- 07 Feb, 2022 10 commits
-
-
Danielle Ratson authored
Spectrum-2 supports an ACL action SIP_DIP, which allows IPv4 and IPv6 source and destination addresses change. Offload suitable mangles to the IPv4 address change action. Signed-off-by:
Danielle Ratson <danieller@nvidia.com> Reviewed-by:
Petr Machata <petrm@nvidia.com> Signed-off-by:
Ido Schimmel <idosch@nvidia.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Danielle Ratson authored
Add fields related to SIP_DIP_ACTION, which is used for changing of SIP and DIP addresses. Signed-off-by:
-
David S. Miller authored
Menglong Dong says: ==================== net: use kfree_skb_reason() for ip/udp packet receive In this series patches, kfree_skb() is replaced with kfree_skb_reason() during ipv4 and udp4 packet receiving path, and following drop reasons are introduced: SKB_DROP_REASON_SOCKET_FILTER SKB_DROP_REASON_NETFILTER_DROP SKB_DROP_REASON_OTHERHOST SKB_DROP_REASON_IP_CSUM SKB_DROP_REASON_IP_INHDR SKB_DROP_REASON_IP_RPFILTER SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST SKB_DROP_REASON_XFRM_POLICY SKB_DROP_REASON_IP_NOPROTO SKB_DROP_REASON_SOCKET_RCVBUFF SKB_DROP_REASON_PROTO_MEM TCP is more complex, so I left it in the next series. I just figure out how __print_symbolic() works. It doesn't base on the array index, but searching for symbols by loop. So I'm a little afraid it's performance. Changes since v3: - fix some small problems in the third patch (net: ipv4: use kfree_skb_reason() in ip_rcv_core()), as David Ahern said Changes since v2: - use SKB_DROP_REASON_PKT_TOO_SMALL for a path in ip_rcv_core() Changes since v1: - add document for all drop reasons, as David advised - remove unreleated cleanup - remove EARLY_DEMUX and IP_ROUTE_INPUT drop reason - replace {UDP, TCP}_FILTER with SOCKET_FILTER ==================== Signed-off-by: -
Menglong Dong authored
Replace kfree_skb() with kfree_skb_reason() in __udp_queue_rcv_skb(). Following new drop reasons are introduced: SKB_DROP_REASON_SOCKET_RCVBUFF SKB_DROP_REASON_PROTO_MEM Signed-off-by:
Menglong Dong <imagedong@tencent.com> Reviewed-by:
David Ahern <dsahern@kernel.org> Signed-off-by:
-
Menglong Dong authored
Replace kfree_skb() with kfree_skb_reason() in udp_queue_rcv_one_skb(). Signed-off-by:
-
Menglong Dong authored
Replace kfree_skb() with kfree_skb_reason() in ip_protocol_deliver_rcu(). Following new drop reasons are introduced: SKB_DROP_REASON_XFRM_POLICY SKB_DROP_REASON_IP_NOPROTO Signed-off-by:
-
Menglong Dong authored
Replace kfree_skb() with kfree_skb_reason() in ip_rcv_finish_core(), following drop reasons are introduced: SKB_DROP_REASON_IP_RPFILTER SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST Signed-off-by:
-
Menglong Dong authored
Replace kfree_skb() with kfree_skb_reason() in ip_rcv_core(). Three new drop reasons are introduced: SKB_DROP_REASON_OTHERHOST SKB_DROP_REASON_IP_CSUM SKB_DROP_REASON_IP_INHDR Signed-off-by:
-
Menglong Dong authored
Replace kfree_skb() with kfree_skb_reason() in nf_hook_slow() when skb is dropped by reason of NF_DROP. Following new drop reasons are introduced: SKB_DROP_REASON_NETFILTER_DROP Signed-off-by:
-
Menglong Dong authored
Add document for following existing drop reasons: SKB_DROP_REASON_NOT_SPECIFIED SKB_DROP_REASON_NO_SOCKET SKB_DROP_REASON_PKT_TOO_SMALL SKB_DROP_REASON_TCP_CSUM SKB_DROP_REASON_SOCKET_FILTER SKB_DROP_REASON_UDP_CSUM Signed-off-by:
-
- 06 Feb, 2022 3 commits
-
-
Eric Dumazet authored
After commit e9400660 ("lib/stackdepot: always do filter_irq_stacks() in stack_depot_save()") it became unnecessary to filter the stack before calling stack_depot_save(). Signed-off-by:
Eric Dumazet <edumazet@google.com> Cc: Marco Elver <elver@google.com> Cc: Alexander Potapenko <glider@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Signed-off-by:
-
Eric Dumazet authored
