GitLab

Motivation:
As COPY operation can be called on files when locks are held, then
instead of open state ids the COPY will be called with lock state ids.

Modification:
Update COPY operation to use open state ids when file open modes are
checked.

Result:
passing NFS compatibility tests.

Acked-by: Lea Morschel
Target: master, 0.23

Acked-by: Lea Morschel
Target: master

Acked-by: Marina Sahakyan
target: master

Acked-by: Marina Sahakyan
Target: master

Motivation:
The cache class uses a regular lock object to guard the access to
cache's internal storage. The same lock is used for read and writes.
As cache's access pattern typically read dominated, the concurrent reads
are appreciated. However, the Cache#get might remove expired entries, thus
a shared read lock should be promoted into a write lock.

As Cache class is used as NFS client session holder, any nfs4.x accesses
(mostly) read or write into it, thus cache throughput has a direct influence
on the NFS server performace in concurrent environments.

Modification:
Use StampedLock to guard the access to Cache's internal storage. Promote
read lock to a write lock if needed or re-lock with a read lock before
modifying.

Result:
Better cache throughput in read dominated concurrent environment.

ReentrantLock
  Benchmark                          Mode  Cnt        Score        Error  Units
  CacheBenchmark.cacheGetBenchmark  thrpt   25  3362040,369 ± 166081,245  ops/s

StampetLock
  Benchmark                          Mode  Cnt        Score        Error  Units
  CacheBenchmark.cacheGetBenchmark  thrpt   25  5280790,414 ± 162177,161  ops/s

Target: master
Acked-by: Lea Morschel

Motivation:
'document' the behavior, avoid miss configurations

Modification:
Add precondition into Cache constructor

Result:
Invalid cache objects can be constructed

Target: master
Acked-by: Albert Rossi
Acked-by: Lea Morschel

future release with improvements and fixes

see: https://github.com/hazelcast/hazelcast/releases/tag/v4.1

Obseletes: PR#111
Acked-by: Paul Millar
Target: master

Less dependency on guava

Acked-by: Lea Morschel
Target: master

Motivation:
For a big files client might request an async copy. Moreover, the server
itself might decide to switch to async mode if requests takes too long.
Thus the server should be ready to notify the client when copy is complete.

Modification:
Update OperationCOPY to (a) use async copy if requests or switch to
async mode if copyFileRange syscall takes too long.

Result:
async behavior for server-side-copy

Acked-by: Paul Millar
Target: master

Acked-by: Lea Morschel
Target: master

Motivation:
A typical (POSIX) filesystem has only one unique identifier for a file,
so called inode number. However, dCache as inumber and a corresponding
pnfsid. As being originally developed as a part of dcache, nfs4j have
inherited that and exposes as Stat#ino and Stat#fileid. After some
iterations, only `fileid` is used, and `ino` kept as 32 bit version of
it.

Modification:
drop Stat#get/setFileId methods in favor of Stat#get/setIno (for better
compliance with POSIX). Update `ino` to 64 bit long. Update codebase to
use `ino`.

Result:
less confusions with a price of small incompatibility with earlier
versions.

Acked-by: Lea Morschel
Acked-by: Paul Millar
Target: master

Acked-by: Lea Morschel
Target: master

Acked-by: Lea Morschel
Target: master

Target: master

Motivation:
the new developments on the client side start to proper
handle files creation time, thus this is a good time for
nfs4j to support it as well.

Modification:
Introduce Stat#set/getBTime. Update NFS code to use btime if provided.

Result:
first steps to support of file creation time.

Acked-by: Paul Millar
Target: master

Feature release with highlights:

    added possibility to control memory pool used by underlying grizzly NIO framework
    improvements in RPC-over-TLS
    support builds with java17
    up-to-date dependencies

Changelog for oncrpc4j-3.2.0...oncrpc4j-3.1.0
* [ada2353] [maven-release-plugin] prepare for next development iteration
* [b42847a] [maven-release-plugin] prepare release oncrpc4j-3.1.0
* [7b531eb] readme: add latest release badge
* [6cd0cab] pom: add profile to sign artifacts
* [1e91d7a] test: improve test coverage of XdrLong
* [9328681] rpc: remove complete timeout task after request is complete
* [c66adbf] rpc: change ABI to have more predictable client behaviour
* [22fdabb] portmap: don't use Boolean object when primitive is required
* [cef527b] portmap: don't create an array when calling vararg method
* [5b97286] Use HTTPS instead of HTTP to resolve dependencies
* [a96f091] readme: describe how to use RPC-over-TLS
* [3c627fa] test: add TLS-over-UDP (DTLS) test
* [d433442] build(deps): bump guava from 24.1-jre to 24.1.1-jre
* [2085136] build(deps): bump junit from 4.12 to 4.13.1
* [3e86824] rpc: fix typo in error message
* [b60d201] pom: require java11
* [b95b59a] portmap: fix unnecessary array creation for logging
* [d4ffc78] Log the address we disconnected from in the exception
* [54cf59c] xdr: remote redundant bracket
* [c4b9961] oncrpc: add ability to control memory allocator
* [95a99b4] gss: remove unused imports
* [2d47552] rpcsvc: use try-with-resource when updating portmap
* [656391d] rpc: fix typo in a comment
* [cab1d81] rpc: move MemoryAllocator into org.dcache.oncrpc4j.rpc package
* [f67311c] rpc: drop SimpleRpcClient and SimpleRpcServer
* [d0f33a0] junit: simplify unit test assertions
* [cf298a5] rpc: add RpcCall#startTLS to simplify the API
* [5b1b405] build(deps-dev): bump bcprov-ext-jdk15on from 1.60 to 1.67
* [50ad1f0] pom: use grizzly-3.0.0
* [ec08f5d] rpc-over-tls: use per-connection STARTTLS attribute
* [e04037d] pom: use java-17 capable jacoco-0.8.7
* [9926a66] rpc: ensure that client on startTLS sends an empty verifier
* [7ecf686] rpc: introduce RpcTransport#isTLS
* [87985be] gss: don't use grizzly buffer without need
* [801cce6] xdr: pass grizzly memory managet to Xdr
* [afb7577] [maven-release-plugin] prepare branch 3.2
* [65accd1] [maven-release-plugin] prepare release oncrpc4j-3.2.0

