X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/d7e50217d7adf6e52786a38bcaa4cd698cb9a79e..e2fac8b15b12a7979f72090454d850e612fc5b13:/bsd/kern/uipc_syscalls.c diff --git a/bsd/kern/uipc_syscalls.c b/bsd/kern/uipc_syscalls.c index 7e73f9ff1..7e9cafa35 100644 --- a/bsd/kern/uipc_syscalls.c +++ b/bsd/kern/uipc_syscalls.c @@ -1,16 +1,19 @@ /* - * Copyright (c) 2000-2001 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2007 Apple Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ - * - * Copyright (c) 1999-2003 Apple Computer, Inc. All Rights Reserved. + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * * This file contains Original Code and/or Modifications of Original Code * as defined in and that are subject to the Apple Public Source License * Version 2.0 (the 'License'). You may not use this file except in - * compliance with the License. Please obtain a copy of the License at - * http://www.opensource.apple.com/apsl/ and read it before using this - * file. + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. + * + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. * * The Original Code and all software distributed under the License are * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER @@ -20,14 +23,14 @@ * Please see the License for the specific language governing rights and * limitations under the License. * - * @APPLE_LICENSE_HEADER_END@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* * Copyright (c) 1982, 1986, 1989, 1990, 1993 * The Regents of the University of California. All rights reserved. * * sendfile(2) and related extensions: - * Copyright (c) 1998, David Greenman. All rights reserved. + * Copyright (c) 1998, David Greenman. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions @@ -59,250 +62,365 @@ * * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94 */ - - +/* + * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce + * support for mandatory and extensible security protections. This notice + * is included in support of clause 2.2 (b) of the Apple Public License, + * Version 2.0. + */ #include #include #include -#include -#include -#include +#include +#include +#include #include #include +#include +#include #include +#include #include #include -#if KTRACE -#include -#endif #include +#include +#include -#include - -#if KDEBUG - -#define DBG_LAYER_IN_BEG NETDBG_CODE(DBG_NETSOCK, 0) -#define DBG_LAYER_IN_END NETDBG_CODE(DBG_NETSOCK, 2) -#define DBG_LAYER_OUT_BEG NETDBG_CODE(DBG_NETSOCK, 1) -#define DBG_LAYER_OUT_END NETDBG_CODE(DBG_NETSOCK, 3) -#define DBG_FNC_SENDMSG NETDBG_CODE(DBG_NETSOCK, (1 << 8) | 1) -#define DBG_FNC_SENDTO NETDBG_CODE(DBG_NETSOCK, (2 << 8) | 1) -#define DBG_FNC_SENDIT NETDBG_CODE(DBG_NETSOCK, (3 << 8) | 1) -#define DBG_FNC_RECVFROM NETDBG_CODE(DBG_NETSOCK, (5 << 8)) -#define DBG_FNC_RECVMSG NETDBG_CODE(DBG_NETSOCK, (6 << 8)) -#define DBG_FNC_RECVIT NETDBG_CODE(DBG_NETSOCK, (7 << 8)) - -#endif - -struct getsockname_args { - int fdes; - caddr_t asa; - int *alen; -}; - -struct getsockopt_args { - int s; - int level; - int name; - caddr_t val; - int *avalsize; -} ; - -struct accept_args { - int s; - caddr_t name; - int *anamelen; -}; - -struct getpeername_args { - int fdes; - caddr_t asa; - int *alen; -}; - - -/* ARGSUSED */ +#include +#include +#include +#include +#include +#include + +#if CONFIG_MACF_SOCKET_SUBSET +#include +#endif /* MAC_SOCKET_SUBSET */ + +#define f_flag f_fglob->fg_flag +#define f_type f_fglob->fg_type +#define f_msgcount f_fglob->fg_msgcount +#define f_cred f_fglob->fg_cred +#define f_ops f_fglob->fg_ops +#define f_offset f_fglob->fg_offset +#define f_data f_fglob->fg_data + + +#define DBG_LAYER_IN_BEG NETDBG_CODE(DBG_NETSOCK, 0) +#define DBG_LAYER_IN_END NETDBG_CODE(DBG_NETSOCK, 2) +#define DBG_LAYER_OUT_BEG NETDBG_CODE(DBG_NETSOCK, 1) +#define DBG_LAYER_OUT_END NETDBG_CODE(DBG_NETSOCK, 3) +#define DBG_FNC_SENDMSG NETDBG_CODE(DBG_NETSOCK, (1 << 8) | 1) +#define DBG_FNC_SENDTO NETDBG_CODE(DBG_NETSOCK, (2 << 8) | 1) +#define DBG_FNC_SENDIT NETDBG_CODE(DBG_NETSOCK, (3 << 8) | 1) +#define DBG_FNC_RECVFROM NETDBG_CODE(DBG_NETSOCK, (5 << 8)) +#define DBG_FNC_RECVMSG NETDBG_CODE(DBG_NETSOCK, (6 << 8)) +#define DBG_FNC_RECVIT NETDBG_CODE(DBG_NETSOCK, (7 << 8)) +#define DBG_FNC_SENDFILE NETDBG_CODE(DBG_NETSOCK, (10 << 8)) +#define DBG_FNC_SENDFILE_WAIT NETDBG_CODE(DBG_NETSOCK, ((10 << 8) | 1)) +#define DBG_FNC_SENDFILE_READ NETDBG_CODE(DBG_NETSOCK, ((10 << 8) | 2)) +#define DBG_FNC_SENDFILE_SEND NETDBG_CODE(DBG_NETSOCK, ((10 << 8) | 3)) + + +#define HACK_FOR_4056224 1 +#if HACK_FOR_4056224 +static pid_t last_pid_4056224 = 0; +#endif /* HACK_FOR_4056224 */ + +/* TODO: should be in header file */ +int falloc_locked(proc_t, struct fileproc **, int *, vfs_context_t, int); + +static int sendit(struct proc *, int, struct user_msghdr *, uio_t, int, + register_t *); +static int recvit(struct proc *, int, struct user_msghdr *, uio_t, user_addr_t, + register_t *); +static int getsockaddr(struct socket *, struct sockaddr **, user_addr_t, + size_t, boolean_t); +static int getsockaddr_s(struct socket *, struct sockaddr_storage *, + user_addr_t, size_t, boolean_t); #if SENDFILE -static void sf_buf_init(void *arg); -SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL) -static struct sf_buf *sf_buf_alloc(void); -static void sf_buf_ref(caddr_t addr, u_int size); -static void sf_buf_free(caddr_t addr, u_int size); - -static SLIST_HEAD(, sf_buf) sf_freelist; -static vm_offset_t sf_base; -static struct sf_buf *sf_bufs; -static int sf_buf_alloc_want; -#endif - -static int sendit __P((struct proc *p, int s, struct msghdr *mp, int flags, register_t *retval)); -static int recvit __P((struct proc *p, int s, struct msghdr *mp, - caddr_t namelenp, register_t *retval)); - -static int accept1 __P((struct proc *p, struct accept_args *uap, register_t *retval, int compat)); -static int getsockname1 __P((struct proc *p, struct getsockname_args *uap, - register_t *retval, int compat)); -static int getpeername1 __P((struct proc *p, struct getpeername_args *uap, - register_t *retval, int compat)); +static void alloc_sendpkt(int, size_t, unsigned int *, struct mbuf **, + boolean_t); +#endif /* SENDFILE */ /* * System call interface to the socket abstraction. */ -#if COMPAT_43 || defined(COMPAT_SUNOS) -#define COMPAT_OLDSOCK -#endif extern struct fileops socketops; -struct socket_args { - int domain; - int type; - int protocol; -}; +/* + * Returns: 0 Success + * EACCES Mandatory Access Control failure + * falloc:ENFILE + * falloc:EMFILE + * falloc:ENOMEM + * socreate:EAFNOSUPPORT + * socreate:EPROTOTYPE + * socreate:EPROTONOSUPPORT + * socreate:ENOBUFS + * socreate:ENOMEM + * socreate:EISCONN + * socreate:??? [other protocol families, IPSEC] + */ int -socket(p, uap, retval) - struct proc *p; - register struct socket_args *uap; - register_t *retval; +socket(struct proc *p, struct socket_args *uap, register_t *retval) { - struct filedesc *fdp = p->p_fd; struct socket *so; - struct file *fp; + struct fileproc *fp; int fd, error; - thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); - error = falloc(p, &fp, &fd); - thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); + AUDIT_ARG(socket, uap->domain, uap->type, uap->protocol); +#if CONFIG_MACF_SOCKET_SUBSET + if ((error = mac_socket_check_create(kauth_cred_get(), uap->domain, + uap->type, uap->protocol)) != 0) + return (error); +#endif /* MAC_SOCKET_SUBSET */ - if (error) + error = falloc(p, &fp, &fd, vfs_context_current()); + if (error) { return (error); + } fp->f_flag = FREAD|FWRITE; fp->f_type = DTYPE_SOCKET; fp->f_ops = &socketops; - if (error = socreate(uap->domain, &so, uap->type, - uap->protocol)) { - thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); - fdrelse(p, fd); - ffree(fp); - thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); + + error = socreate(uap->domain, &so, uap->type, uap->protocol); + if (error) { + fp_free(p, fd, fp); } else { + thread_t thread; + struct uthread *ut; + + thread = current_thread(); + ut = get_bsdthread_info(thread); + + /* if this is a backgrounded thread then throttle all new sockets */ + if ( (ut->uu_flag & UT_BACKGROUND) != 0 ) { + so->so_traffic_mgt_flags |= TRAFFIC_MGT_SO_BACKGROUND; + so->so_background_thread = thread; + } fp->f_data = (caddr_t)so; - *fdflags(p, fd) &= ~UF_RESERVED; + + proc_fdlock(p); + procfdtbl_releasefd(p, fd, NULL); + + fp_drop(p, fd, fp, 1); + proc_fdunlock(p); + *retval = fd; } return (error); } -struct bind_args { - int s; - caddr_t name; - int namelen; -}; - +/* + * Returns: 0 Success + * EDESTADDRREQ Destination address required + * EBADF Bad file descriptor + * EACCES Mandatory Access Control failure + * file_socket:ENOTSOCK + * file_socket:EBADF + * getsockaddr:ENAMETOOLONG Filename too long + * getsockaddr:EINVAL Invalid argument + * getsockaddr:ENOMEM Not enough space + * getsockaddr:EFAULT Bad address + * sobind:??? + */ /* ARGSUSED */ int -bind(p, uap, retval) - struct proc *p; - register struct bind_args *uap; - register_t *retval; +bind(__unused proc_t p, struct bind_args *uap, __unused register_t *retval) { - struct file *fp; - struct sockaddr *sa; + struct sockaddr_storage ss; + struct sockaddr *sa = NULL; + struct socket *so; + boolean_t want_free = TRUE; int error; - error = getsock(p->p_fd, uap->s, &fp); - if (error) - return (error); - error = getsockaddr(&sa, uap->name, uap->namelen); - if (error) + AUDIT_ARG(fd, uap->s); + error = file_socket(uap->s, &so); + if (error != 0) return (error); - error = sobind((struct socket *)fp->f_data, sa); - FREE(sa, M_SONAME); + if (so == NULL) { + error = EBADF; + goto out; + } + if (uap->name == USER_ADDR_NULL) { + error = EDESTADDRREQ; + goto out; + } + if (uap->namelen > sizeof (ss)) { + error = getsockaddr(so, &sa, uap->name, uap->namelen, TRUE); + } else { + error = getsockaddr_s(so, &ss, uap->name, uap->namelen, TRUE); + if (error == 0) { + sa = (struct sockaddr *)&ss; + want_free = FALSE; + } + } + if (error != 0) + goto out; + AUDIT_ARG(sockaddr, vfs_context_cwd(vfs_context_current()), sa); +#if CONFIG_MACF_SOCKET_SUBSET + if ((error = mac_socket_check_bind(kauth_cred_get(), so, sa)) == 0) + error = sobind(so, sa); +#else + error = sobind(so, sa); +#endif /* MAC_SOCKET_SUBSET */ + if (want_free) + FREE(sa, M_SONAME); +out: + file_drop(uap->s); return (error); } -struct listen_args { - int s; - int backlog; -}; - - - +/* + * Returns: 0 Success + * EBADF + * EACCES Mandatory Access Control failure + * file_socket:ENOTSOCK + * file_socket:EBADF + * solisten:EINVAL + * solisten:EOPNOTSUPP + * solisten:??? + */ int -listen(p, uap, retval) - struct proc *p; - register struct listen_args *uap; - register_t *retval; +listen(__unused struct proc *p, struct listen_args *uap, + __unused register_t *retval) { - struct file *fp; int error; + struct socket *so; - error = getsock(p->p_fd, uap->s, &fp); + AUDIT_ARG(fd, uap->s); + error = file_socket(uap->s, &so); if (error) return (error); - return (solisten((struct socket *)fp->f_data, uap->backlog)); -} - -#ifndef COMPAT_OLDSOCK -#define accept1 accept -#endif - + if (so != NULL) +#if CONFIG_MACF_SOCKET_SUBSET + { + error = mac_socket_check_listen(kauth_cred_get(), so); + if (error == 0) + error = solisten(so, uap->backlog); + } +#else + error = solisten(so, uap->backlog); +#endif /* MAC_SOCKET_SUBSET */ + else + error = EBADF; + file_drop(uap->s); + return (error); +} +/* + * Returns: fp_getfsock:EBADF Bad file descriptor + * fp_getfsock:EOPNOTSUPP ... + * xlate => :ENOTSOCK Socket