While testing a patch that will follow later ("net: add netns refcount tracker to struct nsproxy") I found that devtmpfs_init() was called before init_net was initialized. This is a bug, because devtmpfs_setup() calls ksys_unshare(CLONE_NEWNS); This has the effect of increasing init_net refcount, which will be later overwritten to 1, as part of setup_net(&init_net) We had too many prior patches [1] trying to work around the root cause. Really, make sure init_net is in BSS section, and that net_ns_init() is called earlier at boot time. Note that another patch ("vfs: add netns refcount tracker to struct fs_context") also will need net_ns_init() being called before vfs_caches_init() As a bonus, this patch saves around 4KB in .data section. [1] f8c46cb3 ("netns: do not call pernet ops for not yet set up init_net namespace") b5082df8 ("net: Initialise init_net.count to 1") 734b6541 ("net: Statically initialize init_net.dev_base_head") v2: fixed a build error reported by kernel build bots (CONFIG_NET=n) Signed-off-by: -
Juhee Kang authored
Currently, HSR manages mac addresses of known HSR nodes by using list_head. It takes a lot of time when there are a lot of registered nodes due to finding specific mac address nodes by using linear search. We can be reducing the time by using hlist. Thus, this patch moves list_head to hlist_head for mac addresses and this allows for further improvement of network performance. Condition: registered 10,000 known HSR nodes Before: # iperf3 -c 192.168.10.1 -i 1 -t 10 Connecting to host 192.168.10.1, port 5201 [ 5] local 192.168.10.2 port 59442 connected to 192.168.10.1 port 5201 [ ID] Interval Transfer Bitrate Retr Cwnd [ 5] 0.00-1.49 sec 3.75 MBytes 21.1 Mbits/sec 0 158 KBytes [ 5] 1.49-2.05 sec 1.25 MBytes 18.7 Mbits/sec 0 166 KBytes [ 5] 2.05-3.06 sec 2.44 MBytes 20.3 Mbits/sec 56 16.9 KBytes [ 5] 3.06-4.08 sec 1.43 MBytes 11.7 Mbits/sec 11 38.0 KBytes [ 5] 4.08-5.00 sec 951 KBytes 8.49 Mbits/sec 0 56.3 KBytes After: # iperf3 -c 192.168.10.1 -i 1 -t 10 Connecting to host 192.168.10.1, port 5201 [ 5] local 192.168.10.2 port 36460 connected to 192.168.10.1 port 5201 [ ID] Interval Transfer Bitrate Retr Cwnd [ 5] 0.00-1.00 sec 7.39 MBytes 62.0 Mbits/sec 3 130 KBytes [ 5] 1.00-2.00 sec 5.06 MBytes 42.4 Mbits/sec 16 113 KBytes [ 5] 2.00-3.00 sec 8.58 MBytes 72.0 Mbits/sec 42 94.3 KBytes [ 5] 3.00-4.00 sec 7.44 MBytes 62.4 Mbits/sec 2 131 KBytes [ 5] 4.00-5.07 sec 8.13 MBytes 63.5 Mbits/sec 38 92.9 KBytes Signed-off-by:Juhee Kang <claudiajkang@gmail.com> Signed-off-by:
-
- 05 Feb, 2022 27 commits
-
-
Eric Dumazet authored
We have plans for increasing MAX_SKB_FRAGS, but sk_msg_sg::copy is currently an unsigned long, limiting MAX_SKB_FRAGS to 30 on 32bit arches. Convert it to a bitmap, as Jakub suggested. Signed-off-by:
-
Eric Dumazet authored
Instead of disabling TSO at compile time if MAX_SKB_FRAGS > 32, implement ndo_features_check() method for this driver for a more dynamic handling. If skb has more than 32 frags and is a GSO packet, force software segmentation. Most locally generated packets will use a small number of fragments anyway. For forwarding workloads, we can limit gro_max_size at ingress, we might also implement gro_max_segs if needed. Signed-off-by:
-
Gustavo A. R. Silva authored
It seems this one-element array is not actually being used as an array of variable size, so we can just replace it with just a non-array object of type struct desc_frag and refactor a bit the rest of the code. This helps with the ongoing efforts to globally enable -Warray-bounds and get us closer to being able to tighten the FORTIFY_SOURCE routines on memcpy(). This issue was found with the help of Coccinelle and audited and fixed, manually. [1] https://en.wikipedia.org/wiki/Flexible_array_member [2] https://www.kernel.org/doc/html/v5.16/process/deprecated.html#zero-length-and-one-element-arrays Link: https://github.com/KSPP/linux/issues/79Signed-off-by:
Gustavo A. R. Silva <gustavoars@kernel.org> Reviewed-by:
Jakub Kicinski <kuba@kernel.org> Signed-off-by:
-
Gustavo A. R. Silva authored
There is a regular need in the kernel to provide a way to declare having a dynamically sized set of trailing elements in a structure. Kernel code should always use “flexible array members”[1] for these cases. The older style of one-element or zero-length arrays should no longer be used[2]. This helps with the ongoing efforts to globally enable -Warray-bounds and get us closer to being able to tighten the FORTIFY_SOURCE routines on memcpy(). This issue was found with the help of Coccinelle and audited and fixed, manually. [1] https://en.wikipedia.org/wiki/Flexible_array_member [2] https://www.kernel.org/doc/html/v5.16/process/deprecated.html#zero-length-and-one-element-arrays Link: https://github.com/KSPP/linux/issues/79Signed-off-by:
-
David S. Miller authored
