Eric Dumazet
e80054ea0c
tcp: fix potential xmit stalls caused by TCP_NOTSENT_LOWAT ...
[ Upstream commit 4bfe744ff1644fbc0a991a2677dc874475dd6776 ]
I had this bug sitting for too long in my pile, it is time to fix it.
Thanks to Doug Porter for reminding me of it!
We had various attempts in the past, including commit
0cbe6a8f089e ("tcp: remove SOCK_QUEUE_SHRUNK"),
but the issue is that TCP stack currently only generates
EPOLLOUT from input path, when tp->snd_una has advanced
and skb(s) cleaned from rtx queue.
If a flow has a big RTT, and/or receives SACKs, it is possible
that the notsent part (tp->write_seq - tp->snd_nxt) reaches 0
and no more data can be sent until tp->snd_una finally advances.
What is needed is to also check if POLLOUT needs to be generated
whenever tp->snd_nxt is advanced, from output path.
This bug triggers more often after an idle period, as
we do not receive ACK for at least one RTT. tcp_notsent_lowat
could be a fraction of what CWND and pacing rate would allow to
send during this RTT.
In a followup patch, I will remove the bogus call
to tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED)
from tcp_check_space(). Fact that we have decided to generate
an EPOLLOUT does not mean the application has immediately
refilled the transmit queue. This optimistic call
might have been the reason the bug seemed not too serious.
Tested:
200 ms rtt, 1% packet loss, 32 MB tcp_rmem[2] and tcp_wmem[2]
$ echo 500000 >/proc/sys/net/ipv4/tcp_notsent_lowat
$ cat bench_rr.sh
SUM=0
for i in {1..10}
do
V=`netperf -H remote_host -l30 -t TCP_RR -- -r 10000000,10000 -o LOCAL_BYTES_SENT | egrep -v "MIGRATED|Bytes"`
echo $V
SUM=$(($SUM + $V))
done
echo SUM=$SUM
Before patch:
$ bench_rr.sh
130000000
80000000
140000000
140000000
140000000
140000000
130000000
40000000
90000000
110000000
SUM=1140000000
After patch:
$ bench_rr.sh
430000000
590000000
530000000
450000000
450000000
350000000
450000000
490000000
480000000
460000000
SUM=4680000000 # This is 410 % of the value before patch.
Fixes: c9bee3b7fd
2022-05-09 09:03:25 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 188
2019-05-30 11:29:21 -07:00
Bluetooth: defer cleanup of resources in hci_unregister_dev()
2021-08-12 13:20:58 +02:00
net: caif: add proper error handling
2021-06-10 13:37:09 +02:00
netfilter: nf_queue: fix possible use-after-free
2022-03-08 19:07:47 +01:00
ipv4: convert fib_num_tclassid_users to atomic_t
2021-12-08 09:01:12 +01:00
NFC: add NCI_UNREG flag to eliminate the race
2021-12-01 09:23:35 +01:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 336
2019-06-05 17:37:07 +02:00
sctp: use call_rcu to free endpoint
2022-01-05 12:37:44 +01:00
net: sched: Fix suspicious RCU usage while accessing tcf_tunnel_info
2021-12-22 09:29:40 +01:00
net: sched: fix police ext initialization
2021-03-04 10:26:53 +01:00
net: bridge: mcast: fix broken length + header check for MRDv6 Adv.
