PageRenderTime 98ms CodeModel.GetById 20ms app.highlight 65ms RepoModel.GetById 1ms app.codeStats 1ms

/lib/netlink-socket.c

https://github.com/noironetworks/ovs
C | 1817 lines | 1321 code | 194 blank | 302 comment | 238 complexity | b55adf649101602b9dd80fe6ae7a23a4 MD5 | raw file

Large files files are truncated, but you can click here to view the full file

   1/*
   2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
   3 *
   4 * Licensed under the Apache License, Version 2.0 (the "License");
   5 * you may not use this file except in compliance with the License.
   6 * You may obtain a copy of the License at:
   7 *
   8 *     http://www.apache.org/licenses/LICENSE-2.0
   9 *
  10 * Unless required by applicable law or agreed to in writing, software
  11 * distributed under the License is distributed on an "AS IS" BASIS,
  12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13 * See the License for the specific language governing permissions and
  14 * limitations under the License.
  15 */
  16
  17#include <config.h>
  18#include "netlink-socket.h"
  19#include <errno.h>
  20#include <inttypes.h>
  21#include <stdlib.h>
  22#include <sys/types.h>
  23#include <sys/uio.h>
  24#include <unistd.h>
  25#include "coverage.h"
  26#include "dynamic-string.h"
  27#include "hash.h"
  28#include "hmap.h"
  29#include "netlink.h"
  30#include "netlink-protocol.h"
  31#include "odp-netlink.h"
  32#include "ofpbuf.h"
  33#include "ovs-thread.h"
  34#include "poll-loop.h"
  35#include "seq.h"
  36#include "socket-util.h"
  37#include "util.h"
  38#include "openvswitch/vlog.h"
  39
  40VLOG_DEFINE_THIS_MODULE(netlink_socket);
  41
  42COVERAGE_DEFINE(netlink_overflow);
  43COVERAGE_DEFINE(netlink_received);
  44COVERAGE_DEFINE(netlink_recv_jumbo);
  45COVERAGE_DEFINE(netlink_sent);
  46
  47/* Linux header file confusion causes this to be undefined. */
  48#ifndef SOL_NETLINK
  49#define SOL_NETLINK 270
  50#endif
  51
  52#ifdef _WIN32
  53static struct ovs_mutex portid_mutex = OVS_MUTEX_INITIALIZER;
  54static uint32_t g_last_portid = 0;
  55
  56/* Port IDs must be unique! */
  57static uint32_t
  58portid_next(void)
  59    OVS_GUARDED_BY(portid_mutex)
  60{
  61    g_last_portid++;
  62    return g_last_portid;
  63}
  64#endif /* _WIN32 */
  65
  66/* A single (bad) Netlink message can in theory dump out many, many log
  67 * messages, so the burst size is set quite high here to avoid missing useful
  68 * information.  Also, at high logging levels we log *all* Netlink messages. */
  69static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600);
  70
  71static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n);
  72static void log_nlmsg(const char *function, int error,
  73                      const void *message, size_t size, int protocol);
  74#ifdef _WIN32
  75static int get_sock_pid_from_kernel(struct nl_sock *sock);
  76#endif
  77
  78/* Netlink sockets. */
  79
  80struct nl_sock {
  81#ifdef _WIN32
  82    HANDLE handle;
  83    OVERLAPPED overlapped;
  84    DWORD read_ioctl;
  85#else
  86    int fd;
  87#endif
  88    uint32_t next_seq;
  89    uint32_t pid;
  90    int protocol;
  91    unsigned int rcvbuf;        /* Receive buffer size (SO_RCVBUF). */
  92};
  93
  94/* Compile-time limit on iovecs, so that we can allocate a maximum-size array
  95 * of iovecs on the stack. */
  96#define MAX_IOVS 128
  97
  98/* Maximum number of iovecs that may be passed to sendmsg, capped at a
  99 * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
 100 *
 101 * Initialized by nl_sock_create(). */
 102static int max_iovs;
 103
 104static int nl_pool_alloc(int protocol, struct nl_sock **sockp);
 105static void nl_pool_release(struct nl_sock *);
 106
 107/* Creates a new netlink socket for the given netlink 'protocol'
 108 * (NETLINK_ROUTE, NETLINK_GENERIC, ...).  Returns 0 and sets '*sockp' to the
 109 * new socket if successful, otherwise returns a positive errno value. */
 110int
 111nl_sock_create(int protocol, struct nl_sock **sockp)
 112{
 113    static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
 114    struct nl_sock *sock;
 115#ifndef _WIN32
 116    struct sockaddr_nl local, remote;
 117#endif
 118    socklen_t local_size;
 119    int rcvbuf;
 120    int retval = 0;
 121
 122    if (ovsthread_once_start(&once)) {
 123        int save_errno = errno;
 124        errno = 0;
 125
 126        max_iovs = sysconf(_SC_UIO_MAXIOV);
 127        if (max_iovs < _XOPEN_IOV_MAX) {
 128            if (max_iovs == -1 && errno) {
 129                VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", ovs_strerror(errno));
 130            }
 131            max_iovs = _XOPEN_IOV_MAX;
 132        } else if (max_iovs > MAX_IOVS) {
 133            max_iovs = MAX_IOVS;
 134        }
 135
 136        errno = save_errno;
 137        ovsthread_once_done(&once);
 138    }
 139
 140    *sockp = NULL;
 141    sock = xmalloc(sizeof *sock);
 142
 143#ifdef _WIN32
 144    sock->handle = CreateFile(OVS_DEVICE_NAME_USER,
 145                              GENERIC_READ | GENERIC_WRITE,
 146                              FILE_SHARE_READ | FILE_SHARE_WRITE,
 147                              NULL, OPEN_EXISTING,
 148                              FILE_FLAG_OVERLAPPED, NULL);
 149
 150    if (sock->handle == INVALID_HANDLE_VALUE) {
 151        VLOG_ERR("fcntl: %s", ovs_lasterror_to_string());
 152        goto error;
 153    }
 154
 155    memset(&sock->overlapped, 0, sizeof sock->overlapped);
 156    sock->overlapped.hEvent = CreateEvent(NULL, FALSE, FALSE, NULL);
 157    if (sock->overlapped.hEvent == NULL) {
 158        VLOG_ERR("fcntl: %s", ovs_lasterror_to_string());
 159        goto error;
 160    }
 161    /* Initialize the type/ioctl to Generic */
 162    sock->read_ioctl = OVS_IOCTL_READ;
 163#else
 164    sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
 165    if (sock->fd < 0) {
 166        VLOG_ERR("fcntl: %s", ovs_strerror(errno));
 167        goto error;
 168    }
 169#endif
 170
 171    sock->protocol = protocol;
 172    sock->next_seq = 1;
 173
 174    rcvbuf = 1024 * 1024;
 175#ifdef _WIN32
 176    sock->rcvbuf = rcvbuf;
 177    retval = get_sock_pid_from_kernel(sock);
 178    if (retval != 0) {
 179        goto error;
 180    }
 181#else
 182    if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
 183                   &rcvbuf, sizeof rcvbuf)) {
 184        /* Only root can use SO_RCVBUFFORCE.  Everyone else gets EPERM.
