serval-dna/socket.c

/*
Serval DNA named sockets
Copyright 2013 Serval Project Inc.

This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License
as published by the Free Software Foundation; either version 2
of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
*/

#include <limits.h>
#include <stdlib.h>
#include <assert.h>
#include <libgen.h>
#include <sys/socket.h>
#include <netdb.h>

#ifdef __APPLE__
#include "TargetConditionals.h"
#endif

#include "instance.h"
#include "str.h"
#include "conf.h"
#include "log.h"
#include "debug.h"
#include "strbuf_helpers.h"
#include "socket.h"

/* Form the name of an AF_UNIX (local) socket in the /var/run/serval (or instance) directory as an
 * absolute path.  Under Linux, this will create a socket name in the abstract namespace.  This
 * permits us to use local sockets on Android despite its lack of a shared writeable directory on a
 * UFS partition.
 *
 * The absolute file name is resolved to its real path using realpath(3), to ensure that name
 * comparisons of addresses returned by recvmsg(2) can reliably be used on systems where the
 * instance path may have a symbolic link in it.
 *
 * Returns -1 if the path name overruns the size of a sockaddr_un structure, or if realpath(3) fails
 * with an error.  The contents of *addr and *addrlen are undefined in this case.
 *
 * @author Andrew Bettison <andrew@servalproject.com>
 * @author Daniel O'Connor <daniel@servalproject.com>
 */
int _make_local_sockaddr(struct __sourceloc __whence, struct socket_address *addr, const char *fmt, ...)
{
  bzero(addr, sizeof(*addr));
  addr->local.sun_family = AF_UNIX;
  va_list ap;
  va_start(ap, fmt);
#ifdef TARGET_OS_IPHONE
 // Do what we need to make the path short enough to fit in the BSD named socket maximum length of
 // 104 chars, even if we have a horribly long container path name.
 int r = snprintf(addr->local.sun_path,100,"%s/%s", getenv("HOME")?getenv("HOME"):"/tmp",fmt);
#else
 int r = VFORMF_SERVAL_RUN_PATH(addr->local.sun_path, fmt, ap);
#endif
  va_end(ap);
  if (!r)
    return WHY("socket name overflow");
  addr->addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(addr->local.sun_path) + 1;
// TODO perform real path transformation in making the serval instance path
//  if (real_sockaddr(addr, addr) == -1)
//    return -1;

#ifdef USE_ABSTRACT_NAMESPACE
  // For the abstract name we use the absolute path name with the initial '/' replaced by the
  // leading nul.  This ensures that different instances of the Serval daemon have different socket
  // names.
  addr->local.sun_path[0] = '\0'; // mark as Linux abstract socket
  --addr->addrlen; // do not count trailing nul in abstract socket name
#endif // USE_ABSTRACT_NAMESPACE
  return 0;
}

/* Converts an AF_UNIX local socket file name to contain a real path name using realpath(3), leaves
 * all other socket types intact, including abstract local socket names.  Returns -1 in case of an
 * error from realpath(3) or a buffer overflow, without modifying *dst_addr or *dst_addrlen.
 * Returns 1 if the path is changed and puts the modified path in *dst_addr and *dst_addrlen.
 * Returns 0 if not the path is not changed and copies from *src_addr to *dst_addr, src_addrlen to
 * *dst_addrlen.
 *
 * Can safely be used to perform an in-place conversion by using src_addr == dst_addr and
 * dst_addrlen == &src_addrlen.
 *
 * @author Andrew Bettison <andrew@servalproject.com>
 */
int real_sockaddr(const struct socket_address *src_addr, struct socket_address *dst_addr)
{
  DEBUGF2(io, verbose_io, "real_sockaddr(src_addr=%p %s, dst_addr=%p)", src_addr, alloca_socket_address(src_addr), dst_addr);
  assert(src_addr->addrlen > (socklen_t)offsetof(struct sockaddr_un, sun_path));
  size_t src_path_len = src_addr->addrlen - offsetof(struct sockaddr_un, sun_path);
  if (	 src_path_len > 1
      && src_addr->local.sun_family == AF_UNIX
      && src_addr->local.sun_path[0] != '\0'
      && src_addr->local.sun_path[src_path_len - 1] == '\0'
  ) {
    char real_path[PATH_MAX];
    size_t real_path_len;
    if (realpath(src_addr->local.sun_path, real_path) == NULL)
      return WHYF_perror("realpath(%s)", alloca_str_toprint(src_addr->local.sun_path));
    else if ((real_path_len = strlen(real_path) + 1) > sizeof dst_addr->local.sun_path)
      return WHYF("sockaddr overrun: realpath(%s) returned %s", 
	  alloca_str_toprint(src_addr->local.sun_path), alloca_str_toprint(real_path));
    else if (   real_path_len != src_path_len
	     || memcmp(real_path, src_addr->local.sun_path, src_path_len) != 0
    ) {
      memcpy(dst_addr->local.sun_path, real_path, real_path_len);
      dst_addr->addrlen = real_path_len + offsetof(struct sockaddr_un, sun_path);
      DEBUGF2(io, verbose_io, "   --> return %s", alloca_socket_address(dst_addr));
      return 1;
    }
    DEBUGF2(io, verbose_io, "real_path=%s", alloca_str_toprint(real_path));
  }
  if (dst_addr != src_addr){
    memcpy(&dst_addr->addr, &src_addr->addr, src_addr->addrlen);
    dst_addr->addrlen = src_addr->addrlen;
  }
  return 0;
}