operation on non-socket + * :EFAULT Bad address on copyin/copyout + * :EBADF Bad file descriptor + * :EOPNOTSUPP Operation not supported on socket + * :EINVAL Invalid argument + * :EWOULDBLOCK Operation would block + * :ECONNABORTED Connection aborted + * :EINTR Interrupted function + * :EACCES Mandatory Access Control failure + * falloc_locked:ENFILE Too many files open in system + * falloc_locked::EMFILE Too many open files + * falloc_locked::ENOMEM Not enough space + * 0 Success + */ int -accept1(p, uap, retval, compat) - struct proc *p; - register struct accept_args *uap; - register_t *retval; - int compat; +accept_nocancel(struct proc *p, struct accept_nocancel_args *uap, + register_t *retval) { - struct file *fp; - struct sockaddr *sa; - u_int namelen; - int error, s; - struct socket *head, *so; - int fd; + struct fileproc *fp; + struct sockaddr *sa = NULL; + socklen_t namelen; + int error; + struct socket *head, *so = NULL; + lck_mtx_t *mutex_held; + int fd = uap->s; + int newfd; short fflag; /* type must match fp->f_flag */ - int tmpfd; + int dosocklock = 0; + + *retval = -1; + + AUDIT_ARG(fd, uap->s); if (uap->name) { - error = copyin((caddr_t)uap->anamelen, (caddr_t)&namelen, - sizeof (namelen)); - if(error) + error = copyin(uap->anamelen, (caddr_t)&namelen, + sizeof (socklen_t)); + if (error) return (error); } - error = getsock(p->p_fd, uap->s, &fp); - if (error) + error = fp_getfsock(p, fd, &fp, &head); + if (error) { + if (error == EOPNOTSUPP) + error = ENOTSOCK; return (error); - s = splnet(); - head = (struct socket *)fp->f_data; + } + if (head == NULL) { + error = EBADF; + goto out; + } +#if CONFIG_MACF_SOCKET_SUBSET + if ((error = mac_socket_check_accept(kauth_cred_get(), head)) != 0) + goto out; +#endif /* MAC_SOCKET_SUBSET */ + + socket_lock(head, 1); + + if (head->so_proto->pr_getlock != NULL) { + mutex_held = (*head->so_proto->pr_getlock)(head, 0); + dosocklock = 1; + } else { + mutex_held = head->so_proto->pr_domain->dom_mtx; + dosocklock = 0; + } + if ((head->so_options & SO_ACCEPTCONN) == 0) { - splx(s); - return (EINVAL); + if ((head->so_proto->pr_flags & PR_CONNREQUIRED) == 0) { + error = EOPNOTSUPP; + } else { + /* POSIX: The socket is not accepting connections */ + error = EINVAL; + } + socket_unlock(head, 1); + goto out; } if ((head->so_state & SS_NBIO) && head->so_comp.tqh_first == NULL) { - splx(s); - return (EWOULDBLOCK); + socket_unlock(head, 1); + error = EWOULDBLOCK; + goto out; } - while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) { + while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) { if (head->so_state & SS_CANTRCVMORE) { head->so_error = ECONNABORTED; break; } - error = tsleep((caddr_t)&head->so_timeo, PSOCK | PCATCH, - "accept", 0); + if (head->so_usecount < 1) + panic("accept: head=%p refcount=%d\n", head, + head->so_usecount); + error = msleep((caddr_t)&head->so_timeo, mutex_held, + PSOCK | PCATCH, "accept", 0); + if (head->so_usecount < 1) + panic("accept: 2 head=%p refcount=%d\n", head, + head->so_usecount); + if ((head->so_state & SS_DRAINING)) { + error = ECONNABORTED; + } if (error) { - splx(s); - return (error); + socket_unlock(head, 1); + goto out; } } if (head->so_error) { error = head->so_error; head->so_error = 0; - splx(s); - return (error); + socket_unlock(head, 1); + goto out; } @@ -313,14 +431,43 @@ accept1(p, uap, retval, compat) * block allowing another process to accept the connection * instead. */ + lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); so = TAILQ_FIRST(&head->so_comp); TAILQ_REMOVE(&head->so_comp, so, so_list); head->so_qlen--; + /* unlock head to avoid deadlock with select, keep a ref on head */ + socket_unlock(head, 0); + +#if CONFIG_MACF_SOCKET_SUBSET + /* + * Pass the pre-accepted socket to the MAC framework. This is + * cheaper than allocating a file descriptor for the socket, + * calling the protocol accept callback, and possibly freeing + * the file descriptor should the MAC check fails. + */ + if ((error = mac_socket_check_accepted(kauth_cred_get(), so)) != 0) { + so->so_state &= ~(SS_NOFDREF | SS_COMP); + so->so_head = NULL; + soclose(so); + /* Drop reference on listening socket */ + sodereference(head); + goto out; + } +#endif /* MAC_SOCKET_SUBSET */ + + /* + * Pass the pre-accepted socket to any interested socket filter(s). + * Upon failure, the socket would have been closed by the callee. + */ + if (so->so_filt != NULL && (error = soacceptfilter(so)) != 0) { + /* Drop reference on listening socket */ + sodereference(head); + /* Propagate socket filter's error code to the caller */ + goto out; + } fflag = fp->f_flag; - thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); - error = falloc(p, &fp, &fd); - thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); + error = falloc(p, &fp, &newfd, vfs_context_current()); if (error) { /* * Probably ran out of file descriptors. Put the @@ -328,109 +475,193 @@ accept1(p, uap, retval, compat) * do another wakeup so some other process might * have a chance at it. */ + socket_lock(head, 0); TAILQ_INSERT_HEAD(&head->so_comp, so, so_list); head->so_qlen++; - wakeup_one(&head->so_timeo); - splx(s); - return (error); - } else { - *fdflags(p, fd) &= ~UF_RESERVED; - *retval = fd; + wakeup_one((caddr_t)&head->so_timeo); + socket_unlock(head, 1); + goto out; } - - so->so_state &= ~SS_COMP; - so->so_head = NULL; + *retval = newfd; fp->f_type = DTYPE_SOCKET; fp->f_flag = fflag; fp->f_ops = &socketops; fp->f_data = (caddr_t)so; - sa = 0; - (void) soaccept(so, &sa); - if (sa == 0) { + socket_lock(head, 0); + if (dosocklock) + socket_lock(so, 1); + so->so_state &= ~SS_COMP; + so->so_head = NULL; + (void) soacceptlock(so, &sa, 0); + socket_unlock(head, 1); + if (sa == NULL) { namelen = 0; if (uap->name) goto gotnoname; - return 0; + if (dosocklock) + socket_unlock(so, 1); + error = 0; + goto releasefd; } + AUDIT_ARG(sockaddr, vfs_context_cwd(vfs_context_current()), sa); + if (uap->name) { - /* check sa_len before it is destroyed */ - if (namelen > sa->sa_len) - namelen = sa->sa_len; -#ifdef COMPAT_OLDSOCK - if (compat) - ((struct osockaddr *)sa)->sa_family = - sa->sa_family; -#endif - error = copyout(sa, (caddr_t)uap->name, (u_int)namelen); + socklen_t sa_len; + + /* save sa_len before it is destroyed */ + sa_len = sa->sa_len; + namelen = MIN(namelen, sa_len); + error = copyout(sa, uap->name, namelen); if (!error) + /* return the actual, untruncated address length */ + namelen = sa_len; gotnoname: - error = copyout((caddr_t)&namelen, - (caddr_t)uap->anamelen, sizeof (*uap->anamelen)); + error = copyout((caddr_t)&namelen, uap->anamelen, + sizeof (socklen_t)); } FREE(sa, M_SONAME); - splx(s); + + /* + * If the socket has been marked as inactive by soacceptfilter(), + * disallow further operations on it. We explicitly call shutdown + * on both data directions to ensure that SS_CANT{RCV,SEND}MORE + * states are set for the socket. This would also flush out data + * hanging off the receive list of this socket. + */ + if (so->so_flags & SOF_DEFUNCT) { + (void) soshutdownlock(so, SHUT_RD); + (void) soshutdownlock(so, SHUT_WR); + (void) sodisconnectlocked(so); + } + + if (dosocklock) + socket_unlock(so, 1); + +releasefd: + proc_fdlock(p); + procfdtbl_releasefd(p, newfd, NULL); + fp_drop(p, newfd, fp, 1); + proc_fdunlock(p); + +out: + file_drop(fd); return (error); } int -accept(p, uap, retval) - struct proc *p; - struct accept_args *uap; - register_t *retval; +accept(struct proc *p, struct accept_args *uap, register_t *retval) { - - return (accept1(p, uap, retval, 0)); + __pthread_testcancel(1); + return(accept_nocancel(p, (struct accept_nocancel_args *)uap, retval)); } -#ifdef COMPAT_OLDSOCK +/* + * Returns: 0 Success + * EBADF Bad file descriptor + * EALREADY Connection already in progress + * EINPROGRESS Operation in progress + * ECONNABORTED Connection aborted + * EINTR Interrupted function + * EACCES Mandatory Access Control failure + * file_socket:ENOTSOCK + * file_socket:EBADF + * getsockaddr:ENAMETOOLONG Filename too long + * getsockaddr:EINVAL Invalid argument + * getsockaddr:ENOMEM Not enough space + * getsockaddr:EFAULT Bad address + * soconnectlock:EOPNOTSUPP + * soconnectlock:EISCONN + * soconnectlock:??? [depends on protocol, filters] + * msleep:EINTR + * + * Imputed: so_error error may be set from so_error, which + * may have been set by soconnectlock. + */ +/* ARGSUSED */ int -oaccept(p, uap, retval) - struct proc *p; - struct accept_args *uap; - register_t *retval; +connect(struct proc *p, struct connect_args *uap, register_t *retval) { - - return (accept1(p, uap, retval, 1)); + __pthread_testcancel(1); + return(connect_nocancel(p, (struct connect_nocancel_args *)uap, retval)); } -#endif /* COMPAT_OLDSOCK */ -struct connect_args { - int s; - caddr_t name; - int namelen; -}; -/* ARGSUSED */ int -connect(p, uap, retval) - struct proc *p; - register struct connect_args *uap; - register_t *retval; +connect_nocancel(__unused proc_t p, struct connect_nocancel_args *uap, __unused register_t *retval) { - struct file *fp; - register struct socket *so; - struct sockaddr *sa; - int error, s; + struct socket *so; + struct sockaddr_storage ss; + struct sockaddr *sa = NULL; + lck_mtx_t *mutex_held; + boolean_t want_free = TRUE; + int error; + int fd = uap->s; + boolean_t dgram; - error = getsock(p->p_fd, uap->s, &fp); - if (error) + AUDIT_ARG(fd, uap->s); + error = file_socket(fd, &so); + if (error != 0) return (error); - so = (struct socket *)fp->f_data; - if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) - return (EALREADY); - error = getsockaddr(&sa, uap->name, uap->namelen); - if (error) - return (error); - error = soconnect(so, sa); + if (so == NULL) { + error = EBADF; + goto out; + } + + /* + * Ask getsockaddr{_s} to not translate AF_UNSPEC to AF_INET + * if this is a datagram socket; translate for other types. + */ + dgram = (so->so_type == SOCK_DGRAM); + + /* Get socket address now before we obtain socket lock */ + if (uap->namelen > sizeof (ss)) { + error = getsockaddr(so, &sa, uap->name, uap->namelen, !dgram); + } else { + error = getsockaddr_s(so, &ss, uap->name, uap->namelen, !dgram); + if (error == 0) { + sa = (struct sockaddr *)&ss; + want_free = FALSE; + } + } + if (error != 0) + goto out; + + AUDIT_ARG(sockaddr, vfs_context_cwd(vfs_context_current()), sa); +#if CONFIG_MACF_SOCKET_SUBSET + if ((error = mac_socket_check_connect(kauth_cred_get(), so, sa)) != 0) { + if (want_free) + FREE(sa, M_SONAME); + goto out; + } +#endif /* MAC_SOCKET_SUBSET */ + socket_lock(so, 1); + + if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) { + if (want_free) + FREE(sa, M_SONAME); + socket_unlock(so, 1); + error = EALREADY; + goto out; + } + error = soconnectlock(so, sa, 0); if (error) goto bad; if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) { - FREE(sa, M_SONAME); - return (EINPROGRESS); + if (want_free) + FREE(sa, M_SONAME); + socket_unlock(so, 1); + error = EINPROGRESS; + goto out; } - s = splnet(); while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { - error = tsleep((caddr_t)&so->so_timeo, PSOCK | PCATCH, - "connec", 0); + if (so->so_proto->pr_getlock != NULL) + mutex_held = (*so->so_proto->pr_getlock)(so, 0); + else + mutex_held = so->so_proto->pr_domain->dom_mtx; + error = msleep((caddr_t)&so->so_timeo, mutex_held, + PSOCK | PCATCH, "connect", 0); + if ((so->so_state & SS_DRAINING)) { + error = ECONNABORTED; + } if (error) break; } @@ -438,542 +669,546 @@ connect(p, uap, retval) error = so->so_error; so->so_error = 