accept(2)

Purpose

accept() is used with connection-based socket types (SOCK_STREAM, SOCK_SEQPACKET): it extracts the first pending connection request from the queue of the listening socket sockfd, creates a new connected socket, and returns a new fd for it. The new socket is not in the listening state and sockfd is unaffected. accept4() is identical when flags is 0 and adds the SOCK_* flags behavior.

Synopsis

#include <sys/socket.h>
int accept(int sockfd, struct sockaddr *restrict addr,
           socklen_t *restrict addr_len);
#define _GNU_SOURCE
#include <sys/socket.h>
int accept4(int sockfd, struct sockaddr *restrict addr,
            socklen_t *restrict addr_len, int flags);

Arguments and behavior

addr/addrlen are a value-result pair: initialize addrlen to the size of the buffer, and on return it holds the actual peer-address size; a too-small buffer truncates the address and addrlen returns the size that would have been needed. addr may be NULL (then addrlen must also be NULL). With no pending connections, accept() blocks unless the socket is nonblocking (O_NONBLOCK), in which case it fails with EAGAIN/EWOULDBLOCK. Linux does NOT propagate file status flags such as O_NONBLOCK/O_ASYNC from the listening socket to the new socket – this differs from canonical BSD, so set flags explicitly (accept.2 NOTES). accept4 flags: SOCK_NONBLOCK, SOCK_CLOEXEC (same meaning as on socket, accepted since 2.6.27 there and at accept4 introduction here).

Return value and errors

New fd on success; -1 with errno, addrlen unchanged on error. Linux passes already-pending network errors on the new socket through accept() – unlike other BSD implementations – so treat ENETDOWN, EPROTO, ENOPROTOOPT, EHOSTDOWN, ENONET, EHOSTUNREACH, EOPNOTSUPP, ENETUNREACH like EAGAIN and retry (accept.2 Error handling). Other errors: EAGAIN/EWOULDBLOCK, EBADF, ECONNABORTED, EFAULT, EINTR, EINVAL (not listening; negative addrlen; accept4 bad flags), EMFILE, ENFILE, ENOBUFS/ENOMEM, ENOTSOCK, EPERM (firewall), EPROTO.

Flags and constants

Security and notes

Numbering: i386 has no direct accept slot (hub shows “-”); 32-bit code reached accept through socketcall SYS_ACCEPT 5 until Linux 4.3 added direct calls (socketcall.2 NOTES; ledger 1-g); accept4 has i386 364. Readiness APIs (select/poll/epoll) only signal “maybe”: the connection can disappear before accept(), so keep O_NONBLOCK on the listener to avoid blocking (accept.2 NOTES). socklen_t replaced int*/size_t* through the POSIX.1g drafts (accept.2). Old slots: x64 214/215 are the never-enabled epoll_ctl_old/epoll_wait_old (see epoll_wait).

References

• Cached man page: source-cache/man-pages-sparse/man2/accept.2 (accept4.2 .so).
• Numbers: hub overview.md; source-cache/kernel/syscall_{64,32,generic,arm}.tbl.
• glibc symbol: nm -D libc.so.6 (glibc 2.41, worklog 3-f).
• Worklog Task 1-g (i386 socketcall routing, old slots).