Reading and storing the RPC direction is a three step process.
1. xs_tcp_read_calldir() reads the RPC direction, but it will not store it
in the XDR buffer since the 'struct rpc_rqst' is not yet available.
2. The 'struct rpc_rqst' is obtained during the TCP_RCV_COPY_DATA state.
This state need not necessarily be preceeded by the TCP_RCV_READ_CALLDIR.
For example, we may be reading a continuation packet to a large reply.
Therefore, we can't simply obtain the 'struct rpc_rqst' during the
TCP_RCV_READ_CALLDIR state and assume it's available during TCP_RCV_COPY_DATA.
This patch adds a new TCP_RCV_READ_CALLDIR flag to indicate the need to
read the RPC direction. It then uses TCP_RCV_COPY_CALLDIR to indicate the
RPC direction needs to be saved after the 'struct rpc_rqst' has been allocated.
3. The 'struct rpc_rqst' is obtained by the xs_tcp_read_data() helper
functions. xs_tcp_read_common() then saves the RPC direction in the XDR
buffer if TCP_RCV_COPY_CALLDIR is set. This will happen when we're reading
the data immediately after the direction was read. xs_tcp_read_common()
then clears this flag.
[was nfs41: Skip past the RPC call direction]
Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[nfs41: sunrpc: Add RPC direction back into the XDR buffer]
Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[nfs41: sunrpc: Don't skip past the RPC call direction]
Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
NFSv4.1 callbacks can arrive over an existing connection. This patch adds
the logic to read the RPC call direction (call or reply). It does this by
updating the state machine to look for the call direction invoking
xs_tcp_read_calldir(...) after reading the XID.
[nfs41: Keep track of RPC call/reply direction with a flag]
As per 11/14/08 review of RFC 53/85.
Add a new flag to track whether the incoming message is an RPC call or an
RPC reply. TCP_RPC_REPLY is set in the 'struct sock_xprt' tcp_flags in
xs_tcp_read_calldir() if the message is an RPC reply sent on the forechannel.
It is cleared if the message is an RPC request sent on the back channel.
Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
See http://bugzilla.kernel.org/show_bug.cgi?id=13034
If the port gets into a TIME_WAIT state, then we cannot reconnect without
binding to a new port.
Tested-by: Petr Vandrovec <petr@vandrovec.name>
Tested-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
As long as one task is holding the socket lock, then calls to
xprt_force_disconnect(xprt) will not succeed in shutting down the socket.
In particular, this would mean that a server initiated shutdown will not
succeed until the lock is relinquished.
In order to avoid the deadlock, we should ensure that xs_tcp_send_request()
closes the socket on EPIPE errors too.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
This fixes a regression against FreeBSD servers as reported by Tomas
Kasparek. Apparently when using RPC over a TCP socket, the FreeBSD servers
don't ever react to the client closing the socket, and so commit
e06799f958 (SUNRPC: Use shutdown() instead of
close() when disconnecting a TCP socket) causes the setup to hang forever
whenever the client attempts to close and then reconnect.
We break the deadlock by adding a 'linger2' style timeout to the socket,
after which, the client will abort the connection using a TCP 'RST'.
The default timeout is set to 15 seconds. A subsequent patch will put it
under user control by means of a systctl.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
If xs_nospace() finds that the socket has disconnected, it attempts to
return ENOTCONN, however that value is then squashed by the callers.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Enforce the comment in xs_tcp_connect_worker4/xs_tcp_connect_worker6 that
we should delay, then retry on certain connection errors.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
While we should definitely return socket errors to the task that is
currently trying to send data, there is no need to propagate the same error
to all the other tasks on xprt->pending. Doing so actually slows down
recovery, since it causes more than one tasks to attempt socket recovery.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
If we get an ECONNREFUSED error, we currently go to sleep on the
'xprt->sending' wait queue. The problem is that no timeout is set there,
and there is nothing else that will wake the task up later.
