- 27 Jun, 2019 23 commits
-
-
Stanislav Fomichev authored
Add sockopt selftests: * require proper expected_attach_type * enforce context field read/write access * test bpf_sockopt_handled handler * test EPERM * test limiting optlen from getsockopt * test out-of-bounds access v9: * add tests for setsockopt argument mangling v7: * remove return 2; test retval=0 and optlen=-1 v3: * use DW for optval{,_end} loads v2: * use return code 2 for kernel bypass Cc: Andrii Nakryiko <andriin@fb.com> Cc: Martin Lau <kafai@fb.com> Signed-off-by:Stanislav Fomichev <sdf@google.com> Signed-off-by:
Alexei Starovoitov <ast@kernel.org>
-
Stanislav Fomichev authored
Add tests that make sure libbpf section detection works. Cc: Andrii Nakryiko <andriin@fb.com> Cc: Martin Lau <kafai@fb.com> Signed-off-by:
-
Stanislav Fomichev authored
Make libbpf aware of new sockopt hooks so it can derive prog type and hook point from the section names. Cc: Andrii Nakryiko <andriin@fb.com> Cc: Martin Lau <kafai@fb.com> Signed-off-by:
-
Stanislav Fomichev authored
Export new prog type and hook points to the libbpf. Cc: Andrii Nakryiko <andriin@fb.com> Cc: Martin Lau <kafai@fb.com> Signed-off-by:
-
Stanislav Fomichev authored
Implement new BPF_PROG_TYPE_CGROUP_SOCKOPT program type and BPF_CGROUP_{G,S}ETSOCKOPT cgroup hooks. BPF_CGROUP_SETSOCKOPT can modify user setsockopt arguments before passing them down to the kernel or bypass kernel completely. BPF_CGROUP_GETSOCKOPT can can inspect/modify getsockopt arguments that kernel returns. Both hooks reuse existing PTR_TO_PACKET{,_END} infrastructure. The buffer memory is pre-allocated (because I don't think there is a precedent for working with __user memory from bpf). This might be slow to do for each {s,g}etsockopt call, that's why I've added __cgroup_bpf_prog_array_is_empty that exits early if there is nothing attached to a cgroup. Note, however, that there is a race between __cgroup_bpf_prog_array_is_empty and BPF_PROG_RUN_ARRAY where cgroup program layout might have changed; this should not be a problem because in general there is a race between multiple calls to {s,g}etsocktop and user adding/removing bpf progs from a cgroup. The return code of the BPF program is handled as follows: * 0: EPERM * 1: success, continue with next BPF program in the cgroup chain v9: * allow overwriting setsockopt arguments (Alexei Starovoitov): * use set_fs (same as kernel_setsockopt) * buffer is always kzalloc'd (no small on-stack buffer) v8: * use s32 for optlen (Andrii Nakryiko) v7: * return only 0 or 1 (Alexei Starovoitov) * always run all progs (Alexei Starovoitov) * use optval=0 as kernel bypass in setsockopt (Alexei Starovoitov) (decided to use optval=-1 instead, optval=0 might be a valid input) * call getsockopt hook after kernel handlers (Alexei Starovoitov) v6: * rework cgroup chaining; stop as soon as bpf program returns 0 or 2; see patch with the documentation for the details * drop Andrii's and Martin's Acked-by (not sure they are comfortable with the new state of things) v5: * skip copy_to_user() and put_user() when ret == 0 (Martin Lau) v4: * don't export bpf_sk_fullsock helper (Martin Lau) * size != sizeof(__u64) for uapi pointers (Martin Lau) * offsetof instead of bpf_ctx_range when checking ctx access (Martin Lau) v3: * typos in BPF_PROG_CGROUP_SOCKOPT_RUN_ARRAY comments (Andrii Nakryiko) * reverse christmas tree in BPF_PROG_CGROUP_SOCKOPT_RUN_ARRAY (Andrii Nakryiko) * use __bpf_md_ptr instead of __u32 for optval{,_end} (Martin Lau) * use BPF_FIELD_SIZEOF() for consistency (Martin Lau) * new CG_SOCKOPT_ACCESS macro to wrap repeated parts v2: * moved bpf_sockopt_kern fields around to remove a hole (Martin Lau) * aligned bpf_sockopt_kern->buf to 8 bytes (Martin Lau) * bpf_prog_array_is_empty instead of bpf_prog_array_length (Martin Lau) * added [0,2] return code check to verifier (Martin Lau) * dropped unused buf[64] from the stack (Martin Lau) * use PTR_TO_SOCKET for bpf_sockopt->sk (Martin Lau) * dropped bpf_target_off from ctx rewrites (Martin Lau) * use return code for kernel bypass (Martin Lau & Andrii Nakryiko) Cc: Andrii Nakryiko <andriin@fb.com> Cc: Martin Lau <kafai@fb.com> Signed-off-by: -
