netfs read failures with current mainline

netfs read failures with current mainline

[Steve French]

I see many "Couldn't get user pages" netfs errors running cifs/102
regression test (which tests for leaked file handles) when using
current mainline from today (which includes the recent netfs read
retry fixes).  For example many of:

[Mon Feb 17 14:08:31 2025] netfs: Couldn't get user pages (rc=-4)
[Mon Feb 17 14:08:31 2025] netfs: Zero-sized read [R=32f]
[Mon Feb 17 14:08:31 2025] netfs: Couldn't get user pages (rc=-4)

Anyone else see netfs problems when running with current mainline?

The test does the following (checking for handle leaks):

# Create a file to test with
echo "hello world" > $TEST_DIR/leak
# Try to kill a 'cat' when it is opening/closing a file
(for i in {1..5000} ; do cat $TEST_DIR/leak & sleep 0.0001 ; kill -9
$! ; done) >/dev/null 2>&1
sleep 3
# and verify if we have any leaked filehandles
smbstatus | grep -i Locked -A1000

-- 
Thanks,

Steve

Re: netfs read failures with current mainline

[David Howells]

Steve French <smfrench@gmail.com> wrote:

> cifs/102

What's cifs/102?

David

Re: netfs read failures with current mainline

[David Howells]

Steve French <smfrench@gmail.com> wrote:

> I see many "Couldn't get user pages" netfs errors running cifs/102
> regression test (which tests for leaked file handles) when using
> current mainline from today (which includes the recent netfs read
> retry fixes).  For example many of:
> 
> [Mon Feb 17 14:08:31 2025] netfs: Couldn't get user pages (rc=-4)
> [Mon Feb 17 14:08:31 2025] netfs: Zero-sized read [R=32f]
> [Mon Feb 17 14:08:31 2025] netfs: Couldn't get user pages (rc=-4)
> 
> Anyone else see netfs problems when running with current mainline?

rc=-4 is -EINTR.  I guess this is because your test is hitting a request with
SIGKILL and iov_iter_extract_pages() is failing.  The above is probably not a
bug, per se, but rather some logging that wants cleaning up.

David

Re: netfs read failures with current mainline

[Steve French]

test cifs/102 (a test for handle leaks) does appear to pass with
current mainline (it should return "no locked files" in the output)
but here is the test:

# Test for Open leaks
#
seq=`basename $0`
seqres=$RESULT_DIR/$seq
echo "QA output created by $seq"

here=`pwd`
tmp=/tmp/$$
status=1        # failure is the default!

_cleanup()
{
        rm -f $tmp.*
}

trap "_cleanup ; exit \$status" 0 1 2 3 15

# get standard environment, filters and checks
. ./common/rc
. ./common/filter

# real QA test starts here
_supported_fs cifs
_require_test

mkdir -p $TEST_DIR/$$ || _fail "failed to create test dir"

# Create a file to test with
echo "hello world" > $TEST_DIR/leak
# Try to kill a 'cat' when it is opening/closing a file
(for i in {1..5000} ; do cat $TEST_DIR/leak & sleep 0.0001 ; kill -9
$! ; done) >/dev/null 2>&1
sleep 3
# and verify if we have any leaked filehandles
smbstatus | grep -i Locked -A1000

status=0
exit

On Wed, Feb 19, 2025 at 10:26 AM David Howells <dhowells@redhat.com> wrote:
>
> Steve French <smfrench@gmail.com> wrote:
>
> > cifs/102
>
> What's cifs/102?
>
> David
>


-- 
Thanks,

Steve