Currently the underlying RDMA device is chosen at transport set-up
time. But it will soon be at connect time instead.
The maximum size of a transport header is based on device
capabilities. Thus transport header buffers have to be allocated
_after_ the underlying device has been chosen (via address and route
resolution); ie, in the connect worker.
Thus, move the allocation of transport header buffers to the connect
worker, after the point at which the underlying RDMA device has been
chosen.
This also means the RDMA device is available to do a DMA mapping of
these buffers at connect time, instead of in the hot I/O path. Make
that optimization as well.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Refactor: Perform the "is supported" check in rpcrdma_ep_create()
instead of in rpcrdma_ia_open(). frwr_open() is where most of the
logic to query device attributes is already located.
The current code displays a redundant error message when the device
does not support FRWR. As an additional clean-up, this patch removes
the extra message.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
To support device hotplug and migrating a connection between devices
of different capabilities, we have to guarantee that all in-kernel
devices can support the same max NFS payload size (1 megabyte).
This means that possibly one or two in-tree devices are no longer
supported for NFS/RDMA because they cannot support 1MB rsize/wsize.
The only one I confirmed was cxgb3, but it has already been removed
from the kernel.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up: there is no need to keep two copies of the same value.
Also, in subsequent patches, rpcrdma_ep_create() will be called in
the connect worker rather than at set-up time.
Minor fix: Initialize the transport's sendctx to the value based on
the capabilities of the underlying device, not the maximum setting.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The size of the sendctx queue depends on the value stored in
ia->ri_max_send_sges. This value is determined by querying the
underlying device.
Eventually, rpcrdma_ia_open() and rpcrdma_ep_create() will be called
in the connect worker rather than at transport set-up time. The
underlying device will not have been chosen device set-up time.
The sendctx queue will thus have to be created after the underlying
device has been chosen via address and route resolution; in other
words, in the connect worker.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean-up. The max_send_sge value also happens to be stored in
ep->rep_attr. Let's keep just a single copy.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The empty_iov structure is only copied into another structure,
so make it const.
The opportunity for this change was found using Coccinelle.
Signed-off-by: Julia Lawall <Julia.Lawall@inria.fr>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The xprtrdma connect logic can return -EPROTO if the underlying
device or network path does not support RDMA. This can happen
after a device removal/insertion.
- When SOFTCONN is set, EPROTO is a permanent error.
- When SOFTCONN is not set, EPROTO is treated as a temporary error.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Using signed 32-bit types for UTC time leads to the y2038 overflow,
which is what happens in the sunrpc code at the moment.
This changes the sunrpc code over to use time64_t where possible.
The one exception is the gss_import_v{1,2}_context() function for
kerberos5, which uses 32-bit timestamps in the protocol. Here,
we can at least treat the numbers as 'unsigned', which extends the
range from 2038 to 2106.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Since v5.4, a device removal occasionally triggered this oops:
Dec 2 17:13:53 manet kernel: BUG: unable to handle page fault for address: 0000000c00000219
Dec 2 17:13:53 manet kernel: #PF: supervisor read access in kernel mode
Dec 2 17:13:53 manet kernel: #PF: error_code(0x0000) - not-present page
Dec 2 17:13:53 manet kernel: PGD 0 P4D 0
Dec 2 17:13:53 manet kernel: Oops: 0000 [#1] SMP
Dec 2 17:13:53 manet kernel: CPU: 2 PID: 468 Comm: kworker/2:1H Tainted: G W 5.4.0-00050-g53717e43af61 #883
