Frank Kellermann reported a kernel crash with 4.5.0 when IPv6 is
disabled at boot using the kernel option ipv6.disable=1. Using
current net-next with the boot option:
$ ip link add red type vrf table 1001
Generates:
[12210.919584] BUG: unable to handle kernel NULL pointer dereference at 0000000000000748
[12210.921341] IP: [<ffffffff814b30e3>] fib6_get_table+0x2c/0x5a
[12210.922537] PGD b79e3067 PUD bb32b067 PMD 0
[12210.923479] Oops: 0000 [#1] SMP
[12210.924001] Modules linked in: ipvlan 8021q garp mrp stp llc
[12210.925130] CPU: 3 PID: 1177 Comm: ip Not tainted 4.7.0-rc1+ #235
[12210.926168] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140531_083030-gandalf 04/01/2014
[12210.928065] task: ffff8800b9ac4640 ti: ffff8800bacac000 task.ti: ffff8800bacac000
[12210.929328] RIP: 0010:[<ffffffff814b30e3>] [<ffffffff814b30e3>] fib6_get_table+0x2c/0x5a
[12210.930697] RSP: 0018:ffff8800bacaf888 EFLAGS: 00010202
[12210.931563] RAX: 0000000000000748 RBX: ffffffff81a9e280 RCX: ffff8800b9ac4e28
[12210.932688] RDX: 00000000000000e9 RSI: 0000000000000002 RDI: 0000000000000286
[12210.933820] RBP: ffff8800bacaf898 R08: ffff8800b9ac4df0 R09: 000000000052001b
[12210.934941] R10: 00000000657c0000 R11: 000000000000c649 R12: 00000000000003e9
[12210.936032] R13: 00000000000003e9 R14: ffff8800bace7800 R15: ffff8800bb3ec000
[12210.937103] FS: 00007faa1766c700(0000) GS:ffff88013ac00000(0000) knlGS:0000000000000000
[12210.938321] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[12210.939166] CR2: 0000000000000748 CR3: 00000000b79d6000 CR4: 00000000000406e0
[12210.940278] Stack:
[12210.940603] ffff8800bb3ec000 ffffffff81a9e280 ffff8800bacaf8c8 ffffffff814b3135
[12210.941818] ffff8800bb3ec000 ffffffff81a9e280 ffffffff81a9e280 ffff8800bace7800
[12210.943040] ffff8800bacaf8f0 ffffffff81397c88 ffff8800bb3ec000 ffffffff81a9e280
[12210.944288] Call Trace:
[12210.944688] [<ffffffff814b3135>] fib6_new_table+0x24/0x8a
[12210.945516] [<ffffffff81397c88>] vrf_dev_init+0xd4/0x162
[12210.946328] [<ffffffff814091e1>] register_netdevice+0x100/0x396
[12210.947209] [<ffffffff8139823d>] vrf_newlink+0x40/0xb3
[12210.948001] [<ffffffff814187f0>] rtnl_newlink+0x5d3/0x6d5
...
The problem above is due to the fact that the fib hash table is not
allocated when IPv6 is disabled at boot.
As for the VRF driver it should not do any IPv6 initializations if IPv6
is disabled, so it needs to know if IPv6 is disabled at boot. The disable
parameter is private to the IPv6 module, so provide an accessor for
modules to determine if IPv6 was disabled at boot time.
Fixes: 35402e3136 ("net: Add IPv6 support to VRF device")
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Simplify the RxRPC connect() implementation. It will just note the
destination address it is given, and if a sendmsg() comes along with no
address, this will be assigned as the address. No transport struct will be
held internally, which will allow us to remove this later.
Simplify sendmsg() also. Whilst a call is active, userspace refers to it
by a private unique user ID specified in a control message. When sendmsg()
sees a user ID that doesn't map to an extant call, it creates a new call
for that user ID and attempts to add it. If, when we try to add it, the
user ID is now registered, we now reject the message with -EEXIST. We
should never see this situation unless two threads are racing, trying to
create a call with the same ID - which would be an error.