 185         * Warn only if the failure is therefore unexpected. */
 186        if (errno != EPERM) {
 187            VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed "
 188                         "(%s)", rcvbuf, ovs_strerror(errno));
 189        }
 190    }
 191
 192    retval = get_socket_rcvbuf(sock->fd);
 193    if (retval < 0) {
 194        retval = -retval;
 195        goto error;
 196    }
 197    sock->rcvbuf = retval;
 198
 199    /* Connect to kernel (pid 0) as remote address. */
 200    memset(&remote, 0, sizeof remote);
 201    remote.nl_family = AF_NETLINK;
 202    remote.nl_pid = 0;
 203    if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
 204        VLOG_ERR("connect(0): %s", ovs_strerror(errno));
 205        goto error;
 206    }
 207
 208    /* Obtain pid assigned by kernel. */
 209    local_size = sizeof local;
 210    if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
 211        VLOG_ERR("getsockname: %s", ovs_strerror(errno));
 212        goto error;
 213    }
 214    if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
 215        VLOG_ERR("getsockname returned bad Netlink name");
 216        retval = EINVAL;
 217        goto error;
 218    }
 219    sock->pid = local.nl_pid;
 220#endif
 221
 222    *sockp = sock;
 223    return 0;
 224
 225error:
 226    if (retval == 0) {
 227        retval = errno;
 228        if (retval == 0) {
 229            retval = EINVAL;
 230        }
 231    }
 232#ifdef _WIN32
 233    if (sock->overlapped.hEvent) {
 234        CloseHandle(sock->overlapped.hEvent);
 235    }
 236    if (sock->handle != INVALID_HANDLE_VALUE) {
 237        CloseHandle(sock->handle);
 238    }
 239#else
 240    if (sock->fd >= 0) {
 241        close(sock->fd);
 242    }
 243#endif
 244    free(sock);
 245    return retval;
 246}
 247
 248/* Creates a new netlink socket for the same protocol as 'src'.  Returns 0 and
 249 * sets '*sockp' to the new socket if successful, otherwise returns a positive
 250 * errno value.  */
 251int
 252nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp)
 253{
 254    return nl_sock_create(src->protocol, sockp);
 255}
 256
 257/* Destroys netlink socket 'sock'. */
 258void
 259nl_sock_destroy(struct nl_sock *sock)
 260{
 261    if (sock) {
 262#ifdef _WIN32
 263        if (sock->overlapped.hEvent) {
 264            CloseHandle(sock->overlapped.hEvent);
 265        }
 266        CloseHandle(sock->handle);
 267#else
 268        close(sock->fd);
 269#endif
 270        free(sock);
 271    }
 272}
 273
 274#ifdef _WIN32
 275/* Reads the pid for 'sock' generated in the kernel datapath. The function
 276 * follows a transaction semantic. Eventually this function should call into
 277 * nl_transact. */
 278static int
 279get_sock_pid_from_kernel(struct nl_sock *sock)
 280{
 281    struct nl_transaction txn;
 282    struct ofpbuf request;
 283    uint64_t request_stub[128];
 284    struct ofpbuf reply;
 285    uint64_t reply_stub[128];
 286    struct ovs_header *ovs_header;
 287    struct nlmsghdr *nlmsg;
 288    uint32_t seq;
 289    int retval;
 290    DWORD bytes;
 291    int ovs_msg_size = sizeof (struct nlmsghdr) + sizeof (struct genlmsghdr) +
 292                       sizeof (struct ovs_header);
 293
 294    ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
 295    txn.request = &request;
 296    ofpbuf_use_stub(&reply, reply_stub, sizeof reply_stub);
 297    txn.reply = &reply;
 298
 299    seq = nl_sock_allocate_seq(sock, 1);
 300    nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
 301                          OVS_CTRL_CMD_WIN_GET_PID, OVS_WIN_CONTROL_VERSION);
 302    nlmsg = nl_msg_nlmsghdr(txn.request);
 303    nlmsg->nlmsg_seq = seq;
 304
 305    ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
 306    ovs_header->dp_ifindex = 0;
 307    ovs_header = ofpbuf_put_uninit(&reply, ovs_msg_size);
 308
 309    if (!DeviceIoControl(sock->handle, OVS_IOCTL_TRANSACT,
 310                         ofpbuf_data(txn.request), ofpbuf_size(txn.request),
 311                         ofpbuf_data(txn.reply), ofpbuf_size(txn.reply),
 312                         &bytes, NULL)) {
 313        retval = EINVAL;
 314        goto done;
 315    } else {
 316        if (bytes < ovs_msg_size) {
 317            retval = EINVAL;
 318            goto done;
 319        }
 320
 321        nlmsg = nl_msg_nlmsghdr(txn.reply);
 322        if (nlmsg->nlmsg_seq != seq) {
 323            retval = EINVAL;
 324            goto done;
 325        }
 326        sock->pid = nlmsg->nlmsg_pid;
 327    }
 328    retval = 0;
 329
 330done:
 331    ofpbuf_uninit(&request);
 332    ofpbuf_uninit(&reply);
 333    return retval;
 334}
 335#endif  /* _WIN32 */
 336
 337#ifdef _WIN32
 338static int __inline
 339nl_sock_mcgroup(struct nl_sock *sock, unsigned int multicast_group, bool join)
 340{
 341    struct ofpbuf request;
 342    uint64_t request_stub[128];
 343    struct ovs_header *ovs_header;
 344    struct nlmsghdr *nlmsg;
 345    int error;
 346
 347    ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
 348
 349    nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
 350                          OVS_CTRL_CMD_MC_SUBSCRIBE_REQ,
 351                          OVS_WIN_CONTROL_VERSION);
 352
 353    ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
 354    ovs_header->dp_ifindex = 0;
 355
 356    nl_msg_put_u32(&request, OVS_NL_ATTR_MCAST_GRP, multicast_group);
 357    nl_msg_put_u8(&request, OVS_NL_ATTR_MCAST_JOIN, join ? 1 : 0);
 358
 359    error = nl_sock_send(sock, &request, true);
 360    ofpbuf_uninit(&request);
 361    return error;
 362}
 363#endif
 364/* Tries to add 'sock' as a listener for 'multicast_group'.  Returns 0 if
 365 * successful, otherwise a positive errno value.
 366 *
 367 * A socket that is subscribed to a multicast group that receives asynchronous
 368 * notifications must not be used for Netlink transactions or dumps, because
 369 * transactions and dumps can cause notifications to be lost.
 370 *
 371 * Multicast group numbers are always positive.