We should deal with ECONNREFUSED in call_status, given that is where we
also deal with -EHOSTDOWN, and friends.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
...so that we can distinguish between when we need to shutdown and when we
don't. Also remove the call to xs_tcp_shutdown() from xs_tcp_connect(),
since xprt_connect() makes the same test.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
To clear out old state, the UDP connect workers unconditionally invoke
xs_close() before proceeding with a new connect. Nowadays this causes
a spurious wake-up of the task waiting for the connect to complete.
This is a little racey, but usually harmless. The waiting task
immediately retries the connect via a call_bind/call_connect sequence,
which usually finds the transport already in the connected state
because the connect worker has finished in the background.
To avoid a spurious wake-up, factor the xs_close() logic that resets
the underlying socket into a helper, and have the UDP connect workers
call that helper instead of xs_close().
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
If the transport isn't bound, then we should just return ENOTCONN, letting
call_connect_status() and/or call_status() deal with retrying. Currently,
we appear to abort all pending tasks with an EIO error.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
We can Oops in both xs_udp_send_request() and xs_tcp_send_request() if the
call to xs_sendpages() returns an error due to the socket not yet being
set up.
Deal with that situation by returning a new error: ENOTSOCK, so that we
know to avoid dereferencing transport->sock.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Using NIPQUAD() with NIPQUAD_FMT, %d.%d.%d.%d or %u.%u.%u.%u
can be replaced with %pI4
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
gcc warns when using the # modifier with the %p format specifier,
so we can't use this to omit the colons when needed, introduces
%pi6 instead.
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The define in kernel.h can be done away with at a later time.
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the server sends us an RST error while we're in the TCP_ESTABLISHED
state, then that will not result in a state change, and so the RPC client
ends up hanging forever (see
http://bugzilla.kernel.org/show_bug.cgi?id=11154)
We can intercept the reset by setting up an sk->sk_error_report callback,
which will then allow us to initiate a proper shutdown and retry...
We also make sure that if the send request receives an ECONNRESET, then we
shutdown too...
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Clean up the various different email addresses of mine listed in the code
to a single current and valid address. As Dave says his network merges
for 2.6.28 are now done this seems a good point to send them in where
they won't risk disrupting real changes.
Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The RPC client uses the rq_xtime field in each RPC request to determine the
round-trip time of the request. Currently, the rq_xtime field is
initialized by each transport just before it starts enqueing a request to
be sent. However, transports do not handle initializing this value
consistently; sometimes they don't initialize it at all.
To make the measurement of request round-trip time consistent for all
RPC client transport capabilities, pull rq_xtime initialization into the
RPC client's generic transport logic. Now all transports will get a
standardized RTT measure automatically, from:
xprt_transmit()
to
xprt_complete_rqst()
This makes round-trip time calculation more accurate for the TCP transport.
The socket ->sendmsg() method can return "-EAGAIN" if the socket's output
buffer is full, so the TCP transport's ->send_request() method may call
the ->sendmsg() method repeatedly until it gets all of the request's bytes
queued in the socket's buffer.
Currently, the TCP transport sets the rq_xtime field every time through
that loop so the final value is the timestamp just before the *last* call
to the underlying socket's ->sendmsg() method. After this patch, the
rq_xtime field contains a timestamp that reflects the time just before the
*first* call to ->sendmsg().
This is consequential under heavy workloads because large requests often
take multiple ->sendmsg() calls to get all the bytes of a request queued.
The TCP transport causes the request to sleep until the remote end of the
socket has received enough bytes to clear space in the socket's local
output buffer. This delay can be quite significant.
The method introduced by this patch is a more accurate measure of RTT
for stream transports, since the server can cause enough back pressure
to delay (ie increase the latency of) requests from the client.
