In the years since the max readahead size was fixed in NFS, a number of
things have happened:
- Users can now set the value directly using /sys/class/bdi
- NFS max supported block sizes have increased by several orders of
magnitude from 64K to 1MB.
- Disk access latencies are orders of magnitude faster due to SSD + NVME.
In particular note that if the server is advertising 1MB as the optimal
read size, as that will set the readahead size to 15MB.
Let's therefore adjust down, and try to default to VM_READAHEAD_PAGES.
However let's inform the VM about our preferred block size so that it
can choose to round up in cases where that makes sense.
Reported-by: Alkis Georgopoulos <alkisg@gmail.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If a LOCKU request receives a NFS4ERR_OLD_STATEID, then bump the
seqid before resending. Ensure we only bump the seqid by 1.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If a CLOSE or OPEN_DOWNGRADE operation receives a NFS4ERR_OLD_STATEID
then bump the seqid before resending. Ensure we only bump the seqid
by 1.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If OPEN_DOWNGRADE returns a state error, then we want to initiate
state recovery in addition to marking the stateid as closed.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If a LAYOUTRETURN receives a reply of NFS4ERR_OLD_STATEID then assume we've
missed an update, and just bump the stateid.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Add a helper function to increment stateid seqids according to the
rules specified in RFC5661 Section 8.2.2.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Handle RPC level errors by assuming that the RPC call was successful.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If the server sends a NFS4ERR_DELAY, then allow the caller to retry.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Both close and delegreturn have identical code to handle pNFS
return-on-close. This patch refactors that code and places it
in pnfs.c
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
IF the server rejected our layout return with a state error such as
NFS4ERR_BAD_STATEID, or even a stale inode error, then we do want
to clear out all the remaining layout segments and mark that stateid
as invalid.
Fixes: 1c5bd76d17 ("pNFS: Enable layoutreturn operation for...")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
This check has been hanging out since we used to have parallel paths to add
dentry in nfs_create(), but that hasn't been the case for some years.
Signed-off-by: Benjamin Coddington <bcodding@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Since commit b0c6108ecf ("nfs_instantiate(): prevent multiple aliases for
directory inode"), nfs_instantiate() may succeed without actually
instantiating the dentry that was passed in. That can be problematic for
some callers in NFSv3, so this patch breaks things up so we can get the
actual dentry obtained.
Signed-off-by: Benjamin Coddington <bcodding@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If the congestion window closes just as the transport disconnects,
a reconnect is never driven because:
1. The XPRT_CONG_WAIT flag prevents tasks from taking the write lock
2. There's no wake-up of the first task on the xprt->sending queue
To address this, clear the congestion wait flag as part of
completing a disconnect.
Fixes: 75891f502f ("SUNRPC: Support for congestion control ... ")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If the copy of the RPC reply into our buffers did not complete, and
we could end up with a truncated message. In that case, just resend
the call.
Fixes: a0584ee9ae ("SUNRPC: Use struct xdr_stream when decoding...")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Let the name reflect the single use. The function now assumes the GSS MIC
is the last object in the buffer.
Signed-off-by: Benjamin Coddington <bcodding@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The GSS Message Integrity Check data for krb5i may lie partially in the XDR
reply buffer's pages and tail. If so, we try to copy the entire MIC into
free space in the tail. But as the estimations of the slack space required
for authentication and verification have improved there may be less free
space in the tail to complete this copy -- see commit 2c94b8eca1
("SUNRPC: Use au_rslack when computing reply buffer size"). In fact, there
may only be room in the tail for a single copy of the MIC, and not part of
the MIC and then another complete copy.
The real world failure reported is that `ls` of a directory on NFS may
sometimes return -EIO, which can be traced back to xdr_buf_read_netobj()
failing to find available free space in the tail to copy the MIC.
