unix(4) on freebsd

NAME

unix – UNIX-domain protocol family

SYNOPSIS

#include <sys/types.h>

#include <sys/un.h>

DESCRIPTION

The UNIX-domain protocol family is a collection of protocols that provides local (on-machine) interprocess communication through the normal socket(2) mechanisms. The UNIX-domain family supports the SOCK_STREAM, SOCK_SEQPACKET, and SOCK_DGRAM socket types and uses file system pathnames for addressing.

ADDRESSING

UNIX-domain addresses are variable-length file system pathnames of at most 104 characters. The include file ⟨sys/un.h⟩ defines this address:

struct sockaddr_un {
	u_char	sun_len;
	u_char	sun_family;
	char	sun_path[104];
};

Binding a name to a UNIX-domain socket with bind(2) causes a socket file to be created in the file system. This file is not removed when the socket is closed — unlink(2) must be used to remove the file.

Prior to binding a socket, fchmod(2) can be used to set the permissions of the socket file. This avoids the race that would otherwise occur between creation of the file and a subsequent call to chmod(2). Once the socket is bound to a file name, the permissions of the file can not be changed this way.

Unlike with most other protocols, binding is not a prerequisite for listen(2). An unbound socket may listen, in which case it can only be reached by descriptor, as described in Naming a peer below. A socket may equally be bound after it listens, which lets a listener be published only once it is ready to accept, rather than leaving a window in which the socket file exists but connections to it are refused.

The length of UNIX-domain address, required by bind(2) and connect(2), can be calculated by the macro SUN_LEN() defined in ⟨sys/un.h⟩. The sun_path field must be terminated by a NUL character to be used with SUN_LEN(), but the terminating NUL is not part of the address.

The UNIX-domain protocol family does not support broadcast addressing or any form of “wildcard” matching on incoming messages. All addresses are absolute- or relative-pathnames of other UNIX-domain sockets. Normal file system access-control mechanisms are also applied when referencing pathnames; e.g., the destination of a connect(2) or sendto(2) must be writable.

Naming a peer

A peer can be named either by a pathname or, using connectat(2), directly by a descriptor. Across these two forms a peer can be named five ways, all of which reach the same socket:

Target	By descriptor	By pathname
The peer socket	the socket's fd	(none)
A bound socket's file	an O_PATH descriptor	an ordinary pathname
A /dev/fd/N node	an O_PATH descriptor	/dev/fd/N

A pathname, accepted by both connect(2) and connectat(2), names either a bound socket's file created by bind(2), or the /dev/fd/N node that fdescfs(5) provides for descriptor N. In the latter case the /dev/fd/N node must name the socket descriptor itself; the named descriptor is resolved a single level and not chased further. With a standard fdescfs(5) mount this means a node naming a traditional socket file, or another /dev/fd/N node, does not resolve to a peer; a nodup mount, however, dereferences such a descriptor before the lookup completes.

connectat(2) also supports both of the above cases. The path is allowed to be empty — that is, a namelen equal to offsetof(struct sockaddr_un, sun_path) — in which case the given file descriptor will be used directly, like with most *at system calls. (Note: /dev/fd/N nodes are not recommended to be used for the connectat(2), since the descriptor can just be named directly, and this is both more efficient and less userland code, but since connect(2) is implemented in terms of connectat(2) this is provided for free.)

Besides supporting cases that connect(2) also supports, connectat(2) additionally supports directly naming the peer by the fd descriptor (and an empty path). The descriptor may be the peer socket itself. In all cases, the descriptor passed must carry the CAP_CONNECTAT capability right, and in the /dev/fd/N form the target descriptor must also carry that capability. A descriptor limited to CAP_CONNECTAT alone is thus a pure “connect-to-me” token, usable as a connection target but not listened on, accepted from, or read.

For the descriptor forms, possession of the descriptor is the authorization, as with descriptor passing over SCM_RIGHTS; the file system access-control and mac(4) checks that apply to the pathname forms are not repeated. Because a descriptor alone suffices, a peer need never be bound at all: a datagram socket may connect to an unbound peer, and a listening socket may itself be unbound, neither of which a pathname could name.

CONTROL MESSAGES

The UNIX-domain sockets support the communication of UNIX file descriptors and process credentials through the use of the msg_control field in the msg argument to sendmsg(2) and recvmsg(2). The items to be passed are described using a struct cmsghdr that is defined in the include file ⟨sys/socket.h⟩.

To send file descriptors, the type of the message is SCM_RIGHTS, and the data portion of the messages is an array of integers representing the file descriptors to be passed. The number of descriptors being passed is defined by the length field of the message; the length field is the sum of the size of the header plus the size of the array of file descriptors.

The received descriptor is a duplicate of the sender's descriptor, as if it were created via dup(fd) or fcntl(fd, F_DUPFD_CLOEXEC, 0) depending on whether MSG_CMSG_CLOEXEC is passed in the recvmsg(2) call.

Descriptors that are awaiting delivery, or that are purposely not received, are automatically closed by the system when the destination socket is closed. The receiving socket can reject SCM_RIGHTS at the sendmsg(2) call with the socket option SO_PASSRIGHTS, which can be set with setsockopt(2) and tested with getsockopt(2).

Credentials of the sending process can be transmitted explicitly using a control message of type SCM_CREDS with a data portion of type struct cmsgcred, defined in ⟨sys/socket.h⟩ as follows:

struct cmsgcred {
  pid_t	cmcred_pid;		/* PID of sending process */
  uid_t	cmcred_uid;		/* real UID of sending process */
  uid_t	cmcred_euid;		/* effective UID of sending process */
  gid_t	cmcred_gid;		/* real GID of sending process */
  short	cmcred_ngroups;		/* number of groups */
  gid_t	cmcred_groups[CMGROUP_MAX];	/* groups */
};

The sender should pass a zeroed buffer which will be filled in by the system.

The group list is truncated to at most CMGROUP_MAX GIDs.

The process ID cmcred_pid should not be looked up (such as via the KERN_PROC_PID sysctl) for making security decisions. The sending process could have exited and its process ID already been reused for a new process.

SOCKET OPTIONS

UNIX domain sockets support a number of socket options for the options level SOL_LOCAL, which can be set with setsockopt(2) and tested with getsockopt(2):

LOCAL_CREDS

This option may be enabled on SOCK_DGRAM, SOCK_SEQPACKET, or a SOCK_STREAM socket. This option provides a mechanism for the receiver to receive the credentials of the process calling write(2), send(2), sendto(2) or sendmsg(2) as a recvmsg(2) control message. The msg_control field in the msghdr structure points to a buffer that contains a cmsghdr structure followed by a variable length sockcred structure, defined in ⟨sys/socket.h⟩ as follows:

struct sockcred {
  uid_t	sc_uid;		/* real user id */
  uid_t	sc_euid;	/* effective user id */
  gid_t	sc_gid;		/* real group id */
  gid_t	sc_egid;	/* effective group id */
  int	sc_ngroups;	/* number of supplemental groups */
  gid_t	sc_groups[1];	/* variable length */
};

The current implementation truncates the group list to at most CMGROUP_MAX groups.

The SOCKCREDSIZE() macro computes the size of the sockcred structure for a specified number of groups. The cmsghdr fields have the following values:

cmsg_len = CMSG_LEN(SOCKCREDSIZE(ngroups))
cmsg_level = SOL_SOCKET
cmsg_type = SCM_CREDS

On SOCK_STREAM and SOCK_SEQPACKET sockets credentials are passed only on the first read from a socket, then the system clears the option on the socket.

This option and the above explicit struct cmsgcred both use the same value SCM_CREDS but incompatible control messages. If this option is enabled and the sender attached a SCM_CREDS control message with a struct cmsgcred, it will be discarded and a struct sockcred will be included.

Many setuid programs will write(2) data at least partially controlled by the invoker, such as error messages. Therefore, a message accompanied by a particular sc_euid value should not be trusted as being from that user.

LOCAL_CREDS_PERSISTENT

This option is similar to LOCAL_CREDS, except that socket credentials are passed on every read from a SOCK_STREAM or SOCK_SEQPACKET socket, instead of just the first read. Additionally, the msg_control field in the msghdr structure points to a buffer that contains a cmsghdr structure followed by a variable length sockcred2 structure, defined in ⟨sys/socket.h⟩ as follows:

struct sockcred2 {
  int	sc_version;	/* version of this structure */
  pid_t	sc_pid;		/* PID of sending process */
  uid_t	sc_uid;		/* real user id */
  uid_t	sc_euid;	/* effective user id */
  gid_t	sc_gid;		/* real group id */
  gid_t	sc_egid;	/* effective group id */
  int	sc_ngroups;	/* number of supplemental groups */
  gid_t	sc_groups[1];	/* variable length */
};

The current version is zero.

The cmsghdr fields have the following values:

cmsg_len = CMSG_LEN(SOCKCRED2SIZE(ngroups))
cmsg_level = SOL_SOCKET
cmsg_type = SCM_CREDS2

The LOCAL_CREDS and LOCAL_CREDS_PERSISTENT options are mutually exclusive.

LOCAL_PEERCRED

Requested via getsockopt(2) on a SOCK_STREAM or SOCK_SEQPACKET socket returns credentials of the remote side. These will arrive in the form of a filled in xucred structure, defined in ⟨sys/ucred.h⟩ as follows:

struct xucred {
  u_int	cr_version;		/* structure layout version */
  uid_t	cr_uid;			/* effective user id */
  short	cr_ngroups;		/* number of groups */
  gid_t	cr_groups[XU_NGROUPS];	/* groups */
  pid_t	cr_pid;			/* process id of the sending process */
};

The cr_version fields should be checked against XUCRED_VERSION define.

The credentials presented to the server (the listen(2) caller) are those of the client when it called connect(2); the credentials presented to the client (the connect(2) caller) are those of the server when it called listen(2). This mechanism is reliable; there is no way for either party to influence the credentials presented to its peer except by calling the appropriate system call (e.g., connect(2) or listen(2)) under different effective credentials.

To reliably obtain peer credentials on a SOCK_DGRAM socket refer to the LOCAL_CREDS socket option.

BUFFERING

Due to the local nature of the UNIX-domain sockets, they do not implement send buffers. The send(2) and write(2) families of system calls attempt to write data to the receive buffer of the destination socket.

The default buffer sizes for SOCK_STREAM and SOCK_SEQPACKET UNIX-domain sockets can be configured with net.local.stream and net.local.seqpacket branches of sysctl(3) MIB respectively. Note that setting the send buffer size (sendspace) affects only the maximum write size.

The UNIX-domain sockets of type SOCK_DGRAM are unreliable and always non-blocking for write operations. The default receive buffer can be configured with net.local.dgram.recvspace. The maximum allowed datagram size is limited by net.local.dgram.maxdgram. A SOCK_DGRAM socket that has been bound with bind(2) can have multiple peers connected at the same time. The modern FreeBSD implementation will allocate net.local.dgram.recvspace sized private buffers in the receive buffer of the bound socket for every connected socket, preventing a situation when a single writer can exhaust all of buffer space. Messages coming from unconnected sends using sendto(2) land on the shared buffer of the receiving socket, which has the same size limit. A side effect of the implementation is that it doesn't guarantee that writes from different senders will arrive at the receiver in the same chronological order they were sent. The order is preserved for writes coming through a particular connection.

unix(4)