/*
- * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <sys/resource.h>
#include <sys/socket.h>
#include <sys/stat.h>
+#include <sys/uio.h>
#include <sys/un.h>
#include <unistd.h>
#include "dynamic-string.h"
#include "fatal-signal.h"
#include "packets.h"
+#include "poll-loop.h"
#include "util.h"
#include "vlog.h"
-#if AF_PACKET && __linux__
+#if AF_PACKET && LINUX_DATAPATH
#include <linux/if_packet.h>
#endif
#ifdef HAVE_NETLINK
/* #ifdefs make it a pain to maintain code: you have to try to build both ways.
* Thus, this file compiles all of the code regardless of the target, by
- * writing "if (LINUX)" instead of "#ifdef __linux__". */
-#ifdef __linux__
-#define LINUX 1
-#else
-#define LINUX 0
+ * writing "if (LINUX_DATAPATH)" instead of "#ifdef __linux__". */
+#ifndef LINUX_DATAPATH
+#define LINUX_DATAPATH 0
#endif
#ifndef O_DIRECTORY
}
}
+void
+xset_nonblocking(int fd)
+{
+ if (set_nonblocking(fd)) {
+ exit(EXIT_FAILURE);
+ }
+}
+
+int
+set_dscp(int fd, uint8_t dscp)
+{
+ int val;
+
+ if (dscp > 63) {
+ return EINVAL;
+ }
+
+ val = dscp << 2;
+ if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
+ return errno;
+ }
+
+ return 0;
+}
+
static bool
rlim_is_finite(rlim_t limit)
{
* Most Open vSwitch code should not use this because it causes deadlocks:
* gethostbyname() sends out a DNS request but that starts a new flow for which
* OVS must set up a flow, but it can't because it's waiting for a DNS reply.
- * The synchronous lookup also delays other activty. (Of course we can solve
+ * The synchronous lookup also delays other activity. (Of course we can solve
* this but it doesn't seem worthwhile quite yet.) */
int
lookup_hostname(const char *host_name, struct in_addr *addr)
: EINVAL);
}
-/* Returns the error condition associated with socket 'fd' and resets the
- * socket's error status. */
-int
-get_socket_error(int fd)
-{
- int error;
-
- if (getsockopt_int(fd, SOL_SOCKET, SO_ERROR, "SO_ERROR", &error)) {
- error = errno;
- }
- return error;
-}
-
int
check_connection_completion(int fd)
{
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
struct pollfd pfd;
int retval;
retval = poll(&pfd, 1, 0);
} while (retval < 0 && errno == EINTR);
if (retval == 1) {
- return get_socket_error(fd);
+ if (pfd.revents & POLLERR) {
+ ssize_t n = send(fd, "", 1, MSG_DONTWAIT);
+ if (n < 0) {
+ return errno;
+ } else {
+ VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
+ return EPROTO;
+ }
+ }
+ return 0;
} else if (retval < 0) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
VLOG_ERR_RL(&rl, "poll: %s", strerror(errno));
return errno;
} else {
*
* On other Unix-like OSes, MSG_TRUNC has no effect in the flags
* argument. */
- char buffer[LINUX ? 1 : 2048];
+ char buffer[LINUX_DATAPATH ? 1 : 2048];
ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
MSG_TRUNC | MSG_DONTWAIT);
if (n_bytes <= 0 || n_bytes >= rcvbuf) {
if (strlen(name) > MAX_UN_LEN) {
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
- if (LINUX) {
+ if (LINUX_DATAPATH) {
/* 'name' is too long to fit in a sockaddr_un, but we have a
* workaround for that on Linux: shorten it by opening a file
* descriptor for the directory part of the name and indirecting
/* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
* SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
* connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
- * is true, the socket is made non-blocking. If 'passcred' is true, the socket
- * is configured to receive SCM_CREDENTIALS control messages.
+ * is true, the socket is made non-blocking.
*
* Returns the socket's fd if successful, otherwise a negative errno value. */
int
-make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED,
+make_unix_socket(int style, bool nonblock,
const char *bind_path, const char *connect_path)
{
int error;
}
}
-#ifdef SCM_CREDENTIALS
- if (passcred) {
- int enable = 1;
- if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) {
- error = errno;
- goto error;
- }
- }
-#endif
-
return fd;
error:
error = EPROTO;
}
if (bind_path) {
- fatal_signal_remove_file_to_unlink(bind_path);
+ fatal_signal_unlink_file_now(bind_path);
}
close(fd);
return -error;
* and stores -1 into '*fdp'.