2021-05-14 09:44:32 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 174
2019-05-30 11:26:41 -07:00
rxrpc: Make rxrpc_kernel_get_srtt() indicate validity
2020-09-09 19:12:23 +02:00
net: split out functions related to registering inflight socket files
2019-02-28 08:24:23 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 321
2019-06-05 17:37:05 +02:00
ipv4: Invalidate neighbour for broadcast address upon address addition
2022-04-15 14:18:33 +02:00
ax25: fix reference count leaks of ax25_dev
2022-04-20 09:19:39 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 500
2019-06-19 17:09:55 +02:00
bonding: fix data-races around agg_select_timer
2022-02-23 11:59:58 +01:00
bonding: make tx_rebalance_counter an atomic
2021-12-14 14:49:00 +01:00
bonding: add an option to specify a delay between peer notifications
2019-07-04 12:30:48 -07:00
bonding: wait for sysfs kobject destruction before freeing struct slave
2020-12-08 10:40:24 +01:00
bpf: support cloning sk storage on accept()
2019-08-17 23:18:54 +02:00
net: annotate data race around sk_ll_usec
2021-07-31 08:19:38 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 13
2019-05-21 11:28:45 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 500
2019-06-19 17:09:55 +02:00
mac80211: properly handle A-MSDUs that start with an RFC 1042 header
2021-06-03 08:59:01 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 174
2019-05-30 11:26:41 -07:00
net: Force inlining of checksum functions in net/checksum.h
2022-03-02 11:41:08 +01:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 13
2019-05-21 11:28:45 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
net: rework SIOCGSTAMP ioctl handling
2019-04-19 14:07:40 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 201
2019-05-30 11:29:52 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 201
2019-05-30 11:29:52 -07:00
devlink: disallow reload operation during device cleanup
2019-11-09 19:38:36 -08:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 24
2019-05-21 11:52:39 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 24
2019-05-21 11:52:39 +02:00
net: drop_monitor: use IS_REACHABLE() to guard net_dm_hw_report()
2020-05-27 17:46:24 +02:00
net: dsa: Pass ndo_setup_tc slave callback to drivers
2019-09-16 21:32:57 +02:00
net: Added pointer check for dst->ops->neigh_lookup in dst_neigh_lookup_skb
2020-07-22 09:32:47 +02:00
net: fix a memleak when uncloning an skb dst and its metadata
2022-02-16 12:52:51 +01:00
net: add bool confirm_neigh parameter for dst_ops.update_pmtu
2020-01-04 19:18:58 +01:00
esp: Fix possible buffer overflow in ESP transformation
2022-03-28 08:46:44 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 500
2019-06-19 17:09:55 +02:00
net: fib: drop include of module.h from fib_notifier.h
2019-04-22 21:50:53 -07:00
ipv6: fix memory leak in fib6_rule_suppress
2021-12-08 09:01:13 +01:00
ipv4: Initialize flowi4_multipath_hash in data path
2020-09-26 18:03:12 +02:00
net/sched: flower: fix parsing of ethertype following VLAN header
2022-04-20 09:19:34 +02:00
net: core: rename indirect block ingress cb function
2019-12-18 16:08:47 +01:00
net/flow_dissector: switch to siphash
2019-10-23 20:13:22 -07:00
net/fq_impl: Switch to kvmalloc() for memory allocation
2019-11-08 09:11:49 +01:00
genetlink: remove genl_bind
2020-07-22 09:32:46 +02:00
net: Move the definition of the default Geneve udp port to public header file
2019-03-22 12:09:31 -07:00
net:gue.h:Fix shifting signed 32-bit value by 31 bits problem
2019-07-01 10:58:23 -07:00
net: hwbm: if CONFIG_NET_HWBM unset, make stub functions static
2019-10-25 16:24:32 -07:00
net: icmp: pass zeroed opts from icmp{,v6}_ndo_send before sending
2021-03-04 10:26:53 +01:00
wireless-drivers-next patches for 5.1
2019-02-22 12:56:24 -08:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 174
2019-05-30 11:26:41 -07:00
ipv6: shrink struct ipv6_mc_socklist
2019-08-28 14:43:03 -07:00
net: ife: drop include of module.h from net/ife.h
2019-04-22 21:50:53 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
inet: factor out inet_send_prepare()
2019-07-03 13:51:54 -07:00
tcp: relookup sock for RST+ACK packets handled by obsolete req sock
2021-03-30 14:35:26 +02:00
inet_ecn: Fix endianness of checksum update when setting ECT(1)
2020-12-08 10:40:25 +01:00
inet: frags: annotate races around fqdir->dead and fqdir->high_thresh
2022-01-27 09:19:54 +01:00
tcp: fix race condition when creating child sockets from syncookies
2022-04-27 13:50:45 +02:00
ip: support SO_MARK cmsg
2019-09-13 21:44:19 +02:00
tcp: honor SO_PRIORITY in TIME_WAIT state
2019-09-27 12:05:02 +02:00
net: ipv4: use a dedicated counter for icmp_v4 redirect packets
2019-02-08 21:50:15 -08:00
ipv4: avoid using shared IP generator for connected sockets
