src/socket_wrapper.c: implement recvmmsg and sendmmsg
[socket_wrapper.git] / src / socket_wrapper.c
index 61dce978ce64c3242e8c3b1c92f8f443487fd631..90ffc25805495b62adeed4147b59f354e5d3699d 100644 (file)
@@ -2,8 +2,8 @@
  * BSD 3-Clause License
  *
  * Copyright (c) 2005-2008, Jelmer Vernooij <jelmer@samba.org>
- * Copyright (c) 2006-2018, Stefan Metzmacher <metze@samba.org>
- * Copyright (c) 2013-2018, Andreas Schneider <asn@samba.org>
+ * Copyright (c) 2006-2021, Stefan Metzmacher <metze@samba.org>
+ * Copyright (c) 2013-2021, Andreas Schneider <asn@samba.org>
  * Copyright (c) 2014-2017, Michael Adam <obnox@samba.org>
  * Copyright (c) 2016-2018, Anoop C S <anoopcs@redhat.com>
  * All rights reserved.
@@ -86,6 +86,8 @@
 #endif
 #include <pthread.h>
 
+#include "socket_wrapper.h"
+
 enum swrap_dbglvl_e {
        SWRAP_LOG_ERROR = 0,
        SWRAP_LOG_WARN,
@@ -370,7 +372,7 @@ static pthread_mutex_t autobind_start_mutex = PTHREAD_MUTEX_INITIALIZER;
 /* Mutex to guard the initialization of array of socket_info structures */
 static pthread_mutex_t sockets_mutex = PTHREAD_MUTEX_INITIALIZER;
 
-/* Mutex to guard the socket reset in swrap_close() and swrap_remove_stale() */
+/* Mutex to guard the socket reset in swrap_remove_wrapper() */
 static pthread_mutex_t socket_reset_mutex = PTHREAD_MUTEX_INITIALIZER;
 
 /* Mutex to synchronize access to first free index in socket_info array */
@@ -392,8 +394,6 @@ static pthread_mutex_t mtu_update_mutex = PTHREAD_MUTEX_INITIALIZER;
 
 /* Function prototypes */
 
-bool socket_wrapper_enabled(void);
-
 #if ! defined(HAVE_CONSTRUCTOR_ATTRIBUTE) && defined(HAVE_PRAGMA_INIT)
 /* xlC and other oldschool compilers support (only) this */
 #pragma init (swrap_constructor)
@@ -492,6 +492,9 @@ typedef int (*__libc_bind)(int sockfd,
                           const struct sockaddr *addr,
                           socklen_t addrlen);
 typedef int (*__libc_close)(int fd);
+#ifdef HAVE___CLOSE_NOCANCEL
+typedef int (*__libc___close_nocancel)(int fd);
+#endif
 typedef int (*__libc_connect)(int sockfd,
                              const struct sockaddr *addr,
                              socklen_t addrlen);
@@ -534,8 +537,29 @@ typedef int (*__libc_recvfrom)(int sockfd,
                             struct sockaddr *src_addr,
                             socklen_t *addrlen);
 typedef int (*__libc_recvmsg)(int sockfd, const struct msghdr *msg, int flags);
+#ifdef HAVE_RECVMMSG
+#if defined(HAVE_RECVMMSG_SSIZE_T_CONST_TIMEOUT)
+/* FreeBSD */
+typedef ssize_t (*__libc_recvmmsg)(int sockfd, struct mmsghdr *msgvec, size_t vlen, int flags, const struct timespec *timeout);
+#elif defined(HAVE_RECVMMSG_CONST_TIMEOUT)
+/* Linux legacy glibc < 2.21 */
+typedef int (*__libc_recvmmsg)(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, int flags, const struct timespec *timeout);
+#else
+/* Linux glibc >= 2.21 */
+typedef int (*__libc_recvmmsg)(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, int flags, struct timespec *timeout);
+#endif
+#endif /* HAVE_RECVMMSG */
 typedef int (*__libc_send)(int sockfd, const void *buf, size_t len, int flags);
 typedef int (*__libc_sendmsg)(int sockfd, const struct msghdr *msg, int flags);
+#ifdef HAVE_SENDMMSG
+#if defined(HAVE_SENDMMSG_SSIZE_T)
+/* FreeBSD */
+typedef ssize_t (*__libc_sendmmsg)(int sockfd, struct mmsghdr *msgvec, size_t vlen, int flags);
+#else
+/* Linux */
+typedef int (*__libc_sendmmsg)(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, int flags);
+#endif
+#endif /* HAVE_SENDMMSG */
 typedef int (*__libc_sendto)(int sockfd,
                           const void *buf,
                           size_t len,
@@ -572,6 +596,9 @@ struct swrap_libc_symbols {
 #endif
        SWRAP_SYMBOL_ENTRY(bind);
        SWRAP_SYMBOL_ENTRY(close);
+#ifdef HAVE___CLOSE_NOCANCEL
+       SWRAP_SYMBOL_ENTRY(__close_nocancel);
+#endif
        SWRAP_SYMBOL_ENTRY(connect);
        SWRAP_SYMBOL_ENTRY(dup);
        SWRAP_SYMBOL_ENTRY(dup2);
@@ -599,8 +626,14 @@ struct swrap_libc_symbols {
        SWRAP_SYMBOL_ENTRY(recv);
        SWRAP_SYMBOL_ENTRY(recvfrom);
        SWRAP_SYMBOL_ENTRY(recvmsg);
+#ifdef HAVE_RECVMMSG
+       SWRAP_SYMBOL_ENTRY(recvmmsg);
+#endif
        SWRAP_SYMBOL_ENTRY(send);
        SWRAP_SYMBOL_ENTRY(sendmsg);
+#ifdef HAVE_SENDMMSG
+       SWRAP_SYMBOL_ENTRY(sendmmsg);
+#endif
        SWRAP_SYMBOL_ENTRY(sendto);
        SWRAP_SYMBOL_ENTRY(setsockopt);
 #ifdef HAVE_SIGNALFD
@@ -851,6 +884,15 @@ static int libc_close(int fd)
        return swrap.libc.symbols._libc_close.f(fd);
 }
 
