summaryrefslogtreecommitdiff
path: root/net
AgeCommit message (Collapse)Author
2019-01-18netfilter: conntrack: remove l4proto destroy hookFlorian Westphal
Only one user (gre), add a direct call and remove this facility. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: remove l4proto init and get_net callbacksFlorian Westphal
Those were needed we still had modular trackers. As we don't have those anymore, prefer direct calls and remove all the (un)register infrastructure associated with this. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: remove sysctl registration helpersFlorian Westphal
After previous patch these are not used anymore. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: unify sysctl handlingFlorian Westphal
Due to historical reasons, all l4 trackers register their own sysctls. This leads to copy&pasted boilerplate code, that does exactly same thing, just with different data structure. Place all of this in a single file. This allows to remove the various ctl_table pointers from the ct_netns structure and reduces overall code size. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: avoid unneeded nf_conntrack_l4proto lookupsFlorian Westphal
after removal of the packet and invert function pointers, several places do not need to lookup the l4proto structure anymore. Remove those lookups. The function nf_ct_invert_tuplepr becomes redundant, replace it with nf_ct_invert_tuple everywhere. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: remove pernet l4 proto register interfaceFlorian Westphal
No used anymore. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: remove remaining l4proto indirect packet callsFlorian Westphal
Now that all l4trackers are builtin, no need to use a mix of direct and indirect calls. This removes the last two users: gre and the generic l4 protocol tracker. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: remove module owner fieldFlorian Westphal
No need to get/put module owner reference, none of these can be removed anymore. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: remove invert_tuple callbackFlorian Westphal
Only used by icmp(v6). Prefer a direct call and remove this function from the l4proto struct. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: remove pkt_to_tuple callbackFlorian Westphal
GRE is now builtin, so we can handle it via direct call and remove the callback. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: remove net_idFlorian Westphal
No users anymore. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: gre: switch module to be built-inFlorian Westphal
This makes the last of the modular l4 trackers 'bool'. After this, all infrastructure to handle dynamic l4 protocol registration becomes obsolete and can be removed in followup patches. Old: 302824 net/netfilter/nf_conntrack.ko 21504 net/netfilter/nf_conntrack_proto_gre.ko New: 313728 net/netfilter/nf_conntrack.ko Old: text data bss dec hex filename 6281 1732 4 8017 1f51 nf_conntrack_proto_gre.ko 108356 20613 236 129205 1f8b5 nf_conntrack.ko New: 112095 21381 240 133716 20a54 nf_conntrack.ko The size increase is only temporary. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: gre: convert rwlock to rcuFlorian Westphal
We can use gre. Lock is only needed when a new expectation is added. In case a single spinlock proves to be problematic we can either add one per netns or use an array of locks combined with net_hash_mix() or similar to pick the 'correct' one. But given this is only needed for an expectation rather than per packet a single one should be ok. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: handle icmp pkt_to_tuple helper via direct callsFlorian Westphal
rather than handling them via indirect call, use a direct one instead. This leaves GRE as the last user of this indirect call facility. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: handle builtin l4proto packet functions via direct callsFlorian Westphal
The l4 protocol trackers are invoked via indirect call: l4proto->packet(). With one exception (gre), all l4trackers are builtin, so we can make .packet optional and use a direct call for most protocols. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: nf_tables: Support RULE_ID reference in new rulePhil Sutter
To allow for a batch to contain rules in arbitrary ordering, introduce NFTA_RULE_POSITION_ID attribute which works just like NFTA_RULE_POSITION but contains the ID of another rule within the same batch. This helps iptables-nft-restore handling dumps with mixed insert/append commands correctly. Note that NFTA_RULE_POSITION takes precedence over NFTA_RULE_POSITION_ID, so if the former is present, the latter is ignored. Signed-off-by: Phil Sutter <phil@nwl.cc> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: physdev: relax br_netfilter dependencyFlorian Westphal
