RPCRDMA requests that specify a read-list are fetched with RDMA_READ. Using
an FRMR to map the data sink improves NFSRDMA security on transports that
place the RDMA_READ data sink LKEY on the wire because the valid lifetime
of the MR is only the duration of the RDMA_READ. The LKEY is invalidated
when the last RDMA_READ WR completes.
Mapping the data sink also allows for very large amounts to data to be
fetched with a single WR, so if the client is also using FRMR, the entire
RPC read-list can be fetched with a single WR.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
WR can be submitted as linked lists of WR. Update the svc_rdma_send
routine to handle WR chains. This will be used to submit a WR that
uses an FRMR with another WR that invalidates the FRMR.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Update the svc_rdma_post_recv routine to use the adapter's global LKEY
instead of sc_phys_mr which is only valid when using a DMA MR.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Fast Reg MR introduces a new WR type. Add a service to register the
region with the adapter and update the completion handling to support
completions with a NULL WR context.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Query the device capabilities in the svc_rdma_accept function to determine
what advanced memory management capabilities are supported by the device.
Based on the query, select the most secure model available given the
requirements of the transport and capabilities of the adapter.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Add services for the allocating, freeing, and unmapping Fast Reg MR. These
services will be used by the transport connection setup, send and receive
routines.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Vegard Nossum reported
----------------------
> I noticed that something weird is going on with /proc/sys/sunrpc/transports.
> This file is generated in net/sunrpc/sysctl.c, function proc_do_xprt(). When
> I "cat" this file, I get the expected output:
> $ cat /proc/sys/sunrpc/transports
> tcp 1048576
> udp 32768
> But I think that it does not check the length of the buffer supplied by
> userspace to read(). With my original program, I found that the stack was
> being overwritten by the characters above, even when the length given to
> read() was just 1.
David Wagner added (among other things) that copy_to_user could be
probably used here.
Ingo Oeser suggested to use simple_read_from_buffer() here.
The conclusion is that proc_do_xprt doesn't check for userside buffer
size indeed so fix this by using Ingo's suggestion.
Reported-by: Vegard Nossum <vegard.nossum@gmail.com>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
CC: Ingo Oeser <ioe-lkml@rameria.de>
Cc: Neil Brown <neilb@suse.de>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Greg Banks <gnb@sgi.com>
Cc: Tom Tucker <tom@opengridcomputing.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
RDMA_READ completions are kept on a separate queue from the general
I/O request queue. Since a separate lock is used to protect the RDMA_READ
completion queue, a race exists between the dto_tasklet and the
svc_rdma_recvfrom thread where the dto_tasklet sets the XPT_DATA
bit and adds I/O to the read-completion queue. Concurrently, the
recvfrom thread checks the generic queue, finds it empty and resets
the XPT_DATA bit. A subsequent svc_xprt_enqueue will fail to enqueue
the transport for I/O and cause the transport to "stall".
The fix is to protect both lists with the same lock and set the XPT_DATA
bit with this lock held.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Kmem cache passed to constructor is only needed for constructors that are
themselves multiplexeres. Nobody uses this "feature", nor does anybody uses
passed kmem cache in non-trivial way, so pass only pointer to object.
Non-trivial places are:
arch/powerpc/mm/init_64.c
arch/powerpc/mm/hugetlbpage.c
This is flag day, yes.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Jon Tollefson <kniht@linux.vnet.ibm.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Matt Mackall <mpm@selenic.com>
[akpm@linux-foundation.org: fix arch/powerpc/mm/hugetlbpage.c]
[akpm@linux-foundation.org: fix mm/slab.c]
[akpm@linux-foundation.org: fix ubifs]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add per-device dma_mapping_ops support for CONFIG_X86_64 as POWER
architecture does:
This enables us to cleanly fix the Calgary IOMMU issue that some devices
are not behind the IOMMU (http://lkml.org/lkml/2008/5/8/423).
I think that per-device dma_mapping_ops support would be also helpful for
KVM people to support PCI passthrough but Andi thinks that this makes it
difficult to support the PCI passthrough (see the above thread). So I
CC'ed this to KVM camp. Comments are appreciated.
A pointer to dma_mapping_ops to struct dev_archdata is added. If the
pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If it's
NULL, the system-wide dma_ops pointer is used as before.
