- 16 Nov, 2019 17 commits
-
-
Matteo Croce authored
A bonding with layer2+3 or layer3+4 hashing uses the IP addresses and the ports to balance packets between slaves. With some network errors, we receive an ICMP error packet by the remote host or a router. If sent by a router, the source IP can differ from the remote host one. Additionally the ICMP protocol has no port numbers, so a layer3+4 bonding will get a different hash than the previous one. These two conditions could let the packet go through a different interface than the other packets of the same flow: # tcpdump -qltnni veth0 |sed 's/^/0: /' & # tcpdump -qltnni veth1 |sed 's/^/1: /' & # hping3 -2 192.168.0.2 -p 9 0: IP 192.168.0.1.2251 > 192.168.0.2.9: UDP, length 0 1: IP 192.168.0.2 > 192.168.0.1: ICMP 192.168.0.2 udp port 9 unreachable, length 36 1: IP 192.168.0.1.2252 > 192.168.0.2.9: UDP, length 0 1: IP 192.168.0.2 > 192.168.0.1: ICMP 192.168.0.2 udp port 9 unreachable, length 36 1: IP 192.168.0.1.2253 > 192.168.0.2.9: UDP, length 0 1: IP 192.168.0.2 > 192.168.0.1: ICMP 192.168.0.2 udp port 9 unreachable, length 36 0: IP 192.168.0.1.2254 > 192.168.0.2.9: UDP, length 0 1: IP 192.168.0.2 > 192.168.0.1: ICMP 192.168.0.2 udp port 9 unreachable, length 36 An ICMP error packet contains the header of the packet which caused the network error, so inspect it and match the flow against it, so we can send the ICMP via the same interface of the previous packet in the flow. Move the IP and port dissect code into a generic function bond_flow_ip() and if we are dissecting an ICMP error packet, call it again with the adjusted offset. # hping3 -2 192.168.0.2 -p 9 1: IP 192.168.0.1.1224 > 192.168.0.2.9: UDP, length 0 1: IP 192.168.0.2 > 192.168.0.1: ICMP 192.168.0.2 udp port 9 unreachable, length 36 1: IP 192.168.0.1.1225 > 192.168.0.2.9: UDP, length 0 1: IP 192.168.0.2 > 192.168.0.1: ICMP 192.168.0.2 udp port 9 unreachable, length 36 0: IP 192.168.0.1.1226 > 192.168.0.2.9: UDP, length 0 0: IP 192.168.0.2 > 192.168.0.1: ICMP 192.168.0.2 udp port 9 unreachable, length 36 0: IP 192.168.0.1.1227 > 192.168.0.2.9: UDP, length 0 0: IP 192.168.0.2 > 192.168.0.1: ICMP 192.168.0.2 udp port 9 unreachable, length 36 Signed-off-by:Matteo Croce <mcroce@redhat.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Horatiu Vultur authored
The commit 0c65b2b9 ("net: of_get_phy_mode: Change API to solve int/unit warnings") updated the function of_get_phy_mode declaration. Now it returns an error code and in case the node doesn't contain the property 'phy-mode' or 'phy-connection-type' it returns -EINVAL and would set the phy_interface_t to PHY_INTERFACE_MODE_NA. Ocelot VSC7514 has 4 internal phys which have the phy interface PHY_INTERFACE_MODE_NA. So because of_get_phy_mode would assign PHY_INTERFACE_MODE_NA to phy_mode when there is an error, there is no need to add the error check. Updates for v2: - drop error check because of_get_phy_mode already assigns phy_interface to PHY_INTERFACE_MODE in case of error. Signed-off-by:
Horatiu Vultur <horatiu.vultur@microchip.com> Signed-off-by:
-
Alexander Lobakin authored
Commit 78d3fd0b ("gro: Only use skb_gro_header for completely non-linear packets") back in May'09 (v2.6.31-rc1) has changed the original condition '!skb_headlen(skb)' to 'skb->mac_header == skb->tail' in gro_reset_offset() saying: "Since the drivers that need this optimisation all provide completely non-linear packets" (note that this condition has become the current 'skb_mac_header(skb) == skb_tail_pointer(skb)' later with commmit ced14f68 ("net: Correct comparisons and calculations using skb->tail and skb-transport_header") without any functional changes). For now, we have the following rough statistics for v5.4-rc7: 1) napi_gro_frags: 14 2) napi_gro_receive with skb->head containing (most of) payload: 83 3) napi_gro_receive with skb->head containing all the headers: 20 4) napi_gro_receive with skb->head containing only Ethernet header: 2 With the current condition, fast GRO with the usage