xref: /linux/Documentation/userspace-api/netlink/intro.rst (revision 6e7fd890f1d6ac83805409e9c346240de2705584)
1.. SPDX-License-Identifier: BSD-3-Clause
2
3=======================
4Introduction to Netlink
5=======================
6
7Netlink is often described as an ioctl() replacement.
8It aims to replace fixed-format C structures as supplied
9to ioctl() with a format which allows an easy way to add
10or extended the arguments.
11
12To achieve this Netlink uses a minimal fixed-format metadata header
13followed by multiple attributes in the TLV (type, length, value) format.
14
15Unfortunately the protocol has evolved over the years, in an organic
16and undocumented fashion, making it hard to coherently explain.
17To make the most practical sense this document starts by describing
18netlink as it is used today and dives into more "historical" uses
19in later sections.
20
21Opening a socket
22================
23
24Netlink communication happens over sockets, a socket needs to be
25opened first:
26
27.. code-block:: c
28
29  fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC);
30
31The use of sockets allows for a natural way of exchanging information
32in both directions (to and from the kernel). The operations are still
33performed synchronously when applications send() the request but
34a separate recv() system call is needed to read the reply.
35
36A very simplified flow of a Netlink "call" will therefore look
37something like:
38
39.. code-block:: c
40
41  fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC);
42
43  /* format the request */
44  send(fd, &request, sizeof(request));
45  n = recv(fd, &response, RSP_BUFFER_SIZE);
46  /* interpret the response */
47
48Netlink also provides natural support for "dumping", i.e. communicating
49to user space all objects of a certain type (e.g. dumping all network
50interfaces).
51
52.. code-block:: c
53
54  fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC);
55
56  /* format the dump request */
57  send(fd, &request, sizeof(request));
58  while (1) {
59    n = recv(fd, &buffer, RSP_BUFFER_SIZE);
60    /* one recv() call can read multiple messages, hence the loop below */
61    for (nl_msg in buffer) {
62      if (nl_msg.nlmsg_type == NLMSG_DONE)
63        goto dump_finished;
64      /* process the object */
65    }
66  }
67  dump_finished:
68
69The first two arguments of the socket() call require little explanation -
70it is opening a Netlink socket, with all headers provided by the user
71(hence NETLINK, RAW). The last argument is the protocol within Netlink.
72This field used to identify the subsystem with which the socket will
73communicate.
74
75Classic vs Generic Netlink
76--------------------------
77
78Initial implementation of Netlink depended on a static allocation
79of IDs to subsystems and provided little supporting infrastructure.
80Let us refer to those protocols collectively as **Classic Netlink**.
81The list of them is defined on top of the ``include/uapi/linux/netlink.h``
82file, they include among others - general networking (NETLINK_ROUTE),
83iSCSI (NETLINK_ISCSI), and audit (NETLINK_AUDIT).
84
85**Generic Netlink** (introduced in 2005) allows for dynamic registration of
86subsystems (and subsystem ID allocation), introspection and simplifies
87implementing the kernel side of the interface.
88
89The following section describes how to use Generic Netlink, as the
90number of subsystems using Generic Netlink outnumbers the older
91protocols by an order of magnitude. There are also no plans for adding
92more Classic Netlink protocols to the kernel.
93Basic information on how communicating with core networking parts of
94the Linux kernel (or another of the 20 subsystems using Classic
95Netlink) differs from Generic Netlink is provided later in this document.
96
97Generic Netlink
98===============
99
100In addition to the Netlink fixed metadata header each Netlink protocol
101defines its own fixed metadata header. (Similarly to how network
102headers stack - Ethernet > IP > TCP we have Netlink > Generic N. > Family.)
103
104A Netlink message always starts with struct nlmsghdr, which is followed
105by a protocol-specific header. In case of Generic Netlink the protocol
106header is struct genlmsghdr.
