virnetlink.c 27.4 KB
Newer Older
1
/*
L
Leno Hou 已提交
2 3
 * Copyright (C) 2010-2016 Red Hat, Inc.
 * Copyright (C) 2010-2012, 2016 IBM Corporation
4 5 6 7 8 9 10 11 12 13 14 15
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
16
 * License along with this library.  If not, see
O
Osier Yang 已提交
17
 * <http://www.gnu.org/licenses/>.
18 19 20
 *
 * Authors:
 *     Stefan Berger <stefanb@us.ibm.com>
21
 *     Dirk Herrendoerfer <herrend[at]de[dot]ibm[dot]com>
22 23 24 25 26
 *
 * Notes:
 * netlink: http://lovezutto.googlepages.com/netlink.pdf
 *          iproute2 package
 *
27 28
 * 2012/02: Renamed from netlink.[ch] to virnetlink.[ch]
 *
29 30 31 32 33 34 35
 */

#include <config.h>

#include <errno.h>
#include <unistd.h>
#include <sys/types.h>
36
#include <sys/socket.h>
37

38
#include "virnetlink.h"
39
#include "virlog.h"
40
#include "viralloc.h"
41
#include "virthread.h"
42
#include "virmacaddr.h"
43
#include "virerror.h"
44 45 46

#define VIR_FROM_THIS VIR_FROM_NET

47 48
VIR_LOG_INIT("util.netlink");

49
#define NETLINK_ACK_TIMEOUT_S  (2*1000)
50

51 52 53 54 55 56 57
#if defined(__linux__) && defined(HAVE_LIBNL)
/* State for a single netlink event handle */
struct virNetlinkEventHandle {
    int watch;
    virNetlinkEventHandleCallback handleCB;
    virNetlinkEventRemoveCallback removeCB;
    void *opaque;
58
    virMacAddr macaddr;
59 60 61
    int deleted;
};

S
Serge Hallyn 已提交
62 63
# ifdef HAVE_LIBNL1
#  define virNetlinkAlloc nl_handle_alloc
L
Leno Hou 已提交
64
#  define virNetlinkSetBufferSize nl_set_buffer_size
S
Serge Hallyn 已提交
65 66 67 68
#  define virNetlinkFree nl_handle_destroy
typedef struct nl_handle virNetlinkHandle;
# else
#  define virNetlinkAlloc nl_socket_alloc
L
Leno Hou 已提交
69
#  define virNetlinkSetBufferSize nl_socket_set_buffer_size
S
Serge Hallyn 已提交
70 71 72 73
#  define virNetlinkFree nl_socket_free
typedef struct nl_sock virNetlinkHandle;
# endif

74 75 76 77 78 79 80
typedef struct _virNetlinkEventSrvPrivate virNetlinkEventSrvPrivate;
typedef virNetlinkEventSrvPrivate *virNetlinkEventSrvPrivatePtr;
struct _virNetlinkEventSrvPrivate {
    /*Server*/
    virMutex lock;
    int eventwatch;
    int netlinkfd;
S
Serge Hallyn 已提交
81
    virNetlinkHandle *netlinknh;
82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
    /*Events*/
    int handled;
    size_t handlesCount;
    size_t handlesAlloc;
    struct virNetlinkEventHandle *handles;
};

enum virNetlinkDeleteMode {
    VIR_NETLINK_HANDLE_VALID,
    VIR_NETLINK_HANDLE_DELETED,
};

/* Unique ID for the next netlink watch to be registered */
static int nextWatch = 1;

/* Allocate extra slots for virEventPollHandle/virEventPollTimeout
 records in this multiple */
# define NETLINK_EVENT_ALLOC_EXTENT 10

101 102 103
/* Linux kernel supports up to MAX_LINKS (32 at the time) individual
 * netlink protocols. */
static virNetlinkEventSrvPrivatePtr server[MAX_LINKS] = {NULL};
104
static virNetlinkHandle *placeholder_nlhandle;
105 106 107