If it's useful for KVM people, I plan to implement a mechanism to register
a hook called when a new pci (or dma capable) device is created (it works
with hot plugging). It enables IOMMUs to set up an appropriate
dma_mapping_ops per device.
The major obstacle is that dma_mapping_error doesn't take a pointer to the
device unlike other DMA operations. So x86 can't have dma_mapping_ops per
device. Note all the POWER IOMMUs use the same dma_mapping_error function
so this is not a problem for POWER but x86 IOMMUs use different
dma_mapping_error functions.
The first patch adds the device argument to dma_mapping_error. The patch
is trivial but large since it touches lots of drivers and dma-mapping.h in
all the architecture.
This patch:
dma_mapping_error() doesn't take a pointer to the device unlike other DMA
operations. So we can't have dma_mapping_ops per device.
Note that POWER already has dma_mapping_ops per device but all the POWER
IOMMUs use the same dma_mapping_error function. x86 IOMMUs use device
argument.
[akpm@linux-foundation.org: fix sge]
[akpm@linux-foundation.org: fix svc_rdma]
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: fix bnx2x]
[akpm@linux-foundation.org: fix s2io]
[akpm@linux-foundation.org: fix pasemi_mac]
[akpm@linux-foundation.org: fix sdhci]
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: fix sparc]
[akpm@linux-foundation.org: fix ibmvscsi]
Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Muli Ben-Yehuda <muli@il.ibm.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* Replace previous instances of the cpumask_of_cpu_ptr* macros
with a the new (lvalue capable) generic cpumask_of_cpu().
Signed-off-by: Mike Travis <travis@sgi.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jack Steiner <steiner@sgi.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'for-2.6.27' of git://linux-nfs.org/~bfields/linux: (51 commits)
nfsd: nfs4xdr.c do-while is not a compound statement
nfsd: Use C99 initializers in fs/nfsd/nfs4xdr.c
lockd: Pass "struct sockaddr *" to new failover-by-IP function
lockd: get host reference in nlmsvc_create_block() instead of callers
lockd: minor svclock.c style fixes
lockd: eliminate duplicate nlmsvc_lookup_host call from nlmsvc_lock
lockd: eliminate duplicate nlmsvc_lookup_host call from nlmsvc_testlock
lockd: nlm_release_host() checks for NULL, caller needn't
file lock: reorder struct file_lock to save space on 64 bit builds
nfsd: take file and mnt write in nfs4_upgrade_open
nfsd: document open share bit tracking
nfsd: tabulate nfs4 xdr encoding functions
nfsd: dprint operation names
svcrdma: Change WR context get/put to use the kmem cache
svcrdma: Create a kmem cache for the WR contexts
svcrdma: Add flush_scheduled_work to module exit function
svcrdma: Limit ORD based on client's advertised IRD
svcrdma: Remove unused wait q from svcrdma_xprt structure
svcrdma: Remove unneeded spin locks from __svc_rdma_free
svcrdma: Add dma map count and WARN_ON
...
* This patch replaces the dangerous lvalue version of cpumask_of_cpu
with new cpumask_of_cpu_ptr macros. These are patterned after the
node_to_cpumask_ptr macros.
In general terms, if there is a cpumask_of_cpu_map[] then a pointer to
the cpumask_of_cpu_map[cpu] entry is used. The cpumask_of_cpu_map
is provided when there is a large NR_CPUS count, reducing
greatly the amount of code generated and stack space used for
cpumask_of_cpu(). The pointer to the cpumask_t value is needed for
calling set_cpus_allowed_ptr() to reduce the amount of stack space
needed to pass the cpumask_t value.
If there isn't a cpumask_of_cpu_map[], then a temporary variable is
declared and filled in with value from cpumask_of_cpu(cpu) as well as
a pointer variable pointing to this temporary variable. Afterwards,
the pointer is used to reference the cpumask value. The compiler
will optimize out the extra dereference through the pointer as well
as the stack space used for the pointer, resulting in identical code.
A good example of the orthogonal usages is in net/sunrpc/svc.c:
case SVC_POOL_PERCPU:
{
unsigned int cpu = m->pool_to[pidx];
cpumask_of_cpu_ptr(cpumask, cpu);
*oldmask = current->cpus_allowed;
set_cpus_allowed_ptr(current, cpumask);
return 1;
}
case SVC_POOL_PERNODE:
{
unsigned int node = m->pool_to[pidx];
node_to_cpumask_ptr(nodecpumask, node);
*oldmask = current->cpus_allowed;
set_cpus_allowed_ptr(current, nodecpumask);
return 1;
}
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Push it into those callback functions that actually need it.