Dec 2 17:13:53 manet kernel: Hardware name: Supermicro SYS-6028R-T/X10DRi, BIOS 1.1a 10/16/2015
Dec 2 17:13:53 manet kernel: Workqueue: ib-comp-wq ib_cq_poll_work [ib_core]
Dec 2 17:13:53 manet kernel: RIP: 0010:rpcrdma_wc_receive+0x7c/0xf6 [rpcrdma]
Dec 2 17:13:53 manet kernel: Code: 6d 8b 43 14 89 c1 89 45 78 48 89 4d 40 8b 43 2c 89 45 14 8b 43 20 89 45 18 48 8b 45 20 8b 53 14 48 8b 30 48 8b 40 10 48 8b 38 <48> 8b 87 18 02 00 00 48 85 c0 75 18 48 8b 05 1e 24 c4 e1 48 85 c0
Dec 2 17:13:53 manet kernel: RSP: 0018:ffffc900035dfe00 EFLAGS: 00010246
Dec 2 17:13:53 manet kernel: RAX: ffff888467290000 RBX: ffff88846c638400 RCX: 0000000000000048
Dec 2 17:13:53 manet kernel: RDX: 0000000000000048 RSI: 00000000f942e000 RDI: 0000000c00000001
Dec 2 17:13:53 manet kernel: RBP: ffff888467611b00 R08: ffff888464e4a3c4 R09: 0000000000000000
Dec 2 17:13:53 manet kernel: R10: ffffc900035dfc88 R11: fefefefefefefeff R12: ffff888865af4428
Dec 2 17:13:53 manet kernel: R13: ffff888466023000 R14: ffff88846c63f000 R15: 0000000000000010
Dec 2 17:13:53 manet kernel: FS: 0000000000000000(0000) GS:ffff88846fa80000(0000) knlGS:0000000000000000
Dec 2 17:13:53 manet kernel: CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
Dec 2 17:13:53 manet kernel: CR2: 0000000c00000219 CR3: 0000000002009002 CR4: 00000000001606e0
Dec 2 17:13:53 manet kernel: Call Trace:
Dec 2 17:13:53 manet kernel: __ib_process_cq+0x5c/0x14e [ib_core]
Dec 2 17:13:53 manet kernel: ib_cq_poll_work+0x26/0x70 [ib_core]
Dec 2 17:13:53 manet kernel: process_one_work+0x19d/0x2cd
Dec 2 17:13:53 manet kernel: ? cancel_delayed_work_sync+0xf/0xf
Dec 2 17:13:53 manet kernel: worker_thread+0x1a6/0x25a
Dec 2 17:13:53 manet kernel: ? cancel_delayed_work_sync+0xf/0xf
Dec 2 17:13:53 manet kernel: kthread+0xf4/0xf9
Dec 2 17:13:53 manet kernel: ? kthread_queue_delayed_work+0x74/0x74
Dec 2 17:13:53 manet kernel: ret_from_fork+0x24/0x30
The proximal cause is that this rpcrdma_rep has a rr_rdmabuf that
is still pointing to the old ib_device, which has been freed. The
only way that is possible is if this rpcrdma_rep was not destroyed
by rpcrdma_ia_remove.
Debugging showed that was indeed the case: this rpcrdma_rep was
still in use by a completing RPC at the time of the device removal,
and thus wasn't on the rep free list. So, it was not found by
rpcrdma_reps_destroy().
The fix is to introduce a list of all rpcrdma_reps so that they all
can be found when a device is removed. That list is used to perform
only regbuf DMA unmapping, replacing that call to
rpcrdma_reps_destroy().
Meanwhile, to prevent corruption of this list, I've moved the
destruction of temp rpcrdma_rep objects to rpcrdma_post_recvs().
rpcrdma_xprt_drain() ensures that post_recvs (and thus rep_destroy) is
not invoked while rpcrdma_reps_unmap is walking rb_all_reps, thus
protecting the rb_all_reps list.
Fixes: b0b227f071 ("xprtrdma: Use an llist to manage free rpcrdma_reps")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
I've found that on occasion, "rmmod <dev>" will hang while if an NFS
is under load.
Ensure that ri_remove_done is initialized only just before the
transport is woken up to force a close. This avoids the completion
possibly getting initialized again while the CM event handler is
waiting for a wake-up.
Fixes: bebd031866 ("xprtrdma: Support unplugging an HCA from under an NFS mount")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The timestamps for the cache are all in boottime seconds, so they
don't overflow 32-bit values, but the use of time_t is deprecated
because it generally does overflow when used with wall-clock time.
There are multiple possible ways of avoiding it:
- leave time_t, which is safe here, but forces others to
look into this code to determine that it is over and over.
- use a more generic type, like 'int' or 'long', which is known
to be sufficient here but loses the documentation of referring
to timestamps
- use ktime_t everywhere, and convert into seconds in the few
places where we want realtime-seconds. The conversion is
sometimes expensive, but not more so than the conversion we
do today.
- use time64_t to clarify that this code is safe. Nothing would
change for 64-bit architectures, but it is slightly less
efficient on 32-bit architectures.
