Loading...
1Documentation 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>
6
7For general info and legal blurb, please look in README.
8
9==============================================================
10
11This file contains the documentation for the sysctl files in
12/proc/sys/net
13
14The interface to the networking parts of the kernel is located in
15/proc/sys/net. The following table shows all possible subdirectories. You may
16see only some of them, depending on your kernel's configuration.
17
18
19Table : 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 bridge Bridging decnet DEC net
29 ipv6 IP version 6 tipc TIPC
30..............................................................................
31
321. /proc/sys/net/core - Network core options
33-------------------------------------------------------
34
35bpf_jit_enable
36--------------
37
38This enables the BPF Just in Time (JIT) compiler. BPF is a flexible
39and efficient infrastructure allowing to execute bytecode at various
40hook points. It is used in a number of Linux kernel subsystems such
41as networking (e.g. XDP, tc), tracing (e.g. kprobes, uprobes, tracepoints)
42and security (e.g. seccomp). LLVM has a BPF back end that can compile
43restricted C into a sequence of BPF instructions. After program load
44through bpf(2) and passing a verifier in the kernel, a JIT will then
45translate these BPF proglets into native CPU instructions. There are
46two flavors of JITs, the newer eBPF JIT currently supported on:
47 - x86_64
48 - arm64
49 - arm32
50 - ppc64
51 - sparc64
52 - mips64
53 - s390x
54
55And the older cBPF JIT supported on the following archs:
56 - mips
57 - ppc
58 - sparc
59
60eBPF JITs are a superset of cBPF JITs, meaning the kernel will
61migrate cBPF instructions into eBPF instructions and then JIT
62compile them transparently. Older cBPF JITs can only translate
63tcpdump filters, seccomp rules, etc, but not mentioned eBPF
64programs loaded through bpf(2).
65
66Values :
67 0 - disable the JIT (default value)
68 1 - enable the JIT
69 2 - enable the JIT and ask the compiler to emit traces on kernel log.
70
71bpf_jit_harden
72--------------
73
74This enables hardening for the BPF JIT compiler. Supported are eBPF
75JIT backends. Enabling hardening trades off performance, but can
76mitigate JIT spraying.
77Values :
78 0 - disable JIT hardening (default value)
79 1 - enable JIT hardening for unprivileged users only
80 2 - enable JIT hardening for all users
81
82bpf_jit_kallsyms
83----------------
84
85When BPF JIT compiler is enabled, then compiled images are unknown
86addresses to the kernel, meaning they neither show up in traces nor
87in /proc/kallsyms. This enables export of these addresses, which can
88be used for debugging/tracing. If bpf_jit_harden is enabled, this
89feature is disabled.
90Values :
91 0 - disable JIT kallsyms export (default value)
92 1 - enable JIT kallsyms export for privileged users only
93
94dev_weight
95--------------
96
97The maximum number of packets that kernel can handle on a NAPI interrupt,
98it's a Per-CPU variable. For drivers that support LRO or GRO_HW, a hardware
99aggregated packet is counted as one packet in this context.
100
101Default: 64
102
103dev_weight_rx_bias
104--------------
105
106RPS (e.g. RFS, aRFS) processing is competing with the registered NAPI poll function
107of the driver for the per softirq cycle netdev_budget. This parameter influences
108the proportion of the configured netdev_budget that is spent on RPS based packet
109processing during RX softirq cycles. It is further meant for making current
110dev_weight adaptable for asymmetric CPU needs on RX/TX side of the network stack.
111(see dev_weight_tx_bias) It is effective on a per CPU basis. Determination is based
112on dev_weight and is calculated multiplicative (dev_weight * dev_weight_rx_bias).
113Default: 1
114
115dev_weight_tx_bias
116--------------
117
118Scales the maximum number of packets that can be processed during a TX softirq cycle.
119Effective on a per CPU basis. Allows scaling of current dev_weight for asymmetric
120net stack processing needs. Be careful to avoid making TX softirq processing a CPU hog.
121Calculation is based on dev_weight (dev_weight * dev_weight_tx_bias).
122Default: 1
123
124default_qdisc
125--------------
126
127The default queuing discipline to use for network devices. This allows
128overriding the default of pfifo_fast with an alternative. Since the default
129queuing discipline is created without additional parameters so is best suited
130to queuing disciplines that work well without configuration like stochastic
131fair queue (sfq), CoDel (codel) or fair queue CoDel (fq_codel). Don't use
132queuing disciplines like Hierarchical Token Bucket or Deficit Round Robin
133which require setting up classes and bandwidths. Note that physical multiqueue
134interfaces still use mq as root qdisc, which in turn uses this default for its
135leaves. Virtual devices (like e.g. lo or veth) ignore this setting and instead
136default to noqueue.