Daniel Borkmann authored
Tariq Toukan says: ==================== This series contains improvements to the AF_XDP kernel infrastructure and AF_XDP support in mlx5e. The infrastructure improvements are required for mlx5e, but also some of them benefit to all drivers, and some can be useful for other drivers that want to implement AF_XDP. The performance testing was performed on a machine with the following configuration: - 24 cores of Intel Xeon E5-2620 v3 @ 2.40 GHz - Mellanox ConnectX-5 Ex with 100 Gbit/s link The results with retpoline disabled, single stream: txonly: 33.3 Mpps (21.5 Mpps with queue and app pinned to the same CPU) rxdrop: 12.2 Mpps l2fwd: 9.4 Mpps The results with retpoline enabled, single stream: txonly: 21.3 Mpps (14.1 Mpps with queue and app pinned to the same CPU) rxdrop: 9.9 Mpps l2fwd: 6.8 Mpps v2 changes: Added patches for mlx5e and addressed the comments for v1. Rebased for bpf-next. v3 changes: Rebased for the newer bpf-next, resolved conflicts in libbpf. Addressed Björn's comments for coding style. Fixed a bug in error handling flow in mlx5e_open_xsk. v4 changes: UAPI is not changed, XSK RX queues are exposed to the kernel. The lower half of the available amount of RX queues are regular queues, and the upper half are XSK RX queues. The patch "xsk: Extend channels to support combined XSK/non-XSK traffic" was dropped. The final patch was reworked accordingly. Added "net/mlx5e: Attach/detach XDP program safely", as the changes introduced in the XSK patch base on the stuff from this one. Added "libbpf: Support drivers with non-combined channels", which aligns the condition in libbpf with the condition in the kernel. Rebased over the newer bpf-next. v5 changes: In v4, ethtool reports the number of channels as 'combined' and the number of XSK RX queues as 'rx' for mlx5e. It was changed, so that 'rx' is 0, and 'combined' reports the double amount of channels if there is an active UMEM - to make libbpf happy. The patch for libbpf was dropped. Although it's still useful and fixes things, it raises some disagreement, so I'm dropping it - it's no longer useful for mlx5e anymore after the change above. v6 changes: As Maxim is out of office, I rebased the series on behalf of him, solved some conflicts, and re-spinned. ==================== Acked-by:
Björn Töpel <bjorn.topel@intel.com> Tested-by:
Jonathan Lemon <jonathan.lemon@gmail.com> Signed-off-by:
Daniel Borkmann <daniel@iogearbox.net>
-
Maxim Mikityanskiy authored
This commit adds support for AF_XDP zero-copy RX and TX. We create a dedicated XSK RQ inside the channel, it means that two RQs are running simultaneously: one for non-XSK traffic and the other for XSK traffic. The regular and XSK RQs use a single ID namespace split into two halves: the lower half is regular RQs, and the upper half is XSK RQs. When any zero-copy AF_XDP socket is active, changing the number of channels is not allowed, because it would break to mapping between XSK RQ IDs and channels. XSK requires different page allocation and release routines. Such functions as mlx5e_{alloc,free}_rx_mpwqe and mlx5e_{get,put}_rx_frag are generic enough to be used for both regular and XSK RQs, and they use the mlx5e_page_{alloc,release} wrappers around the real allocation functions. Function pointers are not used to avoid losing the performance with retpolines. Wherever it's certain that the regular (non-XSK) page release function should be used, it's called directly. Only the stats that could be meaningful for XSK are exposed to the userspace. Those that don't take part in the XSK flow are not considered. Note that we don't wait for WQEs on the XSK RQ (unlike the regular RQ), because the newer xdpsock sample doesn't provide any Fill Ring entries at the setup stage. We create a dedicated XSK SQ in the channel. This separation has its advantages: 1. When the UMEM is closed, the XSK SQ can also be closed and stop receiving completions. If an existing SQ was used for XSK, it would continue receiving completions for the packets of the closed socket. If a new UMEM was opened at that point, it