/* Compare any two struct sockaddr.  Return -1, 0 or 1.  Copes with invalid and truncated sockaddr
 * structures.
 *
 * @author Andrew Bettison <andrew@servalproject.com>
 */
int cmp_sockaddr(const struct socket_address *addrA, const struct socket_address *addrB)
{
  // Two zero-length sockaddrs are equal.
  if (addrA->addrlen == 0 && addrB->addrlen == 0)
    return 0;
  // If either sockaddr is truncated, then we compare the bytes we have.
  if ((size_t)addrA->addrlen < sizeof addrA->addr.sa_family || (size_t)addrB->addrlen < sizeof addrB->addr.sa_family) {
    int c = memcmp(addrA, addrB, addrA->addrlen < addrB->addrlen ? addrA->addrlen : addrB->addrlen);
    if (c == 0)
      c = addrA->addrlen < addrB->addrlen ? -1 : addrA->addrlen > addrB->addrlen ? 1 : 0;
    return c;
  }
  // Order first by address family.
  if (addrA->addr.sa_family < addrB->addr.sa_family)
    return -1;
  if (addrA->addr.sa_family > addrB->addr.sa_family)
    return 1;
  // Both addresses are in the same family...
  switch (addrA->addr.sa_family) {
  case AF_INET: {
      if (addrA->inet.sin_addr.s_addr < addrB->inet.sin_addr.s_addr)
	return -1;
      if (addrA->inet.sin_addr.s_addr > addrB->inet.sin_addr.s_addr)
	return 1;
      if (addrA->inet.sin_port < addrB->inet.sin_port)
	return -1;
      if (addrA->inet.sin_port > addrB->inet.sin_port)
	return 1;
      return 0;
    }break;
  case AF_UNIX: {
      unsigned pathlenA = addrA->addrlen - offsetof(struct sockaddr_un, sun_path);
      unsigned pathlenB = addrB->addrlen - offsetof(struct sockaddr_un, sun_path);
      int c;
      if (   pathlenA > 1 && pathlenB > 1
	  && addrA->local.sun_path[0] == '\0'
	  && addrB->local.sun_path[0] == '\0'
      ) {
	// Both abstract sockets - just compare names, nul bytes are not terminators.
	c = memcmp(&addrA->local.sun_path[1],
		   &addrB->local.sun_path[1],
		   (pathlenA < pathlenB ? pathlenA : pathlenB) - 1);
      } else {
	// Either or both are named local file sockets.  If the file names are identical up to the
	// first nul, then the addresses are equal.  This collates abstract socket names, whose first
	// character is a nul, ahead of all non-empty file socket names.
	c = strncmp(addrA->local.sun_path,
		    addrB->local.sun_path,
		    (pathlenA < pathlenB ? pathlenA : pathlenB));
      }
      if (c == 0)
	c = pathlenA < pathlenB ? -1 : pathlenA > pathlenB ? 1 : 0;
      return c;
    }
    break;
  }
  // Fall back to comparing raw data bytes.
  int c = memcmp(addrA->addr.sa_data, addrB->addr.sa_data, 
      (addrA->addrlen < addrB->addrlen ? addrA->addrlen : addrB->addrlen) - sizeof addrA->addr.sa_family);
  if (c == 0)
    c = addrA->addrlen < addrB->addrlen ? -1 : addrA->addrlen > addrB->addrlen ? 1 : 0;
  
  return c;
}

int is_sockaddr_local(const struct socket_address *addr)
{
  if ((size_t)addr->addrlen < sizeof addr->addr.sa_family)
    return 0;
  switch (addr->addr.sa_family) {
    case AF_INET:
      return ((unsigned char*)&addr->inet.sin_addr.s_addr)[0] == IN_LOOPBACKNET ? 1 : 0;
    case AF_UNIX:
      return 1;
  }
  return 0;
}