0; } - splx(s); bad: so->so_state &= ~SS_ISCONNECTING; - FREE(sa, M_SONAME); + socket_unlock(so, 1); + if (want_free) + FREE(sa, M_SONAME); if (error == ERESTART) error = EINTR; +out: + file_drop(fd); return (error); } -struct socketpair_args { - int domain; - int type; - int protocol; - int *rsv; -}; +/* + * Returns: 0 Success + * socreate:EAFNOSUPPORT + * socreate:EPROTOTYPE + * socreate:EPROTONOSUPPORT + * socreate:ENOBUFS + * socreate:ENOMEM + * socreate:EISCONN + * socreate:??? [other protocol families, IPSEC] + * falloc:ENFILE + * falloc:EMFILE + * falloc:ENOMEM + * copyout:EFAULT + * soconnect2:EINVAL + * soconnect2:EPROTOTYPE + * soconnect2:??? [other protocol families[ + */ int -socketpair(p, uap, retval) - struct proc *p; - register struct socketpair_args *uap; - register_t *retval; +socketpair(struct proc *p, struct socketpair_args *uap, + __unused register_t *retval) { - register struct filedesc *fdp = p->p_fd; - struct file *fp1, *fp2; + struct fileproc *fp1, *fp2; struct socket *so1, *so2; int fd, error, sv[2]; + AUDIT_ARG(socket, uap->domain, uap->type, uap->protocol); error = socreate(uap->domain, &so1, uap->type, uap->protocol); if (error) return (error); error = socreate(uap->domain, &so2, uap->type, uap->protocol); if (error) goto free1; - thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); - error = falloc(p, &fp1, &fd); - if (error) + + error = falloc(p, &fp1, &fd, vfs_context_current()); + if (error) { goto free2; - sv[0] = fd; + } fp1->f_flag = FREAD|FWRITE; fp1->f_type = DTYPE_SOCKET; fp1->f_ops = &socketops; fp1->f_data = (caddr_t)so1; - error = falloc(p, &fp2, &fd); - if (error) + sv[0] = fd; + + error = falloc(p, &fp2, &fd, vfs_context_current()); + if (error) { goto free3; + } fp2->f_flag = FREAD|FWRITE; fp2->f_type = DTYPE_SOCKET; fp2->f_ops = &socketops; fp2->f_data = (caddr_t)so2; sv[1] = fd; - thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); + error = soconnect2(so1, so2); if (error) { - thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); goto free4; } - if (uap->type == SOCK_DGRAM) { /* * Datagram socket connection is asymmetric. */ - error = soconnect2(so2, so1); - if (error) { - thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); - goto free4; - } - } - *fdflags(p, sv[0]) &= ~UF_RESERVED; - *fdflags(p, sv[1]) &= ~UF_RESERVED; - error = copyout((caddr_t)sv, (caddr_t)uap->rsv, - 2 * sizeof (int)); -#if 0 /* old pipe(2) syscall compatability, unused these days */ - retval[0] = sv[0]; /* XXX ??? */ - retval[1] = sv[1]; /* XXX ??? */ -#endif /* 0 */ + error = soconnect2(so2, so1); + if (error) { + goto free4; + } + } + + proc_fdlock(p); + procfdtbl_releasefd(p, sv[0], NULL); + procfdtbl_releasefd(p, sv[1], NULL); + fp_drop(p, sv[0], fp1, 1); + fp_drop(p, sv[1], fp2, 1); + proc_fdunlock(p); + + error = copyout((caddr_t)sv, uap->rsv, 2 * sizeof (int)); return (error); free4: - fdrelse(p, sv[1]); - ffree(fp2); + fp_free(p, sv[1], fp2); free3: - fdrelse(p, sv[0]); - ffree(fp1); + fp_free(p, sv[0], fp1); free2: - thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); - (void)soclose(so2); + (void) soclose(so2); free1: - (void)soclose(so1); + (void) soclose(so1); return (error); } +/* + * Returns: 0 Success + * EINVAL + * ENOBUFS + * EBADF + * EPIPE + * EACCES Mandatory Access Control failure + * file_socket:ENOTSOCK + * file_socket:EBADF + * getsockaddr:ENAMETOOLONG Filename too long + * getsockaddr:EINVAL Invalid argument + * getsockaddr:ENOMEM Not enough space + * getsockaddr:EFAULT Bad address + * :EACCES[TCP] + * :EADDRINUSE[TCP] + * :EADDRNOTAVAIL[TCP] + * :EAFNOSUPPORT[TCP] + * :EAGAIN[TCP] + * :EBADF + * :ECONNRESET[TCP] + * :EFAULT + * :EHOSTUNREACH[TCP] + * :EINTR + * :EINVAL + * :EISCONN[AF_INET] + * :EMSGSIZE[TCP] + * :ENETDOWN[TCP] + * :ENETUNREACH[TCP] + * :ENOBUFS + * :ENOMEM[TCP] + * :ENOTCONN[AF_INET] + * :EOPNOTSUPP + * :EPERM[TCP] + * :EPIPE + * :EWOULDBLOCK + * :???[TCP] [ignorable: mostly IPSEC/firewall/DLIL] + * :???[AF_INET] [whatever a filter author chooses] + * :??? [value from so_error] + * sockargs:??? + */ static int -sendit(p, s, mp, flags, retsize) - register struct proc *p; - int s; - register struct msghdr *mp; - int flags; - register_t *retsize; +sendit(struct proc *p, int s, struct user_msghdr *mp, uio_t uiop, + int flags, register_t *retval) { - struct file *fp; - struct uio auio; - register struct iovec *iov; - register int i; - struct mbuf *control; - struct sockaddr *to; - int len, error; + struct mbuf *control = NULL; + struct sockaddr_storage ss; + struct sockaddr *to = NULL; + boolean_t want_free = TRUE; + int error; struct socket *so; -#if KTRACE - struct iovec *ktriov = NULL; - struct uio ktruio; -#endif - - KERNEL_DEBUG(DBG_FNC_SENDIT | DBG_FUNC_START, 0,0,0,0,0); + user_ssize_t len; - if (error = getsock(p->p_fd, s, &fp)) - { - KERNEL_DEBUG(DBG_FNC_SENDIT | DBG_FUNC_END, error,0,0,0,0); - return (error); - } - - auio.uio_iov = mp->msg_iov; - auio.uio_iovcnt = mp->msg_iovlen; - auio.uio_segflg = UIO_USERSPACE; - auio.uio_rw = UIO_WRITE; - auio.uio_procp = p; - auio.uio_offset = 0; /* XXX */ - auio.uio_resid = 0; - iov = mp->msg_iov; - for (i = 0; i < mp->msg_iovlen; i++, iov++) { - if (iov->iov_len < 0) - { - KERNEL_DEBUG(DBG_FNC_SENDIT | DBG_FUNC_END, EINVAL,0,0,0,0); - return (EINVAL); - } + KERNEL_DEBUG(DBG_FNC_SENDIT | DBG_FUNC_START, 0, 0, 0, 0, 0); - if ((auio.uio_resid += iov->iov_len) < 0) - { - KERNEL_DEBUG(DBG_FNC_SENDIT | DBG_FUNC_END, EINVAL,0,0,0,0); - return (EINVAL); - } + error = file_socket(s, &so); + if (error) { + KERNEL_DEBUG(DBG_FNC_SENDIT | DBG_FUNC_END, error, 0, 0, 0, 0); + return (error); } - if (mp->msg_name) { - error = getsockaddr(&to, mp->msg_name, mp->msg_namelen); - if (error) { - KERNEL_DEBUG(DBG_FNC_SENDIT | DBG_FUNC_END, error,0,0,0,0); - return (error); + if (so == NULL) { + error = EBADF; + goto out; + } + if (mp->msg_name != USER_ADDR_NULL) { + if (mp->msg_namelen > sizeof (ss)) { + error = getsockaddr(so, &to, mp->msg_name, + mp->msg_namelen, TRUE); + } else { + error = getsockaddr_s(so, &ss, mp->msg_name, + mp->msg_namelen, TRUE); + if (error == 0) { + to = (struct sockaddr *)&ss; + want_free = FALSE; + } } - } else - to = 0; - if (mp->msg_control) { - if (mp->msg_controllen < sizeof(struct cmsghdr) -#ifdef COMPAT_OLDSOCK - && mp->msg_flags != MSG_COMPAT -#endif - ) { + if (error != 0) + goto out; + AUDIT_ARG(sockaddr, vfs_context_cwd(vfs_context_current()), to); + } + if (mp->msg_control != USER_ADDR_NULL) { + if (mp->msg_controllen < sizeof (struct cmsghdr)) { error = EINVAL; goto bad; } error = sockargs(&control, mp->msg_control, mp->msg_controllen, MT_CONTROL); - if (error) + if (error != 0) goto bad; -#ifdef COMPAT_OLDSOCK - if (mp->msg_flags == MSG_COMPAT) { - register struct cmsghdr *cm; - - M_PREPEND(control, sizeof(*cm), M_WAIT); - if (control == 0) { - error = ENOBUFS; - goto bad; - } else { - cm = mtod(control, struct cmsghdr *); - cm->cmsg_len = control->m_len; - cm->cmsg_level = SOL_SOCKET; - cm->cmsg_type = SCM_RIGHTS; - } - } -#endif - } else - control = 0; + } -#if KTRACE - if (KTRPOINT(p, KTR_GENIO)) { - int iovlen = auio.uio_iovcnt * sizeof (struct iovec); +#if CONFIG_MACF_SOCKET_SUBSET + /* + * We check the state without holding the socket lock; + * if a race condition occurs, it would simply result + * in an extra call to the MAC check function. + */ + if (!(so->so_state & SS_ISCONNECTED) && + (error = mac_socket_check_send(kauth_cred_get(), so, to)) != 0) + goto bad; +#endif /* MAC_SOCKET_SUBSET */ - MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); - bcopy((caddr_t)auio.uio_iov, (caddr_t)ktriov, iovlen); - ktruio = auio; - } -#endif - len = auio.uio_resid; - so = (struct socket *)fp->f_data; - error = so->so_proto->pr_usrreqs->pru_sosend(so, to, &auio, 0, control, - flags); - if (error) { - if (auio.uio_resid != len && (error == ERESTART || + len = uio_resid(uiop); + error = so->so_proto->pr_usrreqs->pru_sosend(so, to, uiop, 0, control, + flags); + if (error != 0) { + if (uio_resid(uiop) != len && (error == ERESTART || error == EINTR || error == EWOULDBLOCK)) error = 0; - /* Generation of SIGPIPE can be controlled per socket */ + /* Generation of SIGPIPE can be controlled per socket */ if (error == EPIPE && !(so->so_flags & SOF_NOSIGPIPE)) psignal(p, SIGPIPE); } if (error == 0) - *retsize = len - auio.uio_resid; -#if KTRACE - if (ktriov != NULL) { - if (error == 0) { - ktruio.uio_iov = ktriov; - ktruio.uio_resid = retsize[0]; - ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error, -1); - } - FREE(ktriov, M_TEMP); - } -#endif + *retval = (int)(len - uio_resid(uiop)); bad: - if (to) + if (to != NULL && want_free) FREE(to, M_SONAME); - KERNEL_DEBUG(DBG_FNC_SENDIT | DBG_FUNC_END, error,0,0,0,0); +out: + KERNEL_DEBUG(DBG_FNC_SENDIT | DBG_FUNC_END, error, 0, 0, 0, 0); + file_drop(s); return (error); } - -struct sendto_args { - int s; - caddr_t buf; - size_t len; - int flags; - caddr_t to; - int tolen; -}; - +/* + * Returns: 0 Success + * ENOMEM + * sendit:??? [see sendit definition in this file] + * write:??? [4056224: applicable for pipes] + */ int -sendto(p, uap, retval) - struct proc *p; - register struct sendto_args /* { - int s; - caddr_t buf; - size_t len; - int flags; - caddr_t to; - int tolen; - } */ *uap; - register_t *retval; +sendto(struct proc *p, struct sendto_args *uap, register_t *retval) +{ + __pthread_testcancel(1); + return(sendto_nocancel(p, (struct sendto_nocancel_args *)uap, retval)); +} +int +sendto_nocancel(struct proc *p, struct sendto_nocancel_args *uap, register_t *retval) { - struct msghdr msg; - struct iovec aiov; - int stat; + struct user_msghdr msg; + int error; + uio_t auio = NULL; + + KERNEL_DEBUG(DBG_FNC_SENDTO | DBG_FUNC_START, 0, 0, 0, 0, 0); + AUDIT_ARG(fd, uap->s); - KERNEL_DEBUG(DBG_FNC_SENDTO | DBG_FUNC_START, 0,0,0,0,0); + auio = uio_create(1, 0, + (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32), + UIO_WRITE); + if (auio == NULL) { + return (ENOMEM); + } + uio_addiov(auio, uap->buf, uap->len); msg.msg_name = uap->to; msg.msg_namelen = uap->tolen; - msg.msg_iov = &aiov; - msg.msg_iovlen = 1; + /* no need to set up msg_iov. sendit uses uio_t we send it */ + msg.msg_iov = 0; + msg.msg_iovlen = 0; msg.msg_control = 0; -#ifdef COMPAT_OLDSOCK msg.msg_flags = 0; -#endif - aiov.iov_base = uap->buf; - aiov.iov_len = uap->len; - stat = sendit(p, uap->s, &msg, uap->flags, retval); - KERNEL_DEBUG(DBG_FNC_SENDTO | DBG_FUNC_END, stat, *retval,0,0,0); - return(stat); -} -#ifdef COMPAT_OLDSOCK -struct osend_args { - int s; - caddr_t buf; - int len; - int flags; -}; + error = sendit(p, uap->s, &msg, auio, uap->flags, retval); -int -osend(p, uap, retval) - struct proc *p; - register struct osend_args /* { - int s; - caddr_t buf; - int len; - int flags; - } */ *uap; - register_t *retval; + if (auio != NULL) { + uio_free(auio); + } -{ - struct msghdr msg; - struct iovec aiov; - - msg.msg_name = 0; - msg.msg_namelen = 0; - msg.msg_iov = &aiov; - msg.msg_iovlen = 1; - aiov.iov_base = uap->buf; - aiov.iov_len = uap->len; - msg.msg_control = 0; - msg.msg_flags = 0; - return (sendit(p, uap->s, &msg, uap->flags, retval)); -} -struct osendmsg_args { - int s; - caddr_t msg; - int flags; -}; +#if HACK_FOR_4056224 + /* + * Radar 4056224 + * Temporary workaround to let send() and recv() work over + * a pipe for binary compatibility + * This will be removed in the release following Tiger + */ + if (error == ENOTSOCK) { + struct fileproc *fp; -int -osendmsg(p, uap, retval) - struct proc *p; - register