would start getting completions that don't belong to it. 2. Calculating statistics separately. When the userspace kicks the TX, the driver triggers a hardware interrupt by posting a NOP to a dedicated XSK ICO (internal control operations) SQ, in order to trigger NAPI on the right CPU core. This XSK ICO SQ is protected by a spinlock, as the userspace application may kick the TX from any core. Store the pointers to the UMEMs in the net device private context, independently from the kernel. This way the driver can distinguish between the zero-copy and non-zero-copy UMEMs. The kernel function xdp_get_umem_from_qid does not care about this difference, but the driver is only interested in zero-copy UMEMs, particularly, on the cleanup it determines whether to close the XSK RQ and SQ or not by looking at the presence of the UMEM. Use state_lock to protect the access to this area of UMEM pointers. LRO isn't compatible with XDP, but there may be active UMEMs while XDP is off. If this is the case, don't allow LRO to ensure XDP can be reenabled at any time. The validation of XSK parameters typically happens when XSK queues open. However, when the interface is down or the XDP program isn't set, it's still possible to have active AF_XDP sockets and even to open new, but the XSK queues will be closed. To cover these cases, perform the validation also in these flows: 1. A new UMEM is registered, but the XSK queues aren't going to be created due to missing XDP program or interface being down. 2. MTU changes while there are UMEMs registered. Having this early check prevents mlx5e_open_channels from failing at a later stage, where recovery is impossible and the application has no chance to handle the error, because it got the successful return value for an MTU change or XSK open operation. The performance testing was performed on a machine with the following configuration: - 24 cores of Intel Xeon E5-2620 v3 @ 2.40 GHz - Mellanox ConnectX-5 Ex with 100 Gbit/s link The results with retpoline disabled, single stream: txonly: 33.3 Mpps (21.5 Mpps with queue and app pinned to the same CPU) rxdrop: 12.2 Mpps l2fwd: 9.4 Mpps The results with retpoline enabled, single stream: txonly: 21.3 Mpps (14.1 Mpps with queue and app pinned to the same CPU) rxdrop: 9.9 Mpps l2fwd: 6.8 Mpps Signed-off-by:Maxim Mikityanskiy <maximmi@mellanox.com> Signed-off-by:
Tariq Toukan <tariqt@mellanox.com> Acked-by:
Saeed Mahameed <saeedm@mellanox.com> Signed-off-by:
-
Maxim Mikityanskiy authored
structs mlx5e_{rq,sq,cq,channel}_param are going to be used in the upcoming XSK RX and TX patches. Move them to a header file to make them accessible from other C files. Signed-off-by: -
Maxim Mikityanskiy authored
Create new functions mlx5e_{open,close}_queues to encapsulate opening and closing RQs and SQs, and call the new functions from mlx5e_{open,close}_channel. It simplifies the existing functions a bit and prepares them for the upcoming AF_XDP changes. Signed-off-by: -
Maxim Mikityanskiy authored
Use the existing mlx5e_get_linear_rq_headroom function to calculate the headroom for mlx5e_xdp_max_mtu. This function takes the XSK headroom into consideration, which will be used in the following patches. Signed-off-by:
-
Maxim Mikityanskiy authored
When an XDP program returns XDP_TX, and the RQ is XSK-enabled, it requires careful handling, because convert_to_xdp_frame creates a new page and copies the data there, while our driver expects the xdp_frame to point to the same memory as the xdp_buff. Handle this case separately: map the page, and in the end unmap it and call xdp_return_frame. Signed-off-by:
-
Maxim Mikityanskiy authored
Put the XDP SQ that is used for XDP_TX into the channel. It used to be a part of the RQ, but with introduction of AF_XDP there will be one more RQ that could share the same XDP SQ. This patch is a preparation for that change. Separate XDP_TX statistics per RQ were implemented in one of the previous patches. Signed-off-by:
-
Maxim Mikityanskiy authored
Currently, struct mlx5e_xdp_info has some issues that have to be cleaned up before the upcoming AF_XDP support makes things too complicated and messy. This structure is used both when sending the packet and on completion. Moreover, the cleanup procedure on completion depends on the origin of the packet (XDP_REDIRECT, XDP_TX). Adding AF_XDP support will add new flows that use this structure even differently. To avoid overcomplicating the code, this commit refactors the usage of this structure in the following ways: 1. struct mlx5e_xdp_info is split into two different structures. One is struct mlx5e_xdp_xmit_data, a transient structure that doesn't need to be stored and is only used while sending the packet. The other is still struct mlx5e_xdp_info that is stored in a FIFO and contains the fields needed on completion. 2. The fields of struct mlx5e_xdp_info that are used in different flows are put into a union. A special enum indicates the cleanup mode and helps choose the right union member. This approach is clear and explicit. Although it could be possible to "guess" the mode by looking at the values of the fields and at the XDP SQ type, it wouldn't be that clear and extendable and would require looking through the whole chain to understand what's going on. For the reference, there are the fields of struct mlx5e_xdp_info that are used in different flows (including AF_XDP ones): Packet origin | Fields used on completion | Cleanup steps -----------------------+---------------------------+------------------ XDP_REDIRECT, | xdpf, dma_addr | DMA unmap and XDP_TX from XSK RQ | | xdp_return_frame. -----------------------+---------------------------+------------------ XDP_TX from regular RQ | di | Recycle page. -----------------------+---------------------------+------------------ AF_XDP TX | (none) | Increment the | | producer index in | | Completion Ring. On send, the same set of mlx5e_xdp_xmit_data fields is used in all flows: DMA and virtual addresses and length. Signed-off-by: -
Maxim Mikityanskiy authored
Prepare to creation of the XSK RQ, which will require posting UMRs, too. The same ICO SQ will be used for both RQs and also to trigger interrupts by posting NOPs. UMR WQEs can't be reused any more. Optimization introduced in commit ab966d7e ("net/mlx5e: RX, Recycle buffer of UMR WQEs") is reverted. Signed-off-by:
-
Maxim Mikityanskiy authored
Additional conditions introduced: - XSK implies XDP. - Headroom includes the XSK headroom if it exists. - No space is reserved for struct shared_skb_info in XSK mode. - Fragment size smaller than the XSK chunk size is not allowed. A new auxiliary function mlx5e_get_linear_rq_headroom with the support for XSK is introduced. Use this function in the implementation of mlx5e_get_rq_headroom. Change headroom to u32 to match the headroom field in struct xdp_umem. Signed-off-by:
-
Maxim Mikityanskiy authored
The PCI API for DMA is deprecated, and PCI_DMA_TODEVICE is just defined to DMA_TO_DEVICE for backward compatibility. Just use DMA_TO_DEVICE. Signed-off-by:
-
Maxim Mikityanskiy authored
Some drivers want to access the data transmitted in order to implement acceleration features of the NICs. It is also useful in AF_XDP TX flow. Change the xsk_umem_consume_tx API to return the whole xdp_desc, that contains the data pointer, length and DMA address, instead of only the latter two. Adapt the implementation of i40e and ixgbe to this change. Signed-off-by:
-
Maxim Mikityanskiy authored
The typical XDP memory scheme is one packet per page. Change the AF_XDP frame size in libbpf to 4096, which is the page size on x86, to allow libbpf to be used with the drivers with the packet-per-page scheme. Add a command line option -f to xdpsock to allow to specify a custom frame size. Signed-off-by:
-
Maxim Mikityanskiy authored
Query XDP_OPTIONS in libbpf to determine if the zero-copy mode is active or not. Signed-off-by:
-
Maxim Mikityanskiy authored
Make it possible for the application to determine whether the AF_XDP socket is running in zero-copy mode. To achieve this, add a new getsockopt option XDP_OPTIONS that returns flags. The only flag supported for now is the zero-copy mode indicator. Signed-off-by:
-
Maxim Mikityanskiy authored
