xref: /linux/Documentation/admin-guide/sysctl/net.rst (revision 7255fcc80d4b525cc10cfaaf7f485830d4ed2000)
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
76And the older cBPF JIT supported on the following archs:
77
78  - mips
79  - sparc
80
81eBPF JITs are a superset of cBPF JITs, meaning the kernel will
82migrate cBPF instructions into eBPF instructions and then JIT
83compile them transparently. Older cBPF JITs can only translate
84tcpdump filters, seccomp rules, etc, but not mentioned eBPF
85programs loaded through bpf(2).
86
87Values:
88
89	- 0 - disable the JIT (default value)
90	- 1 - enable the JIT
91	- 2 - enable the JIT and ask the compiler to emit traces on kernel log.
92
93bpf_jit_harden
94--------------
95
96This enables hardening for the BPF JIT compiler. Supported are eBPF
97JIT backends. Enabling hardening trades off performance, but can
98mitigate JIT spraying.
99
100Values:
101
102	- 0 - disable JIT hardening (default value)
103	- 1 - enable JIT hardening for unprivileged users only
104	- 2 - enable JIT hardening for all users
105
106where "privileged user" in this context means a process having
107CAP_BPF or CAP_SYS_ADMIN in the root user name space.
108
109bpf_jit_kallsyms
110----------------
111
112When BPF JIT compiler is enabled, then compiled images are unknown
113addresses to the kernel, meaning they neither show up in traces nor
114in /proc/kallsyms. This enables export of these addresses, which can
115be used for debugging/tracing. If bpf_jit_harden is enabled, this
116feature is disabled.
117
118Values :
119
120	- 0 - disable JIT kallsyms export (default value)
121	- 1 - enable JIT kallsyms export for privileged users only
122
123bpf_jit_limit
124-------------
125
126This enforces a global limit for memory allocations to the BPF JIT
127compiler in order to reject unprivileged JIT requests once it has
128been surpassed. bpf_jit_limit contains the value of the global limit
129in bytes.
130
131dev_weight
132----------
133
134The maximum number of packets that kernel can handle on a NAPI interrupt,
135it's a Per-CPU variable. For drivers that support LRO or GRO_HW, a hardware
136aggregated packet is counted as one packet in this context.
137
138Default: 64
139
140dev_weight_rx_bias
141------------------
142
143RPS (e.g. RFS, aRFS) processing is competing with the registered NAPI poll function
144of the driver for the per softirq cycle netdev_budget. This parameter influences
145the proportion of the configured netdev_budget that is spent on RPS based packet
146processing during RX softirq cycles. It is further meant for making current
147dev_weight adaptable for asymmetric CPU needs on RX/TX side of the network stack.
148(see dev_weight_tx_bias) It is effective on a per CPU basis. Determination is based
149on dev_weight and is calculated multiplicative (dev_weight * dev_weight_rx_bias).
150
151Default: 1
152
153dev_weight_tx_bias
154------------------
155
156Scales the maximum number of packets that can be processed during a TX softirq cycle.
157Effective on a per CPU basis. Allows scaling of current dev_weight for asymmetric
158net stack processing needs. Be careful to avoid making TX softirq processing a CPU hog.
159
160Calculation is based on dev_weight (dev_weight * dev_weight_tx_bias).
161
162Default: 1
163
164default_qdisc
165-------------
166
167The default queuing discipline to use for network devices. This allows
168overriding the default of pfifo_fast with an alternative. Since the default
169queuing discipline is created without additional parameters so is best suited
170to queuing disciplines that work well without configuration like stochastic
171fair queue (sfq), CoDel (codel) or fair queue CoDel (fq_codel). Don't use
172queuing disciplines like Hierarchical Token Bucket or Deficit Round Robin
173which require setting up classes and bandwidths. Note that physical multiqueue
174interfaces still use mq as root qdisc, which in turn uses this default for its
175leaves. Virtual devices (like e.g. lo or veth) ignore this setting and instead
176default to noqueue.
177
178Default: pfifo_fast
179
180busy_read
181---------
182
183Low latency busy poll timeout for socket reads. (needs CONFIG_NET_RX_BUSY_POLL)
184Approximate time in us to busy loop waiting for packets on the device queue.
185This sets the default value of the SO_BUSY_POLL socket option.
186Can be set or overridden per socket by setting socket option SO_BUSY_POLL,
187which is the preferred method of enabling. If you need to enable the feature
188globally via sysctl, a value of 50 is recommended.
189
190Will increase power usage.
191
192Default: 0 (off)
193
194busy_poll
195----------------
196Low latency busy poll timeout for poll and select. (needs CONFIG_NET_RX_BUSY_POLL)
197Approximate time in us to busy loop waiting for events.
198Recommended value depends on the number of sockets you poll on.
199For several sockets 50, for several hundreds 100.
200For more than that you probably want to use epoll.
201Note that only sockets with SO_BUSY_POLL set will be busy polled,
202so you want to either selectively set SO_BUSY_POLL on those sockets or set
203sysctl.net.busy_read globally.
204
205Will increase power usage.