Without a clear winner of the three approaches above, this picks
the last one, favouring readability over a small performance
loss on 32-bit architectures.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Using signed 32-bit types for UTC time leads to the y2038 overflow,
which is what happens in the sunrpc code at the moment.
This changes the sunrpc code over to use time64_t where possible.
The one exception is the gss_import_v{1,2}_context() function for
kerberos5, which uses 32-bit timestamps in the protocol. Here,
we can at least treat the numbers as 'unsigned', which extends the
range from 2038 to 2106.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Pull nfsd updates from Bruce Fields:
"This is a relatively quiet cycle for nfsd, mainly various bugfixes.
Possibly most interesting is Trond's fixes for some callback races
that were due to my incomplete understanding of rpc client shutdown.
Unfortunately at the last minute I've started noticing a new
intermittent failure to send callbacks. As the logic seems basically
correct, I'm leaving Trond's patches in for now, and hope to find a
fix in the next week so I don't have to revert those patches"
* tag 'nfsd-5.5' of git://linux-nfs.org/~bfields/linux: (24 commits)
nfsd: depend on CRYPTO_MD5 for legacy client tracking
NFSD fixing possible null pointer derefering in copy offload
nfsd: check for EBUSY from vfs_rmdir/vfs_unink.
nfsd: Ensure CLONE persists data and metadata changes to the target file
SUNRPC: Fix backchannel latency metrics
nfsd: restore NFSv3 ACL support
nfsd: v4 support requires CRYPTO_SHA256
nfsd: Fix cld_net->cn_tfm initialization
lockd: remove __KERNEL__ ifdefs
sunrpc: remove __KERNEL__ ifdefs
race in exportfs_decode_fh()
nfsd: Drop LIST_HEAD where the variable it declares is never used.
nfsd: document callback_wq serialization of callback code
nfsd: mark cb path down on unknown errors
nfsd: Fix races between nfsd4_cb_release() and nfsd4_shutdown_callback()
nfsd: minor 4.1 callback cleanup
SUNRPC: Fix svcauth_gss_proxy_init()
SUNRPC: Trace gssproxy upcall results
sunrpc: fix crash when cache_head become valid before update
nfsd: remove private bin2hex implementation
...
Pull NFS client updates from Trond Myklebust:
"Highlights include:
Features:
- NFSv4.2 now supports cross device offloaded copy (i.e. offloaded
copy of a file from one source server to a different target
server).
- New RDMA tracepoints for debugging congestion control and Local
Invalidate WRs.
Bugfixes and cleanups
- Drop the NFSv4.1 session slot if nfs4_delegreturn_prepare waits for
layoutreturn
- Handle bad/dead sessions correctly in nfs41_sequence_process()
- Various bugfixes to the delegation return operation.
- Various bugfixes pertaining to delegations that have been revoked.
- Cleanups to the NFS timespec code to avoid unnecessary conversions
between timespec and timespec64.
- Fix unstable RDMA connections after a reconnect
- Close race between waking an RDMA sender and posting a receive
- Wake pending RDMA tasks if connection fails
- Fix MR list corruption, and clean up MR usage
- Fix another RPCSEC_GSS issue with MIC buffer space"
* tag 'nfs-for-5.5-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs: (79 commits)
SUNRPC: Capture completion of all RPC tasks
SUNRPC: Fix another issue with MIC buffer space
NFS4: Trace lock reclaims
NFS4: Trace state recovery operation
NFSv4.2 fix memory leak in nfs42_ssc_open
NFSv4.2 fix kfree in __nfs42_copy_file_range
NFS: remove duplicated include from nfs4file.c
NFSv4: Make _nfs42_proc_copy_notify() static
NFS: Fallocate should use the nfs4_fattr_bitmap
NFS: Return -ETXTBSY when attempting to write to a swapfile
fs: nfs: sysfs: Remove NULL check before kfree
NFS: remove unneeded semicolon
NFSv4: add declaration of current_stateid
NFSv4.x: Drop the slot if nfs4_delegreturn_prepare waits for layoutreturn
NFSv4.x: Handle bad/dead sessions correctly in nfs41_sequence_process()
nfsv4: Move NFSPROC4_CLNT_COPY_NOTIFY to end of list
SUNRPC: Avoid RPC delays when exiting suspend
NFS: Add a tracepoint in nfs_fh_to_dentry()
NFSv4: Don't retry the GETATTR on old stateid in nfs4_delegreturn_done()
NFSv4: Handle NFS4ERR_OLD_STATEID in delegreturn
...