It also isn't required to provide sendmsg() with an address - provided the
control message data holds a user ID that maps to a currently active call.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is time to add netdev_lockdep_set_classes() helper
so that lockdep annotations per device type are easier to manage.
This removes a lot of copies and missing annotations.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
1) qdisc_run_begin() is really using the equivalent of a trylock.
Instead of using write_seqcount_begin(), use a combination of
raw_write_seqcount_begin() and correct lockdep annotation.
2) sch_direct_xmit() should use regular spin_lock(root_lock)
Fixes: f9eb8aea2a ("net_sched: transform qdisc running bit into a seqcount")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ipip6_tunnel_xmit() is called immediately after checking that
skb->protocol is htons(ETH_P_IPV6) so there is no need
to check it a second time.
Found by inspection.
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Dinan Gunawardena <dinan.gunawardena@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
after removal of TCA_CBQ_OVL_STRATEGY from cbq scheduler, there are no
more callers of ->drop() outside of other ->drop functions, i.e.
nothing calls them.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
After the removal of TCA_CBQ_POLICE in cbq scheduler qdisc->reshape_fail
is always NULL, i.e. qdisc_rehape_fail is now the same as qdisc_drop.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
iproute2 doesn't implement any cbq option that results in this attribute
being sent to kernel.
To make use of it, user would have to
- patch iproute2
- add a class
- attach a qdisc to the class (default pfifo doesn't work as
q->handle is 0 and cbq_set_police() is a no-op in this case)
- re-'add' the same class (tc class change ...) again
- user must also specifiy a defmap (e.g. 'split 1:0 defmap 3f'), since
this 'police' feature relies on its presence
- the added qdisc must be one of bfifo, pfifo or netem
If all of these conditions are met and _some_ leaf qdiscs, namely
p/bfifo, netem, plug or tbf would drop a packet, kernel calls back into
cbq, which will attempt to re-queue the skb into a different class
as indicated by the parents' defmap entry for TC_PRIO_BESTEFFORT.
[ i.e. we behave as if tc_classify returned TC_ACT_RECLASSIFY ].
This feature, which isn't documented or implemented in iproute2,
and isn't implemented consistently (most qdiscs like sfq, codel, etc
drop right away instead of attempting this reclassification) is the
sole reason for the reshape_fail and __parent member in Qdisc struct.
So remove TCA_CBQ_POLICE support from the kernel, reject it via EOPNOTSUPP
so userspace knows we don't support it, and then remove no-longer needed
infrastructure in followup commit.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
since initial revision of cbq in 2004 iproute 2 has never implemented
support for TCA_CBQ_OVL_STRATEGY, which is what needs to be set to
activate the class->drop() call (TC_CBQ_OVL_DROP strategy must be
set by userspace value must be set by userspace).
David Miller says:
It seems really safe to kill this thing off, flag an error if someone
tries to set the attribute, and therefore kill off all of the
non-default cbq_ovl_*() functions.
A followup commit can then remove all .drop qdisc methods since this
removed the only caller.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ip6 GRE tap device should not be forced to down state to change
the mac address and should allow live address change for tap device
similar to ipv4 gre.
Signed-off-by: Shweta Choudaha <schoudah@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, VRFs require 1 oif and 1 iif rule per address family per
VRF. As the number of VRF devices increases it brings scalability
issues with the increasing rule list. All of the VRF rules have the
same format with the exception of the specific table id to direct the
lookup. Since the table id is available from the oif or iif in the
loopup, the VRF rules can be consolidated to a single rule that pulls
the table from the VRF device.
This patch introduces a new rule attribute l3mdev. The l3mdev rule
means the table id used for the lookup is pulled from the L3 master
device (e.g., VRF) rather than being statically defined. With the
l3mdev rule all of the basic VRF FIB rules are reduced to 1 l3mdev
rule per address family (IPv4 and IPv6).
If an admin wishes to insert higher priority rules for specific VRFs
those rules will co-exist with the l3mdev rule. This capability means
current VRF scripts will co-exist with this new simpler implementation.