struct osendmsg_args /* { - int s; - caddr_t msg; - int flags; - } */ *uap; - register_t *retval; + if (fp_lookup(p, uap->s, &fp, 0) == 0) { + (void) fp_drop(p, uap->s, fp, 0); -{ - struct msghdr msg; - struct iovec aiov[UIO_SMALLIOV], *iov; - int error; + if (fp->f_type == DTYPE_PIPE) { + struct write_args write_uap; + user_ssize_t write_retval; + + if (p->p_pid > last_pid_4056224) { + last_pid_4056224 = p->p_pid; + + printf("%s[%d] uses send/recv " + "on a pipe\n", p->p_comm, p->p_pid); + } + + bzero(&write_uap, sizeof (struct write_args)); + write_uap.fd = uap->s; + write_uap.cbuf = uap->buf; + write_uap.nbyte = uap->len; + + error = write(p, &write_uap, &write_retval); + *retval = (int)write_retval; + } + } + } +#endif /* HACK_FOR_4056224 */ + + KERNEL_DEBUG(DBG_FNC_SENDTO | DBG_FUNC_END, error, *retval, 0, 0, 0); - error = copyin(uap->msg, (caddr_t)&msg, sizeof (struct omsghdr)); - if (error) - return (error); - if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { - if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) - return (EMSGSIZE); - MALLOC(iov, struct iovec *, - sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, - M_WAITOK); - } else - iov = aiov; - error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov, - (unsigned)(msg.msg_iovlen * sizeof (struct iovec))); - if (error) - goto done; - msg.msg_flags = MSG_COMPAT; - msg.msg_iov = iov; - error = sendit(p, uap->s, &msg, uap->flags, retval); -done: - if (iov != aiov) - FREE(iov, M_IOV); return (error); } -#endif -struct sendmsg_args { - int s; - caddr_t msg; - int flags; -}; +/* + * Returns: 0 Success + * ENOBUFS + * copyin:EFAULT + * sendit:??? [see sendit definition in this file] + */ +int +sendmsg(struct proc *p, struct sendmsg_args *uap, register_t *retval) +{ + __pthread_testcancel(1); + return(sendmsg_nocancel(p, (struct sendmsg_nocancel_args *)uap, retval)); +} int -sendmsg(p, uap, retval) - struct proc *p; - register struct sendmsg_args *uap; - register_t *retval; +sendmsg_nocancel(struct proc *p, struct sendmsg_nocancel_args *uap, register_t *retval) { struct msghdr msg; - struct iovec aiov[UIO_SMALLIOV], *iov; + struct user_msghdr user_msg; + caddr_t msghdrp; + int size_of_msghdr; int error; + int size_of_iovec; + uio_t auio = NULL; + struct user_iovec *iovp; + + KERNEL_DEBUG(DBG_FNC_SENDMSG | DBG_FUNC_START, 0, 0, 0, 0, 0); + AUDIT_ARG(fd, uap->s); + if (IS_64BIT_PROCESS(p)) { + msghdrp = (caddr_t)&user_msg; + size_of_msghdr = sizeof (user_msg); + size_of_iovec = sizeof (struct user_iovec); + } else { + msghdrp = (caddr_t)&msg; + size_of_msghdr = sizeof (msg); + size_of_iovec = sizeof (struct iovec); + } + error = copyin(uap->msg, msghdrp, size_of_msghdr); + if (error) { + KERNEL_DEBUG(DBG_FNC_SENDMSG | DBG_FUNC_END, error, 0, 0, 0, 0); + return (error); + } - KERNEL_DEBUG(DBG_FNC_SENDMSG | DBG_FUNC_START, 0,0,0,0,0); - if (error = copyin(uap->msg, (caddr_t)&msg, sizeof (msg))) - { - KERNEL_DEBUG(DBG_FNC_SENDMSG | DBG_FUNC_END, error,0,0,0,0); - return (error); - } - - if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { - if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) { - KERNEL_DEBUG(DBG_FNC_SENDMSG | DBG_FUNC_END, EMSGSIZE,0,0,0,0); - return (EMSGSIZE); - } - MALLOC(iov, struct iovec *, - sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, - M_WAITOK); - } else - iov = aiov; - if (msg.msg_iovlen && - (error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov, - (unsigned)(msg.msg_iovlen * sizeof (struct iovec))))) + /* only need to copy if user process is not 64-bit */ + if (!IS_64BIT_PROCESS(p)) { + user_msg.msg_flags = msg.msg_flags; + user_msg.msg_controllen = msg.msg_controllen; + user_msg.msg_control = CAST_USER_ADDR_T(msg.msg_control); + user_msg.msg_iovlen = msg.msg_iovlen; + user_msg.msg_iov = CAST_USER_ADDR_T(msg.msg_iov); + user_msg.msg_namelen = msg.msg_namelen; + user_msg.msg_name = CAST_USER_ADDR_T(msg.msg_name); + } + + if (user_msg.msg_iovlen <= 0 || user_msg.msg_iovlen > UIO_MAXIOV) { + KERNEL_DEBUG(DBG_FNC_SENDMSG | DBG_FUNC_END, EMSGSIZE, + 0, 0, 0, 0); + return (EMSGSIZE); + } + + /* allocate a uio large enough to hold the number of iovecs passed */ + auio = uio_create(user_msg.msg_iovlen, 0, + (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32), + UIO_WRITE); + if (auio == NULL) { + error = ENOBUFS; goto done; - msg.msg_iov = iov; -#ifdef COMPAT_OLDSOCK - msg.msg_flags = 0; -#endif - error = sendit(p, uap->s, &msg, uap->flags, retval); + } + + if (user_msg.msg_iovlen) { + /* + * get location of iovecs within the uio. + * then copyin the iovecs from user space. + */ + iovp = uio_iovsaddr(auio); + if (iovp == NULL) { + error = ENOBUFS; + goto done; + } + error = copyin(user_msg.msg_iov, (caddr_t)iovp, + (user_msg.msg_iovlen * size_of_iovec)); + if (error) + goto done; + user_msg.msg_iov = CAST_USER_ADDR_T(iovp); + + /* finish setup of uio_t */ + uio_calculateresid(auio); + } else { + user_msg.msg_iov = 0; + } + + /* msg_flags is ignored for send */ + user_msg.msg_flags = 0; + + error = sendit(p, uap->s, &user_msg, auio, uap->flags, retval); done: - if (iov != aiov) - FREE(iov, M_IOV); - KERNEL_DEBUG(DBG_FNC_SENDMSG | DBG_FUNC_END, error,0,0,0,0); + if (auio != NULL) { + uio_free(auio); + } + KERNEL_DEBUG(DBG_FNC_SENDMSG | DBG_FUNC_END, error, 0, 0, 0, 0); + return (error); } +/* + * Returns: 0 Success + * ENOTSOCK + * EINVAL + * EBADF + * EACCES Mandatory Access Control failure + * copyout:EFAULT + * fp_lookup:EBADF + * :ENOBUFS + * :ENOTCONN + * :EWOULDBLOCK + * :EFAULT + * :EINTR + * :EBADF + * :EINVAL + * :EMSGSIZE + * :??? + * + * Notes: Additional return values from calls through + * depend on protocols other than TCP or AF_UNIX, which are + * documented above. + */ static int -recvit(p, s, mp, namelenp, retval) - register struct proc *p; - int s; - register struct msghdr *mp; - caddr_t namelenp; - register_t *retval; +recvit(struct proc *p, int s, struct user_msghdr *mp, uio_t uiop, + user_addr_t namelenp, register_t *retval) { - struct file *fp; - struct uio auio; - register struct iovec *iov; - register int i; int len, error; struct mbuf *m, *control = 0; - caddr_t ctlbuf; + user_addr_t ctlbuf; struct socket *so; struct sockaddr *fromsa = 0; -#if KTRACE - struct iovec *ktriov = NULL; - struct uio ktruio; -#endif + struct fileproc *fp; - KERNEL_DEBUG(DBG_FNC_RECVIT | DBG_FUNC_START, 0,0,0,0,0); - if (error = getsock(p->p_fd, s, &fp)) - { - KERNEL_DEBUG(DBG_FNC_RECVIT | DBG_FUNC_END, error,0,0,0,0); - return (error); - } - - auio.uio_iov = mp->msg_iov; - auio.uio_iovcnt = mp->msg_iovlen; - auio.uio_segflg = UIO_USERSPACE; - auio.uio_rw = UIO_READ; - auio.uio_procp = p; - auio.uio_offset = 0; /* XXX */ - auio.uio_resid = 0; - iov = mp->msg_iov; - for (i = 0; i < mp->msg_iovlen; i++, iov++) { - if ((auio.uio_resid += iov->iov_len) < 0) { - KERNEL_DEBUG(DBG_FNC_RECVIT | DBG_FUNC_END, EINVAL,0,0,0,0); - return (EINVAL); - } - } -#if KTRACE - if (KTRPOINT(p, KTR_GENIO)) { - int iovlen = auio.uio_iovcnt * sizeof (struct iovec); - - MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); - bcopy((caddr_t)auio.uio_iov, (caddr_t)ktriov, iovlen); - ktruio = auio; + KERNEL_DEBUG(DBG_FNC_RECVIT | DBG_FUNC_START, 0, 0, 0, 0, 0); + proc_fdlock(p); + if ((error = fp_lookup(p, s, &fp, 1))) { + KERNEL_DEBUG(DBG_FNC_RECVIT | DBG_FUNC_END, error, 0, 0, 0, 0); + proc_fdunlock(p); + return (error); } -#endif - len = auio.uio_resid; + if (fp->f_type != DTYPE_SOCKET) { + fp_drop(p, s, fp, 1); + proc_fdunlock(p); + return (ENOTSOCK); + } + so = (struct socket *)fp->f_data; - error = so->so_proto->pr_usrreqs->pru_soreceive(so, &fromsa, &auio, + if (so == NULL) { + fp_drop(p, s, fp, 1); + proc_fdunlock(p); + return (EBADF); + } + + proc_fdunlock(p); + +#if CONFIG_MACF_SOCKET_SUBSET + /* + * We check the state without holding the socket lock; + * if a race condition occurs, it would simply result + * in an extra call to the MAC check function. + */ + if (!(so->so_state & SS_ISCONNECTED) && + (error = mac_socket_check_receive(kauth_cred_get(), so)) != 0) + goto out1; +#endif /* MAC_SOCKET_SUBSET */ + if (uio_resid(uiop) < 0) { + KERNEL_DEBUG(DBG_FNC_RECVIT | DBG_FUNC_END, EINVAL, 0, 0, 0, 0); + error = EINVAL; + goto out1; + } + + len = uio_resid(uiop); + error = so->so_proto->pr_usrreqs->pru_soreceive(so, &fromsa, uiop, (struct mbuf **)0, mp->msg_control ? &control : (struct mbuf **)0, &mp->msg_flags); + AUDIT_ARG(sockaddr, vfs_context_cwd(vfs_context_current()), fromsa); if (error) { - if (auio.uio_resid != len && (error == ERESTART || + if (uio_resid(uiop) != len && (error == ERESTART || error == EINTR || error == EWOULDBLOCK)) error = 0; } -#if KTRACE - if (ktriov != NULL) { - if (error == 0) { - ktruio.uio_iov = ktriov; - ktruio.uio_resid = len - auio.uio_resid; - ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error, -1); - } - FREE(ktriov, M_TEMP); - } -#endif + if (error) goto out; - *retval = len - auio.uio_resid; + + *retval = len - uio_resid(uiop); if (mp->msg_name) { + socklen_t sa_len = 0; + len = mp->msg_namelen; - if (len <= 0 || fromsa == 0) + if (len <= 0 || fromsa == 0) { len = 0; - else { + } else { #ifndef MIN -#define MIN(a,b) ((a)>(b)?(b):(a)) -#endif - /* save sa_len before it is destroyed by MSG_COMPAT */ - len = MIN(len, fromsa->sa_len); -#ifdef COMPAT_OLDSOCK - if (mp->msg_flags & MSG_COMPAT) - ((struct osockaddr *)fromsa)->sa_family = - fromsa->sa_family; +#define MIN(a, b) ((a) > (b) ? (b) : (a)) #endif - error = copyout(fromsa, - (caddr_t)mp->msg_name, (unsigned)len); + sa_len = fromsa->sa_len; + len = MIN((unsigned int)len, sa_len); + error = copyout(fromsa, mp->msg_name, (unsigned)len); if (error) goto out; } - mp->msg_namelen = len; + mp->msg_namelen = sa_len; + /* return the actual, untruncated address length */ if (namelenp && - (error = copyout((caddr_t)&len, namelenp, sizeof (int)))) { -#ifdef COMPAT_OLDSOCK - if (mp->msg_flags & MSG_COMPAT) - error = 0; /* old recvfrom didn't check */ - else -#endif + (error = copyout((caddr_t)&sa_len, namelenp, + sizeof (int)))) { goto out; } } if (mp->msg_control) { -#ifdef COMPAT_OLDSOCK - /* - * We assume that old recvmsg calls won't receive access - * rights and other control info, esp. as control info - * is always optional and those options didn't exist in 4.3. - * If we receive rights, trim the cmsghdr; anything else - * is tossed. - */ - if (control && mp->msg_flags & MSG_COMPAT) { - if (mtod(control, struct cmsghdr *)->cmsg_level != - SOL_SOCKET || - mtod(control, struct cmsghdr *)->cmsg_type != - SCM_RIGHTS) { - mp->msg_controllen = 0; - goto out; - } - control->m_len -= sizeof (struct cmsghdr); - control->m_data += sizeof (struct cmsghdr); - } -#endif len = mp->msg_controllen; m = control; mp->msg_controllen = 0; - ctlbuf = (caddr_t) mp->msg_control; + ctlbuf = mp->msg_control; while (m && len > 0) { unsigned int tocopy; - if (len >= m->m_len) + if (len >= m->m_len) { tocopy = m->m_len; - else { + } else { mp->msg_flags |= MSG_CTRUNC; tocopy = len; } - - if (error = copyout((caddr_t)mtod(m, caddr_t), - ctlbuf, tocopy)) + + error = copyout((caddr_t)mtod(m, caddr_t), ctlbuf, + tocopy); + if (error) goto out; ctlbuf += tocopy; @@ -987,257 +1222,307 @@ out: FREE(fromsa, M_SONAME); if (control) m_freem(control); - KERNEL_DEBUG(DBG_FNC_RECVIT | DBG_FUNC_END, error,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_RECVIT | DBG_FUNC_END, error, 0, 0, 0, 0); +out1: + fp_drop(p, s, fp, 0); return (error); } -struct recvfrom_args { - int s; - caddr_t buf; - size_t len; - int flags; - caddr_t