1/ rcu_dereference isn't correct: that field isn't
   RCU protected.   It could potentially change at any time
   so ACCESS_ONCE might be justified.

   changes to ->d_parent are protected by ->d_seq.  However
   that isn't always checked after ->d_revalidate is called,
   so it is safest to keep the double-check that ->d_parent
   hasn't changed at the end of these functions.

2/ in nfs4_lookup_revalidate, "->d_parent" was forgotten.
   So 'parent' was not the parent of 'dentry'.
   This fails safe is the context is that dentry->d_inode is
   NULL, and the result of parent->d_inode being NULL is
   that ECHILD is returned, which is always safe.
Reported-by: Nkbuild test robot <fengguang.wu@intel.com>
Signed-off-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

The access cache is used during RCU-walk path lookups, so it is best
to avoid locking if possible as taking a lock kills concurrency.

The rbtree is not rcu-safe and cannot easily be made so.
Instead we simply check the last (i.e. most recent) entry on the LRU
list.  If this doesn't match, then we return -ECHILD and retry in
lock/refcount mode.

This requires freeing the nfs_access_entry struct with rcu, and
requires using rcu access primatives when adding entries to the lru, and
when examining the last entry.

Calling put_rpccred before kfree_rcu looks a bit odd, but as
put_rpccred already provides rcu protection, we know that the cred will
not actually be freed until the next grace period, so any concurrent
access will be safe.

This patch provides about 5% performance improvement on a stat-heavy
synthetic work load with 4 threads on a 2-core CPU.
Signed-off-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

It fails with -ECHILD rather than make an RPC call.

This allows nfs_lookup_revalidate to call it in RCU-walk mode.
Signed-off-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

This requires nfs_check_verifier to take an rcu_walk flag, and requires
an rcu version of nfs_revalidate_inode which returns -ECHILD rather
than making an RPC call.

With this, nfs_lookup_revalidate can call nfs_neg_need_reval in
RCU-walk mode.

We can also move the LOOKUP_RCU check past the nfs_check_verifier()
call in nfs_lookup_revalidate.

If RCU_WALK prevents nfs_check_verifier or nfs_neg_need_reval from
doing a full check, they return a status indicating that a revalidation
is required.  As this revalidation will not be possible in RCU_WALK
mode, -ECHILD will ultimately be returned, which is the desired result.
Signed-off-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

nfs_permission makes two calls which are not always safe in RCU_WALK,
rpc_lookup_cred and nfs_do_access.

The second can easily be made rcu-safe by aborting with -ECHILD before
making the RPC call.

The former can be made rcu-safe by calling rpc_lookup_cred_nonblock()
instead.
As this will almost always succeed, we use it even when RCU_WALK
isn't being used as it still saves some spinlocks in a common case.
We only fall back to rpc_lookup_cred() if rpc_lookup_cred_nonblock()
fails and MAY_NOT_BLOCK isn't set.

This optimisation (always trying rpc_lookup_cred_nonblock()) is
particularly important when a security module is active.
In that case inode_permission() may return -ECHILD from
security_inode_permission() even though ->permission() succeeded in
RCU_WALK mode.
This leads to may_lookup() retrying inode_permission after performing
unlazy_walk().  The spinlock that rpc_lookup_cred() takes is often
more expensive than anything security_inode_permission() does, so that
spinlock becomes the main bottleneck.
Signed-off-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

The new flag RPCAUTH_LOOKUP_RCU to credential lookup avoids locking,
does not take a reference on the returned credential, and returns
-ECHILD if a simple lookup was not possible.

The returned value can only be used within an rcu_read_lock protected
region.

The main user of this is the new rpc_lookup_cred_nonblock() which
returns a pointer to the current credential which is only rcu-safe (no
ref-count held), and might return -ECHILD if allocation was required.
Signed-off-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

nfs_lookup_revalidate, nfs4_lookup_revalidate, and nfs_permission
all need to understand and handle RCU-walk for NFS to gain the
benefits of RCU-walk for cached information.

