net.rst (16487B)
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 core General parameter appletalk Appletalk protocol 38 unix Unix domain sockets netrom NET/ROM 39 802 E802 protocol ax25 AX25 40 ethernet Ethernet protocol rose X.25 PLP layer 41 ipv4 IP version 4 x25 X.25 protocol 42 bridge Bridging decnet DEC net 43 ipv6 IP version 6 tipc TIPC 44 ========= =================== = ========== ================== 45 461. /proc/sys/net/core - Network core options 47============================================ 48 49bpf_jit_enable 50-------------- 51 52This enables the BPF Just in Time (JIT) compiler. BPF is a flexible 53and efficient infrastructure allowing to execute bytecode at various 54hook points. It is used in a number of Linux kernel subsystems such 55as networking (e.g. XDP, tc), tracing (e.g. kprobes, uprobes, tracepoints) 56and security (e.g. seccomp). LLVM has a BPF back end that can compile 57restricted C into a sequence of BPF instructions. After program load 58through bpf(2) and passing a verifier in the kernel, a JIT will then 59translate these BPF proglets into native CPU instructions. There are 60two flavors of JITs, the newer eBPF JIT currently supported on: 61 62 - x86_64 63 - x86_32 64 - arm64 65 - arm32 66 - ppc64 67 - ppc32 68 - sparc64 69 - mips64 70 - s390x 71 - riscv64 72 - riscv32 73 74And the older cBPF JIT supported on the following archs: 75 76 - mips 77 - sparc 78 79eBPF JITs are a superset of cBPF JITs, meaning the kernel will 80migrate cBPF instructions into eBPF instructions and then JIT 81compile them transparently. Older cBPF JITs can only translate 82tcpdump filters, seccomp rules, etc, but not mentioned eBPF 83programs loaded through bpf(2). 84 85Values: 86 87 - 0 - disable the JIT (default value) 88 - 1 - enable the JIT 89 - 2 - enable the JIT and ask the compiler to emit traces on kernel log. 90 91bpf_jit_harden 92-------------- 93 94This enables hardening for the BPF JIT compiler. Supported are eBPF 95JIT backends. Enabling hardening trades off performance, but can 96mitigate JIT spraying. 97 98Values: 99 100 - 0 - disable JIT hardening (default value) 101 - 1 - enable JIT hardening for unprivileged users only 102 - 2 - enable JIT hardening for all users 103 104bpf_jit_kallsyms 105---------------- 106 107When BPF JIT compiler is enabled, then compiled images are unknown 108addresses to the kernel, meaning they neither show up in traces nor 109in /proc/kallsyms. This enables export of these addresses, which can 110be used for debugging/tracing. If bpf_jit_harden is enabled, this 111feature is disabled. 112 113Values : 114 115 - 0 - disable JIT kallsyms export (default value) 116 - 1 - enable JIT kallsyms export for privileged users only 117 118bpf_jit_limit 119------------- 120 121This enforces a global limit for memory allocations to the BPF JIT 122compiler in order to reject unprivileged JIT requests once it has 123been surpassed. bpf_jit_limit contains the value of the global limit 124in bytes. 125 126dev_weight 127---------- 128 129The maximum number of packets that kernel can handle on a NAPI interrupt, 130it's a Per-CPU variable. For drivers that support LRO or GRO_HW, a hardware 131aggregated packet is counted as one packet in this context. 132 133Default: 64 134 135dev_weight_rx_bias 136------------------ 137 138RPS (e.g. RFS, aRFS) processing is competing with the registered NAPI poll function 139of the driver for the per softirq cycle netdev_budget. This parameter influences 140the proportion of the configured netdev_budget that is spent on RPS based packet 141processing during RX softirq cycles. It is further meant for making current 142dev_weight adaptable for asymmetric CPU needs on RX/TX side of the network stack. 143(see dev_weight_tx_bias) It is effective on a per CPU basis. Determination is based 144on dev_weight and is calculated multiplicative (dev_weight * dev_weight_rx_bias). 145 146Default: 1 147 148dev_weight_tx_bias 149------------------ 150 151Scales the maximum number of packets that can be processed during a TX softirq cycle. 