This patch moves the transport sockaddr to the svc_xprt
structure. Convenience functions are added to set and
get the local and remote addresses of a transport from
the transport provider as well as determine the length
of a sockaddr.
A transport is responsible for setting the xpt_local
and xpt_remote addresses in the svc_xprt structure as
part of transport creation and xpo_accept processing. This
cannot be done in a generic way and in fact varies
between TCP, UDP and RDMA. A set of xpo_ functions
(e.g. getlocalname, getremotename) could have been
added but this would have resulted in additional
caching and copying of the addresses around. Note that
the xpt_local address should also be set on listening
endpoints; for TCP/RDMA this is done as part of
endpoint creation.
For connected transports like TCP and RDMA, the addresses
never change and can be set once and copied into the
rqstp structure for each request. For UDP, however, the
local and remote addresses may change for each request. In
this case, the address information is obtained from the
UDP recvmsg info and copied into the rqstp structure from
there.
A svc_xprt_local_port function was also added that returns
the local port given a transport. This is used by
svc_create_xprt when returning the port associated with
a newly created transport, and later when creating a
generic find transport service to check if a service is
already listening on a given port.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
This patch moves the transport independent sk_deferred list to the svc_xprt
structure and updates the svc_deferred_req structure to keep pointers to
svc_xprt's directly. The deferral processing code is also moved out of the
transport dependent recvfrom functions and into the generic svc_recv path.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Move the authinfo cache to svc_xprt. This allows both the TCP and RDMA
transports to share this logic. A flag bit is used to determine if
auth information is to be cached or not. Previously, this code looked
at the transport protocol.
I've also changed the spin_lock/unlock logic so that a lock is not taken for
transports that are not caching auth info.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
With the implementation of the new mark and sweep algorithm for shutting
down old connections, the sk_lastrecv field is no longer needed.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Move the sk_mutex field to the transport independent svc_xprt structure.
Now all the fields that svc_send touches are transport neutral. Change the
svc_send function to use the transport independent svc_xprt directly instead
of the transport dependent svc_sock structure.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
This functionally trivial patch moves the sk_reserved field to the
transport independent svc_xprt structure.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Move sk_list and sk_ready to svc_xprt. This involves close because these
lists are walked by svcs when closing all their transports. So I combined
the moving of these lists to svc_xprt with making close transport independent.
The svc_force_sock_close has been changed to svc_close_all and takes a list
as an argument. This removes some svc internals knowledge from the svcs.
This code races with module removal and transport addition.
Thanks to Simon Holm Thøgersen for a compile fix.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Cc: Simon Holm Thøgersen <odie@cs.aau.dk>
This is another incremental change that moves transport independent
fields from svc_sock to the svc_xprt structure. The changes
should be functionally null.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
This functionally trivial change moves the transport independent sk_flags
field to the transport independent svc_xprt structure.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Change the atomic_t reference count to a kref and move it to the
transport indepenent svc_xprt structure. Change the reference count
wrapper names to be generic.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Modify the various kernel RPC svcs to use the svc_create_xprt service.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Previously, the accept logic looked into the socket state to determine
whether to call accept or recv when data-ready was indicated on an endpoint.
Since some transports don't use sockets, this logic now uses a flag
bit (SK_LISTENER) to identify listening endpoints. A transport function
(xpo_accept) allows each transport to define its own accept processing.
A transport's initialization logic is reponsible for setting the
SK_LISTENER bit. I didn't see any way to do this in transport independent
logic since the passive side of a UDP connection doesn't listen and
always recv's.
In the svc_recv function, if the SK_LISTENER bit is set, the transport
xpo_accept function is called to handle accept processing.
Note that all functions are defined even if they don't make sense
for a given transport. For example, accept doesn't mean anything for
UDP. The function is defined anyway and bug checks if called. The
UDP transport should never set the SK_LISTENER bit.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
The sk_sendto and sk_recvfrom are function pointers that allow svc_sock
to be used for both UDP and TCP. Move these function pointers to the
svc_xprt_ops structure.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Make TCP and UDP svc_sock transports, and register them
with the svc transport core.
