percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Add the "__must_check" tag to sk_add_backlog() so that any failure to
check and drop packets will be warned about.
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sk_add_backlog -> __sk_add_backlog
sk_add_backlog_limited -> sk_add_backlog
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We got system OOM while running some UDP netperf testing on the loopback
device. The case is multiple senders sent stream UDP packets to a single
receiver via loopback on local host. Of course, the receiver is not able
to handle all the packets in time. But we surprisingly found that these
packets were not discarded due to the receiver's sk->sk_rcvbuf limit.
Instead, they are kept queuing to sk->sk_backlog and finally ate up all
the memory. We believe this is a secure hole that a none privileged user
can crash the system.
The root cause for this problem is, when the receiver is doing
__release_sock() (i.e. after userspace recv, kernel udp_recvmsg ->
skb_free_datagram_locked -> release_sock), it moves skbs from backlog to
sk_receive_queue with the softirq enabled. In the above case, multiple
busy senders will almost make it an endless loop. The skbs in the
backlog end up eat all the system memory.
The issue is not only for UDP. Any protocols using socket backlog is
potentially affected. The patch adds limit for socket backlog so that
the backlog size cannot be expanded endlessly.
Reported-by: Alex Shi <alex.shi@intel.com>
Cc: David Miller <davem@davemloft.net>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru
Cc: "Pekka Savola (ipv6)" <pekkas@netcore.fi>
Cc: Patrick McHardy <kaber@trash.net>
Cc: Vlad Yasevich <vladislav.yasevich@hp.com>
Cc: Sridhar Samudrala <sri@us.ibm.com>
Cc: Jon Maloy <jon.maloy@ericsson.com>
Cc: Allan Stephens <allan.stephens@windriver.com>
Cc: Andrew Hendry <andrew.hendry@gmail.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Convert AF_PACKET to use RCU, eliminating one more reader/writer lock.
There is no need for a real sk_del_node_init_rcu(), because sk_del_node_init
is doing the equivalent thing to hlst_del_init_rcu already; but added
some comments to try and make that obvious.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The function name must be followed by a space, hypen, space, and a
short description.
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extends udp_table to contain a secondary hash table.
socket anchor for this second hash is free, because UDP
doesnt use skc_bind_node : We define an union to hold
both skc_bind_node & a new hlist_nulls_node udp_portaddr_node
udp_lib_get_port() inserts sockets into second hash chain
(additional cost of one atomic op)
udp_lib_unhash() deletes socket from second hash chain
(additional cost of one atomic op)
Note : No spinlock lockdep annotation is needed, because
lock for the secondary hash chain is always get after
lock for primary hash chain.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Union sk_hash with two u16 hashes for udp (no extra memory taken)
One 16 bits hash on (local port) value (the previous udp 'hash')
One 16 bits hash on (local address, local port) values, initialized
but not yet used. This second hash is using jenkin hash for better
distribution.
Because the 'port' is xored later, a partial hash is performed
on local address + net_hash_mix(net)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce sk_tx_queue_mapping; and functions that set, test and
get this value. Reset sk_tx_queue_mapping to -1 whenever the dst
cache is set/reset, and in socket alloc. Setting txq to -1 and
using valid txq=<0 to n-1> allows the tx path to use the value
of sk_tx_queue_mapping directly instead of subtracting 1 on every
tx.
Signed-off-by: Krishna Kumar <krkumar2@in.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Create a new socket level option to report number of queue overflows
Recently I augmented the AF_PACKET protocol to report the number of frames lost
on the socket receive queue between any two enqueued frames. This value was
exported via a SOL_PACKET level cmsg. AFter I completed that work it was
requested that this feature be generalized so that any datagram oriented socket
could make use of this option. As such I've created this patch, It creates a
new SOL_SOCKET level option called SO_RXQ_OVFL, which when enabled exports a
SOL_SOCKET level cmsg that reports the nubmer of times the sk_receive_queue
overflowed between any two given frames. It also augments the AF_PACKET
protocol to take advantage of this new feature (as it previously did not touch
sk->sk_drops, which this patch uses to record the overflow count). Tested
successfully by me.
Notes:
1) Unlike my previous patch, this patch simply records the sk_drops value, which
is not a number of drops between packets, but rather a total number of drops.
Deltas must be computed in user space.