+#ifdef HAVE___CLOSE_NOCANCEL
+static int libc___close_nocancel(int fd)
+{
+       swrap_bind_symbol_all();
+
+       return swrap.libc.symbols._libc___close_nocancel.f(fd);
+}
+#endif /* HAVE___CLOSE_NOCANCEL */
+
 static int libc_connect(int sockfd,
                        const struct sockaddr *addr,
                        socklen_t addrlen)
@@ -969,6 +1011,19 @@ static FILE *libc_fopen64(const char *name, const char *mode)
 }
 #endif /* HAVE_FOPEN64 */
 
+static void swrap_inject_o_largefile(int *flags)
+{
+       (void)*flags; /* maybe unused */
+#if SIZE_MAX == 0xffffffffUL && defined(O_LARGEFILE)
+#ifdef O_PATH
+       if (((*flags) & O_PATH) == 0)
+#endif
+       {
+               *flags |= O_LARGEFILE;
+       }
+#endif
+}
+
 static int libc_vopen(const char *pathname, int flags, va_list ap)
 {
        int mode = 0;
@@ -976,6 +1031,8 @@ static int libc_vopen(const char *pathname, int flags, va_list ap)
 
        swrap_bind_symbol_all();
 
+       swrap_inject_o_largefile(&flags);
+
        if (flags & O_CREAT) {
                mode = va_arg(ap, int);
        }
@@ -1004,6 +1061,8 @@ static int libc_vopen64(const char *pathname, int flags, va_list ap)
 
        swrap_bind_symbol_all();
 
+       swrap_inject_o_largefile(&flags);
+
        if (flags & O_CREAT) {
                mode = va_arg(ap, int);
        }
@@ -1020,6 +1079,8 @@ static int libc_vopenat(int dirfd, const char *path, int flags, va_list ap)
 
        swrap_bind_symbol_all();
 
+       swrap_inject_o_largefile(&flags);
+
        if (flags & O_CREAT) {
                mode = va_arg(ap, int);
        }
@@ -1097,6 +1158,24 @@ static int libc_recvmsg(int sockfd, struct msghdr *msg, int flags)
        return swrap.libc.symbols._libc_recvmsg.f(sockfd, msg, flags);
 }
 
+#ifdef HAVE_RECVMMSG
+#if defined(HAVE_RECVMMSG_SSIZE_T_CONST_TIMEOUT)
+/* FreeBSD */
+static ssize_t libc_recvmmsg(int sockfd, struct mmsghdr *msgvec, size_t vlen, int flags, const struct timespec *timeout)
+#elif defined(HAVE_RECVMMSG_CONST_TIMEOUT)
+/* Linux legacy glibc < 2.21 */
+static int libc_recvmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, int flags, const struct timespec *timeout)
+#else
+/* Linux glibc >= 2.21 */
+static int libc_recvmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, int flags, struct timespec *timeout)
+#endif
+{
+       swrap_bind_symbol_all();
+
+       return swrap.libc.symbols._libc_recvmmsg.f(sockfd, msgvec, vlen, flags, timeout);
+}
+#endif
+
 static int libc_send(int sockfd, const void *buf, size_t len, int flags)
 {
        swrap_bind_symbol_all();
@@ -1111,6 +1190,21 @@ static int libc_sendmsg(int sockfd, const struct msghdr *msg, int flags)
        return swrap.libc.symbols._libc_sendmsg.f(sockfd, msg, flags);
 }
 
+#ifdef HAVE_SENDMMSG
+#if defined(HAVE_SENDMMSG_SSIZE_T)
+/* FreeBSD */
+static ssize_t libc_sendmmsg(int sockfd, struct mmsghdr *msgvec, size_t vlen, int flags)
+#else
+/* Linux */
+static int libc_sendmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, int flags)
+#endif
+{
+       swrap_bind_symbol_all();
+
+       return swrap.libc.symbols._libc_sendmmsg.f(sockfd, msgvec, vlen, flags);
+}
+#endif
+
 static int libc_sendto(int sockfd,
                       const void *buf,
                       size_t len,
@@ -1199,6 +1293,9 @@ static void __swrap_bind_symbol_all_once(void)
 #endif
        swrap_bind_symbol_libsocket(bind);
        swrap_bind_symbol_libc(close);
+#ifdef HAVE___CLOSE_NOCANCEL
+       swrap_bind_symbol_libc(__close_nocancel);
+#endif
        swrap_bind_symbol_libsocket(connect);
        swrap_bind_symbol_libc(dup);
        swrap_bind_symbol_libc(dup2);
@@ -1226,8 +1323,14 @@ static void __swrap_bind_symbol_all_once(void)
        swrap_bind_symbol_libsocket(recv);
        swrap_bind_symbol_libsocket(recvfrom);
        swrap_bind_symbol_libsocket(recvmsg);
+#ifdef HAVE_RECVMMSG
+       swrap_bind_symbol_libsocket(recvmmsg);
+#endif
        swrap_bind_symbol_libsocket(send);
        swrap_bind_symbol_libsocket(sendmsg);
+#ifdef HAVE_SENDMMSG
+       swrap_bind_symbol_libsocket(sendmmsg);
+#endif
        swrap_bind_symbol_libsocket(sendto);
        swrap_bind_symbol_libsocket(setsockopt);
 #ifdef HAVE_SIGNALFD
@@ -1392,6 +1495,55 @@ static size_t socket_length(int family)
        return 0;
 }
 