Additionally, this patch corrects the behavior of the RDMA transport, which
entirely neglected to initialize the rq_xtime field. RPC performance
metrics for RDMA transports now display correct RPC request round trip
times.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Tom Talpey <thomas.talpey@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
NFSv4 requires us to ensure that we break the TCP connection before we're
allowed to retransmit a request. However in the case where we're
retransmitting several requests that have been sent on the same
connection, we need to ensure that we don't interfere with the attempt to
reconnect and/or break the connection again once it has been established.
We therefore introduce a 'connection' cookie that is bumped every time a
connection is broken. This allows requests to track if they need to force a
disconnection.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The condition for exiting from the loop in xs_tcp_send_request() should be
that we find we're not making progress (i.e. number of bytes sent is 0).
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The rest of the networking layer uses SOCK_ASYNC_NOSPACE to signal whether
or not we have someone waiting for buffer memory. Convert the SUNRPC layer
to use the same idiom.
Remove the unlikely()s in xs_udp_write_space and xs_tcp_write_space. In
fact, the most common case will be that there is nobody waiting for buffer
space.
SOCK_NOSPACE is there to tell the TCP layer whether or not the cwnd was
limited by the application window. Ensure that we follow the same idiom as
the rest of the networking layer here too.
Finally, ensure that we clear SOCK_ASYNC_NOSPACE once we wake up, so that
write_space() doesn't keep waking things up on xprt->pending.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
__FUNCTION__ is gcc-specific, use __func__
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Clean up: document the rule (kfree) and the exceptions
(RPC_DISPLAY_PROTO and RPC_DISPLAY_NETID) when freeing the objects in
a transport's address_strings array.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Clean up: have the set up routines explicitly pass the strings to be used
for the transport name and NETID. This removes a number of conditionals
and dependencies on rpc_xprt.prot, which is overloaded.
Tighten up type checking on the address_strings array while we're at it.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
In order to be able to support setting the timeo and retrans parameters on
a per-mountpoint basis, we move the rpc_timeout structure into the
rpc_clnt.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
If we've disconnected from the server, rather than the other way round,
then it makes little sense to wait 3 seconds before reconnecting.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
xprt_disconnect() should really only be called when the transport shutdown
is completed, and it is time to wake up any pending tasks. Rename it to
xprt_disconnect_done() in order to reflect the semantical change.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Move the calls to xprt_disconnect() over to xprt_force_disconnect() in
order to enable the transport layer to manage the state of the
XPRT_CONNECTED flag.
Ditto in xs_tcp_read_fraghdr().
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The transport layer should do that itself whenever appropriate.
Note that the RDMA transport already assumes that it needs to call
xprt_disconnect in xprt_rdma_close().
For TCP sockets, we want to call xprt_disconnect() only after the
connection has been closed by both ends.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
By using shutdown() rather than close() we allow the RPC client to wait
for the TCP close handshake to complete before we start trying to reconnect
using the same port.
We use shutdown(SHUT_WR) only instead of shutting down both directions,
however we wait until the server has closed the connection on its side.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Add an xprt->state bit to enable the TCP ->state_change() method to signal
whether or not the TCP connection is in the process of closing down.
This will to be used by the reconnection logic in a separate patch.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Currently the TCP rebinding logic assumes that if we're not using a
reserved port, then we don't need to reconnect on the same port if a
disconnection event occurs. This breaks most RPC duplicate reply cache
implementations.
Also take into account the fact that xprt_min_resvport and
xprt_max_resvport may change while we're reconnecting, since the user may
change them at any time via the sysctls. Ensure that we check the port
boundaries every time we loop in xs_bind4/xs_bind6. Also ensure that if the
boundaries change, we only scan the ports a maximum of 2 times.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
When scheduling the autoclose RPC call, we want to ensure that we don't
race against the test_bit() call in xprt_clear_locked().
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The previous move of the the UDP inDatagrams counter caused the
counting of encapsulated packets, SUNRPC data (as opposed to call)
packets and RXRPC packets to go missing.
This patch restores all of these.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
