nfs mount disappears due to inode revalidate failure

nfs mount disappears due to inode revalidate failure

[Sylvain Menu]

Hello,

I am writing to report an issue on a nfs mount that disappears due to
an inode revalide failure (already sent in January but probably banned
with html format...).
This very old commit
(https://github.com/torvalds/linux/commit/cc89684c9a265828ce061037f1f79f4a68ccd3f7)
exactly show the problem I have and this old resolved issue
(https://bugzilla.kernel.org/show_bug.cgi?id=117651) is probably
failing again today

To sum up, I have a NFS mount inside another NFS mount (for example:
/opt/nfs/mount1 & /opt/nfs/mount1/mount2).
If I kill a task trying to get a file descriptor on
/opt/nfs/mount1/mount2 then it will be unmounted. My simple test code
to reproduce very easily:

int main(int argc, char *argv[]) {
    while (1) {
        close(open(argv[1], O_RDONLY));
    }
}

In logs, I have: "nfs_revalidate_inode: (0:62/845965) getattr failed,
error=-512"

Tested on 5.19 and 6.1 kernel

Best regards,
Sylvain Menu

Re: nfs mount disappears due to inode revalidate failure

[Greg KH]

On Thu, Mar 09, 2023 at 10:42:41AM +0100, Sylvain Menu wrote:
> Hello,
> 
> I am writing to report an issue on a nfs mount that disappears due to
> an inode revalide failure (already sent in January but probably banned
> with html format...).
> This very old commit
> (https://github.com/torvalds/linux/commit/cc89684c9a265828ce061037f1f79f4a68ccd3f7)
> exactly show the problem I have and this old resolved issue
> (https://bugzilla.kernel.org/show_bug.cgi?id=117651) is probably
> failing again today
> 
> To sum up, I have a NFS mount inside another NFS mount (for example:
> /opt/nfs/mount1 & /opt/nfs/mount1/mount2).
> If I kill a task trying to get a file descriptor on
> /opt/nfs/mount1/mount2 then it will be unmounted. My simple test code
> to reproduce very easily:
> 
> int main(int argc, char *argv[]) {
>     while (1) {
>         close(open(argv[1], O_RDONLY));
>     }
> }
> 
> In logs, I have: "nfs_revalidate_inode: (0:62/845965) getattr failed,
> error=-512"
> 
> Tested on 5.19 and 6.1 kernel

So is this a regression or something that has always been present?

thanks,

greg k-h

Re: nfs mount disappears due to inode revalidate failure

[Sylvain Menu]

I think it's a regression according to the old resolved bugs/tickets
but no idea since when it's broken again

Le jeu. 9 mars 2023 à 11:07, Greg KH <gregkh@linuxfoundation.org> a écrit :
>
> On Thu, Mar 09, 2023 at 10:42:41AM +0100, Sylvain Menu wrote:
> > Hello,
> >
> > I am writing to report an issue on a nfs mount that disappears due to
> > an inode revalide failure (already sent in January but probably banned
> > with html format...).
> > This very old commit
> > (https://github.com/torvalds/linux/commit/cc89684c9a265828ce061037f1f79f4a68ccd3f7)
> > exactly show the problem I have and this old resolved issue
> > (https://bugzilla.kernel.org/show_bug.cgi?id=117651) is probably
> > failing again today
> >
> > To sum up, I have a NFS mount inside another NFS mount (for example:
> > /opt/nfs/mount1 & /opt/nfs/mount1/mount2).
> > If I kill a task trying to get a file descriptor on
> > /opt/nfs/mount1/mount2 then it will be unmounted. My simple test code
> > to reproduce very easily:
> >
> > int main(int argc, char *argv[]) {
> >     while (1) {
> >         close(open(argv[1], O_RDONLY));
> >     }
> > }
> >
> > In logs, I have: "nfs_revalidate_inode: (0:62/845965) getattr failed,
> > error=-512"
> >
> > Tested on 5.19 and 6.1 kernel
>
> So is this a regression or something that has always been present?
>
> thanks,
>
> greg k-h

Re: nfs mount disappears due to inode revalidate failure

[Greg KH]

On Thu, Mar 09, 2023 at 11:17:30AM +0100, Sylvain Menu wrote:
> I think it's a regression according to the old resolved bugs/tickets
> but no idea since when it's broken again

Any chance you can do 'git bisect' to find where it broke and what
commit broke it?

thanks,

greg k-h

Re: nfs mount disappears due to inode revalidate failure

[Sylvain Menu]

No I don't have that, I found the bug in production by no chance.
I tried to dive into the code but it quickly becomes complex for me,
at least it's easy to reproduce with a little script (while(1) timeout
my_c.code)

thanks
sylvain menu

Le jeu. 9 mars 2023 à 11:22, Greg KH <gregkh@linuxfoundation.org> a écrit :
>
> On Thu, Mar 09, 2023 at 11:17:30AM +0100, Sylvain Menu wrote:
> > I think it's a regression according to the old resolved bugs/tickets
> > but no idea since when it's broken again
>
> Any chance you can do 'git bisect' to find where it broke and what
> commit broke it?
>
> thanks,
>
> greg k-h

Re: nfs mount disappears due to inode revalidate failure

[NeilBrown]

On Fri, 10 Mar 2023, Sylvain Menu wrote:
> No I don't have that, I found the bug in production by no chance.
> I tried to dive into the code but it quickly becomes complex for me,
> at least it's easy to reproduce with a little script (while(1) timeout
> my_c.code)
> 
> thanks
> sylvain menu
> 
> Le jeu. 9 mars 2023 à 11:22, Greg KH <gregkh@linuxfoundation.org> a écrit :
> >
> > On Thu, Mar 09, 2023 at 11:17:30AM +0100, Sylvain Menu wrote:
> > > I think it's a regression according to the old resolved bugs/tickets
> > > but no idea since when it's broken again
> >
> > Any chance you can do 'git bisect' to find where it broke and what
> > commit broke it?

Please see
   https://lore.kernel.org/linux-nfs/87361aovm3.fsf@notabene.neil.brown.name/

I posted a patch for this a couple of years ago, but Trond wouldn't take
it.

NeilBrown


> >
> > thanks,
> >
> > greg k-h
> 

nfs mount disappears due to inode revalidate failure

[sylvain.menu]

[-- Attachment #1.1: Type: text/plain, Size: 890 bytes --]

Hello,

 

I am writing to report an issue on a nfs mount that disappears due to an
inode revalide failure.

 
<https://github.com/torvalds/linux/commit/cc89684c9a265828ce061037f1f79f4a68
ccd3f7> This very old commit exactly show the problem I have and
<https://bugzilla.kernel.org/show_bug.cgi?id=117651> this old resolved issue
is probably failing again today

 

To sum up, I have a NFS mount inside another NFS mount (for example:
/opt/nfs/mount1 & /opt/nfs/mount1/mount2).

If I kill a task trying to get a file descriptor on /opt/nfs/mount1/mount2
then it will be unmounted. My simple test code to reproduce very easily:

 

int main(int argc, char *argv[]) {

    while (1) {

        close(open(argv[1], O_RDONLY));

    }

}

 

In logs, I have: "nfs_revalidate_inode: (0:62/845965) getattr failed,
error=-512"

 

Tested on 5.19 and 6.1 kernel

 

Best regards,

Sylvain Menu

 


[-- Attachment #1.2: Type: text/html, Size: 5482 bytes --]

[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 7608 bytes --]