from; - int *fromlenaddr; -}; +/* + * Returns: 0 Success + * ENOMEM + * copyin:EFAULT + * recvit:??? + * read:??? [4056224: applicable for pipes] + * + * Notes: The read entry point is only called as part of support for + * binary backward compatability; new code should use read + * instead of recv or recvfrom when attempting to read data + * from pipes. + * + * For full documentation of the return codes from recvit, see + * the block header for the recvit function. + */ +int +recvfrom(struct proc *p, struct recvfrom_args *uap, register_t *retval) +{ + __pthread_testcancel(1); + return(recvfrom_nocancel(p, (struct recvfrom_nocancel_args *)uap, retval)); +} int -recvfrom(p, uap, retval) - struct proc *p; - register struct recvfrom_args /* { - int s; - caddr_t buf; - size_t len; - int flags; - caddr_t from; - int *fromlenaddr; - } */ *uap; - register_t *retval; +recvfrom_nocancel(struct proc *p, struct recvfrom_nocancel_args *uap, register_t *retval) { - struct msghdr msg; - struct iovec aiov; + struct user_msghdr msg; int error; + uio_t auio = NULL; - KERNEL_DEBUG(DBG_FNC_RECVFROM | DBG_FUNC_START, 0,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_RECVFROM | DBG_FUNC_START, 0, 0, 0, 0, 0); + AUDIT_ARG(fd, uap->s); if (uap->fromlenaddr) { - error = copyin((caddr_t)uap->fromlenaddr, + error = copyin(uap->fromlenaddr, (caddr_t)&msg.msg_namelen, sizeof (msg.msg_namelen)); if (error) return (error); - } else + } else { msg.msg_namelen = 0; + } msg.msg_name = uap->from; - msg.msg_iov = &aiov; - msg.msg_iovlen = 1; - aiov.iov_base = uap->buf; - aiov.iov_len = uap->len; + auio = uio_create(1, 0, + (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32), + UIO_READ); + if (auio == NULL) { + return (ENOMEM); + } + + uio_addiov(auio, uap->buf, uap->len); + /* no need to set up msg_iov. recvit uses uio_t we send it */ + msg.msg_iov = 0; + msg.msg_iovlen = 0; msg.msg_control = 0; + msg.msg_controllen = 0; msg.msg_flags = uap->flags; - KERNEL_DEBUG(DBG_FNC_RECVFROM | DBG_FUNC_END, error,0,0,0,0); - return (recvit(p, uap->s, &msg, (caddr_t)uap->fromlenaddr, retval)); -} + error = recvit(p, uap->s, &msg, auio, uap->fromlenaddr, retval); + if (auio != NULL) { + uio_free(auio); + } -#ifdef COMPAT_OLDSOCK -int -orecvfrom(p, uap, retval) - struct proc *p; - struct recvfrom_args *uap; - register_t *retval; -{ +#if HACK_FOR_4056224 + /* + * Radar 4056224 + * Temporary workaround to let send() and recv() work over + * a pipe for binary compatibility + * This will be removed in the release following Tiger + */ + if (error == ENOTSOCK && proc_is64bit(p) == 0) { + struct fileproc *fp; - uap->flags |= MSG_COMPAT; - return (recvfrom(p, uap)); -} -#endif + if (fp_lookup(p, uap->s, &fp, 0) == 0) { + (void) fp_drop(p, uap->s, fp, 0); + if (fp->f_type == DTYPE_PIPE) { + struct read_args read_uap; + user_ssize_t read_retval; -#ifdef COMPAT_OLDSOCK -struct orecv_args { - int s; - caddr_t buf; - int len; - int flags; -}; + if (p->p_pid > last_pid_4056224) { + last_pid_4056224 = p->p_pid; -int -orecv(p, uap, retval) - struct proc *p; - struct orecv_args *uap; - register_t *retval; -{ - struct msghdr msg; - struct iovec aiov; - - msg.msg_name = 0; - msg.msg_namelen = 0; - msg.msg_iov = &aiov; - msg.msg_iovlen = 1; - aiov.iov_base = uap->buf; - aiov.iov_len = uap->len; - msg.msg_control = 0; - msg.msg_flags = uap->flags; - return (recvit(p, uap->s, &msg, (caddr_t)0, retval)); + printf("%s[%d] uses send/recv on " + "a pipe\n", p->p_comm, p->p_pid); + } + + bzero(&read_uap, sizeof (struct read_args)); + read_uap.fd = uap->s; + read_uap.cbuf = uap->buf; + read_uap.nbyte = uap->len; + + error = read(p, &read_uap, &read_retval); + *retval = (int)read_retval; + } + } + } +#endif /* HACK_FOR_4056224 */ + + KERNEL_DEBUG(DBG_FNC_RECVFROM | DBG_FUNC_END, error, 0, 0, 0, 0); + + return (error); } /* - * Old recvmsg. This code takes advantage of the fact that the old msghdr - * overlays the new one, missing only the flags, and with the (old) access - * rights where the control fields are now. + * Returns: 0 Success + * EMSGSIZE + * ENOMEM + * copyin:EFAULT + * copyout:EFAULT + * recvit:??? + * + * Notes: For full documentation of the return codes from recvit, see + * the block header for the recvit function. */ -struct orecvmsg_args { - int s; - struct omsghdr *msg; - int flags; -}; +int +recvmsg(struct proc *p, struct recvmsg_args *uap, register_t *retval) +{ + __pthread_testcancel(1); + return(recvmsg_nocancel(p, (struct recvmsg_nocancel_args *)uap, retval)); +} int -orecvmsg(p, uap, retval) - struct proc *p; - struct orecvmsg_args *uap; - register_t *retval; +recvmsg_nocancel(struct proc *p, struct recvmsg_nocancel_args *uap, register_t *retval) { struct msghdr msg; - struct iovec aiov[UIO_SMALLIOV], *iov; + struct user_msghdr user_msg; + caddr_t msghdrp; + int size_of_msghdr; + user_addr_t uiov; int error; - - error = copyin((caddr_t)uap->msg, (caddr_t)&msg, - sizeof (struct omsghdr)); - if (error) + int size_of_iovec; + uio_t auio = NULL; + struct user_iovec *iovp; + + KERNEL_DEBUG(DBG_FNC_RECVMSG | DBG_FUNC_START, 0, 0, 0, 0, 0); + AUDIT_ARG(fd, uap->s); + if (IS_64BIT_PROCESS(p)) { + msghdrp = (caddr_t)&user_msg; + size_of_msghdr = sizeof (user_msg); + size_of_iovec = sizeof (struct user_iovec); + } else { + msghdrp = (caddr_t)&msg; + size_of_msghdr = sizeof (msg); + size_of_iovec = sizeof (struct iovec); + } + error = copyin(uap->msg, msghdrp, size_of_msghdr); + if (error) { + KERNEL_DEBUG(DBG_FNC_RECVMSG | DBG_FUNC_END, error, 0, 0, 0, 0); return (error); - if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { - if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) - return (EMSGSIZE); - MALLOC(iov, struct iovec *, - sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, - M_WAITOK); - } else - iov = aiov; - msg.msg_flags = uap->flags | MSG_COMPAT; - error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov, - (unsigned)(msg.msg_iovlen * sizeof (struct iovec))); - if (error) - goto done; - msg.msg_iov = iov; - error = recvit(p, uap->s, &msg, (caddr_t)&uap->msg->msg_namelen, retval); + } - if (msg.msg_controllen && error == 0) - error = copyout((caddr_t)&msg.msg_controllen, - (caddr_t)&uap->msg->msg_accrightslen, sizeof (int)); -done: - if (iov != aiov) - FREE(iov, M_IOV); - return (error); -} -#endif + /* only need to copy if user process is not 64-bit */ + if (!IS_64BIT_PROCESS(p)) { + user_msg.msg_flags = msg.msg_flags; + user_msg.msg_controllen = msg.msg_controllen; + user_msg.msg_control = CAST_USER_ADDR_T(msg.msg_control); + user_msg.msg_iovlen = msg.msg_iovlen; + user_msg.msg_iov = CAST_USER_ADDR_T(msg.msg_iov); + user_msg.msg_namelen = msg.msg_namelen; + user_msg.msg_name = CAST_USER_ADDR_T(msg.msg_name); + } -struct recvmsg_args { - int s; - struct msghdr *msg; - int flags; -}; + if (user_msg.msg_iovlen <= 0 || user_msg.msg_iovlen > UIO_MAXIOV) { + KERNEL_DEBUG(DBG_FNC_RECVMSG | DBG_FUNC_END, EMSGSIZE, + 0, 0, 0, 0); + return (EMSGSIZE); + } -int -recvmsg(p, uap, retval) - struct proc *p; - struct recvmsg_args *uap; - register_t *retval; -{ - struct msghdr msg; - struct iovec aiov[UIO_SMALLIOV], *uiov, *iov; - register int error; + user_msg.msg_flags = uap->flags; - KERNEL_DEBUG(DBG_FNC_RECVMSG | DBG_FUNC_START, 0,0,0,0,0); - if (error = copyin((caddr_t)uap->msg, (caddr_t)&msg, - sizeof (msg))) - { - KERNEL_DEBUG(DBG_FNC_RECVMSG | DBG_FUNC_END, error,0,0,0,0); - return (error); + /* allocate a uio large enough to hold the number of iovecs passed */ + auio = uio_create(user_msg.msg_iovlen, 0, + (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32), + UIO_READ); + if (auio == NULL) { + error = ENOMEM; + goto done; } - if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { - if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) { - KERNEL_DEBUG(DBG_FNC_RECVMSG | DBG_FUNC_END, EMSGSIZE,0,0,0,0); - return (EMSGSIZE); - } - MALLOC(iov, struct iovec *, - sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, - M_WAITOK); - } else - iov = aiov; -#ifdef COMPAT_OLDSOCK - msg.msg_flags = uap->flags &~ MSG_COMPAT; -#else - msg.msg_flags = uap->flags; -#endif - uiov = msg.msg_iov; - msg.msg_iov = iov; - error = copyin((caddr_t)uiov, (caddr_t)iov, - (unsigned)(msg.msg_iovlen * sizeof (struct iovec))); + /* + * get location of iovecs within the uio. then copyin the iovecs from + * user space. + */ + iovp = uio_iovsaddr(auio); + if (iovp == NULL) { + error = ENOMEM; + goto done; + } + uiov = user_msg.msg_iov; + user_msg.msg_iov = CAST_USER_ADDR_T(iovp); + error = copyin(uiov, (caddr_t)iovp, + (user_msg.msg_iovlen * size_of_iovec)); if (error) goto done; - error = recvit(p, uap->s, &msg, (caddr_t)0, retval); + + /* finish setup of uio_t */ + uio_calculateresid(auio); + + error = recvit(p, uap->s, &user_msg, auio, 0, retval); if (!error) { - msg.msg_iov = uiov; - error = copyout((caddr_t)&msg, (caddr_t)uap->msg, sizeof(msg)); + user_msg.msg_iov = uiov; + /* only need to copy if user process is not 64-bit */ + if (!IS_64BIT_PROCESS(p)) { + // LP64todo - do all these change? if not, then no need to copy all of them! + msg.msg_flags = user_msg.msg_flags; + msg.msg_controllen = user_msg.msg_controllen; + msg.msg_control = + CAST_DOWN(caddr_t, user_msg.msg_control); + msg.msg_iovlen = user_msg.msg_iovlen; + msg.msg_iov = (struct iovec *) + CAST_DOWN(caddr_t, user_msg.msg_iov); + msg.msg_namelen = user_msg.msg_namelen; + msg.msg_name = CAST_DOWN(caddr_t, user_msg.msg_name); + } + error = copyout(msghdrp, uap->msg, size_of_msghdr); } done: - if (iov != aiov) - FREE(iov, M_IOV); - KERNEL_DEBUG(DBG_FNC_RECVMSG | DBG_FUNC_END, error,0,0,0,0); + if (auio != NULL) { + uio_free(auio); + } + KERNEL_DEBUG(DBG_FNC_RECVMSG | DBG_FUNC_END, error, 0, 0, 0, 0); return (error); } +/* + * Returns: 0 Success + * EBADF + * file_socket:ENOTSOCK + * file_socket:EBADF + * soshutdown:EINVAL + * soshutdown:ENOTCONN + * soshutdown:EADDRNOTAVAIL[TCP] + * soshutdown:ENOBUFS[TCP] + * soshutdown:EMSGSIZE[TCP] + * soshutdown:EHOSTUNREACH[TCP] + * soshutdown:ENETUNREACH[TCP] + * soshutdown:ENETDOWN[TCP] + * soshutdown:ENOMEM[TCP] + * soshutdown:EACCES[TCP] + * soshutdown:EMSGSIZE[TCP] + * soshutdown:ENOBUFS[TCP] + * soshutdown:???