*
* If 'sinp' is non-null, then on success the target address is stored into
- * '*sinp'. */
+ * '*sinp'.
+ *
+ * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
+ * should be in the range [0, 63] and will automatically be shifted to the
+ * appropriately place in the IP tos field. */
int
inet_open_active(int style, const char *target, uint16_t default_port,
- struct sockaddr_in *sinp, int *fdp)
+ struct sockaddr_in *sinp, int *fdp, uint8_t dscp)
{
struct sockaddr_in sin;
int fd = -1;
}
error = set_nonblocking(fd);
if (error) {
- goto exit_close;
+ goto exit;
+ }
+
+ /* The dscp bits must be configured before connect() to ensure that the TOS
+ * field is set during the connection establishment. If set after
+ * connect(), the handshake SYN frames will be sent with a TOS of 0. */
+ error = set_dscp(fd, dscp);
+ if (error) {
+ VLOG_ERR("%s: socket: %s", target, strerror(error));
+ goto exit;
}
/* Connect. */
error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
if (error == EINPROGRESS) {
error = EAGAIN;
- } else if (error && error != EAGAIN) {
- goto exit_close;
}
- /* Success: error is 0 or EAGAIN. */
- goto exit;
-
-exit_close:
- close(fd);
exit:
if (!error || error == EAGAIN) {
if (sinp) {
*sinp = sin;
}
- *fdp = fd;
- } else {
- *fdp = -1;
+ } else if (fd >= 0) {
+ close(fd);
+ fd = -1;
}
+ *fdp = fd;
return error;
}
* negative errno value.
*
* If 'sinp' is non-null, then on success the bound address is stored into
- * '*sinp'. */
+ * '*sinp'.
+ *
+ * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
+ * should be in the range [0, 63] and will automatically be shifted to the
+ * appropriately place in the IP tos field. */
int
inet_open_passive(int style, const char *target, int default_port,
- struct sockaddr_in *sinp)
+ struct sockaddr_in *sinp, uint8_t dscp)
{
struct sockaddr_in sin;
int fd = 0, error;
goto error;
}
+ /* The dscp bits must be configured before connect() to ensure that the TOS
+ * field is set during the connection establishment. If set after
+ * connect(), the handshake SYN frames will be sent with a TOS of 0. */
+ error = set_dscp(fd, dscp);
+ if (error) {
+ VLOG_ERR("%s: socket: %s", target, strerror(error));
+ goto error;
+ }
+
/* Listen. */
- if (listen(fd, 10) < 0) {
+ if (style == SOCK_STREAM && listen(fd, 10) < 0) {
error = errno;
VLOG_ERR("%s: listen: %s", target, strerror(error));
goto error;
}
}
+void
+xpipe_nonblocking(int fds[2])
+{
+ xpipe(fds);
+ xset_nonblocking(fds[0]);
+ xset_nonblocking(fds[1]);
+}
+
+void
+xsocketpair(int domain, int type, int protocol, int fds[2])
+{
+ if (socketpair(domain, type, protocol, fds)) {
+ VLOG_FATAL("failed to create socketpair (%s)", strerror(errno));
+ }
+}
+
static int
getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
{
memcpy(&sin, &ss, sizeof sin);
ds_put_format(string, IP_FMT":%"PRIu16,
- IP_ARGS(&sin.sin_addr.s_addr), ntohs(sin.sin_port));
+ IP_ARGS(sin.sin_addr.s_addr), ntohs(sin.sin_port));
} else if (ss.ss_family == AF_UNIX) {
struct sockaddr_un sun;
const char *null;
}
}
#endif
-#if AF_PACKET && __linux__
+#if AF_PACKET && LINUX_DATAPATH
else if (ss.ss_family == AF_PACKET) {
struct sockaddr_ll sll;
}
-#ifdef __linux__
+#ifdef LINUX_DATAPATH
static void
put_fd_filename(struct ds *string, int fd)
{
: S_ISFIFO(s.st_mode) ? "FIFO"
: S_ISLNK(s.st_mode) ? "symbolic link"
: "unknown"));
-#ifdef __linux__
+#ifdef LINUX_DATAPATH
put_fd_filename(&string, fd);
#endif
}
return ds_steal_cstr(&string);
}
+
+/* Returns the total of the 'iov_len' members of the 'n_iovs' in 'iovs'.