Following command: iptables -D FORWARD -m physdev ... causes connectivity loss in some setups. Reason is that iptables userspace will probe kernel for the module revision of the physdev patch, and physdev has an artificial dependency on br_netfilter (xt_physdev use makes no sense unless a br_netfilter module is loaded). This causes the "phydev" module to be loaded, which in turn enables the "call-iptables" infrastructure. bridged packets might then get dropped by the iptables ruleset. The better fix would be to change the "call-iptables" defaults to 0 and enforce explicit setting to 1, but that breaks backwards compatibility. This does the next best thing: add a request_module call to checkentry. This was a stray '-D ... -m physdev' won't activate br_netfilter anymore. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: conntrack: remove helper hook againFlorian Westphal
place them into the confirm one. Old: hook (300): ipv4/6_help() first call helper, then seqadj. hook (INT_MAX): confirm Now: hook (INT_MAX): confirm, first call helper, then seqadj, then confirm Not having the extra call is noticeable in bechmarks. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: nf_tables: add direct calls for all builtin expressionsFlorian Westphal
With CONFIG_RETPOLINE its faster to add an if (ptr == &foo_func) check and and use direct calls for all the built-in expressions. ~15% improvement in pathological cases. checkpatch doesn't like the X macro due to the embedded return statement, but the macro has a very limited scope so I don't think its a problem. I would like to avoid bugs of the form If (e->ops->eval == (unsigned long)nft_foo_eval) nft_bar_eval(); and open-coded if ()/else if()/else cascade, thus the macro. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: nf_tables: handle nft_object lookups via rhltableFlorian Westphal
Instead of linear search, use rhlist interface to look up the objects. This fixes rulesets with thousands of named objects (quota, counters and the like). We only use a single table for this and consider the address of the table we're doing the lookup in as a part of the key. This reduces restore time of a sample ruleset with ~20k named counters from 37 seconds to 0.8 seconds. Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: nf_tables: prepare nft_object for lookups via hashtableFlorian Westphal
Add a 'key' structure for object, so we can look them up by name + table combination (the name can be the same in each table). Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-17tcp: move rx_opt & syn_data_acked init to tcp_disconnect()Eric Dumazet
If we make sure all listeners have these fields cleared, then a clone will also inherit zero values. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: move tp->rack init to tcp_disconnect()Eric Dumazet
If we make sure all listeners have proper tp->rack value, then a clone will also inherit proper initial value. Note that fresh sockets init tp->rack from tcp_init_sock() Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: move app_limited init to tcp_disconnect()Eric Dumazet
If we make sure all listeners have app_limited set to ~0U, then a clone will also inherit proper initial value. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: move retrans_out, sacked_out, tlp_high_seq, last_oow_ack_time init to ↵Eric Dumazet
tcp_disconnect() If we make sure all listeners have these fields cleared, then a clone will also inherit zero values. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: do not clear urg_data in tcp_create_openreq_childEric Dumazet
All listeners have this field cleared already, since tcp_disconnect() clears it and newly created sockets have also a zero value here. So a clone will inherit a zero value here. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: move snd_cwnd & snd_cwnd_cnt init to tcp_disconnect()Eric Dumazet
Passive connections can inherit proper value by cloning, if we make sure all listeners have the proper values there. tcp_disconnect() was setting snd_cwnd to 2, which seems quite obsolete since IW10 adoption. Also remove an obsolete comment. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: move mdev_us init to tcp_disconnect()Eric Dumazet
If we make sure a listener always has its mdev_us field set to TCP_TIMEOUT_INIT, we do not need to rewrite this field after a new clone is created. tcp_disconnect() is very seldom used in real applications. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: do not clear srtt_us in tcp_create_openreq_childEric Dumazet
All listeners have this field cleared already, since tcp_disconnect() clears it and newly created sockets have also a zero value here. So a clone will inherit a zero value here. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: do not clear packets_out in tcp_create_openreq_child()Eric Dumazet