107
108The practical meaning of the fields in case of Generic Netlink is as follows:
109
110.. code-block:: c
111
112  struct nlmsghdr {
113	__u32	nlmsg_len;	/* Length of message including headers */
114	__u16	nlmsg_type;	/* Generic Netlink Family (subsystem) ID */
115	__u16	nlmsg_flags;	/* Flags - request or dump */
116	__u32	nlmsg_seq;	/* Sequence number */
117	__u32	nlmsg_pid;	/* Port ID, set to 0 */
118  };
119  struct genlmsghdr {
120	__u8	cmd;		/* Command, as defined by the Family */
121	__u8	version;	/* Irrelevant, set to 1 */
122	__u16	reserved;	/* Reserved, set to 0 */
123  };
124  /* TLV attributes follow... */
125
126In Classic Netlink :c:member:`nlmsghdr.nlmsg_type` used to identify
127which operation within the subsystem the message was referring to
128(e.g. get information about a netdev). Generic Netlink needs to mux
129multiple subsystems in a single protocol so it uses this field to
130identify the subsystem, and :c:member:`genlmsghdr.cmd` identifies
131the operation instead. (See :ref:`res_fam` for
132information on how to find the Family ID of the subsystem of interest.)
133Note that the first 16 values (0 - 15) of this field are reserved for
134control messages both in Classic Netlink and Generic Netlink.
135See :ref:`nl_msg_type` for more details.
136
137There are 3 usual types of message exchanges on a Netlink socket:
138
139 - performing a single action (``do``);
140 - dumping information (``dump``);
141 - getting asynchronous notifications (``multicast``).
142
143Classic Netlink is very flexible and presumably allows other types
144of exchanges to happen, but in practice those are the three that get
145used.
146
147Asynchronous notifications are sent by the kernel and received by
148the user sockets which subscribed to them. ``do`` and ``dump`` requests
149are initiated by the user. :c:member:`nlmsghdr.nlmsg_flags` should
150be set as follows:
151
152 - for ``do``: ``NLM_F_REQUEST | NLM_F_ACK``
153 - for ``dump``: ``NLM_F_REQUEST | NLM_F_ACK | NLM_F_DUMP``
154
155:c:member:`nlmsghdr.nlmsg_seq` should be a set to a monotonically
156increasing value. The value gets echoed back in responses and doesn't
157matter in practice, but setting it to an increasing value for each
158message sent is considered good hygiene. The purpose of the field is
159matching responses to requests. Asynchronous notifications will have
160:c:member:`nlmsghdr.nlmsg_seq` of ``0``.
161
162:c:member:`nlmsghdr.nlmsg_pid` is the Netlink equivalent of an address.
163This field can be set to ``0`` when talking to the kernel.
164See :ref:`nlmsg_pid` for the (uncommon) uses of the field.
165
166The expected use for :c:member:`genlmsghdr.version` was to allow
167versioning of the APIs provided by the subsystems. No subsystem to
168date made significant use of this field, so setting it to ``1`` seems
169like a safe bet.
170
171.. _nl_msg_type:
172
173Netlink message types
174---------------------
175
176As previously mentioned :c:member:`nlmsghdr.nlmsg_type` carries
177protocol specific values but the first 16 identifiers are reserved
178(first subsystem specific message type should be equal to
179``NLMSG_MIN_TYPE`` which is ``0x10``).
180
181There are only 4 Netlink control messages defined:
182
183 - ``NLMSG_NOOP`` - ignore the message, not used in practice;
184 - ``NLMSG_ERROR`` - carries the return code of an operation;
185 - ``NLMSG_DONE`` - marks the end of a dump;
186 - ``NLMSG_OVERRUN`` - socket buffer has overflown, not used to date.
187
188``NLMSG_ERROR`` and ``NLMSG_DONE`` are of practical importance.
189They carry return codes for operations. Note that unless
190the ``NLM_F_ACK`` flag is set on the request Netlink will not respond
191with ``NLMSG_ERROR`` if there is no error. To avoid having to special-case
192this quirk it is recommended to always set ``NLM_F_ACK``.