206
207Default: 0 (off)
208
209mem_pcpu_rsv
210------------
211
212Per-cpu reserved forward alloc cache size in page units. Default 1MB per CPU.
213
214rmem_default
215------------
216
217The default setting of the socket receive buffer in bytes.
218
219rmem_max
220--------
221
222The maximum receive socket buffer size in bytes.
223
224rps_default_mask
225----------------
226
227The default RPS CPU mask used on newly created network devices. An empty
228mask means RPS disabled by default.
229
230tstamp_allow_data
231-----------------
232Allow processes to receive tx timestamps looped together with the original
233packet contents. If disabled, transmit timestamp requests from unprivileged
234processes are dropped unless socket option SOF_TIMESTAMPING_OPT_TSONLY is set.
235
236Default: 1 (on)
237
238
239wmem_default
240------------
241
242The default setting (in bytes) of the socket send buffer.
243
244wmem_max
245--------
246
247The maximum send socket buffer size in bytes.
248
249message_burst and message_cost
250------------------------------
251
252These parameters  are used to limit the warning messages written to the kernel
253log from  the  networking  code.  They  enforce  a  rate  limit  to  make  a
254denial-of-service attack  impossible. A higher message_cost factor, results in
255fewer messages that will be written. Message_burst controls when messages will
256be dropped.  The  default  settings  limit  warning messages to one every five
257seconds.
258
259warnings
260--------
261
262This sysctl is now unused.
263
264This was used to control console messages from the networking stack that
265occur because of problems on the network like duplicate address or bad
266checksums.
267
268These messages are now emitted at KERN_DEBUG and can generally be enabled
269and controlled by the dynamic_debug facility.
270
271netdev_budget
272-------------
273
274Maximum number of packets taken from all interfaces in one polling cycle (NAPI
275poll). In one polling cycle interfaces which are registered to polling are
276probed in a round-robin manner. Also, a polling cycle may not exceed
277netdev_budget_usecs microseconds, even if netdev_budget has not been
278exhausted.
279
280netdev_budget_usecs
281---------------------
282
283Maximum number of microseconds in one NAPI polling cycle. Polling
284will exit when either netdev_budget_usecs have elapsed during the
285poll cycle or the number of packets processed reaches netdev_budget.
286
287netdev_max_backlog
288------------------
289
290Maximum number of packets, queued on the INPUT side, when the interface
291receives packets faster than kernel can process them.
292
293netdev_rss_key
294--------------
295
296RSS (Receive Side Scaling) enabled drivers use a 40 bytes host key that is
297randomly generated.
298Some user space might need to gather its content even if drivers do not
299provide ethtool -x support yet.
300
301::
302
303  myhost:~# cat /proc/sys/net/core/netdev_rss_key
304  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
306File contains nul bytes if no driver ever called netdev_rss_key_fill() function.
307
308Note:
309  /proc/sys/net/core/netdev_rss_key contains 52 bytes of key,
310  but most drivers only use 40 bytes of it.
311
312::
313
314  myhost:~# ethtool -x eth0
315  RX flow hash indirection table for eth0 with 8 RX ring(s):
316      0:    0     1     2     3     4     5     6     7
317  RSS hash key:
318  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
320netdev_tstamp_prequeue
321----------------------
322
323If set to 0, RX packet timestamps can be sampled after RPS processing, when
324the target CPU processes packets. It might give some delay on timestamps, but
325permit to distribute the load on several cpus.
326
327If set to 1 (default), timestamps are sampled as soon as possible, before
328queueing.
329
330netdev_unregister_timeout_secs
331------------------------------
332
333Unregister network device timeout in seconds.
334This option controls the timeout (in seconds) used to issue a warning while
335waiting for a network device refcount to drop to 0 during device
336unregistration. A lower value may be useful during bisection to detect
337a leaked reference faster. A larger value may be useful to prevent false
338warnings on slow/loaded systems.
339Default value is 10, minimum 1, maximum 3600.
340
341skb_defer_max
342-------------
343
344Max size (in skbs) of the per-cpu list of skbs being freed
345by the cpu which allocated them. Used by TCP stack so far.
346
347Default: 64
348
349optmem_max
350----------
351
352Maximum ancillary buffer size allowed per socket. Ancillary data is a sequence
353of struct cmsghdr structures with appended data. TCP tx zerocopy also uses
354optmem_max as a limit for its internal structures.
355
356Default : 128 KB
357
358fb_tunnels_only_for_init_net
359----------------------------
360
361Controls if fallback tunnels (like tunl0, gre0, gretap0, erspan0,
362sit0, ip6tnl0, ip6gre0) are automatically created. There are 3 possibilities
363(a) value = 0; respective fallback tunnels are created when module is
364loaded in every net namespaces (backward compatible behavior).
365(b) value = 1; [kcmd value: initns] respective fallback tunnels are
366created only in init net namespace and every other net namespace will
367not have them.
368(c) value = 2; [kcmd value: none] fallback tunnels are not created
369when a module is loaded in any of the net namespace. Setting value to
370"2" is pointless after boot if these modules are built-in, so there is
371a kernel command-line option that can change this default. Please refer to
372Documentation/admin-guide/kernel-parameters.txt for additional details.
