Change the selection of a multipath route to use a flow-based
hash. This more suitable for traffic sensitive to reordering within a
flow (e.g. TCP, L2VPN) and whilst still allowing a good distribution
of traffic given enough flows.
Selection of the path for a multipath route is done using a hash of:
1. Label stack up to MAX_MP_SELECT_LABELS labels or up to and
including entropy label, whichever is first.
2. 3-tuple of (L3 src, L3 dst, proto) from IPv4/IPv6 header in MPLS
payload, if present.
Naturally, a 5-tuple hash using L4 information in addition would be
possible and be better in some scenarios, but there is a tradeoff
between looking deeper into the packet to achieve good distribution,
and packet forwarding performance, and I have erred on the side of the
latter as the default.
Signed-off-by: Robert Shearman <rshearma@brocade.com>
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support for MPLS multipath routes.
Includes following changes to support multipath:
- splits struct mpls_route into 'struct mpls_route + struct mpls_nh'
- 'struct mpls_nh' represents a mpls nexthop label forwarding entry
- moves mpls route and nexthop structures into internal.h
- A mpls_route can point to multiple mpls_nh structs
- the nexthops are maintained as a array (similar to ipv4 fib)
- In the process of restructuring, this patch also consistently changes
all labels to u8
- Adds support to parse/fill RTA_MULTIPATH netlink attribute for
multipath routes similar to ipv4/v6 fib
- In this patch, the multipath route nexthop selection algorithm
simply returns the first nexthop. It is replaced by a
hash based algorithm from Robert Shearman in the next patch
- mpls_route_update cleanup: remove 'dev' handling in mpls_route_update.
mpls_route_update though implemented to update based on dev, it was
never used that way. And the dev handling gets tricky with multiple
nexthops. Cannot match against any single nexthops dev. So, this patch
removes the unused 'dev' handling in mpls_route_update.
- dead route/path handling will be implemented in a subsequent patch
Example:
$ip -f mpls route add 100 nexthop as 200 via inet 10.1.1.2 dev swp1 \
nexthop as 700 via inet 10.1.1.6 dev swp2 \
nexthop as 800 via inet 40.1.1.2 dev swp3
$ip -f mpls route show
100
nexthop as to 200 via inet 10.1.1.2 dev swp1
nexthop as to 700 via inet 10.1.1.6 dev swp2
nexthop as to 800 via inet 40.1.1.2 dev swp3
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Acked-by: Robert Shearman <rshearma@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Neil Armstrong says:
====================
Refactor nested mdiobus read/write functions
In order to avoid locked signal false positive for nested mdiobus
read/write calls, nested code was introduced in mv88e6xxx and
mdio-mux.
But mv88e6060 also needs such nested mdiobus read/write calls.
For sake of refactoring, introduce nested variants of mdiobus read/write
and make them used by mv88e6xxx and mv88e6060.
In a next patch, mdio-mux should also use these variant calls.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Like mv88e6xxx and mdio-mux, to avoid lockdep give false positives
because of nested MDIO busses, switch to previously introduced
nested mdiobus_read/write variants.
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make the mv88e6xxx driver use the previously introduced nested
variants of mdiobus_read/write functions.
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since nested variants of mdiobus_read/write are used in multiple
drivers, add nested variants in the mdiobus core.
Suggested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no reasons to probe cpsw from late_initcall level
and it's not recommended. Hence, use module_platform_driver()
to register and probe cpsw driver from module_init() level.
Cc: Tony Lindgren <tony@atomide.com>
Acked-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.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>
Currently adding a new ipv4 address always cause the creation of the
related network route, with default metric. When a host has multiple
interfaces on the same network, multiple routes with the same metric
are created.
If the userspace wants to set specific metric on each routes, i.e.
giving better metric to ethernet links in respect to Wi-Fi ones,
the network routes must be deleted and recreated, which is error-prone.
This patch implements the support for IFA_F_NOPREFIXROUTE for ipv4
address. When an address is added with such flag set, no associated
network route is created, no network route is deleted when
said IP is gone and it's up to the user space manage such route.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Broadcom ethernet driver for the new family of NetXtreme-C/E
ethernet devices.
v5:
- Removed empty blank lines at end of files (noted by David Miller).