of NAPI_GRO_CB(skb)->frag0 is available only in the [1] case. Packets pushed by [2] and [3] go through the 'slow' path, but it's not a problem for them as they already contain all the needed headers in skb->head, so pskb_may_pull() only moves skb->data. The layout of skbs in the fourth [4] case at the moment of dev_gro_receive() is identical to skbs that have come through [1], as napi_frags_skb() pulls Ethernet header to skb->head. The only difference is that the mentioned condition is always false for them, because skb_put() and friends irreversibly alter the tail pointer. They also go through the 'slow' path, but now every single pskb_may_pull() in every single .gro_receive() will call the *really* slow __pskb_pull_tail() to pull headers to head. This significantly decreases the overall performance for no visible reasons. The only two users of method [4] is: * drivers/staging/qlge * drivers/net/wireless/iwlwifi (all three variants: dvm, mvm, mvm-mq) Note that in case with wireless drivers we can't use [1] (napi_gro_frags()) at least for now and mac80211 stack always performs pushes and pulls anyways, so performance hit is inavoidable. At the moment of v2.6.31 the mentioned change was necessary (that's why I don't add the "Fixes:" tag), but it became obsolete since skb_gro_mac_header() has gone in commit a50e233c ("net-gro: restore frag0 optimization"), so we can simply revert the condition in gro_reset_offset() to allow skbs from [4] go through the 'fast' path just like in case [1]. This was tested on a 600 MHz MIPS CPU and a custom driver and this patch gave boosts up to 40 Mbps to method [4] in both directions comparing to net-next, which made overall performance relatively close to [1] (without it, [4] is the slowest). v2: - Add more references and explanations to commit message - Fix some typos ibid - No functional changes Signed-off-by:
Alexander Lobakin <alobakin@dlink.ru> Signed-off-by:
-
David S. Miller authored
Kees Cook says: ==================== bnx2x: Remove function casts In order to make the entire kernel usable under Clang's Control Flow Integrity protections, function prototype casts need to be avoided because this will trip CFI checks at runtime (i.e. a mismatch between the caller's expected function prototype and the destination function's prototype). Many of these cases can be found with -Wcast-function-type, which found that bnx2x had a bunch of needless (or at least confusing) function casts. This series removes them all. ==================== Signed-off-by: -
Kees Cook authored
All .rw_reset callbacks except bnx2x_84833_hw_reset_phy() use a void return type. No callers of .hw_reset check a return value and bnx2x_84833_hw_reset_phy() unconditionally returns 0. Remove all hw_reset_t casts and fix the return type to void. Signed-off-by:
Kees Cook <keescook@chromium.org> Signed-off-by:
-
Kees Cook authored
The return values for format_fw_ver_t callbacks are supposed to be "int", not "u8". Ultimately, the top-level caller doesn't actually check the return value at all, but just clean this all up anyway and fix the prototypes so that casts are no longer needed. Signed-off-by:
-
Kees Cook authored
No callers of .config_init check return values. Remove the casting and change all callbacks to have the correct function prototype. Signed-off-by:
-
Kees Cook authored
The function casts for .read_status callbacks end up casting some int return values to u8. This seems to be bug-prone (-EINVAL being returned into something that appears to be true/false), but fixing the function prototypes doesn't change the existing behavior. Fix the return values to remove the casts. Signed-off-by:
-
Kees Cook authored
NULL is already "void *" so it will auto-cast in assignments and initializers. Additionally, all the callbacks for .link_reset, .config_loopback, .set_link_led, and .phy_specific_func are already correct. No casting is needed for these, so remove them. Signed-off-by:
-
Po Liu authored
