- 14 Aug, 2023 24 commits
-
-
William Tu authored
The patch adds native-mode XDP support: XDP DROP, PASS, TX, and REDIRECT. Background: The vmxnet3 rx consists of three rings: ring0, ring1, and dataring. For r0 and r1, buffers at r0 are allocated using alloc_skb APIs and dma mapped to the ring's descriptor. If LRO is enabled and packet size larger than 3K, VMXNET3_MAX_SKB_BUF_SIZE, then r1 is used to mapped the rest of the buffer larger than VMXNET3_MAX_SKB_BUF_SIZE. Each buffer in r1 is allocated using alloc_page. So for LRO packets, the payload will be in one buffer from r0 and multiple from r1, for non-LRO packets, only one descriptor in r0 is used for packet size less than 3k. When receiving a packet, the first descriptor will have the sop (start of packet) bit set, and the last descriptor will have the eop (end of packet) bit set. Non-LRO packets will have only one descriptor with both sop and eop set. Other than r0 and r1, vmxnet3 dataring is specifically designed for handling packets with small size, usually 128 bytes, defined in VMXNET3_DEF_RXDATA_DESC_SIZE, by simply copying the packet from the backend driver in ESXi to the ring's memory region at front-end vmxnet3 driver, in order to avoid memory mapping/unmapping overhead. In summary, packet size: A. < 128B: use dataring B. 128B - 3K: use ring0 (VMXNET3_RX_BUF_SKB) C. > 3K: use ring0 and ring1 (VMXNET3_RX_BUF_SKB + VMXNET3_RX_BUF_PAGE) As a result, the patch adds XDP support for packets using dataring and r0 (case A and B), not the large packet size when LRO is enabled. XDP Implementation: When user loads and XDP prog, vmxnet3 driver checks configurations, such as mtu, lro, and re-allocate the rx buffer size for reserving the extra headroom, XDP_PACKET_HEADROOM, for XDP frame. The XDP prog will then be associated with every rx queue of the device. Note that when using dataring for small packet size, vmxnet3 (front-end driver) doesn't control the buffer allocation, as a result we allocate a new page and copy packet from the dataring to XDP frame. The receive side of XDP is implemented for case A and B, by invoking the bpf program at vmxnet3_rq_rx_complete and handle its returned action. The vmxnet3_process_xdp(), vmxnet3_process_xdp_small() function handles the ring0 and dataring case separately, and decides the next journey of the packet afterward. For TX, vmxnet3 has split header design. Outgoing packets are parsed first and protocol headers (L2/L3/L4) are copied to the backend. The rest of the payload are dma mapped. Since XDP_TX does not parse the packet protocol, the entire XDP frame is dma mapped for transmission and transmitted in a batch. Later on, the frame is freed and recycled back to the memory pool. Performance: Tested using two VMs inside one ESXi vSphere 7.0 machine, using single core on each vmxnet3 device, sender using DPDK testpmd tx-mode attached to vmxnet3 device, sending 64B or 512B UDP packet. VM1 txgen: $ dpdk-testpmd -l 0-3 -n 1 -- -i --nb-cores=3 \ --forward-mode=txonly --eth-peer=0,<mac addr of vm2> option: add "--txonly-multi-flow" option: use --txpkts=512 or 64 byte VM2 running XDP: $ ./samples/bpf/xdp_rxq_info -d ens160 -a <options> --skb-mode $ ./samples/bpf/xdp_rxq_info -d ens160 -a <options> options: XDP_DROP, XDP_PASS, XDP_TX To test REDIRECT to cpu 0, use $ ./samples/bpf/xdp_redirect_cpu -d ens160 -c 0 -e drop Single core performance comparison with skb-mode. 64B: skb-mode -> native-mode XDP_DROP: 1.6Mpps -> 2.4Mpps XDP_PASS: 338Kpps -> 367Kpps XDP_TX: 1.1Mpps -> 2.3Mpps REDIRECT-drop: 1.3Mpps -> 2.3Mpps 512B: skb-mode -> native-mode XDP_DROP: 863Kpps -> 1.3Mpps XDP_PASS: 275Kpps -> 376Kpps XDP_TX: 554Kpps -> 1.2Mpps REDIRECT-drop: 659Kpps -> 1.2Mpps Demo: https://youtu.be/4lm1CSCi78Q Future work: - XDP frag support - use napi_consume_skb() instead of dev_kfree_skb_any at unmap - stats using u64_stats_t - using bitfield macro BIT() - optimization for DMA synchronization using actual frame length, instead of always max_len Signed-off-by:William Tu <u9012063@gmail.com> Reviewed-by:
Alexander Duyck <alexanderduyck@fb.com> Reviewed-by:
Alexander Lobakin <alexandr.lobakin@intel.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
David S. Miller authored