+struct swrap_sockaddr_buf {
+       char str[128];
+};
+
+static const char *swrap_sockaddr_string(struct swrap_sockaddr_buf *buf,
+                                        const struct sockaddr *saddr)
+{
+       unsigned int port = 0;
+       char addr[64] = {0,};
+
+       switch (saddr->sa_family) {
+       case AF_INET: {
+               const struct sockaddr_in *in =
+                   (const struct sockaddr_in *)(const void *)saddr;
+
+               port = ntohs(in->sin_port);
+
+               inet_ntop(saddr->sa_family,
+                         &in->sin_addr,
+                         addr, sizeof(addr));
+               break;
+       }
+#ifdef HAVE_IPV6
+       case AF_INET6: {
+               const struct sockaddr_in6 *in6 =
+                   (const struct sockaddr_in6 *)(const void *)saddr;
+
+               port = ntohs(in6->sin6_port);
+
+               inet_ntop(saddr->sa_family,
+                         &in6->sin6_addr,
+                         addr, sizeof(addr));
+               break;
+       }
+#endif
+       default:
+               snprintf(addr, sizeof(addr),
+                        "<Unknown address family %u>",
+                        saddr->sa_family);
+               break;
+       }
+
+       snprintf(buf->str, sizeof(buf->str),
+                "addr[%s]/port[%u]",
+                addr, port);
+
+       return buf->str;
+}
+
 static struct socket_info *swrap_get_socket_info(int si_index)
 {
        return (struct socket_info *)(&(sockets[si_index].info));
@@ -2027,13 +2179,10 @@ static int convert_in_un_remote(struct socket_info *si, const struct sockaddr *i
                        type = u_type;
                        iface = (addr & 0x000000FF);
                } else {
-                       char str[256] = {0,};
-                       inet_ntop(inaddr->sa_family,
-                                 &in->sin_addr,
-                                 str, sizeof(str));
+                       struct swrap_sockaddr_buf buf = {};
                        SWRAP_LOG(SWRAP_LOG_WARN,
-                                 "str[%s] prt[%u]",
-                                 str, (unsigned)prt);
+                                 "%s",
+                                 swrap_sockaddr_string(&buf, inaddr));
                        errno = ENETUNREACH;
                        return -1;
                }
@@ -2069,13 +2218,10 @@ static int convert_in_un_remote(struct socket_info *si, const struct sockaddr *i
                if (IN6_ARE_ADDR_EQUAL(&cmp1, &cmp2)) {
                        iface = in->sin6_addr.s6_addr[15];
                } else {
-                       char str[256] = {0,};
-                       inet_ntop(inaddr->sa_family,
-                                 &in->sin6_addr,
-                                 str, sizeof(str));
+                       struct swrap_sockaddr_buf buf = {};
                        SWRAP_LOG(SWRAP_LOG_WARN,
-                                 "str[%s] prt[%u]",
-                                 str, (unsigned)prt);
+                                 "%s",
+                                 swrap_sockaddr_string(&buf, inaddr));
                        errno = ENETUNREACH;
                        return -1;
                }
@@ -2404,46 +2550,7 @@ static bool check_addr_port_in_use(const struct sockaddr *sa, socklen_t len)
 }
 #endif
 
-static void swrap_remove_stale(int fd)
-{
-       struct socket_info *si;
-       int si_index;
-
-       SWRAP_LOG(SWRAP_LOG_TRACE, "remove stale wrapper for %d", fd);
-
-       swrap_mutex_lock(&socket_reset_mutex);
-
-       si_index = find_socket_info_index(fd);
-       if (si_index == -1) {
-               swrap_mutex_unlock(&socket_reset_mutex);
-               return;
-       }
-
-       reset_socket_info_index(fd);
-
-       si = swrap_get_socket_info(si_index);
-
-       swrap_mutex_lock(&first_free_mutex);
-       SWRAP_LOCK_SI(si);
-
-       swrap_dec_refcount(si);
-
-       if (swrap_get_refcount(si) > 0) {
-               goto out;
-       }
-
-       if (si->un_addr.sun_path[0] != '\0') {
-               unlink(si->un_addr.sun_path);
-       }
-
-       swrap_set_next_free(si, first_free);
-       first_free = si_index;
-
-out:
-       SWRAP_UNLOCK_SI(si);
-       swrap_mutex_unlock(&first_free_mutex);
-       swrap_mutex_unlock(&socket_reset_mutex);
-}
+static void swrap_remove_stale(int fd);
 
 static int sockaddr_convert_to_un(struct socket_info *si,
                                  const struct sockaddr *in_addr,
@@ -3836,7 +3943,6 @@ static int swrap_auto_bind(int fd, struct socket_info *si, int family)
        char type;
        int ret;
        int port;
-       struct stat st;
        char *swrap_dir = NULL;
 
        swrap_mutex_lock(&autobind_start_mutex);
@@ -3937,10 +4043,12 @@ static int swrap_auto_bind(int fd, struct socket_info *si, int family)
                              type,
                              socket_wrapper_default_iface(),
                              port);
-               if (stat(un_addr.sa.un.sun_path, &st) == 0) continue;
 
                ret = libc_bind(fd, &un_addr.sa.s, un_addr.sa_socklen);
                if (ret == -1) {
+                       if (errno == EALREADY || errno == EADDRINUSE) {
+                               continue;
+                       }
                        goto done;
                }
 
@@ -3985,6 +4093,7 @@ static int swrap_connect(int s, const struct sockaddr *serv_addr,
                .sa_socklen = sizeof(struct sockaddr_un),
        };
        struct socket_info *si = find_socket_info(s);
+       struct swrap_sockaddr_buf buf = {};
        int bcast = 0;
 
        if (!si) {
@@ -4033,7 +4142,8 @@ static int swrap_connect(int s, const struct sockaddr *serv_addr,
        }
 
