FMR is not supported on most recent RDMA devices. It is also less
secure than FRWR because an FMR memory registration can expose
adjacent bytes to remote reading or writing. As discussed during the
RDMA BoF at LPC 2018, it is time to remove support for FMR in the
NFS/RDMA client stack.

Note that NFS/RDMA server-side uses either local memory registration
or FRWR. FMR is not used.

There are a few Infiniband/RoCE devices in the kernel tree that do
not appear to support MEM_MGT_EXTENSIONS (FRWR), and therefore will
not support client-side NFS/RDMA after this patch. These are:

 - mthca
 - qib
 - hns (RoCE)

Users of these devices can use NFS/TCP on IPoIB instead.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Transport disconnect processing does a "wake pending tasks" at
various points.

Suppose an RPC Reply is being processed. The RPC task that Reply
goes with is waiting on the pending queue. If a disconnect wake-up
happens before reply processing is done, that reply, even if it is
good, is thrown away, and the RPC has to be sent again.

This window apparently does not exist for socket transports because
there is a lock held while a reply is being received which prevents
the wake-up call until after reply processing is done.

To resolve this, all RPC replies being processed on an RPC-over-RDMA
transport have to complete before pending tasks are awoken due to a
transport disconnect.

Callers that already hold the transport write lock may invoke
->ops->close directly. Others use a generic helper that schedules
a close when the write lock can be taken safely.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

After thinking about this more, and auditing other kernel ULP imple-
mentations, I believe that a DISCONNECT cm_event will occur after a
fatal QP event. If that's the case, there's no need for an explicit
disconnect in the QP event handler.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

To address a connection-close ordering problem, we need the ability
to drain the RPC completions running on rpcrdma_receive_wq for just
one transport. Give each transport its own RPC completion workqueue,
and drain that workqueue when disconnecting the transport.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Divide the work cleanly:

- rpcrdma_wc_receive is responsible only for RDMA Receives
- rpcrdma_reply_handler is responsible only for RPC Replies
- the posted send and receive counts both belong in rpcrdma_ep
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

While chasing yet another set of DMAR fault reports, I noticed that
the frwr recycler conflates whether or not an MR has been DMA
unmapped with frwr->fr_state. Actually the two have only an indirect
relationship. It's in fact impossible to guess reliably whether the
MR has been DMA unmapped based on its fr_state field, especially as
the surrounding code and its assumptions have changed over time.

A better approach is to track the DMA mapping status explicitly so
that the recycler is less brittle to unexpected situations, and
attempts to DMA-unmap a second time are prevented.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Cc: stable@vger.kernel.org # v4.20
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

To show that a caller did attempt to allocate and post more Receive
buffers, the trace point in rpcrdma_post_recvs() should report when
rpcrdma_post_recvs() was invoked but no new Receive buffers were
posted.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: rb_flags might be used for other things besides
RPCRDMA_BUF_F_EMPTY_SCQ, so initialize it in a generic spot
instead of in a send-completion-queue-related helper.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Replace the hashed memory address of the target rpcrdma_ep
with the server's IP address and port. The server address is more
useful in an administrative error message.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Use a function name that is consistent with the RDMA core
API and with other consumers. Because this is a function that is
invoked from outside the rpcrdma.ko module, add an appropriate
documenting comment.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Currently, when a connection is established, rpcrdma_conn_upcall
invokes rpcrdma_conn_func and then
wake_up_all(&ep->rep_connect_wait). The former wakes waiting RPCs,
but the connect worker is not done yet, and that leads to races,
double wakes, and difficulty understanding how this logic is
supposed to work.

Instead, collect all the "connection established" logic in the
connect worker (xprt_rdma_connect_worker). A disconnect worker is
retained to handle provider upcalls safely.

Fixes: 254f91e2 ("xprtrdma: RPC/RDMA must invoke ... ")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Eliminate the FALLTHROUGH into the default arm to make the
switch easier to understand.

Also, as long as I'm here, do not display the memory address of the
target rpcrdma_ep. A hashed memory address is of marginal use here.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up.

