Re: Infinit LAYOUTGET with mmap

[Jeff Layton <jlayton@poochiereds.net>]

On Mon, 2016-05-30 at 20:29 +0200, Mkrtchyan, Tigran wrote:
> the 4.7 have the same behavior...
> 
> capture file: https://desycloud.desy.de/public.php?service=files&t=5e26a8d73ed6aacbdbf49e4c8f1cc01a&download
> 
> Tigran.
> 

Hmm...looks like the server in this case is not changing the stateid.
The client gets back an open stateid with a hash of 0x4567 in frame 6,
which it then uses in all of the LAYOUTGET requests. The server then
sends back the same stateid in the responses to all of the LAYOUTGETs
(including with the same seqid).


> ----- Original Message -----
> > 
> > From: "Jeff Layton" <jlayton@poochiereds.net>
> > To: "Mkrtchyan, Tigran" <tigran.mkrtchyan@desy.de>, "Linux NFS Mailing List" <linux-nfs@vger.kernel.org>
> > Cc: "Trond Myklebust" <trond.myklebust@primarydata.com>
> > Sent: Monday, May 30, 2016 7:58:15 PM
> > Subject: Re: Infinit LAYOUTGET with mmap
> > 
> > On Mon, 2016-05-30 at 17:51 +0200, Mkrtchyan, Tigran wrote:
> > > 
> > > 
> > > re-sending, as it never arrived....
> > > 
> > > ----- Original Message -----
> > > > 
> > > > 
> > > > From: "Mkrtchyan, Tigran" <tigran.mkrtchyan@desy.de>
> > > > To: "Linux NFS Mailing List" <linux-nfs@vger.kernel.org>
> > > > Sent: Monday, May 30, 2016 5:05:59 PM
> > > > Subject: Infinit LAYOUTGET with mmap
> > > > 
> > > > Dear NFS developers,
> > > > 
> > > > On my test system I hit a behavior of the nfs client with
> > > > kernel 4.4 which I haven't seen before.
> > > > 
> > > > 
> > > > Here is a minimal example to reproduce the issue:
> > > > 
> > > > ============= bug.c =====================
> > > > 
> > > > #include 
> > > > #include 
> > > > #include 
> > > > #include 
> > > > #include 
> > > > 
> > > > 
> > > > int main(int argc, char* argv[])
> > > > {
> > > > 
> > > >    int *m;
> > > >    int fd;
> > > >    int err;
> > > > 
> > > >    fd = open(argv[1], O_CREAT| O_RDWR, O_TRUNC, 0666);
> > > >    if (fd < 0) {
> > > >        perror("failed to open file");
> > > >        exit(1);
> > > >    }
> > > > 
> > > >    err = ftruncate(fd, 4096);
> > > >    if (err) {
> > > >        perror("cant set filesize");
> > > >        exit(2);
> > > >    }
> > > > 
> > > >    m = mmap(NULL, 4096, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
> > > >    if (m == MAP_FAILED) {
> > > >        perror("failed to map the file");
> > > >        exit(2);
> > > >    }
> > > > 
> > > >    m[1] = 0;
> > > > 
> > > >    err = munmap(m, 4096);
> > > >    if (err) {
> > > >        perror("failed to unmap the file");
> > > >    }
> > > > 
> > > >    close(fd);
> > > > }
> > > > =============== end of example ================
> > > > 
> > > > 
> > > > 
> > > > When  running this code, client send an OPEN with share_access
> > > > OPEN4_SHARE_ACCESS_BOTH. But later calient sends LAYOUTGET with
> > > > IOMODE_READ
> > > > followed by GETDEVICEINFO.
> > > > 
> > > > This combination of LG+GDI remain for ever. The capture file is
> > > > attached.
> > > > 
> > > > My guess it that client expects RW layout.
> > > > 
> > > > Kernel: 4.4.9-300.fc23.x86_64. I will try with upstream as well.
> > > > 
> > > > Thanks a lot,
> > > >    Tigran.
> > No capture attached, but 4.7 should be getting a major overhaul to the
> > LAYOUTGET retry handling code. You may want to re-test with a bleeding
> > edge kernel and see if the problem is still present.
> > 
> > Cheers,
> > --
> > Jeff Layton <jlayton@poochiereds.net>
> > 
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html

-- 
Jeff Layton <jlayton@poochiereds.net>