2) While this patch currently works with datagram oriented protocols, it will
also be accepted by non-datagram oriented protocols. I'm not sure if thats
agreeable to everyone, but my argument in favor of doing so is that, for those
protocols which aren't applicable to this option, sk_drops will always be zero,
and reporting no drops on a receive queue that isn't used for those
non-participating protocols seems reasonable to me. This also saves us having
to code in a per-protocol opt in mechanism.
3) This applies cleanly to net-next assuming that commit
977750076d (my af packet cmsg patch) is reverted
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since commit a98b65a3 (net: annotate struct sock bitfield), we lost
8 bytes in struct sock on 64bit arches because of
kmemcheck_bitfield_end(flags) misplacement.
Fix this by putting together sk_shutdown, sk_no_check, sk_userlocks,
sk_protocol and sk_type in the 'flags' 32bits bitfield
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
An incoming datagram must bring into cpu cache *lot* of cache lines,
in particular : (other parts omitted (hash chains, ip route cache...))
On 32bit arches :
offsetof(struct sock, sk_rcvbuf) =0x30 (read)
offsetof(struct sock, sk_lock) =0x34 (rw)
offsetof(struct sock, sk_sleep) =0x50 (read)
offsetof(struct sock, sk_rmem_alloc) =0x64 (rw)
offsetof(struct sock, sk_receive_queue)=0x74 (rw)
offsetof(struct sock, sk_forward_alloc)=0x98 (rw)
offsetof(struct sock, sk_callback_lock)=0xcc (rw)
offsetof(struct sock, sk_drops) =0xd8 (read if we add dropcount support, rw if frame dropped)
offsetof(struct sock, sk_filter) =0xf8 (read)
offsetof(struct sock, sk_socket) =0x138 (read)
offsetof(struct sock, sk_data_ready) =0x15c (read)
We can avoid sk->sk_socket and socket->fasync_list referencing on sockets
with no fasync() structures. (socket->fasync_list ptr is probably already in cache
because it shares a cache line with socket->wait, ie location pointed by sk->sk_sleep)
This avoids one cache line load per incoming packet for common cases (no fasync())
We can leave (or even move in a future patch) sk->sk_socket in a cold location
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This provides safety against negative optlen at the type
level instead of depending upon (sometimes non-trivial)
checks against this sprinkled all over the the place, in
each and every implementation.
Based upon work done by Arjan van de Ven and feedback
from Linus Torvalds.
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit e912b1142b
(net: sk_prot_alloc() should not blindly overwrite memory)
took care of not zeroing whole new socket at allocation time.
sock_copy() is another spot where we should be very careful.
We should not set refcnt to a non null value, until
we are sure other fields are correctly setup, or
a lockless reader could catch this socket by mistake,
while not fully (re)initialized.
This patch puts sk_node & sk_refcnt to the very beginning
of struct sock to ease sock_copy() & sk_prot_alloc() job.
We add appropriate smp_wmb() before sk_refcnt initializations
to match our RCU requirements (changes to sock keys should
be committed to memory before sk_refcnt setting)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adding smp_mb__after_lock define to be used as a smp_mb call after
a lock.
Making it nop for x86, since {read|write|spin}_lock() on x86 are
full memory barriers.
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adding memory barrier after the poll_wait function, paired with
receive callbacks. Adding fuctions sock_poll_wait and sk_has_sleeper
to wrap the memory barrier.
Without the memory barrier, following race can happen.
The race fires, when following code paths meet, and the tp->rcv_nxt
and __add_wait_queue updates stay in CPU caches.
CPU1 CPU2
sys_select receive packet
... ...
__add_wait_queue update tp->rcv_nxt
... ...
tp->rcv_nxt check sock_def_readable
... {
schedule ...
if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
wake_up_interruptible(sk->sk_sleep)
...
}
If there was no cache the code would work ok, since the wait_queue and
rcv_nxt are opposit to each other.
Meaning that once tp->rcv_nxt is updated by CPU2, the CPU1 either already
passed the tp->rcv_nxt check and sleeps, or will get the new value for
tp->rcv_nxt and will return with new data mask.
In both cases the process (CPU1) is being added to the wait queue, so the
waitqueue_active (CPU2) call cannot miss and will wake up CPU1.
The bad case is when the __add_wait_queue changes done by CPU1 stay in its
cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 will then
endup calling schedule and sleep forever if there are no more data on the
socket.