Note that all the NFS operations use their own locking, so don't need the
BKL. Ditto for the rpcbind client.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Introduce a new API to register RPC services on IPv6 interfaces to allow
the NFS server and lockd to advertise on IPv6 networks.
Unlike rpcb_register(), the new rpcb_v4_register() function uses rpcbind
protocol version 4 to contact the local rpcbind daemon. The version 4
SET/UNSET procedures allow services to register address families besides
AF_INET, register at specific network interfaces, and register transport
protocols besides UDP and TCP. All of this functionality is exposed via
the new rpcb_v4_register() kernel API.
A user-space rpcbind daemon implementation that supports version 4 of the
rpcbind protocol is required in order to make use of this new API.
Note that rpcbind version 3 is sufficient to support the new rpcbind
facilities listed above, but most extant implementations use version 4.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
rpcbind version 4 registration will reuse part of rpcb_register, so just
split it out into a separate function now.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Clean up: Callers that required a privileged source port now use
rpcb_create_local(), so we can remove the @privileged argument from
rpcb_create().
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Add rpcb_create_local() for use by rpcb_register() and upcoming IPv6
registration functions.
Ensure any errors encountered by rpcb_create_local() are properly
reported.
We can also use a statically allocated constant loopback socket address
instead of one allocated on the stack and initialized every time the
function is called.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The rpcbind versions 3 and 4 SET and UNSET procedures use the same
arguments as the GETADDR procedure.
While definitely a bug, this hasn't been a problem so far since the
kernel hasn't used version 3 or 4 SET and UNSET. But this will change
in just a moment.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
If another task is busy in rpcb_getport_async number, it is more efficient
to have it wake us up when it has finished instead of arbitrarily sleeping
for 5 seconds.
Also ensure that rpcb_wake_rpcbind_waiters() is called regardless of
whether or not rpcb_getport_done() gets called.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Some server vendors support the higher versions of rpcbind only for
AF_INET6. The kernel doesn't need to use v3 or v4 for AF_INET anyway,
so change the kernel's rpcbind client to query AF_INET servers over
rpcbind v2 only.
This has a few interesting benefits:
1. If the rpcbind request is going over TCP, and the server doesn't
support rpcbind versions 3 or 4, the client reduces by two the number
of ephemeral ports left in TIME_WAIT for each rpcbind request. This
will help during NFS mount storms.
2. The rpcbind interaction with servers that don't support rpcbind
versions 3 or 4 will use less network traffic. Also helpful
during mount storms.
3. We can eliminate the kernel build option that controls whether the
kernel's rpcbind client uses rpcbind version 3 and 4 for AF_INET
servers. Less complicated kernel configuration...
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Some rpcbind servers that do support rpcbind version 4 do not support
the GETVERSADDR procedure. Use GETADDR for querying rpcbind servers
via rpcbind version 4 instead of GETVERSADDR.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Clean up: Change the version 2 procedure name to GETPORT. It's the same
procedure number as GETADDR, but version 2 implementations usually refer
to it as GETPORT.
This also now matches the procedure name used in the version 2 procedure
entry in the rpcb_next_version[] array, making it slightly less confusing.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The RPC client uses the rq_xtime field in each RPC request to determine the
round-trip time of the request. Currently, the rq_xtime field is
initialized by each transport just before it starts enqueing a request to
be sent. However, transports do not handle initializing this value
consistently; sometimes they don't initialize it at all.
To make the measurement of request round-trip time consistent for all
RPC client transport capabilities, pull rq_xtime initialization into the
RPC client's generic transport logic. Now all transports will get a
standardized RTT measure automatically, from:
xprt_transmit()
to
xprt_complete_rqst()
This makes round-trip time calculation more accurate for the TCP transport.
The socket ->sendmsg() method can return "-EAGAIN" if the socket's output
buffer is full, so the TCP transport's ->send_request() method may call
the ->sendmsg() method repeatedly until it gets all of the request's bytes
queued in the socket's buffer.
Currently, the TCP transport sets the rq_xtime field every time through
that loop so the final value is the timestamp just before the *last* call
to the underlying socket's ->sendmsg() method. After this patch, the
rq_xtime field contains a timestamp that reflects the time just before the
*first* call to ->sendmsg().