New sockets have this field cleared, and tcp_disconnect() calls tcp_write_queue_purge() which among other things also clear tp->packets_out So a listener is guaranteed to have this field cleared. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: move icsk_rto init to tcp_disconnect()Eric Dumazet
If we make sure a listener always has its icsk_rto field set to TCP_TIMEOUT_INIT, we do not need to rewrite this field after a new clone is created. tcp_disconnect() is very seldom used in real applications. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: do not set snd_ssthresh in tcp_create_openreq_child()Eric Dumazet
New sockets get the field set to TCP_INFINITE_SSTHRESH in tcp_init_sock() In case a socket had this field changed and transitions to TCP_LISTEN state, tcp_disconnect() also makes sure snd_ssthresh is set to TCP_INFINITE_SSTHRESH. So a listener has this field set to TCP_INFINITE_SSTHRESH already. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17neighbour: Do not perturb drop profiles when neigh_probeYang Wei
Replace the kfree_skb() by consume_skb() to be drop monitor(dropwatch, perf) friendly. Signed-off-by: Yang Wei <yang.wei9@zte.com.cn> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tipc: remove unneeded semicolon in trace.cYueHaibing
Remove unneeded semicolon Signed-off-by: YueHaibing <yuehaibing@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17net: bridge: Fix ethernet header pointer before check skb forwardableYunjian Wang
The skb header should be set to ethernet header before using is_skb_forwardable. Because the ethernet header length has been considered in is_skb_forwardable(including dev->hard_header_len length). To reproduce the issue: 1, add 2 ports on linux bridge br using following commands: $ brctl addbr br $ brctl addif br eth0 $ brctl addif br eth1 2, the MTU of eth0 and eth1 is 1500 3, send a packet(Data 1480, UDP 8, IP 20, Ethernet 14, VLAN 4) from eth0 to eth1 So the expect result is packet larger than 1500 cannot pass through eth0 and eth1. But currently, the packet passes through success, it means eth1's MTU limit doesn't take effect. Fixes: f6367b4660dd ("bridge: use is_skb_forwardable in forward path") Cc: bridge@lists.linux-foundation.org Cc: Nkolay Aleksandrov <nikolay@cumulusnetworks.com> Cc: Roopa Prabhu <roopa@cumulusnetworks.com> Cc: Stephen Hemminger <stephen@networkplumber.org> Signed-off-by: Yunjian Wang <wangyunjian@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-18netfilter: nft_compat: destroy function must not have side effectsFlorian Westphal
The nft_compat destroy function deletes the nft_xt object from a list. This isn't allowed anymore. Destroy functions are called asynchronously, i.e. next batch can find the object that has a pending ->destroy() invocation: cpu0 cpu1 worker ->destroy for_each_entry() if (x == ... return x->ops; list_del(x) kfree_rcu(x) expr->ops->... // ops was free'd To resolve this, the list_del needs to occur before the transaction mutex gets released. nf_tables has a 'deactivate' hook for this purpose, so use that to unlink the object from the list. Fixes: 0935d5588400 ("netfilter: nf_tables: asynchronous release") Reported-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: nft_compat: make lists per netnsFlorian Westphal
There are two problems with nft_compat since the netlink config plane uses a per-netns mutex: 1. Concurrent add/del accesses to the same list 2. accesses to a list element after it has been free'd already. This patch fixes the first problem. Freeing occurs from a work queue, after transaction mutexes have been released, i.e., it still possible for a new transaction (even from same net ns) to find the to-be-deleted expression in the list. The ->destroy functions are not allowed to have any such side effects, i.e. the list_del() in the destroy function is not allowed. This part of the problem is solved in the next patch. I tried to make this work by serializing list access via mutex and by moving list_del() to a deactivate callback, but Taehee spotted following race on this approach: NET #0 NET #1 >select_ops() ->init() ->select_ops() ->deactivate() ->destroy() nft_xt_put() kfree_rcu(xt, rcu_head); ->init() <-- use-after-free occurred. Unfortunately, we can't increment reference count in select_ops(), because we can't undo the refcount increase in case a different expression fails in the same batch. (The destroy hook will only be called in case the expression was initialized successfully). Fixes: f102d66b335a ("netfilter: nf_tables: use dedicated mutex to guard transactions") Reported-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-18netfilter: nft_compat: use refcnt_t type for nft_xt reference countFlorian Westphal