RPC tasks on the backchannel never invoke xprt_complete_rqst(), so
there is no way to report their tk_status at completion. Also, any
RPC task that exits via rpc_exit_task() before it is replied to will
also disappear without a trace.
Introduce a trace point that is symmetrical with rpc_task_begin that
captures the termination status of each RPC task.
Sample trace output for callback requests initiated on the server:
kworker/u8:12-448 [003] 127.025240: rpc_task_end: task:50@3 flags=ASYNC|DYNAMIC|SOFT|SOFTCONN|SENT runstate=RUNNING|ACTIVE status=0 action=rpc_exit_task
kworker/u8:12-448 [002] 127.567310: rpc_task_end: task:51@3 flags=ASYNC|DYNAMIC|SOFT|SOFTCONN|SENT runstate=RUNNING|ACTIVE status=0 action=rpc_exit_task
kworker/u8:12-448 [001] 130.506817: rpc_task_end: task:52@3 flags=ASYNC|DYNAMIC|SOFT|SOFTCONN|SENT runstate=RUNNING|ACTIVE status=0 action=rpc_exit_task
Odd, though, that I never see trace_rpc_task_complete, either in the
forward or backchannel. Should it be removed?
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
I noticed that for callback requests, the reported backlog latency
is always zero, and the rtt value is crazy big. The problem was that
rqst->rq_xtime is never set for backchannel requests.
Fixes: 78215759e2 ("SUNRPC: Make RTT measurement more ... ")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
xdr_shrink_pagelen() BUG's when @len is larger than buf->page_len.
This can happen when xdr_buf_read_mic() is given an xdr_buf with
a small page array (like, only a few bytes).
Instead, just cap the number of bytes that xdr_shrink_pagelen()
will move.
Fixes: 5f1bc39979 ("SUNRPC: Fix buffer handling of GSS MIC ... ")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Benjamin Coddington <bcodding@redhat.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
NFSoRDMA Client Updates for Linux 5.5
New Features:
- New tracepoints for congestion control and Local Invalidate WRs
Bugfixes and Cleanups:
- Eliminate log noise in call_reserveresult
- Fix unstable connections after a reconnect
- Clean up some code duplication
- Close race between waking a sender and posting a receive
- Fix MR list corruption, and clean up MR usage
- Remove unused rpcrdma_sendctx fields
- Try to avoid DMA mapping pages if it is too costly
- Wake pending tasks if connection fails
- Replace some dprintk()s with tracepoints
Jon Hunter: "I have been tracking down another suspend/NFS related
issue where again I am seeing random delays exiting suspend. The delays
can be up to a couple minutes in the worst case and this is causing a
suspend test we have to fail."
Change the use of a deferrable work to a standard delayed one.
Reported-by: Jon Hunter <jonathanh@nvidia.com>
Tested-by: Jon Hunter <jonathanh@nvidia.com>
Fixes: 7e0a0e38fc ("SUNRPC: Replace the queue timer with a delayed work function")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
NFSv2, v3 and NFSv4 servers often have duplicate replay caches that look
at the source port when deciding whether or not an RPC call is a replay
of a previous call. This requires clients to perform strange TCP gymnastics
in order to ensure that when they reconnect to the server, they bind
to the same source port.
NFSv4.1 and NFSv4.2 have sessions that provide proper replay semantics,
that do not look at the source port of the connection. This patch therefore
ensures they can ignore the rebind requirement.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
gss_read_proxy_verf() assumes things about the XDR buffer containing
the RPC Call that are not true for buffers generated by
svc_rdma_recv().
RDMA's buffers look more like what the upper layer generates for
sending: head is a kmalloc'd buffer; it does not point to a page
whose contents are contiguous with the first page in the buffers'
page array. The result is that ACCEPT_SEC_CONTEXT via RPC/RDMA has
stopped working on Linux NFS servers that use gssproxy.
This does not affect clients that use only TCP to send their
ACCEPT_SEC_CONTEXT operation (that's all Linux clients). Other
clients, like Solaris NFS clients, send ACCEPT_SEC_CONTEXT on the
same transport as they send all other NFS operations. Such clients
can send ACCEPT_SEC_CONTEXT via RPC/RDMA.