A transport type (svc_sock) has an svc_xprt as its first member,
and calls svc_xprt_init to initialize this field.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Save the destination address of an incoming request over TCP like is
done already for UDP. It is necessary later for callbacks by the server.
Signed-off-by: Frank van Maarseveen <frankvm@frankvm.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Now that sk_defer_lock protects two different things, make the name more
generic.
Also don't bother with disabling _bh as the lock is only ever taken from
process context.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When the last thread of nfsd exits, it shuts down all related sockets. It
currently uses svc_close_socket to do this, but that only is immediately
effective if the socket is not SK_BUSY.
If the socket is busy - i.e. if a request has arrived that has not yet been
processes - svc_close_socket is not effective and the shutdown process spins.
So create a new svc_force_close_socket which removes the SK_BUSY flag is set
and then calls svc_close_socket.
Also change some open-codes loops in svc_destroy to use
list_for_each_entry_safe.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The remote peer's address won't change after the socket has been accepted. We
don't need to call ->getname on every incoming request.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Cc: Aurelien Charbon <aurelien.charbon@ext.bull.net>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Sometimes we need to create an RPC service but not register it with the local
portmapper. NFSv4 delegation callback, for example.
Change the svc_makesock() API to allow optionally creating temporary or
permanent sockets, optionally registering with the local portmapper, and make
it return the ephemeral port of the new socket.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Cc: Aurelien Charbon <aurelien.charbon@ext.bull.net>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently in the RPC server, registering with the local portmapper and
creating "permanent" sockets are tied together. Expand the internal APIs to
allow these two socket characteristics to be separately specified.
This will be externalized in the next patch.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Cc: Aurelien Charbon <aurelien.charbon@ext.bull.net>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If you lose this race, it can iput a socket inode twice and you get a BUG
in fs/inode.c
When I added the option for user-space to close a socket, I added some
cruft to svc_delete_socket so that I could call that function when closing
a socket per user-space request.
This was the wrong thing to do. I should have just set SK_CLOSE and let
normal mechanisms do the work.
Not only wrong, but buggy. The locking is all wrong and it openned up a
race where-by a socket could be closed twice.
So this patch:
Introduces svc_close_socket which sets SK_CLOSE then either leave
the close up to a thread, or calls svc_delete_socket if it can
get SK_BUSY.
Adds a bias to sk_busy which is removed when SK_DEAD is set,
This avoid races around shutting down the socket.
Changes several 'spin_lock' to 'spin_lock_bh' where the _bh
was missing.
Bugzilla-url: http://bugzilla.kernel.org/show_bug.cgi?id=7916
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Speed up high call-rate workloads by caching the struct ip_map for the peer on
the connected struct svc_sock instead of looking it up in the ip_map cache
hashtable on every call. This helps workloads using AUTH_SYS authentication
over TCP.
Testing was on a 4 CPU 4 NIC Altix using 4 IRIX clients, each with 16
synthetic client threads simulating an rsync (i.e. recursive directory
listing) workload reading from an i386 RH9 install image (161480 regular files
in 10841 directories) on the server. That tree is small enough to fill in the
server's RAM so no disk traffic was involved. This setup gives a sustained
call rate in excess of 60000 calls/sec before being CPU-bound on the server.
Profiling showed strcmp(), called from ip_map_match(), was taking 4.8% of each
CPU, and ip_map_lookup() was taking 2.9%. This patch drops both contribution
into the profile noise.
Note that the above result overstates this value of this patch for most
workloads. The synthetic clients are all using separate IP addresses, so
there are 64 entries in the ip_map cache hash. Because the kernel measured
contained the bug fixed in commit
commit 1f1e030bf7
and was running on 64bit little-endian machine, probably all of those 64
entries were on a single chain, thus increasing the cost of ip_map_lookup().