Calls to poll_wait in following modules were ommited:
net/bluetooth/af_bluetooth.c
net/irda/af_irda.c
net/irda/irnet/irnet_ppp.c
net/mac80211/rc80211_pid_debugfs.c
net/phonet/socket.c
net/rds/af_rds.c
net/rfkill/core.c
net/sunrpc/cache.c
net/sunrpc/rpc_pipe.c
net/tipc/socket.c
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next-2.6:
bnx2: Fix the behavior of ethtool when ONBOOT=no
qla3xxx: Don't sleep while holding lock.
qla3xxx: Give the PHY time to come out of reset.
ipv4 routing: Ensure that route cache entries are usable and reclaimable with caching is off
net: Move rx skb_orphan call to where needed
ipv6: Use correct data types for ICMPv6 type and code
net: let KS8842 driver depend on HAS_IOMEM
can: let SJA1000 driver depend on HAS_IOMEM
netxen: fix firmware init handshake
netxen: fix build with without CONFIG_PM
netfilter: xt_rateest: fix comparison with self
netfilter: xt_quota: fix incomplete initialization
netfilter: nf_log: fix direct userspace memory access in proc handler
netfilter: fix some sparse endianess warnings
netfilter: nf_conntrack: fix conntrack lookup race
netfilter: nf_conntrack: fix confirmation race condition
netfilter: nf_conntrack: death_by_timeout() fix
In order to get the tun driver to account packets, we need to be
able to receive packets with destructors set. To be on the safe
side, I added an skb_orphan call for all protocols by default since
some of them (IP in particular) cannot handle receiving packets
destructors properly.
Now it seems that at least one protocol (CAN) expects to be able
to pass skb->sk through the rx path without getting clobbered.
So this patch attempts to fix this properly by moving the skb_orphan
call to where it's actually needed. In particular, I've added it
to skb_set_owner_[rw] which is what most users of skb->destructor
call.
This is actually an improvement for tun too since it means that
we only give back the amount charged to the socket when the skb
is passed to another socket that will also be charged accordingly.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Tested-by: Oliver Hartkopp <olver@hartkopp.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next-2.6: (55 commits)
netxen: fix tx ring accounting
netxen: fix detection of cut-thru firmware mode
forcedeth: fix dma api mismatches
atm: sk_wmem_alloc initial value is one
net: correct off-by-one write allocations reports
via-velocity : fix no link detection on boot
Net / e100: Fix suspend of devices that cannot be power managed
TI DaVinci EMAC : Fix rmmod error
net: group address list and its count
ipv4: Fix fib_trie rebalancing, part 2
pkt_sched: Update drops stats in act_police
sky2: version 1.23
sky2: add GRO support
sky2: skb recycling
sky2: reduce default transmit ring
sky2: receive counter update
sky2: fix shutdown synchronization
sky2: PCI irq issues
sky2: more receive shutdown
sky2: turn off pause during shutdown
...
Manually fix trivial conflict in net/core/skbuff.c due to kmemcheck
commit 2b85a34e91
(net: No more expensive sock_hold()/sock_put() on each tx)
changed initial sk_wmem_alloc value.
Some protocols check sk_wmem_alloc value to determine if a timer
must delay socket deallocation. We must take care of the sk_wmem_alloc
value being one instead of zero when no write allocations are pending.
Reported by Ingo Molnar, and full diagnostic from David Miller.
This patch introduces three helpers to get read/write allocations
and a followup patch will use these helpers to report correct
write allocations to user.
Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
One of the problem with sock memory accounting is it uses
a pair of sock_hold()/sock_put() for each transmitted packet.
This slows down bidirectional flows because the receive path
also needs to take a refcount on socket and might use a different
cpu than transmit path or transmit completion path. So these
two atomic operations also trigger cache line bounces.
We can see this in tx or tx/rx workloads (media gateways for example),
where sock_wfree() can be in top five functions in profiles.
We use this sock_hold()/sock_put() so that sock freeing
is delayed until all tx packets are completed.
As we also update sk_wmem_alloc, we could offset sk_wmem_alloc
by one unit at init time, until sk_free() is called.
Once sk_free() is called, we atomic_dec_and_test(sk_wmem_alloc)
to decrement initial offset and atomicaly check if any packets
are in flight.
skb_set_owner_w() doesnt call sock_hold() anymore
sock_wfree() doesnt call sock_put() anymore, but check if sk_wmem_alloc
reached 0 to perform the final freeing.
Drawback is that a skb->truesize error could lead to unfreeable sockets, or
even worse, prematurely calling __sk_free() on a live socket.