Currently, the rules list for both ipv4 and ipv6 look like this:
$ ip ru ls
1000: from all oif vrf1 lookup 1001
1000: from all iif vrf1 lookup 1001
1000: from all oif vrf2 lookup 1002
1000: from all iif vrf2 lookup 1002
1000: from all oif vrf3 lookup 1003
1000: from all iif vrf3 lookup 1003
1000: from all oif vrf4 lookup 1004
1000: from all iif vrf4 lookup 1004
1000: from all oif vrf5 lookup 1005
1000: from all iif vrf5 lookup 1005
1000: from all oif vrf6 lookup 1006
1000: from all iif vrf6 lookup 1006
1000: from all oif vrf7 lookup 1007
1000: from all iif vrf7 lookup 1007
1000: from all oif vrf8 lookup 1008
1000: from all iif vrf8 lookup 1008
...
32765: from all lookup local
32766: from all lookup main
32767: from all lookup default
With the l3mdev rule the list is just the following regardless of the
number of VRFs:
$ ip ru ls
1000: from all lookup [l3mdev table]
32765: from all lookup local
32766: from all lookup main
32767: from all lookup default
(Note: the above pretty print of the rule is based on an iproute2
prototype. Actual verbage may change)
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The node keepalive interval is recalculated at each timer expiration
to catch any changes in the link tolerance, and stored in a field in
struct tipc_node. We use jiffies as unit for the stored value.
This is suboptimal, because it makes the calculation unnecessary
complex, including two unit conversions. The conversions also lead to
a rounding error that causes the link "abort limit" to be 3 in the
normal case, instead of 4, as intended. This again leads to unnecessary
link resets when the network is pushed close to its limit, e.g., in an
environment with hundreds of nodes or namesapces.
In this commit, we do instead let the keepalive value be calculated and
stored in milliseconds, so that there is only one conversion and the
rounding error is eliminated.
We also remove a redundant "keepalive" field in struct tipc_link. This
is remnant from the previous implementation.
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
commit 88e8ac7000 ("tipc: reduce transmission rate of reset messages
when link is down") revealed a flaw in the node FSM, as defined in
the log of commit 66996b6c47 ("tipc: extend node FSM").
We see the following scenario:
1: Node B receives a RESET message from node A before its link endpoint
is fully up, i.e., the node FSM is in state SELF_UP_PEER_COMING. This
event will not change the node FSM state, but the (distinct) link FSM
will move to state RESETTING.
2: As an effect of the previous event, the local endpoint on B will
declare node A lost, and post the event SELF_DOWN to the its node
FSM. This moves the FSM state to SELF_DOWN_PEER_LEAVING, meaning
that no messages will be accepted from A until it receives another
RESET message that confirms that A's endpoint has been reset. This
is wasteful, since we know this as a fact already from the first
received RESET, but worse is that the link instance's FSM has not
wasted this information, but instead moved on to state ESTABLISHING,
meaning that it repeatedly sends out ACTIVATE messages to the reset
peer A.
3: Node A will receive one of the ACTIVATE messages, move its link FSM
to state ESTABLISHED, and start repeatedly sending out STATE messages
to node B.
4: Node B will consistently drop these messages, since it can only accept
accept a RESET according to its node FSM.
5: After four lost STATE messages node A will reset its link and start
repeatedly sending out RESET messages to B.
6: Because of the reduced send rate for RESET messages, it is very
likely that A will receive an ACTIVATE (which is sent out at a much
higher frequency) before it gets the chance to send a RESET, and A
may hence quickly move back to state ESTABLISHED and continue sending
out STATE messages, which will again be dropped by B.
7: GOTO 5.
8: After having repeated the cycle 5-7 a number of times, node A will
by chance get in between with sending a RESET, and the situation is
resolved.
Unfortunately, we have seen that it may take a substantial amount of
time before this vicious loop is broken, sometimes in the order of
minutes.
We correct this by making a small correction to the node FSM: When a
node in state SELF_UP_PEER_COMING receives a SELF_DOWN event, it now
moves directly back to state SELF_DOWN_PEER_DOWN, instead of as now
SELF_DOWN_PEER_LEAVING. This is logically consistent, since we don't
need to wait for RESET confirmation from of an endpoint that we alread
know has been reset. It also means that node B in the scenario above
will not be dropping incoming STATE messages, and the link can come up
immediately.