Add a function that checks whether the Fill Ring has the specified amount of descriptors available. It will be useful for mlx5e that wants to check in advance, whether it can allocate a bulk of RX descriptors, to get the best performance. Signed-off-by:
-
Maxim Mikityanskiy authored
When an XDP program is set, a full reopen of all channels happens in two cases: 1. When there was no program set, and a new one is being set. 2. When there was a program set, but it's being unset. The full reopen is necessary, because the channel parameters may change if XDP is enabled or disabled. However, it's performed in an unsafe way: if the new channels fail to open, the old ones are already closed, and the interface goes down. Use the safe way to switch channels instead. The same way is already used for other configuration changes. Signed-off-by:
-
Roman Gushchin authored
Since commit 4bfc0bb2 ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself"), cgroup_bpf release occurs asynchronously (from a worker context), and before the release of the cgroup itself. This introduced a previously non-existing race between the release and update paths. E.g. if a leaf's cgroup_bpf is released and a new bpf program is attached to the one of ancestor cgroups at the same time. The race may result in double-free and other memory corruptions. To fix the problem, let's protect the body of cgroup_bpf_release() with cgroup_mutex, as it was effectively previously, when all this code was called from the cgroup release path with cgroup mutex held. Also let's skip cgroups, which have no chances to invoke a bpf program, on the update path. If the cgroup bpf refcnt reached 0, it means that the cgroup is offline (no attached processes), and there are no associated sockets left. It means there is no point in updating effective progs array! And it can lead to a leak, if it happens after the release. So, let's skip such cgroups. Big thanks for Tejun Heo for discovering and debugging of this problem! Fixes: 4bfc0bb2 ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") Reported-by:
Tejun Heo <tj@kernel.org> Signed-off-by:
Roman Gushchin <guro@fb.com> Signed-off-by:
-
- 26 Jun, 2019 5 commits
-
-
YueHaibing authored
Fix sparse warning: net/core/xdp.c:88:6: warning: symbol '__mem_id_disconnect' was not declared. Should it be static? Reported-by:
Hulk Robot <hulkci@huawei.com> Signed-off-by:
YueHaibing <yuehaibing@huawei.com> Acked-by:
Jesper Dangaard Brouer <brouer@redhat.com> Acked-by:
Song Liu <songliubraving@fb.com> Signed-off-by:
-
Daniel T. Lee authored
Currently, each xdp samples are inconsistent in the use. Most of the samples fetch the interface with it's name. (ex. xdp1, xdp2skb, xdp_redirect_cpu, xdp_sample_pkts, etc.) But some of the xdp samples are fetching the interface with ifindex by command argument. This commit enables xdp samples to fetch interface with it's name without changing the original index interface fetching. (<ifname|ifindex> fetching in the same way as xdp_sample_pkts_user.c does.) Signed-off-by:
Daniel T. Lee <danieltimlee@gmail.com> Acked-by:
Toke Høiland-Jørgensen <toke@redhat.com> Acked-by:
-
Yonghong Song authored
With CONFIG_MODULES=n, the following compiler warning occurs: /data/users/yhs/work/net-next/kernel/trace/bpf_trace.c:605:13: warning: ‘do_bpf_send_signal’ defined but not used [-Wunused-function] static void do_bpf_send_signal(struct irq_work *entry) The __init function send_signal_irq_work_init(), which calls do_bpf_send_signal(), is defined under CONFIG_MODULES. Hence, when CONFIG_MODULES=n, nobody calls static function do_bpf_send_signal(), hence the warning. The init function send_signal_irq_work_init() should work without CONFIG_MODULES. Moving it out of CONFIG_MODULES code section fixed the compiler warning, and also make bpf_send_signal() helper work without CONFIG_MODULES. Fixes: 8b401f9e ("bpf: implement bpf_send_signal() helper") Reported-By:Arnd Bergmann <arnd@arndb.de> Signed-off-by:
Yonghong Song <yhs@fb.com> Acked-by:
-
Andrii Nakryiko authored
Non-BPF (user land) part of selftests is built without debug info making occasional debugging with gdb terrible. Build with debug info always. Signed-off-by:
Andrii Nakryiko <andriin@fb.com> Signed-off-by:
-
Ivan Khoronzhuk authored
It fixes build error for 32bit caused by type mismatch size_t/unsigned long. Fixes: bf829271 ("libbpf: refactor map initialization") Acked-by:
Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org> Signed-off-by:
-
- 25 Jun, 2019 6 commits
-
-
Toshiaki Makita authored
XDP_TX is similar to XDP_REDIRECT as it essentially redirects packets to the device itself. XDP_REDIRECT has bulk transmit mechanism to avoid the heavy cost of indirect call but it also reduces lock acquisition on the destination device that needs locks like veth and tun. XDP_TX does not use indirect calls but drivers which require locks can benefit from the bulk transmit for XDP_TX as well. This patch introduces bulk transmit mechanism in veth using bulk queue on stack, and improves XDP_TX performance by about 9%. Here are single-core/single-flow XDP_TX test results. CPU consumptions are taken from "perf report --no-child". - Before: 7.26 Mpps _raw_spin_lock 7.83% veth_xdp_xmit 12.23% - After: 7.94 Mpps _raw_spin_lock 1.08% veth_xdp_xmit 6.10% v2: - Use stack for bulk queue instead of a global variable. Signed-off-by:
Toshiaki Makita <toshiaki.makita1@gmail.com> Acked-by:
-
Toshiaki Makita authored
This is introduced for admins to check what is happening on XDP_TX when bulk XDP_TX is in use, which will be first introduced in veth in next commit. v3: - Add act field to be in line with other XDP tracepoints. Signed-off-by:
-
Toshiaki Makita authored
Add a test case for veth native XDP. It checks if XDP_PASS, XDP_TX and XDP_REDIRECT work properly. $ cd tools/testing/selftests/bpf $ make \ TEST_CUSTOM_PROGS= \ TEST_GEN_PROGS= \ TEST_GEN_PROGS_EXTENDED= \ TEST_PROGS_EXTENDED= \ TEST_PROGS="test_xdp_veth.sh" \ run_tests TAP version 13 1..1 # selftests: bpf: test_xdp_veth.sh # PING 10.1.1.33 (10.1.1.33) 56(84) bytes of data. # 64 bytes from 10.1.1.33: icmp_seq=1 ttl=64 time=0.073 ms # # --- 10.1.1.33 ping statistics --- # 1 packets transmitted, 1 received, 0% packet loss, time 0ms # rtt min/avg/max/mdev = 0.073/0.073/0.073/0.000 ms # selftests: xdp_veth [PASS] ok 1 selftests: bpf: test_xdp_veth.sh Signed-off-by:
-
Eric Leblond authored
Fix documentation that mention xdpsock_kern.c which has been replaced by code embedded in libbpf. Signed-off-by:
Eric Leblond <eric@regit.org> Acked-by:
-
Colin Ian King authored
There are several spelling mistakes in pr_warning messages. Fix these. Signed-off-by:
Colin Ian King <colin.king@canonical.com> Acked-by:
-
Michal Rostecki authored
ibumad example was implementing the bpf_debug macro which is exactly the same as the bpf_printk macro available in bpf_helpers.h. This change makes use of bpf_printk instead of bpf_debug. Signed-off-by:
Michal Rostecki <mrostecki@opensuse.org> Acked-by:
-
- 23 Jun, 2019 6 commits
-
-
David S. Miller authored
Wei Wang says: ==================== ipv6: avoid taking refcnt on dst during route lookup Ipv6 route lookup code always grabs refcnt on the dst for the caller. But for certain cases, grabbing refcnt is not always necessary if the call path is rcu protected and the caller does not cache the dst. Another issue in the route lookup logic is: When there are multiple custom rules, we have to do the lookup into each table associated to each rule individually. And when we can't find the route in one table, we grab and release refcnt on net->ipv6.ip6_null_entry before going to the next table. This operation is completely redundant, and causes false issue because net->ipv6.ip6_null_entry is a shared object. This patch set introduces a new flag RT6_LOOKUP_F_DST_NOREF for route lookup callers to set, to avoid any manipulation on the dst refcnt. And it converts the major input and output path to use it. The performance gain is noticable. I ran synflood tests between 2 hosts under the same switch. Both hosts have 20G mlx NIC, and 8 tx/rx queues. Sender sends pure SYN flood with random src IPs and ports using trafgen. Receiver has a simple TCP listener on the target port. Both hosts have multiple custom rules: - For incoming packets, only