1 ================================
2 Documentation for /proc/sys/net/
3 ================================
9 - Terrehon Bowden <terrehon@pacbell.net>
10 - Bodo Bauer <bb@ricochet.net>
14 - Jorge Nerin <comandante@zaralinux.com>
18 - Shen Feng <shen@cn.fujitsu.com>
20 For general info and legal blurb, please look in index.rst.
22 ------------------------------------------------------------------------------
24 This file contains the documentation for the sysctl files in
27 The interface to the networking parts of the kernel is located in
28 /proc/sys/net. The following table shows all possible subdirectories. You may
29 see only some of them, depending on your kernel's configuration.
32 Table : Subdirectories in /proc/sys/net
34 ========= =================== = ========== ==================
35 Directory Content Directory Content
36 ========= =================== = ========== ==================
37 core General parameter appletalk Appletalk protocol
38 unix Unix domain sockets netrom NET/ROM
39 802 E802 protocol ax25 AX25
40 ethernet Ethernet protocol rose X.25 PLP layer
41 ipv4 IP version 4 x25 X.25 protocol
42 bridge Bridging decnet DEC net
43 ipv6 IP version 6 tipc TIPC
44 ========= =================== = ========== ==================
46 1. /proc/sys/net/core - Network core options
47 ============================================
52 This enables the BPF Just in Time (JIT) compiler. BPF is a flexible
53 and efficient infrastructure allowing to execute bytecode at various
54 hook points. It is used in a number of Linux kernel subsystems such
55 as networking (e.g. XDP, tc), tracing (e.g. kprobes, uprobes, tracepoints)
56 and security (e.g. seccomp). LLVM has a BPF back end that can compile
57 restricted C into a sequence of BPF instructions. After program load
58 through bpf(2) and passing a verifier in the kernel, a JIT will then
59 translate these BPF proglets into native CPU instructions. There are
60 two flavors of JITs, the newer eBPF JIT currently supported on:
73 And the older cBPF JIT supported on the following archs:
79 eBPF JITs are a superset of cBPF JITs, meaning the kernel will
80 migrate cBPF instructions into eBPF instructions and then JIT
81 compile them transparently. Older cBPF JITs can only translate
82 tcpdump filters, seccomp rules, etc, but not mentioned eBPF
83 programs loaded through bpf(2).
87 - 0 - disable the JIT (default value)
89 - 2 - enable the JIT and ask the compiler to emit traces on kernel log.
94 This enables hardening for the BPF JIT compiler. Supported are eBPF
95 JIT backends. Enabling hardening trades off performance, but can
96 mitigate JIT spraying.
100 - 0 - disable JIT hardening (default value)
101 - 1 - enable JIT hardening for unprivileged users only
102 - 2 - enable JIT hardening for all users
107 When BPF JIT compiler is enabled, then compiled images are unknown
108 addresses to the kernel, meaning they neither show up in traces nor
109 in /proc/kallsyms. This enables export of these addresses, which can
110 be used for debugging/tracing. If bpf_jit_harden is enabled, this
115 - 0 - disable JIT kallsyms export (default value)
116 - 1 - enable JIT kallsyms export for privileged users only
121 This enforces a global limit for memory allocations to the BPF JIT
122 compiler in order to reject unprivileged JIT requests once it has
123 been surpassed. bpf_jit_limit contains the value of the global limit
129 The maximum number of packets that kernel can handle on a NAPI interrupt,
130 it's a Per-CPU variable. For drivers that support LRO or GRO_HW, a hardware
131 aggregated packet is counted as one packet in this context.
138 RPS (e.g. RFS, aRFS) processing is competing with the registered NAPI poll function
139 of the driver for the per softirq cycle netdev_budget. This parameter influences
140 the proportion of the configured netdev_budget that is spent on RPS based packet
141 processing during RX softirq cycles. It is further meant for making current
142 dev_weight adaptable for asymmetric CPU needs on RX/TX side of the network stack.
143 (see dev_weight_tx_bias) It is effective on a per CPU basis. Determination is based
144 on dev_weight and is calculated multiplicative (dev_weight * dev_weight_rx_bias).
