xref: /linux/Documentation/admin-guide/sysctl/net.rst (revision f14aa5ea415b8add245e976bfab96a12986c6843)
1================================
2Documentation for /proc/sys/net/
3================================
4
5Copyright
6
7Copyright (c) 1999
8
9	- Terrehon Bowden <terrehon@pacbell.net>
10	- Bodo Bauer <bb@ricochet.net>
11
12Copyright (c) 2000
13
14	- Jorge Nerin <comandante@zaralinux.com>
15
16Copyright (c) 2009
17
18	- Shen Feng <shen@cn.fujitsu.com>
19
20For general info and legal blurb, please look in index.rst.
21
22------------------------------------------------------------------------------
23
24This file contains the documentation for the sysctl files in
25/proc/sys/net
26
27The interface  to  the  networking  parts  of  the  kernel  is  located  in
28/proc/sys/net. The following table shows all possible subdirectories.  You may
29see only some of them, depending on your kernel's configuration.
30
31
32Table : Subdirectories in /proc/sys/net
33
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
44 ipv6      IP version 6
45 ========= =================== = ========== ===================
46
471. /proc/sys/net/core - Network core options
48============================================
49
50bpf_jit_enable
51--------------
52
53This enables the BPF Just in Time (JIT) compiler. BPF is a flexible
54and efficient infrastructure allowing to execute bytecode at various
55hook points. It is used in a number of Linux kernel subsystems such
56as networking (e.g. XDP, tc), tracing (e.g. kprobes, uprobes, tracepoints)
57and security (e.g. seccomp). LLVM has a BPF back end that can compile
58restricted C into a sequence of BPF instructions. After program load
59through bpf(2) and passing a verifier in the kernel, a JIT will then
60translate these BPF proglets into native CPU instructions. There are
61two flavors of JITs, the newer eBPF JIT currently supported on:
62
63  - x86_64
64  - x86_32
65  - arm64
66  - arm32
67  - ppc64
68  - ppc32
69  - sparc64
70  - mips64
71  - s390x
72  - riscv64
73  - riscv32
74  - loongarch64
75  - arc
76
77And the older cBPF JIT supported on the following archs:
78
79  - mips
80  - sparc
81
82eBPF JITs are a superset of cBPF JITs, meaning the kernel will
83migrate cBPF instructions into eBPF instructions and then JIT
84compile them transparently. Older cBPF JITs can only translate
85tcpdump filters, seccomp rules, etc, but not mentioned eBPF
86programs loaded through bpf(2).
87
88Values:
89
90	- 0 - disable the JIT (default value)
91	- 1 - enable the JIT
92	- 2 - enable the JIT and ask the compiler to emit traces on kernel log.
93
94bpf_jit_harden
95--------------
96
97This enables hardening for the BPF JIT compiler. Supported are eBPF
98JIT backends. Enabling hardening trades off performance, but can
99mitigate JIT spraying.
100
101Values:
102
103	- 0 - disable JIT hardening (default value)
104	- 1 - enable JIT hardening for unprivileged users only
105	- 2 - enable JIT hardening for all users
106
107where "privileged user" in this context means a process having
108CAP_BPF or CAP_SYS_ADMIN in the root user name space.
109
110bpf_jit_kallsyms
111----------------
112
113When BPF JIT compiler is enabled, then compiled images are unknown
114addresses to the kernel, meaning they neither show up in traces nor
115in /proc/kallsyms. This enables export of these addresses, which can
116be used for debugging/tracing. If bpf_jit_harden is enabled, this
117feature is disabled.
118
119Values :
120
121	- 0 - disable JIT kallsyms export (default value)
122	- 1 - enable JIT kallsyms export for privileged users only
123
124bpf_jit_limit
125-------------
126
127This enforces a global limit for memory allocations to the BPF JIT
128compiler in order to reject unprivileged JIT requests once it has
129been surpassed. bpf_jit_limit contains the value of the global limit
130in bytes.
131
132dev_weight
133----------
134
135The maximum number of packets that kernel can handle on a NAPI interrupt,
136it's a Per-CPU variable. For drivers that support LRO or GRO_HW, a hardware
137aggregated packet is counted as one packet in this context.
