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: The logic to wait for write space is common to a bunch of
the encoding helper functions. Lift it out and put it in the tail
of rpcrdma_marshal_req().
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

The use of -EAGAIN in rpcrdma_convert_iovs() is a latent bug: the
transport never calls xprt_write_space() when more pages become
available. -ENOBUFS will trigger the correct "delay briefly and call
again" logic.

Fixes: 7a89f9c6 ("xprtrdma: Honor ->send_request API contract")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Cc: stable@vger.kernel.org # 4.8+
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.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>

This simplifies allocation of the generic RPC slot and xprtrdma
specific per-RPC resources.

It also makes xprtrdma more like the socket-based transports:
->buf_alloc and ->buf_free are now responsible only for send and
receive buffers.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

rpcrdma_buffer_get acquires an rpcrdma_req and rep for each RPC.
Currently this is done in the call_allocate action, and sometimes it
can fail if there are many outstanding RPCs.

When call_allocate fails, the RPC task is put on the delayq. It is
awoken a few milliseconds later, but there's no guarantee it will
get a buffer at that time. The RPC task can be repeatedly put back
to sleep or even starved.

The call_allocate action should rarely fail. The delayq mechanism is
not meant to deal with transport congestion.

In the current sunrpc stack, there is a friendlier way to deal with
this situation. These objects are actually tantamount to an RPC
slot (rpc_rqst) and there is a separate FSM action, distinct from
call_allocate, for allocating slot resources. This is the
call_reserve action.

When allocation fails during this action, the RPC is placed on the
transport's backlog queue. The backlog mechanism provides a stronger
guarantee that when the RPC is awoken, a buffer will be available
for it; and backlogged RPCs are awoken one-at-a-time.

To make slot resource allocation occur in the call_reserve action,
create special ->alloc_slot and ->free_slot call-outs for xprtrdma.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Refactor: xprtrdma needs to have better control over when RPCs are
awoken from the backlog queue, so replace xprt_free_slot with a
transport op callout.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

alloc_slot is a transport-specific op, but initializing an rpc_rqst
is common to all transports. In addition, the only part of initial-
izing an rpc_rqst that needs serialization is getting a fresh XID.

Move rpc_rqst initialization to common code in preparation for
adding a transport-specific alloc_slot to xprtrdma.
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>

if we ever hit rpc_gssd_dummy_depopulate() dentry passed to
it has refcount equal to 1.  __rpc_rmpipe() drops it and
dput() done after that hits an already freed dentry.

Cc: stable@kernel.org
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

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

This information can help track down local misconfiguration issues
as well as network partitions and unresponsive servers.

There are several ways to send a ping, and with transport multi-
plexing, the exact rpc_xprt that is used is sometimes not known by
the upper layer. The rpc_xprt pointer passed to the trace point
call also has to be RCU-safe.

I found a spot inside the client FSM where an rpc_xprt pointer is
always available and safe to use.
Suggested-by: NBill Baker <Bill.Baker@oracle.com>
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Introduce a low-overhead mechanism to report information about
latencies of individual RPCs. The goal is to enable user space to
filter the trace record for latency outliers, or build histograms,
etc.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: struct rpc_task carries a pointer to a struct rpc_clnt,
and in fact task->tk_client is always what is passed into trace
points that are already passing @task.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

If recording xprt->stat.max_slots is moved into xprt_alloc_slot,
then xprt->num_reqs is never manipulated outside
xprt->reserve_lock. There's no longer a need for xprt->num_reqs to
be atomic.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Some RPC transports have more overhead in their send_request
callouts than others. For example, for RPC-over-RDMA:

- Marshaling an RPC often has to DMA map the RPC arguments

- Registration methods perform memory registration as part of
  marshaling

To capture just server and network latencies more precisely: when
sending a Call, capture the rq_xtime timestamp _after_ the transport
header has been marshaled.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Some RPC transports have more overhead in their reply handlers
than others. For example, for RPC-over-RDMA:

- RPC completion has to wait for memory invalidation, which is
  not a part of the server/network round trip

- Recently a context switch was introduced into the reply handler,
  which further artificially inflates the measure of RPC RTT

To capture just server and network latencies more precisely: when
receiving a reply, compute the RTT as soon as the XID is recognized
rather than at RPC completion time.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Since commit 33849792 ("xprtrdma: Detect unreachable NFS/RDMA
servers more reliably"), the xprtrdma transport now has a ->timer
callout. But xprtrdma does not need to compute RTT data, only UDP
needs that. Move the xprt_update_rtt call into the UDP transport
implementation.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
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>

RPC-over-RDMA version 1 credit accounting relies on there being a
response message for every RPC Call. This means that RPC procedures
that have no reply will disrupt credit accounting, just in the same
way as a retransmit would (since it is sent because no reply has
arrived). Deal with the "no reply" case the same way.
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>

Clean up: We need to check only that the value does not exceed the
range of the u8 field it's going into.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

With v4.15, on one of my NFS/RDMA clients I measured a nearly
doubling in the latency of small read and write system calls. There
was no change in server round trip time. The extra latency appears
in the whole RPC execution path.