151 Scales the maximum number of packets that can be processed during a TX softirq cycle.
152 Effective on a per CPU basis. Allows scaling of current dev_weight for asymmetric
153 net stack processing needs. Be careful to avoid making TX softirq processing a CPU hog.
155 Calculation is based on dev_weight (dev_weight * dev_weight_tx_bias).
162 The default queuing discipline to use for network devices. This allows
163 overriding the default of pfifo_fast with an alternative. Since the default
164 queuing discipline is created without additional parameters so is best suited
165 to queuing disciplines that work well without configuration like stochastic
166 fair queue (sfq), CoDel (codel) or fair queue CoDel (fq_codel). Don't use
167 queuing disciplines like Hierarchical Token Bucket or Deficit Round Robin
168 which require setting up classes and bandwidths. Note that physical multiqueue
169 interfaces still use mq as root qdisc, which in turn uses this default for its
170 leaves. Virtual devices (like e.g. lo or veth) ignore this setting and instead
178 Low latency busy poll timeout for socket reads. (needs CONFIG_NET_RX_BUSY_POLL)
179 Approximate time in us to busy loop waiting for packets on the device queue.
180 This sets the default value of the SO_BUSY_POLL socket option.
181 Can be set or overridden per socket by setting socket option SO_BUSY_POLL,
182 which is the preferred method of enabling. If you need to enable the feature
183 globally via sysctl, a value of 50 is recommended.
185 Will increase power usage.
191 Low latency busy poll timeout for poll and select. (needs CONFIG_NET_RX_BUSY_POLL)
192 Approximate time in us to busy loop waiting for events.
193 Recommended value depends on the number of sockets you poll on.
194 For several sockets 50, for several hundreds 100.
195 For more than that you probably want to use epoll.
196 Note that only sockets with SO_BUSY_POLL set will be busy polled,
197 so you want to either selectively set SO_BUSY_POLL on those sockets or set
198 sysctl.net.busy_read globally.
200 Will increase power usage.
207 The default setting of the socket receive buffer in bytes.
212 The maximum receive socket buffer size in bytes.
216 Allow processes to receive tx timestamps looped together with the original
217 packet contents. If disabled, transmit timestamp requests from unprivileged
218 processes are dropped unless socket option SOF_TIMESTAMPING_OPT_TSONLY is set.
226 The default setting (in bytes) of the socket send buffer.
231 The maximum send socket buffer size in bytes.
233 message_burst and message_cost
234 ------------------------------
236 These parameters are used to limit the warning messages written to the kernel
237 log from the networking code. They enforce a rate limit to make a
238 denial-of-service attack impossible. A higher message_cost factor, results in
239 fewer messages that will be written. Message_burst controls when messages will
240 be dropped. The default settings limit warning messages to one every five
246 This sysctl is now unused.
248 This was used to control console messages from the networking stack that
249 occur because of problems on the network like duplicate address or bad
252 These messages are now emitted at KERN_DEBUG and can generally be enabled
253 and controlled by the dynamic_debug facility.
258 Maximum number of packets taken from all interfaces in one polling cycle (NAPI
259 poll). In one polling cycle interfaces which are registered to polling are
260 probed in a round-robin manner. Also, a polling cycle may not exceed
261 netdev_budget_usecs microseconds, even if netdev_budget has not been
265 ---------------------
267 Maximum number of microseconds in one NAPI polling cycle. Polling
268 will exit when either netdev_budget_usecs have elapsed during the
269 poll cycle or the number of packets processed reaches netdev_budget.
274 Maximum number of packets, queued on the INPUT side, when the interface
275 receives packets faster than kernel can process them.
280 RSS (Receive Side Scaling) enabled drivers use a 40 bytes host key that is
282 Some user space might need to gather its content even if drivers do not
283 provide ethtool -x support yet.
287 myhost:~# cat /proc/sys/net/core/netdev_rss_key
288 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)
290 File contains nul bytes if no driver ever called netdev_rss_key_fill() function.
293 /proc/sys/net/core/netdev_rss_key contains 52 bytes of key,
294 but most drivers only use 40 bytes of it.