193
194The format of ``NLMSG_ERROR`` is described by struct nlmsgerr::
195
196  ----------------------------------------------
197  | struct nlmsghdr - response header          |
198  ----------------------------------------------
199  |    int error                               |
200  ----------------------------------------------
201  | struct nlmsghdr - original request header |
202  ----------------------------------------------
203  | ** optionally (1) payload of the request   |
204  ----------------------------------------------
205  | ** optionally (2) extended ACK             |
206  ----------------------------------------------
207
208There are two instances of struct nlmsghdr here, first of the response
209and second of the request. ``NLMSG_ERROR`` carries the information about
210the request which led to the error. This could be useful when trying
211to match requests to responses or re-parse the request to dump it into
212logs.
213
214The payload of the request is not echoed in messages reporting success
215(``error == 0``) or if ``NETLINK_CAP_ACK`` setsockopt() was set.
216The latter is common
217and perhaps recommended as having to read a copy of every request back
218from the kernel is rather wasteful. The absence of request payload
219is indicated by ``NLM_F_CAPPED`` in :c:member:`nlmsghdr.nlmsg_flags`.
220
221The second optional element of ``NLMSG_ERROR`` are the extended ACK
222attributes. See :ref:`ext_ack` for more details. The presence
223of extended ACK is indicated by ``NLM_F_ACK_TLVS`` in
224:c:member:`nlmsghdr.nlmsg_flags`.
225
226``NLMSG_DONE`` is simpler, the request is never echoed but the extended
227ACK attributes may be present::
228
229  ----------------------------------------------
230  | struct nlmsghdr - response header          |
231  ----------------------------------------------
232  |    int error                               |
233  ----------------------------------------------
234  | ** optionally extended ACK                 |
235  ----------------------------------------------
236
237Note that some implementations may issue custom ``NLMSG_DONE`` messages
238in reply to ``do`` action requests. In that case the payload is
239implementation-specific and may also be absent.
240
241.. _res_fam:
242
243Resolving the Family ID
244-----------------------
245
246This section explains how to find the Family ID of a subsystem.
247It also serves as an example of Generic Netlink communication.
248
249Generic Netlink is itself a subsystem exposed via the Generic Netlink API.
250To avoid a circular dependency Generic Netlink has a statically allocated
251Family ID (``GENL_ID_CTRL`` which is equal to ``NLMSG_MIN_TYPE``).
252The Generic Netlink family implements a command used to find out information
253about other families (``CTRL_CMD_GETFAMILY``).
254
255To get information about the Generic Netlink family named for example
256``"test1"`` we need to send a message on the previously opened Generic Netlink
257socket. The message should target the Generic Netlink Family (1), be a
258``do`` (2) call to ``CTRL_CMD_GETFAMILY`` (3). A ``dump`` version of this
259call would make the kernel respond with information about *all* the families
260it knows about. Last but not least the name of the family in question has
261to be specified (4) as an attribute with the appropriate type::
262
263  struct nlmsghdr:
264    __u32 nlmsg_len:	32
265    __u16 nlmsg_type:	GENL_ID_CTRL               // (1)
266    __u16 nlmsg_flags:	NLM_F_REQUEST | NLM_F_ACK  // (2)
267    __u32 nlmsg_seq:	1
268    __u32 nlmsg_pid:	0
269
270  struct genlmsghdr:
271    __u8 cmd:		CTRL_CMD_GETFAMILY         // (3)
272    __u8 version:	2 /* or 1, doesn't matter */
273    __u16 reserved:	0
274
275  struct nlattr:                                   // (4)
276    __u16 nla_len:	10
277    __u16 nla_type:	CTRL_ATTR_FAMILY_NAME
278    char data: 		test1\0
279
280  (padding:)
281    char data:		\0\0
282
283The length fields in Netlink (:c:member:`nlmsghdr.nlmsg_len`
284and :c:member:`nlattr.nla_len`) always *include* the header.
285Attribute headers in netlink must be aligned to 4 bytes from the start
286of the message, hence the extra ``\0\0`` after ``CTRL_ATTR_FAMILY_NAME``.