Fix this by checking for the case of the MIC crossing the boundaries of
head, pages, and tail. If so, shift the buffer until the MIC is contained
completely within the pages or tail. This allows the remainder of the
function to create a sub buffer that directly address the complete MIC.
Signed-off-by: Benjamin Coddington <bcodding@redhat.com>
Cc: stable@vger.kernel.org # v5.1
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Ensure that we set task->tk_rpc_status for all RPC level errors so that
the caller can distinguish between those and server reply status errors.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If we've removed the request from the receive list, and have added
it back after resetting the request receive buffer, then we should
only receive message data if it is a new reply (i.e. if
transport->recv.copied is zero).
Fixes: 277e4ab7d5 ("SUNRPC: Simplify TCP receive code by switching...")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Ensure that we dequeue the request from the transport receive queue
while we're re-encoding to prevent issues like use-after-free when
we release the bvec.
Fixes: 7536908982 ("SUNRPC: Ensure the bvecs are reset when we re-encode...")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Cc: stable@vger.kernel.org # v4.20+
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Eli Dorfman reports that after a series of idle disconnects, an
RPC/RDMA transport becomes unusable (rdma_create_qp returns
-ENOMEM). Problem was tracked down to increasing Send Queue size
after each reconnect.
The rdma_create_qp() API does not promise to leave its @qp_init_attr
parameter unaltered. In fact, some drivers do modify one or more of
its fields. Thus our calls to rdma_create_qp must use a fresh copy
of ib_qp_init_attr each time.
This fix is appropriate for kernels dating back to late 2007, though
it will have to be adapted, as the connect code has changed over the
years.
Reported-by: Eli Dorfman <eli@vastdata.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Ensure that the re-establishment delay does not grow exponentially
on each good reconnect. This probably should have been part of
commit 675dd90ad0 ("xprtrdma: Modernize ops->connect").
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The optimization done in "xprtrdma: Simplify rpcrdma_mr_pop" was a
bit too optimistic. MRs left over after a reconnect still need to
be recycled, not added back to the free list, since they could be
in flight or actually fully registered.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
We need to use the custom rpc_task_setup here to set the
RPC_TASK_NO_ROUND_ROBIN flag on the RPC call.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
An async call followed by an rpc_wait_for_completion() is basically the
same as a synchronous call, so we can use nfs4_call_sync_custom() to
keep our custom callback ops and the RPC_TASK_NO_ROUND_ROBIN flag.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
There are a few cases where we need to manually configure the
rpc_task_setup structure to get the behavior we want.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
In nfs4_try_migration(), if nfs4_begin_drain_session() fails, the
previously allocated 'page' and 'locations' are not deallocated, leading to
memory leaks. To fix this issue, go to the 'out' label to free 'page' and
'locations' before returning the error.
Signed-off-by: Wenwen Wang <wenwen@cs.uga.edu>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Micro-optimization: In rpcrdma_post_recvs, since commit e340c2d6ef
("xprtrdma: Reduce the doorbell rate (Receive)"), the common case is
to return without doing anything. Found with perf.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Micro-optimization: Save the cost of three function calls during
transport header encoding.
These were "noinline" before to generate more meaningful call stacks
during debugging, but this code is now pretty stable.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
For the moment the returned value just happens to be correct because
the current backchannel server implementation does not vary the
number of credits it offers. The spec does permit this value to
change during the lifetime of a connection, however.
The actual maximum is fixed for all RPC/RDMA transports, because
each transport instance has to pre-allocate the resources for
processing BC requests. That's the value that should be returned.
Fixes: 7402a4fedc ("SUNRPC: Fix up backchannel slot table ... ")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up: The function name should match the documenting comment.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
rpcrdma_rep objects are removed from their free list by only a
single thread: the Receive completion handler. Thus that free list
can be converted to an llist, where a single-threaded consumer and
a multi-threaded producer (rpcrdma_buffer_put) can both access the
llist without the need for any serialization.