 372 *
 373 * It is not an error to attempt to join a multicast group to which a socket
 374 * already belongs. */
 375int
 376nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
 377{
 378#ifdef _WIN32
 379    /* Set the socket type as a "multicast" socket */
 380    sock->read_ioctl = OVS_IOCTL_READ_EVENT;
 381    int error = nl_sock_mcgroup(sock, multicast_group, true);
 382    if (error) {
 383        sock->read_ioctl = OVS_IOCTL_READ;
 384        VLOG_WARN("could not join multicast group %u (%s)",
 385                  multicast_group, ovs_strerror(error));
 386        return error;
 387    }
 388#else
 389    if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
 390                   &multicast_group, sizeof multicast_group) < 0) {
 391        VLOG_WARN("could not join multicast group %u (%s)",
 392                  multicast_group, ovs_strerror(errno));
 393        return errno;
 394    }
 395#endif
 396    return 0;
 397}
 398
 399#ifdef _WIN32
 400int
 401nl_sock_subscribe_packets(struct nl_sock *sock)
 402{
 403    int error;
 404
 405    if (sock->read_ioctl != OVS_IOCTL_READ) {
 406        return EINVAL;
 407    }
 408
 409    error = nl_sock_subscribe_packet__(sock, true);
 410    if (error) {
 411        VLOG_WARN("could not unsubscribe packets (%s)",
 412                  ovs_strerror(errno));
 413        return error;
 414    }
 415    sock->read_ioctl = OVS_IOCTL_READ_PACKET;
 416
 417    return 0;
 418}
 419
 420int
 421nl_sock_unsubscribe_packets(struct nl_sock *sock)
 422{
 423    ovs_assert(sock->read_ioctl == OVS_IOCTL_READ_PACKET);
 424
 425    int error = nl_sock_subscribe_packet__(sock, false);
 426    if (error) {
 427        VLOG_WARN("could not subscribe to packets (%s)",
 428                  ovs_strerror(errno));
 429        return error;
 430    }
 431
 432    sock->read_ioctl = OVS_IOCTL_READ;
 433    return 0;
 434}
 435
 436int
 437nl_sock_subscribe_packet__(struct nl_sock *sock, bool subscribe)
 438{
 439    struct ofpbuf request;
 440    uint64_t request_stub[128];
 441    struct ovs_header *ovs_header;
 442    struct nlmsghdr *nlmsg;
 443    int error;
 444
 445    ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
 446    nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
 447                          OVS_CTRL_CMD_PACKET_SUBSCRIBE_REQ,
 448                          OVS_WIN_CONTROL_VERSION);
 449
 450    ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
 451    ovs_header->dp_ifindex = 0;
 452    nl_msg_put_u8(&request, OVS_NL_ATTR_PACKET_SUBSCRIBE, subscribe ? 1 : 0);
 453    nl_msg_put_u32(&request, OVS_NL_ATTR_PACKET_PID, sock->pid);
 454
 455    error = nl_sock_send(sock, &request, true);
 456    ofpbuf_uninit(&request);
 457    return error;
 458}
 459#endif
 460
 461/* Tries to make 'sock' stop listening to 'multicast_group'.  Returns 0 if
 462 * successful, otherwise a positive errno value.
 463 *
 464 * Multicast group numbers are always positive.
 465 *
 466 * It is not an error to attempt to leave a multicast group to which a socket
 467 * does not belong.
 468 *
 469 * On success, reading from 'sock' will still return any messages that were
 470 * received on 'multicast_group' before the group was left. */
 471int
 472nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
 473{
 474#ifdef _WIN32
 475    int error = nl_sock_mcgroup(sock, multicast_group, false);
 476    if (error) {
 477        VLOG_WARN("could not leave multicast group %u (%s)",
 478                   multicast_group, ovs_strerror(error));
 479        return error;
 480    }
 481    sock->read_ioctl = OVS_IOCTL_READ;
 482#else
 483    if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
 484                   &multicast_group, sizeof multicast_group) < 0) {
 485        VLOG_WARN("could not leave multicast group %u (%s)",
 486                  multicast_group, ovs_strerror(errno));
 487        return errno;
 488    }
 489#endif
 490    return 0;
 491}
 492
 493static int
 494nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
 495               uint32_t nlmsg_seq, bool wait)
 496{
 497    struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
 498    int error;
 499
 500    nlmsg->nlmsg_len = ofpbuf_size(msg);
 501    nlmsg->nlmsg_seq = nlmsg_seq;
 502    nlmsg->nlmsg_pid = sock->pid;
 503    do {
 504        int retval;
 505#ifdef _WIN32
 506        DWORD bytes;
 507
 508        if (!DeviceIoControl(sock->handle, OVS_IOCTL_WRITE,
 509                             ofpbuf_data(msg), ofpbuf_size(msg), NULL, 0,
 510                             &bytes, NULL)) {
 511            retval = -1;
 512            /* XXX: Map to a more appropriate error based on GetLastError(). */
 513            errno = EINVAL;
 514        } else {
 515            retval = ofpbuf_size(msg);
 516        }
 517#else
 518        retval = send(sock->fd, ofpbuf_data(msg), ofpbuf_size(msg),
 519                      wait ? 0 : MSG_DONTWAIT);
 520#endif
 521        error = retval < 0 ? errno : 0;
 522    } while (error == EINTR);
 523    log_nlmsg(__func__, error, ofpbuf_data(msg), ofpbuf_size(msg), sock->protocol);
 524    if (!error) {
 525        COVERAGE_INC(netlink_sent);
 526    }
 527    return error;
 528}
 529
 530/* Tries to send 'msg', which must contain a Netlink message, to the kernel on
 531 * 'sock'.  nlmsg_len in 'msg' will be finalized to match ofpbuf_size(msg), nlmsg_pid
 532 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
 533 * sequence number, before the message is sent.
 534 *
 535 * Returns 0 if successful, otherwise a positive errno value.  If
 536 * 'wait' is true, then the send will wait until buffer space is ready;
 537 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
 538int
 539nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
 540{
 541    return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
 542}
 543
 544/* Tries to send 'msg', which must contain a Netlink message, to the kernel on
 545 * 'sock'.  nlmsg_len in 'msg' will be finalized to match ofpbuf_size(msg), nlmsg_pid
 546 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
 547 * 'nlmsg_seq', before the message is sent.
 548 *
 549 * Returns 0 if successful, otherwise a positive errno value.  If
 550 * 'wait' is true, then the send will wait until buffer space is ready;
 551 * otherwise, returns EAGAIN if the 'sock' send buffer is full.
 552 *
 553 * This function is suitable for sending a reply to a request that was received
 554 * with sequence number 'nlmsg_seq'.  Otherwise, use nl_sock_send() instead. */
 555int
 556nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
 557                 uint32_t nlmsg_seq, bool wait)
 558{
 559    return nl_sock_send__(sock, msg, nlmsg_seq, wait);
 560}
 561
 562static int
 563nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
 564{
 565    /* We can't accurately predict the size of the data to be received.  The
 566     * caller is supposed to have allocated enough space in 'buf' to handle the
 567     * "typical" case.  To handle exceptions, we make available enough space in
 568     * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
 569     * figure since that's the maximum length of a Netlink attribute). */
 570    struct nlmsghdr *nlmsghdr;
 571    uint8_t tail[65536];
 572    struct iovec iov[2];
 573    struct msghdr msg;
 574    ssize_t retval;
 575    int error;
 576
 577    ovs_assert(buf->allocated >= sizeof *nlmsghdr);
 578    ofpbuf_clear(buf);
 579
 580    iov[0].iov_base = ofpbuf_base(buf);
 581    iov[0].iov_len = buf->allocated;
 582    iov[1].iov_base = tail;
 583    iov[1].iov_len = sizeof tail;
 584
 585    memset(&msg, 0, sizeof msg);
 586    msg.msg_iov = iov;
 587    msg.msg_iovlen = 2;
 588
 589    /* Receive a Netlink message from the kernel.