int _esocket(struct __sourceloc __whence, int domain, int type, int protocol)
{
  int fd;
  if ((fd = socket(domain, type, protocol)) == -1)
    return WHYF_perror("socket(%s, %s, 0)", alloca_socket_domain(domain), alloca_socket_type(type));
  DEBUGF2(io, verbose_io, "socket(%s, %s, 0) -> %d", alloca_socket_domain(domain), alloca_socket_type(type), fd);
  return fd;
}

int _socket_connect(struct __sourceloc __whence, int sock, const struct socket_address *addr)
{
  if (connect(sock, &addr->addr, addr->addrlen) == -1)
    return WHYF_perror("connect(%d,%s,%lu)", sock, alloca_socket_address(addr), (unsigned long)addr->addrlen);
  DEBUGF2(io, verbose_io, "connect(%d, %s, %lu)", sock, alloca_socket_address(addr), (unsigned long)addr->addrlen);
  return 0;
}

int _socket_bind(struct __sourceloc __whence, int sock, const struct socket_address *addr)
{
  assert(addr->addrlen > (socklen_t)sizeof addr->addr.sa_family);
  if (addr->addr.sa_family == AF_UNIX && addr->local.sun_path[0] != '\0') {
    assert(addr->local.sun_path[addr->addrlen - offsetof(struct sockaddr_un, sun_path) -1] == '\0');
    // make sure the path exists, create it if we can
    size_t dirsiz = strlen(addr->local.sun_path) + 1;
    char dir_buf[dirsiz];
    strcpy(dir_buf, addr->local.sun_path);
    const char *dir = dirname(dir_buf); // modifies dir_buf[]
    if (mkdirs_info(dir, 0700) == -1)
      return WHY_perror("mkdirs()");
    // remove a previous socket
    if (unlink(addr->local.sun_path) == -1 && errno != ENOENT)
      WARNF_perror("unlink(%s)", alloca_str_toprint(addr->local.sun_path));
    DEBUGF2(io, verbose_io, "unlink(%s)", alloca_str_toprint(addr->local.sun_path));
  }
  if (bind(sock, &addr->addr, addr->addrlen) == -1)
    return WHYF_perror("bind(%d,%s,%lu)", sock, alloca_socket_address(addr), (unsigned long)addr->addrlen);
  DEBUGF2(io, verbose_io, "bind(%d, %s, %lu)", sock, alloca_socket_address(addr), (unsigned long)addr->addrlen);
  return 0;
}

int _socket_listen(struct __sourceloc __whence, int sock, int backlog)
{
  if (listen(sock, backlog) == -1)
    return WHYF_perror("listen(%d,%d)", sock, backlog);
  DEBUGF2(io, verbose_io, "listen(%d, %d)", sock, backlog);
  return 0;
}

int _socket_set_reuseaddr(struct __sourceloc __whence, int sock, int reuseP)
{
  if (setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &reuseP, sizeof reuseP) == -1) {
    WARNF_perror("setsockopt(%d,SOL_SOCKET,SO_REUSEADDR,&%d,%u)", sock, reuseP, (unsigned)sizeof reuseP);
    return -1;
  }
  DEBUGF2(io, verbose_io, "setsockopt(%d, SOL_SOCKET, SO_REUSEADDR, &%d, %u)", sock, reuseP, (unsigned)sizeof reuseP);
  return 0;
}

int _socket_set_rcvbufsize(struct __sourceloc __whence, int sock, unsigned buffer_size)
{
  if (setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &buffer_size, sizeof buffer_size) == -1) {
    WARNF_perror("setsockopt(%d,SOL_SOCKET,SO_RCVBUF,&%u,%u)", sock, buffer_size, (unsigned)sizeof buffer_size);
    return -1;
  }
  DEBUGF2(io, verbose_io, "setsockopt(%d, SOL_SOCKET, SO_RCVBUF, &%u, %u)", sock, buffer_size, (unsigned)sizeof buffer_size);
  return 0;
}

int socket_unlink_close(int sock)
{
  // get the socket name and unlink it from the filesystem if not abstract
  struct socket_address addr;
  addr.addrlen = sizeof addr.store;
  if (getsockname(sock, &addr.addr, &addr.addrlen))
    WHYF_perror("getsockname(%d)", sock);
  else if (addr.addr.sa_family==AF_UNIX 
    && (size_t)addr.addrlen >= offsetof(struct sockaddr_un, sun_path)
    && (size_t)addr.addrlen <= sizeof addr.local
    && addr.local.sun_path[0] != '\0') {
    if (unlink(addr.local.sun_path) == -1)
      WARNF_perror("unlink(%s)", alloca_str_toprint(addr.local.sun_path));
  }
  close(sock);
  return 0;
}

