xref: /aosp_15_r20/external/ltp/doc/old/C-Test-Network-API.asciidoc (revision 49cdfc7efb34551c7342be41a7384b9c40d7cab7)

LTP C Test Network API
======================

NOTE: See also
      https://github.com/linux-test-project/ltp/wiki/Test-Writing-Guidelines[Test Writing Guidelines],
      https://github.com/linux-test-project/ltp/wiki/C-Test-Case-Tutorial[C Test Case Tutorial],
      https://github.com/linux-test-project/ltp/wiki/C-Test-API[C Test API],
      https://github.com/linux-test-project/ltp/wiki/Shell-Test-API[Shell Test API].

LTP library includes helper functions for configuring sockets and setting up
network devices.

1 Configuring sockets
---------------------

1.1 Safe syscall variants
~~~~~~~~~~~~~~~~~~~~~~~~~

+#include "tst_safe_net.h"+

Most common standard syscalls and libc functions for configuring sockets have a
"safe" variant in LTP library which will call +tst_brk()+ if the underlying
system function fails. See
https://github.com/linux-test-project/ltp/wiki/C-Test-API[C Test API]. The
safe function names are in uppercase with the +SAFE_+ prefix (e.g. the safe
variant of +socket()+ is called +SAFE_SOCKET()+). For most safe functions, the
parameters and return type are identical to the standard system function:

- +SAFE_SOCKET()+
- +SAFE_SOCKETPAIR()+
- +SAFE_GETSOCKOPT()+
- +SAFE_SETSOCKOPT()+
- +SAFE_BIND()+
- +SAFE_LISTEN()+
- +SAFE_ACCEPT()+
- +SAFE_CONNECT()+
- +SAFE_GETSOCKNAME()+
- +SAFE_GETHOSTNAME()+
- +SAFE_GETADDRINFO()+

A few safe functions have extra parameters for quick return value validation.
The ellipsis (+...+) represents the standard parameters of the underlying system
function:

* +SAFE_SEND(char strict, ...)+
* +SAFE_SENDTO(char strict, ...)+
** If +strict+ is non-zero, the return value must be equal to the data length
   argument. Otherwise the test will fail and exit.

* +SAFE_SENDMSG(size_t msg_len, ...)+
* +SAFE_RECV(size_t msg_len, ...)+
* +SAFE_RECVMSG(size_t msg_len, ...)+
** If +msg_len+ is non-zero, the return value must be equal to the +msg_len+
   argument. Otherwise the test will fail and exit.

There are also some custom functions for simpler configuration and queries:

- +int SAFE_SETSOCKOPT_INT(int sockfd, int level, int optname, int value)+ –
  Simple setsockopt() variant for passing integers by value.

- +int TST_GETSOCKPORT(int sockfd)+ – Get port number (in host byte order) of a
  bound socket.

- +unsigned short TST_GET_UNUSED_PORT(int family, int type)+ – Get a random
  port number (in network byte order) which is currently closed for the given
  socket family and type. Note that another application may open the port while
  the test is still running. The test user is responsible for setting up test
  environment without such interference.

1.2 Address conversion functions
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

+#include "tst_net.h"+

LTP library also provides helper functions for quick initialization of socket
address structures:

- +void tst_get_in_addr(const char *ip_str, struct in_addr *ip)+ – Convert
  human-readable IPv4 address string +ip_str+ to binary representation in
  network byte order. The converted value will be stored in the second argument.

- +void tst_get_in6_addr(const char *ip_str, struct in6_addr *ip6)+ – Convert
  human-readable IPv6 address string +ip_str+ to binary representation in
  network byte order. The converted value will be stored in the second argument.

- +socklen_t tst_get_connect_address(int sock, struct sockaddr_storage *addr)+ –
  Find the address which can be used to send data to bound socket +sock+ from
  another socket. The address will be stored in the second argument. This
  function automatically converts wildcard bind address to localhost. Returns
  size of the address in bytes.

- +void tst_init_sockaddr_inet(struct sockaddr_in *sa, const char *ip_str,
  uint16_t port)+ – Initialize socket address structure +sa+ using
  human-readable IPv4 address +ip_str+ and port number +port+ in host byte
  order.