+ * The caller must ensure that the total does not exceed SIZE_MAX. */
+size_t
+iovec_len(const struct iovec iovs[], size_t n_iovs)
+{
+ size_t len = 0;
+ size_t i;
+
+ for (i = 0; i < n_iovs; i++) {
+ len += iovs[i].iov_len;
+ }
+ return len;
+}
+
+/* Returns true if all of the 'n_iovs' iovecs in 'iovs' have length zero. */
+bool
+iovec_is_empty(const struct iovec iovs[], size_t n_iovs)
+{
+ size_t i;
+
+ for (i = 0; i < n_iovs; i++) {
+ if (iovs[i].iov_len) {
+ return false;
+ }
+ }
+ return true;
+}
+
+/* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
+ * in 'fds' on Unix domain socket 'sock'. Returns the number of bytes
+ * successfully sent or -1 if an error occurred. On error, sets errno
+ * appropriately. */
+int
+send_iovec_and_fds(int sock,
+ const struct iovec *iovs, size_t n_iovs,
+ const int fds[], size_t n_fds)
+{
+ ovs_assert(sock >= 0);
+ if (n_fds > 0) {
+ union {
+ struct cmsghdr cm;
+ char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
+ } cmsg;
+ struct msghdr msg;
+
+ ovs_assert(!iovec_is_empty(iovs, n_iovs));
+ ovs_assert(n_fds <= SOUTIL_MAX_FDS);
+
+ memset(&cmsg, 0, sizeof cmsg);
+ cmsg.cm.cmsg_len = CMSG_LEN(n_fds * sizeof *fds);
+ cmsg.cm.cmsg_level = SOL_SOCKET;
+ cmsg.cm.cmsg_type = SCM_RIGHTS;
+ memcpy(CMSG_DATA(&cmsg.cm), fds, n_fds * sizeof *fds);
+
+ msg.msg_name = NULL;
+ msg.msg_namelen = 0;
+ msg.msg_iov = CONST_CAST(struct iovec *, iovs);
+ msg.msg_iovlen = n_iovs;
+ msg.msg_control = &cmsg.cm;
+ msg.msg_controllen = CMSG_SPACE(n_fds * sizeof *fds);
+ msg.msg_flags = 0;
+
+ return sendmsg(sock, &msg, 0);
+ } else {
+ return writev(sock, iovs, n_iovs);
+ }
+}
+
+/* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
+ * in 'fds' on Unix domain socket 'sock'. If 'skip_bytes' is nonzero, then the
+ * first 'skip_bytes' of data in the iovecs are not sent, and none of the file
+ * descriptors are sent. The function continues to retry sending until an
+ * error (other than EINTR) occurs or all the data and fds are sent.
+ *
+ * Returns 0 if all the data and fds were successfully sent, otherwise a
+ * positive errno value. Regardless of success, stores the number of bytes
+ * sent (always at least 'skip_bytes') in '*bytes_sent'. (If at least one byte
+ * is sent, then all the fds have been sent.)
+ *
+ * 'skip_bytes' must be less than or equal to iovec_len(iovs, n_iovs). */
+int
+send_iovec_and_fds_fully(int sock,
+ const struct iovec iovs[], size_t n_iovs,
+ const int fds[], size_t n_fds,
+ size_t skip_bytes, size_t *bytes_sent)
+{
+ *bytes_sent = 0;
+ while (n_iovs > 0) {
+ int retval;
+
+ if (skip_bytes) {
+ retval = skip_bytes;
+ skip_bytes = 0;
+ } else if (!*bytes_sent) {
+ retval = send_iovec_and_fds(sock, iovs, n_iovs, fds, n_fds);
+ } else {
+ retval = writev(sock, iovs, n_iovs);
+ }
+
+ if (retval > 0) {
+ *bytes_sent += retval;
+ while (retval > 0) {
+ const uint8_t *base = iovs->iov_base;
+ size_t len = iovs->iov_len;
+
+ if (retval < len) {
+ size_t sent;
+ int error;
+
+ error = write_fully(sock, base + retval, len - retval,
+ &sent);
+ *bytes_sent += sent;
+ retval += sent;
+ if (error) {
+ return error;
+ }
+ }
+ retval -= len;
+ iovs++;
+ n_iovs--;
+ }
+ } else if (retval == 0) {
+ if (iovec_is_empty(iovs, n_iovs)) {
+ break;
+ }
+ VLOG_WARN("send returned 0");
+ return EPROTO;
+ } else if (errno != EINTR) {
+ return errno;
+ }
+ }
+
+ return 0;
+}
+
+/* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
+ * in 'fds' on Unix domain socket 'sock'. The function continues to retry
+ * sending until an error (other than EAGAIN or EINTR) occurs or all the data
+ * and fds are sent. Upon EAGAIN, the function blocks until the socket is
+ * ready for more data.