/* Function definitions */

108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
/**
 * virNetlinkStartup:
 *
 * Perform any initialization that needs to take place before the
 * program starts up worker threads. This is currently used to assure
 * that an nl_handle is allocated prior to any attempts to bind a
 * netlink socket. For a discussion of why this is necessary, please
 * see the following email message:
 *
 *   https://www.redhat.com/archives/libvir-list/2012-May/msg00202.html
 *
 * The short version is that, without this placeholder allocation of
 * an nl_handle that is never used, it is possible for nl_connect() in
 * one thread to collide with a direct bind() of a netlink socket in
 * another thread, leading to failure of the operation (which could
 * lead to failure of libvirtd to start). Since getaddrinfo() (used by
 * libvirtd in virSocketAddrParse, which is called quite frequently
 * during startup) directly calls bind() on a netlink socket, this is
 * actually a very common occurrence (15-20% failure rate on some
 * hardware).
 *
 * Returns 0 on success, -1 on failure.
 */
int
virNetlinkStartup(void)
{
    if (placeholder_nlhandle)
        return 0;
136
    VIR_DEBUG("Running global netlink initialization");
J
Jiri Denemark 已提交
137
    placeholder_nlhandle = virNetlinkAlloc();
138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155
    if (!placeholder_nlhandle) {
        virReportSystemError(errno, "%s",
                             _("cannot allocate placeholder nlhandle for netlink"));
        return -1;
    }
    return 0;
}

/**
 * virNetlinkShutdown:
 *
 * Undo any initialization done by virNetlinkStartup. This currently
 * destroys the placeholder nl_handle.
 */
void
virNetlinkShutdown(void)
{
    if (placeholder_nlhandle) {
J
Jiri Denemark 已提交
156
        virNetlinkFree(placeholder_nlhandle);
157 158 159 160
        placeholder_nlhandle = NULL;
    }
}

L
Leno Hou 已提交
161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211

/**
 * virNetLinkCreateSocket:
 *
 * @protocol: which protocol to connect to (e.g. NETLINK_ROUTE,
 *
 * Create a netlink socket, set its buffer size, and turn on message
 * peeking (so the buffer size can be dynamically increased if
 * needed).
 *
 * Returns a handle to the new netlink socket, or 0 if there was a failure.
 *
 */
static virNetlinkHandle *
virNetlinkCreateSocket(int protocol)
{
    virNetlinkHandle *nlhandle = NULL;

    if (!(nlhandle = virNetlinkAlloc())) {
        virReportSystemError(errno, "%s",
                             _("cannot allocate nlhandle for netlink"));
        goto error;
    }
    if (nl_connect(nlhandle, protocol) < 0) {
        virReportSystemError(errno,
                             _("cannot connect to netlink socket "
                               "with protocol %d"), protocol);
        goto error;
    }

    if (virNetlinkSetBufferSize(nlhandle, 131702, 0) < 0) {
        virReportSystemError(errno, "%s",
                             _("cannot set netlink socket buffer "
                               "size to 128k"));
        goto error;
    }
    nl_socket_enable_msg_peek(nlhandle);

 cleanup:
    return nlhandle;

 error:
    if (nlhandle) {
        nl_close(nlhandle);
        virNetlinkFree(nlhandle);
        nlhandle = NULL;
    }
    goto cleanup;
}


