netfilter: nf_nat: better error handling of nf_ct_expect_related() in helpers
This patch improves the situation in which the expectation table is
full for conntrack NAT helpers. Basically, we give up if we don't
find a place in the table instead of looping over nf_ct_expect_related()
with a different port (we should only do this if it returns -EBUSY, for
-EMFILE or -ESHUTDOWN I think that it's better to skip this).
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Patrick McHardy <kaber@trash.net>
Change the usage of svc usecnt during command execution:
- we check if svc is registered but we do not need to hold usecnt
reference while under __ip_vs_mutex, only the packet handling needs
it during scheduling
- change __ip_vs_service_get to __ip_vs_service_find and
__ip_vs_svc_fwm_get to __ip_vs_svc_fwm_find because now caller
will increase svc->usecnt
- put common code that calls update_service in __ip_vs_update_dest
- put common code in ip_vs_unlink_service() and use it to unregister
the service
- add comment that svc should not be accessed after ip_vs_del_service
anymore
- all IP_VS_WAIT_WHILE calls are now unified: usecnt > 0
- Properly log the app ports
As result, some problems are fixed:
- possible use-after-free of svc in ip_vs_genl_set_cmd after
ip_vs_del_service because our usecnt reference does not guarantee that
svc is not freed on refcnt==0, eg. when no dests are moved to trash
- possible usecnt leak in do_ip_vs_set_ctl after ip_vs_del_service
when the service is not freed now, for example, when some
destionations are moved into trash and svc->refcnt remains above 0.
It is harmless because svc is not in hash anymore.
Signed-off-by: Julian Anastasov <ja@ssi.bg> Acked-by: Simon Horman <horms@verge.net.au> Signed-off-by: Patrick McHardy <kaber@trash.net>
netfilter: save the hash of the tuple in the original direction for latter use
Since we don't change the tuple in the original direction, we can save it
in ct->tuplehash[IP_CT_DIR_REPLY].hnode.pprev for __nf_conntrack_confirm()
use.
__hash_conntrack() is split into two steps: hash_conntrack_raw() is used
to get the raw hash, and __hash_bucket() is used to get the bucket id.
In SYN-flood case, early_drop() doesn't need to recompute the hash again.
Signed-off-by: Changli Gao <xiaosuo@gmail.com> Signed-off-by: Patrick McHardy <kaber@trash.net>
Add new sysctl flag "snat_reroute". Recent kernels use
ip_route_me_harder() to route LVS-NAT responses properly by
VIP when there are multiple paths to client. But setups
that do not have alternative default routes can skip this
routing lookup by using snat_reroute=0.
Signed-off-by: Julian Anastasov <ja@ssi.bg> Signed-off-by: Patrick McHardy <kaber@trash.net>
Add more code to IPVS to work with Netfilter connection
tracking and fix some problems.
- Allow IPVS to be compiled without connection tracking as in
2.6.35 and before. This can avoid keeping conntracks for all
IPVS connections because this costs memory. ip_vs_ftp still
depends on connection tracking and NAT as implemented for 2.6.36.
- Add sysctl var "conntrack" to enable connection tracking for
all IPVS connections. For loaded IPVS directors it needs
tuning of nf_conntrack_max limit.
- Add IP_VS_CONN_F_NFCT connection flag to request the connection
to use connection tracking. This allows user space to provide this
flag, for example, in dest->conn_flags. This can be useful to
request connection tracking per real server instead of forcing it
for all connections with the "conntrack" sysctl. This flag is
set currently only by ip_vs_ftp and of course by "conntrack" sysctl.
- Add ip_vs_nfct.c file to hold all connection tracking code,
by this way main code should not depend of netfilter conntrack
support.
- Return back the ip_vs_post_routing handler as in 2.6.35 and use
skb->ipvs_property=1 to allow IPVS to work without connection
tracking
Connection tracking:
- most of the code is already in 2.6.36-rc
- alter conntrack reply tuple for LVS-NAT connections when first packet
from client is forwarded and conntrack state is NEW or RELATED.