 590     *
 591     * This works around a kernel bug in which the kernel returns an error code
 592     * as if it were the number of bytes read.  It doesn't actually modify
 593     * anything in the receive buffer in that case, so we can initialize the
 594     * Netlink header with an impossible message length and then, upon success,
 595     * check whether it changed. */
 596    nlmsghdr = ofpbuf_base(buf);
 597    do {
 598        nlmsghdr->nlmsg_len = UINT32_MAX;
 599#ifdef _WIN32
 600        DWORD bytes;
 601        if (!DeviceIoControl(sock->handle, sock->read_ioctl,
 602                             NULL, 0, tail, sizeof tail, &bytes, NULL)) {
 603            retval = -1;
 604            errno = EINVAL;
 605        } else {
 606            retval = bytes;
 607            if (retval == 0) {
 608                retval = -1;
 609                errno = EAGAIN;
 610            } else {
 611                if (retval >= buf->allocated) {
 612                    ofpbuf_reinit(buf, retval);
 613                    nlmsghdr = ofpbuf_base(buf);
 614                    nlmsghdr->nlmsg_len = UINT32_MAX;
 615                }
 616                memcpy(ofpbuf_data(buf), tail, retval);
 617                ofpbuf_set_size(buf, retval);
 618            }
 619        }
 620#else
 621        retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
 622#endif
 623        error = (retval < 0 ? errno
 624                 : retval == 0 ? ECONNRESET /* not possible? */
 625                 : nlmsghdr->nlmsg_len != UINT32_MAX ? 0
 626                 : retval);
 627    } while (error == EINTR);
 628    if (error) {
 629        if (error == ENOBUFS) {
 630            /* Socket receive buffer overflow dropped one or more messages that
 631             * the kernel tried to send to us. */
 632            COVERAGE_INC(netlink_overflow);
 633        }
 634        return error;
 635    }
 636
 637    if (msg.msg_flags & MSG_TRUNC) {
 638        VLOG_ERR_RL(&rl, "truncated message (longer than %"PRIuSIZE" bytes)",
 639                    sizeof tail);
 640        return E2BIG;
 641    }
 642
 643    if (retval < sizeof *nlmsghdr
 644        || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
 645        || nlmsghdr->nlmsg_len > retval) {
 646        VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE" bytes < %"PRIuSIZE")",
 647                    retval, sizeof *nlmsghdr);
 648        return EPROTO;
 649    }
 650#ifndef _WIN32
 651    ofpbuf_set_size(buf, MIN(retval, buf->allocated));
 652    if (retval > buf->allocated) {
 653        COVERAGE_INC(netlink_recv_jumbo);
 654        ofpbuf_put(buf, tail, retval - buf->allocated);
 655    }
 656#endif
 657
 658    log_nlmsg(__func__, 0, ofpbuf_data(buf), ofpbuf_size(buf), sock->protocol);
 659    COVERAGE_INC(netlink_received);
 660
 661    return 0;
 662}
 663
 664/* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'.  If
 665 * 'wait' is true, waits for a message to be ready.  Otherwise, fails with
 666 * EAGAIN if the 'sock' receive buffer is empty.
 667 *
 668 * The caller must have initialized 'buf' with an allocation of at least
 669 * NLMSG_HDRLEN bytes.  For best performance, the caller should allocate enough
 670 * space for a "typical" message.
 671 *
 672 * On success, returns 0 and replaces 'buf''s previous content by the received
 673 * message.  This function expands 'buf''s allocated memory, as necessary, to
 674 * hold the actual size of the received message.
 675 *
 676 * On failure, returns a positive errno value and clears 'buf' to zero length.
 677 * 'buf' retains its previous memory allocation.
 678 *
 679 * Regardless of success or failure, this function resets 'buf''s headroom to
 680 * 0. */
 681int
 682nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
 683{
 684    return nl_sock_recv__(sock, buf, wait);
 685}
 686
 687static void
 688nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
 689                        int error)
 690{
 691    size_t i;
 692
 693    for (i = 0; i < n; i++) {
 694        struct nl_transaction *txn = transactions[i];
 695
 696        txn->error = error;
 697        if (txn->reply) {
 698            ofpbuf_clear(txn->reply);
 699        }
 700    }
 701}
 702
 703static int
 704nl_sock_transact_multiple__(struct nl_sock *sock,
 705                            struct nl_transaction **transactions, size_t n,
 706                            size_t *done)
 707{
 708    uint64_t tmp_reply_stub[1024 / 8];
 709    struct nl_transaction tmp_txn;
 710    struct ofpbuf tmp_reply;
 711
 712    uint32_t base_seq;
 713    struct iovec iovs[MAX_IOVS];
 714    struct msghdr msg;
 715    int error;
 716    int i;
 717
 718    base_seq = nl_sock_allocate_seq(sock, n);
 719    *done = 0;
 720    for (i = 0; i < n; i++) {
 721        struct nl_transaction *txn = transactions[i];
 722        struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
 723
 724        nlmsg->nlmsg_len = ofpbuf_size(txn->request);
 725        nlmsg->nlmsg_seq = base_seq + i;
 726        nlmsg->nlmsg_pid = sock->pid;
 727
 728        iovs[i].iov_base = ofpbuf_data(txn->request);
 729        iovs[i].iov_len = ofpbuf_size(txn->request);
 730    }
 731
 732#ifndef _WIN32
 733    memset(&msg, 0, sizeof msg);
 734    msg.msg_iov = iovs;
 735    msg.msg_iovlen = n;
 736    do {
 737        error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
 738    } while (error == EINTR);
 739
 740    for (i = 0; i < n; i++) {
 741        struct nl_transaction *txn = transactions[i];
 742
 743        log_nlmsg(__func__, error, ofpbuf_data(txn->request),
 744                  ofpbuf_size(txn->request), sock->protocol);
 745    }
 746    if (!error) {
 747        COVERAGE_ADD(netlink_sent, n);
 748    }
 749
 750    if (error) {
 751        return error;
 752    }
 753
 754    ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
 755    tmp_txn.request = NULL;
 756    tmp_txn.reply = &tmp_reply;
 757    tmp_txn.error = 0;
 758    while (n > 0) {
 759        struct nl_transaction *buf_txn, *txn;
 760        uint32_t seq;
 761
 762        /* Find a transaction whose buffer we can use for receiving a reply.