137Default: pfifo_fast
138
139busy_read
140----------------
141Low latency busy poll timeout for socket reads. (needs CONFIG_NET_RX_BUSY_POLL)
142Approximate time in us to busy loop waiting for packets on the device queue.
143This sets the default value of the SO_BUSY_POLL socket option.
144Can be set or overridden per socket by setting socket option SO_BUSY_POLL,
145which is the preferred method of enabling. If you need to enable the feature
146globally via sysctl, a value of 50 is recommended.
147Will increase power usage.
148Default: 0 (off)
149
150busy_poll
151----------------
152Low latency busy poll timeout for poll and select. (needs CONFIG_NET_RX_BUSY_POLL)
153Approximate time in us to busy loop waiting for events.
154Recommended value depends on the number of sockets you poll on.
155For several sockets 50, for several hundreds 100.
156For more than that you probably want to use epoll.
157Note that only sockets with SO_BUSY_POLL set will be busy polled,
158so you want to either selectively set SO_BUSY_POLL on those sockets or set
159sysctl.net.busy_read globally.
160Will increase power usage.
161Default: 0 (off)
162
163rmem_default
164------------
165
166The default setting of the socket receive buffer in bytes.
167
168rmem_max
169--------
170
171The maximum receive socket buffer size in bytes.
172
173tstamp_allow_data
174-----------------
175Allow processes to receive tx timestamps looped together with the original
176packet contents. If disabled, transmit timestamp requests from unprivileged
177processes are dropped unless socket option SOF_TIMESTAMPING_OPT_TSONLY is set.
178Default: 1 (on)
179
180
181wmem_default
182------------
183
184The default setting (in bytes) of the socket send buffer.
185
186wmem_max
187--------
188
189The maximum send socket buffer size in bytes.
190
191message_burst and message_cost
192------------------------------
193
194These parameters are used to limit the warning messages written to the kernel
195log from the networking code. They enforce a rate limit to make a
196denial-of-service attack impossible. A higher message_cost factor, results in
197fewer messages that will be written. Message_burst controls when messages will
198be dropped. The default settings limit warning messages to one every five
199seconds.
200
201warnings
202--------
203
204This sysctl is now unused.
205
206This was used to control console messages from the networking stack that
207occur because of problems on the network like duplicate address or bad
208checksums.
209
210These messages are now emitted at KERN_DEBUG and can generally be enabled
211and controlled by the dynamic_debug facility.
212
213netdev_budget
214-------------
215
216Maximum number of packets taken from all interfaces in one polling cycle (NAPI
217poll). In one polling cycle interfaces which are registered to polling are
218probed in a round-robin manner. Also, a polling cycle may not exceed
219netdev_budget_usecs microseconds, even if netdev_budget has not been
220exhausted.
221
222netdev_budget_usecs
223---------------------
224
225Maximum number of microseconds in one NAPI polling cycle. Polling
226will exit when either netdev_budget_usecs have elapsed during the
227poll cycle or the number of packets processed reaches netdev_budget.
228
229netdev_max_backlog
230------------------
231
232Maximum number of packets, queued on the INPUT side, when the interface
233receives packets faster than kernel can process them.
234
235netdev_rss_key
236--------------
237
238RSS (Receive Side Scaling) enabled drivers use a 40 bytes host key that is
239randomly generated.
240Some user space might need to gather its content even if drivers do not
241provide ethtool -x support yet.
242
243myhost:~# cat /proc/sys/net/core/netdev_rss_key
24484:50: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)
245
246File contains nul bytes if no driver ever called netdev_rss_key_fill() function.
247Note:
248/proc/sys/net/core/netdev_rss_key contains 52 bytes of key,
249but most drivers only use 40 bytes of it.
250
251myhost:~# ethtool -x eth0
252RX flow hash indirection table for eth0 with 8 RX ring(s):
253 0: 0 1 2 3 4 5 6 7
254RSS hash key:
25584:50: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
256
257netdev_tstamp_prequeue
258----------------------
259
260If set to 0, RX packet timestamps can be sampled after RPS processing, when
261the target CPU processes packets. It might give some delay on timestamps, but
262permit to distribute the load on several cpus.
263
264If set to 1 (default), timestamps are sampled as soon as possible, before
265queueing.
266
267optmem_max
268----------
269
270Maximum ancillary buffer size allowed per socket. Ancillary data is a sequence
271of struct cmsghdr structures with appended data.
