Allow one-packet scheduling for UDP connections. When the fwmark-based or
normal virtual service is marked with '-o' or '--ops' options all
connections are created only to schedule one packet. Useful to schedule UDP
packets from same client port to different real servers. Recommended with
RR or WRR schedulers (the connections are not visible with ipvsadm -L).
Signed-off-by: Nick Chalk <nick@loadbalancer.org>
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
remove useless union keyword in rtable, rt6_info and dn_route.
Since there is only one member in a union, the union keyword isn't useful.
Signed-off-by: Changli Gao <xiaosuo@gmail.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The code that hashes and unhashes connections from the connection table
is missing locking of the connection being modified, which opens up a
race condition and results in memory corruption when this race condition
is hit.
Here is what happens in pretty verbose form:
CPU 0 CPU 1
------------ ------------
An active connection is terminated and
we schedule ip_vs_conn_expire() on this
CPU to expire this connection.
IRQ assignment is changed to this CPU,
but the expire timer stays scheduled on
the other CPU.
New connection from same ip:port comes
in right before the timer expires, we
find the inactive connection in our
connection table and get a reference to
it. We proper lock the connection in
tcp_state_transition() and read the
connection flags in set_tcp_state().
ip_vs_conn_expire() gets called, we
unhash the connection from our
connection table and remove the hashed
flag in ip_vs_conn_unhash(), without
proper locking!
While still holding proper locks we
write the connection flags in
set_tcp_state() and this sets the hashed
flag again.
ip_vs_conn_expire() fails to expire the
connection, because the other CPU has
incremented the reference count. We try
to re-insert the connection into our
connection table, but this fails in
ip_vs_conn_hash(), because the hashed
flag has been set by the other CPU. We
re-schedule execution of
ip_vs_conn_expire(). Now this connection
has the hashed flag set, but isn't
actually hashed in our connection table
and has a dangling list_head.
We drop the reference we held on the
connection and schedule the expire timer
for timeouting the connection on this
CPU. Further packets won't be able to
find this connection in our connection
table.
ip_vs_conn_expire() gets called again,
we think it's already hashed, but the
list_head is dangling and while removing
the connection from our connection table
we write to the memory location where
this list_head points to.
The result will probably be a kernel oops at some other point in time.
This race condition is pretty subtle, but it can be triggered remotely.
It needs the IRQ assignment change or another circumstance where packets
coming from the same ip:port for the same service are being processed on
different CPUs. And it involves hitting the exact time at which
ip_vs_conn_expire() gets called. It can be avoided by making sure that
all packets from one connection are always processed on the same CPU and
can be made harder to exploit by changing the connection timeouts to
some custom values.
Signed-off-by: Sven Wegener <sven.wegener@stealer.net>
Cc: stable@kernel.org
Acked-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Define a new function to return the waitqueue of a "struct sock".
static inline wait_queue_head_t *sk_sleep(struct sock *sk)
{
return sk->sk_sleep;
}
Change all read occurrences of sk_sleep by a call to this function.
Needed for a future RCU conversion. sk_sleep wont be a field directly
available.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When protocols use very long names, the sprintf calls might overflow
the on-stack buffer. No protocol in the kernel does this however.
Print the protocol name in the pr_debug statement directly to avoid
this.
Based on patch by Zhitong Wang <zhitong.wangzt@alibaba-inc.com>
Acked-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
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>
NIPQUAD has very few uses left.
Remove this use and make the code have the identical form of the only
other use of "%u,%u,%u,%u,%u,%u" in net/ipv4/netfilter/nf_nat_ftp.c
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Use list_head rather than a custom list implementation.
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Dunno, what was the idea, it wasn't used for a long time.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Enhance IPVS to load balance SCTP transport protocol packets. This is done
based on the SCTP rfc 4960. All possible control chunks have been taken
care. The state machine used in this code looks some what lengthy. I tried
to make the state machine easy to understand.
Signed-off-by: Venkata Mohan Reddy Koppula <mohanreddykv@gmail.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Use the same format string as net/ipv4/netfilter/nf_nat_ftp.c
to encode an ipv4 address and port.
Both uses should be a single common function.
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
I was very frustrated about the fact that I have to recompile the kernel
to change the hash size. So, I created this patch.
If IPVS is built-in you can append ip_vs.conn_tab_bits=?? to kernel
command line, or, if you built IPVS as modules, you can add
options ip_vs conn_tab_bits=??.
To keep everything backward compatible, you still can select the size at
compile time, and that will be used as default.
It has been about a year since this patch was originally posted
and subsequently dropped on the basis of insufficient test data.
Mark Bergsma has provided the following test results which seem
to strongly support the need for larger hash table sizes:
We do however run into the same problem with the default setting (212 =
4096 entries), as most of our LVS balancers handle around a million
connections/SLAB entries at any point in time (around 100-150 kpps
load). With only 4096 hash table entries this implies that each entry
consists of a linked list of 256 connections *on average*.