"git bisect" settled on commit ccede759 ("xprtrdma: Spread reply
processing over more CPUs") .

After some experimentation, I found that leaving the WQ bound and
allowing the scheduler to pick the dispatch CPU seems to eliminate
the long latencies, and it does not introduce any new regressions.

The fix is implemented by reverting only the part of
commit ccede759 ("xprtrdma: Spread reply processing over more
CPUs") that dispatches RPC replies specifically on the CPU where the
matching RPC call was made.

Interestingly, saving the CPU number and later queuing reply
processing there was effective _only_ for a NFS READ and WRITE
request. On my NUMA client, in-kernel RPC reply processing for
asynchronous RPCs was dispatched on the same CPU where the RPC call
was made, as expected. However synchronous RPCs seem to get their
reply dispatched on some other CPU than where the call was placed,
every time.

Fixes: ccede759 ("xprtrdma: Spread reply processing over ... ")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Cc: stable@vger.kernel.org # v4.15+
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

make_checksum_hmac_md5() is allocating an HMAC transform and doing
crypto API calls in the following order:

    crypto_ahash_init()
    crypto_ahash_setkey()
    crypto_ahash_digest()

This is wrong because it makes no sense to init() the request before a
key has been set, given that the initial state depends on the key.  And
digest() is short for init() + update() + final(), so in this case
there's no need to explicitly call init() at all.

Before commit 9fa68f62 ("crypto: hash - prevent using keyed hashes
without setting key") the extra init() had no real effect, at least for
the software HMAC implementation.  (There are also hardware drivers that
implement HMAC-MD5, and it's not immediately obvious how gracefully they
handle init() before setkey().)  But now the crypto API detects this
incorrect initialization and returns -ENOKEY.  This is breaking NFS
mounts in some cases.

Fix it by removing the incorrect call to crypto_ahash_init().
Reported-by: NMichael Young <m.a.young@durham.ac.uk>
Fixes: 9fa68f62 ("crypto: hash - prevent using keyed hashes without setting key")
Fixes: fffdaef2 ("gss_krb5: Add support for rc4-hmac encryption")
Cc: stable@vger.kernel.org
Signed-off-by: NEric Biggers <ebiggers@google.com>
Signed-off-by: NJ. Bruce Fields <bfields@redhat.com>

Move common code in NFSD's legacy SYMLINK decoders into a helper.
The immediate benefits include:

 - one fewer data copies on transports that support DDP
 - consistent error checking across all versions
 - reduction of code duplication
 - support for both legal forms of SYMLINK requests on RDMA
   transports for all versions of NFS (in particular, NFSv2, for
   completeness)

In the long term, this helper is an appropriate spot to perform a
per-transport call-out to fill the pathname argument using, say,
RDMA Reads.

Filling the pathname in the proc function also means that eventually
the incoming filehandle can be interpreted so that filesystem-
specific memory can be allocated as a sink for the pathname
argument, rather than using anonymous pages.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NJ. Bruce Fields <bfields@redhat.com>

Move common code in NFSD's legacy NFS WRITE decoders into a helper.
The immediate benefit is reduction of code duplication and some nice
micro-optimizations (see below).

In the long term, this helper can perform a per-transport call-out
to fill the rq_vec (say, using RDMA Reads).

The legacy WRITE decoders and procs are changed to work like NFSv4,
which constructs the rq_vec just before it is about to call
vfs_writev.

Why? Calling a transport call-out from the proc instead of the XDR
decoder means that the incoming FH can be resolved to a particular
filesystem and file. This would allow pages from the backing file to
be presented to the transport to be filled, rather than presenting
anonymous pages and copying or flipping them into the file's page
cache later.

I also prefer using the pages in rq_arg.pages, instead of pulling
the data pages directly out of the rqstp::rq_pages array. This is
currently the way the NFSv3 write decoder works, but the other two
do not seem to take this approach. Fixing this removes the only
reference to rq_pages found in NFSD, eliminating an NFSD assumption
about how transports use the pages in rq_pages.

Lastly, avoid setting up the first element of rq_vec as a zero-
length buffer. This happens with an RDMA transport when a normal
Read chunk is present because the data payload is in rq_arg's
page list (none of it is in the head buffer).
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NJ. Bruce Fields <bfields@redhat.com>

Record the time between when a rqstp is enqueued on a transport
and when it is dequeued. This includes how long the rqstp waits on
the queue and how long it takes the kernel scheduler to wake a
nfsd thread to service it.

The svc_xprt_dequeue trace point is altered to include the number
of microseconds between xprt_enqueue and xprt_dequeue.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NJ. Bruce Fields <bfields@redhat.com>

Introduce a mechanism to report the server-side execution latency of
each RPC. The goal is to enable user space to filter the trace
record for latency outliers, build histograms, etc.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NJ. Bruce Fields <bfields@redhat.com>