I thought I had found every direct reference in the server RPC code
to the rqstp->rq_pages field.
Bug found at the 2019 Westford NFS bake-a-thon.
Fixes: 3316f06311 ("svcrdma: Persistently allocate and DMA- ... ")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Tested-by: Bill Baker <bill.baker@oracle.com>
Reviewed-by: Simo Sorce <simo@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Record results of a GSS proxy ACCEPT_SEC_CONTEXT upcall and the
svc_authenticate() function to make field debugging of NFS server
Kerberos issues easier.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Bill Baker <bill.baker@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
When we're destroying the host transport mechanism, we should ensure
that we do not leak memory by failing to release any back channel
slots that might still exist.
Reported-by: Neil Brown <neilb@suse.de>
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If there are RDMA back channel requests being processed by the
server threads, then we should hold a reference to the transport
to ensure it doesn't get freed from underneath us.
Reported-by: Neil Brown <neilb@suse.de>
Fixes: 63cae47005 ("xprtrdma: Handle incoming backward direction RPC calls")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If there are TCP back channel requests being processed by the
server threads, then we should hold a reference to the transport
to ensure it doesn't get freed from underneath us.
Reported-by: Neil Brown <neilb@suse.de>
Fixes: 2ea24497a1 ("SUNRPC: RPC callbacks may be split across several..")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up: Use a single trace point to record each connection's
negotiated inline thresholds and the computed maximum byte size
of transport headers.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
For debugging, the op_connect trace point should report the computed
connect delay. We can then ensure that the delay is computed at the
proper times, for example.
As a further clean-up, remove a few low-value "heartbeat" trace
points in the connect path.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Pending tasks are currently never awoken when the connect worker
fails. The reason is that XPRT_CONNECTED is always clear after a
failure return of rpcrdma_ep_connect, thus the
xprt_test_and_clear_connected() check in xprt_rdma_connect_worker()
always fails.
- xprt_rdma_close always clears XPRT_CONNECTED.
- rpcrdma_ep_connect always clears XPRT_CONNECTED.
After reviewing the TCP connect worker, it appears that there's no
need for extra test_and_set paranoia in xprt_rdma_connect_worker.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
On some platforms, DMA mapping part of a page is more costly than
copying bytes. Restore the pull-up code and use that when we
think it's going to be faster. The heuristic for now is to pull-up
when the size of the RPC message body fits in the buffer underlying
the head iovec.
Indeed, not involving the I/O MMU can help the RPC/RDMA transport
scale better for tiny I/Os across more RDMA devices. This is because
interaction with the I/O MMU is eliminated, as is handling a Send
completion, for each of these small I/Os. Without the explicit
unmapping, the NIC no longer needs to do a costly internal TLB shoot
down for buffers that are just a handful of bytes.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Refactor: Replace spaghetti with code that makes it plain what needs
to be done for each rtype. This makes it easier to add features and
optimizations.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up: This field is not needed in the Send completion handler,
so it can be moved to struct rpcrdma_req to reduce the size of
struct rpcrdma_sendctx, and to reduce the amount of memory that
is sloshed between the sending process and the Send completion
process.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
ia->ri_id is replaced during a reconnect. The connect_worker runs
with the transport send lock held to prevent ri_id from being
dereferenced by the send_request path during this process.
Currently, however, there is no guarantee that ia->ri_id is stable
in the MR recycling worker, which operates in the background and is
not serialized with the connect_worker in any way.
But now that Local_Inv completions are being done in process
context, we can handle the recycling operation there instead of
deferring the recycling work to another process. Because the
disconnect path drains all work before allowing tear down to
proceed, it is guaranteed that Local Invalidations complete only
while the ri_id pointer is stable.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
MRs are now allocated on demand so we can safely throw them away on
disconnect. This way an idle transport can disconnect and it won't
pin hardware MR resources.
Two additional changes:
- Now that all MRs are destroyed on disconnect, there's no need to
check during header marshaling if a req has MRs to recycle. Each
req is sent only once per connection, and now rl_registered is
guaranteed to be empty when rpcrdma_marshal_req is invoked.
- Because MRs are now destroyed in a WQ_MEM_RECLAIM context, they
also must be allocated in a WQ_MEM_RECLAIM context. This reduces
the likelihood that device driver memory allocation will trigger
memory reclaim during NFS writeback.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Close some holes introduced by commit 6dc6ec9e04 ("xprtrdma: Cache
free MRs in each rpcrdma_req") that could result in list corruption.
