Allow each memory registration mode to plug in a callout that handles
the completion of a memory registration operation.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Reviewed-by: NSagi Grimberg <sagig@mellanox.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

The open op determines the size of various transport data structures
based on device capabilities and memory registration mode.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Memory Region objects associated with a transport instance are
destroyed before the instance is shutdown and destroyed.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Reviewed-by: NSagi Grimberg <sagig@mellanox.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

This method is invoked when a transport instance is about to be
reconnected. Each Memory Region object is reset to its initial
state.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Reviewed-by: NSagi Grimberg <sagig@mellanox.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

This method is used when setting up a new transport instance to
create a pool of Memory Region objects that will be used to register
memory during operation.

Memory Regions are not needed for "physical" registration, since
->prepare and ->release are no-ops for that mode.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Reviewed-by: NSagi Grimberg <sagig@mellanox.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

There is very little common processing among the different external
memory deregistration functions.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

There is very little common processing among the different external
memory registration functions. Have rpcrdma_create_chunks() call
the registration method directly. This removes a stack frame and a
switch statement from the external registration path.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

The max_payload computation is generalized to ensure that the
payload maximum is the lesser of RPC_MAX_DATA_SEGS and the number of
data segments that can be transmitted in an inline buffer.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Reviewed-by: NSagi Grimberg <sagig@mellanox.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Instead of employing switch() statements, let's use the typical
Linux kernel idiom for handling behavioral variation: virtual
functions.

Start by defining a vector of operations for each supported memory
registration mode, and by adding a source file for each mode.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Reviewed-by: NSagi Grimberg <sagig@mellanox.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

If a provider advertizes a zero max_fast_reg_page_list_len, FRWR
depth detection loops forever. Instead of just failing the mount,
try other memory registration modes.

Fixes: 0fc6c4e7 ("xprtrdma: mind the device's max fast . . .")
Reported-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

The RPC/RDMA transport's FRWR registration logic registers whole
pages. This means areas in the first and last pages that are not
involved in the RDMA I/O are needlessly exposed to the server.

Buffered I/O is typically page-aligned, so not a problem there. But
for direct I/O, which can be byte-aligned, and for reply chunks,
which are nearly always smaller than a page, the transport could
expose memory outside the I/O buffer.

FRWR allows byte-aligned memory registration, so let's use it as
it was intended.
Reported-by: NSagi Grimberg <sagig@mellanox.com>
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Commit 6ab59945 ("xprtrdma: Update rkeys after transport
reconnect" added logic in the ->send_request path to update the
chunk list when an RPC/RDMA request is retransmitted.

Note that rpc_xdr_encode() resets and re-encodes the entire RPC
send buffer for each retransmit of an RPC. The RPC send buffer
is not preserved from the previous transmission of an RPC.

Revert 6ab59945, and instead, just force each request to be
fully marshaled every time through ->send_request. This should
preserve the fix from 6ab59945, while also performing pullup
during retransmits.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Acked-by: NSagi Grimberg <sagig@mellanox.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Reviewed-by: NSagi Grimberg <sagig@mellanox.com>
Tested-by: NDevesh Sharma <Devesh.Sharma@Emulex.Com>
Tested-by: NMeghana Cheripady <Meghana.Cheripady@Emulex.Com>
Tested-by: NVeeresh U. Kokatnur <veereshuk@chelsio.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

POLL_OUT isn't what callers of ->poll() are expecting to see; it's
actually __SI_POLL | 2 and it's a siginfo code, not a poll bitmap
bit...
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Cc: stable@vger.kernel.org
Cc: Bruce Fields <bfields@fieldses.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

If we call groups_alloc() with invalid values then it's might lead to
memory corruption.  For example, with a negative value then we might not
allocate enough for sizeof(struct group_info).

(We're doing this in the caller for consistency with other callers of
groups_alloc().  The other alternative might be to move the check out of
all the callers into groups_alloc().)
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: NSimo Sorce <simo@redhat.com>
Signed-off-by: NJ. Bruce Fields <bfields@redhat.com>

Dan Carpenter's static checker pointed out:

   net/sunrpc/xprtrdma/rpc_rdma.c:879 rpcrdma_reply_handler()
   warn: can 'credits' be negative?

"credits" is defined as an int. The credits value comes from the
server as a 32-bit unsigned integer.

A malicious or broken server can plant a large unsigned integer in
that field which would result in an underflow in the following
logic, potentially triggering a deadlock of the mount point by
blocking the client from issuing more RPC requests.

net/sunrpc/xprtrdma/rpc_rdma.c:

  876          credits = be32_to_cpu(headerp->rm_credit);
  877          if (credits == 0)
  878                  credits = 1;    /* don't deadlock */
  879          else if (credits > r_xprt->rx_buf.rb_max_requests)
  880                  credits = r_xprt->rx_buf.rb_max_requests;
  881
  882          cwnd = xprt->cwnd;
  883          xprt->cwnd = credits << RPC_CWNDSHIFT;
  884          if (xprt->cwnd > cwnd)
  885                  xprt_release_rqst_cong(rqst->rq_task);
Reported-by: NDan Carpenter <dan.carpenter@oracle.com>
Fixes: eba8ff66 ("xprtrdma: Move credit update to RPC . . .")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Our UC-KLEE tool found a kernel memory leak of 512 bytes (on x86_64) for
each call to gssp_accept_sec_context_upcall()
(net/sunrpc/auth_gss/gss_rpc_upcall.c). Since it appears that this call
can be triggered by remote connections (at least, from a cursory a
glance at the call chain), it may be exploitable to cause kernel memory
exhaustion. We found the bug in kernel 3.16.3, but it appears to date
back to commit 9dfd87da (2013-08-20).