ENETC has a register PSPEED to indicate the link speed of hardware. It is need to update accordingly. PSPEED field needs to be updated with the port speed for QBV scheduling purposes. Or else there is chance for gate slot not free by frame taking the MAC if PSPEED and phy speed not match. So update PSPEED when link adjust. This is implement by the adjust_link. Signed-off-by:
Po Liu <Po.Liu@nxp.com> Signed-off-by:
Claudiu Manoil <claudiu.manoil@nxp.com> Signed-off-by:
Vladimir Oltean <vladimir.oltean@nxp.com> Signed-off-by:
-
Po Liu authored
ENETC supports in hardware for time-based egress shaping according to IEEE 802.1Qbv. This patch implement the Qbv enablement by the hardware offload method qdisc tc-taprio method. Also update cbdr writeback to up level since control bd ring may writeback data to control bd ring. Signed-off-by:
-
Jonathan Lemon authored
The page pool keeps track of the number of pages in flight, and it isn't safe to remove the pool until all pages are returned. Disallow removing the pool until all pages are back, so the pool is always available for page producers. Make the page pool responsible for its own delayed destruction instead of relying on XDP, so the page pool can be used without the xdp memory model. When all pages are returned, free the pool and notify xdp if the pool is registered with the xdp memory system. Have the callback perform a table walk since some drivers (cpsw) may share the pool among multiple xdp_rxq_info. Note that the increment of pages_state_release_cnt may result in inflight == 0, resulting in the pool being released. Fixes: d956a048 ("xdp: force mem allocator removal and periodic warning") Signed-off-by:
Jonathan Lemon <jonathan.lemon@gmail.com> Acked-by:
Jesper Dangaard Brouer <brouer@redhat.com> Acked-by:
Ilias Apalodimas <ilias.apalodimas@linaro.org> Signed-off-by:
-
David S. Miller authored
Karsten Graul says: ==================== last part of termination improvements Patches 1 and 2 finish the set of termination patches, introducing a reboot handler that terminates all link groups. Patch 3 adds an rcu_barrier before the module is unloaded, and patch 4 is cleanup. ==================== Signed-off-by: -
Ursula Braun authored
Constant SMC_CLOSE_WAIT_LISTEN_CLCSOCK_TIME is defined, but since commit 3d502067 ("net/smc: simplify wait when closing listen socket") no longer used. Remove it. Signed-off-by:
Ursula Braun <ubraun@linux.ibm.com> Signed-off-by:
Karsten Graul <kgraul@linux.ibm.com> Signed-off-by:
-
Ursula Braun authored
Add rcu_barrier() to make sure no RCU readers or callbacks are pending when the module is unloaded. Signed-off-by:
-
Ursula Braun authored
When rebooting it should be guaranteed all link groups are cleaned up and freed. Signed-off-by:
-
Ursula Braun authored
If the smc module is unloaded return control from exit routine only, if all link groups are freed. If an IB device is thrown away return control from device removal only, if all link groups belonging to this device are freed. Counters for the total number of SMCR link groups and for the total number of SMCR links per IB device are introduced. smc module unloading continues only if the total number of SMCR link groups is zero. IB device removal continues only it the total number of SMCR links per IB device has decreased to zero. Signed-off-by:
-
- 15 Nov, 2019 23 commits
-
-
Petar Penkov authored
There is a race in the TUN driver between napi_busy_loop and napi_gro_frags. This commit resolves the race by adding the NAPI struct via netif_tx_napi_add, instead of netif_napi_add, which disables polling for the NAPI struct. KCSAN reported: BUG: KCSAN: data-race in gro_normal_list.part.0 / napi_busy_loop write to 0xffff8880b5d474b0 of 4 bytes by task 11205 on cpu 0: gro_normal_list.part.0+0x77/0xb0 net/core/dev.c:5682 gro_normal_list net/core/dev.c:5678 [inline] gro_normal_one net/core/dev.c:5692 [inline] napi_frags_finish net/core/dev.c:5705 [inline] napi_gro_frags+0x625/0x770 net/core/dev.c:5778 tun_get_user+0x2150/0x26a0 drivers/net/tun.c:1976 tun_chr_write_iter+0x79/0xd0 drivers/net/tun.c:2022 call_write_iter include/linux/fs.h:1895 [inline] do_iter_readv_writev+0x487/0x5b0 fs/read_write.c:693 do_iter_write fs/read_write.c:970 [inline] do_iter_write+0x13b/0x3c0 fs/read_write.c:951 vfs_writev+0x118/0x1c0 fs/read_write.c:1015 do_writev+0xe3/0x250 fs/read_write.c:1058 __do_sys_writev fs/read_write.c:1131 [inline] __se_sys_writev fs/read_write.c:1128 [inline] __x64_sys_writev+0x4e/0x60 fs/read_write.c:1128 do_syscall_64+0xcc/0x370 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x44/0xa9 read to 0xffff8880b5d474b0 of 4 bytes by task 11168 on cpu 1: gro_normal_list net/core/dev.c:5678 [inline] napi_busy_loop+0xda/0x4f0 net/core/dev.c:6126 sk_busy_loop include/net/busy_poll.h:108 [inline] __skb_recv_udp+0x4ad/0x560 net/ipv4/udp.c:1689 udpv6_recvmsg+0x29e/0xe90 net/ipv6/udp.c:288 inet6_recvmsg+0xbb/0x240 net/ipv6/af_inet6.c:592 sock_recvmsg_nosec net/socket.c:871 [inline] sock_recvmsg net/socket.c:889 [inline] sock_recvmsg+0x92/0xb0 net/socket.c:885 sock_read_iter+0x15f/0x1e0 net/socket.c:967 call_read_iter include/linux/fs.h:1889 [inline] new_sync_read+0x389/0x4f0 fs/read_write.c:414 __vfs_read+0xb1/0xc0 fs/read_write.c:427 vfs_read fs/read_write.c:461 [inline] vfs_read+0x143/0x2c0 fs/read_write.c:446 ksys_read+0xd5/0x1b0 fs/read_write.c:587 __do_sys_read fs/read_write.c:597 [inline] __se_sys_read fs/read_write.c:595 [inline] __x64_sys_read+0x4c/0x60 fs/read_write.c:595 do_syscall_64+0xcc/0x370 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x44/0xa9 Reported by Kernel Concurrency Sanitizer on: CPU: 1 PID: 11168 Comm: syz-executor.0 Not tainted 5.4.0-rc6+ #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Fixes: 94317099 ("tun: enable NAPI for TUN/TAP driver") Signed-off-by:
Petar Penkov <ppenkov@google.com> Reported-by:
syzbot <syzkaller@googlegroups.com> Reviewed-by:
Eric Dumazet <edumazet@google.com> Signed-off-by:
-
Eric Dumazet authored
Since we do not plan using pthread_join() in the server do_accept() loop, we better create detached threads, or risk increasing memory footprint over time. Fixes: 192dc405 ("selftests: net: add tcp_mmap program") Signed-off-by:
-
Tonghao Zhang authored
The nla_put_u16/nla_put_u32 makes sure that *attrlen is align. The call tree is that: nla_put_u16/nla_put_u32 -> nla_put attrlen = sizeof(u16) or sizeof(u32) -> __nla_put attrlen -> __nla_reserve attrlen -> skb_put(skb, nla_total_size(attrlen)) nla_total_size returns the total length of attribute including padding. Cc: Joe Stringer <joe@ovn.org> Cc: William Tu <u9012063@gmail.com> Signed-off-by:
Tonghao Zhang <xiangxia.m.yue@gmail.com> Acked-by:
Pravin B Shelar <pshelar@ovn.org> Signed-off-by:
-
David S. Miller authored
Vladimir Oltean says: ==================== DSA driver for Vitesse Felix switch This series builds upon the previous "Accomodate DSA front-end into Ocelot" topic and does the following: - Reworks the Ocelot (VSC7514) driver to support one more switching core (VSC9959), used in NPI mode. Some code which was thought to be SoC-specific (ocelot_board.c) wasn't, and vice versa, so it is being accordingly moved. - Exports ocelot driver structures and functions to include/soc/mscc. - Adds a DSA ocelot front-end for VSC9959, which is a PCI device and uses the exported ocelot functionality for hardware configuration. - Adds a tagger driver for the Vitesse injection/extraction DSA headers. This is known to be compatible with at least Ocelot and Felix. ==================== Signed-off-by: -
Vladimir Oltean authored