- Moved busy poll helper functions to bnxt.h to at least make the
.c file look less cluttered with #ifdef (noted by Stephen Hemminger).
v4:
- Broke up 2 long message strings with "\n" (suggested by John Linville)
- Constify an array of strings (suggested by Stephen Hemminger)
- Improve bnxt_vf_pciid() (suggested by Stephen Hemminger)
- Use PCI_VDEVICE() to populate pci_device_id table for more compact
source.
v3:
- Fixed 2 more sparse warnings.
- Removed some unused structures in .h files.
v2:
- Fixed all kbuild test robot reported warnings.
- Fixed many of the checkpatch.pl errors and warnings.
- Fixed the Kconfig description (noted by Dmitry Kravkov).
Acked-by: Eddie Wai <eddie.wai@broadcom.com>
Acked-by: Jeffrey Huang <huangjw@broadcom.com>
Signed-off-by: Prashant Sreedharan <prashant@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is preferable to have a common debugfs interface for DSA or switchdev
instead of a driver specific one. Thus remove the mv88e6xxx debug code.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Vivien Didelot says:
====================
net: dsa: implement port_fdb_dump in drivers
Not all switch chips provide a Get Next kind of operation to dump FDB entries.
It is preferred to let the driver handle the dump operation the way it works
best for the chip. Thus, drop port_fdb_getnext and implement the port_fdb_dump
operation in DSA, which pushes the switchdev FDB dump callback down to the
drivers. mv88e6xxx is the only driver affected and is updated accordingly.
v3 -> v4: fix rejects on latest net-next
v2 -> v3: opencode switchdev_obj_dump_cb_t to avoid multiple typedef;
use ether_addr_copy in fdb_dump
v1 -> v2: fix a few "return err" instead of "goto unlock" in mv88e6xxx.c
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
No driver implements port_fdb_getnext anymore, and port_fdb_dump is
preferred anyway, so remove this function from DSA.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that port_fdb_dump is implemented and even simpler, get rid of
port_fdb_getnext.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement the port_fdb_dump DSA operation.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no need to write the MAC address before every Get Next
operation, since ATU MAC registers are not cleared between calls.
Move the _mv88e6xxx_atu_mac_write call outside of _mv88e6xxx_atu_getnext
so future code could call ATU Get Next multiple times and save a few
register access.
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no need to write the VLAN ID before every Get Next operation,
since the VTU VID register is not cleared between calls.
Move the VID write call in a _mv88e6xxx_vtu_vid_write function outside
of _mv88e6xxx_vtu_getnext so future code could call VTU Get Next
multiple times and save a few register accesses.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Not all switch chips support a Get Next operation to iterate on its FDB.
So add a more simple port_fdb_dump function for them.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
* I merged net-next back to avoid a conflict with the
* cfg80211 scheduled scan API extensions
* preparations for better scan result timestamping
* regulatory cleanups
* mac80211 statistics cleanups
* a few other small cleanups and fixes
-----BEGIN PGP SIGNATURE-----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=uZ5N
-----END PGP SIGNATURE-----
Merge tag 'mac80211-next-for-davem-2015-10-21' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next
Johannes Berg says:
====================
Here's another set of patches for the current cycle:
* I merged net-next back to avoid a conflict with the
* cfg80211 scheduled scan API extensions
* preparations for better scan result timestamping
* regulatory cleanups
* mac80211 statistics cleanups
* a few other small cleanups and fixes
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
the global Soc configuration is treated by syscon, and sub ctrl bus is
Soc bus. it has to be treated by syscon.
Signed-off-by: yankejian <yankejian@huawei.com>
Signed-off-by: lisheng <lisheng011@huawei.com>
Signed-off-by: lipeng <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Hariprasad Shenai says:
====================
Trivial fixes for cxgb4 driver
This patch series updates driver description for next gen. adapters, updates
firmware info., returns error for setup_rss error case, restores L1
configuration in case of FW rejects new config, updates and aligns ethtool
get stats settings, etc
This patch series has been created against net-next tree and includes
patches on cxgb4 and cxgb4vf driver.