Additionally, alter reply for RELATED connections from real server,
again for packet in original direction.
- add IP_VS_XMIT_TUNNEL to confirm conntrack (without altering
reply) for LVS-TUN early because we want to call nf_reset. It is
needed because we add IPIP header and the original conntrack
should be preserved, not destroyed. The transmitted IPIP packets
can reuse same conntrack, so we do not set skb->ipvs_property.
- try to destroy conntrack when the IPVS connection is destroyed.
It is not fatal if conntrack disappears before that, it depends
on the used timers.
Fix problems from long time:
- add skb->ip_summed = CHECKSUM_NONE for the LVS-TUN transmitters
Signed-off-by: Julian Anastasov <ja@ssi.bg> Signed-off-by: Patrick McHardy <kaber@trash.net>
- the sync protocol supports 16 bits only, so bits 0..15 should be
used only for flags that should go to backup server, bits 16 and
above should be allocated for flags not sent to backup.
- use IP_VS_CONN_F_DEST_MASK as mask of connection flags in
destination that can be changed by user space
- allow IP_VS_CONN_F_ONE_PACKET to be set in destination
Signed-off-by: Julian Anastasov <ja@ssi.bg> Signed-off-by: Patrick McHardy <kaber@trash.net>
netfilter: nf_nat: no IP_NAT_RANGE_MAP_IPS flags when alloc_null_binding()
When alloc_null_binding(), no IP_NAT_RNAGE_MAP_IPS in flags means no IP address
translation is needed. It isn't necessary to specify the address explicitly.
Signed-off-by: Changli Gao <xiaosuo@gmail.com> Signed-off-by: Patrick McHardy <kaber@trash.net>
Jean Delvare [Sun, 12 Sep 2010 22:45:39 +0000 (22:45 +0000)]
e1000e: Simplify MSI interrupt testing
The code is quite convoluted, simplify it. This also avoids calling
e1000_request_irq() without testing the value it returned, which was
bad.
Signed-off-by: Jean Delvare <jdelvare@suse.de> Cc: Bruce Allan <bruce.w.allan@intel.com> Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com> Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Default number of rx buffers will be divided equally
between allocated queues. This will decrease amount of
pre-allocated buffers on systems with multiple CPUs.
User can override this behavior with ethtool -G.
Minimum amount of rx buffers per queue set to 128.
Reported-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Steve Hodgson [Fri, 10 Sep 2010 06:42:22 +0000 (06:42 +0000)]
sfc: Make the dmaq size a run-time setting (rather than compile-time)
- Allow the ring size to be specified in non
power-of-two sizes (for instance to limit
the amount of receive buffers).
- Automatically size the event queue.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ben Hutchings [Fri, 10 Sep 2010 06:41:47 +0000 (06:41 +0000)]
sfc: Refactor channel and queue lookup and iteration
In preparation for changes to the way channels and queue structures
are allocated, revise the macros and functions used to look up and
iterator over them.
- Replace efx_for_each_tx_queue() with iteration over channels then TX
queues
- Replace efx_for_each_rx_queue() with iteration over channels then RX
queues (with one exception, shortly to be removed)
- Introduce efx_get_{channel,rx_queue,tx_queue}() functions to look up
channels and queues by index
- Introduce efx_channel_get_{rx,tx}_queue() functions to look up a
channel's queues
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ben Hutchings [Fri, 10 Sep 2010 06:41:26 +0000 (06:41 +0000)]
sfc: Allocate DMA and event rings using GFP_KERNEL
Currently we allocate DMA descriptor rings and event rings using
pci_alloc_consistent() which selects non-blocking behaviour from the
page allocator (GFP_ATOMIC). This is unnecessary, and since we
currently allocate a single contiguous block for each ring (up to 32
pages!) these allocations are likely to fail if there is any
significant memory pressure. Use dma_alloc_coherent() and GFP_KERNEL
instead.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ben Hutchings [Fri, 10 Sep 2010 06:41:06 +0000 (06:41 +0000)]
sfc: Accumulate RX_NODESC_DROP count in rx_dropped, not rx_over_errors
rx_over_errors appears to be intended as a count of packets that
overflow a packet buffer in the NIC. Given that we implement a
cut-through receive path, this should always be 0.