152Effective on a per CPU basis. Allows scaling of current dev_weight for asymmetric 153net stack processing needs. Be careful to avoid making TX softirq processing a CPU hog. 154 155Calculation is based on dev_weight (dev_weight * dev_weight_tx_bias). 156 157Default: 1 158 159default_qdisc 160------------- 161 162The default queuing discipline to use for network devices. This allows 163overriding the default of pfifo_fast with an alternative. Since the default 164queuing discipline is created without additional parameters so is best suited 165to queuing disciplines that work well without configuration like stochastic 166fair queue (sfq), CoDel (codel) or fair queue CoDel (fq_codel). Don't use 167queuing disciplines like Hierarchical Token Bucket or Deficit Round Robin 168which require setting up classes and bandwidths. Note that physical multiqueue 169interfaces still use mq as root qdisc, which in turn uses this default for its 170leaves. Virtual devices (like e.g. lo or veth) ignore this setting and instead 171default to noqueue. 172 173Default: pfifo_fast 174 175busy_read 176--------- 177 178Low latency busy poll timeout for socket reads. (needs CONFIG_NET_RX_BUSY_POLL) 179Approximate time in us to busy loop waiting for packets on the device queue. 180This sets the default value of the SO_BUSY_POLL socket option. 181Can be set or overridden per socket by setting socket option SO_BUSY_POLL, 182which is the preferred method of enabling. If you need to enable the feature 183globally via sysctl, a value of 50 is recommended. 184 185Will increase power usage. 186 187Default: 0 (off) 188 189busy_poll 190---------------- 191Low latency busy poll timeout for poll and select. (needs CONFIG_NET_RX_BUSY_POLL) 192Approximate time in us to busy loop waiting for events. 193Recommended value depends on the number of sockets you poll on. 194For several sockets 50, for several hundreds 100. 195For more than that you probably want to use epoll. 196Note that only sockets with SO_BUSY_POLL set will be busy polled, 197so you want to either selectively set SO_BUSY_POLL on those sockets or set 198sysctl.net.busy_read globally. 199 200Will increase power usage. 201 202Default: 0 (off) 203 204rmem_default 205------------ 206 207The default setting of the socket receive buffer in bytes. 208 209rmem_max 210-------- 211 212The maximum receive socket buffer size in bytes. 213 214tstamp_allow_data 215----------------- 216Allow processes to receive tx timestamps looped together with the original 217packet contents. If disabled, transmit timestamp requests from unprivileged 218processes are dropped unless socket option SOF_TIMESTAMPING_OPT_TSONLY is set. 219 220Default: 1 (on) 221 222 223wmem_default 224------------ 225 226The default setting (in bytes) of the socket send buffer. 227 228wmem_max 229-------- 230 231The maximum send socket buffer size in bytes. 232 233message_burst and message_cost 234------------------------------ 235 236These parameters are used to limit the warning messages written to the kernel 237log from the networking code. They enforce a rate limit to make a 238denial-of-service attack impossible. A higher message_cost factor, results in 239fewer messages that will be written. Message_burst controls when messages will 240be dropped. The default settings limit warning messages to one every five 241seconds. 242 243warnings 244-------- 245 246This sysctl is now unused. 247 248This was used to control console messages from the networking stack that 249occur because of problems on the network like duplicate address or bad 250checksums. 251 252These messages are now emitted at KERN_DEBUG and can generally be enabled 253and controlled by the dynamic_debug facility. 254 255netdev_budget 256------------- 257 258Maximum number of packets taken from all interfaces in one polling cycle (NAPI 259poll). In one polling cycle interfaces which are registered to polling are 260probed in a round-robin manner. Also, a polling cycle may not exceed 261netdev_budget_usecs microseconds, even if netdev_budget has not been 262exhausted. 263 264netdev_budget_usecs 265--------------------- 266 267Maximum number of microseconds in one NAPI polling cycle. Polling 268will exit when either netdev_budget_usecs have elapsed during the 269poll cycle or the number of packets processed reaches netdev_budget. 270 271netdev_max_backlog 272------------------ 273 274Maximum number of packets, queued on the INPUT side, when the interface 275receives packets faster than kernel can process them. 276 277netdev_rss_key 278-------------- 279 280RSS (Receive Side Scaling) enabled drivers use a 40 bytes host key that is 281randomly generated. 282Some user space might need to gather its content even if drivers do not 283provide ethtool -x support yet. 284 285:: 286 287 myhost:~# cat /proc/sys/net/core/netdev_rss_key 288 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) 289 290File contains nul bytes if no driver ever called netdev_rss_key_fill() function. 291 292Note: 293 /proc/sys/net/core/netdev_rss_key contains 52 bytes of key, 294 but most drivers only use 40 bytes of it. 295 296:: 297 298 myhost:~# ethtool -x eth0 299 RX flow hash indirection table for eth0 with 8 RX ring(s): 300 0: 0 1 2 3 4 5 6 7 301 RSS hash key: 302 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:f1:42:89 303 304netdev_tstamp_prequeue 305---------------------- 306 307If set to 0, RX packet timestamps can be sampled after RPS processing, when 308the target CPU processes packets. It might give some delay on timestamps, but 309permit to distribute the load on several cpus. 310 311If set to 1 (default), timestamps are sampled as soon as possible, before 312queueing. 313 314netdev_unregister_timeout_secs 315------------------------------ 316 317Unregister network device timeout in seconds. 318This option controls the timeout (in seconds) used to issue a warning while 319waiting for a network device refcount to drop to 0 during device 320unregistration. A lower value may be useful during bisection to detect 321a leaked reference faster. A larger value may be useful to prevent false 322warnings on slow/loaded systems. 323Default value is 10, minimum 1, maximum 3600. 324 325skb_defer_max 326------------- 327 328Max size (in skbs) of the per-cpu list of skbs being freed 329by the cpu which allocated them. Used by TCP stack so far. 330 331Default: 64 332 333optmem_max 334---------- 335 336Maximum ancillary buffer size allowed per socket. Ancillary data is a sequence 337of struct cmsghdr structures with appended data. 338 339fb_tunnels_only_for_init_net 340---------------------------- 341 342Controls if fallback tunnels (like tunl0, gre0, gretap0, erspan0, 343sit0, ip6tnl0, ip6gre0) are automatically created. There are 3 possibilities 344(a) value = 0; respective fallback tunnels are created when module is 345loaded in every net namespaces (backward compatible behavior). 346(b) value = 1; [kcmd value: initns] respective fallback tunnels are 347created only in init net namespace and every other net namespace will 348not have them. 349(c) value = 2; [kcmd value: none] fallback tunnels are not created 350when a module is loaded in any of the net namespace. Setting value to 351"2" is pointless after boot if these modules are built-in, so there is 352a kernel command-line option that can change this default. Please refer to 353Documentation/admin-guide/kernel-parameters.txt for additional details. 354 355Not creating fallback tunnels gives control to userspace to create 356whatever is needed only and avoid creating devices which are redundant. 357 358Default : 0 (for compatibility reasons) 359 360devconf_inherit_init_net 361------------------------ 362 363Controls if a new network namespace should inherit all current 364settings under /proc/sys/net/{ipv4,ipv6}/conf/{all,default}/. By 365default, we keep the current behavior: for IPv4 we inherit all current 366settings from init_net and for IPv6 we reset all settings to default. 367 368If set to 1, both IPv4 and IPv6 settings are forced to inherit from 369current ones in init_net. If set to 2, both IPv4 and IPv6 settings are 370forced to reset to their default values. If set to 3, both IPv4 and IPv6 371settings are forced to inherit from current ones in the netns where this 372new netns has been created. 373 374Default : 0 (for compatibility reasons) 375 376txrehash 377-------- 378 379Controls default hash rethink behaviour on listening socket when SO_TXREHASH 380option is set to SOCK_TXREHASH_DEFAULT (i. e. not overridden by setsockopt). 381 382If set to 1 (default), hash rethink is performed on listening socket. 383If set to 0, hash rethink is not performed. 384 385gro_normal_batch 386---------------- 387 388Maximum number of the segments to batch up on output of GRO. When a packet 389exits GRO, either as a coalesced superframe or as an original packet which 390GRO has decided not to coalesce, it is placed on a per-NAPI list. This 391list is then passed to the stack when the number of segments reaches the 392gro_normal_batch limit. 393 3942. /proc/sys/net/unix - Parameters for Unix domain sockets 395---------------------------------------------------------- 396 397There is only one file in this directory. 398unix_dgram_qlen limits the max number of datagrams queued in Unix domain 399socket's buffer. It will not take effect unless PF_UNIX flag is specified. 400 401 4023. /proc/sys/net/ipv4 - IPV4 settings 403------------------------------------- 404Please see: Documentation/networking/ip-sysctl.rst and 405Documentation/admin-guide/sysctl/net.rst for descriptions of these entries. 406 407 4084. Appletalk 409------------ 410 411The /proc/sys/net/appletalk directory holds the Appletalk configuration data 412when Appletalk is loaded. The configurable parameters are: 413 414aarp-expiry-time 415---------------- 416 417The amount of time we keep an ARP entry before expiring it. Used to age out 418old hosts. 419 420aarp-resolve-time 421----------------- 422 423The amount of time we will spend trying to resolve an Appletalk address. 424 425aarp-retransmit-limit 426--------------------- 427 428The number of times we will retransmit a query before giving up. 429 430aarp-tick-time 431-------------- 432 433Controls the rate at which expires are checked. 434 435The directory /proc/net/appletalk holds the list of active Appletalk sockets 436on a machine. 437 438The fields indicate the DDP type, the local address (in network:node format) 439the remote address, the size of the transmit pending queue, the size of the 440received queue (bytes waiting for applications to read) the state and the uid 441owning the socket. 442 443/proc/net/atalk_iface lists all the interfaces configured for appletalk.It 444shows the name of the interface, its Appletalk address, the network range on 445that address (or network number for phase 1 networks), and the status of the 446interface. 447 448/proc/net/atalk_route lists each known network route. It lists the target 449(network) that the route leads to, the router (may be directly connected), the 450route flags, and the device the route is using. 451 4525. TIPC 453------- 454 455tipc_rmem 456--------- 457 458The TIPC protocol now has a tunable for the receive memory, similar to the 459tcp_rmem - i.e. a vector of 3 INTEGERs: (min, default, max) 460 461:: 462 463 # cat /proc/sys/net/tipc/tipc_rmem 464 4252725 34021800 68043600 465 # 466 467The max value is set to CONN_OVERLOAD_LIMIT, and the default and min values 468are scaled (shifted) versions of that same value. Note that the min value 469is not at this point in time used in any meaningful way, but the triplet is 470preserved in order to be consistent with things like tcp_rmem. 471 472named_timeout 473------------- 474 475TIPC name table updates are distributed asynchronously in a cluster, without 476any form of transaction handling. This means that different race scenarios are 477possible. One such is that a name withdrawal sent out by one node and received 478by another node may arrive after a second, overlapping name publication already 479has been accepted from a third node, although the conflicting updates 480originally may have been issued in the correct sequential order. 481If named_timeout is nonzero, failed topology updates will be placed on a defer 482queue until another event arrives that clears the error, or until the timeout 483expires. Value is in milliseconds.