With a modern kernel you would need more clients to see the same amount of
performance improvement. This patch has helped to scale knfsd to handle a
deployment with 2000 NFS clients.
Signed-off-by: Greg Banks <gnb@melbourne.sgi.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Split out the list of idle threads and pending sockets from svc_serv into a
new svc_pool structure, and allocate a fixed number (in this patch, 1) of
pools per svc_serv. The new structure contains a lock which takes over
several of the duties of svc_serv->sv_lock, which is now relegated to
protecting only sv_tempsocks, sv_permsocks, and sv_tmpcnt in svc_serv.
The point is to move the hottest fields out of svc_serv and into svc_pool,
allowing a following patch to arrange for a svc_pool per NUMA node or per CPU.
This is a major step towards making the NFS server NUMA-friendly.
Signed-off-by: Greg Banks <gnb@melbourne.sgi.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Convert the svc_sock->sk_reserved variable from an int protected by
svc_serv->sv_lock, to an atomic. This reduces (by 1) the number of places we
need to take the (effectively global) svc_serv->sv_lock.
Signed-off-by: Greg Banks <gnb@melbourne.sgi.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Protect the svc_sock->sk_deferred list with a new lock svc_sock->sk_defer_lock
instead of svc_serv->sv_lock. Using the more fine-grained lock reduces the
number of places we need to take the svc_serv lock.
Signed-off-by: Greg Banks <gnb@melbourne.sgi.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Convert the svc_sock->sk_inuse counter from an int protected by
svc_serv->sv_lock, to an atomic. This reduces the number of places we need to
take the (effectively global) svc_serv->sv_lock.
Signed-off-by: Greg Banks <gnb@melbourne.sgi.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Following are 11 patches from Greg Banks which combine to make knfsd more
Numa-aware. They reduce hitting on 'global' data structures, and create some
data-structures that can be node-local.
knfsd threads are bound to a particular node, and the thread to handle a new
request is chosen from the threads that are attach to the node that received
the interrupt.
The distribution of threads across nodes can be controlled by a new file in
the 'nfsd' filesystem, though the default approach of an even spread is
probably fine for most sites.
Some (old) numbers that show the efficacy of these patches: N == number of
NICs == number of CPUs == nmber of clients. Number of NUMA nodes == N/2
N Throughput, MiB/s CPU usage, % (max=N*100)
Before After Before After
--- ------ ---- ----- -----
4 312 435 350 228
6 500 656 501 418
8 562 804 690 589
This patch:
Move the aging of RPC/TCP connection sockets from the main svc_recv() loop to
a timer which uses a mark-and-sweep algorithm every 6 minutes. This reduces
the amount of work that needs to be done in the main RPC loop and the length
of time we need to hold the (effectively global) svc_serv->sv_lock.
[akpm@osdl.org: cleanup]
Signed-off-by: Greg Banks <gnb@melbourne.sgi.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
It isn't needed as it is available in rqstp->rq_server, and dropping it allows
some local vars to be dropped.
[akpm@osdl.org: build fix]
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Userspace should create and bind a socket (but not connectted) and write the
'fd' to portlist. This will cause the nfs server to listen on that socket.
To close a socket, the name of the socket - as read from 'portlist' can be
written to 'portlist' with a preceding '-'.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This file will list all ports that nfsd has open.
Default when TCP enabled will be
ipv4 udp 0.0.0.0 2049
ipv4 tcp 0.0.0.0 2049
Later, the list of ports will be settable.
'portlist' chosen rather than 'ports', to avoid unnecessary confusion with
non-mainline patches which created 'ports' with different semantics.
[akpm@osdl.org: cleanups, build fix]
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Semaphore to mutex conversion.
The conversion was generated via scripts, and the result was validated
automatically via a script as well.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Tom Tucker