rx_dropped appears to be the correct counter for packets dropped due
to lack of host buffers.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ben Hutchings [Fri, 10 Sep 2010 06:41:00 +0000 (06:41 +0000)]
sfc: Use MCDI RX_BAD_FCS_PKTS count as MAC rx_bad count
Calculating rx_bad as rx_packets - rx_good is unnecessary and
incorrect, since rx_good does not include control frames (e.g.
pause frames) and rx_packets does.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Dan Williams [Wed, 8 Sep 2010 07:50:47 +0000 (07:50 +0000)]
ipheth: remove incorrect devtype to WWAN
The 'wwan' devtype is meant for devices that require preconfiguration
and *every* time setup before the ethernet interface can be used, like
cellular modems which require a series of setup commands on serial ports
or other mechanisms before the ethernet interface will handle packets.
As ipheth only requires one-per-hotplug pairing setup with no
preconfiguration (like APN, phone #, etc) and the network interface is
usable at any time after that initial setup, remove the incorrect
devtype wwan.
Signed-off-by: Dan Williams <dcbw@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes the following checkpatch.pl warnings:
- spaces after tabs
- space between function and arguments
- one-line statement braces
- tabs instead of spaces
Signed-off-by: Florian Fainelli <florian@openwrt.org> Signed-off-by: David S. Miller <davem@davemloft.net>
Paul Gortmaker [Wed, 8 Sep 2010 13:31:24 +0000 (13:31 +0000)]
tipc: Optimize handling excess content on incoming messages
Remove code that trimmed excess trailing info from incoming messages
arriving over an Ethernet interface. TIPC now ignores the extra info
while the message is being processed by the node, and only trims it off
if the message is retransmitted to another node. (This latter step is
done to ensure the extra info doesn't cause the sk_buff to exceed the
outgoing interface's MTU limit.) The outgoing buffer is guaranteed to
be linear.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com> Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet [Thu, 9 Sep 2010 05:33:43 +0000 (05:33 +0000)]
tunnels: missing rcu_assign_pointer()
xfrm4_tunnel_register() & xfrm6_tunnel_register() should
use rcu_assign_pointer() to make sure previous writes
(to handler->next) are committed to memory before chain
insertion.
deregister functions dont need a particular barrier.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Namhyung Kim [Wed, 8 Sep 2010 03:48:48 +0000 (03:48 +0000)]
net/core: add lock context change annotations in net/core/sock.c
__lock_sock() and __release_sock() releases and regrabs lock but
were missing proper annotations. Add it. This removes following
warning from sparse. (Currently __lock_sock() does not emit any
warning about it but I think it is better to add also.)
net/core/sock.c:1580:17: warning: context imbalance in '__release_sock' - unexpected unlock
Signed-off-by: Namhyung Kim <namhyung@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Namhyung Kim [Wed, 8 Sep 2010 03:48:47 +0000 (03:48 +0000)]
net/core: remove address space warnings on verify_iovec()
move_addr_to_kernel() and copy_from_user() requires their argument
as __user pointer but were missing proper markups. Add it.
This removes following warnings from sparse.
net/core/iovec.c:44:52: warning: incorrect type in argument 1 (different address spaces)
net/core/iovec.c:44:52: expected void [noderef] <asn:1>*uaddr
net/core/iovec.c:44:52: got void *msg_name
net/core/iovec.c:55:34: warning: incorrect type in argument 2 (different address spaces)
net/core/iovec.c:55:34: expected void const [noderef] <asn:1>*from
net/core/iovec.c:55:34: got struct iovec *msg_iov
Signed-off-by: Namhyung Kim <namhyung@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet [Wed, 8 Sep 2010 05:08:44 +0000 (05:08 +0000)]
udp: add rehash on connect()
commit 30fff923 introduced in linux-2.6.33 (udp: bind() optimisation)
added a secondary hash on UDP, hashed on (local addr, local port).
