linux_dsm_epyc7002/Documentation/sysctl/net.txt
Wang YanQing 03f5781be2 bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.

Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW

It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.

IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.

The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
  for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
  on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
  will degrade jit performance heavily.

Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.

Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2

Below are some numbers for this jit implementation:
Note:
  I run test_progs in kselftest 100 times continuously for every condition,
  the numbers are in format: total/times=avg.
  The numbers that test_bpf reports show almost the same relation.

a:jit_enable=0 and jit_harden=0            b:jit_enable=1 and jit_harden=0
  test_pkt_access:PASS:ipv4:15622/100=156    test_pkt_access:PASS:ipv4:10674/100=106
  test_pkt_access:PASS:ipv6:9130/100=91      test_pkt_access:PASS:ipv6:4855/100=48
  test_xdp:PASS:ipv4:240198/100=2401         test_xdp:PASS:ipv4:138912/100=1389
  test_xdp:PASS:ipv6:137326/100=1373         test_xdp:PASS:ipv6:68542/100=685
  test_l4lb:PASS:ipv4:61100/100=611          test_l4lb:PASS:ipv4:37302/100=373
  test_l4lb:PASS:ipv6:101000/100=1010        test_l4lb:PASS:ipv6:55030/100=550

c:jit_enable=1 and jit_harden=2
  test_pkt_access:PASS:ipv4:10558/100=105
  test_pkt_access:PASS:ipv6:5092/100=50
  test_xdp:PASS:ipv4:131902/100=1319
  test_xdp:PASS:ipv6:77932/100=779
  test_l4lb:PASS:ipv4:38924/100=389
  test_l4lb:PASS:ipv6:57520/100=575

The numbers show we get 30%~50% improvement.

See Documentation/networking/filter.txt for more information.

Changelog:

 Changes v5-v6:
 1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
   consistence reason.
 2:Clean up non-standard comments, reported by Daniel Borkmann.
 3:Fix a memory leak issue, repoted by Daniel Borkmann.

 Changes v4-v5:
 1:Delete is_on_stack, BPF_REG_AX is the only one
   on real hardware registers, so just check with
   it.
 2:Apply commit 1612a981b7 ("bpf, x64: fix JIT emission
   for dead code"), suggested by Daniel Borkmann.

 Changes v3-v4:
 1:Fix changelog in commit.
   I install llvm-6.0, then test_progs willn't report errors.
   I submit another patch:
   "bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
   to fix another problem, after that patch, test_verifier willn't report errors too.
 2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.

 Changes v2-v3:
 1:Move BPF_REG_AX to real hardware registers for performance reason.
 3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
 4:Delete partial codes in 1c2a088a66, suggested by Daniel Borkmann.
 5:Some bug fixes and comments improvement.

 Changes v1-v2:
 1:Fix bug in emit_ia32_neg64.
 2:Fix bug in emit_ia32_arsh_r64.
 3:Delete filename in top level comment, suggested by Thomas Gleixner.
 4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
 5:Rewrite some words in changelog.
 6:CodingSytle improvement and a little more comments.

Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 18:15:25 +02:00

400 lines
14 KiB
Plaintext

Documentation for /proc/sys/net/*
(c) 1999 Terrehon Bowden <terrehon@pacbell.net>
Bodo Bauer <bb@ricochet.net>
(c) 2000 Jorge Nerin <comandante@zaralinux.com>
(c) 2009 Shen Feng <shen@cn.fujitsu.com>
For general info and legal blurb, please look in README.
==============================================================
This file contains the documentation for the sysctl files in
/proc/sys/net
The interface to the networking parts of the kernel is located in
/proc/sys/net. The following table shows all possible subdirectories. You may
see only some of them, depending on your kernel's configuration.
Table : Subdirectories in /proc/sys/net
..............................................................................
Directory Content Directory Content
core General parameter appletalk Appletalk protocol
unix Unix domain sockets netrom NET/ROM
802 E802 protocol ax25 AX25
ethernet Ethernet protocol rose X.25 PLP layer
ipv4 IP version 4 x25 X.25 protocol
ipx IPX token-ring IBM token ring
bridge Bridging decnet DEC net
ipv6 IP version 6 tipc TIPC
..............................................................................
1. /proc/sys/net/core - Network core options
-------------------------------------------------------
bpf_jit_enable
--------------
This enables the BPF Just in Time (JIT) compiler. BPF is a flexible
and efficient infrastructure allowing to execute bytecode at various
hook points. It is used in a number of Linux kernel subsystems such
as networking (e.g. XDP, tc), tracing (e.g. kprobes, uprobes, tracepoints)
and security (e.g. seccomp). LLVM has a BPF back end that can compile
restricted C into a sequence of BPF instructions. After program load
through bpf(2) and passing a verifier in the kernel, a JIT will then
translate these BPF proglets into native CPU instructions. There are
two flavors of JITs, the newer eBPF JIT currently supported on:
- x86_64
- x86_32
- arm64
- arm32
- ppc64
- sparc64
- mips64
- s390x
And the older cBPF JIT supported on the following archs:
- mips
- ppc
- sparc
eBPF JITs are a superset of cBPF JITs, meaning the kernel will
migrate cBPF instructions into eBPF instructions and then JIT
compile them transparently. Older cBPF JITs can only translate
tcpdump filters, seccomp rules, etc, but not mentioned eBPF
programs loaded through bpf(2).
Values :
0 - disable the JIT (default value)
1 - enable the JIT
2 - enable the JIT and ask the compiler to emit traces on kernel log.
bpf_jit_harden
--------------
This enables hardening for the BPF JIT compiler. Supported are eBPF
JIT backends. Enabling hardening trades off performance, but can
mitigate JIT spraying.
Values :
0 - disable JIT hardening (default value)
1 - enable JIT hardening for unprivileged users only
2 - enable JIT hardening for all users
bpf_jit_kallsyms
----------------
When BPF JIT compiler is enabled, then compiled images are unknown
addresses to the kernel, meaning they neither show up in traces nor
in /proc/kallsyms. This enables export of these addresses, which can
be used for debugging/tracing. If bpf_jit_harden is enabled, this
feature is disabled.
Values :
0 - disable JIT kallsyms export (default value)
1 - enable JIT kallsyms export for privileged users only
dev_weight
--------------
The maximum number of packets that kernel can handle on a NAPI interrupt,
it's a Per-CPU variable. For drivers that support LRO or GRO_HW, a hardware
aggregated packet is counted as one packet in this context.
Default: 64
dev_weight_rx_bias
--------------
RPS (e.g. RFS, aRFS) processing is competing with the registered NAPI poll function
of the driver for the per softirq cycle netdev_budget. This parameter influences
the proportion of the configured netdev_budget that is spent on RPS based packet
processing during RX softirq cycles. It is further meant for making current
dev_weight adaptable for asymmetric CPU needs on RX/TX side of the network stack.
(see dev_weight_tx_bias) It is effective on a per CPU basis. Determination is based
on dev_weight and is calculated multiplicative (dev_weight * dev_weight_rx_bias).
Default: 1
dev_weight_tx_bias
--------------
Scales the maximum number of packets that can be processed during a TX softirq cycle.
Effective on a per CPU basis. Allows scaling of current dev_weight for asymmetric
net stack processing needs. Be careful to avoid making TX softirq processing a CPU hog.
Calculation is based on dev_weight (dev_weight * dev_weight_tx_bias).
Default: 1
default_qdisc
--------------
The default queuing discipline to use for network devices. This allows
overriding the default of pfifo_fast with an alternative. Since the default
queuing discipline is created without additional parameters so is best suited
to queuing disciplines that work well without configuration like stochastic