This supports an Ethernet switching core from Vitesse / Microsemi / Microchip (VSC9959) which is part of the Ocelot family (a brand name), and whose code name is Felix. The switch can be (and is) integrated on different SoCs as a PCIe endpoint device. The functionality is provided by the core of the Ocelot switch driver (drivers/net/ethernet/mscc). In this regard, the current driver is an instance of Microsemi's Ocelot core driver, with a DSA front-end. It inherits its name from VSC9959's code name, to distinguish itself from the switchdev ocelot driver. The patch adds the logic for probing a PCI device and defines the register map for the VSC9959 switch core, since it has some differences in register addresses and bitfield mappings compared to the other Ocelot switches (VSC7511, VSC7512, VSC7513, VSC7514). The Felix driver declares the register map as part of the "instance table". Currently the VSC9959 inside NXP LS1028A is the only instance, but presumably it can support other switches in the Ocelot family, when used in DSA mode (Linux running on the external CPU, and not on the embedded MIPS). In a few cases, some h/w operations have to be done differently on VSC9959 due to missing bitfields. This is the case for the switch core reset and init. Because for this operation Ocelot uses some bits that are not present on Felix, the latter has to use a register from the global registers block (GCB) instead. Although it is a PCI driver, it relies on DT bindings for compatibility with DSA (CPU port link, PHY library). It does not have any custom device tree bindings, since we would like to minimize its dependency on device tree though. Signed-off-by:
-
Vladimir Oltean authored
While it is entirely possible that this tagger format is in fact more generic than just these 2 switch families, I don't have that knowledge. The Seville switch in NXP T1040 has a similar frame format, but there are enough differences (e.g. DEST field starts at bit 57 instead of 56) that calling this file tag_vitesse.c is a bit of a stretch at the moment. The frame format has been listed in a comment so that people who add support for further Vitesse switches can rework this tagger while keeping compatibility with Felix. The "ocelot" name was chosen instead of "felix" because even the Ocelot switch can act as a DSA device when it is used in NPI mode, and the Felix tagger format is almost identical. Currently it is only used for the Felix switch embedded in the NXP LS1028A chip. The ABI for this tagger should be considered "not stable" at the moment. The DSA tag is always placed before the Ethernet header and therefore, we are using the long prefix for RX tags to avoid putting the DSA master port in promiscuous mode. Once there will be an API in DSA for drivers to request DSA masters to be in promiscuous mode unconditionally, we will switch to the "no prefix" extraction frame header, which will save 16 padding bytes for each RX frame. Signed-off-by:
Florian Fainelli <f.fainelli@gmail.com> Signed-off-by:
-
Vladimir Oltean authored
The Felix DSA driver needs to write to SYS_RAM_INIT_RAM_INIT for its own chip initialization process. Also update the MAINTAINERS file such that the headers exported by the ocelot driver are under the same maintainers' umbrella as the driver itself. Signed-off-by:
-
Vladimir Oltean authored
We will be registering another switch driver based on ocelot, which lives under drivers/net/dsa. Make sure the Felix DSA front-end has the necessary abstractions to implement a new Ocelot driver instantiation. This includes the function prototypes for implementing DSA callbacks. Signed-off-by:
-
Vladimir Oltean authored
The Felix switch has a different reset procedure, so a function pointer needs to be created and added to the ocelot_ops structure. The reset procedure has been moved into ocelot_init. Signed-off-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
-
Vladimir Oltean authored
When using the NPI port, the DSA tag is passed through Ethernet, so the switch's MAC needs to accept it as it comes from the DSA master. Increase the MTU on the external CPU port to account for the length of the injection header. Without this patch, MTU-sized frames are dropped by the switch's CPU port on xmit, which is especially obvious in TCP sessions. Signed-off-by:
-
Vladimir Oltean authored
This constant will be used in a future patch to increase the MTU on NPI ports, and will also be used in the tagger driver for Felix. Signed-off-by:
-
Vladimir Oltean authored
Since in an NPI/DSA setup, not all ports will have the same MTU, we need to make sure the watermarks for pause frames and/or tail dropping logic that existed in the driver is still coherent for the new MTU values. We need to do this because the NPI (aka external CPU) port needs an increased MTU for the DSA tag. This will be done in a future patch. Signed-off-by:
-
Vladimir Oltean authored
It doesn't make sense to rewrite all these registers every time the PHY library notifies us about a link state change. In a future patch we will customize the MTU for the CPU port, and since the MTU was previously configured from adjust_link, if we don't make this change, its value would have got overridden. Signed-off-by:
-
Claudiu Manoil authored
The adjust_link routine should be generic enough to be (re)used by any SoC that integrates a switch core compatible with the Ocelot core switch driver. Currently all configurations are generic except for the PCS settings that are SoC specific. Move these out to the Ocelot SoC/board instance. Signed-off-by:
-
Claudiu Manoil authored
Let's make this ioremap and regmap init code common. It should not be platform dependent as it should be usable by PCI devices too. Use better names where necessary to avoid clashes. Signed-off-by:
-
David S. Miller authored
Karsten Graul says: ==================== net/smc: improve termination handling (part 3) Part 3 of the SMC termination patches improves the link group termination processing and introduces the ability to immediately terminate a link group. ==================== Signed-off-by: -
Ursula Braun authored
If the SMC module is unloaded or an IB device is thrown away, the immediate link group freeing introduced for SMCD is exploited for SMCR as well. That means SMCR-specifics are added to smc_conn_kill(). Signed-off-by:
-
Ursula Braun authored
Make sure all pending work requests are completed before freeing a link. Dismiss tx pending slots already when terminating a link group to exploit termination shortcut in tx completion queue handler. And kill the completion queue tasklets after destroy of the completion queues, otherwise there is a time window for another tasklet schedule of an already killed tasklet. Signed-off-by:
-
Ursula Braun authored
For abnormal termination issue an LLC DELETE_LINK without the orderly flag. Signed-off-by:
-
Ursula Braun authored
Avoid waiting for a free work request buffer, if the link group is already terminating. Signed-off-by:
-
Ursula Braun authored
If the ism module is unloaded return control from exit routine only, if all link groups are freed. If an IB device is thrown away return control from device removal only, if all link groups belonging to this device are freed. A counters for the total number of SMCD link groups per ISM device is introduced. ism module unloading continues only if the total number of SMCD link groups for all ISM devices is zero. ISM device removal continues only it the total number of SMCD link groups per ISM device has decreased to zero. Signed-off-by:
-
Ursula Braun authored
A final cleanup due to SMCD device removal means immediate freeing of all link groups belonging to this device in interrupt context. This patch introduces a separate SMCD link group termination routine, which terminates all link groups of an SMCD device. This new routine smcd_terminate_all ()is reused if the smc module is unloaded. Signed-off-by:
-
Ursula Braun authored
SMCD link group termination is called when peer signals its shutdown of its corresponding link group. For regular shutdowns no connections exist anymore. For abnormal shutdowns connections must be killed and their DMBs must be unregistered immediately. That means the SMCR method to delay the link group freeing several seconds does not fit. This patch adds immediate termination of a link group and its SMCD connections and makes sure all SMCD link group related cleanup steps are finished. Signed-off-by:
-