287The attribute lengths *exclude* the padding.
288
289If the family is found kernel will reply with two messages, the response
290with all the information about the family::
291
292  /* Message #1 - reply */
293  struct nlmsghdr:
294    __u32 nlmsg_len:	136
295    __u16 nlmsg_type:	GENL_ID_CTRL
296    __u16 nlmsg_flags:	0
297    __u32 nlmsg_seq:	1    /* echoed from our request */
298    __u32 nlmsg_pid:	5831 /* The PID of our user space process */
299
300  struct genlmsghdr:
301    __u8 cmd:		CTRL_CMD_GETFAMILY
302    __u8 version:	2
303    __u16 reserved:	0
304
305  struct nlattr:
306    __u16 nla_len:	10
307    __u16 nla_type:	CTRL_ATTR_FAMILY_NAME
308    char data: 		test1\0
309
310  (padding:)
311    data:		\0\0
312
313  struct nlattr:
314    __u16 nla_len:	6
315    __u16 nla_type:	CTRL_ATTR_FAMILY_ID
316    __u16: 		123  /* The Family ID we are after */
317
318  (padding:)
319    char data:		\0\0
320
321  struct nlattr:
322    __u16 nla_len:	9
323    __u16 nla_type:	CTRL_ATTR_FAMILY_VERSION
324    __u16: 		1
325
326  /* ... etc, more attributes will follow. */
327
328And the error code (success) since ``NLM_F_ACK`` had been set on the request::
329
330  /* Message #2 - the ACK */
331  struct nlmsghdr:
332    __u32 nlmsg_len:	36
333    __u16 nlmsg_type:	NLMSG_ERROR
334    __u16 nlmsg_flags:	NLM_F_CAPPED /* There won't be a payload */
335    __u32 nlmsg_seq:	1    /* echoed from our request */
336    __u32 nlmsg_pid:	5831 /* The PID of our user space process */
337
338  int error:		0
339
340  struct nlmsghdr: /* Copy of the request header as we sent it */
341    __u32 nlmsg_len:	32
342    __u16 nlmsg_type:	GENL_ID_CTRL
343    __u16 nlmsg_flags:	NLM_F_REQUEST | NLM_F_ACK
344    __u32 nlmsg_seq:	1
345    __u32 nlmsg_pid:	0
346
347The order of attributes (struct nlattr) is not guaranteed so the user
348has to walk the attributes and parse them.
349
350Note that Generic Netlink sockets are not associated or bound to a single
351family. A socket can be used to exchange messages with many different
352families, selecting the recipient family on message-by-message basis using
353the :c:member:`nlmsghdr.nlmsg_type` field.
354
355.. _ext_ack:
356
357Extended ACK
358------------
359
360Extended ACK controls reporting of additional error/warning TLVs
361in ``NLMSG_ERROR`` and ``NLMSG_DONE`` messages. To maintain backward
362compatibility this feature has to be explicitly enabled by setting
363the ``NETLINK_EXT_ACK`` setsockopt() to ``1``.
364
365Types of extended ack attributes are defined in enum nlmsgerr_attrs.
366The most commonly used attributes are ``NLMSGERR_ATTR_MSG``,
367``NLMSGERR_ATTR_OFFS`` and ``NLMSGERR_ATTR_MISS_*``.
368
369``NLMSGERR_ATTR_MSG`` carries a message in English describing
370the encountered problem. These messages are far more detailed
371than what can be expressed thru standard UNIX error codes.
372
373``NLMSGERR_ATTR_OFFS`` points to the attribute which caused the problem.
374
375``NLMSGERR_ATTR_MISS_TYPE`` and ``NLMSGERR_ATTR_MISS_NEST``
376inform about a missing attribute.
377
378Extended ACKs can be reported on errors as well as in case of success.
379The latter should be treated as a warning.
380
381Extended ACKs greatly improve the usability of Netlink and should
382always be enabled, appropriately parsed and reported to the user.