Haiyang Zhang says: ==================== net: mana: Add handling of CQE_RX_TRUNCATED and a cleanup Add handling of CQE_RX_TRUNCATED and a cleanup patch ==================== Signed-off-by: -
Haiyang Zhang authored
The switch statement already ensures cqe_type == CQE_RX_OKAY at that point. Signed-off-by:
Haiyang Zhang <haiyangz@microsoft.com> Reviewed-by:
Dexuan Cui <decui@microsoft.com> Signed-off-by:
-
Haiyang Zhang authored
The proper way to drop this kind of CQE is advancing rxq tail without indicating the packet to the upper network layer. Signed-off-by:
-
David S. Miller authored
Eric Dumazet says: ==================== net: device tracking improvements Main goal of this series is to be able to detect the following case which apparently is still haunting us. dev_hold_track(dev, tracker_1, GFP_ATOMIC); dev_hold(dev); dev_put(dev); dev_put(dev); // Should complain loudly here. dev_put_track(dev, tracker_1); // instead of here (as before this series) v2: third patch: I replaced the dev_put() in linkwatch_do_dev() with __dev_put(). ==================== Signed-off-by: -
Eric Dumazet authored
We are still chasing some syzbot reports where we think a rogue dev_put() is called with no corresponding prior dev_hold(). Unfortunately it eats a reference on dev->dev_refcnt taken by innocent dev_hold_track(), meaning that the refcount saturation splat comes too late to be useful. Make sure that 'not tracked' dev_put() and dev_hold() better use CONFIG_NET_DEV_REFCNT_TRACKER=y debug infrastructure: Prior patch in the series allowed ref_tracker_alloc() and ref_tracker_free() to be called with a NULL @trackerp parameter, and to use a separate refcount only to detect too many put() even in the following case: dev_hold_track(dev, tracker_1, GFP_ATOMIC); dev_hold(dev); dev_put(dev); dev_put(dev); // Should complain loudly here. dev_put_track(dev, tracker_1); // instead of here Add clarification about netdev_tracker_alloc() role. v2: I replaced the dev_put() in linkwatch_do_dev() with __dev_put() because callers called netdev_tracker_free(). Signed-off-by: -
Eric Dumazet authored
We are still chasing a netdev refcount imbalance, and we suspect we have one rogue dev_put() that is consuming a reference taken from a dev_hold_track() To detect this case, allow ref_tracker_alloc() and ref_tracker_free() to be called with a NULL @trackerp parameter, and use a dedicated refcount_t just for them. Signed-off-by:
-
Eric Dumazet authored
Whenever ref_tracker_dir_init() is called, mark the struct ref_tracker_dir as dead. Test the dead status from ref_tracker_alloc() and ref_tracker_free() This should detect buggy dev_put()/dev_hold() happening too late in netdevice dismantle process. Signed-off-by:
-
David S. Miller authored
Eric Dumazet says: ==================== ipv6: mc_forwarding changes First patch removes minor data-races, as mc_forwarding can be locklessly read in fast path. Second patch adds a short cut in ip6mr_sk_done() ==================== Signed-off-by: -
Eric Dumazet authored
In many cases, ip6mr_sk_done() is called while no ipmr socket has been registered. This removes 4 rtnl acquisitions per netns dismantle, with following callers: igmp6_net_exit(), tcpv6_net_exit(), ndisc_net_exit() Signed-off-by:
-
Eric Dumazet authored
This fixes minor data-races in ip6_mc_input() and batadv_mcast_mla_rtr_flags_softif_get_ipv6() Signed-off-by:
-
Jakub Kicinski authored
We generally default the vendor to y and the drivers itself to n. NET_DSA_REALTEK, however, selects a whole bunch of things, so it's not a pure "vendor selection" knob. Let's default it all to n. Signed-off-by:
Florian Fainelli <f.fainelli@gmail.com> Acked-by:
Arınç ÜNAL <arinc.unal@arinc9.com> Reviewed-by:
Linus Walleij <linus.walleij@linaro.org> Signed-off-by:
-
Russell King (Oracle) authored
Phylink will use PCS polling whenever phylink_config.pcs_poll or the phylink_pcs poll member is set. As this driver sets both, remove the former. Signed-off-by:
Russell King (Oracle) <rmk+kernel@armlinux.org.uk> Signed-off-by:
-
Russell King (Oracle) authored
phylink_set_10g_modes() is no longer used with the conversion of drivers to phylink_generic_validate(), so we can remove it. Signed-off-by:
-
David S. Miller authored