373
374Not creating fallback tunnels gives control to userspace to create
375whatever is needed only and avoid creating devices which are redundant.
376
377Default : 0  (for compatibility reasons)
378
379devconf_inherit_init_net
380------------------------
381
382Controls if a new network namespace should inherit all current
383settings under /proc/sys/net/{ipv4,ipv6}/conf/{all,default}/. By
384default, we keep the current behavior: for IPv4 we inherit all current
385settings from init_net and for IPv6 we reset all settings to default.
386
387If set to 1, both IPv4 and IPv6 settings are forced to inherit from
388current ones in init_net. If set to 2, both IPv4 and IPv6 settings are
389forced to reset to their default values. If set to 3, both IPv4 and IPv6
390settings are forced to inherit from current ones in the netns where this
391new netns has been created.
392
393Default : 0  (for compatibility reasons)
394
395txrehash
396--------
397
398Controls default hash rethink behaviour on socket when SO_TXREHASH option is set
399to SOCK_TXREHASH_DEFAULT (i. e. not overridden by setsockopt).
400
401If set to 1 (default), hash rethink is performed on listening socket.
402If set to 0, hash rethink is not performed.
403
404gro_normal_batch
405----------------
406
407Maximum number of the segments to batch up on output of GRO. When a packet
408exits GRO, either as a coalesced superframe or as an original packet which
409GRO has decided not to coalesce, it is placed on a per-NAPI list. This
410list is then passed to the stack when the number of segments reaches the
411gro_normal_batch limit.
412
413high_order_alloc_disable
414------------------------
415
416By default the allocator for page frags tries to use high order pages (order-3
417on x86). While the default behavior gives good results in most cases, some users
418might have hit a contention in page allocations/freeing. This was especially
419true on older kernels (< 5.14) when high-order pages were not stored on per-cpu
420lists. This allows to opt-in for order-0 allocation instead but is now mostly of
421historical importance.
422
423Default: 0
424
4252. /proc/sys/net/unix - Parameters for Unix domain sockets
426----------------------------------------------------------
427
428There is only one file in this directory.
429unix_dgram_qlen limits the max number of datagrams queued in Unix domain
430socket's buffer. It will not take effect unless PF_UNIX flag is specified.
431
432
4333. /proc/sys/net/ipv4 - IPV4 settings
434-------------------------------------
435Please see: Documentation/networking/ip-sysctl.rst and
436Documentation/admin-guide/sysctl/net.rst for descriptions of these entries.
437
438
4394. Appletalk
440------------
441
442The /proc/sys/net/appletalk  directory  holds the Appletalk configuration data
443when Appletalk is loaded. The configurable parameters are:
444
445aarp-expiry-time
446----------------
447
448The amount  of  time  we keep an ARP entry before expiring it. Used to age out
449old hosts.
450
451aarp-resolve-time
452-----------------
453
454The amount of time we will spend trying to resolve an Appletalk address.
455
456aarp-retransmit-limit
457---------------------
458
459The number of times we will retransmit a query before giving up.
460
461aarp-tick-time
462--------------
463
464Controls the rate at which expires are checked.
465
466The directory  /proc/net/appletalk  holds the list of active Appletalk sockets
467on a machine.
468
469The fields  indicate  the DDP type, the local address (in network:node format)
470the remote  address,  the  size of the transmit pending queue, the size of the
471received queue  (bytes waiting for applications to read) the state and the uid
472owning the socket.
473
474/proc/net/atalk_iface lists  all  the  interfaces  configured for appletalk.It
475shows the  name  of the interface, its Appletalk address, the network range on
476that address  (or  network number for phase 1 networks), and the status of the
477interface.
478
479/proc/net/atalk_route lists  each  known  network  route.  It lists the target
480(network) that the route leads to, the router (may be directly connected), the
481route flags, and the device the route is using.
482
4835. TIPC
484-------
485
486tipc_rmem
487---------
488
489The TIPC protocol now has a tunable for the receive memory, similar to the
490tcp_rmem - i.e. a vector of 3 INTEGERs: (min, default, max)
491
492::
493
494    # cat /proc/sys/net/tipc/tipc_rmem
495    4252725 34021800        68043600
496    #
497
498The max value is set to CONN_OVERLOAD_LIMIT, and the default and min values
499are scaled (shifted) versions of that same value.  Note that the min value
500is not at this point in time used in any meaningful way, but the triplet is
501preserved in order to be consistent with things like tcp_rmem.
502
503named_timeout
504-------------
505
506TIPC name table updates are distributed asynchronously in a cluster, without
507any form of transaction handling. This means that different race scenarios are
508possible. One such is that a name withdrawal sent out by one node and received
509by another node may arrive after a second, overlapping name publication already
510has been accepted from a third node, although the conflicting updates
511originally may have been issued in the correct sequential order.
512If named_timeout is nonzero, failed topology updates will be placed on a defer
513queue until another event arrives that clears the error, or until the timeout
514expires. Value is in milliseconds.
515