local table is traversed. - For outgoing packets, 3 tables are traversed to find the route. The packet processing rate on the receiver is as follows: - Before the fix: 3.78Mpps - After the fix: 5.50Mpps v2->v3: - Handled fib6_rule_lookup() when CONFIG_IPV6_MULTIPLE_TABLES is not configured in patch 03 (suggested by David Ahern) - Removed the renaming of l3mdev_link_scope_lookup() in patch 05 (suggested by David Ahern) - Moved definition of ip6_route_output_flags() from an inline function in /net/ipv6/route.c to net/ipv6/route.c in order to address kbuild error in patch 05 v1->v2: - Added a helper ip6_rt_put_flags() in patch 3 suggested by David Miller ==================== Reviewed-by:
David Ahern <dsahern@gmail.com> Reviewed-by:
Eric Dumazet <edumazet@google.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Wei Wang authored
For tx path, in most cases, we still have to take refcnt on the dst cause the caller is caching the dst somewhere. But it still is beneficial to make use of RT6_LOOKUP_F_DST_NOREF flag while doing the route lookup. It is cause this flag prevents manipulating refcnt on net->ipv6.ip6_null_entry when doing fib6_rule_lookup() to traverse each routing table. The null_entry is a shared object and constant updates on it cause false sharing. We converted the current major lookup function ip6_route_output_flags() to make use of RT6_LOOKUP_F_DST_NOREF. Together with the change in the rx path, we see noticable performance boost: I ran synflood tests between 2 hosts under the same switch. Both hosts have 20G mlx NIC, and 8 tx/rx queues. Sender sends pure SYN flood with random src IPs and ports using trafgen. Receiver has a simple TCP listener on the target port. Both hosts have multiple custom rules: - For incoming packets, only local table is traversed. - For outgoing packets, 3 tables are traversed to find the route. The packet processing rate on the receiver is as follows: - Before the fix: 3.78Mpps - After the fix: 5.50Mpps Signed-off-by:
Wei Wang <weiwan@google.com> Signed-off-by:
-
Wei Wang authored
ip6_route_input() is the key function to do the route lookup in the rx data path. All the callers to this function are already holding rcu lock. So it is fairly easy to convert it to not take refcnt on the dst: We pass in flag RT6_LOOKUP_F_DST_NOREF and do skb_dst_set_noref(). This saves a few atomic inc or dec operations and should boost performance overall. This also makes the logic more aligned with v4. Signed-off-by:
Mahesh Bandewar <maheshb@google.com> Signed-off-by:
-
Wei Wang authored
This patch specifically converts the rule lookup logic to honor this flag and not release refcnt when traversing each rule and calling lookup() on each routing table. Similar to previous patch, we also need some special handling of dst entries in uncached list because there is always 1 refcnt taken for them even if RT6_LOOKUP_F_DST_NOREF flag is set. Signed-off-by:
-
Wei Wang authored
Initialize rt6->rt6i_uncached on the following pre-allocated dsts: net->ipv6.ip6_null_entry net->ipv6.ip6_prohibit_entry net->ipv6.ip6_blk_hole_entry This is a preparation patch for later commits to be able to distinguish dst entries in uncached list by doing: !list_empty(rt6->rt6i_uncached) Signed-off-by:
-
Wei Wang authored
This new flag is to instruct the route lookup function to not take refcnt on the dst entry. The user which does route lookup with this flag must properly use rcu protection. ip6_pol_route() is the major route lookup function for both tx and rx path. In this function: Do not take refcnt on dst if RT6_LOOKUP_F_DST_NOREF flag is set, and directly return the route entry. The caller should be holding rcu lock when using this flag, and decide whether to take refcnt or not. One note on the dst cache in the uncached_list: As uncached_list does not consume refcnt, one refcnt is always returned back to the caller even if RT6_LOOKUP_F_DST_NOREF flag is set. Uncached dst is only possible in the output path. So in such call path, caller MUST check if the dst is in the uncached_list before assuming that there is no refcnt taken on the returned dst. Signed-off-by:
-