This is consequential under heavy workloads because large requests often
take multiple ->sendmsg() calls to get all the bytes of a request queued.
The TCP transport causes the request to sleep until the remote end of the
socket has received enough bytes to clear space in the socket's local
output buffer. This delay can be quite significant.
The method introduced by this patch is a more accurate measure of RTT
for stream transports, since the server can cause enough back pressure
to delay (ie increase the latency of) requests from the client.
Additionally, this patch corrects the behavior of the RDMA transport, which
entirely neglected to initialize the rq_xtime field. RPC performance
metrics for RDMA transports now display correct RPC request round trip
times.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Tom Talpey <thomas.talpey@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Try to make the comment here a little more clear and concise.
Also, this macro definition seems unnecessary.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
There used to be a print_hexl() function that used isprint(), now gone.
I don't know why NFS_NGROUPS and CA_RUN_AS_MACHINE were here.
I also don't know why another #define that's actually used was marked
"unused".
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Also, a minor comment grammar fix in the same file.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The cl_chatty flag alows us to control whether a given rpc client leaves
"server X not responding, timed out"
messages in the syslog. Such messages make sense for ordinary nfs
clients (where an unresponsive server means applications on the
mountpoint are probably hanging), but not for the callback client (which
can fail more commonly, with the only result just of disabling some
optimizations).
Previously cl_chatty was removed, do to lack of users; reinstate it, and
use it for the nfsd's callback client.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Recent changes to the RPC client's transport connect logic make connect
status values ECONNREFUSED and ECONNRESET impossible.
Clean up xprt_connect_status() to account for these changes.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Clean up: move the logic that displays each task to its own function.
This removes indentation and makes future changes easier.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Clean up: don't display the rpc_show_tasks column header unless there is at
least one task to display. As far as I can tell, it is safe to let the
list_for_each_entry macro decide that each list is empty.
scripts/checkpatch.pl also wants a KERN_FOO at the start of any newly added
printk() calls, so this and subsequent patches will also add KERN_INFO.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The RPC client uses a finite state machine to move RPC tasks through each
step of an RPC request. Each state is contained in a function in
net/sunrpc/clnt.c, and named call_foo.
Some of the functions named call_foo have changed over the past few years and
are no longer states in the FSM. These include: call_encode, call_header,
and call_verify. As a clean up, rename the functions that have changed.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Improve debugging messages in call_start() and call_verify() by having
them show the RPC procedure name instead of the procedure number.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Since the credentials may be allocated during the call to rpc_new_task(),
which again may be called by a memory allocator...
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The special 'ENOMEM' case that was previously flagged as non-fatal is
bogus: auth_gss always returns EAGAIN for non-fatal errors, and may in fact
return ENOMEM in the special case where xdr_buf_read_netobj runs out of
preallocated buffer space (invariably a _fatal_ error, since there is no
provision for preallocating larger buffers).
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
All errors from call_encode(), with exception of EAGAIN are fatal, so we
should immediately return instead of proceeding to xprt_transmit().
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Now that rpcb_next_version has been split into an IPv4 version and an IPv6
version, we Oops when rpcb_call_async attempts to look up the IPv6-specific
RPC procedure in rpcb_next_version.
Fix the Oops simply by having rpcb_getport_async pass the correct RPC
procedure as an argument.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
It is wrong to be freeing up the rpcbind arguments if the call to
rpcb_call_async() fails, since they should already have been freed up by
rpcb_map_release().
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
To return garbage_args, the accept_stat must be 0, and we must have a
verifier. So we shouldn't be resetting the write pointer as we reject
the call.
Also, we must add the two placeholder words here regardless of success
of the unwrap, to ensure the output buffer is left in a consistent state
for svcauth_gss_release().
This fixes a BUG() in svcauth_gss.c:svcauth_gss_release().
Thanks to Aime Le Rouzic for bug report, debugging help, and testing.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Tested-by: Aime Le Rouzic <aime.le-rouzic@bull.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Change the WR context pool to be shared across mount points. This
reduces the RDMA transport memory footprint significantly since
idle mounts don't consume WR context memory.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
![](/avatar/5fc124e8ccd7b94d35e5c23f8c0b7c63?size=56 "\\\\"J. Bruce Fields\\\"){kind=small}