Adrian Moreno says: ==================== openvswitch: add drop reasons There is currently a gap in drop visibility in the openvswitch module. This series tries to improve this by adding a new drop reason subsystem for OVS. Apart from adding a new drop reasson subsystem and some common drop reasons, this series takes Eric's preliminary work [1] on adding an explicit drop action and integrates it into the same subsystem. A limitation of this series is that it does not report upcall errors. The reason is that there could be many sources of upcall drops and the most common one, which is the netlink buffer overflow, cannot be reported via kfree_skb() because the skb is freed in the netlink layer (see [2]). Therefore, using a reason for the rare events and not the common one would be even more misleading. I'd propose we add (in a follow up patch) a tracepoint to better report upcall errors. [1] https://lore.kernel.org/netdev/202306300609.tdRdZscy-lkp@intel.com/T/ [2] commit 1100248a ("openvswitch: Fix double reporting of drops in dropwatch") --- v4 -> v5: - Rebased - Added a helper function to explicitly convert drop reason enum types v3 -> v4: - Changed names of errors following Ilya's suggestions - Moved the ovs-dpctl.py changes from patch 7/7 to 3/7 - Added a test to ensure actions following a drop are rejected rfc2 -> v3: - Rebased on top of latest net-next rfc1 -> rfc2: - Fail when an explicit drop is not the last - Added a drop reason for action errors - Added braces around macros - Dropped patch that added support for masks in ovs-dpctl.py as it's now included in Aaron's series [2]. ==================== Signed-off-by:
-
Adrian Moreno authored
Test explicit drops generate the right drop reason. Also, verify that the kernel rejects flows with actions following an explicit drop. Acked-by:
Aaron Conole <aconole@redhat.com> Signed-off-by:
Adrian Moreno <amorenoz@redhat.com> Signed-off-by:
-
Adrian Moreno authored
Test if the correct drop reason is reported when OVS drops a packet due to an explicit flow. Acked-by:
-
Adrian Moreno authored
Use drop reasons from include/net/dropreason-core.h when a reasonable candidate exists. Acked-by:
-
Adrian Moreno authored
By using an independent drop reason it makes it easy to distinguish between QoS-triggered or flow-triggered drop. Acked-by:
-
Eric Garver authored
From: Eric Garver <eric@garver.life> This adds an explicit drop action. This is used by OVS to drop packets for which it cannot determine what to do. An explicit action in the kernel allows passing the reason _why_ the packet is being dropped or zero to indicate no particular error happened (i.e: OVS intentionally dropped the packet). Since the error codes coming from userspace mean nothing for the kernel, we squash all of them into only two drop reasons: - OVS_DROP_EXPLICIT_WITH_ERROR to indicate a non-zero value was passed - OVS_DROP_EXPLICIT to indicate a zero value was passed (no error) e.g. trace all OVS dropped skbs # perf trace -e skb:kfree_skb --filter="reason >= 0x30000" [..] 106.023 ping/2465 skb:kfree_skb(skbaddr: 0xffffa0e8765f2000, \ location:0xffffffffc0d9b462, protocol: 2048, reason: 196611) reason: 196611 --> 0x30003 (OVS_DROP_EXPLICIT) Also, this patch allows ovs-dpctl.py to add explicit drop actions as: "drop" -> implicit empty-action drop "drop(0)" -> explicit non-error action drop "drop(42)" -> explicit error action drop Signed-off-by:
Eric Garver <eric@garver.life> Co-developed-by:
-
Adrian Moreno authored
Add a drop reason for packets that are dropped because an action returns a non-zero error code. Acked-by:
-
Adrian Moreno authored
Create a new drop reason subsystem for openvswitch and add the first drop reason to represent last-action drops. Last-action drops happen when a flow has an empty action list or there is no action that consumes the packet (output, userspace, recirc, etc). It is the most common way in which OVS drops packets. Implementation-wise, most of these skb-consuming actions already call "consume_skb" internally and return directly from within the do_execute_actions() loop so with minimal changes we can assume that any skb that exits the loop normally is a packet drop. Signed-off-by:
-
David S. Miller authored
Matthieu Baerts says: ==================== mptcp: get rid of msk->subflow The MPTCP protocol maintains an additional struct socket per connection, mainly to be able to easily use tcp-level struct socket operations. This leads to several side effects, beyond the quite unfortunate / confusing 'subflow' field name: - active and passive sockets behaviour is inconsistent: only active ones have a not NULL msk->subflow, leading to different error handling and different error code returned to the user-space in several places. - active sockets uses an unneeded, larger amount of memory - passive sockets can't successfully go through accept(), disconnect(), accept() sequence, see [1] for more details. The 13 first patches of this series are from Paolo and address all the above, finally getting rid of the blamed field: - The first patch is a minor clean-up. - In the next 11 patches, msk->subflow usage is systematically removed from the MPTCP protocol, replacing it with direct msk->first usage, eventually introducing new core helpers when needed. - The 13th patch finally disposes the field, and it's the only patch in the series intended to produce functional changes. The last and 14th patch is from Kuniyuki and it is not linked to the previous ones: it is a small clean-up to get rid of an unnecessary check in mptcp_init_sock(). [1] https://github.com/multipath-tcp/mptcp_net-next/issues/290 ==================== Signed-off-by:
Matthieu Baerts <matthieu.baerts@tessares.net>
-
Kuniyuki Iwashima authored
__mptcp_init_sock() always returns 0 because mptcp_init_sock() used to return the value directly. But after commit 18b683bf ("mptcp: queue data for mptcp level retransmission"), __mptcp_init_sock() need not return value anymore. Let's remove the unnecessary test for __mptcp_init_sock() and make it return void. Signed-off-by:
Kuniyuki Iwashima <kuniyu@amazon.com> Reviewed-by:
-
Paolo Abeni authored
Such field is now unused just as a flag to control the first subflow deletion at close() time. Introduce a new bit flag for that and finally drop the mentioned field. As an intended side effect, now the first subflow sock is not freed before close() even for passive sockets. The msk has no open/active subflows if the first one is closed and the subflow list is singular, update accordingly the state check in mptcp_stream_accept(). Among other benefits, the subflow removal, reduces the amount of memory used on the client side for each mptcp connection, allows passive sockets to go through successful accept()/disconnect()/connect() and makes return error code consistent for failing both passive and active sockets. Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/290Signed-off-by:
Paolo Abeni <pabeni@redhat.com> Reviewed-by:
Mat Martineau <martineau@kernel.org> Signed-off-by:
-
Paolo Abeni authored
After the previous patch the __mptcp_nmpc_socket helper is used only to ensure that the MPTCP socket is a suitable status - that is, the mptcp capable handshake is not started yet. Change the return value to the relevant subflow sock, to finally remove the last references to first subflow socket in the MPTCP stack. As a bonus, we can get rid of a few local variables in different functions. No functional change intended. Signed-off-by:
-
Paolo Abeni authored
This is one of the few remaining spots actually manipulating the first subflow socket. We can leverage the recently introduced inet helpers to get rid of ssock there. No functional changes intended. Signed-off-by:
-
Paolo Abeni authored
The mptcp sockopt infrastructure unneedly uses the first subflow socket struct in a few spots. We are going to remove such field soon, so use directly the first subflow sock instead. No functional changes intended. Signed-off-by:
-
Paolo Abeni authored
We are going to remove the first subflow socket soon, so avoid the additional indirection at accept() time. Instead access directly the first subflow sock, and update mptcp_accept() to operate on it. This allows dropping a duplicated check in mptcp_accept(). No functional changes intended. Signed-off-by:
-
Paolo Abeni authored
We are going to remove the first subflow socket soon, so avoid the additional indirection at poll() time. Instead access directly the first subflow sock. No functional changes intended. Signed-off-by:
-
Paolo Abeni authored
We are going to remove the first subflow socket soon, so avoid the additional indirection via at listen() time. Instead call directly the recently introduced helper on the first subflow sock. No functional changes intended. Signed-off-by:
-
Paolo Abeni authored
The mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is inet_listen(). Factor out an helper operating directly on the (locked) struct sock, to allow get rid of the above dependency in the next patch without duplicating the existing code. No functional changes intended. Signed-off-by:
-
Paolo Abeni authored
We are going to remove the first subflow socket soon, so avoid the additional indirection via at bind() time. Instead call directly the recently introduced helpers on the first subflow sock. Signed-off-by:
-
Paolo Abeni authored
The mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is bind(). Factor out the helpers operating directly on the struct sock, to allow get rid of the above dependency in the next patch without duplicating the existing code. No functional changes intended. Signed-off-by:
-
Paolo Abeni authored
We are going to remove the first subflow socket soon, so avoid accessing it in mptcp_get_port(). Instead, access directly the first subflow sock. Signed-off-by:
-
Paolo Abeni authored
The mptcp protocol maintains an additional socket just to easily invoke a few stream operations on the first subflow. One of them is __inet_stream_connect(). We are going to remove the first subflow socket soon, so avoid the additional indirection via at connect time, calling directly into the sock-level connect() ops. The sk-level connect never return -EINPROGRESS, cleanup the error path accordingly. Additionally, the ssk status on error is always TCP_CLOSE. Avoid unneeded access to the subflow sk state. No functional change intended. Signed-off-by:
-
Paolo Abeni authored
The MPTCP protocol currently clears the msk token both at connect() and listen() time. That is needed to deal with failing connect() calls that can create a new token while leaving the sk in TCP_CLOSE,SS_UNCONNECTED status and thus allowing later connect() and/or listen() calls. Let's deal with such failures explicitly, cleaning the token in a timely manner and avoid the confusing early mptcp_token_destroy(). Signed-off-by:
-
- 13 Aug, 2023 16 commits
-
-
Jörn-Thorben Hinz authored
Commit db3685b4 ("net: remove obsolete members from struct net") removed the uses of struct list_head from this header, without removing the corresponding included header. Signed-off-by:
Jörn-Thorben Hinz <jthinz@mailbox.tu-berlin.de> Reviewed-by:
Simon Horman <horms@kernel.org> Signed-off-by:
-
David S. Miller authored
Merge tag 'for-net-next-2023-08-11' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth-next bluetooth-next pull request for net-next: - Add new VID/PID for Mediatek MT7922 - Add support multiple BIS/BIG - Add support for Intel Gale Peak - Add support for Qualcomm WCN3988 - Add support for BT_PKT_STATUS for ISO sockets - Various fixes for experimental ISO support - Load FW v2 for RTL8852C - Add support for NXP AW693 chipset - Add support for Mediatek MT2925
-
Vladimir Oltean authored
lynx_pcs_link_up_sgmii() is supposed to update the PCS speed and duplex for the non-inband operating modes, and prior to the blamed commit, it did just that, but a mistake sneaked into the conversion and reversed the condition. It is easy for this to go undetected on platforms that also initialize the PCS in the bootloader, because Linux doesn't reset it (although maybe it should). The nature of the bug is that phylink will not touch the IF_MODE_HALF_DUPLEX | IF_MODE_SPEED_MSK fields when it should, and it will apparently keep working if the previous values set by the bootloader were correct. Fixes: c689a652 ("net: pcs: lynx: update PCS driver to use neg_mode") Signed-off-by:
Vladimir Oltean <vladimir.oltean@nxp.com> Reviewed-by:
Russell King (Oracle) <rmk+kernel@armlinux.org.uk> Signed-off-by:
-
David S. Miller authored
Zheng Zengkai says: ==================== net: Use pci_dev_id() to simplify the code PCI core API pci_dev_id() can be used to get the BDF number for a pci device. Use the API to simplify the code. ==================== Signed-off-by: -
Zheng Zengkai authored
PCI core API pci_dev_id() can be used to get the BDF number for a pci device. We don't need to compose it manually. Use pci_dev_id() to simplify the code a little bit. Signed-off-by:
Zheng Zengkai <zhengzengkai@huawei.com> Signed-off-by:
-
Zheng Zengkai authored
PCI core API pci_dev_id() can be used to get the BDF number for a pci device. We don't need to compose it manually. Use pci_dev_id() to simplify the code a little bit. Signed-off-by:
-
Zheng Zengkai authored
PCI core API pci_dev_id() can be used to get the BDF number for a pci device. We don't need to compose it manually. Use pci_dev_id() to simplify the code a little bit. Signed-off-by:
-
Zheng Zengkai authored
PCI core API pci_dev_id() can be used to get the BDF number for a pci device. We don't need to compose it manually. Use pci_dev_id() to simplify the code a little bit. Signed-off-by:
-
Zheng Zengkai authored
PCI core API pci_dev_id() can be used to get the BDF number for a pci device. We don't need to compose it manually. Use pci_dev_id() to simplify the code a little bit. Signed-off-by:
-
Yue Haibing authored
Commit 675ad473 ("e1000: Use netdev_<level>, pr_<level> and dev_<level>") declared but never implemented e1000_get_hw_dev_name(). Commit 1532ecea ("e1000: drop dead pcie code from e1000") removed e1000_check_mng_mode()/e1000_blink_led_start() but not the declarations. Commit c46b59b2 ("e1000: Remove unused function e1000_mta_set.") removed e1000_mta_set() but not its declaration. Signed-off-by:
Yue Haibing <yuehaibing@huawei.com> Reviewed-by:
-
Yue Haibing authored