To provide some statistics, I did an oprofile run on an 2.6.31 kernel,
with both the default 4096 table size, and the same kernel recompiled
with IP_VS_CONN_TAB_BITS set to 18 (218 = 262144 entries). I built a
quick test setup with a part of Wikimedia/Wikipedia's live traffic
mirrored by the switch to the test host.
With the default setting, at ~ 120 kpps packet load we saw a typical %si
CPU usage of around 30-35%, and oprofile reported a hot spot in
ip_vs_conn_in_get:
samples % image name app name
symbol name
1719761 42.3741 ip_vs.ko ip_vs.ko ip_vs_conn_in_get
302577 7.4554 bnx2 bnx2 /bnx2
181984 4.4840 vmlinux vmlinux __ticket_spin_lock
128636 3.1695 vmlinux vmlinux ip_route_input
74345 1.8318 ip_vs.ko ip_vs.ko ip_vs_conn_out_get
68482 1.6874 vmlinux vmlinux mwait_idle
After loading the recompiled kernel with 218 entries, %si CPU usage
dropped in half to around 12-18%, and oprofile looks much healthier,
with only 7% spent in ip_vs_conn_in_get:
samples % image name app name
symbol name
265641 14.4616 bnx2 bnx2 /bnx2
143251 7.7986 vmlinux vmlinux __ticket_spin_lock
140661 7.6576 ip_vs.ko ip_vs.ko ip_vs_conn_in_get
94364 5.1372 vmlinux vmlinux mwait_idle
86267 4.6964 vmlinux vmlinux ip_route_input
[ horms@verge.net.au: trivial up-port and minor style fixes ]
Signed-off-by: Catalin(ux) M. BOIE <catab@embedromix.ro>
Cc: Mark Bergsma <mark@wikimedia.org>
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
The ipvs code has a nifty system for doing the size of ioctl command
copies; it defines an array with values into which it indexes the cmd
to find the right length.
Unfortunately, the ipvs code forgot to check if the cmd was in the
range that the array provides, allowing for an index outside of the
array, which then gives a "garbage" result into the length, which
then gets used for copying into a stack buffer.
Fix this by adding sanity checks on these as well as the copy size.
[ horms@verge.net.au: adjusted limit to IP_VS_SO_GET_MAX ]
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Acked-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Remove the private version of the greatest common divider to use
lib/gcd.c, the latter also implementing the a < b case.
[akpm@linux-foundation.org: repair neighboring whitespace because the diff looked odd]
Signed-off-by: Florian Fainelli <florian@openwrt.org>
Cc: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Cc: Takashi Iwai <tiwai@suse.de>
Acked-by: Simon Horman <horms@verge.net.au>
Cc: Julius Volz <juliusv@google.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Make sure that any otherwise uninitialised fields of usvc are zero.
This has been obvserved to cause a problem whereby the port of
fwmark services may end up as a non-zero value which causes
scheduling of a destination server to fail for persisitent services.
As observed by Deon van der Merwe <dvdm@truteq.co.za>.
This fix suggested by Julian Anastasov <ja@ssi.bg>.
For good measure also zero udest.
Cc: Deon van der Merwe <dvdm@truteq.co.za>
Acked-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
Cc: stable@kernel.org
Signed-off-by: Patrick McHardy <kaber@trash.net>
commit 9d3a0de makes slaves expire as they would do on the master
with much shorter timeouts. But it introduces another problem:
When we close a connection, on master server the connection became
CLOSE_WAIT/TIME_WAIT, it was synced to slaves, but if master is
finished within it's timeouts (CLOSE), it will not be synced to
slaves. Then slaves will be kept on CLOSE_WAIT/TIME_WAIT until
timeout reaches. Thus we should also sync with CLOSE.
Cc: Wensong Zhang <wensong@linux-vs.org>
Cc: Simon Horman <horms@verge.net.au>
Cc: Julian Anastasov <ja@ssi.bg>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Xiaotian Feng <dfeng@redhat.com>
Acked-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Now that sys_sysctl is a compatiblity wrapper around /proc/sys
all sysctl strategy routines, and all ctl_name and strategy
entries in the sysctl tables are unused, and can be
revmoed.
In addition neigh_sysctl_register has been modified to no longer
take a strategy argument and it's callers have been modified not
to pass one.
Cc: "David Miller" <davem@davemloft.net>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Cc: netdev@vger.kernel.org
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
David S. Miller
Nick Chalk
Changli Gao
Sven Wegener
Patrick McHardy
Eric Dumazet
Tejun Heo
Jan Engelhardt
Joe Perches
Simon Horman
Alexey Dobriyan
Venkata Mohan Reddy
Catalin(ux) M. BOIE
Arjan van de Ven
Florian Fainelli
Xiaotian Feng
Eric W. Biederman