Using standard integer type was fine while all operations on it were guarded by the nftnl subsys mutex. This isn't true anymore: 1. transactions are guarded only by a pernet mutex, so concurrent rule manipulation in different netns is racy 2. the ->destroy hook runs from a work queue after the transaction mutex has been released already. cpu0 cpu1 (net 1) cpu2 (net 2) kworker nft_compat->destroy nft_compat->init nft_compat->init if (--nft_xt->ref == 0) nft_xt->ref++ nft_xt->ref++ Switch to refcount_t. Doing this however only fixes a minor aspect, nft_compat also performs linked-list operations in an unsafe way. This is addressed in the next two patches. Fixes: f102d66b335a ("netfilter: nf_tables: use dedicated mutex to guard transactions") Fixes: 0935d5588400 ("netfilter: nf_tables: asynchronous release") Reported-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
2019-01-17af_packet: fix raw sockets over 6in4 tunnelNicolas Dichtel
Since commit cb9f1b783850, scapy (which uses an AF_PACKET socket in SOCK_RAW mode) is unable to send a basic icmp packet over a sit tunnel: Here is a example of the setup: $ ip link set ntfp2 up $ ip addr add 10.125.0.1/24 dev ntfp2 $ ip tunnel add tun1 mode sit ttl 64 local 10.125.0.1 remote 10.125.0.2 dev ntfp2 $ ip addr add fd00:cafe:cafe::1/128 dev tun1 $ ip link set dev tun1 up $ ip route add fd00:200::/64 dev tun1 $ scapy >>> p = [] >>> p += IPv6(src='fd00:100::1', dst='fd00:200::1')/ICMPv6EchoRequest() >>> send(p, count=1, inter=0.1) >>> quit() $ ip -s link ls dev tun1 | grep -A1 "TX.*errors" TX: bytes packets errors dropped carrier collsns 0 0 1 0 0 0 The problem is that the network offset is set to the hard_header_len of the output device (tun1, ie 14 + 20) and in our case, because the packet is small (48 bytes) the pskb_inet_may_pull() fails (it tries to pull 40 bytes (ipv6 header) starting from the network offset). This problem is more generally related to device with variable hard header length. To avoid a too intrusive patch in the current release, a (ugly) workaround is proposed in this patch. It has to be cleaned up in net-next. Link: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=993675a3100b1 Link: http://patchwork.ozlabs.org/patch/1024489/ Fixes: cb9f1b783850 ("ip: validate header length on virtual device xmit") CC: Willem de Bruijn <willemb@google.com> CC: Maxim Mikityanskiy <maximmi@mellanox.com> Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com> Acked-by: Willem de Bruijn <willemb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17net: add a route cache full diagnostic messagePeter Oskolkov
In some testing scenarios, dst/route cache can fill up so quickly that even an explicit GC call occasionally fails to clean it up. This leads to sporadically failing calls to dst_alloc and "network unreachable" errors to the user, which is confusing. This patch adds a diagnostic message to make the cause of the failure easier to determine. Signed-off-by: Peter Oskolkov <posk@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-18bpf: fix SO_MAX_PACING_RATE to support TCP internal pacingYuchung Cheng
If sch_fq packet scheduler is not used, TCP can fallback to internal pacing, but this requires sk_pacing_status to be properly set. Fixes: 8c4b4c7e9ff0 ("bpf: Add setsockopt helper function to bpf") Signed-off-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Lawrence Brakmo <brakmo@fb.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-01-18bpf: bpf_setsockopt: reset sock dst on SO_MARK changesPeter Oskolkov
In sock_setsockopt() (net/core/sock.h), when SO_MARK option is used to change sk_mark, sk_dst_reset(sk) is called. The same should be done in bpf_setsockopt(). Fixes: 8c4b4c7e9ff0 ("bpf: Add setsockopt helper function to bpf") Reported-by: Maciej Żenczykowski <maze@google.com> Signed-off-by: Peter Oskolkov <posk@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Reviewed-by: Maciej Żenczykowski <maze@google.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-01-17switchdev: Add extack argument to call_switchdev_notifiers()Petr Machata
A follow-up patch will enable vetoing of FDB entries. Make it possible to communicate details of why an FDB entry is not acceptable back to the user. Signed-off-by: Petr Machata <petrm@mellanox.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17net: Add extack argument to ndo_fdb_add()Petr Machata
Drivers may not be able to support certain FDB entries, and an error code is insufficient to give clear hints as to the reasons of rejection. In order to make it possible to communicate the rejection reason, extend ndo_fdb_add() with an extack argument. Adapt the existing implementations of ndo_fdb_add() to take the parameter (and ignore it). Pass the extack parameter when invoking ndo_fdb_add() from rtnl_fdb_add(). Signed-off-by: Petr Machata <petrm@mellanox.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: less aggressive window probing on local congestionYuchung Cheng