Paolo Abeni says: ==================== gro: a couple of minor optimization This series collects a couple of small optimizations for the GRO engine, reducing slightly the number of cycles for dev_gro_receive(). The delta is within noise range in tput tests, but with big TCP coming every cycle saved from the GRO engine will count - I hope ;) v1 -> v2: - a few cleanup suggested from Alexander(s) - moved away the more controversial 3rd patch ==================== Signed-off-by: -
Paolo Abeni authored
While inspecting some perf report, I noticed that the compiler emits suboptimal code for the napi CB initialization, fetching and storing multiple times the memory for flags bitfield. This is with gcc 10.3.1, but I observed the same with older compiler versions. We can help the compiler to do a nicer work clearing several fields at once using an u32 alias. The generated code is quite smaller, with the same number of conditional. Before: objdump -t net/core/gro.o | grep " F .text" 0000000000000bb0 l F .text 0000000000000357 dev_gro_receive After: 0000000000000bb0 l F .text 000000000000033c dev_gro_receive v1 -> v2: - use struct_group (Alexander and Alex) RFC -> v1: - use __struct_group to delimit the zeroed area (Alexander) Signed-off-by:
Paolo Abeni <pabeni@redhat.com> Signed-off-by:
-
Paolo Abeni authored
After commit 5e10da53 ("skbuff: allow 'slow_gro' for skb carring sock reference") and commit af352460 ("net: fix GRO skb truesize update") the truesize of the skb with stolen head is properly updated by the GRO engine, we don't need anymore resetting it at recycle time. v1 -> v2: - clarify the commit message (Alexander) Signed-off-by:
-
Dan Carpenter authored
This is a copy and paste bug. It was supposed to check "clear_skb" instead of "write_skb". Fixes: 2cd54856 ("net: dsa: qca8k: add support for phy read/write with mgmt Ethernet") Signed-off-by:
Dan Carpenter <dan.carpenter@oracle.com> Reviewed-by:
-
David S. Miller authored
Horatiu Vultur says: ==================== net: lan966x: add support for mcast snooping Implement the switchdev callback SWITCHDEV_ATTR_ID_BRIDGE_MC_DISABLED to allow to enable/disable multicast snooping. ==================== Signed-off-by: -
Horatiu Vultur authored
When the multicast snooping is disabled, the mdb entries should be removed from the HW, but they still need to be kept in memory for when the mcast_snooping will be enabled again. Signed-off-by:
Horatiu Vultur <horatiu.vultur@microchip.com> Signed-off-by:
-
Horatiu Vultur authored
The callback allows to enable/disable multicast snooping. When the snooping is enabled, all IGMP and MLD frames are redirected to the CPU, therefore make sure not to set the skb flag 'offload_fwd_mark'. The HW will not flood multicast ipv4/ipv6 data frames. When the snooping is disabled, the HW will flood IGMP, MLD and multicast ipv4/ipv6 frames according to the mcast_flood flag. Signed-off-by:
-
Horatiu Vultur authored
When enabling the multicast snooping, the forwarding of the IPV6 frames has it's own forwarding mask. Signed-off-by:
-
Paul Blakey authored
Currently tc skb extension is used to send miss info from tc to ovs datapath module, and driver to tc. For the tc to ovs miss it is currently always allocated even if it will not be used by ovs datapath (as it depends on a requested feature). Export the static key which is used by openvswitch module to guard this code path as well, so it will be skipped if ovs datapath doesn't need it. Enable this code path once ovs datapath needs it. Signed-off-by:
Paul Blakey <paulb@nvidia.com> Reviewed-by:
Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by:
-
Jakub Kicinski authored
Mat Martineau says: ==================== mptcp: Improve set-flags command and update self tests Patches 1-3 allow more flexibility in the combinations of features and flags allowed with the MPTCP_PM_CMD_SET_FLAGS netlink command, and add self test case coverage for the new functionality. Patches 4-6 and 9 refactor the mptcp_join.sh self tests to allow them to configure all of the test cases using either the pm_nl_ctl utility (part of the mptcp self tests) or the 'ip mptcp' command (from iproute2). The default remains to use pm_nl_ctl. Patches 7 and 8 update the pm_netlink.sh self tests to cover the use of endpoint ids to set endpoint flags (instead of just addresses). ==================== Link: https://lore.kernel.org/r/20220205000337.187292-1-mathew.j.martineau@linux.intel.comSigned-off-by:
-