        SWRAP_LOG(SWRAP_LOG_TRACE,
-                 "connect() path=%s, fd=%d",
+                 "connect(%s) path=%s, fd=%d",
+                 swrap_sockaddr_string(&buf, serv_addr),
                  un_addr.sa.un.sun_path, s);
 
 
@@ -4099,6 +4209,8 @@ static int swrap_bind(int s, const struct sockaddr *myaddr, socklen_t addrlen)
                .sa_socklen = sizeof(struct sockaddr_un),
        };
        struct socket_info *si = find_socket_info(s);
+       struct swrap_sockaddr_buf buf = {};
+       int ret_errno = errno;
        int bind_error = 0;
 #if 0 /* FIXME */
        bool in_use;
@@ -4156,7 +4268,7 @@ static int swrap_bind(int s, const struct sockaddr *myaddr, socklen_t addrlen)
        }
 
        if (bind_error != 0) {
-               errno = bind_error;
+               ret_errno = bind_error;
                ret = -1;
                goto out;
        }
@@ -4180,16 +4292,21 @@ static int swrap_bind(int s, const struct sockaddr *myaddr, socklen_t addrlen)
                                     1,
                                     &si->bcast);
        if (ret == -1) {
+               ret_errno = errno;
                goto out;
        }
 
        unlink(un_addr.sa.un.sun_path);
 
        ret = libc_bind(s, &un_addr.sa.s, un_addr.sa_socklen);
+       if (ret == -1) {
+               ret_errno = errno;
+       }
 
        SWRAP_LOG(SWRAP_LOG_TRACE,
-                 "bind() path=%s, fd=%d",
-                 un_addr.sa.un.sun_path, s);
+                 "bind(%s) path=%s, fd=%d ret=%d ret_errno=%d",
+                 swrap_sockaddr_string(&buf, myaddr),
+                 un_addr.sa.un.sun_path, s, ret, ret_errno);
 
        if (ret == 0) {
                si->bound = 1;
@@ -4197,7 +4314,7 @@ static int swrap_bind(int s, const struct sockaddr *myaddr, socklen_t addrlen)
 
 out:
        SWRAP_UNLOCK_SI(si);
-
+       errno = ret_errno;
        return ret;
 }
 
@@ -5345,7 +5462,7 @@ union __swrap_cmsghdr {
        struct cmsghdr *cmsg;
 };
 
-static int swrap_sendmsg_unix_scm_rights(const struct cmsghdr *cmsg,
+static int swrap_sendmsg_unix_scm_rights(struct cmsghdr *cmsg,
                                         uint8_t **cm_data,
                                         size_t *cm_data_space,
                                         int *scm_rights_pipe_fd)
@@ -5577,7 +5694,7 @@ static int swrap_sendmsg_unix_scm_rights(const struct cmsghdr *cmsg,
        return 0;
 }
 
-static int swrap_sendmsg_unix_sol_socket(const struct cmsghdr *cmsg,
+static int swrap_sendmsg_unix_sol_socket(struct cmsghdr *cmsg,
                                         uint8_t **cm_data,
                                         size_t *cm_data_space,
                                         int *scm_rights_pipe_fd)
@@ -5601,7 +5718,7 @@ static int swrap_sendmsg_unix_sol_socket(const struct cmsghdr *cmsg,
        return rc;
 }
 
-static int swrap_recvmsg_unix_scm_rights(const struct cmsghdr *cmsg,
+static int swrap_recvmsg_unix_scm_rights(struct cmsghdr *cmsg,
                                         uint8_t **cm_data,
                                         size_t *cm_data_space)
 {
@@ -5880,7 +5997,7 @@ static int swrap_recvmsg_unix_scm_rights(const struct cmsghdr *cmsg,
        return 0;
 }
 
-static int swrap_recvmsg_unix_sol_socket(const struct cmsghdr *cmsg,
+static int swrap_recvmsg_unix_sol_socket(struct cmsghdr *cmsg,
                                         uint8_t **cm_data,
                                         size_t *cm_data_space)
 {
@@ -6121,6 +6238,7 @@ static ssize_t swrap_sendmsg_before(int fd,
 {
        size_t i, len = 0;
        ssize_t ret = -1;
+       struct swrap_sockaddr_buf buf = {};
 
        if (to_un) {
                *to_un = NULL;
@@ -6182,6 +6300,10 @@ static ssize_t swrap_sendmsg_before(int fd,
                                msg->msg_name = NULL;
                                msg->msg_namelen = 0;
                        }
+                       SWRAP_LOG(SWRAP_LOG_TRACE,
+                                 "connected(%s) fd=%d",
+                                 swrap_sockaddr_string(&buf, &si->peername.sa.s),
+                                 fd);
                } else {
                        const struct sockaddr *msg_name;
                        msg_name = (const struct sockaddr *)msg->msg_name;
@@ -6236,6 +6358,11 @@ static ssize_t swrap_sendmsg_before(int fd,
                        goto out;
                }
 
+               SWRAP_LOG(SWRAP_LOG_TRACE,
+                         "deferred connect(%s) path=%s, fd=%d",
+                         swrap_sockaddr_string(&buf, &si->peername.sa.s),
+                         tmp_un->sun_path, fd);
+
                ret = libc_connect(fd,
                                   (struct sockaddr *)(void *)tmp_un,
                                   sizeof(*tmp_un));
@@ -6306,9 +6433,11 @@ static void swrap_sendmsg_after(int fd,
 
        for (i = 0; i < (size_t)msg->msg_iovlen; i++) {
                size_t this_time = MIN(remain, (size_t)msg->msg_iov[i].iov_len);
-               memcpy(buf + ofs,
-                      msg->msg_iov[i].iov_base,
-                      this_time);
+               if (this_time > 0) {
+                       memcpy(buf + ofs,
+                              msg->msg_iov[i].iov_base,
+                              this_time);
+               }
                ofs += this_time;
                remain -= this_time;
        }
@@ -7134,6 +7263,219 @@ ssize_t recvmsg(int sockfd, struct msghdr *msg, int flags)
        return swrap_recvmsg(sockfd, msg, flags);
 }
 