The gssp_accept_sec_context_upcall() function performs a pair of calls
to gssp_alloc_receive_pages() and gssp_free_receive_pages().  The first
allocates memory for arg->pages.  The second then frees the pages
pointed to by the arg->pages array, but not the array itself.
Reported-by: NDavid A. Ramos <daramos@stanford.edu>
Fixes: 9dfd87da ("rpc: fix huge kmalloc's in gss-proxy”)
Signed-off-by: NDavid A. Ramos <daramos@stanford.edu>
Signed-off-by: NJ. Bruce Fields <bfields@redhat.com>

Other code that accesses rq_bc_pa_list holds xprt->bc_pa_lock.
xprt_complete_bc_request() should do the same.

Fixes: 2ea24497 ("SUNRPC: RPC callbacks may be split . . .")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

xs_tcp_close() is now just a call to xs_tcp_shutdown(), so remove it,
and replace the entry in xs_tcp_ops.
Suggested-by: NAnna Schumaker <anna.schumaker@netapp.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Yes, kernel_setsockopt() hates you for using a char argument.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Now that the linger code is gone, the xs_tcp_fin_timeout variable has
no real function. Keep it for now, since it is part of the /proc
interface, but only define it if that /proc interface is enabled.
Suggested-by: NAnna Schumaker <Anna.Schumaker@netapp.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

If the connection reset is due to an active call on our side, then
the state change is sometimes not reported. Catch those instances
using xs_error_report() instead.
Also remove the xs_tcp_shutdown() call in xs_tcp_send_request() as
the change in behaviour makes it redundant.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Use of socket shutdown() means that we monitor the shutdown process
through the xs_tcp_state_change() callback, so it is preferable to
a full close in all cases unless we're destroying the transport.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

The previous behaviour left the connection half-open in order to try
to scrape the last replies from the socket. Now that we have more reliable
reconnection, change the behaviour to close down the socket faster.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Now that we no longer use the partial shutdown code when closing the
socket, we no longer need to worry about the TCP linger2 state.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Instead we rely on SO_REUSEPORT to provide the reconnection semantics
that we need for NFSv2/v3.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

It is not safe to call xs_reset_transport() from inside xs_udp_setup_socket()
or xs_tcp_setup_socket(), since they do not own the correct locks. Instead,
do it in xs_connect().
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

The socket lock is currently held by the task that is requesting the
connection be established. While that is efficient in the case where
the connection happens quickly, it is racy in the case where it doesn't.
What we really want is for the connect helper to be able to block access
to the socket while it is being set up.

This patch does so by arranging to transfer the socket lock from the
task that is requesting the connect attempt, and then releasing that
lock once everything is done.
This scheme also gives us automatic protection against collisions with
the RPC close code, so we can kill the cancel_delayed_work_sync()
call in xs_close().
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Otherwise, we may end up looping.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Now that we can reuse bound ports after a close, we never really want to
clear the transport's source port after it has been set. Doing so really
messes up the NFSv3 DRC on the server.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Now that we're setting SO_REUSEPORT, we still need to handle the
case where a connect() is attempted, but the old socket is still
lingering.
Essentially, all we want to do here is handle the error by waiting
a few seconds and then retrying.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

When using TCP, we need the ability to reuse port numbers after
a disconnection, so that the NFSv3 server knows that we're the same
client. Currently we use a hack to work around the TCP socket's
TIME_WAIT: we send an RST instead of closing, which doesn't
always work...
The SO_REUSEPORT option added in Linux 3.9 allows us to bind multiple
TCP connections to the same source address+port combination, and thus
to use ordinary TCP close() instead of the current hack.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

With "make ARCH=x86_64 allmodconfig make C=1 CF=-D__CHECK_ENDIAN__":

linux-2.6/net/sunrpc/xprtrdma/xprt_rdma.h:273:30: warning: incorrect
  type in initializer (different base types)
linux-2.6/net/sunrpc/xprtrdma/xprt_rdma.h:273:30: expected restricted
  __be32 [usertype] *buffer
linux-2.6/net/sunrpc/xprtrdma/xprt_rdma.h:273:30:    got unsigned int
  [usertype] *rq_buffer

As far as I can tell this is a false positive.

Reported-by: kbuild-all@01.org
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Fix an Oopsable condition when nsm_mon_unmon is called as part of the
namespace cleanup, which now apparently happens after the utsname
has been freed.

Link: http://lkml.kernel.org/r/20150125220604.090121ae@neptune.homeReported-by: NBruno Prémont <bonbons@linux-vserver.org>
Cc: stable@vger.kernel.org # 3.18
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Add a call to tally stats for a task under a different statsidx than
what's contained in the task structure.

This is needed to properly account for pnfs reads/writes when the
DS nfs version != the MDS version.
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTom Haynes <Thomas.Haynes@primarydata.com>

Reflect the more conservative approach used in the socket transport's
version of this transport method. An RPC buffer allocation should
avoid forcing not just FS activity, but any I/O.

In particular, two recent changes missed updating xprtrdma:

 - Commit c6c8fe79 ("net, sunrpc: suppress allocation warning ...")
 - Commit a564b8f0 ("nfs: enable swap on NFS")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

rpcrdma_{de}register_internal() are used only in verbs.c now.

MAX_RPCRDMAHDR is no longer used and can be removed.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Reviewed-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Use the new rpcrdma_alloc_regbuf() API to shrink the amount of
contiguous memory needed for a buffer pool by moving the zero
pad buffer into a regbuf.

This is for consistency with the other uses of internally
registered memory.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Reviewed-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>