298 myhost:~# ethtool -x eth0
299 RX flow hash indirection table for eth0 with 8 RX ring(s):
302 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
304 netdev_tstamp_prequeue
305 ----------------------
307 If set to 0, RX packet timestamps can be sampled after RPS processing, when
308 the target CPU processes packets. It might give some delay on timestamps, but
309 permit to distribute the load on several cpus.
311 If set to 1 (default), timestamps are sampled as soon as possible, before
317 Maximum ancillary buffer size allowed per socket. Ancillary data is a sequence
318 of struct cmsghdr structures with appended data.
320 fb_tunnels_only_for_init_net
321 ----------------------------
323 Controls if fallback tunnels (like tunl0, gre0, gretap0, erspan0,
324 sit0, ip6tnl0, ip6gre0) are automatically created when a new
325 network namespace is created, if corresponding tunnel is present
326 in initial network namespace.
327 If set to 1, these devices are not automatically created, and
328 user space is responsible for creating them if needed.
330 Default : 0 (for compatibility reasons)
332 devconf_inherit_init_net
333 ------------------------
335 Controls if a new network namespace should inherit all current
336 settings under /proc/sys/net/{ipv4,ipv6}/conf/{all,default}/. By
337 default, we keep the current behavior: for IPv4 we inherit all current
338 settings from init_net and for IPv6 we reset all settings to default.
340 If set to 1, both IPv4 and IPv6 settings are forced to inherit from
341 current ones in init_net. If set to 2, both IPv4 and IPv6 settings are
342 forced to reset to their default values.
344 Default : 0 (for compatibility reasons)
346 2. /proc/sys/net/unix - Parameters for Unix domain sockets
347 ----------------------------------------------------------
349 There is only one file in this directory.
350 unix_dgram_qlen limits the max number of datagrams queued in Unix domain
351 socket's buffer. It will not take effect unless PF_UNIX flag is specified.
354 3. /proc/sys/net/ipv4 - IPV4 settings
355 -------------------------------------
356 Please see: Documentation/networking/ip-sysctl.txt and ipvs-sysctl.txt for
357 descriptions of these entries.
363 The /proc/sys/net/appletalk directory holds the Appletalk configuration data
364 when Appletalk is loaded. The configurable parameters are:
369 The amount of time we keep an ARP entry before expiring it. Used to age out
375 The amount of time we will spend trying to resolve an Appletalk address.
377 aarp-retransmit-limit
378 ---------------------
380 The number of times we will retransmit a query before giving up.
385 Controls the rate at which expires are checked.
387 The directory /proc/net/appletalk holds the list of active Appletalk sockets
390 The fields indicate the DDP type, the local address (in network:node format)
391 the remote address, the size of the transmit pending queue, the size of the
392 received queue (bytes waiting for applications to read) the state and the uid
395 /proc/net/atalk_iface lists all the interfaces configured for appletalk.It
396 shows the name of the interface, its Appletalk address, the network range on
397 that address (or network number for phase 1 networks), and the status of the
400 /proc/net/atalk_route lists each known network route. It lists the target
401 (network) that the route leads to, the router (may be directly connected), the
402 route flags, and the device the route is using.
410 The TIPC protocol now has a tunable for the receive memory, similar to the
411 tcp_rmem - i.e. a vector of 3 INTEGERs: (min, default, max)
415 # cat /proc/sys/net/tipc/tipc_rmem
416 4252725 34021800 68043600
419 The max value is set to CONN_OVERLOAD_LIMIT, and the default and min values
420 are scaled (shifted) versions of that same value. Note that the min value
421 is not at this point in time used in any meaningful way, but the triplet is
422 preserved in order to be consistent with things like tcp_rmem.
427 TIPC name table updates are distributed asynchronously in a cluster, without
428 any form of transaction handling. This means that different race scenarios are
429 possible. One such is that a name withdrawal sent out by one node and received
430 by another node may arrive after a second, overlapping name publication already
431 has been accepted from a third node, although the conflicting updates
432 originally may have been issued in the correct sequential order.
433 If named_timeout is nonzero, failed topology updates will be placed on a defer
434 queue until another event arrives that clears the error, or until the timeout
435 expires. Value is in milliseconds.