 763         * If no such transaction is left, use tmp_txn. */
 764        buf_txn = &tmp_txn;
 765        for (i = 0; i < n; i++) {
 766            if (transactions[i]->reply) {
 767                buf_txn = transactions[i];
 768                break;
 769            }
 770        }
 771
 772        /* Receive a reply. */
 773        error = nl_sock_recv__(sock, buf_txn->reply, false);
 774        if (error) {
 775            if (error == EAGAIN) {
 776                nl_sock_record_errors__(transactions, n, 0);
 777                *done += n;
 778                error = 0;
 779            }
 780            break;
 781        }
 782
 783        /* Match the reply up with a transaction. */
 784        seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
 785        if (seq < base_seq || seq >= base_seq + n) {
 786            VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
 787            continue;
 788        }
 789        i = seq - base_seq;
 790        txn = transactions[i];
 791
 792        /* Fill in the results for 'txn'. */
 793        if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
 794            if (txn->reply) {
 795                ofpbuf_clear(txn->reply);
 796            }
 797            if (txn->error) {
 798                VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
 799                            error, ovs_strerror(txn->error));
 800            }
 801        } else {
 802            txn->error = 0;
 803            if (txn->reply && txn != buf_txn) {
 804                /* Swap buffers. */
 805                struct ofpbuf *reply = buf_txn->reply;
 806                buf_txn->reply = txn->reply;
 807                txn->reply = reply;
 808            }
 809        }
 810
 811        /* Fill in the results for transactions before 'txn'.  (We have to do
 812         * this after the results for 'txn' itself because of the buffer swap
 813         * above.) */
 814        nl_sock_record_errors__(transactions, i, 0);
 815
 816        /* Advance. */
 817        *done += i + 1;
 818        transactions += i + 1;
 819        n -= i + 1;
 820        base_seq += i + 1;
 821    }
 822    ofpbuf_uninit(&tmp_reply);
 823#else
 824    error = 0;
 825    uint8_t reply_buf[65536];
 826    for (i = 0; i < n; i++) {
 827        DWORD reply_len;
 828        struct nl_transaction *txn = transactions[i];
 829        struct nlmsghdr *request_nlmsg, *reply_nlmsg;
 830
 831        if (!DeviceIoControl(sock->handle, OVS_IOCTL_TRANSACT,
 832                             ofpbuf_data(txn->request),
 833                             ofpbuf_size(txn->request),
 834                             reply_buf, sizeof reply_buf,
 835                             &reply_len, NULL)) {
 836            /* XXX: Map to a more appropriate error. */
 837            error = EINVAL;
 838            break;
 839        }
 840
 841        if (reply_len < sizeof *reply_nlmsg) {
 842            nl_sock_record_errors__(transactions, n, 0);
 843            VLOG_DBG_RL(&rl, "insufficient length of reply %#"PRIu32
 844                " for seq: %#"PRIx32, reply_len, request_nlmsg->nlmsg_seq);
 845            break;
 846        }
 847
 848        /* Validate the sequence number in the reply. */
 849        request_nlmsg = nl_msg_nlmsghdr(txn->request);
 850        reply_nlmsg = (struct nlmsghdr *)reply_buf;
 851
 852        if (request_nlmsg->nlmsg_seq != reply_nlmsg->nlmsg_seq) {
 853            ovs_assert(request_nlmsg->nlmsg_seq == reply_nlmsg->nlmsg_seq);
 854            VLOG_DBG_RL(&rl, "mismatched seq request %#"PRIx32
 855                ", reply %#"PRIx32, request_nlmsg->nlmsg_seq,
 856                reply_nlmsg->nlmsg_seq);
 857            break;
 858        }
 859
 860        /* Handle errors embedded within the netlink message. */
 861        ofpbuf_use_stub(&tmp_reply, reply_buf, sizeof reply_buf);
 862        ofpbuf_set_size(&tmp_reply, sizeof reply_buf);
 863        if (nl_msg_nlmsgerr(&tmp_reply, &txn->error)) {
 864            if (txn->reply) {
 865                ofpbuf_clear(txn->reply);
 866            }
 867            if (txn->error) {
 868                VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
 869                            error, ovs_strerror(txn->error));
 870            }
 871        } else {
 872            txn->error = 0;
 873            if (txn->reply) {
 874                /* Copy the reply to the buffer specified by the caller. */
 875                if (reply_len > txn->reply->allocated) {
 876                    ofpbuf_reinit(txn->reply, reply_len);
 877                }
 878                memcpy(ofpbuf_data(txn->reply), reply_buf, reply_len);
 879                ofpbuf_set_size(txn->reply, reply_len);
 880            }
 881        }
 882        ofpbuf_uninit(&tmp_reply);
 883
 884        /* Count the number of successful transactions. */
 885        (*done)++;
 886
 887    }
 888
 889    if (!error) {
 890        COVERAGE_ADD(netlink_sent, n);
 891    }
 892#endif
 893
 894    return error;
 895}
 896
 897static void
 898nl_sock_transact_multiple(struct nl_sock *sock,
 899                          struct nl_transaction **transactions, size_t n)
 900{
 901    int max_batch_count;
 902    int error;
 903
 904    if (!n) {
 905        return;
 906    }
 907
 908    /* In theory, every request could have a 64 kB reply.  But the default and
 909     * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
 910     * be a bit below 128 kB, so that would only allow a single message in a
 911     * "batch".  So we assume that replies average (at most) 4 kB, which allows
 912     * a good deal of batching.
 913     *
 914     * In practice, most of the requests that we batch either have no reply at
 915     * all or a brief reply. */
 916    max_batch_count = MAX(sock->rcvbuf / 4096, 1);
 917    max_batch_count = MIN(max_batch_count, max_iovs);
 918
 919    while (n > 0) {
 920        size_t count, bytes;
 921        size_t done;
 922
 923        /* Batch up to 'max_batch_count' transactions.  But cap it at about a
 924         * page of requests total because big skbuffs are expensive to
 925         * allocate in the kernel.  */
 926#if defined(PAGESIZE)
 927        enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
 928#else
 929        enum { MAX_BATCH_BYTES = 4096 - 512 };
 930#endif
 931        bytes = ofpbuf_size(transactions[0]->request);
 932        for (count = 1; count < n && count < max_batch_count; count++) {
 933            if (bytes + ofpbuf_size(transactions[count]->request) > MAX_BATCH_BYTES) {
 934                break;
 935            }
 936            bytes += ofpbuf_size(transactions[count]->request);
 937        }
 938
 939        error = nl_sock_transact_multiple__(sock, transactions, count, &done);
 940        transactions += done;
 941        n -= done;
 942
 943        if (error == ENOBUFS) {
 944            VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
 945        } else if (error) {
 946            VLOG_ERR_RL(&rl, "transaction error (%s)", ovs_strerror(error));
 947            nl_sock_record_errors__(transactions, n, error);
 948        }
 949    }
 950}
 951
 952static int
 953nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
 954                 struct ofpbuf **replyp)
 955{
 956    struct nl_transaction *transactionp;
 957    struct nl_transaction transaction;
 958
 959    transaction.request = CONST_CAST(struct ofpbuf *, request);
 960    transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
 961    transactionp = &transaction;
 962
 963    nl_sock_transact_multiple(sock, &transactionp, 1);
 964
 965    if (replyp) {
 966        if (transaction.error) {
 967            ofpbuf_delete(transaction.reply);
 968            *replyp = NULL;
 969        } else {
 970            *replyp = transaction.reply;
 971        }
 972    }
 973
 974    return transaction.error;
 975}
 976
 977/* Drain all the messages currently in 'sock''s receive queue. */
 978int
 979nl_sock_drain(struct nl_sock *sock)
 980{
 981#ifdef _WIN32
 982    return 0;
 983#else
 984    return drain_rcvbuf(sock->fd);
 985#endif
 986}
 987
 988/* Starts a Netlink "dump" operation, by sending 'request' to the kernel on a
 989 * Netlink socket created with the given 'protocol', and initializes 'dump' to
 990 * reflect the state of the operation.
 991 *
 992 * 'request' must contain a Netlink message.  Before sending the message,
 993 * nlmsg_len will be finalized to match request->size, and nlmsg_pid will be
 994 * set to the Netlink socket's pid.  NLM_F_DUMP and NLM_F_ACK will be set in
 995 * nlmsg_flags.
 996 *
 997 * The design of this Netlink socket library ensures that the dump is reliable.
 998 *
 999 * This function provides no status indication.  nl_dump_done() provides an
1000 * error status for the entire dump operation.
1001 *
1002 * The caller must eventually destroy 'request'.