Nice speedups on SMP. tbench for example, going from 2691 MB/s to 2711 MB/s
on my 8 cpu dev machine, even if tbench was not really hitting sk_refcnt
contention point. 5 % speedup on a UDP transmit workload (depends
on number of flows), lowering TX completion cpu usage.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A long time ago we had bugs, primarily in TCP, where we would modify
skb->truesize (for TSO queue collapsing) in ways which would corrupt
the socket memory accounting.
skb_truesize_check() was added in order to try and catch this error
more systematically.
However this debugging check has morphed into a Frankenstein of sorts
and these days it does nothing other than catch false-positives.
Signed-off-by: David S. Miller <davem@davemloft.net>
The overlap with the old SO_TIMESTAMP[NS] options is handled so
that time stamping in software (net_enable_timestamp()) is
enabled when SO_TIMESTAMP[NS] and/or SO_TIMESTAMPING_RX_SOFTWARE
is set. It's disabled if all of these are off.
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The problem is that in_atomic() will return false inside spinlocks if
CONFIG_PREEMPT=n. This will lead to deadlockable GFP_KERNEL allocations
from spinlocked regions.
Secondly, if CONFIG_PREEMPT=y, this bug solves itself because networking
will instead use GFP_ATOMIC from this callsite. Hence we won't get the
might_sleep() debugging warnings which would have informed us of the buggy
callsites.
Solve both these problems by switching to in_interrupt(). Now, if someone
runs a gfp_any() allocation from inside spinlock we will get the warning
if CONFIG_PREEMPT=y.
I reviewed all callsites and most of them were too complex for my little
brain and none of them documented their interface requirements. I have no
idea what this patch will do.
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The function sock_alloc_send_pskb is completely useless if not
exported since most of the code in it won't be used as is. In
fact, this code has already been duplicated in the tun driver.
Now that we need accounting in the tun driver, we can in fact
use this function as is. So this patch marks it for export again.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of using one atomic_t per protocol, use a percpu_counter
for "orphan_count", to reduce cache line contention on
heavy duty network servers.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of using one atomic_t per protocol, use a percpu_counter
for "sockets_allocated", to reduce cache line contention on
heavy duty network servers.
Note : We revert commit (248969ae31
net: af_unix can make unix_nr_socks visbile in /proc),
since it is not anymore used after sock_prot_inuse_add() addition
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is a straightforward patch, using hlist_nulls infrastructure.
RCUification already done on UDP two weeks ago.
Using hlist_nulls permits us to avoid some memory barriers, both
at lookup time and delete time.
Patch is large because it adds new macros to include/net/sock.h.
These macros will be used by TCP & DCCP in next patch.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
fix this warning:
net/bluetooth/af_bluetooth.c:60: warning: ‘bt_key_strings’ defined but not used
net/bluetooth/af_bluetooth.c:71: warning: ‘bt_slock_key_strings’ defined but not used
this is a lockdep macro problem in the !LOCKDEP case.
We cannot convert it to an inline because the macro works on multiple types,
but we can mark the parameter used.
[ also clean up a misaligned tab in sock_lock_init_class_and_name() ]
[ also remove #ifdefs from around af_family_clock_key strings - which
were certainly added to get rid of the ugly build warnings. ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: David S. Miller <davem@davemloft.net>
Every user is under CONFIG_NET_DMA already, so ifdef field as well.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove excess kernel-doc function parameters from networking header
& driver files:
Warning(include/net/sock.h:946): Excess function parameter or struct member 'sk' description in 'sk_filter_release'
Warning(include/linux/netdevice.h:1545): Excess function parameter or struct member 'cpu' description in 'netif_tx_lock'
Warning(drivers/net/wan/z85230.c:712): Excess function parameter or struct member 'regs' description in 'z8530_interrupt'
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Corey Minyard found a race added in commit 271b72c7fa
(udp: RCU handling for Unicast packets.)
"If the socket is moved from one list to another list in-between the
time the hash is calculated and the next field is accessed, and the
socket has moved to the end of the new list, the traversal will not
complete properly on the list it should have, since the socket will
be on the end of the new list and there's not a way to tell it's on a
new list and restart the list traversal. I think that this can be
solved by pre-fetching the "next" field (with proper barriers) before
checking the hash."
This patch corrects this problem, introducing a new
sk_for_each_rcu_safenext() macro.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Goals are :
1) Optimizing handling of incoming Unicast UDP frames, so that no memory
writes should happen in the fast path.
Note: Multicasts and broadcasts still will need to take a lock,
because doing a full lockless lookup in this case is difficult.
2) No expensive operations in the socket bind/unhash phases :
- No expensive synchronize_rcu() calls.