This eliminates spin lock contention between the Receive completion
handler and rpcrdma_buffer_get, and makes the rep consumer wait-
free.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up: Now that the free list is used sparingly, get rid of the
separate spin lock protecting it.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Instead of a globally-contended MR free list, cache MRs in each
rpcrdma_req as they are released. This means acquiring and releasing
an MR will be lock-free in the common case, even outside the
transport send lock.
The original idea of per-rpcrdma_req MR free lists was suggested by
Shirley Ma <shirley.ma@oracle.com> several years ago. I just now
figured out how to make that idea work with on-demand MR allocation.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Probably would be good to also pass GFP flags to ib_alloc_mr.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Refactor: Retrieve an MR and handle error recovery entirely in
rpc_rdma.c, as this is not a device-specific function.
Note that since commit 89f90fe1ad ("SUNRPC: Allow calls to
xprt_transmit() to drain the entire transmit queue"), the
xprt_transmit function handles the cond_resched. The transport no
longer has to do this itself.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up. There is only one remaining rpcrdma_mr_put call site, and
it can be directly replaced with unmap_and_put because mr->mr_dir is
set to DMA_NONE just before the call.
Now all the call sites do a DMA unmap, and we can just rename
mr_unmap_and_put to mr_put, which nicely matches mr_get.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up: rpcrdma_mr_pop call sites check if the list is empty
first. Let's replace the list_empty with less costly logic.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Commit 48be539dd4 ("xprtrdma: Introduce ->alloc_slot call-out for
xprtrdma") added a separate alloc_slot and free_slot to the RPC/RDMA
transport. Later, commit 75891f502f ("SUNRPC: Support for
congestion control when queuing is enabled") modified the generic
alloc/free_slot methods, but neglected the methods in xprtrdma.
Found via code review.
Fixes: 75891f502f ("SUNRPC: Support for congestion control ... ")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up: There are other "all" list heads. For code clarity
distinguish this one as for use only for MRs by renaming it.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Make the field name the same for all trace points that handle
pointers to struct rpcrdma_rep. That makes it easy to grep for
matching rep points in trace output.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
I've heard rumors of an NFS/RDMA server implementation that has a
default credit limit of 1024. The client's default setting remains
at 128.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Although I haven't seen any performance results that justify it,
I've received several complaints that NFS/RDMA no longer supports
a maximum rsize and wsize of 1MB. These days it is somewhat smaller.
To simplify the logic that determines whether a chunk list is
necessary, the implementation uses a fixed maximum size of the
transport header. Currently that maximum size is 256 bytes, one
quarter of the default inline threshold size for RPC/RDMA v1.
Since commit a78868497c ("xprtrdma: Reduce max_frwr_depth"), the
size of chunks is also smaller to take advantage of inline page
lists in device internal MR data structures.
The combination of these two design choices has reduced the maximum
NFS rsize and wsize that can be used for most RNIC/HCAs. Increasing
the maximum transport header size and the maximum number of RDMA
segments it can contain increases the negotiated maximum rsize/wsize
on common RNIC/HCAs.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Commit 302d3deb20 ("xprtrdma: Prevent inline overflow") added this
calculation back in 2016, but got it wrong. I tested only the lower
bound, which is why there is a max_t there. The upper bound should be
rounded up too.
Now, when using DIV_ROUND_UP, that takes care of the lower bound as
well.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Comment was made obsolete by commit 8cec3dba76 ("xprtrdma:
rpcrdma_regbuf alignment").
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Things have changed since this comment was written. In particular,
the reworking of connection closing, on-demand creation of MRs, and
the removal of fr_state all mean that deferring MR recovery to
frwr_map is no longer needed. The description is obsolete.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Micro-optimization: For xdr_commit_encode call sites in
net/sunrpc/xdr.c, eliminate the extra calling sequence. On my
client, this change saves about a microsecond for every 30 calls
to xdr_reserve_space().
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