In addition, the result that is tabulated in @count is no longer
used, so @count is removed.
Fixes: 6dc6ec9e04 ("xprtrdma: Cache free MRs in each rpcrdma_req")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
A recent clean up attempted to separate Receive handling and RPC
Reply processing, in the name of clean layering.
Unfortunately, we can't do this because the Receive Queue has to be
refilled _after_ the most recent credit update from the responder
is parsed from the transport header, but _before_ we wake up the
next RPC sender. That is right in the middle of
rpcrdma_reply_handler().
Usually this isn't a problem because current responder
implementations don't vary their credit grant. The one exception is
when a connection is established: the grant goes from one to a much
larger number on the first Receive. The requester MUST post enough
Receives right then so that any outstanding requests can be sent
without risking RNR and connection loss.
Fixes: 6ceea36890 ("xprtrdma: Refactor Receive accounting")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up/code de-duplication.
Nit: RPC_CWNDSHIFT is incorrect as the initial value for xprt->cwnd.
This mistake does not appear to have operational consequences, since
the cwnd value is replaced with a valid value upon the first Receive
completion.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
This is because xprt_request_get_cong() is allowing more than one
RPC Call to be transmitted before the first Receive on the new
connection. The first Receive fills the Receive Queue based on the
server's credit grant. Before that Receive, there is only a single
Receive WR posted because the client doesn't know the server's
credit grant.
Solution is to clear rq_cong on all outstanding rpc_rqsts when the
the cwnd is reset. This is because an RPC/RDMA credit is good for
one connection instance only.
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>
When adding frwr_unmap_async way back when, I re-used the existing
trace_xprtrdma_post_send() trace point to record the return code
of ib_post_send.
Unfortunately there are some cases where re-using that trace point
causes a crash. Instead, construct a trace point specific to posting
Local Invalidate WRs that will always be safe to use in that context,
and will act as a trace log eye-catcher for Local Invalidation.
Fixes: 847568942f ("xprtrdma: Remove fr_state")
Fixes: d8099feda4 ("xprtrdma: Reduce context switching due ... ")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Tested-by: Bill Baker <bill.baker@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
To help debug problems with RPC/RDMA credit management, replace
dprintk() call sites in the transport send lock paths with trace
events.
Similar trace points are defined for the non-congestion paths.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Sep 11 16:35:20 manet kernel:
call_reserveresult: unrecognized error -512, exiting
Diagnostic error messages such as this likely have no value for NFS
client administrators.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
I was investigating a crash in our Virtuozzo7 kernel which happened in
in svcauth_unix_set_client. I found out that we access m_client field
in ip_map structure, which was received from sunrpc_cache_lookup (we
have a bit older kernel, now the code is in sunrpc_cache_add_entry), and
these field looks uninitialized (m_client == 0x74 don't look like a
pointer) but in the cache_head in flags we see 0x1 which is CACHE_VALID.
It looks like the problem appeared from our previous fix to sunrpc (1):
commit 4ecd55ea07 ("sunrpc: fix cache_head leak due to queued
request")
And we've also found a patch already fixing our patch (2):
commit d58431eacb ("sunrpc: don't mark uninitialised items as VALID.")
Though the crash is eliminated, I think the core of the problem is not
completely fixed:
Neil in the patch (2) makes cache_head CACHE_NEGATIVE, before
cache_fresh_locked which was added in (1) to fix crash. These way
cache_is_valid won't say the cache is valid anymore and in
svcauth_unix_set_client the function cache_check will return error
instead of 0, and we don't count entry as initialized.
But it looks like we need to remove cache_fresh_locked completely in
sunrpc_cache_lookup:
In (1) we've only wanted to make cache_fresh_unlocked->cache_dequeue so
that cache_requests with no readers also release corresponding
cache_head, to fix their leak. We with Vasily were not sure if
cache_fresh_locked and cache_fresh_unlocked should be used in pair or
not, so we've guessed to use them in pair.
Now we see that we don't want the CACHE_VALID bit set here by
cache_fresh_locked, as "valid" means "initialized" and there is no
initialization in sunrpc_cache_add_entry. Both expiry_time and
last_refresh are not used in cache_fresh_unlocked code-path and also not
required for the initial fix.