Problem is that following sequence :
fd = socket(...)
connect(fd, &remote, ...)
not only selects remote end point (address and port), but also sets
local address, while UDP stack stored in secondary hash table the socket
while its local address was INADDR_ANY (or ipv6 equivalent)
Sequence is :
- autobind() : choose a random local port, insert socket in hash tables
[while local address is INADDR_ANY]
- connect() : set remote address and port, change local address to IP
given by a route lookup.
When an incoming UDP frame comes, if more than 10 sockets are found in
primary hash table, we switch to secondary table, and fail to find
socket because its local address changed.
One solution to this problem is to rehash datagram socket if needed.
We add a new rehash(struct socket *) method in "struct proto", and
implement this method for UDP v4 & v6, using a common helper.
This rehashing only takes care of secondary hash table, since primary
hash (based on local port only) is not changed.
Reported-by: Krzysztof Piotr Oledzki <ole@ans.pl> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Tested-by: Krzysztof Piotr Oledzki <ole@ans.pl> Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com> Acked-by: Chris Snook <chris.snook@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ondrej Zary [Thu, 9 Sep 2010 04:29:20 +0000 (21:29 -0700)]
cxacru: ignore cx82310_eth devices
Ignore ADSL routers, which can have the same vendor and product IDs
as ADSL modems but should be handled by the cx82310_eth driver.
This intentionally ignores device IDs that aren't currently handled
by cx82310_eth. There may be other device IDs that perhaps shouldn't
be claimed by cxacru.
Signed-off-by: Ondrej Zary <linux@rainbow-software.org> Signed-off-by: Simon Arlott <simon@fire.lp0.eu> Signed-off-by: David S. Miller <davem@davemloft.net>
Andy Grover [Wed, 25 Aug 2010 12:51:28 +0000 (05:51 -0700)]
RDS: Implement masked atomic operations
Add two CMSGs for masked versions of cswp and fadd. args
struct modified to use a union for different atomic op type's
arguments. Change IB to do masked atomic ops. Atomic op type
in rds_message similarly unionized.
Signed-off-by: Andy Grover <andy.grover@oracle.com>
Zach Brown [Fri, 23 Jul 2010 17:36:58 +0000 (10:36 -0700)]
RDS: don't call rds_conn_shutdown() from rds_conn_destroy()
rds_conn_shutdown() can return before the connection is shut down when
it encounters an existing state that it doesn't understand. This lets
rds_conn_destroy() then start tearing down the conn from under paths
that are still using it.
It's more reliable the shutdown work and wait for krdsd to complete the
shutdown callback. This stopped some hangs I was seeing where krdsd was
trying to shut down a freed conn.
Zach Brown [Fri, 23 Jul 2010 17:32:31 +0000 (10:32 -0700)]
RDS: have sockets get transport module references
Right now there's nothing to stop the various paths that use
rs->rs_transport from racing with rmmod and executing freed transport
code. The simple fix is to have binding to a transport also hold a
reference to the transport's module, removing this class of races.
We already had an unused t_owner field which was set for the modular
transports and which wasn't set for the built-in loop transport.
Zach Brown [Thu, 15 Jul 2010 19:34:33 +0000 (12:34 -0700)]
RDS/IB: protect the list of IB devices
The RDS IB device list wasn't protected by any locking. Traversal in
both the get_mr and FMR flushing paths could race with additon and
removal.
List manipulation is done with RCU primatives and is protected by the
write side of a rwsem. The list traversal in the get_mr fast path is
protected by a rcu read critical section. The FMR list traversal is
more problematic because it can block while traversing the list. We
protect this with the read side of the rwsem.
Zach Brown [Wed, 14 Jul 2010 21:01:21 +0000 (14:01 -0700)]
RDS/IB: print IB event strings as well as their number
It's nice to not have to go digging in the code to see which event
occurred. It's easy to throw together a quick array that maps the ib
event enums to their strings. I didn't see anything in the stack that
does this translation for us, but I also didn't look very hard.