ssize_t _send_message(struct __sourceloc __whence, int fd, const struct socket_address *address, const struct fragmented_data *data)
{
  struct msghdr hdr={
    .msg_name=(void *)&address->addr,
    .msg_namelen=address->addrlen,
    .msg_iov=(struct iovec*)data->iov,
    .msg_iovlen=data->fragment_count,
  };
  ssize_t ret = sendmsg(fd, &hdr, 0);
  if (ret == -1 && errno != EAGAIN)
    WHYF_perror("sendmsg(%d,%s,%lu)", fd, alloca_socket_address(address), (unsigned long)address->addrlen);
  DEBUGF(verbose_io, "sendmsg(%d, %s, %lu)", fd, alloca_socket_address(address), (unsigned long)address->addrlen);
  return ret;
}

ssize_t _recv_message_frag(struct __sourceloc __whence, int fd, struct socket_address *address, int *ttl, struct fragmented_data *data)
{
  uint8_t cmsg_buff[1024];
  struct msghdr msg = {
    .msg_name       = (void *)&address->addr,
    .msg_namelen    = sizeof(address->raw),
    .msg_iov        = data->iov,
    .msg_iovlen     = data->fragment_count,
    .msg_control    = cmsg_buff,
    .msg_controllen = sizeof cmsg_buff,
    .msg_flags      = 0
  };
  bzero(address, sizeof(struct socket_address));
  ssize_t ret = recvmsg(fd, &msg, 0);
  if (ret == -1 && errno != EAGAIN && errno != EWOULDBLOCK)
    WHYF_perror("recvmsg(%d,{name=%p,namelen=%u,iov=%s,control=%p,controllen=%u},0)",
	  fd, &address->addr, (unsigned) address->addrlen,
          alloca_iovec(data->iov, data->fragment_count),
          cmsg_buff, (unsigned) sizeof cmsg_buff);
  address->addrlen = msg.msg_namelen;
  if (ttl && ret > 0) {
    struct cmsghdr *cmsg;
    for (cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
      if (   cmsg->cmsg_level == IPPROTO_IP
	  && ((cmsg->cmsg_type == IP_RECVTTL) || (cmsg->cmsg_type == IP_TTL))
	  && cmsg->cmsg_len
      ) {
	DEBUGF(verbose_io, "  TTL (%p) data location resolves to %p", ttl, CMSG_DATA(cmsg));
	if (CMSG_DATA(cmsg)) {
	  *ttl = *(unsigned char *) CMSG_DATA(cmsg);
	  DEBUGF(verbose_io, "  TTL of packet is %d", *ttl);
	}
      } else {
	DEBUGF(verbose_io, "  unexpected level=%02x, type=%02x", cmsg->cmsg_level, cmsg->cmsg_type);
      }
    }
  }
  DEBUGF(verbose_io, "recvmsg(%d) -> %zd, flags=%x, address=%s ttl=%d",
         fd,
         ret,
         msg.msg_flags,
         alloca_socket_address(address),
         ttl ? *ttl : -1);
  return ret;
}

ssize_t _recv_message(struct __sourceloc __whence, int fd, struct socket_address *address, int *ttl, unsigned char *buffer, size_t buflen)
{
  struct fragmented_data data;
  data.fragment_count = 1;
  data.iov[0].iov_base = buffer;
  data.iov[0].iov_len = buflen;
  return _recv_message_frag(__whence, fd, address, ttl, &data);
}

int socket_resolve_name(int family, const char *name, const char *service, struct socket_address *address){
  int ret=-1;
  struct addrinfo hint={
    .ai_flags = AI_ADDRCONFIG,
    .ai_family = family,
  };
  struct addrinfo *addresses=NULL;
  if (getaddrinfo(name, service, &hint, &addresses))
    return WHYF_perror("Failed to resolve %s",name);

  struct addrinfo *p = addresses;
  while(p){
    if (p->ai_addrlen < (socklen_t)sizeof(address->raw) && (p->ai_addr->sa_family == AF_INET || p->ai_addr->sa_family == AF_INET6)){
      address->addrlen = p->ai_addrlen;
      memcpy(&address->addr, p->ai_addr, p->ai_addrlen);
      ret = 0;
      break;
    }
    p = p->ai_next;
  }

  freeaddrinfo(addresses);
  return ret;
}