fair queue (sfq), CoDel (codel) or fair queue CoDel (fq_codel). Don't use
queuing disciplines like Hierarchical Token Bucket or Deficit Round Robin
which require setting up classes and bandwidths. Note that physical multiqueue
interfaces still use mq as root qdisc, which in turn uses this default for its
leaves. Virtual devices (like e.g. lo or veth) ignore this setting and instead
default to noqueue.
Default: pfifo_fast
busy_read
----------------
Low latency busy poll timeout for socket reads. (needs CONFIG_NET_RX_BUSY_POLL)
Approximate time in us to busy loop waiting for packets on the device queue.
This sets the default value of the SO_BUSY_POLL socket option.
Can be set or overridden per socket by setting socket option SO_BUSY_POLL,
which is the preferred method of enabling. If you need to enable the feature
globally via sysctl, a value of 50 is recommended.
Will increase power usage.
Default: 0 (off)
busy_poll
----------------
Low latency busy poll timeout for poll and select. (needs CONFIG_NET_RX_BUSY_POLL)
Approximate time in us to busy loop waiting for events.
Recommended value depends on the number of sockets you poll on.
For several sockets 50, for several hundreds 100.
For more than that you probably want to use epoll.
Note that only sockets with SO_BUSY_POLL set will be busy polled,
so you want to either selectively set SO_BUSY_POLL on those sockets or set
sysctl.net.busy_read globally.
Will increase power usage.
Default: 0 (off)
rmem_default
------------
The default setting of the socket receive buffer in bytes.
rmem_max
--------
The maximum receive socket buffer size in bytes.
tstamp_allow_data
-----------------
Allow processes to receive tx timestamps looped together with the original
packet contents. If disabled, transmit timestamp requests from unprivileged
processes are dropped unless socket option SOF_TIMESTAMPING_OPT_TSONLY is set.
Default: 1 (on)
wmem_default
------------
The default setting (in bytes) of the socket send buffer.
wmem_max
--------
The maximum send socket buffer size in bytes.
message_burst and message_cost
------------------------------
These parameters are used to limit the warning messages written to the kernel
log from the networking code. They enforce a rate limit to make a
denial-of-service attack impossible. A higher message_cost factor, results in
fewer messages that will be written. Message_burst controls when messages will
be dropped. The default settings limit warning messages to one every five
seconds.
warnings
--------
This sysctl is now unused.
This was used to control console messages from the networking stack that
occur because of problems on the network like duplicate address or bad
checksums.
These messages are now emitted at KERN_DEBUG and can generally be enabled
and controlled by the dynamic_debug facility.
netdev_budget
-------------
Maximum number of packets taken from all interfaces in one polling cycle (NAPI
poll). In one polling cycle interfaces which are registered to polling are
probed in a round-robin manner. Also, a polling cycle may not exceed
netdev_budget_usecs microseconds, even if netdev_budget has not been
exhausted.
netdev_budget_usecs
---------------------
Maximum number of microseconds in one NAPI polling cycle. Polling
will exit when either netdev_budget_usecs have elapsed during the
poll cycle or the number of packets processed reaches netdev_budget.
netdev_max_backlog
------------------
Maximum number of packets, queued on the INPUT side, when the interface
receives packets faster than kernel can process them.
netdev_rss_key
--------------
RSS (Receive Side Scaling) enabled drivers use a 40 bytes host key that is
randomly generated.
Some user space might need to gather its content even if drivers do not
provide ethtool -x support yet.
myhost:~# cat /proc/sys/net/core/netdev_rss_key
84:50:f4:00:a8:15:d1:a7:e9:7f:1d:60:35:c7:47:25:42:97:74:ca:56:bb:b6:a1:d8: ... (52 bytes total)
File contains nul bytes if no driver ever called netdev_rss_key_fill() function.
Note:
/proc/sys/net/core/netdev_rss_key contains 52 bytes of key,
but most drivers only use 40 bytes of it.
myhost:~# ethtool -x eth0
RX flow hash indirection table for eth0 with 8 RX ring(s):
0: 0 1 2 3 4 5 6 7
RSS hash key:
84:50:f4:00:a8:15:d1:a7:e9:7f:1d:60:35:c7:47:25:42:97:74:ca:56:bb:b6:a1:d8:43:e3:c9:0c:fd:17:55:c2:3a:4d:69:ed:f1:42:89
netdev_tstamp_prequeue
----------------------
If set to 0, RX packet timestamps can be sampled after RPS processing, when
the target CPU processes packets. It might give some delay on timestamps, but
permit to distribute the load on several cpus.
If set to 1 (default), timestamps are sampled as soon as possible, before
queueing.
optmem_max
----------
Maximum ancillary buffer size allowed per socket. Ancillary data is a sequence
of struct cmsghdr structures with appended data.
