be
missing something obvious since I have run it various times before.

Tried the following local.config file (placed in root of xfstests directory)

FSTYP=nfs
TEST_DEV=192.168.93.152:/share-1
TEST_DIR=/mnt/test
NFS_MOUNT_OPTIONS='-ov3'

and did
./check -nfs generic/001

which should have worked fine, but it and the various other nfs
capable xfstests returned

     [not run] requires a test device

which is weird because even setting the TEST_DEV as an environment
variable xfstests still thought it wasn't set.

Anyone have a sample NFS local.config file for xfstests which works?

-- 
Thanks,

Steve

Re: xfstests and linux nfs testing

[Anna Schumaker]

I think this is a bug in xfstests.  This fixes it for me, but I'm not sure what needs to be checked in the NFS case:

Anna

diff --git a/common/rc b/common/rc
index 34d7c19..e86bbfb 100644
--- a/common/rc
+++ b/common/rc
@@ -1085,9 +1085,9 @@ _require_scratch()
 _require_test()
 {
     case "$FSTYP" in
-       nfs*)
-                 _notrun "requires a test device"
-                ;;
+       #nfs*)
+        #         _notrun "requires a test device"
+       #        ;;
        cifs)
                echo $TEST_DEV | grep -q "//" > /dev/null 2>&1
                if [ -z "$TEST_DEV" -o "$?" != "0" ]; then
@@ -1097,7 +1097,7 @@ _require_test()
                     _notrun "this test requires a valid \$TEST_DIR"
                fi
                ;;
-       tmpfs)
+       tmpfs|nfs*)
                if [ -z "$TEST_DEV" -o ! -d "$TEST_DIR" ];
                then
                    _notrun "this test requires a valid \$TEST_DIR and unique $TEST_DEV"



On 10/08/14 23:09, Steve French wrote:
> Was running into a strange error with xfstests over nfs - must be
> missing something obvious since I have run it various times before.
>
> Tried the following local.config file (placed in root of xfstests directory)
>
> FSTYP=nfs
> TEST_DEV=192.168.93.152:/share-1
> TEST_DIR=/mnt/test
> NFS_MOUNT_OPTIONS='-ov3'
>
> and did
> ./check -nfs generic/001
>
> which should have worked fine, but it and the various other nfs
> capable xfstests returned
>
>      [not run] requires a test device
>
> which is weird because even setting the TEST_DEV as an environment
> variable xfstests still thought it wasn't set.
>
> Anyone have a sample NFS local.config file for xfstests which works?
>

Re: xfstests and linux nfs testing

[Christoph Hellwig]

On Thu, Oct 09, 2014 at 08:41:54AM -0400, Anna Schumaker wrote:
> I think this is a bug in xfstests.  This fixes it for me, but I'm not sure what needs to be checked in the NFS case:

Lukas broke this with:
 common: Check fs consistency on TEST_DEV only when needed

as far as I can tell it blindly copied the _require_scratch code
to _require_test, which is bad for two reasons:

 a) due to lots of code duplication
 b) because it broke filesystems that don't support a scratch dev
    like nfs.

Anna fix should work, can you resend it to the fstests lists with a
proper signoff?  Even better would be some validation that the
passed in "device" is a valid format for a nfs export, but that's not
strictly required.

Re: xfstests and linux nfs testing

[Anna Schumaker]

On 10/09/14 12:21, Christoph Hellwig wrote:
> On Thu, Oct 09, 2014 at 08:41:54AM -0400, Anna Schumaker wrote:
>> I think this is a bug in xfstests.  This fixes it for me, but I'm not sure what needs to be checked in the NFS case:
> 
> Lukas broke this with:
>  common: Check fs consistency on TEST_DEV only when needed
> 
> as far as I can tell it blindly copied the _require_scratch code
> to _require_test, which is bad for two reasons:
> 
>  a) due to lots of code duplication
>  b) because it broke filesystems that don't support a scratch dev
>     like nfs.
> 
> Anna fix should work, can you resend it to the fstests lists with a
> proper signoff?  Even better would be some validation that the
> passed in "device" is a valid format for a nfs export, but that's not
> strictly required.

Sure!  I'll fix up the patch so it doesn't have commented out lines while I'm at it :).  I'll take a stab at server format validation, and then send something out!

Anna
>