Currently these functions all immediately return -ECHILD
if the relevant flag (LOOKUP_RCU or MAY_NOT_BLOCK) is set.

This patch pushes those tests later in the code so that we only abort
immediately before we enter rcu-unsafe code.  As subsequent patches
make that rcu-unsafe code rcu-safe, several of these new tests will
disappear.

With this patch there are several paths through the code which will no
longer return -ECHILD during an RCU-walk.  However these are mostly
error paths or other uninteresting cases.

A noteworthy change in nfs_lookup_revalidate is that we don't take
(or put) the reference to ->d_parent when LOOKUP_RCU is set.
Rather we rcu_dereference ->d_parent, and check that ->d_inode
is not NULL.  We also check that ->d_parent hasn't changed after
all the tests.

In nfs4_lookup_revalidate we simply avoid testing LOOKUP_RCU on the
path that only calls nfs_lookup_revalidate() as that function
already performs the required test.
Signed-off-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

nfs4_lookup_revalidate only uses 'parent' to get 'dir', and only
uses 'dir' if 'inode == NULL'.

So we don't need to find out what 'parent' or 'dir' is until we
know that 'inode' is NULL.

By moving 'dget_parent' inside the 'if', we can reduce the number of
call sites for 'dput(parent)'.
Signed-off-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

There is a couple of places in client code where returned value
of try_module_get() is ignored. As a result there is a small chance
to premature unload module because of unbalanced refcounting.

The patch adds error handling in that places.

Found by Linux Driver Verification project (linuxtesting.org).
Signed-off-by: NAlexey Khoroshilov <khoroshilov@ispras.ru>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

This is useful when lsegs need to be released while holding locks.
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

nfs_page_find_head_request_locked looks through the regular nfs commit lists
when the page is swapped out, but doesn't look through the pnfs commit lists.

I'm not sure if anyone has hit any issues caused by this.
Suggested-by: NPeng Tao <tao.peng@primarydata.com>
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Fix the comment in nfs_page.h for PG_INODE_REF to reflect that it's no longer
set only on head requests. Also add a WARN_ON_ONCE in nfs_inode_remove_request
as PG_INODE_REF should always be set.
Suggested-by: NPeng Tao <tao.peng@primarydata.com>
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Return errors from wait_on_bit_lock from nfs_page_group_lock.

Add a bool argument @wait to nfs_page_group_lock. If true, loop over
wait_on_bit_lock until it returns cleanly. If false, return the error
from wait_on_bit_lock.
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

It's always 0.
Signed-off-by: NJeff Layton <jlayton@primarydata.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Fix the endianness handling in gss_wrap_kerberos_v1 and drop the memset
call there in favor of setting the filler bytes directly.

In gss_wrap_kerberos_v2, get rid of the "ec" variable which is always
zero, and drop the endianness conversion of 0. Sparse handles 0 as a
special case, so it's not necessary.
Signed-off-by: NJeff Layton <jlayton@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Use u16 pointer in setup_token and setup_token_v2. None of the fields
are actually handled as __be16, so this simplifies the code a bit. Also
get rid of some unneeded pointer increments.
Signed-off-by: NJeff Layton <jlayton@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

The handling of the gc_ctx pointer only seems to be partially RCU-safe.
The assignment and freeing are done using RCU, but many places in the
code seem to dereference that pointer without proper RCU safeguards.

Fix them to use rcu_dereference and to rcu_read_lock/unlock, and to
properly handle the case where the pointer is NULL.

Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: NJeff Layton <jlayton@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Commit 5b22216e (nfs: __rcu annotations) added a __rcu annotation to
the gc_gss_ctx field. I see no rationale for adding that though, as that
field does not seem to be managed via RCU at all.

Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: NJeff Layton <jlayton@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

If you have an NFSv4 mounted directory which does not container 'foo'
and:

  ls -l foo
  ssh $server touch foo
  cat foo

then the 'cat' will fail (usually, depending a bit on the various
cache ages).  This is correct as negative looks are cached by default.
However with the same initial conditions:

  cat foo
  ssh $server touch foo
  cat foo

will usually succeed.  This is because an "open" does not add a
negative dentry to the dcache, while a "lookup" does.

This can have negative performance effects.  When "gcc" searches for
an include file, it will try to "open" the file in every director in
the search path.  Without caching of negative "open" results, this
generates much more traffic to the server than it should (or than
NFSv3 does).

The root of the problem is that _nfs4_open_and_get_state() will call
d_add_unique() on a positive result, but not on a negative result.
Compare with nfs_lookup() which calls d_materialise_unique on both
a positive result and on ENOENT.

This patch adds a call d_add() in the ENOENT case for
_nfs4_open_and_get_state() and also calls nfs_set_verifier().

With it, many fewer "open" requests for known-non-existent files are
sent to the server.
Signed-off-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

There was a check for result being not NULL. But get_acl() may return
NULL, or ERR_PTR, or actual pointer.
The purpose of the function where current change is done is to "list
ACLs only when they are available", so any error condition of get_acl()
mustn't be elevated, and returning 0 there is still valid.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=81111Signed-off-by: NAndrey Utkin <andrey.krieger.utkin@gmail.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Fixes: 74adf83f (nfs: only show Posix ACLs in listxattr if actually...)
Cc: stable@vger.kernel.org # 3.14+
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

* 'nfs-rdma' of git://git.linux-nfs.org/projects/anna/nfs-rdma: (916 commits)
  xprtrdma: Handle additional connection events
  xprtrdma: Remove RPCRDMA_PERSISTENT_REGISTRATION macro
  xprtrdma: Make rpcrdma_ep_disconnect() return void
  xprtrdma: Schedule reply tasklet once per upcall
  xprtrdma: Allocate each struct rpcrdma_mw separately
  xprtrdma: Rename frmr_wr
  xprtrdma: Disable completions for LOCAL_INV Work Requests
  xprtrdma: Disable completions for FAST_REG_MR Work Requests
  xprtrdma: Don't post a LOCAL_INV in rpcrdma_register_frmr_external()
  xprtrdma: Reset FRMRs after a flushed LOCAL_INV Work Request
  xprtrdma: Reset FRMRs when FAST_REG_MR is flushed by a disconnect
  xprtrdma: Properly handle exhaustion of the rb_mws list
  xprtrdma: Chain together all MWs in same buffer pool
  xprtrdma: Back off rkey when FAST_REG_MR fails
  xprtrdma: Unclutter struct rpcrdma_mr_seg
  xprtrdma: Don't invalidate FRMRs if registration fails
  xprtrdma: On disconnect, don't ignore pending CQEs
  xprtrdma: Update rkeys after transport reconnect
  xprtrdma: Limit data payload size for ALLPHYSICAL
  xprtrdma: Protect ia->ri_id when unmapping/invalidating MRs
  ...

This may be used to limit the number of cached credentials building up
inside the access cache.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

In some cases where the credentials are not often reused, we may want
to limit their total number just in order to make the negative lookups
in the hash table more manageable.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Commit 38ca83a5 added RDMA_CM_EVENT_TIMEWAIT_EXIT. But that status
is relevant only for consumers that re-use their QPs on new
connections. xprtrdma creates a fresh QP on reconnection, so that
event should be explicitly ignored.

Squelch the alarming "unexpected CM event" message.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up.

