X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fsocket-util.c;h=fa55480d3e1fbd14a1c9001d77bab47c33b9bd96;hb=bbb8dee92d639331e8bd81823638267dcc895396;hp=249b6f002b19e2303c9b054bc91b04af50b94c3d;hpb=007948177581f3b3dad188221593d0e4bdca6ba0;p=sliver-openvswitch.git diff --git a/lib/socket-util.c b/lib/socket-util.c index 249b6f002..fa55480d3 100644 --- a/lib/socket-util.c +++ b/lib/socket-util.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks. + * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -19,6 +19,7 @@ #include #include #include +#include #include #include #include @@ -28,27 +29,39 @@ #include #include #include +#include #include #include +#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_DATAPATH +#include +#endif +#ifdef HAVE_NETLINK +#include "netlink-protocol.h" +#include "netlink-socket.h" +#endif VLOG_DEFINE_THIS_MODULE(socket_util); /* #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 #define O_DIRECTORY 0 #endif +static int getsockopt_int(int fd, int level, int option, const char *optname, + int *valuep); + /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a * positive errno value. */ int @@ -59,13 +72,38 @@ set_nonblocking(int fd) if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) { return 0; } else { - VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno)); + VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno)); return errno; } } else { - VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno)); + VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno)); + return errno; + } +} + +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 @@ -135,24 +173,72 @@ lookup_ipv6(const char *host_name, struct in6_addr *addr) return 0; } -/* Returns the error condition associated with socket 'fd' and resets the - * socket's error status. */ +/* Translates 'host_name', which must be a host name or a string representation + * of an IP address, into a numeric IP address in '*addr'. Returns 0 if + * successful, otherwise a positive errno value. + * + * Most Open vSwitch code should not use this because it causes deadlocks: + * getaddrinfo() 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 activity. (Of course we can solve + * this but it doesn't seem worthwhile quite yet.) */ int -get_socket_error(int fd) +lookup_hostname(const char *host_name, struct in_addr *addr) { - int error; - socklen_t len = sizeof(error); - if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) < 0) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); - error = errno; - VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error)); + struct addrinfo *result; + struct addrinfo hints; + + if (inet_aton(host_name, addr)) { + return 0; + } + + memset(&hints, 0, sizeof hints); + hints.ai_family = AF_INET; + + switch (getaddrinfo(host_name, NULL, &hints, &result)) { + case 0: + *addr = ((struct sockaddr_in *) result->ai_addr)->sin_addr; + freeaddrinfo(result); + return 0; + +#ifdef EAI_ADDRFAMILY + case EAI_ADDRFAMILY: +#endif + case EAI_NONAME: + case EAI_SERVICE: + return ENOENT; + + case EAI_AGAIN: + return EAGAIN; + + case EAI_BADFLAGS: + case EAI_FAMILY: + case EAI_SOCKTYPE: + return EINVAL; + + case EAI_FAIL: + return EIO; + + case EAI_MEMORY: + return ENOMEM; + +#ifdef EAI_NODATA + case EAI_NODATA: + return ENXIO; +#endif + + case EAI_SYSTEM: + return errno; + + default: + return EPROTO; } - 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; @@ -162,10 +248,18 @@ check_connection_completion(int fd) 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)); + VLOG_ERR_RL(&rl, "poll: %s", ovs_strerror(errno)); return errno; } else { return EAGAIN; @@ -180,15 +274,13 @@ check_connection_completion(int fd) int drain_rcvbuf(int fd) { - socklen_t rcvbuf_len; - size_t rcvbuf; + int rcvbuf; - rcvbuf_len = sizeof rcvbuf; - if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); - VLOG_ERR_RL(&rl, "getsockopt(SO_RCVBUF) failed: %s", strerror(errno)); - return errno; + rcvbuf = get_socket_rcvbuf(fd); + if (rcvbuf < 0) { + return -rcvbuf; } + while (rcvbuf > 0) { /* In Linux, specifying MSG_TRUNC in the