1 Documentation for /proc/sys/net/*
2 (c) 1999 Terrehon Bowden <terrehon@pacbell.net>
3 Bodo Bauer <bb@ricochet.net>
4 (c) 2000 Jorge Nerin <comandante@zaralinux.com>
5 (c) 2009 Shen Feng <shen@cn.fujitsu.com>
7 For general info and legal blurb, please look in README.
9 ==============================================================
11 This file contains the documentation for the sysctl files in
14 The interface to the networking parts of the kernel is located in
15 /proc/sys/net. The following table shows all possible subdirectories. You may
16 see only some of them, depending on your kernel's configuration.
19 Table : Subdirectories in /proc/sys/net
20 ..............................................................................
21 Directory Content Directory Content
22 core General parameter appletalk Appletalk protocol
23 unix Unix domain sockets netrom NET/ROM
24 802 E802 protocol ax25 AX25
25 ethernet Ethernet protocol rose X.25 PLP layer
26 ipv4 IP version 4 x25 X.25 protocol
27 ipx IPX token-ring IBM token ring
28 ipv6 IP version 6 tipc TIPC
29 ..............................................................................
31 1. /proc/sys/net/core - Network core options
32 -------------------------------------------------------
37 This enables the BPF Just in Time (JIT) compiler. BPF is a flexible
38 and efficient infrastructure allowing to execute bytecode at various
39 hook points. It is used in a number of Linux kernel subsystems such
40 as networking (e.g. XDP, tc), tracing (e.g. kprobes, uprobes, tracepoints)
41 and security (e.g. seccomp). LLVM has a BPF back end that can compile
42 restricted C into a sequence of BPF instructions. After program load
43 through bpf(2) and passing a verifier in the kernel, a JIT will then
44 translate these BPF proglets into native CPU instructions. There are
45 two flavors of JITs, the newer eBPF JIT currently supported on:
54 And the older cBPF JIT supported on the following archs:
59 eBPF JITs are a superset of cBPF JITs, meaning the kernel will
60 migrate cBPF instructions into eBPF instructions and then JIT
61 compile them transparently. Older cBPF JITs can only translate
62 tcpdump filters, seccomp rules, etc, but not mentioned eBPF
63 programs loaded through bpf(2).
66 0 - disable the JIT (default value)
68 2 - enable the JIT and ask the compiler to emit traces on kernel log.
73 This enables hardening for the BPF JIT compiler. Supported are eBPF
74 JIT backends. Enabling hardening trades off performance, but can
75 mitigate JIT spraying.
77 0 - disable JIT hardening (default value)
78 1 - enable JIT hardening for unprivileged users only
79 2 - enable JIT hardening for all users
84 When BPF JIT compiler is enabled, then compiled images are unknown
85 addresses to the kernel, meaning they neither show up in traces nor
86 in /proc/kallsyms. This enables export of these addresses, which can
87 be used for debugging/tracing. If bpf_jit_harden is enabled, this
90 0 - disable JIT kallsyms export (default value)
91 1 - enable JIT kallsyms export for privileged users only
96 This enforces a global limit for memory allocations to the BPF JIT
97 compiler in order to reject unprivileged JIT requests once it has
98 been surpassed. bpf_jit_limit contains the value of the global limit
104 The maximum number of packets that kernel can handle on a NAPI interrupt,
105 it's a Per-CPU variable.
111 RPS (e.g. RFS, aRFS) processing is competing with the registered NAPI poll function
112 of the driver for the per softirq cycle netdev_budget. This parameter influences
113 the proportion of the configured netdev_budget that is spent on RPS based packet
114 processing during RX softirq cycles. It is further meant for making current
115 dev_weight adaptable for asymmetric CPU needs on RX/TX side of the network stack.
116 (see dev_weight_tx_bias) It is effective on a per CPU basis. Determination is based
117 on dev_weight and is calculated multiplicative (dev_weight * dev_weight_rx_bias).
123 Scales the maximum number of packets that can be processed during a TX softirq cycle.
124 Effective on a per CPU basis. Allows scaling of current dev_weight for asymmetric
125 net stack processing needs. Be careful to avoid making TX softirq processing a CPU hog.