+/****************************************************************************
+ *   RECVMMSG
+ ***************************************************************************/
+
+#ifdef HAVE_RECVMMSG
+#if defined(HAVE_RECVMMSG_SSIZE_T_CONST_TIMEOUT)
+/* FreeBSD */
+static ssize_t swrap_recvmmsg(int s, struct mmsghdr *omsgvec, size_t _vlen, int flags, const struct timespec *timeout)
+#elif defined(HAVE_RECVMMSG_CONST_TIMEOUT)
+/* Linux legacy glibc < 2.21 */
+static int swrap_recvmmsg(int s, struct mmsghdr *omsgvec, unsigned int _vlen, int flags, const struct timespec *timeout)
+#else
+/* Linux glibc >= 2.21 */
+static int swrap_recvmmsg(int s, struct mmsghdr *omsgvec, unsigned int _vlen, int flags, struct timespec *timeout)
+#endif
+{
+       struct socket_info *si = find_socket_info(s);
+#define __SWRAP_RECVMMSG_MAX_VLEN 16
+       struct mmsghdr msgvec[__SWRAP_RECVMMSG_MAX_VLEN] = {};
+       struct {
+               struct iovec iov;
+               struct swrap_address from_addr;
+               struct swrap_address convert_addr;
+#ifdef HAVE_STRUCT_MSGHDR_MSG_CONTROL
+               size_t msg_ctrllen_filled;
+               size_t msg_ctrllen_left;
+#endif
+       } tmp[__SWRAP_RECVMMSG_MAX_VLEN] = {};
+       int vlen;
+       int i;
+       int ret;
+       int rc;
+       int saved_errno;
+
+       if (_vlen > __SWRAP_RECVMMSG_MAX_VLEN) {
+               vlen = __SWRAP_RECVMMSG_MAX_VLEN;
+       } else {
+               vlen = _vlen;
+       }
+
+       if (si == NULL) {
+               uint8_t *tmp_control[__SWRAP_RECVMMSG_MAX_VLEN] = { NULL, };
+
+               for (i = 0; i < vlen; i++) {
+                       struct msghdr *omsg = &omsgvec[i].msg_hdr;
+                       struct msghdr *msg = &msgvec[i].msg_hdr;
+
+                       rc = swrap_recvmsg_before_unix(omsg, msg,
+                                                      &tmp_control[i]);
+                       if (rc < 0) {
+                               ret = rc;
+                               goto fail_libc;
+                       }
+               }
+
+               ret = libc_recvmmsg(s, msgvec, vlen, flags, timeout);
+               if (ret < 0) {
+                       goto fail_libc;
+               }
+
+               for (i = 0; i < ret; i++) {
+                       omsgvec[i].msg_len = msgvec[i].msg_len;
+               }
+
+fail_libc:
+               saved_errno = errno;
+               for (i = 0; i < vlen; i++) {
+                       struct msghdr *omsg = &omsgvec[i].msg_hdr;
+                       struct msghdr *msg = &msgvec[i].msg_hdr;
+
+                       if (i == 0 || i < ret) {
+                               swrap_recvmsg_after_unix(msg, &tmp_control[i], omsg, ret);
+                       }
+                       SAFE_FREE(tmp_control[i]);
+               }
+               errno = saved_errno;
+
+               return ret;
+       }
+
+       for (i = 0; i < vlen; i++) {
+               struct msghdr *omsg = &omsgvec[i].msg_hdr;
+               struct msghdr *msg = &msgvec[i].msg_hdr;
+
+               tmp[i].from_addr.sa_socklen = sizeof(struct sockaddr_un);
+               tmp[i].convert_addr.sa_socklen = sizeof(struct sockaddr_storage);
+
+               msg->msg_name = &tmp[i].from_addr.sa;              /* optional address */
+               msg->msg_namelen = tmp[i].from_addr.sa_socklen;    /* size of address */
+               msg->msg_iov = omsg->msg_iov;               /* scatter/gather array */
+               msg->msg_iovlen = omsg->msg_iovlen;         /* # elements in msg_iov */
+#ifdef HAVE_STRUCT_MSGHDR_MSG_CONTROL
+               tmp[i].msg_ctrllen_filled = 0;
+               tmp[i].msg_ctrllen_left = omsg->msg_controllen;
+
+               msg->msg_control = omsg->msg_control;       /* ancillary data, see below */
+               msg->msg_controllen = omsg->msg_controllen; /* ancillary data buffer len */
+               msg->msg_flags = omsg->msg_flags;           /* flags on received message */
+#endif
+
+               rc = swrap_recvmsg_before(s, si, msg, &tmp[i].iov);
+               if (rc < 0) {
+                       ret = rc;
+                       goto fail_swrap;
+               }
+       }
+
+       ret = libc_recvmmsg(s, msgvec, vlen, flags, timeout);
+       if (ret < 0) {