So to conclude cache_fresh_locked was called by mistake, and we can just
safely remove it instead of crutching it with CACHE_NEGATIVE. It looks
ideologically better for me. Hope I don't miss something here.
Here is our crash backtrace:
[13108726.326291] BUG: unable to handle kernel NULL pointer dereference at 0000000000000074
[13108726.326365] IP: [<ffffffffc01f79eb>] svcauth_unix_set_client+0x2ab/0x520 [sunrpc]
[13108726.326448] PGD 0
[13108726.326468] Oops: 0002 [#1] SMP
[13108726.326497] Modules linked in: nbd isofs xfs loop kpatch_cumulative_81_0_r1(O) xt_physdev nfnetlink_queue bluetooth rfkill ip6table_nat nf_nat_ipv6 ip_vs_wrr ip_vs_wlc ip_vs_sh nf_conntrack_netlink ip_vs_sed ip_vs_pe_sip nf_conntrack_sip ip_vs_nq ip_vs_lc ip_vs_lblcr ip_vs_lblc ip_vs_ftp ip_vs_dh nf_nat_ftp nf_conntrack_ftp iptable_raw xt_recent nf_log_ipv6 xt_hl ip6t_rt nf_log_ipv4 nf_log_common xt_LOG xt_limit xt_TCPMSS xt_tcpmss vxlan ip6_udp_tunnel udp_tunnel xt_statistic xt_NFLOG nfnetlink_log dummy xt_mark xt_REDIRECT nf_nat_redirect raw_diag udp_diag tcp_diag inet_diag netlink_diag af_packet_diag unix_diag rpcsec_gss_krb5 xt_addrtype ip6t_rpfilter ipt_REJECT nf_reject_ipv4 ip6t_REJECT nf_reject_ipv6 ebtable_nat ebtable_broute nf_conntrack_ipv6 nf_defrag_ipv6 ip6table_mangle ip6table_raw nfsv4
[13108726.327173] dns_resolver cls_u32 binfmt_misc arptable_filter arp_tables ip6table_filter ip6_tables devlink fuse_kio_pcs ipt_MASQUERADE nf_nat_masquerade_ipv4 xt_nat iptable_nat nf_nat_ipv4 xt_comment nf_conntrack_ipv4 nf_defrag_ipv4 xt_wdog_tmo xt_multiport bonding xt_set xt_conntrack iptable_filter iptable_mangle kpatch(O) ebtable_filter ebt_among ebtables ip_set_hash_ip ip_set nfnetlink vfat fat skx_edac intel_powerclamp coretemp intel_rapl iosf_mbi kvm_intel kvm irqbypass fuse pcspkr ses enclosure joydev sg mei_me hpwdt hpilo lpc_ich mei ipmi_si shpchp ipmi_devintf ipmi_msghandler xt_ipvs acpi_power_meter ip_vs_rr nfsv3 nfsd auth_rpcgss nfs_acl nfs lockd grace fscache nf_nat cls_fw sch_htb sch_cbq sch_sfq ip_vs em_u32 nf_conntrack tun br_netfilter veth overlay ip6_vzprivnet ip6_vznetstat ip_vznetstat
[13108726.327817] ip_vzprivnet vziolimit vzevent vzlist vzstat vznetstat vznetdev vzmon vzdev bridge pio_kaio pio_nfs pio_direct pfmt_raw pfmt_ploop1 ploop ip_tables ext4 mbcache jbd2 sd_mod crc_t10dif crct10dif_generic mgag200 i2c_algo_bit drm_kms_helper scsi_transport_iscsi 8021q syscopyarea sysfillrect garp sysimgblt fb_sys_fops mrp stp ttm llc bnx2x crct10dif_pclmul crct10dif_common crc32_pclmul crc32c_intel drm dm_multipath ghash_clmulni_intel uas aesni_intel lrw gf128mul glue_helper ablk_helper cryptd tg3 smartpqi scsi_transport_sas mdio libcrc32c i2c_core usb_storage ptp pps_core wmi sunrpc dm_mirror dm_region_hash dm_log dm_mod [last unloaded: kpatch_cumulative_82_0_r1]
[13108726.328403] CPU: 35 PID: 63742 Comm: nfsd ve: 51332 Kdump: loaded Tainted: G W O ------------ 3.10.0-862.20.2.vz7.73.29 #1 73.29
[13108726.328491] Hardware name: HPE ProLiant DL360 Gen10/ProLiant DL360 Gen10, BIOS U32 10/02/2018
[13108726.328554] task: ffffa0a6a41b1160 ti: ffffa0c2a74bc000 task.ti: ffffa0c2a74bc000