383
384Advanced topics
385===============
386
387Dump consistency
388----------------
389
390Some of the data structures kernel uses for storing objects make
391it hard to provide an atomic snapshot of all the objects in a dump
392(without impacting the fast-paths updating them).
393
394Kernel may set the ``NLM_F_DUMP_INTR`` flag on any message in a dump
395(including the ``NLMSG_DONE`` message) if the dump was interrupted and
396may be inconsistent (e.g. missing objects). User space should retry
397the dump if it sees the flag set.
398
399Introspection
400-------------
401
402The basic introspection abilities are enabled by access to the Family
403object as reported in :ref:`res_fam`. User can query information about
404the Generic Netlink family, including which operations are supported
405by the kernel and what attributes the kernel understands.
406Family information includes the highest ID of an attribute kernel can parse,
407a separate command (``CTRL_CMD_GETPOLICY``) provides detailed information
408about supported attributes, including ranges of values the kernel accepts.
409
410Querying family information is useful in cases when user space needs
411to make sure that the kernel has support for a feature before issuing
412a request.
413
414.. _nlmsg_pid:
415
416nlmsg_pid
417---------
418
419:c:member:`nlmsghdr.nlmsg_pid` is the Netlink equivalent of an address.
420It is referred to as Port ID, sometimes Process ID because for historical
421reasons if the application does not select (bind() to) an explicit Port ID
422kernel will automatically assign it the ID equal to its Process ID
423(as reported by the getpid() system call).
424
425Similarly to the bind() semantics of the TCP/IP network protocols the value
426of zero means "assign automatically", hence it is common for applications
427to leave the :c:member:`nlmsghdr.nlmsg_pid` field initialized to ``0``.
428
429The field is still used today in rare cases when kernel needs to send
430a unicast notification. User space application can use bind() to associate
431its socket with a specific PID, it then communicates its PID to the kernel.
432This way the kernel can reach the specific user space process.
433
434This sort of communication is utilized in UMH (User Mode Helper)-like
435scenarios when kernel needs to trigger user space processing or ask user
436space for a policy decision.
437
438Multicast notifications
439-----------------------
440
441One of the strengths of Netlink is the ability to send event notifications
442to user space. This is a unidirectional form of communication (kernel ->
443user) and does not involve any control messages like ``NLMSG_ERROR`` or
444``NLMSG_DONE``.
445
446For example the Generic Netlink family itself defines a set of multicast
447notifications about registered families. When a new family is added the
448sockets subscribed to the notifications will get the following message::
449
450  struct nlmsghdr:
451    __u32 nlmsg_len:	136
452    __u16 nlmsg_type:	GENL_ID_CTRL
453    __u16 nlmsg_flags:	0
454    __u32 nlmsg_seq:	0
455    __u32 nlmsg_pid:	0
456
457  struct genlmsghdr:
458    __u8 cmd:		CTRL_CMD_NEWFAMILY
459    __u8 version:	2
460    __u16 reserved:	0
461
462  struct nlattr:
463    __u16 nla_len:	10
464    __u16 nla_type:	CTRL_ATTR_FAMILY_NAME
465    char data: 		test1\0
466
467  (padding:)
468    data:		\0\0
469
470  struct nlattr:
471    __u16 nla_len:	6
472    __u16 nla_type:	CTRL_ATTR_FAMILY_ID
473    __u16: 		123  /* The Family ID we are after */
474
475  (padding:)
476    char data:		\0\0
477
478  struct nlattr:
479    __u16 nla_len:	9
480    __u16 nla_type:	CTRL_ATTR_FAMILY_VERSION
481    __u16: 		1
482
483  /* ... etc, more attributes will follow. */
484
485The notification contains the same information as the response
486to the ``CTRL_CMD_GETFAMILY`` request.
487
488The Netlink headers of the notification are mostly 0 and irrelevant.
489The :c:member:`nlmsghdr.nlmsg_seq` may be either zero or a monotonically
490increasing notification sequence number maintained by the family.
491
492To receive notifications the user socket must subscribe to the relevant
493notification group. Much like the Family ID, the Group ID for a given
494multicast group is dynamic and can be found inside the Family information.