2022-02-01 17:24:37 +01:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
ipv6: annotate accesses to fn->fn_sernum
2022-02-01 17:24:38 +01:00
net: ipv6: fix returned variable type in ip6_skb_dst_mtu
2021-08-12 13:20:57 +02:00
ip6_tunnel: allow not to count pkts on tstats by passing dev as NULL
2019-06-18 20:48:45 -04:00
ipv4: convert fib_num_tclassid_users to atomic_t
2021-12-08 09:01:12 +01:00
ip_tunnels: Set tunnel option flag when tunnel metadata is present
2020-11-24 13:29:05 +01:00
ipvs: allow connection reuse for unconfirmed conntrack
2020-08-19 08:16:10 +02:00
net: ipv6: add net argument to ip6_dst_lookup_flow
2019-12-18 16:08:40 +01:00
inet: frags: annotate races around fqdir->dead and fqdir->high_thresh
2022-01-27 09:19:54 +01:00
net: ipv6: add fib6_nh_release_dsts stub
2021-12-01 09:23:33 +01:00
bonding/alb: properly access headers in bond_alb_xmit()
2020-02-11 04:35:48 -08:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 500
2019-06-19 17:09:55 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
llc: fix out-of-bound array index in llc_sk_dev_hash()
2021-11-17 09:48:47 +01:00
llc: fix sk_buff leak in llc_conn_service()
2019-10-08 13:23:05 -07:00
net: llc: fix skb_over_panic
2021-08-04 12:27:39 +02:00
lwtunnel: Pass encap and encap type attributes to lwtunnel_fill_encap
2019-04-23 19:42:29 -07:00
mac80211: populate debugfs only after cfg80211 init
2020-04-29 16:33:18 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 174
2019-05-30 11:26:41 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 13
2019-05-21 11:28:45 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 295
2019-06-05 17:36:38 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 295
2019-06-05 17:36:38 +02:00
net: avoid potential false sharing in neighbor related code
2019-12-31 16:45:03 +01:00
net, neigh: Enable state migration between NUD_PERMANENT and NTF_USE
2021-11-17 09:48:49 +01:00
net: make get_net_ns return error if NET_NS is disabled
2021-06-23 14:41:25 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 13
2019-05-21 11:28:45 +02:00
netlink: Fix nlmsg_parse as a wrapper for strict message parsing
2019-08-13 20:37:16 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
net: netrom: Fix error cleanup path of nr_proto_init
2019-04-11 13:59:49 -07:00
net: ipv4: Fix rtnexthop len when RTA_FLOW is present
2021-10-06 15:42:33 +02:00
net: ieee802154: handle iftypes as u32
2021-12-01 09:23:32 +01:00
mm: fix struct page layout on 32-bit systems
2021-05-19 10:08:31 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
net_sched: fix ops->bind_class() implementations
2020-02-01 09:34:38 +00:00
net: prevent user from passing illegal stab size
2021-10-17 10:42:34 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
psample: Add a fwd declaration for skbuff
2021-08-18 08:56:59 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
sch_red: fix off-by-one checks in red_check_params()
2021-04-14 08:24:14 +02:00
net: add {READ|WRITE}_ONCE() annotations on ->rskq_accept_head
2019-10-09 21:34:31 -07:00
ipv4: remove sparse error in ip_neigh_gw4()
2022-02-01 17:24:39 +01:00
can: dev: Move device back to init netns on owning netns delete
2021-03-30 14:35:29 +02:00
net: Rename net/nexthop.h net/rtnh.h
2019-04-22 21:47:25 -07:00
net_sched: restore "mpu xxx" handling
2022-01-27 09:19:55 +01:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
net/smc: add smcd support to the pnet table
2019-02-21 10:34:37 -08:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
esp: Fix possible buffer overflow in ESP transformation
2022-03-28 08:46:44 +02:00
udp: correct reuseport selection with connected sockets
2019-09-16 09:02:18 +02:00
bpf: sockmap, strparser, and tls are reusing qdisc_skb_cb and colliding
2021-11-17 09:48:48 +01:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
tcp: fix potential xmit stalls caused by TCP_NOTSENT_LOWAT
2022-05-09 09:03:25 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
2019-05-30 11:26:32 -07:00
net/tls: Fix flipped sign in tls_err_abort() calls
2021-11-02 19:46:12 +01:00
udp: call udp_encap_enable for v6 sockets when enabling encap
2022-04-15 14:18:01 +02:00
udp: call udp_encap_enable for v6 sockets when enabling encap
2022-04-15 14:18:01 +02:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 321
2019-06-05 17:37:05 +02:00
vxlan: add adjacent link to limit depth level
2019-10-24 14:53:49 -07:00
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 268
2019-06-05 17:30:29 +02:00
xdp: page_pool related fix to cpumap
2019-06-19 11:23:13 -04:00
page_pool: do not release pool until inflight == 0.
2019-12-18 16:09:07 +01:00
xsk: add support to allow unaligned chunk placement
2019-08-31 01:08:26 +02:00
xfrm: Check if_id in xfrm_migrate
2022-03-19 13:40:16 +01:00
e80054ea0c