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 ========= =================== = ========== ===================
37 802 E802 protocol mptcp Multipath TCP
38 appletalk Appletalk protocol netfilter Network Filter
39 ax25 AX25 netrom NET/ROM
40 bridge Bridging rose X.25 PLP layer
41 core General parameter tipc TIPC
42 ethernet Ethernet protocol unix Unix domain sockets
43 ipv4 IP version 4 x25 X.25 protocol
45 ========= =================== = ========== ===================
47 1. /proc/sys/net/core - Network core options
48 ============================================
53 This enables the BPF Just in Time (JIT) compiler. BPF is a flexible
54 and efficient infrastructure allowing to execute bytecode at various
55 hook points. It is used in a number of Linux kernel subsystems such
56 as networking (e.g. XDP, tc), tracing (e.g. kprobes, uprobes, tracepoints)
57 and security (e.g. seccomp). LLVM has a BPF back end that can compile
58 restricted C into a sequence of BPF instructions. After program load
59 through bpf(2) and passing a verifier in the kernel, a JIT will then
60 translate these BPF proglets into native CPU instructions. There are
61 two flavors of JITs, the newer eBPF JIT currently supported on:
75 And the older cBPF JIT supported on the following archs:
80 eBPF JITs are a superset of cBPF JITs, meaning the kernel will
81 migrate cBPF instructions into eBPF instructions and then JIT
82 compile them transparently. Older cBPF JITs can only translate
83 tcpdump filters, seccomp rules, etc, but not mentioned eBPF
84 programs loaded through bpf(2).
88 - 0 - disable the JIT (default value)
90 - 2 - enable the JIT and ask the compiler to emit traces on kernel log.
95 This enables hardening for the BPF JIT compiler. Supported are eBPF
96 JIT backends. Enabling hardening trades off performance, but can
97 mitigate JIT spraying.
101 - 0 - disable JIT hardening (default value)
102 - 1 - enable JIT hardening for unprivileged users only
103 - 2 - enable JIT hardening for all users
105 where "privileged user" in this context means a process having
106 CAP_BPF or CAP_SYS_ADMIN in the root user name space.
111 When BPF JIT compiler is enabled, then compiled images are unknown
112 addresses to the kernel, meaning they neither show up in traces nor
113 in /proc/kallsyms. This enables export of these addresses, which can
114 be used for debugging/tracing. If bpf_jit_harden is enabled, this
119 - 0 - disable JIT kallsyms export (default value)
120 - 1 - enable JIT kallsyms export for privileged users only
125 This enforces a global limit for memory allocations to the BPF JIT
126 compiler in order to reject unprivileged JIT requests once it has
127 been surpassed. bpf_jit_limit contains the value of the global limit
133 The maximum number of packets that kernel can handle on a NAPI interrupt,
134 it's a Per-CPU variable. For drivers that support LRO or GRO_HW, a hardware
135 aggregated packet is counted as one packet in this context.
142 RPS (e.g. RFS, aRFS) processing is competing with the registered NAPI poll function
143 of the driver for the per softirq cycle netdev_budget. This parameter influences
144 the proportion of the configured netdev_budget that is spent on RPS based packet
145 processing during RX softirq cycles. It is further meant for making current
146 dev_weight adaptable for asymmetric CPU needs on RX/TX side of the network stack.
147 (see dev_weight_tx_bias) It is effective on a per CPU basis. Determination is based
148 on dev_weight and is calculated multiplicative (dev_weight * dev_weight_rx_bias).
155 Scales the maximum number of packets that can be processed during a TX softirq cycle.
156 Effective on a per CPU basis. Allows scaling of current dev_weight for asymmetric
157 net stack processing needs. Be careful to avoid making TX softirq processing a CPU hog.
159 Calculation is based on dev_weight (dev_weight * dev_weight_tx_bias).
166 The default queuing discipline to use for network devices. This allows
167 overriding the default of pfifo_fast with an alternative. Since the default
168 queuing discipline is created without additional parameters so is best suited
169 to queuing disciplines that work well without configuration like stochastic
170 fair queue (sfq), CoDel (codel) or fair queue CoDel (fq_codel). Don't use
171 queuing disciplines like Hierarchical Token Bucket or Deficit Round Robin
172 which require setting up classes and bandwidths. Note that physical multiqueue
173 interfaces still use mq as root qdisc, which in turn uses this default for its
174 leaves. Virtual devices (like e.g. lo or veth) ignore this setting and instead
182 Low latency busy poll timeout for socket reads. (needs CONFIG_NET_RX_BUSY_POLL)
183 Approximate time in us to busy loop waiting for packets on the device queue.
184 This sets the default value of the SO_BUSY_POLL socket option.
185 Can be set or overridden per socket by setting socket option SO_BUSY_POLL,
186 which is the preferred method of enabling. If you need to enable the feature
187 globally via sysctl, a value of 50 is recommended.