138
139Default: 64
140
141dev_weight_rx_bias
142------------------
143
144RPS (e.g. RFS, aRFS) processing is competing with the registered NAPI poll function
145of the driver for the per softirq cycle netdev_budget. This parameter influences
146the proportion of the configured netdev_budget that is spent on RPS based packet
147processing during RX softirq cycles. It is further meant for making current
148dev_weight adaptable for asymmetric CPU needs on RX/TX side of the network stack.
149(see dev_weight_tx_bias) It is effective on a per CPU basis. Determination is based
150on dev_weight and is calculated multiplicative (dev_weight * dev_weight_rx_bias).
151
152Default: 1
153
154dev_weight_tx_bias
155------------------
156
157Scales the maximum number of packets that can be processed during a TX softirq cycle.
158Effective on a per CPU basis. Allows scaling of current dev_weight for asymmetric
159net stack processing needs. Be careful to avoid making TX softirq processing a CPU hog.
160
161Calculation is based on dev_weight (dev_weight * dev_weight_tx_bias).
162
163Default: 1
164
165default_qdisc
166-------------
167
168The default queuing discipline to use for network devices. This allows
169overriding the default of pfifo_fast with an alternative. Since the default
170queuing discipline is created without additional parameters so is best suited
171to queuing disciplines that work well without configuration like stochastic
172fair queue (sfq), CoDel (codel) or fair queue CoDel (fq_codel). Don't use
173queuing disciplines like Hierarchical Token Bucket or Deficit Round Robin
174which require setting up classes and bandwidths. Note that physical multiqueue
175interfaces still use mq as root qdisc, which in turn uses this default for its
176leaves. Virtual devices (like e.g. lo or veth) ignore this setting and instead
177default to noqueue.
178
179Default: pfifo_fast
180
181busy_read
182---------
183
184Low latency busy poll timeout for socket reads. (needs CONFIG_NET_RX_BUSY_POLL)
185Approximate time in us to busy loop waiting for packets on the device queue.
186This sets the default value of the SO_BUSY_POLL socket option.
187Can be set or overridden per socket by setting socket option SO_BUSY_POLL,
188which is the preferred method of enabling. If you need to enable the feature
189globally via sysctl, a value of 50 is recommended.
190
191Will increase power usage.
192
193Default: 0 (off)
194
195busy_poll
196----------------
197Low latency busy poll timeout for poll and select. (needs CONFIG_NET_RX_BUSY_POLL)
198Approximate time in us to busy loop waiting for events.
199Recommended value depends on the number of sockets you poll on.
200For several sockets 50, for several hundreds 100.
201For more than that you probably want to use epoll.
202Note that only sockets with SO_BUSY_POLL set will be busy polled,
203so you want to either selectively set SO_BUSY_POLL on those sockets or set
204sysctl.net.busy_read globally.
205
206Will increase power usage.
207
208Default: 0 (off)
209
210mem_pcpu_rsv
211------------
212
213Per-cpu reserved forward alloc cache size in page units. Default 1MB per CPU.
214
215rmem_default
216------------
217
218The default setting of the socket receive buffer in bytes.
219
220rmem_max
221--------
222
223The maximum receive socket buffer size in bytes.
224
225rps_default_mask
226----------------
227
228The default RPS CPU mask used on newly created network devices. An empty
229mask means RPS disabled by default.
230
231tstamp_allow_data
232-----------------
233Allow processes to receive tx timestamps looped together with the original
234packet contents. If disabled, transmit timestamp requests from unprivileged
235processes are dropped unless socket option SOF_TIMESTAMPING_OPT_TSONLY is set.
236
237Default: 1 (on)
238
239
240wmem_default
241------------
242
243The default setting (in bytes) of the socket send buffer.
244
245wmem_max
246--------
247
248The maximum send socket buffer size in bytes.
249
250message_burst and message_cost
251------------------------------
252
253These parameters  are used to limit the warning messages written to the kernel
254log from  the  networking  code.  They  enforce  a  rate  limit  to  make  a
255denial-of-service attack  impossible. A higher message_cost factor, results in
256fewer messages that will be written. Message_burst controls when messages will
257be dropped.  The  default  settings  limit  warning messages to one every five
258seconds.
259
260warnings
261--------
262
263This sysctl is now unused.
264
265This was used to control console messages from the networking stack that
266occur because of problems on the network like duplicate address or bad
267checksums.