126 Calculation is based on dev_weight (dev_weight * dev_weight_tx_bias).
132 The default queuing discipline to use for network devices. This allows
133 overriding the default of pfifo_fast with an alternative. Since the default
134 queuing discipline is created without additional parameters so is best suited
135 to queuing disciplines that work well without configuration like stochastic
136 fair queue (sfq), CoDel (codel) or fair queue CoDel (fq_codel). Don't use
137 queuing disciplines like Hierarchical Token Bucket or Deficit Round Robin
138 which require setting up classes and bandwidths. Note that physical multiqueue
139 interfaces still use mq as root qdisc, which in turn uses this default for its
140 leaves. Virtual devices (like e.g. lo or veth) ignore this setting and instead
146 Low latency busy poll timeout for socket reads. (needs CONFIG_NET_RX_BUSY_POLL)
147 Approximate time in us to busy loop waiting for packets on the device queue.
148 This sets the default value of the SO_BUSY_POLL socket option.
149 Can be set or overridden per socket by setting socket option SO_BUSY_POLL,
150 which is the preferred method of enabling. If you need to enable the feature
151 globally via sysctl, a value of 50 is recommended.
152 Will increase power usage.
157 Low latency busy poll timeout for poll and select. (needs CONFIG_NET_RX_BUSY_POLL)
158 Approximate time in us to busy loop waiting for events.
159 Recommended value depends on the number of sockets you poll on.
160 For several sockets 50, for several hundreds 100.
161 For more than that you probably want to use epoll.
162 Note that only sockets with SO_BUSY_POLL set will be busy polled,
163 so you want to either selectively set SO_BUSY_POLL on those sockets or set
164 sysctl.net.busy_read globally.
165 Will increase power usage.
171 The default setting of the socket receive buffer in bytes.
176 The maximum receive socket buffer size in bytes.
180 Allow processes to receive tx timestamps looped together with the original
181 packet contents. If disabled, transmit timestamp requests from unprivileged
182 processes are dropped unless socket option SOF_TIMESTAMPING_OPT_TSONLY is set.
189 The default setting (in bytes) of the socket send buffer.
194 The maximum send socket buffer size in bytes.
196 message_burst and message_cost
197 ------------------------------
199 These parameters are used to limit the warning messages written to the kernel
200 log from the networking code. They enforce a rate limit to make a
201 denial-of-service attack impossible. A higher message_cost factor, results in
202 fewer messages that will be written. Message_burst controls when messages will
203 be dropped. The default settings limit warning messages to one every five
209 This sysctl is now unused.
211 This was used to control console messages from the networking stack that
212 occur because of problems on the network like duplicate address or bad
215 These messages are now emitted at KERN_DEBUG and can generally be enabled
216 and controlled by the dynamic_debug facility.
221 Maximum number of packets taken from all interfaces in one polling cycle (NAPI
222 poll). In one polling cycle interfaces which are registered to polling are
223 probed in a round-robin manner. Also, a polling cycle may not exceed
224 netdev_budget_usecs microseconds, even if netdev_budget has not been
228 ---------------------
230 Maximum number of microseconds in one NAPI polling cycle. Polling
231 will exit when either netdev_budget_usecs have elapsed during the
232 poll cycle or the number of packets processed reaches netdev_budget.
237 Maximum number of packets, queued on the INPUT side, when the interface
238 receives packets faster than kernel can process them.
243 RSS (Receive Side Scaling) enabled drivers use a 40 bytes host key that is
245 Some user space might need to gather its content even if drivers do not
246 provide ethtool -x support yet.
248 myhost:~# cat /proc/sys/net/core/netdev_rss_key
249 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)
251 File contains nul bytes if no driver ever called netdev_rss_key_fill() function.
253 /proc/sys/net/core/netdev_rss_key contains 52 bytes of key,
254 but most drivers only use 40 bytes of it.
256 myhost:~# ethtool -x eth0
257 RX flow hash indirection table for eth0 with 8 RX ring(s):
260 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
262 netdev_tstamp_prequeue
263 ----------------------
265 If set to 0, RX packet timestamps can be sampled after RPS processing, when
266 the target CPU processes packets. It might give some delay on timestamps, but
267 permit to distribute the load on several cpus.