Chris Mason [Tue, 20 Jul 2010 00:06:46 +0000 (17:06 -0700)]
RDS: flush fmrs before allocating new ones
Flushing FMRs is somewhat expensive, and is currently kicked off when
the interrupt handler notices that we are getting low. The result of
this is that FMR flushing only happens from the interrupt cpus.
This spreads the load more effectively by triggering flushes just before
we allocate a new FMR.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Zach Brown [Wed, 14 Jul 2010 20:55:35 +0000 (13:55 -0700)]
RDS/IB: track signaled sends
We're seeing bugs today where IB connection shutdown clears the send
ring while the tasklet is processing completed sends. Implementation
details cause this to dereference a null pointer. Shutdown needs to
wait for send completion to stop before tearing down the connection. We
can't simply wait for the ring to empty because it may contain
unsignaled sends that will never be processed.
This patch tracks the number of signaled sends that we've posted and
waits for them to complete. It also makes sure that the tasklet has
finished executing.
Zach Brown [Tue, 6 Jul 2010 22:04:34 +0000 (15:04 -0700)]
RDS/IB: always process recv completions
The recv refill path was leaking fragments because the recv event handler had
marked a ring element as free without freeing its frag. This was happening
because it wasn't processing receives when the conn wasn't marked up or
connecting, as can be the case if it races with rmmod.
Two observations support always processing receives in the callback.
First, buildup should only post receives, thus triggering recv event handler
calls, once it has built up all the state to handle them. Teardown should
destroy the CQ and drain the ring before tearing down the state needed to
process recvs. Both appear to be true today.
Second, this test was fundamentally racy. There is nothing to stop rmmod and
connection destruction from swooping in the moment after the conn state was
sampled but before real receive procesing starts.
Zach Brown [Tue, 6 Jul 2010 22:08:48 +0000 (15:08 -0700)]
RDS: return to a single-threaded krdsd
We were seeing very nasty bugs due to fundamental assumption the current code
makes about concurrent work struct processing. The code simpy isn't able to
handle concurrent connection shutdown work function execution today, for
example, which is very much possible once a multi-threaded krdsd was
introduced. The problem compounds as additional work structs are added to the
mix.
krdsd is no longer perforance critical now that send and receive posting and
FMR flushing are done elsewhere, so the safest fix is to move back to the
single threaded krdsd that the current code was built around.
Zach Brown [Tue, 6 Jul 2010 22:09:56 +0000 (15:09 -0700)]
RDS/IB: create a work queue for FMR flushing
This patch moves the FMR flushing work in to its own mult-threaded work queue.
This is to maintain performance in preparation for returning the main krdsd
work queue back to a single threaded work queue to avoid deep-rooted
concurrency bugs.
This is also good because it further separates FMRs, which might be removed
some day, from the rest of the code base.
Zach Brown [Fri, 25 Jun 2010 21:58:16 +0000 (14:58 -0700)]
RDS/IB: destroy connections on rmmod
IB connections were not being destroyed during rmmod.
First, recently IB device removal callback was changed to disconnect
connections that used the removing device rather than destroying them. So
connections with devices during rmmod were not being destroyed.
Second, rds_ib_destroy_nodev_conns() was being called before connections are
disassociated with devices. It would almost never find connections in the
nodev list.
We first get rid of rds_ib_destroy_conns(), which is no longer called, and
refactor the existing caller into the main body of the function and get rid of
the list and lock wrappers.
Then we call rds_ib_destroy_nodev_conns() *after* ib_unregister_client() has
removed the IB device from all the conns and put the conns on the nodev list.
The result is that IB connections are destroyed by rmmod.
Zach Brown [Fri, 25 Jun 2010 21:59:49 +0000 (14:59 -0700)]
RDS/IB: wait for IB dev freeing work to finish during rmmod
The RDS IB client removal callback can queue work to drop the final reference
to an IB device. We have to make sure that this function has returned before
we complete rmmod or the work threads can try to execute freed code.