+ *
+ * Returns 0 if all the data and fds were successfully sent, otherwise a
+ * positive errno value. */
+int
+send_iovec_and_fds_fully_block(int sock,
+ const struct iovec iovs[], size_t n_iovs,
+ const int fds[], size_t n_fds)
+{
+ size_t sent = 0;
+
+ for (;;) {
+ int error;
+
+ error = send_iovec_and_fds_fully(sock, iovs, n_iovs,
+ fds, n_fds, sent, &sent);
+ if (error != EAGAIN) {
+ return error;
+ }
+ poll_fd_wait(sock, POLLOUT);
+ poll_block();
+ }
+}
+
+/* Attempts to receive from Unix domain socket 'sock' up to 'size' bytes of
+ * data into 'data' and up to SOUTIL_MAX_FDS file descriptors into 'fds'.
+ *
+ * - Upon success, returns the number of bytes of data copied into 'data'
+ * and stores the number of received file descriptors into '*n_fdsp'.
+ *
+ * - On failure, returns a negative errno value and stores 0 in
+ * '*n_fdsp'.
+ *
+ * - On EOF, returns 0 and stores 0 in '*n_fdsp'. */
+int
+recv_data_and_fds(int sock,
+ void *data, size_t size,
+ int fds[SOUTIL_MAX_FDS], size_t *n_fdsp)
+{
+ union {
+ struct cmsghdr cm;
+ char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
+ } cmsg;
+ struct msghdr msg;
+ int retval;
+ struct cmsghdr *p;
+ size_t i;
+
+ *n_fdsp = 0;
+
+ do {
+ struct iovec iov;
+
+ iov.iov_base = data;
+ iov.iov_len = size;
+
+ msg.msg_name = NULL;
+ msg.msg_namelen = 0;
+ msg.msg_iov = &iov;
+ msg.msg_iovlen = 1;
+ msg.msg_control = &cmsg.cm;
+ msg.msg_controllen = sizeof cmsg.control;
+ msg.msg_flags = 0;
+
+ retval = recvmsg(sock, &msg, 0);
+ } while (retval < 0 && errno == EINTR);
+ if (retval <= 0) {
+ return retval < 0 ? -errno : 0;
+ }
+
+ for (p = CMSG_FIRSTHDR(&msg); p; p = CMSG_NXTHDR(&msg, p)) {
+ if (p->cmsg_level != SOL_SOCKET || p->cmsg_type != SCM_RIGHTS) {
+ VLOG_ERR("unexpected control message %d:%d",
+ p->cmsg_level, p->cmsg_type);
+ goto error;
+ } else if (*n_fdsp) {
+ VLOG_ERR("multiple SCM_RIGHTS received");
+ goto error;
+ } else {
+ size_t n_fds = (p->cmsg_len - CMSG_LEN(0)) / sizeof *fds;
+ const int *fds_data = (const int *) CMSG_DATA(p);
+
+ ovs_assert(n_fds > 0);
+ if (n_fds > SOUTIL_MAX_FDS) {
+ VLOG_ERR("%zu fds received but only %d supported",
+ n_fds, SOUTIL_MAX_FDS);
+ for (i = 0; i < n_fds; i++) {
+ close(fds_data[i]);
+ }
+ goto error;
+ }
+
+ *n_fdsp = n_fds;
+ memcpy(fds, fds_data, n_fds * sizeof *fds);
+ }
+ }
+
+ return retval;
+
+error:
+ for (i = 0; i < *n_fdsp; i++) {
+ close(fds[i]);
+ }
+ *n_fdsp = 0;
+ return EPROTO;
+}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff