This change extends the fast SO_REUSEPORT socket lookup implemented
for UDP to TCP. Listener sockets with SO_REUSEPORT and the same
receive address are additionally added to an array for faster
random access. This means that only a single socket from the group
must be found in the listener list before any socket in the group can
be used to receive a packet. Previously, every socket in the group
needed to be considered before handing off the incoming packet.
This feature also exposes the ability to use a BPF program when
selecting a socket from a reuseport group.
Signed-off-by: Craig Gallek <kraig@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to support fast reuseport lookups in TCP, the hash function
defined in struct proto must be capable of returning an error code.
This patch changes the function signature of all related hash functions
to return an integer and handles or propagates this return value at
all call sites.
Signed-off-by: Craig Gallek <kraig@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some functions access TCP sockets without holding a lock and
might output non consistent data, depending on compiler and or
architecture.
tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
Introduce sk_state_load() and sk_state_store() to fix the issues,
and more clearly document where this lack of locking is happening.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Multiple cpus can process duplicates of incoming ACK messages
matching a SYN_RECV request socket. This is a rare event under
normal operations, but definitely can happen.
Only one must win the race, otherwise corruption would occur.
To fix this without adding new atomic ops, we use logic in
inet_ehash_nolisten() to detect the request was present in the same
ehash bucket where we try to insert the new child.
If request socket was not found, we have to undo the child creation.
This actually removes a spin_lock()/spin_unlock() pair in
reqsk_queue_unlink() for the fast path.
Fixes: e994b2f0fb ("tcp: do not lock listener to process SYN packets")
Fixes: 079096f103 ("tcp/dccp: install syn_recv requests into ehash table")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Under stress, a close() on a listener can trigger the
WARN_ON(sk->sk_ack_backlog) in inet_csk_listen_stop()
We need to test if listener is still active before queueing
a child in inet_csk_reqsk_queue_add()
Create a common inet_child_forget() helper, and use it
from inet_csk_reqsk_queue_add() and inet_csk_listen_stop()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Let's reduce the confusion about inet_csk_reqsk_queue_drop() :
In many cases we also need to release reference on request socket,
so add a helper to do this, reducing code size and complexity.
Fixes: 4bdc3d6614 ("tcp/dccp: fix behavior of stale SYN_RECV request sockets")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
At listen() time, there is a small window where listener is visible with
a zero backlog, triggering a spurious "Possible SYN flooding on port"
message.
Nothing prevents us from setting the correct backlog.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some applications use a listen() backlog of 1.
Prior kernels were silently enforcing a qlen_log of 4, so that we were
sending up to /proc/sys/net/ipv4/tcp_synack_retries SYNACK messages.
Fixes: ef547f2ac1 ("tcp: remove max_qlen_log")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If a listener with thousands of children in accept queue
is dismantled, it can take a while to close all of them.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This control variable was set at first listen(fd, backlog)
call, but not updated if application tried to increase or decrease
backlog. It made sense at the time listener had a non resizeable
hash table.
Also rounding to powers of two was not very friendly.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is enough to check listener sk_state, no need for an extra
condition.
max_qlen_log can be moved into struct request_sock_queue
We can remove syn_wait_lock and the alignment it enforced.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If a listen backlog is very big (to avoid syncookies), then
the listener sk->sk_wmem_alloc is the main source of false
sharing, as we need to touch it twice per SYNACK re-transmit
and TX completion.
(One SYN packet takes listener lock once, but up to 6 SYNACK
are generated)
By attaching the skb to the request socket, we remove this
source of contention.
Tested:
listen(fd, 10485760); // single listener (no SO_REUSEPORT)
16 RX/TX queue NIC
Sustain a SYNFLOOD attack of ~320,000 SYN per second,
Sending ~1,400,000 SYNACK per second.
Perf profiles now show listener spinlock being next bottleneck.
20.29% [kernel] [k] queued_spin_lock_slowpath
10.06% [kernel] [k] __inet_lookup_established
5.12% [kernel] [k] reqsk_timer_handler
3.22% [kernel] [k] get_next_timer_interrupt
3.00% [kernel] [k] tcp_make_synack
2.77% [kernel] [k] ipt_do_table
2.70% [kernel] [k] run_timer_softirq
2.50% [kernel] [k] ip_finish_output
2.04% [kernel] [k] cascade
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In this patch, we insert request sockets into TCP/DCCP
regular ehash table (where ESTABLISHED and TIMEWAIT sockets
are) instead of using the per listener hash table.
ACK packets find SYN_RECV pseudo sockets without having
to find and lock the listener.
In nominal conditions, this halves pressure on listener lock.
Note that this will allow for SO_REUSEPORT refinements,
so that we can select a listener using cpu/numa affinities instead
of the prior 'consistent hash', since only SYN packets will
apply this selection logic.
We will shrink listen_sock in the following patch to ease
code review.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ying Cai <ycai@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
qlen_inc & young_inc were protected by listener lock,
while qlen_dec & young_dec were atomic fields.
Everything needs to be atomic for upcoming lockless listener.
Also move qlen/young in request_sock_queue as we'll get rid
of struct listen_sock eventually.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
struct request_sock_queue fields are currently protected
by the listener 'lock' (not a real spinlock)
We need to add a private spinlock instead, so that softirq handlers
creating children do not have to worry with backlog notion
that the listener 'lock' carries.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While auditing TCP stack for upcoming 'lockless' listener changes,
I found I had to change fastopen_init_queue() to properly init the object
before publishing it.
Otherwise an other cpu could try to lock the spinlock before it gets
properly initialized.
Instead of adding appropriate barriers, just remove dynamic memory
allocations :
- Structure is 28 bytes on 64bit arches. Using additional 8 bytes
for holding a pointer seems overkill.
- Two listeners can share same cache line and performance would suffer.
If we really want to save few bytes, we would instead dynamically allocate
whole struct request_sock_queue in the future.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
net/ipv4/arp.c
The net/ipv4/arp.c conflict was one commit adding a new
local variable while another commit was deleting one.
Signed-off-by: David S. Miller <davem@davemloft.net>
SYNACK packets are sent on behalf on unlocked listeners
or fastopen sockets. Mark socket as const to catch future changes
that might break the assumption.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is used by TCP listener core, and listener socket shall
not be modified by inet_csk_route_req().
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Before allowing lockless LISTEN processing, we need to make
sure to arm the SYN_RECV timer before the req socket is visible
in hash tables.
Also, req->rsk_hash should be written before we set rsk_refcnt
to a non zero value.
Fixes: fa76ce7328 ("inet: get rid of central tcp/dccp listener timer")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ying Cai <ycai@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When replacing del_timer() with del_timer_sync(), I introduced
a deadlock condition :
reqsk_queue_unlink() is called from inet_csk_reqsk_queue_drop()
inet_csk_reqsk_queue_drop() can be called from many contexts,
one being the timer handler itself (reqsk_timer_handler()).
In this case, del_timer_sync() loops forever.
Simple fix is to test if timer is pending.
Fixes: 2235f2ac75 ("inet: fix races with reqsk timers")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
reqsk_queue_destroy() and reqsk_queue_unlink() should use
del_timer_sync() instead of del_timer() before calling reqsk_put(),
otherwise we could free a req still used by another cpu.
But before doing so, reqsk_queue_destroy() must release syn_wait_lock
spinlock or risk a dead lock, as reqsk_timer_handler() might
need to take this same spinlock from reqsk_queue_unlink() (called from
inet_csk_reqsk_queue_drop())
Fixes: fa76ce7328 ("inet: get rid of central tcp/dccp listener timer")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
inet_csk_get_port() randomization effort tends to spread
sockets on all the available range (ip_local_port_range)
This is unfortunate because SO_REUSEADDR sockets have
less requirements than non SO_REUSEADDR ones.
If an application uses SO_REUSEADDR hint, it is to try to
allow source ports being shared.
So instead of picking a random port number in ip_local_port_range,
lets try first in first half of the range.
This gives more chances to use upper half of the range for the
sockets with strong requirements (not using SO_REUSEADDR)
Note this patch does not add a new sysctl, and only changes
the way we try to pick port number.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Marcelo Ricardo Leitner <mleitner@redhat.com>
Cc: Flavio Leitner <fbl@redhat.com>
Acked-by: Flavio Leitner <fbl@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We no longer need bsocket atomic counter, as inet_csk_get_port()
calls bind_conflict() regardless of its value, after commit
2b05ad33e1 ("tcp: bind() fix autoselection to share ports")
This patch removes overhead of maintaining this counter and
double inet_csk_get_port() calls under pressure.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Marcelo Ricardo Leitner <mleitner@redhat.com>
Cc: Flavio Leitner <fbl@redhat.com>
Acked-by: Flavio Leitner <fbl@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
[ 3897.923145] BUG: unable to handle kernel NULL pointer dereference at
0000000000000080
[ 3897.931025] IP: [<ffffffffa9f27686>] reqsk_timer_handler+0x1a6/0x243
There is a race when reqsk_timer_handler() and tcp_check_req() call
inet_csk_reqsk_queue_unlink() on the same req at the same time.
Before commit fa76ce7328 ("inet: get rid of central tcp/dccp listener
timer"), listener spinlock was held and race could not happen.
To solve this bug, we change reqsk_queue_unlink() to not assume req
must be found, and we return a status, to conditionally release a
refcount on the request sock.
This also means tcp_check_req() in non fastopen case might or not
consume req refcount, so tcp_v6_hnd_req() & tcp_v4_hnd_req() have
to properly handle this.
(Same remark for dccp_check_req() and its callers)
inet_csk_reqsk_queue_drop() is now too big to be inlined, as it is
called 4 times in tcp and 3 times in dccp.
Fixes: fa76ce7328 ("inet: get rid of central tcp/dccp listener timer")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ipv4 code uses a mixture of coding styles. In some instances check
for non-NULL pointer is done as x != NULL and sometimes as x. x is
preferred according to checkpatch and this patch makes the code
consistent by adopting the latter form.
No changes detected by objdiff.
Signed-off-by: Ian Morris <ipm@chirality.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is a low hanging fruit, as we'll get rid of syn_wait_lock eventually.
We hold syn_wait_lock for such small sections, that it makes no sense to use
a read/write lock. A spin lock is simply faster.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
listener can be source of false sharing. request sock has some
useful information like : ireq->ir_iif, ireq->ir_num, ireq->ireq_net
This patch does not solve the major problem of having to read
sk->sk_protocol which is sharing a cache line with sk->sk_wmem_alloc.
(This same field is read later in ip_build_and_send_pkt())
One idea would be to move sk_protocol close to sk_family
(using 8 bits instead of 16 for sk_family seems enough)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is not needed, and req->sk_listener points to the listener anyway.
request_sock argument can be const.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Cache listen_sock_qlen() to limit false sharing, and read
rskq_defer_accept once as it might change under us.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/emulex/benet/be_main.c
net/core/sysctl_net_core.c
net/ipv4/inet_diag.c
The be_main.c conflict resolution was really tricky. The conflict
hunks generated by GIT were very unhelpful, to say the least. It
split functions in half and moved them around, when the real actual
conflict only existed solely inside of one function, that being
be_map_pci_bars().
So instead, to resolve this, I checked out be_main.c from the top
of net-next, then I applied the be_main.c changes from 'net' since
the last time I merged. And this worked beautifully.
The inet_diag.c and sysctl_net_core.c conflicts were simple
overlapping changes, and were easily to resolve.
Signed-off-by: David S. Miller <davem@davemloft.net>
One of the major issue for TCP is the SYNACK rtx handling,
done by inet_csk_reqsk_queue_prune(), fired by the keepalive
timer of a TCP_LISTEN socket.
This function runs for awful long times, with socket lock held,
meaning that other cpus needing this lock have to spin for hundred of ms.
SYNACK are sent in huge bursts, likely to cause severe drops anyway.
This model was OK 15 years ago when memory was very tight.
We now can afford to have a timer per request sock.
Timer invocations no longer need to lock the listener,
and can be run from all cpus in parallel.
With following patch increasing somaxconn width to 32 bits,
I tested a listener with more than 4 million active request sockets,
and a steady SYNFLOOD of ~200,000 SYN per second.
Host was sending ~830,000 SYNACK per second.
This is ~100 times more what we could achieve before this patch.
Later, we will get rid of the listener hash and use ehash instead.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When request sock are put in ehash table, the whole notion
of having a previous request to update dl_next is pointless.
Also, following patch will get rid of big purge timer,
so we want to delete a request sock without holding listener lock.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is not because a TCP listener is FastOpen ready that
all incoming sockets actually used FastOpen.
Avoid taking queue->fastopenq->lock if not needed.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The listener field in struct tcp_request_sock is a pointer
back to the listener. We now have req->rsk_listener, so TCP
only needs one boolean and not a full pointer.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
reqsk_put() is the generic function that should be used
to release a refcount (and automatically call reqsk_free())
reqsk_free() might be called if refcount is known to be 0
or undefined.
refcnt is set to one in inet_csk_reqsk_queue_add()
As request socks are not yet in global ehash table,
I added temporary debugging checks in reqsk_put() and reqsk_free()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A long standing problem in netlink socket dumps is the use
of kernel socket addresses as cookies.
1) It is a security concern.
2) Sockets can be reused quite quickly, so there is
no guarantee a cookie is used once and identify
a flow.
3) request sock, establish sock, and timewait socks
for a given flow have different cookies.
Part of our effort to bring better TCP statistics requires
to switch to a different allocator.
In this patch, I chose to use a per network namespace 64bit generator,
and to use it only in the case a socket needs to be dumped to netlink.
(This might be refined later if needed)
Note that I tried to carry cookies from request sock, to establish sock,
then timewait sockets.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Eric Salo <salo@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ip_local_port_range is already per netns, so should ip_local_reserved_ports
be. And since it is none by default we don't actually need it when we don't
enable CONFIG_SYSCTL.
By the way, rename inet_is_reserved_local_port() to inet_is_local_reserved_port()
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When using mark-based routing, sockets returned from accept()
may need to be marked differently depending on the incoming
connection request.
This is the case, for example, if different socket marks identify
different networks: a listening socket may want to accept
connections from all networks, but each connection should be
marked with the network that the request came in on, so that
subsequent packets are sent on the correct network.
This patch adds a sysctl to mark TCP sockets based on the fwmark
of the incoming SYN packet. If enabled, and an unmarked socket
receives a SYN, then the SYN packet's fwmark is written to the
connection's inet_request_sock, and later written back to the
accepted socket when the connection is established. If the
socket already has a nonzero mark, then the behaviour is the same
as it is today, i.e., the listening socket's fwmark is used.
Black-box tested using user-mode linux:
- IPv4/IPv6 SYN+ACK, FIN, etc. packets are routed based on the
mark of the incoming SYN packet.
- The socket returned by accept() is marked with the mark of the
incoming SYN packet.
- Tested with syncookies=1 and syncookies=2.
Signed-off-by: Lorenzo Colitti <lorenzo@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When CONFIG_SYSCTL is not set, ip_local_port_range should still work,
just that no one can change it. Therefore we should move it out of sysctl_inet.c.
Also, rename it to ->ip_local_ports instead.
Cc: David S. Miller <davem@davemloft.net>
Cc: Francois Romieu <romieu@fr.zoreil.com>
Reported-by: Stefan de Konink <stefan@konink.de>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes the net_random and net_srandom macros and replaces
them with direct calls to the prandom ones. As new commits only seem to
use prandom_u32 there is no use to keep them around.
This change makes it easier to grep for users of prandom_u32.
Signed-off-by: Aruna-Hewapathirane <aruna.hewapathirane@gmail.com>
Suggested-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 634fb979e8 ("inet: includes a sock_common in request_sock")
I forgot that the two ports in sock_common do not have same byte order :
skc_dport is __be16 (network order), but skc_num is __u16 (host order)
So sparse complains because ir_loc_port (mapped into skc_num) is
considered as __u16 while it should be __be16
Let rename ir_loc_port to ireq->ir_num (analogy with inet->inet_num),
and perform appropriate htons/ntohs conversions.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP listener refactoring, part 5 :
We want to be able to insert request sockets (SYN_RECV) into main
ehash table instead of the per listener hash table to allow RCU
lookups and remove listener lock contention.
This patch includes the needed struct sock_common in front
of struct request_sock
This means there is no more inet6_request_sock IPv6 specific
structure.
Following inet_request_sock fields were renamed as they became
macros to reference fields from struct sock_common.
Prefix ir_ was chosen to avoid name collisions.
loc_port -> ir_loc_port
loc_addr -> ir_loc_addr
rmt_addr -> ir_rmt_addr
rmt_port -> ir_rmt_port
iif -> ir_iif
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP listener refactoring, part 2 :
We can use a generic lookup, sockets being in whatever state, if
we are sure all relevant fields are at the same place in all socket
types (ESTABLISH, TIME_WAIT, SYN_RECV)
This patch removes these macros :
inet_addrpair, inet_addrpair, tw_addrpair, tw_portpair
And adds :
sk_portpair, sk_addrpair, sk_daddr, sk_rcv_saddr
Then, INET_TW_MATCH() is really the same than INET_MATCH()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
- Move sysctl_local_ports from a global variable into struct netns_ipv4.
- Modify inet_get_local_port_range to take a struct net, and update all
of the callers.
- Move the initialization of sysctl_local_ports into
sysctl_net_ipv4.c:ipv4_sysctl_init_net from inet_connection_sock.c
v2:
- Ensure indentation used tabs
- Fixed ip.h so it applies cleanly to todays net-next
v3:
- Compile fixes of strange callers of inet_get_local_port_range.
This patch now successfully passes an allmodconfig build.
Removed manual inlining of inet_get_local_port_range in ipv4_local_port_range
Originally-by: Samya <samya@twitter.com>
Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCPCT uses option-number 253, reserved for experimental use and should
not be used in production environments.
Further, TCPCT does not fully implement RFC 6013.
As a nice side-effect, removing TCPCT increases TCP's performance for
very short flows:
Doing an apache-benchmark with -c 100 -n 100000, sending HTTP-requests
for files of 1KB size.
before this patch:
average (among 7 runs) of 20845.5 Requests/Second
after:
average (among 7 runs) of 21403.6 Requests/Second
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
Craig Gallek