We have included all the maintainers of respective drivers. Kindly review
the change and let us know in case of any review comments.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Update ethtool get_drvinfo to display regdump len and also update
firmware string version print to display N/A in case FW isn't present
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the ethtool set_settings() routine we need to remember our old L1
Configuration in case the firmware rejects the request and then restore
that.
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For {1, 10, 40} Gb/s. Prohibiting turning off autonegotiation isn't anywhere
in the standard.
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Align the ethtool get stats settings with the rest so it looks uniform
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With use of lwtunnel, we can directly call dev_queue_xmit()
rather than calling netdev vport send operation.
Following change make tunnel vport code bit cleaner.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Patch fixes following sparse warning.
net/openvswitch/flow_netlink.c:583:30: warning: incorrect type in assignment (different base types)
net/openvswitch/flow_netlink.c:583:30: expected restricted __be16 [usertype] ipv4
net/openvswitch/flow_netlink.c:583:30: got int
Fixes: 6b26ba3a7d ("openvswitch: netlink attributes for IPv6 tunneling")
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Alexei Starovoitov says:
====================
bpf_perf_event_output helper
Over the last year there were multiple attempts to let eBPF programs
output data into perf events by He Kuang and Wangnan.
The last one was:
https://lkml.org/lkml/2015/7/20/736
It was almost perfect with exception that all bpf programs would sent
data into one global perf_event.
This patch set takes different approach by letting user space
open independent PERF_COUNT_SW_BPF_OUTPUT events, so that program
output won't collide.
Wangnan is working on corresponding perf patches.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Performance test and example of bpf_perf_event_output().
kprobe is attached to sys_write() and trivial bpf program streams
pid+cookie into userspace via PERF_COUNT_SW_BPF_OUTPUT event.
Usage:
$ sudo ./bld_x64/samples/bpf/trace_output
recv 2968913 events per sec
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This helper is used to send raw data from eBPF program into
special PERF_TYPE_SOFTWARE/PERF_COUNT_SW_BPF_OUTPUT perf_event.
User space needs to perf_event_open() it (either for one or all cpus) and
store FD into perf_event_array (similar to bpf_perf_event_read() helper)
before eBPF program can send data into it.
Today the programs triggered by kprobe collect the data and either store
it into the maps or print it via bpf_trace_printk() where latter is the debug
facility and not suitable to stream the data. This new helper replaces
such bpf_trace_printk() usage and allows programs to have dedicated
channel into user space for post-processing of the raw data collected.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of WARN_ON in perf_event_output() on unpaded raw samples,
pad them automatically.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the ipv4 outbound path of an ipvlan device in l3 mode, the ifindex is
being grabbed from dev_get_iflink. This works for the physical device
case, since as the documentation of that function notes: "Physical
interfaces have the same 'ifindex' and 'iflink' values.". However, if
the master device is a veth, and the pairs are in separate net
namespaces, the route lookup will fail with -ENODEV due to outer veth
pair being in a separate namespace from the ipvlan master/routing
namespace.
ns0 | ns1 | ns2
veth0a--|--veth0b--|--ipvl0
In ipvlan_process_v4_outbound(), a packet sent from ipvl0 in the above
configuration will pass fl.flowi4_oif == veth0a to
ip_route_output_flow(), but *net == ns1.
Notice also that ipv6 processing is not using iflink. Since there is a
discrepancy in usage, fixup both v4 and v6 case to use local dev
variable.
Tested this with l3 ipvlan on top of veth, as well as with single
physical interface in the top namespace.
Signed-off-by: Brenden Blanco <bblanco@plumgrid.com>
Reviewed-by: Jiri Benc <jbenc@redhat.com>
Acked-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allowing an application to set whatever limit for
the list of recently RST fastopen sessions [1] is not wise,
as it open ways to deplete kernel memory.
Cap the user provided limit by somaxconn sysctl,
like listen() backlog.
[1] https://tools.ietf.org/html/rfc7413#section-5.1
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This header file only contains the platform data structure definition,
so move it to the include/linux/platform_data/ directory.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove the inclusion of linux/mdio-gpio.h in nas4220b, wbd111 and wbd222
boards since mdio-gpio is not used.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/hisilicon/hns/hnae.c:442:1-3: WARNING: PTR_ERR_OR_ZERO can be used
Use PTR_ERR_OR_ZERO rather than if(IS_ERR(...)) + PTR_ERR
Generated by: scripts/coccinelle/api/ptr_ret.cocci
CC: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tom Herbert added SIT support to GRO with commit
19424e052f ("sit: Add gro callbacks to sit_offload"),
later reverted by Herbert Xu.
