fails when cp to mounted dir

[wushengzhang]

when I execute
cp /tmp/nfs-utils-2.5.4.tar.xz /mnt1/
It say:
cp: cannot stat '/mnt1/nfs-utils-2.5.4.tar.xz': Input/output error

I have tracked the execution path, It fails on LOOKUP operation, this operation further call
the lookup method in LocalVFS, method call sequence is as following:
lookup->path2fh->fNameToHandleAt.invokeExact(fd, str.address(), bytes.address(), mntId.address(), flags)
the fNameToHandleAt returns -1
therefor the code fails on checkError(rc == 0), because of the target file nfs-utils-2.5.4.tar.xz dose not exists.
Are there any idea to fix this?

[kofemann]

Hi. This can be the result of migration to java 16 FFI, some methods doesn't work as expected. Can you please check with last pre java 16 commit 12a71b9?

[wushengzhang]

Thanks for reply.

I have use the JDK-16, it version is as following:

java 16.0.1 2021-04-20
Java(TM) SE Runtime Environment (build 16.0.1+9-24)
Java HotSpot(TM) 64-Bit Server VM (build 16.0.1+9-24, mixed mode, sharing)

and I have recognized that, when I use centos-7.6 as nfs client,
[root@mu01 ~]# mount -V
mount from util-linux 2.23.2 (libmount 2.23.0: selinux, debug, assert)

this can work fine.

When I use centos-8.2 as nfs client,

[root@app05 vfs4j]# mount.nfs4 -V
mount.nfs4: (linux nfs-utils 2.3.3)

this will get the I/O error

Hi. This can be the result of migration to java 16 FFI, some methods doesn't work as expected. Can you please check with last pre java 16 commit 12a71b9?

[wushengzhang]

This may caused by the server not replied the LOOKUP operation properly?

[kofemann]

Hm. Sounds weird. I will check. Did you have tried earlier commit?

[wushengzhang]

I tried with 300c918, it seems work fine.
except It warn:
[OncRpcSvc (1)] INFO org.dcache.nfs.v4.NFSServerV41 - NFS operation failed: NFS4ERR_MINOR_VERS_MISMATCH : Unsupported minor version [2]

[kofemann]

Great. The warning is ok. It tells you that the client have tried to use nfsv4.2 before switching to v4.1.

[wushengzhang]

Great. The warning is ok. It tells you that the client have tried to use nfsv4.2 before switching to v4.1.

I have tested more commits, and finally find out that, it may work until 0bb7a6d, since ed0f0e8 the bug appears.

[kofemann]

Thanks for looking at it. Yeh, the FFI interface it still work-in-progress. To be honest, I didn't expect that someone will use vfs4j. Shall I branch it at pre-java16 commit? Or may be add a stable tag...

[wushengzhang]

I think Java-16 is OK， except some interfaces, such as MemorySegment, may behave in unknown way (I am not sure).
I just want lookup a lightweight nfs implementation to test with linux distributions with NFSROOT， this project give me greate help， thanks a lot！