fb_tunnels_only_for_init_net
----------------------------
Controls if fallback tunnels (like tunl0, gre0, gretap0, erspan0,
sit0, ip6tnl0, ip6gre0) are automatically created when a new
network namespace is created, if corresponding tunnel is present
in initial network namespace.
If set to 1, these devices are not automatically created, and
user space is responsible for creating them if needed.
Default : 0 (for compatibility reasons)
2. /proc/sys/net/unix - Parameters for Unix domain sockets
-------------------------------------------------------
There is only one file in this directory.
unix_dgram_qlen limits the max number of datagrams queued in Unix domain
socket's buffer. It will not take effect unless PF_UNIX flag is specified.
3. /proc/sys/net/ipv4 - IPV4 settings
-------------------------------------------------------
Please see: Documentation/networking/ip-sysctl.txt and ipvs-sysctl.txt for
descriptions of these entries.
4. Appletalk
-------------------------------------------------------
The /proc/sys/net/appletalk directory holds the Appletalk configuration data
when Appletalk is loaded. The configurable parameters are:
aarp-expiry-time
----------------
The amount of time we keep an ARP entry before expiring it. Used to age out
old hosts.
aarp-resolve-time
-----------------
The amount of time we will spend trying to resolve an Appletalk address.
aarp-retransmit-limit
---------------------
The number of times we will retransmit a query before giving up.
aarp-tick-time
--------------
Controls the rate at which expires are checked.
The directory /proc/net/appletalk holds the list of active Appletalk sockets
on a machine.
The fields indicate the DDP type, the local address (in network:node format)
the remote address, the size of the transmit pending queue, the size of the
received queue (bytes waiting for applications to read) the state and the uid
owning the socket.
/proc/net/atalk_iface lists all the interfaces configured for appletalk.It
shows the name of the interface, its Appletalk address, the network range on
that address (or network number for phase 1 networks), and the status of the
interface.
/proc/net/atalk_route lists each known network route. It lists the target
(network) that the route leads to, the router (may be directly connected), the
route flags, and the device the route is using.
5. IPX
-------------------------------------------------------
The IPX protocol has no tunable values in proc/sys/net.
The IPX protocol does, however, provide proc/net/ipx. This lists each IPX
socket giving the local and remote addresses in Novell format (that is
network:node:port). In accordance with the strange Novell tradition,
everything but the port is in hex. Not_Connected is displayed for sockets that
are not tied to a specific remote address. The Tx and Rx queue sizes indicate
the number of bytes pending for transmission and reception. The state
indicates the state the socket is in and the uid is the owning uid of the
socket.
The /proc/net/ipx_interface file lists all IPX interfaces. For each interface
it gives the network number, the node number, and indicates if the network is
the primary network. It also indicates which device it is bound to (or
Internal for internal networks) and the Frame Type if appropriate. Linux
supports 802.3, 802.2, 802.2 SNAP and DIX (Blue Book) ethernet framing for
IPX.
The /proc/net/ipx_route table holds a list of IPX routes. For each route it
gives the destination network, the router node (or Directly) and the network
address of the router (or Connected) for internal networks.
6. TIPC
-------------------------------------------------------
tipc_rmem
----------
The TIPC protocol now has a tunable for the receive memory, similar to the
tcp_rmem - i.e. a vector of 3 INTEGERs: (min, default, max)
# cat /proc/sys/net/tipc/tipc_rmem
4252725 34021800 68043600
#
The max value is set to CONN_OVERLOAD_LIMIT, and the default and min values
are scaled (shifted) versions of that same value. Note that the min value
is not at this point in time used in any meaningful way, but the triplet is
preserved in order to be consistent with things like tcp_rmem.
named_timeout
--------------
TIPC name table updates are distributed asynchronously in a cluster, without
any form of transaction handling. This means that different race scenarios are
possible. One such is that a name withdrawal sent out by one node and received
by another node may arrive after a second, overlapping name publication already
has been accepted from a third node, although the conflicting updates
originally may have been issued in the correct sequential order.
If named_timeout is nonzero, failed topology updates will be placed on a defer
queue until another event arrives that clears the error, or until the timeout
expires. Value is in milliseconds.