Chris Mason [Fri, 11 Jun 2010 18:17:59 +0000 (11:17 -0700)]
rds: recycle FMRs through lockless lists
FRM allocation and recycling is performance critical and fairly lock
intensive. The current code has a per connection lock that all
processes bang on and it becomes a major bottleneck on large systems.
This changes things to use a number of cmpxchg based lists instead,
allowing us to go through the whole FMR lifecycle without locking inside
RDS.
Zach Brown pointed out that our usage of cmpxchg for xlist removal is
racey if someone manages to remove and add back an FMR struct into the list
while another CPU can see the FMR's address at the head of the list.
The second CPU might assume the list hasn't changed when in fact any
number of operations might have happened in between the deletion and
reinsertion.
This commit maintains a per cpu count of CPUs that are currently
in xlist removal, and establishes a grace period to make sure that
nobody can see an entry we have just removed from the list.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Zach Brown [Fri, 4 Jun 2010 21:41:41 +0000 (14:41 -0700)]
rds: fix rds_send_xmit() serialization
rds_send_xmit() was changed to hold an interrupt masking spinlock instead of a
mutex so that it could be called from the IB receive tasklet path. This broke
the TCP transport because its xmit method can block and masks and unmasks
interrupts.
This patch serializes callers to rds_send_xmit() with a simple bit instead of
the current spinlock or previous mutex. This enables rds_send_xmit() to be
called from any context and to call functions which block. Getting rid of the
c_send_lock exposes the bare c_lock acquisitions which are changed to block
interrupts.
A waitqueue is added so that rds_conn_shutdown() can wait for callers to leave
rds_send_xmit() before tearing down partial send state. This lets us get rid
of c_senders.
rds_send_xmit() is changed to check the conn state after acquiring the
RDS_IN_XMIT bit to resolve races with the shutdown path. Previously both
worked with the conn state and then the lock in the same order, allowing them
to race and execute the paths concurrently.
rds_send_reset() isn't racing with rds_send_xmit() now that rds_conn_shutdown()
properly ensures that rds_send_xmit() can't start once the conn state has been
changed. We can remove its previous use of the spinlock.
Finally, c_send_generation is redundant. Callers can race to test the c_flags
bit by simply retrying instead of racing to test the c_send_generation atomic.
Zach Brown [Fri, 4 Jun 2010 21:25:27 +0000 (14:25 -0700)]
rds: block ints when acquiring c_lock in rds_conn_message_info()
conn->c_lock is acquired in interrupt context. rds_conn_message_info() is
called from user context and was acquiring c_lock without blocking interrupts,
leading to possible deadlocks.
Zach Brown [Fri, 4 Jun 2010 21:26:32 +0000 (14:26 -0700)]
rds: remove unused rds_send_acked_before()
rds_send_acked_before() wasn't blocking interrupts when acquiring c_lock from
user context but nothing calls it. Rather than fix its use of c_lock we just
remove the function.
Chris Mason [Thu, 27 May 2010 04:45:06 +0000 (21:45 -0700)]
RDS: use friendly gfp masks for prefill
When prefilling the rds frags, we end up doing a lot of allocations.
We're not in atomic context here, and so there's no reason to dip into
atomic reserves. This changes the prefills to use masks that allow
waiting.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Chris Mason [Thu, 27 May 2010 05:05:37 +0000 (22:05 -0700)]
RDS/IB: Add caching of frags and incs
This patch is based heavily on an initial patch by Chris Mason.
Instead of freeing slab memory and pages, it keeps them, and
funnels them back to be reused.
The lock minimization strategy uses xchg and cmpxchg atomic ops
for manipulation of pointers to list heads. We anchor the lists with a
pointer to a list_head struct instead of a static list_head struct.
We just have to carefully use the existing primitives with
the difference between a pointer and a static head struct.
For example, 'list_empty()' means that our anchor pointer points to a list with
a single item instead of meaning that our static head element doesn't point to
any list items.
Original patch by Chris, with significant mods and fixes by Andy and Zach.