212
/**
213
 * virNetlinkCommand:
214 215 216 217
 * @nlmsg: pointer to netlink message
 * @respbuf: pointer to pointer where response buffer will be allocated
 * @respbuflen: pointer to integer holding the size of the response buffer
 *      on return of the function.
218 219 220 221
 * @src_pid: the pid of the process to send a message
 * @dst_pid: the pid of the process to talk to, i.e., pid = 0 for kernel
 * @protocol: netlink protocol
 * @groups: the group identifier
222 223 224 225 226
 *
 * Send the given message to the netlink layer and receive response.
 * Returns 0 on success, -1 on error. In case of error, no response
 * buffer will be returned.
 */
227
int virNetlinkCommand(struct nl_msg *nl_msg,
228
                      struct nlmsghdr **resp, unsigned int *respbuflen,
229 230
                      uint32_t src_pid, uint32_t dst_pid,
                      unsigned int protocol, unsigned int groups)
231
{
232
    int ret = -1;
233 234
    struct sockaddr_nl nladdr = {
            .nl_family = AF_NETLINK,
235
            .nl_pid    = dst_pid,
236 237 238
            .nl_groups = 0,
    };
    ssize_t nbytes;
239
    struct pollfd fds[1];
240 241 242
    int fd;
    int n;
    struct nlmsghdr *nlmsg = nlmsg_hdr(nl_msg);
243
    virNetlinkHandle *nlhandle = NULL;
244
    int len = 0;
245 246 247 248

    if (protocol >= MAX_LINKS) {
        virReportSystemError(EINVAL,
                             _("invalid protocol argument: %d"), protocol);
249
        goto cleanup;
250
    }
251

L
Leno Hou 已提交
252
    if (!(nlhandle = virNetlinkCreateSocket(protocol)))
253
        goto cleanup;
254 255 256 257 258

    fd = nl_socket_get_fd(nlhandle);
    if (fd < 0) {
        virReportSystemError(errno,
                             "%s", _("cannot get netlink socket fd"));
259
        goto cleanup;
260 261 262
    }

    if (groups && nl_socket_add_membership(nlhandle, groups) < 0) {
263
        virReportSystemError(errno,
264
                             "%s", _("cannot add netlink membership"));
265
        goto cleanup;
266 267 268 269
    }

    nlmsg_set_dst(nl_msg, &nladdr);

270
    nlmsg->nlmsg_pid = src_pid ? src_pid : getpid();
271 272 273 274 275

    nbytes = nl_send_auto_complete(nlhandle, nl_msg);
    if (nbytes < 0) {
        virReportSystemError(errno,
                             "%s", _("cannot send to netlink socket"));
276
        goto cleanup;
277 278
    }

279 280 281
    memset(fds, 0, sizeof(fds));
    fds[0].fd = fd;
    fds[0].events = POLLIN;
282

283
    n = poll(fds, ARRAY_CARDINALITY(fds), NETLINK_ACK_TIMEOUT_S);
284 285 286
    if (n <= 0) {
        if (n < 0)
            virReportSystemError(errno, "%s",
287
                                 _("error in poll call"));
288 289 290
        if (n == 0)
            virReportSystemError(ETIMEDOUT, "%s",
                                 _("no valid netlink response was received"));
291
        goto cleanup;
292 293
    }

294 295 296 297 298
    len = nl_recv(nlhandle, &nladdr, (unsigned char **)resp, NULL);
    if (len == 0) {
        virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
                       _("nl_recv failed - returned 0 bytes"));
        goto cleanup;
299
    }
300 301 302 303 304 305 306 307 308
    if (len < 0) {
        virReportSystemError(errno, "%s", _("nl_recv failed"));
        goto cleanup;
    }

    ret = 0;
    *respbuflen = len;
 cleanup:
    if (ret < 0) {
309
        *resp = NULL;
310 311 312
        *respbuflen = 0;
    }

S
Serge Hallyn 已提交
313
    virNetlinkFree(nlhandle);
314
    return ret;
315 316
}

E
Eric Blake 已提交
317

318 319 320 321
/**
 * virNetlinkDelLink:
 *
 * @ifname: Name of the link
322 323 324 325
 * @fallback: pointer to an alternate function that will
 *            be called to perform the delete if RTM_DELLINK fails
 *            with EOPNOTSUPP (any other error will simply be treated
 *            as an error).
326 327 328 329 330 331 332 333
 *
 * delete a network "link" (aka interface aka device) with the given
 * name. This works for many different types of network devices,
 * including macvtap and bridges.
 *
 * Returns 0 on success, -1 on fatal error.
 */
int
334
virNetlinkDelLink(const char *ifname, virNetlinkDelLinkFallback fallback)
335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369
{
    int rc = -1;
    struct nlmsghdr *resp = NULL;
    struct nlmsgerr *err;
    struct ifinfomsg ifinfo = { .ifi_family = AF_UNSPEC };
    unsigned int recvbuflen;
    struct nl_msg *nl_msg;