1003 */
1004void
1005nl_dump_start(struct nl_dump *dump, int protocol, const struct ofpbuf *request)
1006{
1007    nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
1008
1009    ovs_mutex_init(&dump->mutex);
1010    ovs_mutex_lock(&dump->mutex);
1011    dump->status = nl_pool_alloc(protocol, &dump->sock);
1012    if (!dump->status) {
1013        dump->status = nl_sock_send__(dump->sock, request,
1014                                      nl_sock_allocate_seq(dump->sock, 1),
1015                                      true);
1016    }
1017    dump->nl_seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
1018    ovs_mutex_unlock(&dump->mutex);
1019}
1020
1021static int
1022nl_dump_refill(struct nl_dump *dump, struct ofpbuf *buffer)
1023    OVS_REQUIRES(dump->mutex)
1024{
1025    struct nlmsghdr *nlmsghdr;
1026    int error;
1027
1028    while (!ofpbuf_size(buffer)) {
1029        error = nl_sock_recv__(dump->sock, buffer, false);
1030        if (error) {
1031            /* The kernel never blocks providing the results of a dump, so
1032             * error == EAGAIN means that we've read the whole thing, and
1033             * therefore transform it into EOF.  (The kernel always provides
1034             * NLMSG_DONE as a sentinel.  Some other thread must have received
1035             * that already but not yet signaled it in 'status'.)
1036             *
1037             * Any other error is just an error. */
1038            return error == EAGAIN ? EOF : error;
1039        }
1040
1041        nlmsghdr = nl_msg_nlmsghdr(buffer);
1042        if (dump->nl_seq != nlmsghdr->nlmsg_seq) {
1043            VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
1044                        nlmsghdr->nlmsg_seq, dump->nl_seq);
1045            ofpbuf_clear(buffer);
1046        }
1047    }
1048
1049    if (nl_msg_nlmsgerr(buffer, &error) && error) {
1050        VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
1051                     ovs_strerror(error));
1052        ofpbuf_clear(buffer);
1053        return error;
1054    }
1055
1056    return 0;
1057}
1058
1059static int
1060nl_dump_next__(struct ofpbuf *reply, struct ofpbuf *buffer)
1061{
1062    struct nlmsghdr *nlmsghdr = nl_msg_next(buffer, reply);
1063    if (!nlmsghdr) {
1064        VLOG_WARN_RL(&rl, "netlink dump contains message fragment");
1065        return EPROTO;
1066    } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
1067        return EOF;
1068    } else {
1069        return 0;
1070    }
1071}
1072
1073/* Attempts to retrieve another reply from 'dump' into 'buffer'. 'dump' must
1074 * have been initialized with nl_dump_start(), and 'buffer' must have been
1075 * initialized. 'buffer' should be at least NL_DUMP_BUFSIZE bytes long.
1076 *
1077 * If successful, returns true and points 'reply->data' and
1078 * 'ofpbuf_size(reply)' to the message that was retrieved. The caller must not
1079 * modify 'reply' (because it points within 'buffer', which will be used by
1080 * future calls to this function).
1081 *
1082 * On failure, returns false and sets 'reply->data' to NULL and
1083 * 'ofpbuf_size(reply)' to 0.  Failure might indicate an actual error or merely
1084 * the end of replies.  An error status for the entire dump operation is
1085 * provided when it is completed by calling nl_dump_done().
1086 *
1087 * Multiple threads may call this function, passing the same nl_dump, however
1088 * each must provide independent buffers. This function may cache multiple
1089 * replies in the buffer, and these will be processed before more replies are
1090 * fetched. When this function returns false, other threads may continue to
1091 * process replies in their buffers, but they will not fetch more replies.
1092 */
1093bool
1094nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply, struct ofpbuf *buffer)
1095{
1096    int retval = 0;
1097
1098    /* If the buffer is empty, refill it.
1099     *
1100     * If the buffer is not empty, we don't check the dump's status.
1101     * Otherwise, we could end up skipping some of the dump results if thread A
1102     * hits EOF while thread B is in the midst of processing a batch. */
1103    if (!ofpbuf_size(buffer)) {
1104        ovs_mutex_lock(&dump->mutex);
1105        if (!dump->status) {
1106            /* Take the mutex here to avoid an in-kernel race.  If two threads
1107             * try to read from a Netlink dump socket at once, then the socket
1108             * error can be set to EINVAL, which will be encountered on the
1109             * next recv on that socket, which could be anywhere due to the way
1110             * that we pool Netlink sockets.  Serializing the recv calls avoids
1111             * the issue. */
1112            dump->status = nl_dump_refill(dump, buffer);
1113        }
1114        retval = dump->status;
1115        ovs_mutex_unlock(&dump->mutex);
1116    }
1117
1118    /* Fetch the next message from the buffer. */
1119    if (!retval) {
1120        retval = nl_dump_next__(reply, buffer);
1121        if (retval) {
1122            /* Record 'retval' as the dump status, but don't overwrite an error
1123             * with EOF.  */
1124            ovs_mutex_lock(&dump->mutex);
1125            if (dump->status <= 0) {
1126                dump->status = retval;
1127            }
1128            ovs_mutex_unlock(&dump->mutex);
1129        }
1130    }
1131
1132    if (retval) {
1133        ofpbuf_set_data(reply, NULL);
1134        ofpbuf_set_size(reply, 0);
1135    }
1136    return !retval;
1137}
1138
1139/* Completes Netlink dump operation 'dump', which must have been initialized
1140 * with nl_dump_start().  Returns 0 if the dump operation was error-free,
1141 * otherwise a positive errno value describing the problem. */
1142int
1143nl_dump_done(struct nl_dump *dump)
1144{
1145    int status;
1146
1147    ovs_mutex_lock(&dump->mutex);
1148    status = dump->status;
1149    ovs_mutex_unlock(&dump->mutex);
1150
1151    /* Drain any remaining messages that the client didn't read.  Otherwise the
1152     * kernel will continue to queue them up and waste buffer space.