Signed-off-by: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andy Grover <andy.grover@oracle.com> Signed-off-by: Zach Brown <zach.brown@oracle.com>
Andy Grover [Tue, 25 May 2010 03:28:49 +0000 (20:28 -0700)]
RDS: Assume recv->r_frag is always NULL in refill_one()
refill_one() should never be called on a recv struct that
doesn't need a new r_frag allocated. Add a WARN and remove
conditional around r_frag alloc code.
Also, add a comment to explain why r_ibinc may or may not
need refilling.
Signed-off-by: Andy Grover <andy.grover@oracle.com>
Andy Grover [Tue, 25 May 2010 03:12:41 +0000 (20:12 -0700)]
RDS: Use page_remainder_alloc() for recv bufs
Instead of splitting up a page into RDS_FRAG_SIZE chunks
ourselves, ask rds_page_remainder_alloc() to do it. While it
is possible PAGE_SIZE > FRAG_SIZE, on x86en it isn't, so having
duplicate "carve up a page into buffers" code seems excessive.
The other modification this spawns is the use of a single
struct scatterlist in rds_page_frag instead of a bare page ptr.
This causes verbosity to increase in some places, and decrease
in others.
Finally, I decided to unify the lifetimes and alloc/free of
rds_page_frag and its page. This is a nice simplification in itself,
but will be extra-nice once we come to adding cmason's recycling
patch.
Signed-off-by: Andy Grover <andy.grover@oracle.com>
Zach Brown [Mon, 24 May 2010 20:16:57 +0000 (13:16 -0700)]
RDS/IB: disconnect when IB devices are removed
Currently IB device removal destroys connections which are associated with the
device. This prevents connections from being re-established when replacement
devices are added.
Instead we'll queue shutdown work on the connections as their devices are
removed. When we see that devices are added we triger connection attempts on
all connections that don't currently have a device.
The result is that RDS sockets can resume device-independent work (bcopy, not
RDMA) across IB device removal and restoration.
Zach Brown [Tue, 18 May 2010 22:48:51 +0000 (15:48 -0700)]
RDS/IB: add refcount tracking to struct rds_ib_device
The RDS IB client .remove callback used to free the rds_ibdev for the given
device unconditionally. This could race other users of the struct. This patch
adds refcounting so that we only free the rds_ibdev once all of its users are
done.
Many rds_ibdev users are tied to connections. We give the connection a
reference and change these users to reference the device in the connection
instead of looking it up in the IB client data. The only user of the IB client
data remaining is the first lookup of the device as connections are built up.
Incrementing the reference count of a device found in the IB client data could
race with final freeing so we use an RCU grace period to make sure that freeing
won't happen until those lookups are done.
MRs need the rds_ibdev to get at the pool that they're freed in to. They exist
outside a connection and many MRs can reference different devices from one
socket, so it was natural to have each MR hold a reference. MR refs can be
dropped from interrupt handlers and final device teardown can block so we push
it off to a work struct. Pool teardown had to be fixed to cancel its pending
work instead of deadlocking waiting for all queued work, including itself, to
finish.
MRs get their reference from the global device list, which gets a reference.
It is left unprotected by locks and remains racy. A simple global lock would
be a significant bottleneck. More scalable (complicated) locking should be
done carefully in a later patch.
Zach Brown [Tue, 18 May 2010 22:44:50 +0000 (15:44 -0700)]
RDS/IB: get the xmit max_sge from the RDS IB device on the connection
rds_ib_xmit_rdma() was calling ib_get_client_data() to get at the rds_ibdevice
just to get the max_sge for the transmit. This patch instead has it get it
directly off the rds_ibdev which is stored on the connection.
The current code won't free the rds_ibdev until all the IB connections that use
it are freed. So it's safe to reference the rds_ibdev this way. In the future
it also makes it easier to support proper reference counting of the rds_ibdev
struct.
As an additional bonus, this gets rid of the performance hit of calling in to
the IB stack to look up the rds_ibdev. The current implementation in the IB
stack acquires an interrupt blocking spinlock to protect the registration of
client callback data.
projects / linux-beck.git / log