Acked-by: Lea Morschel
Target: master

Allow virtual filesystem to check user subject.

Signed-off-by: David Kocher <dkocher@iterate.ch>

Motivation:
The attribute bitmap+xdr should provide attributes that applied on
file/directory creation. The current implementation in test client code
is broken as bitmask calculation is wrong.

Modification:
Update client code to use bitmap4#of instead of custom implementation.
add a possibility to specify different attributes for file and
directories.

Result:
valid XDR on the wire.

Fixes: #109
Acked-by: Paul Millar
Target: master

Motivation:
The values use by aceflag4, acemask4 and acetype4 are wrapped by uint32_t
ints.

Modification:
Simplify aceflag4, acemask4 and acetype4 by dropping extra uint32 wrapper.

Result:
less work for GC.

Acked-by: Paul Millar
Acked-by: Lea Morschel
Target: master

Motivation:
The ACE flags are missing in the result of nfsace4#toString method and
introduce a confusion during debugging.

Modification:
include ACE flags in toString method

Result:
consistent output in logs

Acked-by: Paul Millar
Acked-by: Lea Morschel
Target: master

Motivation:
When an async server side copy is performed, then server must notify the
client on completion.

Modification:
Add copy-offload notification to the back channel.

Result:
the ground work for async server side copy

Acked-by: Paul Millar
Target: master

Motivation:
the newer libc make use of copy_file_range function, that directly
mapped to NFSv4.2 COPY request.

Modification:
Add initial implementation on OperationCOPY that synchronously copies
the data between two open files.

Result:
initial step to support intra-server copy.

Acked-by: Paul Millar
Target: master

Guava's implementation of Files#asCharSink s marked @Beta and can be replaced
with java native Files#writeString.

Result:
less dependency on guava's non stable API.

Acked-by: Paul Millar
Acked-by: Lea Morschel
Target: master

the call of Inode#toString for logging is redundant, especially when
logger is disabled.

Acked-by: Lea Morschel
Acked-by: Paul Millar
Target: master

Fixes: #103

Motivation:
There are no reasons to have to utility classes in a single package
(plus Himera vs Chimera)

Modification:
merge NameUtils and HimeraNfsUtils into single class Utils class,
including corresponding tests.

Result:
same code, less classes

Acked-by: Marina Sahakyan
Target: master

Motivation:
uid3 and gid3 object are wraps around uint32, which is by itself wrap around
int.

Modification:
simplify buid3 and gid3y wrapping directly int

Result:
less load on GC.

Acked-by: Paul Millar
Acked-by: Marina Sahakyan
Target: master

Motivation:
size3 object is a wrap around uint64, which is by itself wrap around
long.

Modification:
simplify size3 by wrapping directly long

Result:
less load on GC.

Acked-by: Marina Sahakyan
Acked-by: Paul Millar
Target: master

if client supports multiple layout types, then we need a way to specify which
one is needed.

Acked-by: Paul Millar
Target: master

re-format only

Motivation:
Current client uses LAYOUT4_NFSV4_1_FILES layout types independent from
what server is offering.

Modification:
Update client to show server side supported layout types and disable pNFS
if client doesn't supports any of them.

Result:
more debug possibilities

Acked-by: Albert Rossi
Acked-by: Paul Millar
Target: master

Not used as the functionality is handles by implementations of
LayoutDriver interface.

Acked-by: Lea Morschel
Acked-by: Paul Millar
Acked-by: Albert Rossi
Target: master

as dcache.org hae disabled http->https redirection some builds start to
fail.

Fixes: #98
Target: master, 0.22, 0.21

Motivation:
The state handler keeps track of NFSv4 clients. This duty is not
symmetric - updates happen when a client is added or removed, but
queried almost for any compound operation in a session. Thus on a
busy system we can observe thread serialization.

Modification:
replace synchronized blocks with read/write locks.

Result:
a better concurrency in a multi-client environment.

Acked-by: Paul Millar
Acked-by: Albert Rossi
Target: master

Motivation:
when estimating readdir reply size we should not forget about boolean
flag that indicates the presence of a next entry.

Modification:
fix entry overhead size.

Result:
The returned xdr doesn't overflow expected reply size.

Acked-by: Paul Millar
Target: master, 0.22

The exception is never thrown.

Acked-by: Lea Morschel
Target: master