189 Will increase power usage.
195 Low latency busy poll timeout for poll and select. (needs CONFIG_NET_RX_BUSY_POLL)
196 Approximate time in us to busy loop waiting for events.
197 Recommended value depends on the number of sockets you poll on.
198 For several sockets 50, for several hundreds 100.
199 For more than that you probably want to use epoll.
200 Note that only sockets with SO_BUSY_POLL set will be busy polled,
201 so you want to either selectively set SO_BUSY_POLL on those sockets or set
202 sysctl.net.busy_read globally.
204 Will increase power usage.
211 Per-cpu reserved forward alloc cache size in page units. Default 1MB per CPU.
216 The default setting of the socket receive buffer in bytes.
221 The maximum receive socket buffer size in bytes.
226 The default RPS CPU mask used on newly created network devices. An empty
227 mask means RPS disabled by default.
231 Allow processes to receive tx timestamps looped together with the original
232 packet contents. If disabled, transmit timestamp requests from unprivileged
233 processes are dropped unless socket option SOF_TIMESTAMPING_OPT_TSONLY is set.
241 The default setting (in bytes) of the socket send buffer.
246 The maximum send socket buffer size in bytes.
248 message_burst and message_cost
249 ------------------------------
251 These parameters are used to limit the warning messages written to the kernel
252 log from the networking code. They enforce a rate limit to make a
253 denial-of-service attack impossible. A higher message_cost factor, results in
254 fewer messages that will be written. Message_burst controls when messages will
255 be dropped. The default settings limit warning messages to one every five
261 This sysctl is now unused.
263 This was used to control console messages from the networking stack that
264 occur because of problems on the network like duplicate address or bad
267 These messages are now emitted at KERN_DEBUG and can generally be enabled
268 and controlled by the dynamic_debug facility.
273 Maximum number of packets taken from all interfaces in one polling cycle (NAPI
274 poll). In one polling cycle interfaces which are registered to polling are
275 probed in a round-robin manner. Also, a polling cycle may not exceed
276 netdev_budget_usecs microseconds, even if netdev_budget has not been
280 ---------------------
282 Maximum number of microseconds in one NAPI polling cycle. Polling
283 will exit when either netdev_budget_usecs have elapsed during the
284 poll cycle or the number of packets processed reaches netdev_budget.
289 Maximum number of packets, queued on the INPUT side, when the interface
290 receives packets faster than kernel can process them.
295 RSS (Receive Side Scaling) enabled drivers use a 40 bytes host key that is
297 Some user space might need to gather its content even if drivers do not
298 provide ethtool -x support yet.
302 myhost:~# cat /proc/sys/net/core/netdev_rss_key
303 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)
305 File contains nul bytes if no driver ever called netdev_rss_key_fill() function.
308 /proc/sys/net/core/netdev_rss_key contains 52 bytes of key,
309 but most drivers only use 40 bytes of it.
313 myhost:~# ethtool -x eth0
314 RX flow hash indirection table for eth0 with 8 RX ring(s):
317 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
319 netdev_tstamp_prequeue
320 ----------------------
322 If set to 0, RX packet timestamps can be sampled after RPS processing, when
323 the target CPU processes packets. It might give some delay on timestamps, but
324 permit to distribute the load on several cpus.
326 If set to 1 (default), timestamps are sampled as soon as possible, before
329 netdev_unregister_timeout_secs
330 ------------------------------
332 Unregister network device timeout in seconds.
333 This option controls the timeout (in seconds) used to issue a warning while
334 waiting for a network device refcount to drop to 0 during device
335 unregistration. A lower value may be useful during bisection to detect
336 a leaked reference faster. A larger value may be useful to prevent false
337 warnings on slow/loaded systems.
338 Default value is 10, minimum 1, maximum 3600.
343 Max size (in skbs) of the per-cpu list of skbs being freed
344 by the cpu which allocated them. Used by TCP stack so far.
351 Maximum ancillary buffer size allowed per socket. Ancillary data is a sequence
352 of struct cmsghdr structures with appended data.
354 fb_tunnels_only_for_init_net
355 ----------------------------
357 Controls if fallback tunnels (like tunl0, gre0, gretap0, erspan0,
358 sit0, ip6tnl0, ip6gre0) are automatically created. There are 3 possibilities
359 (a) value = 0; respective fallback tunnels are created when module is
360 loaded in every net namespaces (backward compatible behavior).