Finally, a symmetry comparison reveals that the FSM has a similar
error when receiving the event PEER_DOWN in state PEER_UP_SELF_COMING.
Instead of moving to PERR_DOWN_SELF_LEAVING, it should move directly
to SELF_DOWN_PEER_DOWN. Although we have never seen any negative effect
of this logical error, we choose fix this one, too.
The node FSM looks as follows after those changes:
+----------------------------------------+
| PEER_DOWN_EVT|
| |
+------------------------+----------------+ |
|SELF_DOWN_EVT | | |
| | | |
| +-----------+ +-----------+ |
| |NODE_ | |NODE_ | |
| +----------|FAILINGOVER|<---------|SYNCHING |-----------+ |
| |SELF_ +-----------+ FAILOVER_+-----------+ PEER_ | |
| |DOWN_EVT | A BEGIN_EVT A | DOWN_EVT| |
| | | | | | | |
| | | | | | | |
| | |FAILOVER_ |FAILOVER_ |SYNCH_ |SYNCH_ | |
| | |END_EVT |BEGIN_EVT |BEGIN_EVT|END_EVT | |
| | | | | | | |
| | | | | | | |
| | | +--------------+ | | |
| | +-------->| SELF_UP_ |<-------+ | |
| | +-----------------| PEER_UP |----------------+ | |
| | |SELF_DOWN_EVT +--------------+ PEER_DOWN_EVT| | |
| | | A A | | |
| | | | | | | |
| | | PEER_UP_EVT| |SELF_UP_EVT | | |
| | | | | | | |
V V V | | V V V
+------------+ +-----------+ +-----------+ +------------+
|SELF_DOWN_ | |SELF_UP_ | |PEER_UP_ | |PEER_DOWN |
|PEER_LEAVING| |PEER_COMING| |SELF_COMING| |SELF_LEAVING|
+------------+ +-----------+ +-----------+ +------------+
| | A A | |
| | | | | |
| SELF_ | |SELF_ |PEER_ |PEER_ |
| DOWN_EVT| |UP_EVT |UP_EVT |DOWN_EVT |
| | | | | |
| | | | | |
| | +--------------+ | |
|PEER_DOWN_EVT +--->| SELF_DOWN_ |<---+ SELF_DOWN_EVT|
+------------------->| PEER_DOWN |<--------------------+
+--------------+
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that we can properly support multiple distinct trees in the system,
using a global variable: dsa_cpu_port_ethtool_ops is getting clobbered
as soon as the second switch tree gets probed, and we don't want that.
We need to move this to be dynamically allocated, and since we can't
really be comparing addresses anymore to determine first time
initialization versus any other times, just move this to dsa.c and
dsa2.c where the remainder of the dst/ds initialization happens.
The operations teardown restores the master netdev's ethtool_ops to its
original ethtool_ops pointer (typically within the Ethernet driver)
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a helper function: dsa_cpu_port_ethtool_init() which initializes a
custom ethtool_ops structure with custom DSA ethtool operations for CPU
ports. This is a preliminary change to move the initialization outside
of net/dsa/slave.c.
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Mimic what net/dsa/dsa.c does and provide a slave MII bus by default
which will be created if the driver implements a phy_read method.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some drivers rely on these two bitmasks to contain the correct values
for them to successfully probe and initialize at drv->setup() time,
calculate correct values to put in both masks as early as possible in
dsa_get_ports_dn().
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case we have multiples trees and switches with the same index, we
need to add another discriminating id: the switch tree.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
"make htmldocs" complains otherwise:
.//net/core/gen_stats.c:168: warning: No description found for parameter 'running'
.//include/linux/netdevice.h:1867: warning: No description found for parameter 'qdisc_running_key'
Fixes: f9eb8aea2a ("net_sched: transform qdisc running bit into a seqcount")
Fixes: edb09eb17e ("net: sched: do not acquire qdisc spinlock in qdisc/class stats dump")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When in kdump kernel, reduce memory usage by only using a single Queue
Set for multiqueue devices. So make netif_get_num_default_rss_queues()
return one, when in kdump kernel.