[13108726.328610] RIP: 0010:[<ffffffffc01f79eb>] [<ffffffffc01f79eb>] svcauth_unix_set_client+0x2ab/0x520 [sunrpc]
[13108726.328706] RSP: 0018:ffffa0c2a74bfd80 EFLAGS: 00010246
[13108726.328750] RAX: 0000000000000001 RBX: ffffa0a6183ae000 RCX: 0000000000000000
[13108726.328811] RDX: 0000000000000074 RSI: 0000000000000286 RDI: ffffa0c2a74bfcf0
[13108726.328864] RBP: ffffa0c2a74bfe00 R08: ffffa0bab8c22960 R09: 0000000000000001
[13108726.328916] R10: 0000000000000001 R11: 0000000000000001 R12: ffffa0a32aa7f000
[13108726.328969] R13: ffffa0a6183afac0 R14: ffffa0c233d88d00 R15: ffffa0c2a74bfdb4
[13108726.329022] FS: 0000000000000000(0000) GS:ffffa0e17f9c0000(0000) knlGS:0000000000000000
[13108726.329081] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[13108726.332311] CR2: 0000000000000074 CR3: 00000026a1b28000 CR4: 00000000007607e0
[13108726.334606] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[13108726.336754] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[13108726.338908] PKRU: 00000000
[13108726.341047] Call Trace:
[13108726.343074] [<ffffffff8a2c78b4>] ? groups_alloc+0x34/0x110
[13108726.344837] [<ffffffffc01f5eb4>] svc_set_client+0x24/0x30 [sunrpc]
[13108726.346631] [<ffffffffc01f2ac1>] svc_process_common+0x241/0x710 [sunrpc]
[13108726.348332] [<ffffffffc01f3093>] svc_process+0x103/0x190 [sunrpc]
[13108726.350016] [<ffffffffc07d605f>] nfsd+0xdf/0x150 [nfsd]
[13108726.351735] [<ffffffffc07d5f80>] ? nfsd_destroy+0x80/0x80 [nfsd]
[13108726.353459] [<ffffffff8a2bf741>] kthread+0xd1/0xe0
[13108726.355195] [<ffffffff8a2bf670>] ? create_kthread+0x60/0x60
[13108726.356896] [<ffffffff8a9556dd>] ret_from_fork_nospec_begin+0x7/0x21
[13108726.358577] [<ffffffff8a2bf670>] ? create_kthread+0x60/0x60
[13108726.360240] Code: 4c 8b 45 98 0f 8e 2e 01 00 00 83 f8 fe 0f 84 76 fe ff ff 85 c0 0f 85 2b 01 00 00 49 8b 50 40 b8 01 00 00 00 48 89 93 d0 1a 00 00 <f0> 0f c1 02 83 c0 01 83 f8 01 0f 8e 53 02 00 00 49 8b 44 24 38
[13108726.363769] RIP [<ffffffffc01f79eb>] svcauth_unix_set_client+0x2ab/0x520 [sunrpc]
[13108726.365530] RSP <ffffa0c2a74bfd80>
[13108726.367179] CR2: 0000000000000074
Fixes: d58431eacb ("sunrpc: don't mark uninitialised items as VALID.")
Signed-off-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Acked-by: NeilBrown <neilb@suse.de>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Since commit 4f8943f808 ("SUNRPC: Replace direct task wakeups from
softirq context") there has been a race to the value of the sk_err if both
XPRT_SOCK_WAKE_ERROR and XPRT_SOCK_WAKE_DISCONNECT are set. In that case,
we may end up losing the sk_err value that existed when xs_error_report was
called.
Fix this by reverting to the previous behavior: instead of using SO_ERROR
to retrieve the value at a later time (which might also return sk_err_soft),
copy the sk_err value onto struct sock_xprt, and use that value to wake
pending tasks.
Signed-off-by: Benjamin Coddington <bcodding@redhat.com>
Fixes: 4f8943f808 ("SUNRPC: Replace direct task wakeups from softirq context")
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