The problem came because Tom patch was building GRO
packets without proper meta data : If packets were locally
delivered, we would not care.
But if packets needed to be forwarded, GSO engine was not
able to segment individual segments.
With the following patch, we correctly set skb->encapsulation
and inner network header. We also update gso_type.
Tested:
Server :
netserver
modprobe dummy
ifconfig dummy0 8.0.0.1 netmask 255.255.255.0 up
arp -s 8.0.0.100 4e:32:51:04:47:e5
iptables -I INPUT -s 10.246.7.151 -j TEE --gateway 8.0.0.100
ifconfig sixtofour0
sixtofour0 Link encap:IPv6-in-IPv4
inet6 addr: 2002:af6:798::1/128 Scope:Global
inet6 addr: 2002:af6:798::/128 Scope:Global
UP RUNNING NOARP MTU:1480 Metric:1
RX packets:411169 errors:0 dropped:0 overruns:0 frame:0
TX packets:409414 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:20319631739 (20.3 GB) TX bytes:29529556 (29.5 MB)
Client :
netperf -H 2002:af6:798::1 -l 1000 &
Checked on server traffic copied on dummy0 and verify segments were
properly rebuilt, with proper IP headers, TCP checksums...
tcpdump on eth0 shows proper GRO aggregation takes place.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While testing my SIT/GRO patch using netfilter TEE module and a dummy
device, I found some features were missing :
TSO IPv6, UFO, and encapsulated traffic.
ethtool -k dummy0 now gives :
...
tcp-segmentation-offload: on
tx-tcp-segmentation: on
tx-tcp-ecn-segmentation: on
tx-tcp6-segmentation: on
udp-fragmentation-offload: on
...
tx-gre-segmentation: on
tx-ipip-segmentation: on
tx-sit-segmentation: on
tx-udp_tnl-segmentation: on
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
if_nlmsg_size() overestimates the minimum allocation size of netlink
dump request (when called from rtnl_calcit()) or the size of the
message (when called from rtnl_getlink()). This is because
ext_filter_mask is not supported by rtnl_link_get_af_size() and
rtnl_link_get_size().
The over-estimation is significant when at least one netdev has many
VLANs configured (8 bytes for each configured VLAN).
This patch-set "rightsizes" the protocol specific attribute size
calculation by propagating ext_filter_mask to rtnl_link_get_af_size()
and adding this a argument to get_link_af_size op in rtnl_af_ops.
Bridge module already used filtering aware sizing for notifications.
br_get_link_af_size_filtered() is consistent with the modified
get_link_af_size op so it replaces br_get_link_af_size() in br_af_ops.
br_get_link_af_size() becomes unused and thus removed.
Signed-off-by: Ronen Arad <ronen.arad@intel.com>
Acked-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Configure ageing time to the HW for newly bridged device
CC: Scott Feldman <sfeldma@gmail.com>
CC: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Elad Raz <eladr@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Yuchung Cheng says:
====================
RACK loss detection
RACK (Recent ACK) loss recovery uses the notion of time instead of
packet sequence (FACK) or counts (dupthresh).
It's inspired by the FACK heuristic in tcp_mark_lost_retrans(): when a
limited transmit (new data packet) is sacked in recovery, then any
retransmission sent before that newly sacked packet was sent must have
been lost, since at least one round trip time has elapsed.
But that existing heuristic from tcp_mark_lost_retrans()
has several limitations:
1) it can't detect tail drops since it depends on limited transmit
2) it's disabled upon reordering (assumes no reordering)
3) it's only enabled in fast recovery but not timeout recovery
RACK addresses these limitations with a core idea: an unacknowledged
packet P1 is deemed lost if a packet P2 that was sent later is is
s/acked, since at least one round trip has passed.
Since RACK cares about the time sequence instead of the data sequence
of packets, it can detect tail drops when a later retransmission is
s/acked, while FACK or dupthresh can't. For reordering RACK uses a
dynamically adjusted reordering window ("reo_wnd") to reduce false
positives on ever (small) degree of reordering, similar to the delayed
Early Retransmit.