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When setting up ILA in a router we noticed that the the encapsulation
is invoked twice: once in the route input path and again upon route
output. To resolve this we add a flag set_csum_neutral for the
ila_update_ipv6_locator. If this flag is set and the checksum
neutral bit is also set we assume that checksum-neutral translation
has already been performed and take no further action. The
flag is set only in ila_output path. The flag is not set for ila_input and
ila_xlat.
Tested:
Used 3 netns to set to emulate a router and two hosts. The router
translates SIR addresses between the two destinations in other two netns.
Verified ping and netperf are functional.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RFC 5961 advises to only accept RST packets containing a seq number
matching the next expected seq number instead of the whole receive
window in order to avoid spoofing attacks.
However, this situation is not optimal in the case SACK is in use at the
time the RST is sent. I recently run into a scenario in which packet
losses were high while uploading data to a server, and userspace was
willing to frequently terminate connections by sending a RST. In
this case, the ACK sent on the receiver side (rcv_nxt) is frozen waiting
for a lost packet retransmission and SACK blocks are used to let the
client continue uploading data. At some point later on, the client sends
the RST (snd_nxt), which matches the next expected seq number of the
right-most SACK block on the receiver side which is going forward
receiving data.
In this scenario, as RFC 5961 defines, the RST SEQ doesn't match the
frozen main ACK at receiver side and thus gets dropped and a challenge
ACK is sent, which gets usually lost due to network conditions. The main
consequence is that the connection stays alive for a while even if it
made sense to accept the RST. This can get really bad if lots of
connections like this one are created in few seconds, allocating all the
resources of the server easily.
For security reasons, not all SACK blocks are checked (there could be a
big amount of SACK blocks => acceptable SEQ numbers). Furthermore, it
wouldn't make sense to check for RST in blocks other than the right-most
received one because the sender is not expected to be sending new data
after the RST. For simplicity, only up to the 4 most recently updated
SACK blocks (selective_acks[4] field) are compared to find the
right-most block, as usually those are the ones with bigger probability
to contain it.
This patch was tested in a 3.18 kernel and probed to improve the
situation in the scenario described above.
Signed-off-by: Pau Espin Pedrol <pau.espin@tessares.net>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for locally originated traffic to VRF-local IPv6 addresses.
Similar to IPv4 a local dst is set on the skb and the packet is
reinserted with a call to netif_rx. With this patch, ping, tcp and udp
packets to a local IPv6 address are successfully routed:
$ ip addr show dev eth1
4: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master red state UP group default qlen 1000
link/ether 02:e0:f9:1c:b9:74 brd ff:ff:ff:ff:ff:ff
inet 10.100.1.1/24 brd 10.100.1.255 scope global eth1
valid_lft forever preferred_lft forever
inet6 2100:1::1/120 scope global
valid_lft forever preferred_lft forever
inet6 fe80::e0:f9ff:fe1c:b974/64 scope link
valid_lft forever preferred_lft forever
$ ping6 -c1 -I red 2100:1::1
ping6: Warning: source address might be selected on device other than red.
PING 2100:1::1(2100:1::1) from 2100:1::1 red: 56 data bytes
64 bytes from 2100:1::1: icmp_seq=1 ttl=64 time=0.098 ms
ip6_input is exported so the VRF driver can use it for the dst input
function. The dst_alloc function for IPv4 defaults to setting the input and
output functions; IPv6's does not. VRF does not need to duplicate the Rx path
so just export the ipv6 input function.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch implements direct encapsulation of IPv4 and IPv6 packets
in UDP. This is done a version "1" of GUE and as explained in I-D
draft-ietf-nvo3-gue-03.
Changes here are only in the receive path, fou with IPxIPx already
supports the transmit side. Both the normal receive path and
GRO path are modified to check for GUE version and check for
IP version in the case that GUE version is "1".