+               goto fail_swrap;
+       }
+
+       for (i = 0; i < ret; i++) {
+               omsgvec[i].msg_len = msgvec[i].msg_len;
+       }
+
+fail_swrap:
+
+       saved_errno = errno;
+       for (i = 0; i < vlen; i++) {
+               struct msghdr *omsg = &omsgvec[i].msg_hdr;
+               struct msghdr *msg = &msgvec[i].msg_hdr;
+
+               if (!(i == 0 || i < ret)) {
+                       break;
+               }
+
+#ifdef HAVE_STRUCT_MSGHDR_MSG_CONTROL
+               tmp[i].msg_ctrllen_filled += msg->msg_controllen;
+               tmp[i].msg_ctrllen_left -= msg->msg_controllen;
+
+               if (omsg->msg_control != NULL) {
+                       uint8_t *p;
+
+                       p = omsg->msg_control;
+                       p += tmp[i].msg_ctrllen_filled;
+
+                       msg->msg_control = p;
+                       msg->msg_controllen = tmp[i].msg_ctrllen_left;
+               } else {
+                       msg->msg_control = NULL;
+                       msg->msg_controllen = 0;
+               }
+#endif
+
+               /*
+                * We convert the unix address to a IP address so we need a buffer
+                * which can store the address in case of SOCK_DGRAM, see below.
+                */
+               msg->msg_name = &tmp[i].convert_addr.sa;
+               msg->msg_namelen = tmp[i].convert_addr.sa_socklen;
+
+               swrap_recvmsg_after(s, si, msg,
+                                   &tmp[i].from_addr.sa.un,
+                                   tmp[i].from_addr.sa_socklen,
+                                   ret);
+
+#ifdef HAVE_STRUCT_MSGHDR_MSG_CONTROL
+               if (omsg->msg_control != NULL) {
+                       /* msg->msg_controllen = space left */
+                       tmp[i].msg_ctrllen_left = msg->msg_controllen;
+                       tmp[i].msg_ctrllen_filled = omsg->msg_controllen - tmp[i].msg_ctrllen_left;
+               }
+
+               /* Update the original message length */
+               omsg->msg_controllen = tmp[i].msg_ctrllen_filled;
+               omsg->msg_flags = msg->msg_flags;
+#endif
+               omsg->msg_iovlen = msg->msg_iovlen;
+
+               SWRAP_LOCK_SI(si);
+
+               /*
+                * From the manpage:
+                *
+                * The  msg_name  field  points  to a caller-allocated buffer that is
+                * used to return the source address if the socket is unconnected.  The
+                * caller should set msg_namelen to the size of this buffer before this
+                * call; upon return from a successful call, msg_name will contain the
+                * length of the returned address.  If the application  does  not  need
+                * to know the source address, msg_name can be specified as NULL.
+                */
+               if (si->type == SOCK_STREAM) {
+                       omsg->msg_namelen = 0;
+               } else if (omsg->msg_name != NULL &&
+                          omsg->msg_namelen != 0 &&
+                          omsg->msg_namelen >= msg->msg_namelen) {
+                       memcpy(omsg->msg_name, msg->msg_name, msg->msg_namelen);
+                       omsg->msg_namelen = msg->msg_namelen;
+               }
+
+               SWRAP_UNLOCK_SI(si);
+       }
+       errno = saved_errno;
+
+       return ret;
+}
+
+#if defined(HAVE_RECVMMSG_SSIZE_T_CONST_TIMEOUT)
+/* FreeBSD */
+ssize_t recvmmsg(int sockfd, struct mmsghdr *msgvec, size_t vlen, int flags, const struct timespec *timeout)
+#elif defined(HAVE_RECVMMSG_CONST_TIMEOUT)
+/* Linux legacy glibc < 2.21 */
+int recvmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, int flags, const struct timespec *timeout)
+#else
+/* Linux glibc >= 2.21 */
+int recvmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, int flags, struct timespec *timeout)
+#endif
+{
+       return swrap_recvmmsg(sockfd, msgvec, vlen, flags, timeout);
+}
+#endif /* HAVE_RECVMMSG */
+
 /****************************************************************************
  *   SENDMSG
  ***************************************************************************/
@@ -7305,6 +7647,249 @@ ssize_t sendmsg(int s, const struct msghdr *omsg, int flags)
        return swrap_sendmsg(s, omsg, flags);
 }
 