flags argument causes the * datagram length to be returned, even if that is longer than the @@ -198,7 +290,7 @@ drain_rcvbuf(int fd) * * 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) { @@ -209,6 +301,18 @@ drain_rcvbuf(int fd) return 0; } +/* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a + * negative errno value if an error occurs. */ +int +get_socket_rcvbuf(int sock) +{ + int rcvbuf; + int error; + + error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf); + return error ? -error : rcvbuf; +} + /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no * more data can be immediately read. ('fd' should therefore be in * non-blocking mode.)*/ @@ -255,7 +359,7 @@ make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len, 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 @@ -316,12 +420,11 @@ bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len) /* 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; @@ -337,13 +440,8 @@ make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED, * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only * if a backlog of un-accepted connections has built up in the kernel.) */ if (nonblock) { - int flags = fcntl(fd, F_GETFL, 0); - if (flags == -1) { - error = errno; - goto error; - } - if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) { - error = errno; + error = set_nonblocking(fd); + if (error) { goto error; } } @@ -354,7 +452,8 @@ make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED, int dirfd; if (unlink(bind_path) && errno != ENOENT) { - VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno)); + VLOG_WARN("unlinking \"%s\": %s\n", + bind_path, ovs_strerror(errno)); } fatal_signal_add_file_to_unlink(bind_path); @@ -389,16 +488,6 @@ make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED, } } -#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: @@ -406,7 +495,7 @@ 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; @@ -492,10 +581,14 @@ exit: * 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; @@ -510,43 +603,44 @@ inet_open_active(int style, const char *target, uint16_t default_port, /* Create non-blocking socket. */ fd = socket(AF_INET, style, 0); if (fd < 0) { - VLOG_ERR("%s: socket: %s", target, strerror(errno)); + VLOG_ERR("%s: socket: %s", target, ovs_strerror(errno)); error = errno; goto exit; } 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, ovs_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; } -/* Opens a non-blocking IPv4 socket of the specified 'style', binds to - * 'target', and listens for incoming connections. 'target' should be a string - * in the format "[][:]": +/* Parses 'target', which should be a string in the format "[][:]": * * - If 'default_port' is -1, then is required. Otherwise, if * is omitted, then 'default_port' is used instead. @@ -556,106 +650,150 @@ exit: * * - If is omitted then the IP address is wildcarded. * - * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). - * - * For TCP, the socket will have SO_REUSEADDR turned on. - * - * On success, returns a non-negative file descriptor. On failure, returns a - * negative errno value. - * - * If 'sinp' is non-null, then on success the bound address is stored into - * '*sinp'. */ -int -inet_open_passive(int style, const char *target_, int default_port, - struct sockaddr_in *sinp) + * If successful, stores the address into '*sinp' and returns true; otherwise + * zeros '*sinp' and returns false. */ +bool +inet_parse_passive(const char *target_, int default_port, + struct sockaddr_in *sinp) { char *target = xstrdup(target_); char *string_ptr = target; - struct sockaddr_in sin; const char *host_name; const char *port_string; - int fd = 0, error, port; - unsigned int yes = 1; + bool ok = false; + int port; /* Address defaults. */ - memset(&sin, 0, sizeof sin); - sin.sin_family = AF_INET; - sin.sin_addr.s_addr = htonl(INADDR_ANY); - sin.sin_port = htons(default_port); + memset(sinp, 0, sizeof *sinp); + sinp->sin_family = AF_INET; + sinp->sin_addr.s_addr = htonl(INADDR_ANY); + sinp->sin_port = htons(default_port); /* Parse optional port