495The ``CTRL_ATTR_MCAST_GROUPS`` attribute contains nests with names
496(``CTRL_ATTR_MCAST_GRP_NAME``) and IDs (``CTRL_ATTR_MCAST_GRP_ID``) of
497the groups family.
498
499Once the Group ID is known a setsockopt() call adds the socket to the group:
500
501.. code-block:: c
502
503  unsigned int group_id;
504
505  /* .. find the group ID... */
506
507  setsockopt(fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
508             &group_id, sizeof(group_id));
509
510The socket will now receive notifications.
511
512It is recommended to use separate sockets for receiving notifications
513and sending requests to the kernel. The asynchronous nature of notifications
514means that they may get mixed in with the responses making the message
515handling much harder.
516
517Buffer sizing
518-------------
519
520Netlink sockets are datagram sockets rather than stream sockets,
521meaning that each message must be received in its entirety by a single
522recv()/recvmsg() system call. If the buffer provided by the user is too
523short, the message will be truncated and the ``MSG_TRUNC`` flag set
524in struct msghdr (struct msghdr is the second argument
525of the recvmsg() system call, *not* a Netlink header).
526
527Upon truncation the remaining part of the message is discarded.
528
529Netlink expects that the user buffer will be at least 8kB or a page
530size of the CPU architecture, whichever is bigger. Particular Netlink
531families may, however, require a larger buffer. 32kB buffer is recommended
532for most efficient handling of dumps (larger buffer fits more dumped
533objects and therefore fewer recvmsg() calls are needed).
534
535.. _classic_netlink:
536
537Classic Netlink
538===============
539
540The main differences between Classic and Generic Netlink are the dynamic
541allocation of subsystem identifiers and availability of introspection.
542In theory the protocol does not differ significantly, however, in practice
543Classic Netlink experimented with concepts which were abandoned in Generic
544Netlink (really, they usually only found use in a small corner of a single
545subsystem). This section is meant as an explainer of a few of such concepts,
546with the explicit goal of giving the Generic Netlink
547users the confidence to ignore them when reading the uAPI headers.
548
549Most of the concepts and examples here refer to the ``NETLINK_ROUTE`` family,
550which covers much of the configuration of the Linux networking stack.
551Real documentation of that family, deserves a chapter (or a book) of its own.
552
553Families
554--------
555
556Netlink refers to subsystems as families. This is a remnant of using
557sockets and the concept of protocol families, which are part of message
558demultiplexing in ``NETLINK_ROUTE``.
559
560Sadly every layer of encapsulation likes to refer to whatever it's carrying
561as "families" making the term very confusing:
562
563 1. AF_NETLINK is a bona fide socket protocol family
564 2. AF_NETLINK's documentation refers to what comes after its own
565    header (struct nlmsghdr) in a message as a "Family Header"
566 3. Generic Netlink is a family for AF_NETLINK (struct genlmsghdr follows
567    struct nlmsghdr), yet it also calls its users "Families".
568
569Note that the Generic Netlink Family IDs are in a different "ID space"
570and overlap with Classic Netlink protocol numbers (e.g. ``NETLINK_CRYPTO``
571has the Classic Netlink protocol ID of 21 which Generic Netlink will
572happily allocate to one of its families as well).
573
574Strict checking
575---------------
576
577The ``NETLINK_GET_STRICT_CHK`` socket option enables strict input checking
578in ``NETLINK_ROUTE``. It was needed because historically kernel did not
579validate the fields of structures it didn't process. This made it impossible
580to start using those fields later without risking regressions in applications
581which initialized them incorrectly or not at all.
582
583``NETLINK_GET_STRICT_CHK`` declares that the application is initializing
584all fields correctly. It also opts into validating that message does not
585contain trailing data and requests that kernel rejects attributes with
586type higher than largest attribute type known to the kernel.
587
588``NETLINK_GET_STRICT_CHK`` is not used outside of ``NETLINK_ROUTE``.