Tested:
IPIP with direct GUE encap
1 TCP_STREAM
4530 Mbps
200 TCP_RR
1297625 tps
135/232/444 90/95/99% latencies
IP4IP6 with direct GUE encap
1 TCP_STREAM
4903 Mbps
200 TCP_RR
1184481 tps
149/253/473 90/95/99% latencies
IP6IP6 direct GUE encap
1 TCP_STREAM
5146 Mbps
200 TCP_RR
1202879 tps
146/251/472 90/95/99% latencies
SIT with direct GUE encap
1 TCP_STREAM
6111 Mbps
200 TCP_RR
1250337 tps
139/241/467 90/95/99% latencies
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Large tc dumps (tc -s {qdisc|class} sh dev ethX) done by Google BwE host
agent [1] are problematic at scale :
For each qdisc/class found in the dump, we currently lock the root qdisc
spinlock in order to get stats. Sampling stats every 5 seconds from
thousands of HTB classes is a challenge when the root qdisc spinlock is
under high pressure. Not only the dumps take time, they also slow
down the fast path (queue/dequeue packets) by 10 % to 20 % in some cases.
An audit of existing qdiscs showed that sch_fq_codel is the only qdisc
that might need the qdisc lock in fq_codel_dump_stats() and
fq_codel_dump_class_stats()
In v2 of this patch, I now use the Qdisc running seqcount to provide
consistent reads of packets/bytes counters, regardless of 32/64 bit arches.
I also changed rate estimators to use the same infrastructure
so that they no longer need to lock root qdisc lock.
[1]
http://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43838.pdf
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Kevin Athey <kda@google.com>
Cc: Xiaotian Pei <xiaotian@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of using a single bit (__QDISC___STATE_RUNNING)
in sch->__state, use a seqcount.
This adds lockdep support, but more importantly it will allow us
to sample qdisc/class statistics without having to grab qdisc root lock.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Useful to know when the action was first used for accounting
(and debugging)
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to make a filter processed only by hardware, skip_sw flag
should be supplied. This is an addition to the already existing skip_hw
flag (filter will be processed by software only). If no flag is
specified, filter will be processed by both software and hardware.
If only hardware offloaded filters exist, fl_classify() will return
without doing anything.
A following userspace patch will be sent once kernel patch is accepted.
Example:
tc filter add dev enp0s9 protocol ip prio 20 parent ffff: \
flower \
ip_proto 6 \
indev enp0s9 \
skip_sw \
action skbedit mark 0x1234
Signed-off-by: Amir Vadai <amirva@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Note: Tom Herbert posted almost same patch 3 months back, but for
different reasons.
The reasons we want to get rid of this spin_trylock() are :
1) Under high qdisc pressure, the spin_trylock() has almost no
chance to succeed.
2) We loop multiple times in softirq handler, eventually reaching
the max retry count (10), and we schedule ksoftirqd.
Since we want to adhere more strictly to ksoftirqd being waked up in
the future (https://lwn.net/Articles/687617/), better avoid spurious
wakeups.
3) calls to __netif_reschedule() dirty the cache line containing
q->next_sched, slowing down the owner of qdisc.
4) RT kernels can not use the spin_trylock() here.
With help of busylock, we get the qdisc spinlock fast enough, and
the trylock trick brings only performance penalty.
Depending on qdisc setup, I observed a gain of up to 19 % in qdisc
performance (1016600 pps instead of 853400 pps, using prio+tbf+fq_codel)
("mpstat -I SCPU 1" is much happier now)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <tom@herbertland.com>
Acked-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Before commit 6d7b857d54 ("net: use lib/percpu_counter API for
fragmentation mem accounting"), setting the reassembly high threshold
to 0 prevented fragment reassembly as first fragment would be always
evicted before second could be added to the queue. While inefficient,
some users apparently relied on this method.
Since the commit mentioned above, a percpu counter is used for
reassembly memory accounting and high batch size avoids taking slow path
in most common scenarios. As a result, a whole full sized packet can be
reassembled without the percpu counter's main counter changing its value
so that even with high_thresh set to 0, fragmented packets can be still
reassembled and processed.
Add explicit check preventing reassembly if high threshold is zero.