    nl_msg = nlmsg_alloc_simple(RTM_DELLINK,
                                NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL);
    if (!nl_msg) {
        virReportOOMError();
        return -1;
    }

    if (nlmsg_append(nl_msg,  &ifinfo, sizeof(ifinfo), NLMSG_ALIGNTO) < 0)
        goto buffer_too_small;

    if (nla_put(nl_msg, IFLA_IFNAME, strlen(ifname)+1, ifname) < 0)
        goto buffer_too_small;

    if (virNetlinkCommand(nl_msg, &resp, &recvbuflen, 0, 0,
                          NETLINK_ROUTE, 0) < 0) {
        goto cleanup;
    }

    if (recvbuflen < NLMSG_LENGTH(0) || resp == NULL)
        goto malformed_resp;

    switch (resp->nlmsg_type) {
    case NLMSG_ERROR:
        err = (struct nlmsgerr *)NLMSG_DATA(resp);
        if (resp->nlmsg_len < NLMSG_LENGTH(sizeof(*err)))
            goto malformed_resp;

370 371 372 373
        if (-err->error == EOPNOTSUPP && fallback) {
            rc = fallback(ifname);
            goto cleanup;
        }
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406
        if (err->error) {
            virReportSystemError(-err->error,
                                 _("error destroying network device %s"),
                                 ifname);
            goto cleanup;
        }
        break;

    case NLMSG_DONE:
        break;

    default:
        goto malformed_resp;
    }

    rc = 0;
 cleanup:
    nlmsg_free(nl_msg);
    VIR_FREE(resp);
    return rc;

 malformed_resp:
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
                   _("malformed netlink response message"));
    goto cleanup;

 buffer_too_small:
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
                   _("allocated netlink buffer is too small"));
    goto cleanup;
}


E
Eric Blake 已提交
407 408
int
virNetlinkGetErrorCode(struct nlmsghdr *resp, unsigned int recvbuflen)
409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434
{
    struct nlmsgerr *err;
    int result = 0;

    if (recvbuflen < NLMSG_LENGTH(0) || resp == NULL)
        goto malformed_resp;

    switch (resp->nlmsg_type) {
    case NLMSG_ERROR:
        err = (struct nlmsgerr *)NLMSG_DATA(resp);
        if (resp->nlmsg_len < NLMSG_LENGTH(sizeof(*err)))
            goto malformed_resp;

        switch (err->error) {
        case 0: /* ACK */
            break;

        default:
            result = err->error;
        }
        break;

    case NLMSG_DONE:
        break;

    default:
435 436 437
        /* We allow multipart messages. */
        if (!(resp->nlmsg_flags & NLM_F_MULTI))
            goto malformed_resp;
438 439 440 441 442 443 444 445 446 447
    }

    return result;

 malformed_resp:
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
                   _("malformed netlink response message"));
    return -EINVAL;
}

E
Eric Blake 已提交
448

449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464
static void
virNetlinkEventServerLock(virNetlinkEventSrvPrivatePtr driver)
{
    virMutexLock(&driver->lock);
}

static void
virNetlinkEventServerUnlock(virNetlinkEventSrvPrivatePtr driver)
{
    virMutexUnlock(&driver->lock);
}

