nfstest - Man Page

Testing NFS has mostly been done using test tools like connectathon test suite, filebench, iozone and others. But mostly using the connectathon test suite. These are good tools for testing, but they are outdated and they also cannot be used for testing pNFS thoroughly.

For example, you can run the connectathon test suite on pNFS, it runs, it passes all the tests -- but how can we make sure that pNFS worked properly. How can we verify that a layout is granted, but not only that a layout is granted, what type of layout was granted (read, rw). Did the client sent IO to the data servers or to the metadata server.

The Packet trace module takes a trace file created by tcpdump and unpacks the contents of each packet. You can decode one packet at a time, or do a search for specific packets. The main difference between this modules and other tools used to decode trace files is that you can use this module to completely automate your tests.

How does it work? It opens the trace file and reads one record at a time keeping track where each record starts. This way, very large trace files can be opened without having to wait for the file to load and avoid loading the whole file into memory.

Packet layers supported:
   - ETHERNET II (RFC 894)
   - IP layer (supports IPv4 and IPv6)
   - TCP layer
   - UDP layer
   - RPC layer
   - NFS v4.0
   - NFS v4.1 including pNFS file layouts
   - NFS v4.2
   - PORTMAP v2
   - MOUNT v3
   - NLM v4

Verify correct functionality of space reservations so applications are
able to reserve or unreserve space for a file. The system call fallocate
is used to manipulate the allocated disk space for a file, either to
preallocate or deallocate it. For filesystems which support the fallocate
system call, preallocation is done quickly by allocating blocks and
marking them as uninitialized, requiring no I/O to the data blocks.
This is much faster than creating a file and filling it with zeros.

Basic allocate tests verify the disk space is actually preallocated or
reserved for the given range by filling up the device after the allocation
and make sure data can be written to the allocated range without any
problems. Also, any data written outside the allocated range will fail
with NFS4ERR_NOSPC when there is no more space left on the device.
On the other hand, deallocating space will give the disk space back
so it can be used by either the same file on regions not already
preallocated or by different files without the risk of getting
a no space error.

Valid for NFSv4.2 and beyond

Verify consistency of attribute caching by varying acregmin, acregmax,
acdirmin, acdirmax and actimo. Verify consistency of data caching by varying
acregmin, acregmax, acdirmin, acdirmax and actimo.

Valid for any version of NFS

Basic delegation tests verify that a correct delegation is granted when
opening a file for reading or writing. Also, another OPEN should not be
sent for the same file when the client is holding a delegation. Verify
that the stateid of all I/O operations should be the delegation stateid.
Reads from a different process on the same file should not cause the client
to send additional READ packets when the client is holding a read delegation.
Furthermore, a LOCK packet should not be sent to the server when the client
is holding a delegation.

Recall delegation tests verify the delegation is recalled when a conflicting
operation is sent to the server from a different client. Conflicting operations
are reading, writing and changing the permissions on the same file. Note, that
reading a file from a different client can only recall a read delegation.
Also, verify that a delegation is not recalled when a different client is
granted a read delegation. After a delegation is recalled, the client should
send an OPEN with CLAIM_DELEGATE_CUR before returning the delegation and the
stateid should be the same as the original OPEN stateid. Also, a delegation
should not be granted when re-opening the file right before returning
the delegation. Verify client flushes all written data before returning
the WRITE delegation. The LOCK should be sent as well right before returning
a delegation which has been recalled. A delegation should not be granted
on the second client who cause the delegation recall on the first client.

Valid for any version of NFS granting delegations

Functional direct I/O tests verify that every READ/WRITE is sent to the server
instead of the client caching the requests. Client bypasses read ahead by
sending the READ with only the requested bytes. Verify the client correctly
handles eof marker when reading the whole file. Verify client ignores delegation
while writing a file.

Direct I/O on pNFS tests verify the client sends the READ/WRITE to the correct
DS or the MDS if using a PAGESIZE aligned buffer or not, respectively.

Direct I/O data correctness tests verify that a file written with buffered I/O
is read correctly with direct I/O. Verify that a file written with direct I/O
is read correctly with buffered I/O.

Vectored I/O tests verify coalescence of multiple vectors into one READ/WRITE
packet when all vectors are PAGESIZE aligned. Vectors with different alignments
are sent on separate packets.