[TCP] [ignorable: mostly IPSEC/firewall/DLIL] + * soshutdown:??? [other protocol families] + */ /* ARGSUSED */ -struct shutdown_args { - int s; - int how; -}; - int -shutdown(p, uap, retval) - struct proc *p; - struct shutdown_args *uap; - register_t *retval; +shutdown(__unused struct proc *p, struct shutdown_args *uap, + __unused register_t *retval) { - struct file *fp; + struct socket *so; int error; - error = getsock(p->p_fd, uap->s, &fp); + AUDIT_ARG(fd, uap->s); + error = file_socket(uap->s, &so); if (error) return (error); - return (soshutdown((struct socket *)fp->f_data, uap->how)); + if (so == NULL) { + error = EBADF; + goto out; + } + error = soshutdown((struct socket *)so, uap->how); +out: + file_drop(uap->s); + return (error); } - - - - +/* + * Returns: 0 Success + * EFAULT + * EINVAL + * EACCES Mandatory Access Control failure + * file_socket:ENOTSOCK + * file_socket:EBADF + * sosetopt:EINVAL + * sosetopt:ENOPROTOOPT + * sosetopt:ENOBUFS + * sosetopt:EDOM + * sosetopt:EFAULT + * sosetopt:EOPNOTSUPP[AF_UNIX] + * sosetopt:??? + */ /* ARGSUSED */ -struct setsockopt_args { - int s; - int level; - int name; - caddr_t val; - int valsize; -}; - int -setsockopt(p, uap, retval) - struct proc *p; - struct setsockopt_args *uap; - register_t *retval; +setsockopt(struct proc *p, struct setsockopt_args *uap, + __unused register_t *retval) { - struct file *fp; + struct socket *so; struct sockopt sopt; int error; + AUDIT_ARG(fd, uap->s); if (uap->val == 0 && uap->valsize != 0) return (EFAULT); - if (uap->valsize < 0) - return (EINVAL); + /* No bounds checking on size (it's unsigned) */ - error = getsock(p->p_fd, uap->s, &fp); + error = file_socket(uap->s, &so); if (error) return (error); @@ -1248,34 +1533,55 @@ setsockopt(p, uap, retval) sopt.sopt_valsize = uap->valsize; sopt.sopt_p = p; - return (sosetopt((struct socket *)fp->f_data, &sopt)); + if (so == NULL) { + error = EINVAL; + goto out; + } +#if CONFIG_MACF_SOCKET_SUBSET + if ((error = mac_socket_check_setsockopt(kauth_cred_get(), so, + &sopt)) != 0) + goto out; +#endif /* MAC_SOCKET_SUBSET */ + error = sosetopt(so, &sopt); +out: + file_drop(uap->s); + return (error); } +/* + * Returns: 0 Success + * EINVAL + * EBADF + * EACCES Mandatory Access Control failure + * copyin:EFAULT + * copyout:EFAULT + * file_socket:ENOTSOCK + * file_socket:EBADF + * sogetopt:??? + */ int -getsockopt(p, uap, retval) - struct proc *p; - struct getsockopt_args *uap; - register_t *retval; +getsockopt(struct proc *p, struct getsockopt_args *uap, + __unused register_t *retval) { - int valsize, error; - struct file *fp; - struct sockopt sopt; + int error; + socklen_t valsize; + struct sockopt sopt; + struct socket *so; - error = getsock(p->p_fd, uap->s, &fp); + error = file_socket(uap->s, &so); if (error) return (error); if (uap->val) { - error = copyin((caddr_t)uap->avalsize, (caddr_t)&valsize, + error = copyin(uap->avalsize, (caddr_t)&valsize, sizeof (valsize)); if (error) - return (error); - if (valsize < 0) - return (EINVAL); - } else + goto out; + /* No bounds checking on size (it's unsigned) */ + } else { valsize = 0; - + } sopt.sopt_dir = SOPT_GET; sopt.sopt_level = uap->level; sopt.sopt_name = uap->name; @@ -1283,107 +1589,90 @@ getsockopt(p, uap, retval) sopt.sopt_valsize = (size_t)valsize; /* checked non-negative above */ sopt.sopt_p = p; - error = sogetopt((struct socket *)fp->f_data, &sopt); + if (so == NULL) { + error = EBADF; + goto out; + } +#if CONFIG_MACF_SOCKET_SUBSET + if ((error = mac_socket_check_getsockopt(kauth_cred_get(), so, + &sopt)) != 0) + goto out; +#endif /* MAC_SOCKET_SUBSET */ + error = sogetopt((struct socket *)so, &sopt); if (error == 0) { valsize = sopt.sopt_valsize; - error = copyout((caddr_t)&valsize, - (caddr_t)uap->avalsize, sizeof (valsize)); - } - return (error); -} - - - -struct pipe_args { - int dummy; -}; -/* ARGSUSED */ -int -pipe(p, uap, retval) - struct proc *p; - struct pipe_args *uap; - register_t *retval; -{ - struct file *rf, *wf; - struct socket *rso, *wso; - int fd, error; - - thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); - if (error = socreate(AF_UNIX, &rso, SOCK_STREAM, 0)) { - thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); - return (error); - } - if (error = socreate(AF_UNIX, &wso, SOCK_STREAM, 0)) { - goto free1; + error = copyout((caddr_t)&valsize, uap->avalsize, + sizeof (valsize)); } - thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); - error = falloc(p, &rf, &fd); - if (error) - goto free2; - retval[0] = fd; - rf->f_flag = FREAD; - rf->f_type = DTYPE_SOCKET; - rf->f_ops = &socketops; - rf->f_data = (caddr_t)rso; - if (error = falloc(p, &wf, &fd)) - goto free3; - wf->f_flag = FWRITE; - wf->f_type = DTYPE_SOCKET; - wf->f_ops = &socketops; - wf->f_data = (caddr_t)wso; - retval[1] = fd; - - thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); - error = unp_connect2(wso, rso); - thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); - if (error) - goto free4; - *fdflags(p, retval[0]) &= ~UF_RESERVED; - *fdflags(p, retval[1]) &= ~UF_RESERVED; - return (0); -free4: - fdrelse(p, retval[1]); - ffree(wf); -free3: - fdrelse(p, retval[0]); - ffree(rf); -free2: - thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); - (void)soclose(wso); -free1: - (void)soclose(rso); - - thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); +out: + file_drop(uap->s); return (error); } /* * Get socket name. + * + * Returns: 0 Success + * EBADF + * file_socket:ENOTSOCK + * file_socket:EBADF + * copyin:EFAULT + * copyout:EFAULT + * :ENOBUFS[TCP] + * :ECONNRESET[TCP] + * :EINVAL[AF_UNIX] + * :??? */ /* ARGSUSED */ -static int -getsockname1(p, uap, retval, compat) - struct proc *p; - register struct getsockname_args *uap; - register_t *retval; - int compat; +int +getsockname(__unused struct proc *p, struct getsockname_args *uap, + __unused register_t *retval) { - struct file *fp; - register struct socket *so; + struct socket *so; struct sockaddr *sa; - u_int len; + socklen_t len; + socklen_t sa_len; int error; - error = getsock(p->p_fd, uap->fdes, &fp); + error = file_socket(uap->fdes, &so); if (error) return (error); - error = copyin((caddr_t)uap->alen, (caddr_t)&len, sizeof (len)); + error = copyin(uap->alen, (caddr_t)&len, sizeof (socklen_t)); if (error) - return (error); - so = (struct socket *)fp->f_data; + goto out; + if (so == NULL) { + error = EBADF; + goto out; + } sa = 0; + socket_lock(so, 1); error = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, &sa); + if (error == 0) { + struct socket_filter_entry *filter; + int filtered = 0; + for (filter = so->so_filt; filter && error == 0; + filter = filter->sfe_next_onsocket) { + if (filter->sfe_filter->sf_filter.sf_getsockname) { + if (!filtered) { + filtered = 1; + sflt_use(so); + socket_unlock(so, 0); + } + error = filter->sfe_filter->sf_filter. + sf_getsockname(filter->sfe_cookie, so, &sa); + } + } + + if (error == EJUSTRETURN) + error = 0; + + if (filtered) { + socket_lock(so, 0); + sflt_unuse(so); + } + } + socket_unlock(so, 1); if (error) goto bad; if (sa == 0) { @@ -1391,536 +1680,598 @@ getsockname1(p, uap, retval, compat) goto gotnothing; } - len = MIN(len, sa->sa_len); -#ifdef COMPAT_OLDSOCK - if (compat) - ((struct osockaddr *)sa)->sa_family = sa->sa_family; -#endif - error = copyout(sa, (caddr_t)uap->asa, (u_int)len); - if (error == 0) + sa_len = sa->sa_len; + len = MIN(len, sa_len); + error = copyout((caddr_t)sa, uap->asa, len); + if (error) + goto bad; + /* return the actual, untruncated address length */ + len = sa_len; gotnothing: - error = copyout((caddr_t)&len, (caddr_t)uap->alen, - sizeof (len)); + error = copyout((caddr_t)&len, uap->alen, sizeof (socklen_t)); bad: if (sa) FREE(sa, M_SONAME); +out: + file_drop(uap->fdes); return (error); } -int -getsockname(p, uap, retval) - struct proc *p; - struct getsockname_args *uap; - register_t *retval; -{ - - return (getsockname1(p, uap, retval, 0)); -} - -#ifdef COMPAT_OLDSOCK -int -ogetsockname(p, uap, retval) - struct proc *p; - struct getsockname_args *uap; - register_t *retval; -{ - - return (getsockname1(p, uap, retval, 1)); -} -#endif /* COMPAT_OLDSOCK */ - /* * Get name of peer for connected socket. + * + * Returns: 0 Success + * EBADF + * EINVAL + * ENOTCONN + * file_socket:ENOTSOCK + * file_socket:EBADF + * copyin:EFAULT + * copyout:EFAULT + * :??? + * :??? */ /* ARGSUSED */ int -getpeername1(p, uap, retval, compat) - struct proc *p; - register struct getpeername_args *uap; - register_t *retval; - int compat; +getpeername(__unused struct proc *p, struct getpeername_args *uap, + __unused register_t *retval) { - struct file *fp; - register struct socket *so; + struct socket *so; struct sockaddr *sa; - u_int len; + socklen_t len; + socklen_t sa_len; int error; - error = getsock(p->p_fd, uap->fdes, &fp); - if (error) - return (error); - so = (struct socket *)fp->f_data; - if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) - return (ENOTCONN); - error = copyin((caddr_t)uap->alen, (caddr_t)&len, sizeof (len)); + error = file_socket(uap->fdes, &so); if (error) return (error); + if (so == NULL) { + error = EBADF; + goto out; + } + + socket_lock(so, 1); + + if ((so->so_state & (SS_CANTRCVMORE | SS_CANTSENDMORE)) == + (SS_CANTRCVMORE | SS_CANTSENDMORE)) { + /* the socket has been shutdown, no more getpeername's */ + socket_unlock(so, 1); + error = EINVAL; + goto out; + } + + if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { + socket_unlock(so, 1); + error = ENOTCONN; + goto out; + } + error = copyin(uap->alen, (caddr_t)&len, sizeof (socklen_t)); + if (error) { + socket_unlock(so, 1); + goto out; + } sa = 0; error = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, &sa); + if (error == 0) { + struct socket_filter_entry *filter; + int filtered = 0; + for (filter = so->so_filt; filter && error == 0; + filter = filter->sfe_next_onsocket) { + if (filter->sfe_filter->sf_filter.sf_getpeername) { + if (!filtered) { + filtered = 1; + sflt_use(so); + socket_unlock(so, 0); + } + error = filter->sfe_filter->sf_filter. + sf_getpeername(filter->sfe_cookie, so, &sa); + } + } + + if (error == EJUSTRETURN) + error = 0; + + if (filtered) { + socket_lock(so, 0); + sflt_unuse(so); + } + } + socket_unlock(so, 1); if (error) goto bad; if (sa == 0) { len = 0; goto gotnothing; } - len = MIN(len, sa->sa_len); -#ifdef COMPAT_OLDSOCK - if (compat) - ((struct osockaddr *)sa)->sa_family = - sa->sa_family; -#endif - error = copyout(sa, (caddr_t)uap->asa, (u_int)len); + sa_len = sa->sa_len; + len = MIN(len, sa_len); + error = copyout(sa, uap->asa, len); if (error) goto bad; + /* return the actual, untruncated address length */ + len = sa_len; gotnothing: - error = copyout((caddr_t)&len, (caddr_t)uap->alen, sizeof (len)); + error = copyout((caddr_t)&len, uap->alen, sizeof (socklen_t)); bad: if (sa) FREE(sa, M_SONAME); +out: + file_drop(uap->fdes); return (error); } int -getpeername(p, uap, retval) - struct proc *p; - struct getpeername_args *uap; - register_t *retval; -{ - - return (getpeername1(p, uap, retval, 0)); -} - -#ifdef COMPAT_OLDSOCK -int -ogetpeername(p, uap, retval) - struct proc *p; - struct ogetpeername_args *uap; - register_t *retval; +sockargs(struct mbuf **mp, user_addr_t data, int buflen, int type) { - - /* XXX uap should have type `getpeername_args *' to begin with. */ - return (getpeername1(p, (struct getpeername_args *)uap, retval, 1)); -} -#endif /* COMPAT_OLDSOCK */ - -int -sockargs(mp, buf, buflen, type) - struct mbuf **mp; - caddr_t buf; - int buflen, type; -{ - register struct sockaddr *sa; - register struct mbuf *m; + struct sockaddr *sa; + struct mbuf *m; int error; if ((u_int)buflen > MLEN) { -#ifdef COMPAT_OLDSOCK if (type == MT_SONAME && (u_int)buflen <= 112) buflen = MLEN; /* unix domain compat. hack */ - else -#endif - return (EINVAL); + else if ((u_int)buflen > MCLBYTES) + return (EINVAL); } m = m_get(M_WAIT, type); if (m == NULL) return (ENOBUFS); + if ((u_int)buflen > MLEN) { + MCLGET(m, M_WAIT); + if ((m->m_flags & M_EXT) == 0) { + m_free(m); + return (ENOBUFS); + } + } m->m_len = buflen; - error = copyin(buf, mtod(m, caddr_t), (u_int)buflen); - if (error) + error = copyin(data, mtod(m, caddr_t), (u_int)buflen); + if (error) { (void) m_free(m); - else { + } else { *mp = m; if (type == MT_SONAME) { sa = mtod(m, struct sockaddr *); - -#if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN - if (sa->sa_family == 0 && sa->sa_len < AF_MAX) - sa->sa_family = sa->sa_len; -#endif sa->sa_len = buflen; } } return (error); } -int -getsockaddr(namp, uaddr, len) - struct sockaddr **namp; - caddr_t uaddr; - size_t len; +/* + * Given a user_addr_t of length len, allocate and fill out a *sa. + * + * Returns: 0 Success + * ENAMETOOLONG Filename too long + * EINVAL Invalid argument + * ENOMEM Not enough space + * copyin:EFAULT Bad address + */ +static int +getsockaddr(struct socket *so, struct sockaddr **namp, user_addr_t uaddr, + size_t len, boolean_t translate_unspec) { struct sockaddr *sa; int error; if (len > SOCK_MAXADDRLEN) - return ENAMETOOLONG; + return (ENAMETOOLONG); - if (len == 0) - return EINVAL; + if (len < offsetof(struct sockaddr, sa_data[0])) + return (EINVAL); - MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK); - error = copyin(uaddr, sa, len); + MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK | M_ZERO); + if (sa == NULL) { + return (ENOMEM); + } + error = copyin(uaddr, (caddr_t)sa, len); if (error) { FREE(sa, M_SONAME); } else { -#if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN - if (sa->sa_family == 0 && sa->sa_len < AF_MAX) - sa->sa_family = sa->sa_len; -#endif + /* + * Force sa_family to AF_INET on AF_INET sockets to handle + * legacy applications that use AF_UNSPEC (0). On all other + * sockets we leave it unchanged and let the lower layer + * handle it. + */ + if (translate_unspec && sa->sa_family == AF_UNSPEC && + INP_CHECK_SOCKAF(so, AF_INET) && + len == sizeof (struct sockaddr_in)) + sa->sa_family = AF_INET; + sa->sa_len = len; *namp = sa; } - return error; + return (error); } -int -getsock(fdp, fdes, fpp) - struct filedesc *fdp; - int fdes; - struct file **fpp; +static int +getsockaddr_s(struct socket *so, struct sockaddr_storage *ss, + user_addr_t uaddr, size_t len, boolean_t translate_unspec) { - register struct file *fp; + int error; - if ((unsigned)fdes >= fdp->fd_nfiles || - (fp = fdp->fd_ofiles[fdes]) == NULL || - (fdp->fd_ofileflags[fdes] & UF_RESERVED)) - return (EBADF); - if (fp->f_type != DTYPE_SOCKET) - return (ENOTSOCK); - *fpp = fp; - return (0); -} + if (ss == NULL || uaddr == USER_ADDR_NULL || + len < offsetof(struct sockaddr, sa_data[0])) + return (EINVAL); -#if SENDFILE -/* - * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-)) - * XXX - The sf_buf functions are currently private to sendfile(2), so have - * been made static, but may be useful in the future for doing zero-copy in - * other parts of the networking code. - */ -static void -sf_buf_init(void *arg) -{ - int i; + /* + * sockaddr_storage size is less than SOCK_MAXADDRLEN, + * so the check here is inclusive. + */ + if (len > sizeof (*ss)) + return (ENAMETOOLONG); - SLIST_INIT(&sf_freelist); - sf_base = kmem_alloc_pageable(kernel_map, nsfbufs * PAGE_SIZE); - sf_bufs = _MALLOC(nsfbufs * sizeof(struct sf_buf), M_TEMP, M_NOWAIT); - bzero(sf_bufs, nsfbufs * sizeof(struct sf_buf)); - for (i = 0; i < nsfbufs; i++) { - sf_bufs[i].kva = sf_base + i * PAGE_SIZE; - SLIST_INSERT_HEAD(&sf_freelist, &sf_bufs[i], free_list); + bzero(ss, sizeof (*ss)); + error = copyin(uaddr, (caddr_t)ss, len); + if (error == 0) { + /* + * Force sa_family to AF_INET on AF_INET sockets to handle + * legacy applications that use AF_UNSPEC (0). On all other + * sockets we leave it unchanged and let the lower layer + * handle it. + */ + if (translate_unspec && ss->ss_family == AF_UNSPEC && + INP_CHECK_SOCKAF(so, AF_INET) && + len == sizeof (struct sockaddr_in)) + ss->ss_family = AF_INET; + + ss->ss_len = len; } + return (error); } -/* - * Get an sf_buf from the freelist. Will block if none are available. - */ -static struct sf_buf * -sf_buf_alloc() -{ - struct sf_buf *sf; - int s; - - s = splimp(); - while ((sf = SLIST_FIRST(&sf_freelist)) == NULL) { - sf_buf_alloc_want = 1; - tsleep(&sf_freelist, PVM, "sfbufa", 0); - } - SLIST_REMOVE_HEAD(&sf_freelist, free_list); - splx(s); - sf->refcnt = 1; - return (sf); -} +#if SENDFILE -#define dtosf(x) (&sf_bufs[((uintptr_t)(x) - (uintptr_t)sf_base) >> PAGE_SHIFT]) -static void -sf_buf_ref(caddr_t addr, u_int size) +SYSCTL_DECL(_kern_ipc); + +#define SFUIOBUFS 64 +static int sendfileuiobufs = SFUIOBUFS; +SYSCTL_INT(_kern_ipc, OID_AUTO, sendfileuiobufs, CTLFLAG_RW, &sendfileuiobufs, + 0, ""); + +/* Macros to compute the number of mbufs needed depending on cluster size */ +#define HOWMANY_16K(n) ((((unsigned int)(n) - 1) >> (PGSHIFT + 2)) + 1) +#define HOWMANY_4K(n) ((((unsigned int)(n) - 1) >> PGSHIFT) + 1) + +/* Upper send limit in bytes (sendfileuiobufs * PAGESIZE) */ +#define SENDFILE_MAX_BYTES (sendfileuiobufs << PGSHIFT) + +/* Upper send limit in the number of mbuf clusters */ +#define SENDFILE_MAX_16K HOWMANY_16K(SENDFILE_MAX_BYTES) +#define SENDFILE_MAX_4K HOWMANY_4K(SENDFILE_MAX_BYTES) + +size_t mbuf_pkt_maxlen(mbuf_t m); + +__private_extern__ size_t +mbuf_pkt_maxlen(mbuf_t m) { - struct sf_buf *sf; + size_t maxlen = 0; - sf = dtosf(addr); - if (sf->refcnt == 0) - panic("sf_buf_ref: referencing a free sf_buf"); - sf->refcnt++; + while (m) { + maxlen += mbuf_maxlen(m); + m = mbuf_next(m); + } + return (maxlen); } -/* - * Lose a reference to an sf_buf. When none left, detach mapped page - * and release resources back to the system. - * - * Must be called at splimp. - */ static void -sf_buf_free(caddr_t addr, u_int size) +alloc_sendpkt(int how, size_t pktlen, unsigned int *maxchunks, + struct mbuf **m, boolean_t jumbocl) { - struct sf_buf *sf; - struct vm_page *m; - int s; - - sf = dtosf(addr); - if (sf->refcnt == 0) - panic("sf_buf_free: freeing free sf_buf"); - sf->refcnt--; - if (sf->refcnt == 0) { - pmap_qremove((vm_offset_t)addr, 1); - m = sf->m; - s = splvm(); - vm_page_unwire(m, 0); - /* - * Check for the object going away on us. This can - * happen since we don't hold a reference to it. - * If so, we're responsible for freeing the page. - */ - if (m->wire_count == 0 && m->object == NULL) - vm_page_lock_queues(); - vm_page_free(m); - vm_page_unlock_queues(); - splx(s); - sf->m = NULL; - SLIST_INSERT_HEAD(&sf_freelist, sf, free_list); - if (sf_buf_alloc_want) { - sf_buf_alloc_want = 0; - wakeup(&sf_freelist); - } + unsigned int needed; + + if (pktlen == 0) + panic("%s: pktlen (%ld) must be non-zero\n", __func__, pktlen); + + /* + * Try to allocate for the whole thing. Since we want full control + * over the buffer size and be able to accept partial result, we can't + * use mbuf_allocpacket(). The logic below is similar to sosend(). + */ + *m = NULL; + if (pktlen > NBPG && jumbocl) { + needed = MIN(SENDFILE_MAX_16K, HOWMANY_16K(pktlen)); + *m = m_getpackets_internal(&needed, 1, how, 0, M16KCLBYTES); } + if (*m == NULL) { + needed = MIN(SENDFILE_MAX_4K, HOWMANY_4K(pktlen)); + *m = m_getpackets_internal(&needed, 1, how, 0, NBPG); + } + + /* + * Our previous attempt(s) at allocation had failed; the system + * may be short on mbufs, and we want to block until they are + * available. This time, ask just for 1 mbuf and don't return + * until we get it. + */ + if (*m == NULL) { + needed = 1; + *m = m_getpackets_internal(&needed, 1, M_WAIT, 1, NBPG); + } + if (*m == NULL) + panic("%s: blocking allocation returned NULL\n", __func__); + + *maxchunks = needed; } /* * sendfile(2). - * int sendfile(int fd, int s, off_t offset, size_t nbytes, - * struct sf_hdtr *hdtr, off_t *sbytes, int flags) + * int sendfile(int fd, int s, off_t offset, off_t *nbytes, + * struct sf_hdtr *hdtr, int flags) * * Send a file specified by 'fd' and starting at 'offset' to a socket - * specified by 's'. Send only 'nbytes' of the file or until EOF if - * nbytes == 0. Optionally add a header and/or trailer to the socket - * output. If specified, write the total number of bytes sent into *sbytes. + * specified by 's'. Send only '*nbytes' of the file or until EOF if + * *nbytes == 0. Optionally add a header and/or trailer to the socket + * output. If specified, write the total number of bytes sent into *nbytes. */ int -sendfile(struct proc *p, struct sendfile_args *uap) +sendfile(struct proc *p, struct sendfile_args *uap, __unused int *retval) { - struct file *fp; - struct filedesc *fdp = p->p_fd; + struct fileproc *fp; struct vnode *vp; - struct vm_object *obj; struct socket *so; - struct mbuf *m; - struct sf_buf *sf; - struct vm_page *pg; - struct writev_args nuap; + struct writev_nocancel_args nuap; + user_ssize_t writev_retval; struct sf_hdtr hdtr; - off_t off, xfsize, sbytes = 0; - int error = 0, s; - + struct user_sf_hdtr user_hdtr; + off_t off, xfsize; + off_t nbytes = 0, sbytes = 0; + int error = 0; + size_t sizeof_hdtr; + size_t size_of_iovec; + off_t file_size; + struct vfs_context context = *vfs_context_current(); + + KERNEL_DEBUG_CONSTANT((DBG_FNC_SENDFILE | DBG_FUNC_START), uap->s, + 0, 0, 0, 0); /* * Do argument checking. Must be a regular file in, stream * type and connected socket out, positive offset. */ - if (((u_int)uap->fd) >= fdp->fd_nfiles || - (fp = fdp->fd_ofiles[uap->fd]) == NULL || - (fp->f_flag & FREAD) == 0) { - error = EBADF; + if ((error = fp_getfvp(p, uap->fd, &fp, &vp))) { goto done; } - if (fp->f_type != DTYPE_VNODE) { - error = EINVAL; - goto done; + if ((fp->f_flag & FREAD) == 0) { + error = EBADF; + goto done1; } - vp = (struct vnode *)fp->f_data; - obj = vp->v_object; - if (vp->v_type != VREG || obj == NULL) { - error = EINVAL; - goto done; + if (vnode_isreg(vp) == 0) { + error = ENOTSUP; + goto done1; + } + error = file_socket(uap->s, &so); + if (error) { + goto done1; + } + if (so == NULL) { + error = EBADF; + goto done2; } - error = getsock(p->p_fd, uap->s, &fp); - if (error) - goto done; - so = (struct socket *)fp->f_data; if (so->so_type != SOCK_STREAM) { error = EINVAL; - goto done; + goto done2; } if ((so->so_state & SS_ISCONNECTED) == 0) { error = ENOTCONN; - goto done; + goto done2; } if (uap->offset < 0) { error = EINVAL; - goto done; + goto done2; } + if (uap->nbytes == USER_ADDR_NULL) { + error = EINVAL; + goto done2; + } + if (uap->flags != 0) { + error = EINVAL; + goto done2; + } + + context.vc_ucred = fp->f_fglob->fg_cred; + +#if CONFIG_MACF_SOCKET_SUBSET + /* JMM - fetch connected sockaddr? */ + error = mac_socket_check_send(context.vc_ucred, so, NULL); + if (error) + goto done2; +#endif + + /* + * Get number of bytes to send + * Should it applies to size of header and trailer? + * JMM - error handling? + */ + copyin(uap->nbytes, &nbytes, sizeof (off_t)); /* * If specified, get the pointer to the sf_hdtr struct for * any headers/trailers. */ - if (uap->hdtr != NULL) { - error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); + if (uap->hdtr != USER_ADDR_NULL) { + caddr_t hdtrp; + + bzero(&user_hdtr, sizeof (user_hdtr)); + if (IS_64BIT_PROCESS(p)) { + hdtrp = (caddr_t)&user_hdtr; + sizeof_hdtr = sizeof (user_hdtr); + size_of_iovec = sizeof (struct user_iovec); + } else { + hdtrp = (caddr_t)&hdtr; + sizeof_hdtr = sizeof (hdtr); + size_of_iovec = sizeof (struct iovec); + } + error = copyin(uap->hdtr, hdtrp, sizeof_hdtr); if (error) - goto done; + goto done2; + /* need to copy if user process is not 64-bit */ + if (!IS_64BIT_PROCESS(p)) { + user_hdtr.headers = CAST_USER_ADDR_T(hdtr.headers); + user_hdtr.hdr_cnt = hdtr.hdr_cnt; + user_hdtr.trailers = CAST_USER_ADDR_T(hdtr.trailers); + user_hdtr.trl_cnt = hdtr.trl_cnt; + } + /* * Send any headers. Wimp out and use writev(2). */ - if (hdtr.headers != NULL) { + if (user_hdtr.headers != USER_ADDR_NULL) { + bzero(&nuap, sizeof (struct writev_args)); nuap.fd = uap->s; - nuap.iovp = hdtr.headers; - nuap.iovcnt = hdtr.hdr_cnt; - error = writev(p, &nuap); + nuap.iovp = user_hdtr.headers; + nuap.iovcnt = user_hdtr.hdr_cnt; + error = writev_nocancel(p, &nuap, &writev_retval); if (error) - goto done; - sbytes += p->p_retval[0]; + goto done2; + sbytes += writev_retval; } } /* - * Protect against multiple writers to the socket. + * Get the file size for 2 reasons: + * 1. We don't want to allocate more mbufs than necessary + * 2. We don't