/**
 * virNetlinkEventRemoveClientPrimitive:
 *
 * @i: index of the client to remove from the table
465
 * @protocol: netlink protocol
466 467 468 469 470 471 472 473 474
 *
 * This static function does the low level removal of a client from
 * the table once its index is known, including calling the remove
 * callback (which usually will free resources required by the
 * handler). The event server lock *must* be locked before calling
 * this function.
 *
 * assumes success, returns nothing.
 */
475 476
static int
virNetlinkEventRemoveClientPrimitive(size_t i, unsigned int protocol)
477
{
478 479 480 481
    if (protocol >= MAX_LINKS)
        return -EINVAL;

    virNetlinkEventRemoveCallback removeCB = server[protocol]->handles[i].removeCB;
482 483

    if (removeCB) {
484 485 486
        (removeCB)(server[protocol]->handles[i].watch,
                   &server[protocol]->handles[i].macaddr,
                   server[protocol]->handles[i].opaque);
487
    }
488 489
    server[protocol]->handles[i].deleted = VIR_NETLINK_HANDLE_DELETED;
    return 0;
490 491 492 493 494 495 496 497 498
}

static void
virNetlinkEventCallback(int watch,
                        int fd ATTRIBUTE_UNUSED,
                        int events ATTRIBUTE_UNUSED,
                        void *opaque)
{
    virNetlinkEventSrvPrivatePtr srv = opaque;
499
    struct nlmsghdr *msg;
500 501
    struct sockaddr_nl peer;
    struct ucred *creds = NULL;
502 503
    size_t i;
    int length;
504 505
    bool handled = false;

506 507
    length = nl_recv(srv->netlinknh, &peer,
                     (unsigned char **)&msg, &creds);
508 509 510 511

    if (length == 0)
        return;
    if (length < 0) {
512 513
        virReportSystemError(errno,
                             "%s", _("nl_recv returned with error"));
514 515 516 517 518 519 520 521 522 523 524 525
        return;
    }

    virNetlinkEventServerLock(srv);

    VIR_DEBUG("dispatching to max %d clients, called from event watch %d",
            (int)srv->handlesCount, watch);

    for (i = 0; i < srv->handlesCount; i++) {
        if (srv->handles[i].deleted != VIR_NETLINK_HANDLE_VALID)
            continue;

526
        VIR_DEBUG("dispatching client %zu.", i);
527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543

        (srv->handles[i].handleCB)(msg, length, &peer, &handled,
                                   srv->handles[i].opaque);
    }

    if (!handled)
        VIR_DEBUG("event not handled.");
    VIR_FREE(msg);
    virNetlinkEventServerUnlock(srv);
}

/**
 * virNetlinkEventServiceStop:
 *
 * stop the monitor to receive netlink messages for libvirtd.
 * This removes the netlink socket fd from the event handler.
 *
544 545
 * @protocol: netlink protocol
 *
546
 * Returns -1 if the monitor cannot be unregistered, 0 upon success
547 548
 */
int
549
virNetlinkEventServiceStop(unsigned int protocol)
550
{
551 552 553 554
    if (protocol >= MAX_LINKS)
        return -EINVAL;

    virNetlinkEventSrvPrivatePtr srv = server[protocol];
555
    size_t i;
556 557 558

    VIR_INFO("stopping netlink event service");

559
    if (!server[protocol])
560 561 562 563
        return 0;

    virNetlinkEventServerLock(srv);
    nl_close(srv->netlinknh);
S
Serge Hallyn 已提交
564
    virNetlinkFree(srv->netlinknh);
565 566 567 568 569
    virEventRemoveHandle(srv->eventwatch);

    /* free any remaining clients on the list */
    for (i = 0; i < srv->handlesCount; i++) {
        if (srv->handles[i].deleted == VIR_NETLINK_HANDLE_VALID)
570
            virNetlinkEventRemoveClientPrimitive(i, protocol);
571 572
    }

573
    server[protocol] = NULL;
574 575 576 577 578 579 580
    virNetlinkEventServerUnlock(srv);

    virMutexDestroy(&srv->lock);
    VIR_FREE(srv);
    return 0;
}

581 582 583 584 585 586 587 588 589 590
/**
 * virNetlinkEventServiceStopAll:
 *
 * Stop all the monitors to receive netlink messages for libvirtd.
 *
 * Returns -1 if any monitor cannot be unregistered, 0 upon success
 */
int
virNetlinkEventServiceStopAll(void)
{
591
    size_t i, j;
592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620
    virNetlinkEventSrvPrivatePtr srv = NULL;