number. */ port_string = strsep(&string_ptr, ":"); if (port_string && str_to_int(port_string, 10, &port)) { - sin.sin_port = htons(port); + sinp->sin_port = htons(port); } else if (default_port < 0) { VLOG_ERR("%s: port number must be specified", target_); - error = EAFNOSUPPORT; goto exit; } /* Parse optional bind IP. */ host_name = strsep(&string_ptr, ":"); - if (host_name && host_name[0]) { - error = lookup_ip(host_name, &sin.sin_addr); - if (error) { - goto exit; - } + if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) { + goto exit; + } + + ok = true; + +exit: + if (!ok) { + memset(sinp, 0, sizeof *sinp); + } + free(target); + return ok; +} + + +/* Opens a non-blocking IPv4 socket of the specified 'style', binds to + * 'target', and listens for incoming connections. Parses 'target' in the same + * way was inet_parse_passive(). + * + * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). + * + * For TCP, the socket will have SO_REUSEADDR turned on. + * + * On success, returns a non-negative file descriptor. On failure, returns a + * negative errno value. + * + * If 'sinp' is non-null, then on success the bound address is stored into + * '*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, uint8_t dscp) +{ + bool kernel_chooses_port; + struct sockaddr_in sin; + int fd = 0, error; + unsigned int yes = 1; + + if (!inet_parse_passive(target, default_port, &sin)) { + return -EAFNOSUPPORT; } /* Create non-blocking socket, set SO_REUSEADDR. */ fd = socket(AF_INET, style, 0); if (fd < 0) { error = errno; - VLOG_ERR("%s: socket: %s", target_, strerror(error)); - goto exit; + VLOG_ERR("%s: socket: %s", target, ovs_strerror(error)); + return -error; } error = set_nonblocking(fd); if (error) { - goto exit_close; + goto error; } if (style == SOCK_STREAM && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) { error = errno; - VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error)); - goto exit_close; + VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", + target, ovs_strerror(error)); + goto error; } /* Bind. */ if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) { error = errno; - VLOG_ERR("%s: bind: %s", target_, strerror(error)); - goto exit_close; + VLOG_ERR("%s: bind: %s", target, ovs_strerror(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, ovs_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 exit_close; + VLOG_ERR("%s: listen: %s", target, ovs_strerror(error)); + goto error; } - if (sinp) { + kernel_chooses_port = sin.sin_port == htons(0); + if (sinp || kernel_chooses_port) { socklen_t sin_len = sizeof sin; - if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){ + if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0) { error = errno; - VLOG_ERR("%s: getsockname: %s", target_, strerror(error)); - goto exit_close; + VLOG_ERR("%s: getsockname: %s", target, ovs_strerror(error)); + goto error; } if (sin.sin_family != AF_INET || sin_len != sizeof sin) { - VLOG_ERR("%s: getsockname: invalid socket name", target_); - goto exit_close; + error = EAFNOSUPPORT; + VLOG_ERR("%s: getsockname: invalid socket name", target); + goto error; + } + if (sinp) { + *sinp = sin; + } + if (kernel_chooses_port) { + VLOG_INFO("%s: listening on port %"PRIu16, + target, ntohs(sin.sin_port)); } - *sinp = sin; } - error = 0; - goto exit; + return fd; -exit_close: +error: close(fd); -exit: - free(target); - return error ? -error : fd; + return -error; } /* Returns a readable and writable fd for /dev/null, if successful, otherwise @@ -669,7 +807,7 @@ get_null_fd(void) null_fd = open("/dev/null", O_RDWR); if (null_fd < 0) { int error = errno; - VLOG_ERR("could not open /dev/null: %s", strerror(error)); + VLOG_ERR("could not open /dev/null: %s", ovs_strerror(error)); return -error; } } @@ -737,13 +875,13 @@ fsync_parent_dir(const char *file_name) * really an error. */ } else { error = errno; - VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error)); + VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error)); } } close(fd); } else { error = errno; - VLOG_ERR("%s: open failed (%s)", dir, strerror(error)); + VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error)); } free(dir); @@ -781,6 +919,428 @@ void xpipe(int fds[2]) { if (pipe(fds)) { - VLOG_FATAL("failed to create pipe (%s)", strerror(errno)); + VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno)); + } +} + +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)", ovs_strerror(errno)); + } +} + +static int +getsockopt_int(int fd, int level, int option, const char *optname, int *valuep) +{ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); + socklen_t len; + int value; + int error; + + len = sizeof value; + if (getsockopt(fd, level, option, &value, &len)) { + error = errno; + VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, ovs_strerror(error)); + } else if (len != sizeof value) { + error = EINVAL; + VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %zu)", + optname, (unsigned int) len, sizeof value); + } else { + error = 0; + } + + *valuep = error ? 0 : value; + return error; +} + +static void +describe_sockaddr(struct ds *string, int fd, + int (*getaddr)(int, struct sockaddr *, socklen_t *)) +{ + struct sockaddr_storage ss; + socklen_t len = sizeof ss; + + if (!getaddr(fd, (struct sockaddr *) &ss, &len)) { + if (ss.ss_family == AF_INET) { + struct sockaddr_in sin; + + memcpy(&sin, &ss, sizeof sin); + ds_put_format(string, IP_FMT":%"PRIu16, + 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; + size_t maxlen; + + memcpy(&sun, &ss, sizeof sun); + maxlen = len - offsetof(struct sockaddr_un, sun_path); + null = memchr(sun.sun_path, '\0', maxlen); + ds_put_buffer(string, sun.sun_path, + null ? null - sun.sun_path : maxlen); + } +#ifdef HAVE_NETLINK + else if (ss.ss_family == AF_NETLINK) { + int protocol; + +/* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version + * of the Linux kernel headers in use at build time. */ +#ifndef SO_PROTOCOL +#define SO_PROTOCOL 38 +#endif + + if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL", + &protocol)) { + switch (protocol) { + case NETLINK_ROUTE: + ds_put_cstr(string, "NETLINK_ROUTE"); + break; + + case NETLINK_GENERIC: + ds_put_cstr(string, "NETLINK_GENERIC"); + break; + + default: + ds_put_format(string, "AF_NETLINK family %d", protocol); + break; + } + } else { + ds_put_cstr(string, "AF_NETLINK"); + } + } +#endif +#if AF_PACKET && LINUX_DATAPATH + else if (ss.ss_family == AF_PACKET) { + struct sockaddr_ll sll; + + memcpy(&sll, &ss, sizeof sll); + ds_put_cstr(string, "AF_PACKET"); + if (sll.sll_ifindex) { + char name[IFNAMSIZ]; + + if (if_indextoname(sll.sll_ifindex, name)) { + ds_put_format(string, "(%s)", name); + } else { + ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex); + } + } + if (sll.sll_protocol) { + ds_put_format(string, "(protocol=0x%"PRIu16")", + ntohs(sll.sll_protocol)); + } + } +#endif + else if (ss.ss_family == AF_UNSPEC) { + ds_put_cstr(string, "AF_UNSPEC"); + } else { + ds_put_format(string, "AF_%d", (int) ss.ss_family); + } + } +} + + +#ifdef LINUX_DATAPATH +static void +put_fd_filename(struct ds *string, int fd) +{ + char buf[1024]; + char *linkname; + int n; + + linkname = xasprintf("/proc/self/fd/%d", fd); + n = readlink(linkname, buf, sizeof buf); + if (n > 0) { + ds_put_char(string, ' '); + ds_put_buffer(string, buf, n); + if (n > sizeof buf) { + ds_put_cstr(string, "..."); + } + } + free(linkname); +} +#endif + +/* Returns a malloc()'d string describing 'fd', for use in logging. */ +char * +describe_fd(int fd) +{ + struct ds string; + struct stat s; + + ds_init(&string); + if (fstat(fd, &s)) { + ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno)); + } else if (S_ISSOCK(s.st_mode)) { + describe_sockaddr(&string, fd, getsockname); + ds_put_cstr(&string, "<->"); + describe_sockaddr(&string, fd, getpeername); + } else { + ds_put_cstr(&string, (isatty(fd) ? "tty" + : S_ISDIR(s.st_mode) ? "directory" + : S_ISCHR(s.st_mode) ? "character device" + : S_ISBLK(s.st_mode) ? "block device" + : S_ISREG(s.st_mode) ? "file" + : S_ISFIFO(s.st_mode) ? "FIFO" + : S_ISLNK(s.st_mode) ? "symbolic link" + : "unknown")); +#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; }