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 75And the older cBPF JIT supported on the following archs: 76 77 - mips 78 - sparc 79 80eBPF JITs are a superset of cBPF JITs, meaning the kernel will 81migrate cBPF instructions into eBPF instructions and then JIT 82compile them transparently. Older cBPF JITs can only translate 83tcpdump filters, seccomp rules, etc, but not mentioned eBPF 84programs loaded through bpf(2). 85 86Values: 87 88 - 0 - disable the JIT (default value) 89 - 1 - enable the JIT 90 - 2 - enable the JIT and ask the compiler to emit traces on kernel log. 91 92bpf_jit_harden 93-------------- 94 95This enables hardening for the BPF JIT compiler. Supported are eBPF 96JIT backends. Enabling hardening trades off performance, but can 97mitigate JIT spraying. 98 99Values: 100 101 - 0 - disable JIT hardening (default value) 102 - 1 - enable JIT hardening for unprivileged users only 103 - 2 - enable JIT hardening for all users 104 105where "privileged user" in this context means a process having 106CAP_BPF or CAP_SYS_ADMIN in the root user name space. 107 108bpf_jit_kallsyms 109---------------- 110 111When BPF JIT compiler is enabled, then compiled images are unknown 112addresses to the kernel, meaning they neither show up in traces nor 113in /proc/kallsyms. This enables export of these addresses, which can 114be used for debugging/tracing. If bpf_jit_harden is enabled, this 115feature is disabled. 116 117Values : 118 119 - 0 - disable JIT kallsyms export (default value) 120 - 1 - enable JIT kallsyms export for privileged users only 121 122bpf_jit_limit 123------------- 124 125This enforces a global limit for memory allocations to the BPF JIT 126compiler in order to reject unprivileged JIT requests once it has 127been surpassed. bpf_jit_limit contains the value of the global limit 128in bytes. 129 130dev_weight 131---------- 132 133The maximum number of packets that kernel can handle on a NAPI interrupt, 134it's a Per-CPU variable. For drivers that support LRO or GRO_HW, a hardware 135aggregated packet is counted as one packet in this context. 136 137Default: 64 138 139dev_weight_rx_bias 140------------------ 141 142RPS (e.g. RFS, aRFS) processing is competing with the registered NAPI poll function 143of the driver for the per softirq cycle netdev_budget. This parameter influences 144the proportion of the configured netdev_budget that is spent on RPS based packet 145processing during RX softirq cycles. It is further meant for making current 146dev_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 148on dev_weight and is calculated multiplicative (dev_weight * dev_weight_rx_bias). 149 150Default: 1 151 152dev_weight_tx_bias 153------------------ 154 155Scales the maximum number of packets that can be processed during a TX softirq cycle. 156Effective on a per CPU basis. Allows scaling of current dev_weight for asymmetric 157net stack processing needs. Be careful to avoid making TX softirq processing a CPU hog. 158 159Calculation is based on dev_weight (dev_weight * dev_weight_tx_bias). 160 161Default: 1 162 163default_qdisc 164------------- 165 166The default queuing discipline to use for network devices. This allows 167overriding the default of pfifo_fast with an alternative. Since the default 168queuing discipline is created without additional parameters so is best suited 169to queuing disciplines that work well without configuration like stochastic 170fair queue (sfq), CoDel (codel) or fair queue CoDel (fq_codel). Don't use 171queuing disciplines like Hierarchical Token Bucket or Deficit Round Robin 172which require setting up classes and bandwidths. Note that physical multiqueue 173interfaces still use mq as root qdisc, which in turn uses this default for its 174leaves. Virtual devices (like e.g. lo or veth) ignore this setting and instead 175default to noqueue. 176 177Default: pfifo_fast 178 179busy_read 180--------- 181 182Low latency busy poll timeout for socket reads. (needs CONFIG_NET_RX_BUSY_POLL) 183Approximate time in us to busy loop waiting for packets on the device queue. 184This sets the default value of the SO_BUSY_POLL socket option. 185Can be set or overridden per socket by setting socket option SO_BUSY_POLL, 186which is the preferred method of enabling. If you need to enable the feature 187globally via sysctl, a value of 50 is recommended. 188 189Will increase power usage. 190 191Default: 0 (off) 192 193busy_poll 194---------------- 195Low latency busy poll timeout for poll and select. (needs CONFIG_NET_RX_BUSY_POLL) 196Approximate time in us to busy loop waiting for events. 197Recommended value depends on the number of sockets you poll on. 198For several sockets 50, for several hundreds 100. 199For more than that you probably want to use epoll. 200Note that only sockets with SO_BUSY_POLL set will be busy polled, 201so you want to either selectively set SO_BUSY_POLL on those sockets or set 202sysctl.net.busy_read globally. 203 204Will increase power usage. 205 206Default: 0 (off) 207 208rmem_default 209------------ 210 211The default setting of the socket receive buffer in bytes. 212 213rmem_max 214-------- 215 216The maximum receive socket buffer size in bytes. 217 218rps_default_mask 219---------------- 220 221The default RPS CPU mask used on newly created network devices. An empty 222mask means RPS disabled by default. 223 224tstamp_allow_data 225----------------- 226Allow processes to receive tx timestamps looped together with the original 227packet contents. If disabled, transmit timestamp requests from unprivileged 228processes are dropped unless socket option SOF_TIMESTAMPING_OPT_TSONLY is set. 229 230Default: 1 (on) 231 232 233wmem_default 234------------ 235 236The default setting (in bytes) of the socket send buffer. 237 238wmem_max 239-------- 240 241The maximum send socket buffer size in bytes. 242 243message_burst and message_cost 244------------------------------ 245 246These parameters are used to limit the warning messages written to the kernel 247log from the networking code. They enforce a rate limit to make a 248denial-of-service attack impossible. A higher message_cost factor, results in 249fewer messages that will be written. Message_burst controls when messages will 250be dropped. The default settings limit warning messages to one every five 251seconds. 252 253warnings 254-------- 255 256This sysctl is now unused. 257 258This was used to control console messages from the networking stack that 259occur because of problems on the network like duplicate address or bad 260checksums. 261 262These messages are now emitted at KERN_DEBUG and can generally be enabled 263and controlled by the dynamic_debug facility. 264 265netdev_budget 266------------- 267 268Maximum number of packets taken from all interfaces in one polling cycle (NAPI 269poll). In one polling cycle interfaces which are registered to polling are 270probed in a round-robin manner. Also, a polling cycle may not exceed 271netdev_budget_usecs microseconds, even if netdev_budget has not been 272exhausted. 273 274netdev_budget_usecs 275--------------------- 276 277Maximum number of microseconds in one NAPI polling cycle. Polling 278will exit when either netdev_budget_usecs have elapsed during the 279poll cycle or the number of packets processed reaches netdev_budget. 280 281netdev_max_backlog 282------------------ 283 284Maximum number of packets, queued on the INPUT side, when the interface 285receives packets faster than kernel can process them. 286 287netdev_rss_key 288-------------- 289 290RSS (Receive Side Scaling) enabled drivers use a 40 bytes host key that is 291randomly generated. 292Some user space might need to gather its content even if drivers do not 293provide ethtool -x support yet. 294 295:: 296 297 myhost:~# cat /proc/sys/net/core/netdev_rss_key 298 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) 299 300File contains nul bytes if no driver ever called netdev_rss_key_fill() function. 301 302Note: 303 /proc/sys/net/core/netdev_rss_key contains 52 bytes of key, 304 but most drivers only use 40 bytes of it. 305 306:: 307 308 myhost:~# ethtool -x eth0 309 RX flow hash indirection table for eth0 with 8 RX ring(s): 310 0: 0 1 2 3 4 5 6 7 311 RSS hash key: 312 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:f1:42:89 313 314netdev_tstamp_prequeue 315---------------------- 316 317If set to 0, RX packet timestamps can be sampled after RPS processing, when 318the target CPU processes packets. It might give some delay on timestamps, but 319permit to distribute the load on several cpus. 320 321If set to 1 (default), timestamps are sampled as soon as possible, before 322queueing. 323 324netdev_unregister_timeout_secs 325------------------------------ 326 327Unregister network device timeout in seconds. 328This option controls the timeout (in seconds) used to issue a warning while 329waiting for a network device refcount to drop to 0 during device 330unregistration. A lower value may be useful during bisection to detect 331a leaked reference faster. A larger value may be useful to prevent false 332warnings on slow/loaded systems. 333Default value is 10, minimum 1, maximum 3600. 334 335skb_defer_max 336------------- 337 338Max size (in skbs) of the per-cpu list of skbs being freed 339by the cpu which allocated them. Used by TCP stack so far. 340 341Default: 64 342 343optmem_max 344---------- 345 346Maximum ancillary buffer size allowed per socket. Ancillary data is a sequence 347of struct cmsghdr structures with appended data. 348 349fb_tunnels_only_for_init_net 350---------------------------- 351 352Controls if fallback tunnels (like tunl0, gre0, gretap0, erspan0, 353sit0, ip6tnl0, ip6gre0) are automatically created. There are 3 possibilities 354(a) value = 0; respective fallback tunnels are created when module is 355loaded in every net namespaces (backward compatible behavior). 356(b) value = 1; [kcmd value: initns] respective fallback tunnels are 357created only in init net namespace and every other net namespace will 358not have them. 359(c) value = 2; [kcmd value: none] fallback tunnels are not created 360when a module is loaded in any of the net namespace. Setting value to 361"2" is pointless after boot if these modules are built-in, so there is 362a kernel command-line option that can change this default. Please refer to 363Documentation/admin-guide/kernel-parameters.txt for additional details. 364 365Not creating fallback tunnels gives control to userspace to create 366whatever is needed only and avoid creating devices which are redundant. 367 368Default : 0 (for compatibility reasons) 369 370devconf_inherit_init_net 371------------------------ 372 373Controls if a new network namespace should inherit all current 374settings under /proc/sys/net/{ipv4,ipv6}/conf/{all,default}/. By 375default, we keep the current behavior: for IPv4 we inherit all current 376settings from init_net and for IPv6 we reset all settings to default. 377 378If set to 1, both IPv4 and IPv6 settings are forced to inherit from 379current ones in init_net. If set to 2, both IPv4 and IPv6 settings are 380forced to reset to their default values. If set to 3, both IPv4 and IPv6 381settings are forced to inherit from current ones in the netns where this 382new netns has been created. 383 384Default : 0 (for compatibility reasons) 385 386txrehash 387-------- 388 389Controls default hash rethink behaviour on socket when SO_TXREHASH option is set 390to SOCK_TXREHASH_DEFAULT (i. e. not overridden by setsockopt). 391 392If set to 1 (default), hash rethink is performed on listening socket. 393If set to 0, hash rethink is not performed. 394 395gro_normal_batch 396---------------- 397 398Maximum number of the segments to batch up on output of GRO. When a packet 399exits GRO, either as a coalesced superframe or as an original packet which 400GRO has decided not to coalesce, it is placed on a per-NAPI list. This 401list is then passed to the stack when the number of segments reaches the 402gro_normal_batch limit. 403 404high_order_alloc_disable 405------------------------ 406 407By default the allocator for page frags tries to use high order pages (order-3 408on x86). While the default behavior gives good results in most cases, some users 409might have hit a contention in page allocations/freeing. This was especially 410true on older kernels (< 5.14) when high-order pages were not stored on per-cpu 411lists. This allows to opt-in for order-0 allocation instead but is now mostly of 412historical importance. 413 414Default: 0 415 4162. /proc/sys/net/unix - Parameters for Unix domain sockets 417---------------------------------------------------------- 418 419There is only one file in this directory. 420unix_dgram_qlen limits the max number of datagrams queued in Unix domain 421socket's buffer. It will not take effect unless PF_UNIX flag is specified. 422 423 4243. /proc/sys/net/ipv4 - IPV4 settings 425------------------------------------- 426Please see: Documentation/networking/ip-sysctl.rst and 427Documentation/admin-guide/sysctl/net.rst for descriptions of these entries. 428 429 4304. Appletalk 431------------ 432 433The /proc/sys/net/appletalk directory holds the Appletalk configuration data 434when Appletalk is loaded. The configurable parameters are: 435 436aarp-expiry-time 437---------------- 438 439The amount of time we keep an ARP entry before expiring it. Used to age out 440old hosts. 441 442aarp-resolve-time 443----------------- 444 445The amount of time we will spend trying to resolve an Appletalk address. 446 447aarp-retransmit-limit 448--------------------- 449 450The number of times we will retransmit a query before giving up. 451 452aarp-tick-time 453-------------- 454 455Controls the rate at which expires are checked. 456 457The directory /proc/net/appletalk holds the list of active Appletalk sockets 458on a machine. 459 460The fields indicate the DDP type, the local address (in network:node format) 461the remote address, the size of the transmit pending queue, the size of the 462received queue (bytes waiting for applications to read) the state and the uid 463owning the socket. 464 465/proc/net/atalk_iface lists all the interfaces configured for appletalk.It 466shows the name of the interface, its Appletalk address, the network range on 467that address (or network number for phase 1 networks), and the status of the 468interface. 469 470/proc/net/atalk_route lists each known network route. It lists the target 471(network) that the route leads to, the router (may be directly connected), the 472route flags, and the device the route is using. 473 4745. TIPC 475------- 476 477tipc_rmem 478--------- 479 480The TIPC protocol now has a tunable for the receive memory, similar to the 481tcp_rmem - i.e. a vector of 3 INTEGERs: (min, default, max) 482 483:: 484 485 # cat /proc/sys/net/tipc/tipc_rmem 486 4252725 34021800 68043600 487 # 488 489The max value is set to CONN_OVERLOAD_LIMIT, and the default and min values 490are scaled (shifted) versions of that same value. Note that the min value 491is not at this point in time used in any meaningful way, but the triplet is 492preserved in order to be consistent with things like tcp_rmem. 493 494named_timeout 495------------- 496 497TIPC name table updates are distributed asynchronously in a cluster, without 498any form of transaction handling. This means that different race scenarios are 499possible. One such is that a name withdrawal sent out by one node and received 500by another node may arrive after a second, overlapping name publication already 501has been accepted from a third node, although the conflicting updates 502originally may have been issued in the correct sequential order. 503If named_timeout is nonzero, failed topology updates will be placed on a defer 504queue until another event arrives that clears the error, or until the timeout 505expires. Value is in milliseconds. 506