    VIR_INFO("stopping all netlink event services");

    for (i = 0; i < MAX_LINKS; i++) {
        srv = server[i];
        if (!srv)
            continue;

        virNetlinkEventServerLock(srv);
        nl_close(srv->netlinknh);
        virNetlinkFree(srv->netlinknh);
        virEventRemoveHandle(srv->eventwatch);

        for (j = 0; j < srv->handlesCount; j++) {
            if (srv->handles[j].deleted == VIR_NETLINK_HANDLE_VALID)
                virNetlinkEventRemoveClientPrimitive(j, i);
        }

        server[i] = NULL;
        virNetlinkEventServerUnlock(srv);

        virMutexDestroy(&srv->lock);
        VIR_FREE(srv);
    }

    return 0;
}

621 622 623
/**
 * virNetlinkEventServiceIsRunning:
 *
624
 * Returns if the netlink event service is running.
625
 *
626 627
 * @protocol: netlink protocol
 *
628
 * Returns 'true' if the service is running, 'false' if stopped.
629 630
 */
bool
631
virNetlinkEventServiceIsRunning(unsigned int protocol)
632
{
633 634 635 636 637 638 639
    if (protocol >= MAX_LINKS) {
        virReportSystemError(EINVAL,
                             _("invalid protocol argument: %d"), protocol);
        return false;
    }

    return server[protocol] != NULL;
640 641
}

642 643 644
/**
 * virNetlinkEventServiceLocalPid:
 *
645 646
 * @protocol: netlink protocol
 *
647 648 649 650
 * Returns the nl_pid value that was used to bind() the netlink socket
 * used by the netlink event service, or -1 on error (netlink
 * guarantees that this value will always be > 0).
 */
651
int virNetlinkEventServiceLocalPid(unsigned int protocol)
652
{
653 654 655 656
    if (protocol >= MAX_LINKS)
        return -EINVAL;

    if (!(server[protocol] && server[protocol]->netlinknh)) {
657 658
        virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
                       _("netlink event service not running"));
659 660
        return -1;
    }
661
    return (int)nl_socket_get_local_port(server[protocol]->netlinknh);
662 663 664
}


665 666 667 668 669 670
/**
 * virNetlinkEventServiceStart:
 *
 * start a monitor to receive netlink messages for libvirtd.
 * This registers a netlink socket with the event interface.
 *
671 672
 * @protocol: netlink protocol
 * @groups: broadcast groups to join in
673
 * Returns -1 if the monitor cannot be registered, 0 upon success
674 675
 */
int
676
virNetlinkEventServiceStart(unsigned int protocol, unsigned int groups)
677 678 679 680 681
{
    virNetlinkEventSrvPrivatePtr srv;
    int fd;
    int ret = -1;

682 683 684 685 686 687 688
    if (protocol >= MAX_LINKS) {
        virReportSystemError(EINVAL,
                             _("invalid protocol argument: %d"), protocol);
        return -EINVAL;
    }

    if (server[protocol])
689 690
        return 0;

691
    VIR_INFO("starting netlink event service with protocol %d", protocol);
692

693
    if (VIR_ALLOC(srv) < 0)
694
        return -1;
695

696 697 698 699
    if (virMutexInit(&srv->lock) < 0) {
        VIR_FREE(srv);
        return -1;
    }
700 701 702 703

    virNetlinkEventServerLock(srv);

    /* Allocate a new socket and get fd */
L
Leno Hou 已提交
704
    if (!(srv->netlinknh = virNetlinkCreateSocket(protocol)))
705 706 707 708
        goto error_locked;

    fd = nl_socket_get_fd(srv->netlinknh);
    if (fd < 0) {
709 710
        virReportSystemError(errno,
                             "%s", _("cannot get netlink socket fd"));
711 712 713
        goto error_server;
    }