269 If set to 1 (default), timestamps are sampled as soon as possible, before
275 Maximum ancillary buffer size allowed per socket. Ancillary data is a sequence
276 of struct cmsghdr structures with appended data.
278 2. /proc/sys/net/unix - Parameters for Unix domain sockets
279 -------------------------------------------------------
281 There is only one file in this directory.
282 unix_dgram_qlen limits the max number of datagrams queued in Unix domain
283 socket's buffer. It will not take effect unless PF_UNIX flag is specified.
286 3. /proc/sys/net/ipv4 - IPV4 settings
287 -------------------------------------------------------
288 Please see: Documentation/networking/ip-sysctl.txt and ipvs-sysctl.txt for
289 descriptions of these entries.
293 -------------------------------------------------------
295 The /proc/sys/net/appletalk directory holds the Appletalk configuration data
296 when Appletalk is loaded. The configurable parameters are:
301 The amount of time we keep an ARP entry before expiring it. Used to age out
307 The amount of time we will spend trying to resolve an Appletalk address.
309 aarp-retransmit-limit
310 ---------------------
312 The number of times we will retransmit a query before giving up.
317 Controls the rate at which expires are checked.
319 The directory /proc/net/appletalk holds the list of active Appletalk sockets
322 The fields indicate the DDP type, the local address (in network:node format)
323 the remote address, the size of the transmit pending queue, the size of the
324 received queue (bytes waiting for applications to read) the state and the uid
327 /proc/net/atalk_iface lists all the interfaces configured for appletalk.It
328 shows the name of the interface, its Appletalk address, the network range on
329 that address (or network number for phase 1 networks), and the status of the
332 /proc/net/atalk_route lists each known network route. It lists the target
333 (network) that the route leads to, the router (may be directly connected), the
334 route flags, and the device the route is using.
338 -------------------------------------------------------
340 The IPX protocol has no tunable values in proc/sys/net.
342 The IPX protocol does, however, provide proc/net/ipx. This lists each IPX
343 socket giving the local and remote addresses in Novell format (that is
344 network:node:port). In accordance with the strange Novell tradition,
345 everything but the port is in hex. Not_Connected is displayed for sockets that
346 are not tied to a specific remote address. The Tx and Rx queue sizes indicate
347 the number of bytes pending for transmission and reception. The state
348 indicates the state the socket is in and the uid is the owning uid of the
351 The /proc/net/ipx_interface file lists all IPX interfaces. For each interface
352 it gives the network number, the node number, and indicates if the network is
353 the primary network. It also indicates which device it is bound to (or
354 Internal for internal networks) and the Frame Type if appropriate. Linux
355 supports 802.3, 802.2, 802.2 SNAP and DIX (Blue Book) ethernet framing for
358 The /proc/net/ipx_route table holds a list of IPX routes. For each route it
359 gives the destination network, the router node (or Directly) and the network
360 address of the router (or Connected) for internal networks.
363 -------------------------------------------------------
368 The TIPC protocol now has a tunable for the receive memory, similar to the
369 tcp_rmem - i.e. a vector of 3 INTEGERs: (min, default, max)
371 # cat /proc/sys/net/tipc/tipc_rmem
372 4252725 34021800 68043600
375 The max value is set to CONN_OVERLOAD_LIMIT, and the default and min values
376 are scaled (shifted) versions of that same value. Note that the min value
377 is not at this point in time used in any meaningful way, but the triplet is
378 preserved in order to be consistent with things like tcp_rmem.
383 TIPC name table updates are distributed asynchronously in a cluster, without
384 any form of transaction handling. This means that different race scenarios are
385 possible. One such is that a name withdrawal sent out by one node and received
386 by another node may arrive after a second, overlapping name publication already
387 has been accepted from a third node, although the conflicting updates
388 originally may have been issued in the correct sequential order.
389 If named_timeout is nonzero, failed topology updates will be placed on a defer
390 queue until another event arrives that clears the error, or until the timeout
391 expires. Value is in milliseconds.