272
273fb_tunnels_only_for_init_net
274----------------------------
275
276Controls if fallback tunnels (like tunl0, gre0, gretap0, erspan0,
277sit0, ip6tnl0, ip6gre0) are automatically created when a new
278network namespace is created, if corresponding tunnel is present
279in initial network namespace.
280If set to 1, these devices are not automatically created, and
281user space is responsible for creating them if needed.
282
283Default : 0 (for compatibility reasons)
284
2852. /proc/sys/net/unix - Parameters for Unix domain sockets
286-------------------------------------------------------
287
288There is only one file in this directory.
289unix_dgram_qlen limits the max number of datagrams queued in Unix domain
290socket's buffer. It will not take effect unless PF_UNIX flag is specified.
291
292
2933. /proc/sys/net/ipv4 - IPV4 settings
294-------------------------------------------------------
295Please see: Documentation/networking/ip-sysctl.txt and ipvs-sysctl.txt for
296descriptions of these entries.
297
298
2994. Appletalk
300-------------------------------------------------------
301
302The /proc/sys/net/appletalk directory holds the Appletalk configuration data
303when Appletalk is loaded. The configurable parameters are:
304
305aarp-expiry-time
306----------------
307
308The amount of time we keep an ARP entry before expiring it. Used to age out
309old hosts.
310
311aarp-resolve-time
312-----------------
313
314The amount of time we will spend trying to resolve an Appletalk address.
315
316aarp-retransmit-limit
317---------------------
318
319The number of times we will retransmit a query before giving up.
320
321aarp-tick-time
322--------------
323
324Controls the rate at which expires are checked.
325
326The directory /proc/net/appletalk holds the list of active Appletalk sockets
327on a machine.
328
329The fields indicate the DDP type, the local address (in network:node format)
330the remote address, the size of the transmit pending queue, the size of the
331received queue (bytes waiting for applications to read) the state and the uid
332owning the socket.
333
334/proc/net/atalk_iface lists all the interfaces configured for appletalk.It
335shows the name of the interface, its Appletalk address, the network range on
336that address (or network number for phase 1 networks), and the status of the
337interface.
338
339/proc/net/atalk_route lists each known network route. It lists the target
340(network) that the route leads to, the router (may be directly connected), the
341route flags, and the device the route is using.
342
343
3445. IPX
345-------------------------------------------------------
346
347The IPX protocol has no tunable values in proc/sys/net.
348
349The IPX protocol does, however, provide proc/net/ipx. This lists each IPX
350socket giving the local and remote addresses in Novell format (that is
351network:node:port). In accordance with the strange Novell tradition,
352everything but the port is in hex. Not_Connected is displayed for sockets that
353are not tied to a specific remote address. The Tx and Rx queue sizes indicate
354the number of bytes pending for transmission and reception. The state
355indicates the state the socket is in and the uid is the owning uid of the
356socket.
357
358The /proc/net/ipx_interface file lists all IPX interfaces. For each interface
359it gives the network number, the node number, and indicates if the network is
360the primary network. It also indicates which device it is bound to (or
361Internal for internal networks) and the Frame Type if appropriate. Linux
362supports 802.3, 802.2, 802.2 SNAP and DIX (Blue Book) ethernet framing for
363IPX.
364
365The /proc/net/ipx_route table holds a list of IPX routes. For each route it
366gives the destination network, the router node (or Directly) and the network
367address of the router (or Connected) for internal networks.
368
3696. TIPC
370-------------------------------------------------------
371
372tipc_rmem
373----------
374
375The TIPC protocol now has a tunable for the receive memory, similar to the
376tcp_rmem - i.e. a vector of 3 INTEGERs: (min, default, max)
377
378 # cat /proc/sys/net/tipc/tipc_rmem
379 4252725 34021800 68043600
380 #
381
382The max value is set to CONN_OVERLOAD_LIMIT, and the default and min values
383are scaled (shifted) versions of that same value. Note that the min value
384is not at this point in time used in any meaningful way, but the triplet is
385preserved in order to be consistent with things like tcp_rmem.
386
387named_timeout
388--------------
389
390TIPC name table updates are distributed asynchronously in a cluster, without
391any form of transaction handling. This means that different race scenarios are
392possible. One such is that a name withdrawal sent out by one node and received
393by another node may arrive after a second, overlapping name publication already
394has been accepted from a third node, although the conflicting updates
395originally may have been issued in the correct sequential order.
396If named_timeout is nonzero, failed topology updates will be placed on a defer
397queue until another event arrives that clears the error, or until the timeout
398expires. Value is in milliseconds.