361 (b) value = 1; [kcmd value: initns] respective fallback tunnels are
362 created only in init net namespace and every other net namespace will
364 (c) value = 2; [kcmd value: none] fallback tunnels are not created
365 when a module is loaded in any of the net namespace. Setting value to
366 "2" is pointless after boot if these modules are built-in, so there is
367 a kernel command-line option that can change this default. Please refer to
368 Documentation/admin-guide/kernel-parameters.txt for additional details.
370 Not creating fallback tunnels gives control to userspace to create
371 whatever is needed only and avoid creating devices which are redundant.
373 Default : 0 (for compatibility reasons)
375 devconf_inherit_init_net
376 ------------------------
378 Controls if a new network namespace should inherit all current
379 settings under /proc/sys/net/{ipv4,ipv6}/conf/{all,default}/. By
380 default, we keep the current behavior: for IPv4 we inherit all current
381 settings from init_net and for IPv6 we reset all settings to default.
383 If set to 1, both IPv4 and IPv6 settings are forced to inherit from
384 current ones in init_net. If set to 2, both IPv4 and IPv6 settings are
385 forced to reset to their default values. If set to 3, both IPv4 and IPv6
386 settings are forced to inherit from current ones in the netns where this
387 new netns has been created.
389 Default : 0 (for compatibility reasons)
394 Controls default hash rethink behaviour on socket when SO_TXREHASH option is set
395 to SOCK_TXREHASH_DEFAULT (i. e. not overridden by setsockopt).
397 If set to 1 (default), hash rethink is performed on listening socket.
398 If set to 0, hash rethink is not performed.
403 Maximum number of the segments to batch up on output of GRO. When a packet
404 exits GRO, either as a coalesced superframe or as an original packet which
405 GRO has decided not to coalesce, it is placed on a per-NAPI list. This
406 list is then passed to the stack when the number of segments reaches the
407 gro_normal_batch limit.
409 high_order_alloc_disable
410 ------------------------
412 By default the allocator for page frags tries to use high order pages (order-3
413 on x86). While the default behavior gives good results in most cases, some users
414 might have hit a contention in page allocations/freeing. This was especially
415 true on older kernels (< 5.14) when high-order pages were not stored on per-cpu
416 lists. This allows to opt-in for order-0 allocation instead but is now mostly of
417 historical importance.
421 2. /proc/sys/net/unix - Parameters for Unix domain sockets
422 ----------------------------------------------------------
424 There is only one file in this directory.
425 unix_dgram_qlen limits the max number of datagrams queued in Unix domain
426 socket's buffer. It will not take effect unless PF_UNIX flag is specified.
429 3. /proc/sys/net/ipv4 - IPV4 settings
430 -------------------------------------
431 Please see: Documentation/networking/ip-sysctl.rst and
432 Documentation/admin-guide/sysctl/net.rst for descriptions of these entries.
438 The /proc/sys/net/appletalk directory holds the Appletalk configuration data
439 when Appletalk is loaded. The configurable parameters are:
444 The amount of time we keep an ARP entry before expiring it. Used to age out
450 The amount of time we will spend trying to resolve an Appletalk address.
452 aarp-retransmit-limit
453 ---------------------
455 The number of times we will retransmit a query before giving up.
460 Controls the rate at which expires are checked.
462 The directory /proc/net/appletalk holds the list of active Appletalk sockets
465 The fields indicate the DDP type, the local address (in network:node format)
466 the remote address, the size of the transmit pending queue, the size of the
467 received queue (bytes waiting for applications to read) the state and the uid
470 /proc/net/atalk_iface lists all the interfaces configured for appletalk.It
471 shows the name of the interface, its Appletalk address, the network range on
472 that address (or network number for phase 1 networks), and the status of the
475 /proc/net/atalk_route lists each known network route. It lists the target
476 (network) that the route leads to, the router (may be directly connected), the
477 route flags, and the device the route is using.
485 The TIPC protocol now has a tunable for the receive memory, similar to the
486 tcp_rmem - i.e. a vector of 3 INTEGERs: (min, default, max)
490 # cat /proc/sys/net/tipc/tipc_rmem
491 4252725 34021800 68043600
494 The max value is set to CONN_OVERLOAD_LIMIT, and the default and min values
495 are scaled (shifted) versions of that same value. Note that the min value
496 is not at this point in time used in any meaningful way, but the triplet is
497 preserved in order to be consistent with things like tcp_rmem.
502 TIPC name table updates are distributed asynchronously in a cluster, without
503 any form of transaction handling. This means that different race scenarios are
504 possible. One such is that a name withdrawal sent out by one node and received
505 by another node may arrive after a second, overlapping name publication already
506 has been accepted from a third node, although the conflicting updates
507 originally may have been issued in the correct sequential order.
508 If named_timeout is nonzero, failed topology updates will be placed on a defer
509 queue until another event arrives that clears the error, or until the timeout
510 expires. Value is in milliseconds.