268
269These messages are now emitted at KERN_DEBUG and can generally be enabled
270and controlled by the dynamic_debug facility.
271
272netdev_budget
273-------------
274
275Maximum number of packets taken from all interfaces in one polling cycle (NAPI
276poll). In one polling cycle interfaces which are registered to polling are
277probed in a round-robin manner. Also, a polling cycle may not exceed
278netdev_budget_usecs microseconds, even if netdev_budget has not been
279exhausted.
280
281netdev_budget_usecs
282---------------------
283
284Maximum number of microseconds in one NAPI polling cycle. Polling
285will exit when either netdev_budget_usecs have elapsed during the
286poll cycle or the number of packets processed reaches netdev_budget.
287
288netdev_max_backlog
289------------------
290
291Maximum number of packets, queued on the INPUT side, when the interface
292receives packets faster than kernel can process them.
293
294netdev_rss_key
295--------------
296
297RSS (Receive Side Scaling) enabled drivers use a 40 bytes host key that is
298randomly generated.
299Some user space might need to gather its content even if drivers do not
300provide ethtool -x support yet.
301
302::
303
304  myhost:~# cat /proc/sys/net/core/netdev_rss_key
305  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)
306
307File contains nul bytes if no driver ever called netdev_rss_key_fill() function.
308
309Note:
310  /proc/sys/net/core/netdev_rss_key contains 52 bytes of key,
311  but most drivers only use 40 bytes of it.
312
313::
314
315  myhost:~# ethtool -x eth0
316  RX flow hash indirection table for eth0 with 8 RX ring(s):
317      0:    0     1     2     3     4     5     6     7
318  RSS hash key:
319  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
320
321netdev_tstamp_prequeue
322----------------------
323
324If set to 0, RX packet timestamps can be sampled after RPS processing, when
325the target CPU processes packets. It might give some delay on timestamps, but
326permit to distribute the load on several cpus.
327
328If set to 1 (default), timestamps are sampled as soon as possible, before
329queueing.
330
331netdev_unregister_timeout_secs
332------------------------------
333
334Unregister network device timeout in seconds.
335This option controls the timeout (in seconds) used to issue a warning while
336waiting for a network device refcount to drop to 0 during device
337unregistration. A lower value may be useful during bisection to detect
338a leaked reference faster. A larger value may be useful to prevent false
339warnings on slow/loaded systems.
340Default value is 10, minimum 1, maximum 3600.
341
342skb_defer_max
343-------------
344
345Max size (in skbs) of the per-cpu list of skbs being freed
346by the cpu which allocated them. Used by TCP stack so far.
347
348Default: 64
349
350optmem_max
351----------
352
353Maximum ancillary buffer size allowed per socket. Ancillary data is a sequence
354of struct cmsghdr structures with appended data. TCP tx zerocopy also uses
355optmem_max as a limit for its internal structures.
356
357Default : 128 KB
358
359fb_tunnels_only_for_init_net
360----------------------------
361
362Controls if fallback tunnels (like tunl0, gre0, gretap0, erspan0,
363sit0, ip6tnl0, ip6gre0) are automatically created. There are 3 possibilities
364(a) value = 0; respective fallback tunnels are created when module is
365loaded in every net namespaces (backward compatible behavior).
366(b) value = 1; [kcmd value: initns] respective fallback tunnels are
367created only in init net namespace and every other net namespace will
368not have them.
369(c) value = 2; [kcmd value: none] fallback tunnels are not created
370when a module is loaded in any of the net namespace. Setting value to
371"2" is pointless after boot if these modules are built-in, so there is
372a kernel command-line option that can change this default. Please refer to
373Documentation/admin-guide/kernel-parameters.txt for additional details.
374
375Not creating fallback tunnels gives control to userspace to create
376whatever is needed only and avoid creating devices which are redundant.
377
378Default : 0  (for compatibility reasons)
379
380devconf_inherit_init_net
381------------------------
382
383Controls if a new network namespace should inherit all current
384settings under /proc/sys/net/{ipv4,ipv6}/conf/{all,default}/. By
385default, we keep the current behavior: for IPv4 we inherit all current
386settings from init_net and for IPv6 we reset all settings to default.
387
388If set to 1, both IPv4 and IPv6 settings are forced to inherit from
389current ones in init_net. If set to 2, both IPv4 and IPv6 settings are
390forced to reset to their default values. If set to 3, both IPv4 and IPv6
391settings are forced to inherit from current ones in the netns where this
392new netns has been created.