Since commit 173b8f49 ("xprtrdma: Demote "connect" log messages")
there has been no need to initialize connstat to zero. In fact, in
this code path there's now no reason not to set rep_connected
directly.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: The convention throughout other parts of xprtrdma is to
name variables of type struct rpcrdma_xprt "r_xprt", not "xprt".
This convention enables the use of the name "xprt" for a "struct
rpc_xprt" type variable, as in other parts of the RPC client.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Use a function name that is consistent with the RDMA core
API and with other consumers. Because this is a function that is
invoked from outside the rpcrdma.ko module, add an appropriate
documenting comment.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up the names of trace events related to MRs so that it's
easy to enable these with a glob.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

When a memory operation fails, the MR's driver state might not match
its hardware state. The only reliable recourse is to dereg the MR.
This is done in ->ro_recover_mr, which then attempts to allocate a
fresh MR to replace the released MR.

Since commit e2ac236c ("xprtrdma: Allocate MRs on demand"),
xprtrdma dynamically allocates MRs. It can add more MRs whenever
they are needed.

That makes it possible to simply release an MR when a memory
operation fails, instead of "recovering" it. It will automatically
be replaced by the on-demand MR allocator.

This commit is a little larger than I wanted, but it replaces
->ro_recover_mr, rb_recovery_lock, rb_recovery_worker, and the
rb_stale_mrs list with a generic work queue.

Since MRs are no longer orphaned, the mrs_orphaned metric is no
longer used.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Some devices require more than 3 MRs to build a single 1MB I/O.
Ensure that rpcrdma_mrs_create() will add enough MRs to build that
I/O.

In a subsequent patch I'm changing the MR recovery logic to just
toss out the MRs. In that case it's possible for ->send_request to
loop acquiring some MRs, not getting enough, getting called again,
recycling the previous MRs, then not getting enough, lather rinse
repeat. Thus first we need to ensure enough MRs are created to
prevent that loop.

I'm "reusing" ia->ri_max_segs. All of its accessors seem to want the
maximum number of data segments plus two, so I'm going to bake that
into the initial calculation.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

I found that injecting disconnects with v4.18-rc resulted in
random failures of the multi-threaded git regression test.

The root cause appears to be that, after a reconnect, the
RPC/RDMA transport is waking pending RPCs before the transport has
posted enough Receive buffers to receive the Replies. If a Reply
arrives before enough Receive buffers are posted, the connection
is dropped. A few connection drops happen in quick succession as
the client and server struggle to regain credit synchronization.

This regression was introduced with commit 7c8d9e7c ("xprtrdma:
Move Receive posting to Receive handler"). The client is supposed to
post a single Receive when a connection is established because
it's not supposed to send more than one RPC Call before it gets
a fresh credit grant in the first RPC Reply [RFC 8166, Section
3.3.3].

Unfortunately there appears to be a longstanding bug in the Linux
client's credit accounting mechanism. On connect, it simply dumps
all pending RPC Calls onto the new connection. It's possible it has
done this ever since the RPC/RDMA transport was added to the kernel
ten years ago.

Servers have so far been tolerant of this bad behavior. Currently no
server implementation ever changes its credit grant over reconnects,
and servers always repost enough Receives before connections are
fully established.

The Linux client implementation used to post a Receive before each
of these Calls. This has covered up the flooding send behavior.

I could try to correct this old bug so that the client sends exactly
one RPC Call and waits for a Reply. Since we are so close to the
next merge window, I'm going to instead provide a simple patch to
post enough Receives before a reconnect completes (based on the
number of credits granted to the previous connection).

The spurious disconnects will be gone, but the client will still
send multiple RPC Calls immediately after a reconnect.

Addressing the latter problem will wait for a merge window because
a) I expect it to be a large change requiring lots of testing, and
b) obviously the Linux client has interoperated successfully since
day zero while still being broken.

Fixes: 7c8d9e7c ("xprtrdma: Move Receive posting to ... ")
Cc: stable@vger.kernel.org # v4.18+
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Since neither ib_post_send() nor ib_post_recv() modify the data structure
their second argument points at, declare that argument const. This change
makes it necessary to declare the 'bad_wr' argument const too and also to
modify all ULPs that call ib_post_send(), ib_post_recv() or
ib_post_srq_recv(). This patch does not change any functionality but makes
it possible for the compiler to verify whether the
ib_post_(send|recv|srq_recv) really do not modify the posted work request.