- No added rcu_head in socket structure, increasing memory needs,
but more important, forcing us to use call_rcu() calls,
that have the bad property of making sockets structure cold.
(rcu grace period between socket freeing and its potential reuse
make this socket being cold in CPU cache).
David did a previous patch using call_rcu() and noticed a 20%
impact on TCP connection rates.
Quoting Cristopher Lameter :
"Right. That results in cacheline cooldown. You'd want to recycle
the object as they are cache hot on a per cpu basis. That is screwed
up by the delayed regular rcu processing. We have seen multiple
regressions due to cacheline cooldown.
The only choice in cacheline hot sensitive areas is to deal with the
complexity that comes with SLAB_DESTROY_BY_RCU or give up on RCU."
- Because udp sockets are allocated from dedicated kmem_cache,
use of SLAB_DESTROY_BY_RCU can help here.
Theory of operation :
---------------------
As the lookup is lockfree (using rcu_read_lock()/rcu_read_unlock()),
special attention must be taken by readers and writers.
Use of SLAB_DESTROY_BY_RCU is tricky too, because a socket can be freed,
reused, inserted in a different chain or in worst case in the same chain
while readers could do lookups in the same time.
In order to avoid loops, a reader must check each socket found in a chain
really belongs to the chain the reader was traversing. If it finds a
mismatch, lookup must start again at the begining. This *restart* loop
is the reason we had to use rdlock for the multicast case, because
we dont want to send same message several times to the same socket.
We use RCU only for fast path.
Thus, /proc/net/udp still takes spinlocks.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
UDP sockets are hashed in a 128 slots hash table.
This hash table is protected by *one* rwlock.
This rwlock is readlocked each time an incoming UDP message is handled.
This rwlock is writelocked each time a socket must be inserted in
hash table (bind time), or deleted from this table (close time)
This is not scalable on SMP machines :
1) Even in read mode, lock() and unlock() are atomic operations and
must dirty a contended cache line, shared by all cpus.
2) A writer might be starved if many readers are 'in flight'. This can
happen on a machine with some NIC receiving many UDP messages. User
process can be delayed a long time at socket creation/dismantle time.
This patch prepares RCU migration, by introducing 'struct udp_table
and struct udp_hslot', and using one spinlock per chain, to reduce
contention on central rwlock.
Introducing one spinlock per chain reduces latencies, for port
randomization on heavily loaded UDP servers. This also speedup
bindings to specific ports.
udp_lib_unhash() was uninlined, becoming to big.
Some cleanups were done to ease review of following patch
(RCUification of UDP Unicast lookups)
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Wrap calling sk->sk_backlog_rcv() in a function. This will allow extending the
generic sk_backlog_rcv behaviour.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the socket cached in the skb if it's present.
Signed-off-by: KOVACS Krisztian <hidden@sch.bme.hu>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
All users of struct proto::compat_[gs]etsockopt and
struct inet_connection_sock_af_ops::compat_[gs]etsockopt are under
#ifdef already, so use it in structure definition too.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The tcp_enter_memory_pressure calls NET_INC_STATS, but doesn't
have where to get the net from.
I decided to add a sk argument, not the net itself, only to factor
all the required sock_net(sk) calls inside the enter_memory_pressure
callback itself.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to more easily grep for all things that set
sk->sk_socket, add sk_set_socket() helper inline function.
Suggested (although only half-seriously) by Evgeniy Polyakov.
Signed-off-by: David S. Miller <davem@davemloft.net>
In commits 33c732c361 ([IPV4]: Add raw
drops counter) and a92aa318b4 ([IPV6]:
Add raw drops counter), Wang Chen added raw drops counter for
/proc/net/raw & /proc/net/raw6
This patch adds this capability to UDP sockets too (/proc/net/udp &
/proc/net/udp6).
This means that 'RcvbufErrors' errors found in /proc/net/snmp can be also
be examined for each udp socket.
# grep Udp: /proc/net/snmp
Udp: InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors
Udp: 23971006 75 899420 16390693 146348 0
# cat /proc/net/udp
sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt ---
uid timeout inode ref pointer drops
75: 00000000:02CB 00000000:0000 07 00000000:00000000 00:00000000 00000000 ---
0 0 2358 2 ffff81082a538c80 0
111: 00000000:006F 00000000:0000 07 00000000:00000000 00:00000000 00000000 ---
0 0 2286 2 ffff81042dd35c80 146348
In this example, only port 111 (0x006F) was flooded by messages that
user program could not read fast enough. 146348 messages were lost.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