Valid for NFSv4.0 and NFSv4.1 including pNFS

Basic interoperability tests verify that a file written with different
versions of NFS is written correctly. The contents of the file are verified
by reading the file back using one of the NFS versions.

The tests append different data from different versions of NFS one at a
time then reads the contents of the file to verify it was written
correctly. This is done twice for each test.

Basic locking tests verify that a lock is granted using various arguments
to fcntl. These include blocking and non-blocking locks, read or write locks,
where the file is opened either for reading, writing or both. It also checks
different ranges including limit conditions.

Non-overlapping tests verity that locks are granted on both the client under
test and a second process or a remote client when locking the same file.

Overlapping tests verity that a lock is granted on the client under test
and a second process or a remote client trying to lock the same file will
be denied if a non-blocking lock is issue or will be blocked if a blocking
lock is issue on the second process or remote client.

Valid for any version of NFS

Verify basic pNFS functionality for file (both READ and WRITE), including
opening a second file within the same mount and having a lock on the file.
Also, verify basic pNFS functionality for a file opened for both READ and
WRITE while reading the file first and then writing to it or the other way
around by writing to the file fist and then reading the file. These tests
verify proper functionality of pNFS and NFSv4.1 as well:

    - Verify EXCHANGE_ID is sent to MDS
    - Verify CREATE_SESSION is sent to MDS
    - Verify LAYOUTGET is sent to MDS (check layout type, iomode, layout range)
    - Verify GETDEVICEINFO is sent to MDS
    - Verify EXCHANGE_ID is sent to the correct DS
    - Verify CREATE_SESSION is sent to DS
    - Verify READ/WRITE is sent to DS (check correct stateid, correct offset and size)
    - Verify no GETATTR is sent to DS

Only valid using NFSv4.1 with pNFS enabled and file layout type

Verify POSIX file system level access over the specified path using
positive and negative testing.

Valid for any version of NFS

Verify correct functionality of sparse files. These are files which
have unallocated or uninitialized data blocks as holes. The new NFSv4.2
operation SEEK is used to search for the next hole or data segment
in a file.

Basic tests verify the SEEK operation returns the correct offset of the
next hole or data with respect to the starting offset given to the seek
system call. Verify the SEEK operation is sent to the server with the
correct stateid as a READ call. All files have a virtual hole at the
end of the file so when searching for the next hole, even if the file
does not have a hole, it returns the size of the file.

Valid for NFSv4.2 and beyond

Verify correct functionality of server side copy

Copying a file via NFS the client reads the data from the source file and
then writes the same data to the destination file which is located in the
same server or it could be located in a different server. Either way the
file data is transferred twice, once for reading and the second for writing.
Server side copy allows unnecessary network traffic to be eliminated.
The intra-server copy allows the client to request the server to perform
the copy internally thus avoiding any data being sent through the network
at all. In the case for the inter-server copy where the destination server
is different from the source server, the client authorizes both servers
to interact directly with one another.

Basic server side copy tests verify the actual file range from the source
file(s) are copied correctly to the destination file(s). Most tests deal
with a single source and destination file while verifying the data is copied
correctly. Also it verifies the data is copied starting from the correct
source offset and it is copied to the correct offset on the destination file.
Other tests deal with multiple files: copying multiple source files to a
single destination file, a single source file to multiple destination files,
or N number of source files to M number of destination files.

This I/O tool is used to create and manipulate files of different types.
The arguments allow running for a specified period of time as well as running
multiple processes. Each process modifies a single file at a time and the
file name space is different for each process so there are no collisions
between two different processes modifying the same file.

Decode and display all packets in the packet trace file(s) given.
The match option gives the ability to search for specific packets within
the packet trace file. Other options allow displaying of their corresponding
call or reply when only one or the other is matched. Only a range of packets
can be displayed if the start and/or end options are used.

Display all NFS packets for the specified path. It takes a relative path,
where it searches for each of the directory entries given in the path until
it gets the file handle for the directory where the file is located. Once the
directory file handle is found, a LOOKUP or OPEN/CREATE is searched for the
given file name. If the file lookup or creation is found, all file handles
and state ids associated with that file are searched and all packets found,
including their respective replies are displayed.

Search all the packet traces given for XID inconsistencies. Verify all
operations in the NFSv4.x COMPOUND reply are the same as the operations
given in the call.

Valid for packet traces with NFSv4 and above