Signed-off-by: Michal Kubecek <mkubecek@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
The existing DSA binding has a number of limitations and problems. The
main problem is that it cannot represent a switch as a linux device,
hanging off some bus. It is limited to one CPU port. The DSA platform
device is artificial, and does not really represent hardware.
Implement a new binding which can be embedded into any type of node on
a bus to represent one switch device, and its links to other switches.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The switch may want to instantiate its own MDIO bus. Only do it
centrally if the switch has not already created one, and the read op
is implemented.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace the two switch statements with an array lookup, and store the
result in the dsa tree structure. The drivers no longer need to know
the selected tag protocol, so remove it from the dsa switch structure.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Refactor the code to setup a single DSA/CPU port into a function of
its own, and export it, so it can be used by the new binding.
Similarly, refactor the destroy code into a function. When destroying
the ports, don't put the of node. They should be released at the end
along with the normal ports.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The new binding will not have a chip data structure, it will place the
routing directly into the switch structure. To enable backwards
compatibility, copy the routing from the chip data into the switch
structure.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With a maximum of four switches, the size of the routing table is the
same as the pointer to it. Removing it makes the code simpler.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the port device node structure into the port structure, from the
chip data. This information is needed in the next step of implementing
the new binding.
The chip data structure is used while parsing the whole old binding,
before the individual switch structures exist. With the new bindings,
this is reversed, the switches exist first, and the interconnections
between the switches is derived from the individual switch
bindings. Thus this chip data structure becomes unneeded.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
eviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are going to be more per-port members added to the switch
structure. So add a port structure and move the netdev into it.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The platform data nr_chips is used when validating a received packet,
to ensure it comes from a know switch chip. The number of possible
switches is limited to DSA_MAX_SWITCHES, so use this as the first
validation step. The new binding allows holes in the dst->ds[] array,
so also ensure ensure there is a valid dsa_switch for this packet.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The DSA layer should no longer assume the switch is connected to an
MDIO bus. As a result, we cannot use the address on the MDIO bus when
forming the name of the switches internal MDIO bus for its builtin and
possibly external PHYs. The switch index is sufficient to make the
name unique, so drop the MDIO address.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The new binding does not make use of dsa_chip_data, a.k.a cd. When
retrieving the size of the EEPROM attached to a switch, don't assume
there is a cd attached to the switch structure.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
size_t objects should be printed with %Z printf format.
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the more common kernel logging style and reduce object size.
The logging message prefix changes from a mixture of
"RxRPC:" and "RXRPC:" to "af_rxrpc: ".
$ size net/rxrpc/built-in.o*
text data bss dec hex filename
64172 1972 8304 74448 122d0 net/rxrpc/built-in.o.new
67512 1972 8304 77788 12fdc net/rxrpc/built-in.o.old
Miscellanea:
o Consolidate the ASSERT macros to use a single pr_err call with
decimal and hexadecimal output and a stringified #OP argument
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is useful for debugging packet sizes.
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Tested-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
SCTP has this pecualiarity that its packets cannot be just segmented to
(P)MTU. Its chunks must be contained in IP segments, padding respected.
So we can't just generate a big skb, set gso_size to the fragmentation
point and deliver it to IP layer.
This patch takes a different approach. SCTP will now build a skb as it
would be if it was received using GRO. That is, there will be a cover
skb with protocol headers and children ones containing the actual
segments, already segmented to a way that respects SCTP RFCs.
With that, we can tell skb_segment() to just split based on frag_list,
trusting its sizes are already in accordance.
This way SCTP can benefit from GSO and instead of passing several
packets through the stack, it can pass a single large packet.
v2:
- Added support for receiving GSO frames, as requested by Dave Miller.
- Clear skb->cb if packet is GSO (otherwise it's not used by SCTP)
- Added heuristics similar to what we have in TCP for not generating
single GSO packets that fills cwnd.
v3:
- consider sctphdr size in skb_gso_transport_seglen()
- rebased due to 5c7cdf339a ("gso: Remove arbitrary checks for
unsupported GSO")
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Tested-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