393
394Default : 0  (for compatibility reasons)
395
396txrehash
397--------
398
399Controls default hash rethink behaviour on socket when SO_TXREHASH option is set
400to SOCK_TXREHASH_DEFAULT (i. e. not overridden by setsockopt).
401
402If set to 1 (default), hash rethink is performed on listening socket.
403If set to 0, hash rethink is not performed.
404
405gro_normal_batch
406----------------
407
408Maximum number of the segments to batch up on output of GRO. When a packet
409exits GRO, either as a coalesced superframe or as an original packet which
410GRO has decided not to coalesce, it is placed on a per-NAPI list. This
411list is then passed to the stack when the number of segments reaches the
412gro_normal_batch limit.
413
414high_order_alloc_disable
415------------------------
416
417By default the allocator for page frags tries to use high order pages (order-3
418on x86). While the default behavior gives good results in most cases, some users
419might have hit a contention in page allocations/freeing. This was especially
420true on older kernels (< 5.14) when high-order pages were not stored on per-cpu
421lists. This allows to opt-in for order-0 allocation instead but is now mostly of
422historical importance.
423
424Default: 0
425
4262. /proc/sys/net/unix - Parameters for Unix domain sockets
427----------------------------------------------------------
428
429There is only one file in this directory.
430unix_dgram_qlen limits the max number of datagrams queued in Unix domain
431socket's buffer. It will not take effect unless PF_UNIX flag is specified.
432
433
4343. /proc/sys/net/ipv4 - IPV4 settings
435-------------------------------------
436Please see: Documentation/networking/ip-sysctl.rst and
437Documentation/admin-guide/sysctl/net.rst for descriptions of these entries.
438
439
4404. Appletalk
441------------
442
443The /proc/sys/net/appletalk  directory  holds the Appletalk configuration data
444when Appletalk is loaded. The configurable parameters are:
445
446aarp-expiry-time
447----------------
448
449The amount  of  time  we keep an ARP entry before expiring it. Used to age out
450old hosts.
451
452aarp-resolve-time
453-----------------
454
455The amount of time we will spend trying to resolve an Appletalk address.
456
457aarp-retransmit-limit
458---------------------
459
460The number of times we will retransmit a query before giving up.
461
462aarp-tick-time
463--------------
464
465Controls the rate at which expires are checked.
466
467The directory  /proc/net/appletalk  holds the list of active Appletalk sockets
468on a machine.
469
470The fields  indicate  the DDP type, the local address (in network:node format)
471the remote  address,  the  size of the transmit pending queue, the size of the
472received queue  (bytes waiting for applications to read) the state and the uid
473owning the socket.
474
475/proc/net/atalk_iface lists  all  the  interfaces  configured for appletalk.It
476shows the  name  of the interface, its Appletalk address, the network range on
477that address  (or  network number for phase 1 networks), and the status of the
478interface.
479
480/proc/net/atalk_route lists  each  known  network  route.  It lists the target
481(network) that the route leads to, the router (may be directly connected), the
482route flags, and the device the route is using.
483
4845. TIPC
485-------
486
487tipc_rmem
488---------
489
490The TIPC protocol now has a tunable for the receive memory, similar to the
491tcp_rmem - i.e. a vector of 3 INTEGERs: (min, default, max)
492
493::
494
495    # cat /proc/sys/net/tipc/tipc_rmem
496    4252725 34021800        68043600
497    #
498
499The max value is set to CONN_OVERLOAD_LIMIT, and the default and min values
500are scaled (shifted) versions of that same value.  Note that the min value
501is not at this point in time used in any meaningful way, but the triplet is
502preserved in order to be consistent with things like tcp_rmem.
503
504named_timeout
505-------------
506
507TIPC name table updates are distributed asynchronously in a cluster, without
508any form of transaction handling. This means that different race scenarios are
509possible. One such is that a name withdrawal sent out by one node and received
510by another node may arrive after a second, overlapping name publication already
511has been accepted from a third node, although the conflicting updates
512originally may have been issued in the correct sequential order.
513If named_timeout is nonzero, failed topology updates will be placed on a defer
514queue until another event arrives that clears the error, or until the timeout
515expires. Value is in milliseconds.
516