- +void tst_init_sockaddr_inet_bin(struct sockaddr_in *sa, uint32_t ip_val,
  uint16_t port)+ – Initialize socket address structure +sa+ using binary IPv4
  address +ip_val+ and port number +port+, both in host byte order.

- +void tst_init_sockaddr_inet6(struct sockaddr_in6 *sa, const char *ip_str,
  uint16_t port)+ – Initialize socket address structure +sa+ using
  human-readable IPv6 address +ip_str+ and port number +port+ in host byte
  order.

- +void tst_init_sockaddr_inet6_bin(struct sockaddr_in6 *sa, const struct
  in6_addr *ip_val, uint16_t port)+ – Initialize socket address structure +sa+
  using binary IPv6 address +ip_val+ and port number +port+, both in host byte
  order.

Example Usage
+++++++++++++
[source,c]
-------------------------------------------------------------------------------

#include <sys/socket.h>
#include <netinet/in.h>

#include "tst_test.h"
#include "tst_safe_net.h"
#include "tst_net.h"

static int sockfd = -1;

static void setup(void)
{
	struct sockaddr_in addr;

	tst_init_sockaddr_inet_bin(&addr, INADDR_ANY, 0);
	sockfd = SAFE_SOCKET(AF_INET, SOCK_STREAM, 0);
	SAFE_SETSOCKOPT_INT(sockfd, SOL_SOCKET, SO_SNDBUF, 4096);
	SAFE_BIND(sockfd, (struct sockaddr *)&addr, sizeof(addr));
	SAFE_LISTEN(sockfd, 8);
}

-------------------------------------------------------------------------------

2 Configuring network devices
-----------------------------

+#include "tst_netdevice.h"+

When opening a localhost socket isn't enough and the test needs special device
or routing configuration, the netdevice library can create the required network
setup without calling external programs. Internally, the netdevice functions
use a netlink socket to communicate with the kernel.

All of these functions will call +tst_brk()+ on failure, unless stated
otherwise. Error values described below are returned only during test cleanup
stage.

2.1 Network device management
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- +int NETDEV_INDEX_BY_NAME(const char *ifname)+ – Returns the device index for
  the given device name, or -1 on error.

- +int NETDEV_SET_STATE(const char *ifname, int up)+ – Enable or disable a
  network device +ifname+. Returns 0 on success, -1 on error.

- +int CREATE_VETH_PAIR(const char *ifname1, const char *ifname2)+ – Creates a
  connected pair of virtual network devices with given device names. Returns 1
  on success, 0 on error. Add +"CONFIG_VETH"+ to +test.needs_kconfigs+ if your
  test calls this function.

- +int NETDEV_ADD_DEVICE(const char *ifname, const char *devtype)+ - Creates
  a new network device named +ifname+ of specified device type. Returns 1 on
  success, 0 on error.

- +int NETDEV_REMOVE_DEVICE(const char *ifname)+ – Removes network device
  +ifname+. Returns 1 on success, 0 on error.

2.2 Network address management
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- +int NETDEV_ADD_ADDRESS(const char \*ifname, unsigned int family, const void
  *address, unsigned int prefix, size_t addrlen, unsigned int flags)+ – Adds
  new address to network device +ifname+. This is a low-level function which
  allows setting any type of address. You must specify the protocol +family+,
  address length in bytes (+addrlen+) and network prefix length (+prefix+). The
  +address+ itself must be in binary representation in network byte order. You
  can also pass rtnetlink flags from the +IFA_F_*+ group. Returns 1 on success,
  0 on error.

- +int NETDEV_ADD_ADDRESS_INET(const char *ifname, in_addr_t address, unsigned
  int prefix, unsigned int flags)+ – Adds new IPv4 address to network device
  +ifname+. Parameters have the same meaning as in +NETDEV_ADD_ADDRESS()+.
  Returns 1 on success, 0 on error.

- +int NETDEV_REMOVE_ADDRESS(const char *ifname, unsigned int family, const
  void *address, size_t addrlen)+ – Removes the specified address from network
  device +ifname+. Parameters have the same meaning as in
  +NETDEV_ADD_ADDRESS()+. Returns 1 on success, 0 on error.

- +int NETDEV_REMOVE_ADDRESS_INET(const char *ifname, in_addr_t address)+ –
  Removes specified IPv4 +address+ (in network byte order) from network device
  +ifname+. Returns 1 on success, 0 on error.

2.3 Network namespace device assignment
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

WARNING: Moving a network device to another namespace will erase previous
         configuration. Move the device to the correct namespace first, then
         configure it.

- +int NETDEV_CHANGE_NS_FD(const char *ifname, int nsfd)+ – Moves network
  device +ifname+ to network namespace designated by open file descriptor
  +nsfd+. Returns 1 on success, 0 on error.

- +int NETDEV_CHANGE_NS_PID(const char *ifname, pid_t nspid)+ – Moves network
  device +ifname+ to the network namespace currently used by process +nspid+.
  Returns 1 on success, 0 on error.

2.4 Routing table management
~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- +int NETDEV_ADD_ROUTE(const char *ifname, unsigned int family, const void
  *srcaddr, unsigned int srcprefix, size_t srclen, const void *dstaddr,
  unsigned int dstprefix, size_t dstlen, const void *gateway, size_t
  gatewaylen)+ – Adds new route to the main routing table. This is a low-level
  function which allows creating routes for any protocol. You must specify the
  protocol +family+ and either network device name +ifname+ or +gateway+
  address. Both packet source address +srcaddr+ and destination address
  +dstaddr+ are optional. You must also specify the corresponding length
  and prefix argument for any address which is not +NULL+. All addresses must
  be in binary representation in network byte order. Returns 1 on success,
  0 on error.

- +int NETDEV_ADD_ROUTE_INET(const char *ifname, in_addr_t srcaddr, unsigned
  int srcprefix, in_addr_t dstaddr, unsigned int dstprefix, in_addr_t
  gateway)+ – Adds new IPv4 route to the main routing table. Parameters have
  the same meaning as in +NETDEV_ADD_ROUTE()+. If you do not want to set
  explicit +gateway+ address, set it to 0. If the routing rule should ignore
  the source or destination address, set the corresponding prefix argument
  to 0. Returns 1 on success, 0 on error.

- +int NETDEV_REMOVE_ROUTE(const char *ifname, unsigned int family, const void
  *srcaddr, unsigned int srcprefix, size_t srclen, const void *dstaddr,
  unsigned int dstprefix, size_t dstlen, const void *gateway, size_t
  gatewaylen)+ – Removes a route from the main routing table. Parameters have
  the same meaning as in +NETDEV_ADD_ROUTE()+. Returns 1 on success, 0 on
  error.

- +int NETDEV_REMOVE_ROUTE_INET(const char *ifname, in_addr_t srcaddr,
  unsigned int srcprefix, in_addr_t dstaddr, unsigned int dstprefix, in_addr_t
  gateway)+ – Removes IPv4 route from the main routing table. Parameters have
  the same meaning as in +NETDEV_ADD_ROUTE_INET()+. Returns 1 on success,
  0 on error.

Example Usage
+++++++++++++
[source,c]
-------------------------------------------------------------------------------
#include <arpa/inet.h>
#include <linux/if_addr.h>
#include "tst_test.h"
#include "tst_netdevice.h"

...

static void setup(void)
{
	CREATE_VETH_PAIR("ltp_veth1", "ltp_veth2");
	NETDEV_ADD_ADDRESS_INET("ltp_veth2", htonl(DSTADDR), NETMASK,
		IFA_F_NOPREFIXROUTE);
	NETDEV_SET_STATE("ltp_veth2", 1);
	NETDEV_ADD_ROUTE_INET("ltp_veth2", 0, 0, htonl(SRCNET), NETMASK, 0);

	NETDEV_ADD_ADDRESS_INET("ltp_veth1", htonl(SRCADDR), NETMASK,
		IFA_F_NOPREFIXROUTE);
	NETDEV_SET_STATE("ltp_veth1", 1);
	NETDEV_ADD_ROUTE_INET("ltp_veth1", 0, 0, htonl(DSTNET), NETMASK, 0);
	...
}
-------------------------------------------------------------------------------

3 Netlink API
-------------

+#include "tst_netlink.h"+

The netlink library provides helper functions for constructing and sending
arbitrary messages and parsing kernel responses.

All of the functions below will call +tst_brk()+ on failure, unless stated
otherwise. Error values described below are returned only during test cleanup
stage.

3.1 Data structures
~~~~~~~~~~~~~~~~~~~

[source,c]
-------------------------------------------------------------------------------
struct tst_netlink_context;

struct tst_netlink_attr_list {
	unsigned short type;
	const void *data;
	ssize_t len;
	const struct tst_netlink_attr_list *sublist;
};

struct tst_netlink_message {
	struct nlmsghdr *header;
	struct nlmsgerr *err;
	void *payload;
	size_t payload_size;
};
-------------------------------------------------------------------------------

+struct tst_netlink_context+ is an opaque netlink socket with buffer for
constructing and sending arbitrary messages using the functions described
below. Create a new context using +NETLINK_CREATE_CONTEXT()+, then free it
using +NETLINK_DESTROY_CONTEXT()+ when you're done with it.

+struct tst_netlink_attr_list+ is a helper structure for defining complex
rtnetlink message attribute payloads, including nested attribute lists. Every
list and sublist defined using this structure is terminated by item with
negative +len+.

- +type+ is the attribute type that will be stored in +struct nlattr.nla_type+
  or +struct rtattr.rta_type+.

- +data+ contains arbitrary attribute payload.

- +len+ contains length of the +data+ attribute in bytes. If +data+ is +NULL+,
  set +len+ to 0. The last item in a list or sublist must have negative length.

- +sublist+ contains a nested attribute list which will be appended after
  +data+ as part of the attribute payload. +struct nlattr.nla_len+ or
  +struct rtattr.rta_len+ will be calculated automatically with proper
  alignment, do _not_ add the sublist size to the +len+ field. If you do not
  want to add nested attributes, set +sublist+ to +NULL+.

+struct tst_netlink_message+ is a structure holding partially parsed netlink
messages received from the kernel. +NETLINK_RECV()+ returns an array of these
structures with the last item having +NULL+ in the +header+ field. Call
+NETLINK_FREE_MESSAGE()+ to free a message list returned by +NETLINK_RECV()+.

- +header+ is the netlink header structure of the message. +NULL+ in the header
  field terminates a list of messages.

- +err+ points to the payload of +NLMSG_ERROR+ messages. It is set to +NULL+
  for all other message types.

- +payload+ is a pointer to message data.

- +payload_size+ is the length of +payload+ data in bytes.

3.2 Sending and receiving messages
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- +struct tst_netlink_context *NETLINK_CREATE_CONTEXT(int protocol)+ – Creates
  a new netlink communication context with given netlink protocol for use
  with the functions described below. Returns +NULL+ on error.

- +void NETLINK_FREE_MESSAGE(struct tst_netlink_message *msg)+ – Frees
  an array of messages returned by +NETLINK_RECV()+.

- +void NETLINK_DESTROY_CONTEXT(struct tst_netlink_context *ctx)+ – Closes a
  communication context created by +NETLINK_CREATE_CONTEXT()+.

- +int NETLINK_SEND(struct tst_netlink_context *ctx)+ – Sends all messages
  waiting in +ctx+ buffer to the kernel. If there are multiple messages
  to send, a new +NLMSG_DONE+ message will be added automatically. Returns
  the number of bytes sent on success. Return 0 or negative value on error.

- +int NETLINK_SEND_VALIDATE(struct tst_netlink_context *ctx)+ – Sends all
  messages just like +NETLINK_SEND()+, then receives the response from
  the kernel and validates results of requests sent with the +NLM_F_ACK+ flag.
  This function calls +tst_brk()+ as usual if communication fails but it will
  return error status without terminating the test if one of the received
  messages contains error code. See +NETLINK_CHECK_ACKS()+ below for
  explanation of the return value.

- +int NETLINK_WAIT(struct tst_netlink_context *ctx)+ – Waits until data becomes
  available to read from the netlink socket (timeout: 1 second). Returns 1
  if there is data to read, 0 on timeout or -1 on error.

- +struct tst_netlink_message *NETLINK_RECV(struct tst_netlink_context *ctx)+ –
  Receives netlink messages from the kernel. The messages are received
  in non-blocking mode so calling +NETLINK_WAIT()+ first is recommended.
  Returns an array of partially parsed messages terminated by an item with
  +NULL+ in the +header+ field. On error or when there are no messages
  to receive, returns +NULL+. Call +NETLINK_FREE_MESSAGE()+ to free
  the returned data.

- +int NETLINK_CHECK_ACKS(struct tst_netlink_context *ctx,
  struct tst_netlink_message *response)+ – Validate results of requests sent
  with the +NLM_F_ACK+ flag. Do not call +NETLINK_ADD_MESSAGE()+ between
  +NETLINK_SEND()+ and +NETLINK_CHECK_ACKS()+ because it will reset the state
  of +ctx+ and prevent result validation. Returns 1 if all messages sent
  with the +NLM_F_ACK+ flag have a corresponding message in +response+ and
  the error code is 0. If any of the expected response messages is missing,
  this function will call +tst_brk()+ (or return 0 during test cleanup phase).
  If any of the response messages has non-zero error code, this function will
  return 0 and store the first non-zero error code in global variable
  +tst_netlink_errno+ (sign-flipped just like regular libc +errno+).

3.3 Creating messages
~~~~~~~~~~~~~~~~~~~~~

- +int NETLINK_ADD_MESSAGE(struct tst_netlink_context *ctx, const struct
  nlmsghdr *header, const void *payload, size_t payload_size)+ – Adds new
  netlink message to +ctx+ buffer. You need to provide message +header+ and
  optional +payload+. +payload_size+ is the size of +payload+ data in bytes.
  If you don't want to add any payload data, set +payload+ to +NULL+ and
  +payload_size+ to 0. This function will automatically fill the +nlmsg_len+,
  +nlmsg_seq+ and +nlmsg_pid+ fields of the new message header. You don't need
  to set those. It'll also automatically add +NLM_F_MULTI+ flag when needed.
  Returns 1 on success, 0 on error. Note that the first call of
  +NETLINK_ADD_MESSAGE()+ after +NETLINK_SEND()+ will reset the state of +ctx+
  and +NETLINK_CHECK_ACKS()+ will not work correctly until the next
  +NETLINK_SEND()+.

- +int NETLINK_ADD_ATTR(struct tst_netlink_context *ctx, unsigned short type,
  const void *data, unsigned short len)+ – Adds new +struct nlattr+ attribute
  to the last message in +ctx+ buffer. See +NETLINK_ADD_MESSAGE()+. You need
  to provide attribute +type+ which will be stored in +struct nlattr.nla_type+,
  optional payload +data+ and payload size +len+ in bytes. If you don't want
  to add any payload, set +data+ to +NULL+ and +len+ to 0. Returns 1 on
  success, 0 on error.

- +int NETLINK_ADD_ATTR_STRING(struct tst_netlink_context *ctx, unsigned short
  type, const char *data)+ – Adds new +struct nlattr+ string attribute to the
  last message in +ctx+ buffer. Parameters and return value are the same as
  for +NETLINK_ADD_ATTR()+, except the payload length is calculated using
  +strlen()+.

- +int NETLINK_ADD_ATTR_LIST(struct tst_netlink_context *ctx, const struct
  tst_netlink_attr_list *list)+ – Adds a list of +struct nlattr+ attributes
  to the last message in +ctx+ buffer. See description of
  +struct tst_netlink_attr_list+ and +NETLINK_ADD_MESSAGE()+ above. Returns
  the number of added attributes on success (nested attributes are not
  counted), -1 on error.

- +int RTNL_ADD_ATTR(struct tst_netlink_context *ctx, unsigned short type,
  const void *data, unsigned short len)+ – Adds new +struct rtattr+ attribute
  to the last message in +ctx+ buffer. See +NETLINK_ADD_MESSAGE()+. You need
  to provide attribute +type+ which will be stored in +struct rtattr.rta_type+,
  optional payload +data+ and payload size +len+ in bytes. If you don't want
  to add any payload, set +data+ to +NULL+ and +len+ to 0. Returns 1 on
  success, 0 on error.

- +int RTNL_ADD_ATTR_STRING(struct tst_netlink_context *ctx, unsigned short
  type, const char *data)+ – Adds new +struct rtattr+ string attribute to the
  last message in +ctx+ buffer. Parameters and return value are the same as
  for +RTNL_ADD_ATTR()+, except the payload length is calculated using
  +strlen()+.

- +int RTNL_ADD_ATTR_LIST(struct tst_netlink_context *ctx, const struct
  tst_netlink_attr_list *list)+ – Adds a list of +struct rtattr+ attributes
  to the last message in +ctx+ buffer. See description of
  +struct tst_netlink_attr_list+ and +NETLINK_ADD_MESSAGE()+ above. Returns
  the number of added attributes on success (nested attributes are not
  counted), -1 on error.

Example Usage
+++++++++++++
[source,c]
-------------------------------------------------------------------------------
#include <asm/types.h>
#include <linux/netlink.h>
#include <linux/rtnetlink.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>

#include "tst_test.h"
#include "tst_netlink.h"
#include "tst_netdevice.h"

...

void setup(void)
{
	struct tst_netlink_context *ctx;
	int index, ret;
	in_addr_t addr;

	struct nlmsghdr header = {
		.nlmsg_type = RTM_NEWADDR,
		.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE |
			NLM_F_EXCL
	};

	struct ifaddrmsg info = {
		.ifa_family = AF_INET,
		.ifa_prefixlen = 24
	};

	index = NETDEV_INDEX_BY_NAME("ltp_veth1");
	info.ifa_index = index;

	ctx = NETLINK_CREATE_CONTEXT(NETLINK_ROUTE);
	NETLINK_ADD_MESSAGE(ctx, &header, &info, sizeof(info));
	addr = inet_addr("192.168.123.45");
	RTNL_ADD_ATTR(ctx, IFA_LOCAL, &addr, sizeof(addr));
	ret = NETLINK_SEND_VALIDATE(ctx);
	NETLINK_DESTROY_CONTEXT(ctx);

	if (!ret) {
		tst_brk(TBROK, "Failed to set ltp_veth1 address");
	}
}
-------------------------------------------------------------------------------