+/****************************************************************************
+ *   SENDMMSG
+ ***************************************************************************/
+
+#ifdef HAVE_SENDMMSG
+#if defined(HAVE_SENDMMSG_SSIZE_T)
+/* FreeBSD */
+static ssize_t swrap_sendmmsg(int s, struct mmsghdr *omsgvec, size_t _vlen, int flags)
+#else
+/* Linux */
+static int swrap_sendmmsg(int s, struct mmsghdr *omsgvec, unsigned int _vlen, int flags)
+#endif
+{
+       struct socket_info *si = find_socket_info(s);
+#define __SWRAP_SENDMMSG_MAX_VLEN 16
+       struct mmsghdr msgvec[__SWRAP_SENDMMSG_MAX_VLEN] = {};
+       struct {
+               struct iovec iov;
+               struct sockaddr_un un_addr;
+               const struct sockaddr_un *to_un;
+               const struct sockaddr *to;
+               int bcast;
+       } tmp[__SWRAP_SENDMMSG_MAX_VLEN] = {};
+       int vlen;
+       int i;
+       char *swrap_dir = NULL;
+       int connected = 0;
+       int found_bcast = 0;
+       int ret;
+       int rc;
+       int saved_errno;
+
+       if (_vlen > __SWRAP_SENDMMSG_MAX_VLEN) {
+               vlen = __SWRAP_SENDMMSG_MAX_VLEN;
+       } else {
+               vlen = _vlen;
+       }
+
+       if (!si) {
+               int scm_rights_pipe_fd[__SWRAP_SENDMMSG_MAX_VLEN];
+
+               for (i = 0; i < __SWRAP_SENDMMSG_MAX_VLEN; i++) {
+                       scm_rights_pipe_fd[i] = -1;
+               }
+
+               for (i = 0; i < vlen; i++) {
+                       struct msghdr *omsg = &omsgvec[i].msg_hdr;
+                       struct msghdr *msg = &msgvec[i].msg_hdr;
+
+                       rc = swrap_sendmsg_before_unix(omsg, msg,
+                                                      &scm_rights_pipe_fd[i]);
+                       if (rc < 0) {
+                               ret = rc;
+                               goto fail_libc;
+                       }
+               }
+
+               ret = libc_sendmmsg(s, msgvec, vlen, flags);
+               if (ret < 0) {
+                       goto fail_libc;
+               }
+
+               for (i = 0; i < ret; i++) {
+                       omsgvec[i].msg_len = msgvec[i].msg_len;
+               }
+
+fail_libc:
+               saved_errno = errno;
+               for (i = 0; i < vlen; i++) {
+                       struct msghdr *msg = &msgvec[i].msg_hdr;
+
+                       swrap_sendmsg_after_unix(msg, ret,
+                                                scm_rights_pipe_fd[i]);
+               }
+               errno = saved_errno;
+
+               return ret;
+       }
+
+       SWRAP_LOCK_SI(si);
+       connected = si->connected;
+       SWRAP_UNLOCK_SI(si);
+
+       for (i = 0; i < vlen; i++) {
+               struct msghdr *omsg = &omsgvec[i].msg_hdr;
+               struct msghdr *msg = &msgvec[i].msg_hdr;
+
+               if (connected == 0) {
+                       msg->msg_name = omsg->msg_name;             /* optional address */
+                       msg->msg_namelen = omsg->msg_namelen;       /* size of address */
+               }
+               msg->msg_iov = omsg->msg_iov;               /* scatter/gather array */
+               msg->msg_iovlen = omsg->msg_iovlen;         /* # elements in msg_iov */
+
+#ifdef HAVE_STRUCT_MSGHDR_MSG_CONTROL
+               if (omsg->msg_controllen > 0 && omsg->msg_control != NULL) {
+                       uint8_t *cmbuf = NULL;
+                       size_t cmlen = 0;
+
+                       rc = swrap_sendmsg_filter_cmsghdr(omsg, &cmbuf, &cmlen);
+                       if (rc < 0) {
+                               ret = rc;
+                               goto fail_swrap;
+                       }
+
+                       if (cmlen != 0) {
+                               msg->msg_control = cmbuf;
+                               msg->msg_controllen = cmlen;
+                       }
+               }
+               msg->msg_flags = omsg->msg_flags;           /* flags on received message */
+#endif
+
+               rc = swrap_sendmsg_before(s, si, msg,
+                                         &tmp[i].iov,
+                                         &tmp[i].un_addr,
+                                         &tmp[i].to_un,
+                                         &tmp[i].to,
+                                         &tmp[i].bcast);
+               if (rc < 0) {
+                       ret = rc;
+                       goto fail_swrap;
+               }
+
+               if (tmp[i].bcast) {
+                       found_bcast = 1;
+               }
+       }
+
+       if (found_bcast) {
+
+               swrap_dir = socket_wrapper_dir();
+               if (swrap_dir == NULL) {
+                       ret = -1;
+                       goto fail_swrap;
+               }
+
+               for (i = 0; i < vlen; i++) {
+                       struct msghdr *msg = &msgvec[i].msg_hdr;
+                       struct sockaddr_un *un_addr = &tmp[i].un_addr;
+                       const struct sockaddr *to = tmp[i].to;
+                       struct stat st;
+                       unsigned int iface;
+                       unsigned int prt = ntohs(((const struct sockaddr_in *)(const void *)to)->sin_port);
+                       char type;
+                       size_t l, len = 0;
+                       uint8_t *buf;
+                       off_t ofs = 0;
+                       size_t avail = 0;
+                       size_t remain;
+
+                       for (l = 0; l < (size_t)msg->msg_iovlen; l++) {
+                               avail += msg->msg_iov[l].iov_len;
+                       }
+
+                       len = avail;
+                       remain = avail;
+
+                       /* we capture it as one single packet */
+                       buf = (uint8_t *)malloc(remain);
+                       if (!buf) {
+                               ret = -1;
+                               goto fail_swrap;
+                       }
+
+                       for (l = 0; l < (size_t)msg->msg_iovlen; l++) {
+                               size_t this_time = MIN(remain, (size_t)msg->msg_iov[l].iov_len);
+                               memcpy(buf + ofs,
+                                      msg->msg_iov[l].iov_base,
+                                      this_time);
+                               ofs += this_time;
+                               remain -= this_time;
+                       }
+
+                       type = SOCKET_TYPE_CHAR_UDP;
+
+                       for(iface=0; iface <= MAX_WRAPPED_INTERFACES; iface++) {
+                               swrap_un_path(un_addr, swrap_dir, type, iface, prt);
+                               if (stat(un_addr->sun_path, &st) != 0) continue;
+
+                               msg->msg_name = un_addr;             /* optional address */
+                               msg->msg_namelen = sizeof(*un_addr); /* size of address */
+
+                               /*
+                                * ignore the any errors in broadcast sends and
+                                * do a single sendmsg instead of sendmmsg
+                                */
+                               libc_sendmsg(s, msg, flags);
+                       }
+
+                       SWRAP_LOCK_SI(si);
+                       swrap_pcap_dump_packet(si, to, SWRAP_SENDTO, buf, len);
+                       SWRAP_UNLOCK_SI(si);
+
+                       SAFE_FREE(buf);
+
+                       msgvec[i].msg_len = len;
+               }
+
+               ret = vlen;
+               goto bcast_done;
+       }
+
+       ret = libc_sendmmsg(s, msgvec, vlen, flags);
+       if (ret < 0) {
+               goto fail_swrap;
+       }
+
+bcast_done:
+       for (i = 0; i < ret; i++) {
+               omsgvec[i].msg_len = msgvec[i].msg_len;
+       }
+
+fail_swrap:
+       saved_errno = errno;
+       for (i = 0; i < vlen; i++) {
+               struct msghdr *msg = &msgvec[i].msg_hdr;
+
+               if (i == 0 || i < ret) {
+                       swrap_sendmsg_after(s, si, msg, tmp[i].to, ret);
+               }
+#ifdef HAVE_STRUCT_MSGHDR_MSG_CONTROL
+               SAFE_FREE(msg->msg_control);
+#endif
+       }
+       SAFE_FREE(swrap_dir);
+       errno = saved_errno;
+
+       return ret;
+}
+
+#if defined(HAVE_SENDMMSG_SSIZE_T)
+/* FreeBSD */
+ssize_t sendmmsg(int s, struct mmsghdr *msgvec, size_t vlen, int flags)
+#else
+/* Linux */
+int sendmmsg(int s, struct mmsghdr *msgvec, unsigned int vlen, int flags)
+#endif
+{
+       return swrap_sendmmsg(s, msgvec, vlen, flags);
+}
+#endif /* HAVE_SENDMMSG */
+
 /****************************************************************************
  *   READV
  ***************************************************************************/
@@ -7417,7 +8002,9 @@ ssize_t writev(int s, const struct iovec *vector, int count)
  * CLOSE
  ***************************/
 