In the current patch set RACK is only a supplemental loss detection
and does not trigger fast recovery. However we are developing RACK
to replace or consolidate FACK/dupthresh, early retransmit, and
thin-dupack. These heuristics all implicitly bear the time notion.
For example, the delayed Early Retransmit is simply applying RACK
to trigger the fast recovery with small inflight.
RACK requires measuring the minimum RTT. Tracking a global min is less
robust due to traffic engineering pathing changes. Therefore it uses a
windowed filter by Kathleen Nichols. The min RTT can also be useful
for various other purposes like congestion control or stat monitoring.
This patch has been used on Google servers for well over 1 year. RACK
has also been implemented in the QUIC protocol. We are submitting an
IETF draft as well.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch implements the second half of RACK that uses the the most
recent transmit time among all delivered packets to detect losses.
tcp_rack_mark_lost() is called upon receiving a dubious ACK.
It then checks if an not-yet-sacked packet was sent at least
"reo_wnd" prior to the sent time of the most recently delivered.
If so the packet is deemed lost.
The "reo_wnd" reordering window starts with 1msec for fast loss
detection and changes to min-RTT/4 when reordering is observed.
We found 1msec accommodates well on tiny degree of reordering
(<3 pkts) on faster links. We use min-RTT instead of SRTT because
reordering is more of a path property but SRTT can be inflated by
self-inflicated congestion. The factor of 4 is borrowed from the
delayed early retransmit and seems to work reasonably well.
Since RACK is still experimental, it is now used as a supplemental
loss detection on top of existing algorithms. It is only effective
after the fast recovery starts or after the timeout occurs. The
fast recovery is still triggered by FACK and/or dupack threshold
instead of RACK.
We introduce a new sysctl net.ipv4.tcp_recovery for future
experiments of loss recoveries. For now RACK can be disabled by
setting it to 0.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch is the first half of the RACK loss recovery.
RACK loss recovery uses the notion of time instead
of packet sequence (FACK) or counts (dupthresh). It's inspired by the
previous FACK heuristic in tcp_mark_lost_retrans(): when a limited
transmit (new data packet) is sacked, then current retransmitted
sequence below the newly sacked sequence must been lost,
since at least one round trip time has elapsed.
But it has several limitations:
1) can't detect tail drops since it depends on limited transmit
2) is disabled upon reordering (assumes no reordering)
3) only enabled in fast recovery ut not timeout recovery
RACK (Recently ACK) addresses these limitations with the notion
of time instead: a packet P1 is lost if a later packet P2 is s/acked,
as at least one round trip has passed.
Since RACK cares about the time sequence instead of the data sequence
of packets, it can detect tail drops when later retransmission is
s/acked while FACK or dupthresh can't. For reordering RACK uses a
dynamically adjusted reordering window ("reo_wnd") to reduce false
positives on ever (small) degree of reordering.
This patch implements tcp_advanced_rack() which tracks the
most recent transmission time among the packets that have been
delivered (ACKed or SACKed) in tp->rack.mstamp. This timestamp
is the key to determine which packet has been lost.
Consider an example that the sender sends six packets:
T1: P1 (lost)
T2: P2
T3: P3
T4: P4
T100: sack of P2. rack.mstamp = T2
T101: retransmit P1
T102: sack of P2,P3,P4. rack.mstamp = T4
T205: ACK of P4 since the hole is repaired. rack.mstamp = T101
We need to be careful about spurious retransmission because it may
falsely advance tp->rack.mstamp by an RTT or an RTO, causing RACK
to falsely mark all packets lost, just like a spurious timeout.
We identify spurious retransmission by the ACK's TS echo value.
If TS option is not applicable but the retransmission is acknowledged
less than min-RTT ago, it is likely to be spurious. We refrain from
using the transmission time of these spurious retransmissions.
The second half is implemented in the next patch that marks packet
lost using RACK timestamp.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
a helper to prepare the first main RACK patch.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
a helper to prepare the main RACK patch
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove the existing lost retransmit detection because RACK subsumes
it completely. This also stops the overloading the ack_seq field of
the skb control block.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