To make this possible, only one cast had to be introduce that casts away
constness, namely in rpcrdma_post_recvs(). The only way I can think of to
avoid that cast is to introduce an additional loop in that function or to
change the data type of bad_wr from struct ib_recv_wr ** into int
(an index that refers to an element in the work request list). However,
both approaches would require even more extensive changes than this
patch.
Signed-off-by: NBart Van Assche <bart.vanassche@wdc.com>
Reviewed-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NJason Gunthorpe <jgg@mellanox.com>

This patch replaces the ib_device_attr.max_sge with max_send_sge and
max_recv_sge. It allows ulps to take advantage of devices that have very
different send and recv sge depths.  For example cxgb4 has a max_recv_sge
of 4, yet a max_send_sge of 16.  Splitting out these attributes allows
much more efficient use of the SQ for cxgb4 with ulps that use the RDMA_RW
API. Consider a large RDMA WRITE that has 16 scattergather entries.
With max_sge of 4, the ulp would send 4 WRITE WRs, but with max_sge of
16, it can be done with 1 WRITE WR.
Acked-by: NSagi Grimberg <sagi@grimberg.me>
Acked-by: NChristoph Hellwig <hch@lst.de>
Acked-by: NSelvin Xavier <selvin.xavier@broadcom.com>
Acked-by: NShiraz Saleem <shiraz.saleem@intel.com>
Acked-by: NDennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NJason Gunthorpe <jgg@mellanox.com>

Currently, when the sendctx queue is exhausted during marshaling, the
RPC/RDMA transport places the RPC task on the delayq, which forces a
wait for HZ >> 2 before the marshal and send is retried.

With this change, the transport now places such an RPC task on the
pending queue, and wakes it just as soon as more sendctxs become
available. This typically takes less than a millisecond, and the
write_space waking mechanism is less deadlock-prone.

Moreover, the waiting RPC task is holding the transport's write
lock, which blocks the transport from sending RPCs. Therefore faster
recovery from sendctx queue exhaustion is desirable.

Cf. commit 5804891455d5 ("xprtrdma: ->send_request returns -EAGAIN
when there are no free MRs").
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Move #include <trace/events/rpcrdma.h> into source files,
similar to how it is done with trace/events/sunrpc.h.

Server-side trace points will be part of the rpcrdma subsystem,
just like the client-side trace points.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NJ. Bruce Fields <bfields@redhat.com>

Clean up: The only call site is in the same file as the function's
definition.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: There is only one remaining call site for this helper.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up. There is only one call-site for this helper, and it can be
simplified by using list_first_entry_or_null().
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: These functions are no longer used.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Receive completion and Reply handling are done by a BOUND
workqueue, meaning they run on only one CPU.

Posting receives is currently done in the send_request path, which
on large systems is typically done on a different CPU than the one
handling Receive completions. This results in movement of
Receive-related cachelines between the sending and receiving CPUs.

More importantly, it means that currently Receives are posted while
the transport's write lock is held, which is unnecessary and costly.

Finally, allocation of Receive buffers is performed on-demand in
the Receive completion handler. This helps guarantee that they are
allocated on the same NUMA node as the CPU that handles Receive
completions.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

For clarity, report the posting and completion of Receive CQEs.

Also, the wc->byte_len field contains garbage if wc->status is
non-zero, and the vendor error field contains garbage if wc->status
is zero. For readability, don't save those fields in those cases.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

For FRWR, the computation of max_send_wr is split between
frwr_op_open and rpcrdma_ep_create, which makes it difficult to tell
that the max_send_wr result is currently incorrect if frwr_op_open
has to reduce the credit limit to accommodate a small max_qp_wr.
This is a problem now that extra WRs are needed for backchannel
operations and a drain CQE.

So, refactor the computation so that it is all done in ->ro_open,
and fix the FRWR version of this computation so that it
accommodates HCAs with small max_qp_wr correctly.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Set up RPC/RDMA transport in mount.nfs's network namespace. This
passes the correct namespace information to the RDMA core, similar
to how RPC sockets are created (see xs_create_sock).
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

rdma_resolve_addr(3) says:

> This call is used to map a given destination IP address to a
> usable RDMA address. The IP to RDMA address mapping is done
> using the local routing tables, or via ARP.