-static int swrap_close(int fd)
+static int swrap_remove_wrapper(const char *__func_name,
+                               int (*__close_fd_fn)(int fd),
+                               int fd)
 {
        struct socket_info *si = NULL;
        int si_index;
@@ -7429,10 +8016,10 @@ static int swrap_close(int fd)
        si_index = find_socket_info_index(fd);
        if (si_index == -1) {
                swrap_mutex_unlock(&socket_reset_mutex);
-               return libc_close(fd);
+               return __close_fd_fn(fd);
        }
 
-       SWRAP_LOG(SWRAP_LOG_TRACE, "Close wrapper for fd=%d", fd);
+       swrap_log(SWRAP_LOG_TRACE, __func_name, "Remove wrapper for fd=%d", fd);
        reset_socket_info_index(fd);
 
        si = swrap_get_socket_info(si_index);
@@ -7440,7 +8027,7 @@ static int swrap_close(int fd)
        swrap_mutex_lock(&first_free_mutex);
        SWRAP_LOCK_SI(si);
 
-       ret = libc_close(fd);
+       ret = __close_fd_fn(fd);
        if (ret == -1) {
                ret_errno = errno;
        }
@@ -7482,11 +8069,64 @@ out:
        return ret;
 }
 
+static int swrap_noop_close(int fd)
+{
+       (void)fd; /* unused */
+       return 0;
+}
+
+static void swrap_remove_stale(int fd)
+{
+       swrap_remove_wrapper(__func__, swrap_noop_close, fd);
+}
+
+/*
+ * This allows socket_wrapper aware applications to
+ * indicate that the given fd does not belong to
+ * an inet socket.
+ *
+ * We already overload a lot of unrelated functions
+ * like eventfd(), timerfd_create(), ... in order to
+ * call swrap_remove_stale() on the returned fd, but
+ * we'll never be able to handle all possible syscalls.
+ *
+ * socket_wrapper_indicate_no_inet_fd() gives them a way
+ * to do the same.
+ *
+ * We don't export swrap_remove_stale() in order to
+ * make it easier to analyze SOCKET_WRAPPER_DEBUGLEVEL=3
+ * log files.
+ */
+void socket_wrapper_indicate_no_inet_fd(int fd)
+{
+       swrap_remove_wrapper(__func__, swrap_noop_close, fd);
+}
+
+static int swrap_close(int fd)
+{
+       return swrap_remove_wrapper(__func__, libc_close, fd);
+}
+
 int close(int fd)
 {
        return swrap_close(fd);
 }
 
+#ifdef HAVE___CLOSE_NOCANCEL
+
+static int swrap___close_nocancel(int fd)
+{
+       return swrap_remove_wrapper(__func__, libc___close_nocancel, fd);
+}
+
+int __close_nocancel(int fd);
+int __close_nocancel(int fd)
+{
+       return swrap___close_nocancel(fd);
+}
+
+#endif /* HAVE___CLOSE_NOCANCEL */
+
 /****************************
  * DUP
  ***************************/
@@ -7794,10 +8434,18 @@ void swrap_destructor(void)
 
        SAFE_FREE(sockets);
 
-       if (swrap.libc.handle != NULL) {
+       if (swrap.libc.handle != NULL
+#ifdef RTLD_NEXT
+           && swrap.libc.handle != RTLD_NEXT
+#endif
+                       ) {
                dlclose(swrap.libc.handle);
        }
-       if (swrap.libc.socket_handle) {
+       if (swrap.libc.socket_handle
+#ifdef RTLD_NEXT
+           && swrap.libc.socket_handle != RTLD_NEXT
+#endif
+                       ) {
                dlclose(swrap.libc.socket_handle);
        }
 }
@@ -7815,8 +8463,8 @@ void swrap_destructor(void)
  * related syscalls also with the '_' prefix.
  *
  * This is tested in Samba's 'make test',
- * there we noticed that providing '_read'
- * and '_open' would cause errors, which
+ * there we noticed that providing '_read',
+ * '_open' and '_close' would cause errors, which
  * means we skip '_read', '_write' and
  * all non socket related calls without
  * further analyzing the problem.
@@ -7829,7 +8477,6 @@ SWRAP_SYMBOL_ALIAS(accept4, _accept4);
 #endif
 SWRAP_SYMBOL_ALIAS(accept, _accept);
 SWRAP_SYMBOL_ALIAS(bind, _bind);
-SWRAP_SYMBOL_ALIAS(close, _close);
 SWRAP_SYMBOL_ALIAS(connect, _connect);
 SWRAP_SYMBOL_ALIAS(dup, _dup);
 SWRAP_SYMBOL_ALIAS(dup2, _dup2);