Re: [PATCH net-next 00/18] mlx5 RoCE/RDMA packet sniffer

[David Miller]

From: Saeed Mahameed 
Date: Tue, 21 Jun 2016 16:10:26 +0300

> I would like to drop this series,

Don't worry, I dropped it several days before this.

Re: [PATCH net-next 00/18] mlx5 RoCE/RDMA packet sniffer

[Saeed Mahameed]

On Fri, Jun 17, 2016 at 5:43 PM, Saeed Mahameed  wrote:
> Hi Dave,
>
> This patch set introduces mlx5 RoCE/RDMA packet sniffer, it allows
> mlx5e netdevice to receive RoCE/RDMA or RAW ETH traffic which isn't
> supposed to be passed to the kernel stack, for sniffing and diagnostics
> purposes.  This traffic is still not supposed to go through the whole
> network stack processing and should only go to the non-protocol specific
> handlers (ptype_all). e.g: tcpdump, etc ..

Dave,

I would like to drop this series, we did some evaluation and there is
no straightforward way to tell the kernel to not process traffic that
should only go to tcpdump or (ptype_all).

We are evaluating a semi-pure userspace solution to sniff user space
traffic directly form userspace into libpcap. (verbs/rdma
plugin/extension for libpcap)

Sorry for any inconvenience,

Thanks,
Saeed.

[PATCH net-next 00/18] mlx5 RoCE/RDMA packet sniffer

[Saeed Mahameed]

Hi Dave,

This patch set introduces mlx5 RoCE/RDMA packet sniffer, it allows
mlx5e netdevice to receive RoCE/RDMA or RAW ETH traffic which isn't
supposed to be passed to the kernel stack, for sniffing and diagnostics
purposes.  This traffic is still not supposed to go through the whole 
network stack processing and should only go to the non-protocol specific 
handlers (ptype_all). e.g: tcpdump, etc ..

In order to achieve this, when RoCE/RDMA sniffer is enabled, all RoCE/RDMA
steering rules which are forwarded to user space QPs will be duplicated
and marked with "OFFLOAD" flow tag, then forwarded to mlx5e netdevice receive 
path.
mlx5e receive path will detect sniffer packets by looking at the receive
completion flow tag, and in case it matches the "OFFLOAD" tag, skb->pkt_type is
set to (PACKET_OFFLOAD_KERNEL) so it will go to non-protocol specific handlers
(ptype_all) only.

To duplicate specific steering rules, a new notification mechanism is added.
It allows a consumer to request add/del rule notification on specific steering
namespaces.  Once a consumer registers, it will be notified on all existing 
rules.
Asynchronously, notifications on a dynamically added/deleted rules will be 
generated.

To achieve RoCE/RDMA sniffer, a new steering namespace is introduced 
(SNIFFER_NAMESPACE),
which will host all the duplicated steering rules to be forwarded to mlx5e 
netdevice
receive path.

RoCE sniffer module:
RoCE sniffer module will register to (RoCE/RDMA) user space traffic steering 
namespaces add/del
rules notification.
- On rule add it will generated an identical rule and inject it into the 
SNIFFER_NAMESPACE 
flow table with flow tag = "OFFLOAD" and destination = "mlx5e netdevice"
- On rule delete it will remove the duplicated corresponding sniffer rule.

Thanks,
Saeed.

Huy Nguyen (3):
  net/mlx5e: Set sniffer skbs packet type to offload kernel
  net/mlx5e: Sniffer support for kernel offload (RoCE) traffic
  net/mlx5e: Add netdev hw feature flag offload-sniffer

Maor Gottlieb (15):
  net/mlx5: Refactor mlx5_add_flow_rule
  net/mlx5: Introduce mlx5_flow_steering structure
  net/mlx5: Properly remove all steering objects
  net/mlx5: Add hold/put rules refcount API
  net/mlx5: Add support to add/del flow rule notifiers
  net/mlx5: Introduce table of function pointer steering commands
  net/mlx5: Introduce nop steering commands
  if_ether.h: Add RoCE Ethertype
  IB/mlx5: Create RoCE root namespace
  net/mlx5: Introduce get flow rule match API
  net/mlx5: Add sniffer namespaces
  IB/mlx5: Add kernel offload flow-tag
  net: Add offload kernel net stack packet type
  net/mlx5: Introduce sniffer steering hardware capabilities
  net/mlx5e: Lock device state in set features

 drivers/infiniband/hw/mlx4/qp.c|   6 +-
 drivers/infiniband/hw/mlx5/main.c  | 143 +++-
 drivers/infiniband/hw/mlx5/mlx5_ib.h   |  15 +-
 drivers/infiniband/hw/ocrdma/ocrdma_ah.c   |   4 +-
 drivers/infiniband/hw/ocrdma/ocrdma_hw.c   |   2 +-
 drivers/infiniband/hw/ocrdma/ocrdma_sli.h  |   4 -
 drivers/infiniband/hw/usnic/usnic_common_pkt_hdr.h |   1 -
 drivers/infiniband/hw/usnic/usnic_fwd.h|   2 +-
 drivers/net/ethernet/mellanox/mlx5/core/Makefile   |   3 +-
 drivers/net/ethernet/mellanox/mlx5/core/en.h   |  10 +
 drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c  |  26 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_fs.c|  28 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c  |  44 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_rx.c|   4 +
 .../net/ethernet/mellanox/mlx5/core/en_sniffer.c   | 574 
 drivers/net/ethernet/mellanox/mlx5/core/en_tc.c|   8 +-
 drivers/net/ethernet/mellanox/mlx5/core/eswitch.c  |  51 +-
 drivers/net/ethernet/mellanox/mlx5/core/fs_cmd.c   | 161 -
 drivers/net/ethernet/mellanox/mlx5/core/fs_cmd.h   |  71 +-
 drivers/net/ethernet/mellanox/mlx5/core/fs_core.c  | 746 ++---
 drivers/net/ethernet/mellanox/mlx5/core/fs_core.h  |  31 +
 include/linux/mlx5/device.h|  17 +
 include/linux/mlx5/driver.h|   6 +-
 include/linux/mlx5/fs.h|  62 +-
 include/linux/netdev_features.h|   2 +
 include/linux/skbuff.h |   6 +-
 include/uapi/linux/if_ether.h  |   1 +
 include/uapi/linux/if_packet.h |   1 +
 net/core/dev.c |   4 +
 net/core/ethtool.c |   1 +
 30 files changed, 1626 insertions(+), 408 deletions(-)
 create mode 100644 drivers/net/ethernet/mellanox/mlx5/core/en_sniffer.c

-- 
2.8.0