1153     *
1154     * XXX We could just destroy and discard the socket in this case. */
1155    if (!status) {
1156        uint64_t tmp_reply_stub[NL_DUMP_BUFSIZE / 8];
1157        struct ofpbuf reply, buf;
1158
1159        ofpbuf_use_stub(&buf, tmp_reply_stub, sizeof tmp_reply_stub);
1160        while (nl_dump_next(dump, &reply, &buf)) {
1161            /* Nothing to do. */
1162        }
1163        ofpbuf_uninit(&buf);
1164
1165        ovs_mutex_lock(&dump->mutex);
1166        status = dump->status;
1167        ovs_mutex_unlock(&dump->mutex);
1168        ovs_assert(status);
1169    }
1170
1171    nl_pool_release(dump->sock);
1172    ovs_mutex_destroy(&dump->mutex);
1173
1174    return status == EOF ? 0 : status;
1175}
1176
1177#ifdef _WIN32
1178/* Pend an I/O request in the driver. The driver completes the I/O whenever
1179 * an event or a packet is ready to be read. Once the I/O is completed
1180 * the overlapped structure event associated with the pending I/O will be set
1181 */
1182static int
1183pend_io_request(struct nl_sock *sock)
1184{
1185    struct ofpbuf request;
1186    uint64_t request_stub[128];
1187    struct ovs_header *ovs_header;
1188    struct nlmsghdr *nlmsg;
1189    uint32_t seq;
1190    int retval;
1191    int error;
1192    DWORD bytes;
1193    OVERLAPPED *overlapped = CONST_CAST(OVERLAPPED *, &sock->overlapped);
1194
1195    int ovs_msg_size = sizeof (struct nlmsghdr) + sizeof (struct genlmsghdr) +
1196                               sizeof (struct ovs_header);
1197
1198    ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
1199
1200    seq = nl_sock_allocate_seq(sock, 1);
1201    nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
1202                          OVS_CTRL_CMD_WIN_PEND_REQ, OVS_WIN_CONTROL_VERSION);
1203    nlmsg = nl_msg_nlmsghdr(&request);
1204    nlmsg->nlmsg_seq = seq;
1205    nlmsg->nlmsg_pid = sock->pid;
1206
1207    ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
1208    ovs_header->dp_ifindex = 0;
1209
1210    if (!DeviceIoControl(sock->handle, OVS_IOCTL_WRITE,
1211                         ofpbuf_data(&request), ofpbuf_size(&request),
1212                         NULL, 0, &bytes, overlapped)) {
1213        error = GetLastError();
1214        /* Check if the I/O got pended */
1215        if (error != ERROR_IO_INCOMPLETE && error != ERROR_IO_PENDING) {
1216            VLOG_ERR("nl_sock_wait failed - %s\n", ovs_format_message(error));
1217            retval = EINVAL;
1218            goto done;
1219        }
1220    } else {
1221        /* The I/O was completed synchronously */
1222        poll_immediate_wake();
1223    }
1224    retval = 0;
1225
1226done:
1227    ofpbuf_uninit(&request);
1228    return retval;
1229}
1230#endif  /* _WIN32 */
1231
1232/* Causes poll_block() to wake up when any of the specified 'events' (which is
1233 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'.
1234 * On Windows, 'sock' is not treated as const, and may be modified. */
1235void
1236nl_sock_wait(const struct nl_sock *sock, short int events)
1237{
1238#ifdef _WIN32
1239    if (sock->overlapped.Internal != STATUS_PENDING) {
1240        pend_io_request(CONST_CAST(struct nl_sock *, sock));
1241       /* XXX: poll_wevent_wait(sock->overlapped.hEvent); */
1242    }
1243    poll_immediate_wake(); /* XXX: temporary. */
1244#else
1245    poll_fd_wait(sock->fd, events);
1246#endif
1247}
1248
1249/* Returns the underlying fd for 'sock', for use in "poll()"-like operations
1250 * that can't use nl_sock_wait().
1251 *
1252 * It's a little tricky to use the returned fd correctly, because nl_sock does
1253 * "copy on write" to allow a single nl_sock to be used for notifications,
1254 * transactions, and dumps.  If 'sock' is used only for notifications and
1255 * transactions (and never for dump) then the usage is safe. */
1256int
1257nl_sock_fd(const struct nl_sock *sock)
1258{
1259#ifdef _WIN32
1260    BUILD_ASSERT_DECL(sizeof sock->handle == sizeof(int));
1261    return (int)sock->handle;
1262#else
1263    return sock->fd;
1264#endif
1265}
1266
1267/* Returns the PID associated with this socket. */
1268uint32_t
1269nl_sock_pid(const struct nl_sock *sock)
1270{
1271    return sock->pid;
1272}
1273
1274/* Miscellaneous.  */
1275
1276struct genl_family {
1277    struct hmap_node hmap_node;
1278    uint16_t id;
1279    char *name;
1280};
1281
1282static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
1283
1284static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
1285    [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
1286    [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
1287};
1288
1289static struct genl_family *
1290find_genl_family_by_id(uint16_t id)
1291{
1292    struct genl_family *family;
1293
1294    HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
1295                             &genl_families) {
1296        if (family->id == id) {
1297            return family;
1298        }
1299    }
1300    return NULL;
1301}
1302
1303static void
1304define_genl_family(uint16_t id, const char *name)
1305{
1306    struct genl_family *family = find_genl_family_by_id(id);
1307
1308    if (family) {
1309        if (!strcmp(family->name, name)) {
1310            return;
1311        }
1312        free(family->name);
1313    } else {
1314        family = xmalloc(sizeof *family);
1315        family->id = id;
1316        hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
1317    }
1318    family->name = xstrdup(name);
1319}
1320
1321static const char *
1322genl_family_to_name(uint16_t id)
1323{
1324    if (id == GENL_ID_CTRL) {
1325        return "control";
1326    } else {
1327        struct genl_family *family = find_genl_family_by_id(id);
1328        return family ? family->name : "unknown";
1329    }
1330}
1331
1332#ifndef _WIN32
1333static int
1334do_lookup_genl_family(const char *name, struct nlattr **attrs,
1335                      struct ofpbuf **replyp)
1336{
1337    struct nl_sock *sock;
1338    struct ofpbuf request, *reply;
1339    int error;
1340
1341    *replyp = NULL;
1342    error = nl_sock_create(NETLINK_GENERIC, &sock);
1343    if (error) {
1344        return error;
1345    }
1346
1347    ofpbuf_init(&request, 0);
1348    nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
1349                          CTRL_CMD_GETFAMILY, 1);
1350    nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
1351    error = nl_sock_transact(sock, &request, &reply);
1352    ofpbuf_uninit(&request);
1353    if (error) {
1354        nl_sock_destroy(sock);
1355        return error;
1356    }
1357
1358    if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
1359                         family_policy, attrs, ARRAY_SIZE(family_policy))
1360        || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
1361        nl_sock_destroy(sock);
1362        ofpbuf_delete(reply);
1363        return EPROTO;
1364    }
1365
1366    nl_sock_destroy(sock);
1367    *replyp = reply;
1368    return 0;
1369}
1370#else
1371static int
1372do_lookup_genl_family(const char *name, struct nlattr **attrs,
1373                      struct ofpbuf **replyp)
1374{
1375    struct nlmsghdr *nlmsg;
1376    struct ofpbuf *reply;
1377    int error;
1378    uint16_t family_id;
1379    const char *family_name;
1380    uint32_t family_version;
1381    uint32_t family_attrmax;
1382    uint32_t mcgrp_id = OVS_WIN_NL_INVALID_MCGRP_ID;
1383    const char *mcgrp_name = NULL;