589
590Unknown attributes
591------------------
592
593Historically Netlink ignored all unknown attributes. The thinking was that
594it would free the application from having to probe what kernel supports.
595The application could make a request to change the state and check which
596parts of the request "stuck".
597
598This is no longer the case for new Generic Netlink families and those opting
599in to strict checking. See enum netlink_validation for validation types
600performed.
601
602Fixed metadata and structures
603-----------------------------
604
605Classic Netlink made liberal use of fixed-format structures within
606the messages. Messages would commonly have a structure with
607a considerable number of fields after struct nlmsghdr. It was also
608common to put structures with multiple members inside attributes,
609without breaking each member into an attribute of its own.
610
611This has caused problems with validation and extensibility and
612therefore using binary structures is actively discouraged for new
613attributes.
614
615Request types
616-------------
617
618``NETLINK_ROUTE`` categorized requests into 4 types ``NEW``, ``DEL``, ``GET``,
619and ``SET``. Each object can handle all or some of those requests
620(objects being netdevs, routes, addresses, qdiscs etc.) Request type
621is defined by the 2 lowest bits of the message type, so commands for
622new objects would always be allocated with a stride of 4.
623
624Each object would also have its own fixed metadata shared by all request
625types (e.g. struct ifinfomsg for netdev requests, struct ifaddrmsg for address
626requests, struct tcmsg for qdisc requests).
627
628Even though other protocols and Generic Netlink commands often use
629the same verbs in their message names (``GET``, ``SET``) the concept
630of request types did not find wider adoption.
631
632Notification echo
633-----------------
634
635``NLM_F_ECHO`` requests for notifications resulting from the request
636to be queued onto the requesting socket. This is useful to discover
637the impact of the request.
638
639Note that this feature is not universally implemented.
640
641Other request-type-specific flags
642---------------------------------
643
644Classic Netlink defined various flags for its ``GET``, ``NEW``
645and ``DEL`` requests in the upper byte of nlmsg_flags in struct nlmsghdr.
646Since request types have not been generalized the request type specific
647flags are rarely used (and considered deprecated for new families).
648
649For ``GET`` - ``NLM_F_ROOT`` and ``NLM_F_MATCH`` are combined into
650``NLM_F_DUMP``, and not used separately. ``NLM_F_ATOMIC`` is never used.
651
652For ``DEL`` - ``NLM_F_NONREC`` is only used by nftables and ``NLM_F_BULK``
653only by FDB some operations.
654
655The flags for ``NEW`` are used most commonly in classic Netlink. Unfortunately,
656the meaning is not crystal clear. The following description is based on the
657best guess of the intention of the authors, and in practice all families
658stray from it in one way or another. ``NLM_F_REPLACE`` asks to replace
659an existing object, if no matching object exists the operation should fail.
660``NLM_F_EXCL`` has the opposite semantics and only succeeds if object already
661existed.
662``NLM_F_CREATE`` asks for the object to be created if it does not
663exist, it can be combined with ``NLM_F_REPLACE`` and ``NLM_F_EXCL``.
664
665A comment in the main Netlink uAPI header states::
666
667   4.4BSD ADD		NLM_F_CREATE|NLM_F_EXCL
668   4.4BSD CHANGE	NLM_F_REPLACE
669
670   True CHANGE		NLM_F_CREATE|NLM_F_REPLACE
671   Append		NLM_F_CREATE
672   Check		NLM_F_EXCL
673
674which seems to indicate that those flags predate request types.
675``NLM_F_REPLACE`` without ``NLM_F_CREATE`` was initially used instead
676of ``SET`` commands.
677``NLM_F_EXCL`` without ``NLM_F_CREATE`` was used to check if object exists
678without creating it, presumably predating ``GET`` commands.
679
680``NLM_F_APPEND`` indicates that if one key can have multiple objects associated
681with it (e.g. multiple next-hop objects for a route) the new object should be
682added to the list rather than replacing the entire list.
683
684uAPI reference
685==============
686
687.. kernel-doc:: include/uapi/linux/netlink.h
688