If this can't be done, there's no local device that can be used
to establish an RDMA-capable network path to the remote. In this
case, the RDMA CM very quickly posts an RDMA_CM_EVENT_ADDR_ERROR
upcall.

Currently rpcrdma_conn_upcall() converts RDMA_CM_EVENT_ADDR_ERROR
to EHOSTUNREACH. mount.nfs seems to want to retry EHOSTUNREACH
forever, thinking that this is a temporary situation. This makes
mount.nfs appear to hang if I try to mount with proto=rdma through,
say, a conventional Ethernet device.

If the admin has specified proto=rdma along with a server IP address
that requires a network path that does not support RDMA, instead
let's fail with a permanent error. -EPROTONOSUPPORT is returned when
NFSv4 or one of its minor versions is not supported.

-EPROTO is not (currently) retried by mount.nfs.

There are potentially other similar cases where -EPROTO is an
appropriate return code.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NOlga Kornievskaia <kolga@netapp.com>
Tested-by: NAnna Schumaker <Anna.Schumaker@netapp.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

The ro_release_mr methods check whether mr->mr_list is empty.
Therefore, be sure to always use list_del_init when removing an MR
linked into a list using that field. Otherwise, when recovering from
transport failures or device removal, list corruption can result, or
MRs can get mapped or unmapped an odd number of times, resulting in
IOMMU-related failures.

In general this fix is appropriate back to v4.8. However, code
changes since then make it impossible to apply this patch directly
to stable kernels. The fix would have to be applied by hand or
reworked for kernels earlier than v4.16.

Backport guidance -- there are several cases:
- When creating an MR, initialize mr_list so that using list_empty
  on an as-yet-unused MR is safe.
- When an MR is being handled by the remote invalidation path,
  ensure that mr_list is reinitialized when it is removed from
  rl_registered.
- When an MR is being handled by rpcrdma_destroy_mrs, it is removed
  from mr_all, but it may still be on an rl_registered list. In
  that case, the MR needs to be removed from that list before being
  released.
- Other cases are covered by using list_del_init in rpcrdma_mr_pop.

Fixes: 9d6b0409 ('xprtrdma: Place registered MWs on a ... ')
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Michal Kalderon has found some corner cases around device unload
with active NFS mounts that I didn't have the imagination to test
when xprtrdma device removal was added last year.

- The ULP device removal handler is responsible for deallocating
  the PD. That wasn't clear to me initially, and my own testing
  suggested it was not necessary, but that is incorrect.

- The transport destruction path can no longer assume that there
  is a valid ID.

- When destroying a transport, ensure that ib_free_cq() is not
  invoked on a CQ that was already released.
Reported-by: NMichal Kalderon <Michal.Kalderon@cavium.com>
Fixes: bebd0318 ("xprtrdma: Support unplugging an HCA from ...")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Refactor: Both rpcrdma_create_req call sites have to allocate the
buffer where the transport header is built, so just move that
allocation into rpcrdma_create_req.

This buffer is a fixed size. There's no needed information available
in call_allocate that is not also available when the transport is
created.

The original purpose for allocating these buffers on demand was to
reduce the possibility that an allocation failure during transport
creation will hork the mount operation during low memory scenarios.
Some relief for this rare possibility is coming up in the next few
patches.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

With FRWR, the client transport can perform memory registration and
post a Send with just a single ib_post_send.

This reduces contention between the send_request path and the Send
Completion handlers, and reduces the overhead of registering a chunk
that has multiple segments.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Create fewer MRs on average. Many workloads don't need as many as
32 MRs, and the transport can now quickly restock the MR free list.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Currently, when the MR free list is exhausted during marshaling, the
RPC/RDMA transport places the RPC task on the delayq, which forces a
wait for HZ >> 2 before the marshal and send is retried.

With this change, the transport now places such an RPC task on the
pending queue, and wakes it just as soon as more MRs have been
created. Creating more MRs typically takes less than a millisecond,
and this waking mechanism is less deadlock-prone.

Moreover, the waiting RPC task is holding the transport's write
lock, which blocks the transport from sending RPCs. Therefore faster
recovery from MR exhaustion is desirable.

This is the same mechanism that the TCP transport utilizes when
handling write buffer space exhaustion.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: The generic rq_connect_cookie is sufficient to detect RPC
Call retransmission.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>