Commit 39de8281 ("RDS: Main header file") declared but never implemented rds_trans_init() and rds_trans_exit(), remove it. Commit d37c9359 ("RDS: Move loop-only function to loop.c") removed the implementation rds_message_inc_free() but not the declaration. Since commit 55b7ed0b ("RDS: Common RDMA transport code") rds_rdma_conn_connect() is never implemented and used. rds_tcp_map_seq() is never implemented and used since commit 70041088 ("RDS: Add TCP transport to RDS"). Signed-off-by:
-
Eric Dumazet authored
sk_diag_put_flags(), netlink_setsockopt(), netlink_getsockopt() and others use nlk->flags without correct locking. Use set_bit(), clear_bit(), test_bit(), assign_bit() to remove data-races. Reported-by:
syzbot <syzkaller@googlegroups.com> Signed-off-by:
Eric Dumazet <edumazet@google.com> Reviewed-by:
-
David S. Miller authored
Menglong Dong says: ==================== net: tcp: support probing OOM In this series, we make some small changes to make the tcp retransmission become zero-window probes if the receiver drops the skb because of memory pressure. In the 1st patch, we reply a zero-window ACK if the skb is dropped because out of memory, instead of dropping the skb silently. In the 2nd patch, we allow a zero-window ACK to update the window. In the 3rd patch, fix unexcepted socket die when snd_wnd is 0 in tcp_retransmit_timer(). In the 4th patch, we refactor the debug message in tcp_retransmit_timer() to make it more correct. After these changes, the tcp can probe the OOM of the receiver forever. Changes since v3: - make the timeout "2 * TCP_RTO_MAX" in the 3rd patch - tp->retrans_stamp is not based on jiffies and can't be compared with icsk->icsk_timeout in the 3rd patch. Fix it. - introduce the 4th patch Changes since v2: - refactor the code to avoid code duplication in the 1st patch - use after() instead of max() in tcp_rtx_probe0_timed_out() Changes since v1: - send 0 rwin ACK for the receive queue empty case when necessary in the 1st patch - send the ACK immediately by using the ICSK_ACK_NOW flag in the 1st patch - consider the case of the connection restart from idle, as Neal comment, in the 3rd patch ==================== Signed-off-by: -
Menglong Dong authored
The debug message in tcp_retransmit_timer() is slightly wrong, because they could be printed even if we did not receive a new ACK packet from the remote peer. Change it to probing zero-window, as it is a expected case now. The description may be not correct. Adding the duration since the last ACK we received, and the duration of the retransmission, which are useful for debugging. And the message now like this: Probing zero-window on 127.0.0.1:9999/46946, seq=3737778959:3737791503, recv 209ms ago, lasting 209ms Probing zero-window on 127.0.0.1:9999/46946, seq=3737778959:3737791503, recv 404ms ago, lasting 408ms Probing zero-window on 127.0.0.1:9999/46946, seq=3737778959:3737791503, recv 812ms ago, lasting 1224ms Signed-off-by:
Menglong Dong <imagedong@tencent.com> Reviewed-by:
-
Menglong Dong authored
In tcp_retransmit_timer(), a window shrunk connection will be regarded as timeout if 'tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX'. This is not right all the time. The retransmits will become zero-window probes in tcp_retransmit_timer() if the 'snd_wnd==0'. Therefore, the icsk->icsk_rto will come up to TCP_RTO_MAX sooner or later. However, the timer can be delayed and be triggered after 122877ms, not TCP_RTO_MAX, as I tested. Therefore, 'tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX' is always true once the RTO come up to TCP_RTO_MAX, and the socket will die. Fix this by replacing the 'tcp_jiffies32' with '(u32)icsk->icsk_timeout', which is exact the timestamp of the timeout. However, "tp->rcv_tstamp" can restart from idle, then tp->rcv_tstamp could already be a long time (minutes or hours) in the past even on the first RTO. So we double check the timeout with the duration of the retransmission. Meanwhile, making "2 * TCP_RTO_MAX" as the timeout to avoid the socket dying too soon. Fixes: 1da177e4 ("Linux-2.6.12-rc2") Link: https://lore.kernel.org/netdev/CADxym3YyMiO+zMD4zj03YPM3FBi-1LHi6gSD2XT8pyAMM096pg@mail.gmail.com/Signed-off-by:
-
Menglong Dong authored
Fow now, an ACK can update the window in following case, according to the tcp_may_update_window(): 1. the ACK acknowledged new data 2. the ACK has new data 3. the ACK expand the window and the seq of it is valid Now, we allow the ACK update the window if the window is 0, and the seq/ack of it is valid. This is for the case that the receiver replies an zero-window ACK when it is under memory stress and can't queue the new data. Signed-off-by:
-