1384
1385    *replyp = NULL;
1386    reply = ofpbuf_new(1024);
1387
1388    /* CTRL_ATTR_MCAST_GROUPS is supported only for VPORT family. */
1389    if (!strcmp(name, OVS_WIN_CONTROL_FAMILY)) {
1390        family_id = OVS_WIN_NL_CTRL_FAMILY_ID;
1391        family_name = OVS_WIN_CONTROL_FAMILY;
1392        family_version = OVS_WIN_CONTROL_VERSION;
1393        family_attrmax = OVS_WIN_CONTROL_ATTR_MAX;
1394    } else if (!strcmp(name, OVS_DATAPATH_FAMILY)) {
1395        family_id = OVS_WIN_NL_DATAPATH_FAMILY_ID;
1396        family_name = OVS_DATAPATH_FAMILY;
1397        family_version = OVS_DATAPATH_VERSION;
1398        family_attrmax = OVS_DP_ATTR_MAX;
1399    } else if (!strcmp(name, OVS_PACKET_FAMILY)) {
1400        family_id = OVS_WIN_NL_PACKET_FAMILY_ID;
1401        family_name = OVS_PACKET_FAMILY;
1402        family_version = OVS_PACKET_VERSION;
1403        family_attrmax = OVS_PACKET_ATTR_MAX;
1404    } else if (!strcmp(name, OVS_VPORT_FAMILY)) {
1405        family_id = OVS_WIN_NL_VPORT_FAMILY_ID;
1406        family_name = OVS_VPORT_FAMILY;
1407        family_version = OVS_VPORT_VERSION;
1408        family_attrmax = OVS_VPORT_ATTR_MAX;
1409        mcgrp_id = OVS_WIN_NL_VPORT_MCGRP_ID;
1410        mcgrp_name = OVS_VPORT_MCGROUP;
1411    } else if (!strcmp(name, OVS_FLOW_FAMILY)) {
1412        family_id = OVS_WIN_NL_FLOW_FAMILY_ID;
1413        family_name = OVS_FLOW_FAMILY;
1414        family_version = OVS_FLOW_VERSION;
1415        family_attrmax = OVS_FLOW_ATTR_MAX;
1416    } else if (!strcmp(name, OVS_WIN_NETDEV_FAMILY)) {
1417        family_id = OVS_WIN_NL_NETDEV_FAMILY_ID;
1418        family_name = OVS_WIN_NETDEV_FAMILY;
1419        family_version = OVS_WIN_NETDEV_VERSION;
1420        family_attrmax = OVS_WIN_NETDEV_ATTR_MAX;
1421    } else {
1422        ofpbuf_delete(reply);
1423        return EINVAL;
1424    }
1425
1426    nl_msg_put_genlmsghdr(reply, 0, GENL_ID_CTRL, 0,
1427                          CTRL_CMD_NEWFAMILY, family_version);
1428    /* CTRL_ATTR_HDRSIZE and CTRL_ATTR_OPS are not populated, but the
1429     * callers do not seem to need them. */
1430    nl_msg_put_u16(reply, CTRL_ATTR_FAMILY_ID, family_id);
1431    nl_msg_put_string(reply, CTRL_ATTR_FAMILY_NAME, family_name);
1432    nl_msg_put_u32(reply, CTRL_ATTR_VERSION, family_version);
1433    nl_msg_put_u32(reply, CTRL_ATTR_MAXATTR, family_attrmax);
1434
1435    if (mcgrp_id != OVS_WIN_NL_INVALID_MCGRP_ID) {
1436        size_t mcgrp_ofs1 = nl_msg_start_nested(reply, CTRL_ATTR_MCAST_GROUPS);
1437        size_t mcgrp_ofs2= nl_msg_start_nested(reply,
1438            OVS_WIN_NL_VPORT_MCGRP_ID - OVS_WIN_NL_MCGRP_START_ID);
1439        nl_msg_put_u32(reply, CTRL_ATTR_MCAST_GRP_ID, mcgrp_id);
1440        ovs_assert(mcgrp_name != NULL);
1441        nl_msg_put_string(reply, CTRL_ATTR_MCAST_GRP_NAME, mcgrp_name);
1442        nl_msg_end_nested(reply, mcgrp_ofs2);
1443        nl_msg_end_nested(reply, mcgrp_ofs1);
1444    }
1445
1446    /* Set the total length of the netlink message. */
1447    nlmsg = nl_msg_nlmsghdr(reply);
1448    nlmsg->nlmsg_len = ofpbuf_size(reply);
1449
1450    if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
1451                         family_policy, attrs, ARRAY_SIZE(family_policy))
1452        || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
1453        ofpbuf_delete(reply);
1454        return EPROTO;
1455    }
1456
1457    *replyp = reply;
1458    return 0;
1459}
1460#endif
1461
1462/* Finds the multicast group called 'group_name' in genl family 'family_name'.
1463 * When successful, writes its result to 'multicast_group' and returns 0.
1464 * Otherwise, clears 'multicast_group' and returns a positive error code.
1465 */
1466int
1467nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
1468                       unsigned int *multicast_group)
1469{
1470    struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
1471    const struct nlattr *mc;
1472    struct ofpbuf *reply;
1473    unsigned int left;
1474    int error;
1475
1476    *multicast_group = 0;
1477    error = do_lookup_genl_family(family_name, family_attrs, &reply);
1478    if (error) {
1479        return error;
1480    }
1481
1482    if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1483        error = EPROTO;
1484        goto exit;
1485    }
1486
1487    NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1488        static const struct nl_policy mc_policy[] = {
1489            [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
1490            [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
1491        };
1492
1493        struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
1494        const char *mc_name;
1495
1496        if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
1497            error = EPROTO;
1498            goto exit;
1499        }
1500
1501        mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
1502        if (!strcmp(group_name, mc_name)) {
1503            *multicast_group =
1504                nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
1505            error = 0;
1506            goto exit;
1507        }
1508    }
1509    error = EPROTO;
1510
1511exit:
1512    ofpbuf_delete(reply);
1513    return error;
1514}
1515
1516/* If '*number' is 0, translates the given Generic Netlink family 'name' to a
1517 * number and stores it in '*number'.  If successful, returns 0 and the caller
1518 * may use '*number' as the family number.  On failure, returns a positive
1519 * errno value and '*number' caches the errno value. */
1520int
1521nl_lookup_genl_family(const char *name, int *number)
1522{
1523    if (*number == 0) {
1524        struct nlattr *attrs[ARRAY_SIZE(family_policy)];
1525        struct ofpbuf *reply;
1526        int error;
1527
1528        error = do_lookup_genl_family(name, attrs, &reply);
1529        if (!error) {
1530            *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
1531            define_genl_family(*number, name);
1532        } else {
1533            *number = -error;
1534        }
1535        ofpbuf_delete(reply);
1536
1537        ovs_assert(*number != 0);
1538    }
1539    return *number > 0 ? 0 : -*number;
1540}
1541
1542struct nl_pool {
1543    struct nl_sock *socks[16];
1544    int n;
1545};
1546
1547static struct ovs_mutex pool_mutex = OVS_MUTEX_INITIALIZER;
1548static struct nl_pool pools[MAX_LINKS] OVS_GUARDED_BY(pool_mutex);
1549
1550static int
1551nl_pool_alloc(int protocol, struct nl_sock **sockp)
1552{
1553    struct nl_sock *sock = NULL;
1554    struct nl_pool *pool;
1555
1556    ovs_assert(protocol >= 0 && protocol < ARRAY_SIZE(pools));
1557
1558    ovs_mutex_lock(&pool_mutex);
1559    pool = &pools[protocol];
1560    if (pool->n > 0) {
1561        sock = pool->socks[--pool->n];
1562    }
1563    ovs_mutex_unlock(&pool_mutex);
1564
1565    if (sock) {
1566        *sockp = sock;
1567        return 0;
1568    } else {
1569        return nl_sock_create(protocol, sockp);
1570    }
1571}
1572
1573static void
1574nl_pool_release(struct nl_sock *sock)
1575{
1576    if (sock) {
1577        struct nl_pool *pool = &pools[sock->protocol];
1578
1579        ovs_mutex_lock(&pool_mutex);
1580        if (pool->n < ARRAY_SIZE(pool->socks)) {
1581            pool->socks[pool->n++] = sock;
1582            sock = NULL;
1583        }
1584        ovs_mu

Large files files are truncated, but you can click here to view the full file