want to read past the end of file */ - (void) sblock(&so->so_snd, M_WAIT); + if ((error = vnode_size(vp, &file_size, vfs_context_current())) != 0) + goto done2; /* - * Loop through the pages in the file, starting with the requested - * offset. Get a file page (do I/O if necessary), map the file page - * into an sf_buf, attach an mbuf header to the sf_buf, and queue - * it on the socket. + * Simply read file data into a chain of mbufs that used with scatter + * gather reads. We're not (yet?) setup to use zero copy external + * mbufs that point to the file pages. */ + socket_lock(so, 1); + error = sblock(&so->so_snd, M_WAIT); + if (error) { + socket_unlock(so, 1); + goto done2; + } for (off = uap->offset; ; off += xfsize, sbytes += xfsize) { - vm_object_offset_t pindex; - vm_object_offset_t pgoff; + mbuf_t m0 = NULL, m; + unsigned int nbufs = sendfileuiobufs, i; + uio_t auio; + char uio_buf[UIO_SIZEOF(sendfileuiobufs)]; /* 1 KB !!! */ + size_t uiolen; + user_ssize_t rlen; + off_t pgoff; + size_t pktlen; + boolean_t jumbocl; - pindex = OFF_TO_IDX(off); -retry_lookup: /* - * Calculate the amount to transfer. Not to exceed a page, + * Calculate the amount to transfer. + * Align to round number of pages. + * Not to exceed send socket buffer, * the EOF, or the passed in nbytes. */ - xfsize = obj->un_pager.vnp.vnp_size - off; - if (xfsize > PAGE_SIZE_64) - xfsize = PAGE_SIZE; - pgoff = (vm_object_offset_t)(off & PAGE_MASK_64); - if (PAGE_SIZE - pgoff < xfsize) + xfsize = sbspace(&so->so_snd); + + if (xfsize <= 0) { + if (so->so_state & SS_CANTSENDMORE) { + error = EPIPE; + goto done3; + } else if ((so->so_state & SS_NBIO)) { + error = EAGAIN; + goto done3; + } else { + xfsize = PAGE_SIZE; + } + } + + if (xfsize > SENDFILE_MAX_BYTES) + xfsize = SENDFILE_MAX_BYTES; + else if (xfsize > PAGE_SIZE) + xfsize = trunc_page(xfsize); + pgoff = off & PAGE_MASK_64; + if (pgoff > 0 && PAGE_SIZE - pgoff < xfsize) xfsize = PAGE_SIZE_64 - pgoff; - if (uap->nbytes && xfsize > (uap->nbytes - sbytes)) - xfsize = uap->nbytes - sbytes; + if (nbytes && xfsize > (nbytes - sbytes)) + xfsize = nbytes - sbytes; if (xfsize <= 0) break; + if (off + xfsize > file_size) + xfsize = file_size - off; + if (xfsize <= 0) + break; + /* - * Optimize the non-blocking case by looking at the socket space - * before going to the extra work of constituting the sf_buf. + * Attempt to use larger than system page-size clusters for + * large writes only if there is a jumbo cluster pool and + * if the socket is marked accordingly. */ - if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) { - if (so->so_state & SS_CANTSENDMORE) - error = EPIPE; - else - error = EAGAIN; - sbunlock(&so->so_snd); - goto done; + jumbocl = sosendjcl && njcl > 0 && + ((so->so_flags & SOF_MULTIPAGES) || sosendjcl_ignore_capab); + + socket_unlock(so, 0); + alloc_sendpkt(M_WAIT, xfsize, &nbufs, &m0, jumbocl); + pktlen = mbuf_pkt_maxlen(m0); + if (pktlen < xfsize) + xfsize = pktlen; + + auio = uio_createwithbuffer(nbufs, off, UIO_SYSSPACE, + UIO_READ, &uio_buf[0], sizeof (uio_buf)); + if (auio == NULL) { + //printf("sendfile: uio_createwithbuffer failed\n"); + mbuf_freem(m0); + error = ENXIO; + socket_lock(so, 0); + goto done3; } - /* - * Attempt to look up the page. If the page doesn't exist or the - * part we're interested in isn't valid, then read it from disk. - * If some other part of the kernel has this page (i.e. it's busy), - * then disk I/O may be occuring on it, so wait and retry. - */ - pg = vm_page_lookup(obj, pindex); - if (pg == NULL || (!(pg->flags & PG_BUSY) && !pg->busy && - !vm_page_is_valid(pg, pgoff, xfsize))) { - struct uio auio; - struct iovec aiov; - int bsize; - - if (pg == NULL) { - pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL); - if (pg == NULL) { - VM_WAIT; - goto retry_lookup; - } - /* - * don't just clear PG_BUSY manually - - * vm_page_alloc() should be considered opaque, - * use the VM routine provided to clear - * PG_BUSY. - */ - vm_page_wakeup(pg); + for (i = 0, m = m0, uiolen = 0; + i < nbufs && m != NULL && uiolen < xfsize; + i++, m = mbuf_next(m)) { + size_t mlen = mbuf_maxlen(m); + + if (mlen + uiolen > xfsize) + mlen = xfsize - uiolen; + mbuf_setlen(m, mlen); + uio_addiov(auio, CAST_USER_ADDR_T(mbuf_datastart(m)), + mlen); + uiolen += mlen; + } + + if (xfsize != uio_resid(auio)) + printf("sendfile: xfsize: %lld != uio_resid(auio): " + "%lld\n", xfsize, uio_resid(auio)); + + KERNEL_DEBUG_CONSTANT((DBG_FNC_SENDFILE_READ | DBG_FUNC_START), + uap->s, (unsigned int)((xfsize >> 32) & 0x0ffffffff), + (unsigned int)(xfsize & 0x0ffffffff), 0, 0); + error = fo_read(fp, auio, FOF_OFFSET, &context); + socket_lock(so, 0); + if (error != 0) { + if (uio_resid(auio) != xfsize && (error == ERESTART || + error == EINTR || error == EWOULDBLOCK)) { + error = 0; + } else { + mbuf_freem(m0); + goto done3; } - /* - * Ensure that our page is still around when the I/O completes. - */ - vm_page_io_start(pg); - vm_page_wire(pg); - /* - * Get the page from backing store. - */ - bsize = vp->v_mount->mnt_stat.f_iosize; - auio.uio_iov = &aiov; - auio.uio_iovcnt = 1; - aiov.iov_base = 0; - aiov.iov_len = MAXBSIZE; - auio.uio_resid = MAXBSIZE; - auio.uio_offset = trunc_page(off); - auio.uio_segflg = UIO_NOCOPY; - auio.uio_rw = UIO_READ; - auio.uio_procp = p; - vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, p); - error = VOP_READ(vp, &auio, IO_VMIO | ((MAXBSIZE / bsize) << 16), - p->p_ucred); - VOP_UNLOCK(vp, 0, p); - vm_page_flag_clear(pg, PG_ZERO); - vm_page_io_finish(pg); - if (error) { - vm_page_unwire(pg, 0); - /* - * See if anyone else might know about this page. - * If not and it is not valid, then free it. - */ - if (pg->wire_count == 0 && pg->valid == 0 && - pg->busy == 0 && !(pg->flags & PG_BUSY) && - pg->hold_count == 0) - vm_page_lock_queues(); - vm_page_free(pg); - vm_page_unlock_queues(); - sbunlock(&so->so_snd); - goto done; - } - } else { - if ((pg->flags & PG_BUSY) || pg->busy) { - s = splvm(); - if ((pg->flags & PG_BUSY) || pg->busy) { - /* - * Page is busy. Wait and retry. - */ - vm_page_flag_set(pg, PG_WANTED); - tsleep(pg, PVM, "sfpbsy", 0); - splx(s); - goto retry_lookup; - } - splx(s); - } - /* - * Protect from having the page ripped out from beneath us. - */ - vm_page_wire(pg); } - /* - * Allocate a kernel virtual page and insert the physical page - * into it. - */ - sf = sf_buf_alloc(); - sf->m = pg; - pmap_qenter(sf->kva, &pg, 1); - /* - * Get an mbuf header and set it up as having external storage. - */ - MGETHDR(m, M_WAIT, MT_DATA); - m->m_ext.ext_free = sf_buf_free; - m->m_ext.ext_ref = sf_buf_ref; - m->m_ext.ext_buf = (void *)sf->kva; - m->m_ext.ext_size = PAGE_SIZE; - m->m_data = (char *) sf->kva + pgoff; - m->m_flags |= M_EXT; - m->m_pkthdr.len = m->m_len = xfsize; - /* - * Add the buffer to the socket buffer chain. - */ - s = splnet(); + xfsize -= uio_resid(auio); + KERNEL_DEBUG_CONSTANT((DBG_FNC_SENDFILE_READ | DBG_FUNC_END), + uap->s, (unsigned int)((xfsize >> 32) & 0x0ffffffff), + (unsigned int)(xfsize & 0x0ffffffff), 0, 0); + + if (xfsize == 0) { + //printf("sendfile: fo_read 0 bytes, EOF\n"); + break; + } + if (xfsize + off > file_size) + printf("sendfile: xfsize: %lld + off: %lld > file_size:" + "%lld\n", xfsize, off, file_size); + for (i = 0, m = m0, rlen = 0; + i < nbufs && m != NULL && rlen < xfsize; + i++, m = mbuf_next(m)) { + size_t mlen = mbuf_maxlen(m); + + if (rlen + mlen > xfsize) + mlen = xfsize - rlen; + mbuf_setlen(m, mlen); + + rlen += mlen; + } + mbuf_pkthdr_setlen(m0, xfsize); + retry_space: /* * Make sure that the socket is still able to take more data. @@ -1940,65 +2291,130 @@ retry_space: error = so->so_error; so->so_error = 0; } - m_freem(m); - sbunlock(&so->so_snd); - splx(s); - goto done; + m_freem(m0); + goto done3; } /* * Wait for socket space to become available. We do this just * after checking the connection state above in order to avoid * a race condition with sbwait(). */ - if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) { + if (sbspace(&so->so_snd) < (long)so->so_snd.sb_lowat) { if (so->so_state & SS_NBIO) { - m_freem(m); - sbunlock(&so->so_snd); - splx(s); + m_freem(m0); error = EAGAIN; - goto done; + goto done3; } + KERNEL_DEBUG_CONSTANT((DBG_FNC_SENDFILE_WAIT | + DBG_FUNC_START), uap->s, 0, 0, 0, 0); error = sbwait(&so->so_snd); + KERNEL_DEBUG_CONSTANT((DBG_FNC_SENDFILE_WAIT| + DBG_FUNC_END), uap->s, 0, 0, 0, 0); /* * An error from sbwait usually indicates that we've * been interrupted by a signal. If we've sent anything * then return bytes sent, otherwise return the error. */ if (error) { - m_freem(m); - sbunlock(&so->so_snd); - splx(s); - goto done; + m_freem(m0); + goto done3; } goto retry_space; } - error = (*so->so_proto->pr_usrreqs->pru_send)(so, 0, m, 0, 0, p); - splx(s); + { + /* + * Socket filter processing + */ + struct socket_filter_entry *filter; + int filtered = 0; + struct mbuf *control = NULL; + boolean_t recursive = (so->so_send_filt_thread != NULL); + + error = 0; + for (filter = so->so_filt; filter && (error == 0); + filter = filter->sfe_next_onsocket) { + if (filter->sfe_filter->sf_filter.sf_data_out) { + if (filtered == 0) { + filtered = 1; + so->so_send_filt_thread = + current_thread(); + sflt_use(so); + socket_unlock(so, 0); + } + error = filter->sfe_filter->sf_filter. + sf_data_out(filter->sfe_cookie, so, + NULL, &m0, &control, 0); + } + } + + if (filtered) { + /* + * At this point, we've run at least one filter. + * The socket is unlocked as is the socket + * buffer. Clear the recorded filter thread + * only when we are outside of a filter's + * context. This allows for a filter to issue + * multiple inject calls from its sf_data_out + * callback routine. + */ + socket_lock(so, 0); + sflt_unuse(so); + if (!recursive) + so->so_send_filt_thread = 0; + if (error) { + if (error == EJUSTRETURN) { + error = 0; + continue; + } + goto done3; + } + } + /* + * End Socket filter processing + */ + } + KERNEL_DEBUG_CONSTANT((DBG_FNC_SENDFILE_SEND | DBG_FUNC_START), + uap->s, 0, 0, 0, 0); + error = (*so->so_proto->pr_usrreqs->pru_send)(so, 0, m0, + 0, 0, p); + KERNEL_DEBUG_CONSTANT((DBG_FNC_SENDFILE_SEND | DBG_FUNC_START), + uap->s, 0, 0, 0, 0); if (error) { - sbunlock(&so->so_snd); - goto done; + goto done3; } } - sbunlock(&so->so_snd); - + sbunlock(&so->so_snd, 0); /* will unlock socket */ /* * Send trailers. Wimp out and use writev(2). */ - if (uap->hdtr != NULL && hdtr.trailers != NULL) { - nuap.fd = uap->s; - nuap.iovp = hdtr.trailers; - nuap.iovcnt = hdtr.trl_cnt; - error = writev(p, &nuap); - if (error) - goto done; - sbytes += p->p_retval[0]; + if (uap->hdtr != USER_ADDR_NULL && + user_hdtr.trailers != USER_ADDR_NULL) { + bzero(&nuap, sizeof (struct writev_args)); + nuap.fd = uap->s; + nuap.iovp = user_hdtr.trailers; + nuap.iovcnt = user_hdtr.trl_cnt; + error = writev_nocancel(p, &nuap, &writev_retval); + if (error) + goto done2; + sbytes += writev_retval; } - +done2: + file_drop(uap->s); +done1: + file_drop(uap->fd); done: - if (uap->sbytes != NULL) { - copyout(&sbytes, uap->sbytes, sizeof(off_t)); + if (uap->nbytes != USER_ADDR_NULL) { + /* XXX this appears bogus for some early failure conditions */ + copyout(&sbytes, uap->nbytes, sizeof (off_t)); } + KERNEL_DEBUG_CONSTANT((DBG_FNC_SENDFILE | DBG_FUNC_END), uap->s, + (unsigned int)((sbytes >> 32) & 0x0ffffffff), + (unsigned int)(sbytes & 0x0ffffffff), error, 0); return (error); +done3: + sbunlock(&so->so_snd, 0); /* will unlock socket */ + goto done2; } -#endif + +#endif /* SENDFILE */