Previously when the sender fails to send (original) data packet or window probes due to congestion in the local host (e.g. throttling in qdisc), it'll retry within an RTO or two up to 500ms. In low-RTT networks such as data-centers, RTO is often far below the default minimum 200ms. Then local host congestion could trigger a retry storm pouring gas to the fire. Worse yet, the probe counter (icsk_probes_out) is not properly updated so the aggressive retry may exceed the system limit (15 rounds) until the packet finally slips through. On such rare events, it's wise to retry more conservatively (500ms) and update the stats properly to reflect these incidents and follow the system limit. Note that this is consistent with the behaviors when a keep-alive probe or RTO retry is dropped due to local congestion. Signed-off-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Neal Cardwell <ncardwell@google.com> Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: retry more conservatively on local congestionYuchung Cheng
Previously when the sender fails to retransmit a data packet on timeout due to congestion in the local host (e.g. throttling in qdisc), it'll retry within an RTO up to 500ms. In low-RTT networks such as data-centers, RTO is often far below the default minimum 200ms (and the cap 500ms). Then local host congestion could trigger a retry storm pouring gas to the fire. Worse yet, the retry counter (icsk_retransmits) is not properly updated so the aggressive retry may exceed the system limit (15 rounds) until the packet finally slips through. On such rare events, it's wise to retry more conservatively (500ms) and update the stats properly to reflect these incidents and follow the system limit. Note that this is consistent with the behavior when a keep-alive probe is dropped due to local congestion. Signed-off-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Neal Cardwell <ncardwell@google.com> Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: simplify window probe aborting on USER_TIMEOUTYuchung Cheng
Previously we use the next unsent skb's timestamp to determine when to abort a socket stalling on window probes. This no longer works as skb timestamp reflects the last instead of the first transmission. Instead we can estimate how long the socket has been stalling with the probe count and the exponential backoff behavior. Signed-off-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Neal Cardwell <ncardwell@google.com> Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: create a helper to model exponential backoffYuchung Cheng
Create a helper to model TCP exponential backoff for the next patch. This is pure refactor w no behavior change. Signed-off-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Neal Cardwell <ncardwell@google.com> Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: properly track retry time on passive Fast OpenYuchung Cheng
This patch addresses a corner issue on timeout behavior of a passive Fast Open socket. A passive Fast Open server may write and close the socket when it is re-trying SYN-ACK to complete the handshake. After the handshake is completely, the server does not properly stamp the recovery start time (tp->retrans_stamp is 0), and the socket may abort immediately on the very first FIN timeout, instead of retying until it passes the system or user specified limit. Signed-off-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Neal Cardwell <ncardwell@google.com> Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-17tcp: always set retrans_stamp on recoveryYuchung Cheng
Previously TCP socket's retrans_stamp is not set if the retransmission has failed to send. As a result if a socket is experiencing local issues to retransmit packets, determining when to abort a socket is complicated w/o knowning the starting time of the recovery since retrans_stamp may remain zero. This complication causes sub-optimal behavior that TCP may use the latest, instead of the first, retransmission time to compute the elapsed time of a stalling connection due to local issues. Then TCP may disrecard TCP retries settings and keep retrying until it finally succeed: not a good idea when the local host is already strained. The simple fix is to always timestamp the start of a recovery. It's worth noting that retrans_stamp is also used to compare echo timestamp values to detect spurious recovery. This patch does not break that because retrans_stamp is still later than when the original packet was sent. Signed-off-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Neal Cardwell <ncardwell@google.com> Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
generated by cgit (git 2.34.1) at 2025-09-16 13:14:53 +0000