RPCRDMA_PERSISTENT_REGISTRATION was a compile-time switch between
RPCRDMA_REGISTER mode and RPCRDMA_ALLPHYSICAL mode.  Since
RPCRDMA_REGISTER has been removed, there's no need for the extra
conditional compilation.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: The return code is used only for dprintk's that are
already redundant.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Minor optimization: grab rpcrdma_tk_lock_g and disable hard IRQs
just once after clearing the receive completion queue.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Currently rpcrdma_buffer_create() allocates struct rpcrdma_mw's as
a single contiguous area of memory. It amounts to quite a bit of
memory, and there's no requirement for these to be carved from a
single piece of contiguous memory.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Name frmr_wr after the opcode of the Work Request,
consistent with the send and local invalidation paths.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Instead of relying on a completion to change the state of an FRMR
to FRMR_IS_INVALID, set it in advance. If an error occurs, a completion
will fire anyway and mark the FRMR FRMR_IS_STALE.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Instead of relying on a completion to change the state of an FRMR
to FRMR_IS_VALID, set it in advance. If an error occurs, a completion
will fire anyway and mark the FRMR FRMR_IS_STALE.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Any FRMR arriving in rpcrdma_register_frmr_external() is now
guaranteed to be either invalid, or to be targeted by a queued
LOCAL_INV that will invalidate it before the adapter processes
the FAST_REG_MR being built here.

The problem with current arrangement of chaining a LOCAL_INV to the
FAST_REG_MR is that if the transport is not connected, the LOCAL_INV
is flushed and the FAST_REG_MR is flushed. This leaves the FRMR
valid with the old rkey. But rpcrdma_register_frmr_external() has
already bumped the in-memory rkey.

Next time through rpcrdma_register_frmr_external(), a LOCAL_INV and
FAST_REG_MR is attempted again because the FRMR is still valid. But
the rkey no longer matches the hardware's rkey, and a memory
management operation error occurs.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

When a LOCAL_INV Work Request is flushed, it leaves an FRMR in the
VALID state. This FRMR can be returned by rpcrdma_buffer_get(), and
must be knocked down in rpcrdma_register_frmr_external() before it
can be re-used.

Instead, capture these in rpcrdma_buffer_get(), and reset them.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

FAST_REG_MR Work Requests update a Memory Region's rkey. Rkey's are
used to block unwanted access to the memory controlled by an MR. The
rkey is passed to the receiver (the NFS server, in our case), and is
also used by xprtrdma to invalidate the MR when the RPC is complete.

When a FAST_REG_MR Work Request is flushed after a transport
disconnect, xprtrdma cannot tell whether the WR actually hit the
adapter or not. So it is indeterminant at that point whether the
existing rkey is still valid.

After the transport connection is re-established, the next
FAST_REG_MR or LOCAL_INV Work Request against that MR can sometimes
fail because the rkey value does not match what xprtrdma expects.

The only reliable way to recover in this case is to deregister and
register the MR before it is used again. These operations can be
done only in a process context, so handle it in the transport
connect worker.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

If the rb_mws list is exhausted, clean up and return NULL so that
call_allocate() will delay and try again.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

During connection loss recovery, need to visit every MW in a
buffer pool. Any MW that is in use by an RPC will not be on the
rb_mws list.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

If posting a FAST_REG_MR Work Reqeust fails, revert the rkey update
to avoid subsequent IB_WC_MW_BIND_ERR completions.
Suggested-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean ups:
 - make it obvious that the rl_mw field is a pointer -- allocated
   separately, not as part of struct rpcrdma_mr_seg
 - promote "struct {} frmr;" to a named type
 - promote the state enum to a named type
 - name the MW state field the same way other fields in
   rpcrdma_mw are named
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

If FRMR registration fails, it's likely to transition the QP to the
error state. Or, registration may have failed because the QP is
_already_ in ERROR.

Thus calling rpcrdma_deregister_external() in
rpcrdma_create_chunks() is useless in FRMR mode: the LOCAL_INVs just
get flushed.

It is safe to leave existing registrations: when FRMR registration
is tried again, rpcrdma_register_frmr_external() checks if each FRMR
is already/still VALID, and knocks it down first if it is.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NSteve Wise <swise@opengridcomputing.com>
Tested-by: NShirley Ma <shirley.ma@oracle.com>
Tested-by: NDevesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>