714 715 716 717 718 719
    if (groups && nl_socket_add_membership(srv->netlinknh, groups) < 0) {
        virReportSystemError(errno,
                             "%s", _("cannot add netlink membership"));
        goto error_server;
    }

720
    if (nl_socket_set_nonblocking(srv->netlinknh)) {
721 722
        virReportSystemError(errno, "%s",
                             _("cannot set netlink socket nonblocking"));
723 724 725 726 727 728 729
        goto error_server;
    }

    if ((srv->eventwatch = virEventAddHandle(fd,
                                             VIR_EVENT_HANDLE_READABLE,
                                             virNetlinkEventCallback,
                                             srv, NULL)) < 0) {
730 731
        virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
                       _("Failed to add netlink event handle watch"));
732 733 734 735 736 737 738
        goto error_server;
    }

    srv->netlinkfd = fd;
    VIR_DEBUG("netlink event listener on fd: %i running", fd);

    ret = 0;
739
    server[protocol] = srv;
740

741
 error_server:
742 743
    if (ret < 0) {
        nl_close(srv->netlinknh);
S
Serge Hallyn 已提交
744
        virNetlinkFree(srv->netlinknh);
745
    }
746
 error_locked:
747 748 749 750 751 752 753 754 755 756 757 758 759 760
    virNetlinkEventServerUnlock(srv);
    if (ret < 0) {
        virMutexDestroy(&srv->lock);
        VIR_FREE(srv);
    }
    return ret;
}

/**
 * virNetlinkEventAddClient:
 *
 * @handleCB: callback to invoke when an event occurs
 * @removeCB: callback to invoke when removing a client
 * @opaque: user data to pass to callback
761 762
 * @macaddr: macaddr to store with the data. Used to identify callers.
 *           May be null.
763
 * @protocol: netlink protocol
764 765 766 767 768
 *
 * register a callback for handling of netlink messages. The
 * registered function receives the entire netlink message and
 * may choose to act upon it.
 *
769 770
 * Returns -1 if the file handle cannot be registered, number of
 * monitor upon success.
771 772 773 774
 */
int
virNetlinkEventAddClient(virNetlinkEventHandleCallback handleCB,
                         virNetlinkEventRemoveCallback removeCB,
775
                         void *opaque, const virMacAddr *macaddr,
776
                         unsigned int protocol)
777
{
778 779
    size_t i;
    int r, ret = -1;
780 781 782 783 784 785
    virNetlinkEventSrvPrivatePtr srv = NULL;

    if (protocol >= MAX_LINKS)
        return -EINVAL;

    srv = server[protocol];
786

787
    if (handleCB == NULL) {
788 789
        virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
                       _("Invalid NULL callback provided"));
790
        return -1;
791
    }
792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809

    virNetlinkEventServerLock(srv);

    VIR_DEBUG("adding client: %d.", nextWatch);

    r = 0;
    /* first try to re-use deleted free slots */
    for (i = 0; i < srv->handlesCount; i++) {
        if (srv->handles[i].deleted == VIR_NETLINK_HANDLE_DELETED) {
            r = i;
            goto addentry;
        }
    }
    /* Resize the eventLoop array if needed */
    if (srv->handlesCount == srv->handlesAlloc) {
        VIR_DEBUG("Used %zu handle slots, adding at least %d more",
                  srv->handlesAlloc, NETLINK_EVENT_ALLOC_EXTENT);
        if (VIR_RESIZE_N(srv->handles, srv->handlesAlloc,
810
                        srv->handlesCount, NETLINK_EVENT_ALLOC_EXTENT) < 0)
811 812 813 814
            goto error;
    }
    r = srv->handlesCount++;

815
 addentry:
816 817 818 819 820 821
    srv->handles[r].watch    = nextWatch;
    srv->handles[r].handleCB = handleCB;
    srv->handles[r].removeCB = removeCB;
    srv->handles[r].opaque   = opaque;
    srv->handles[r].deleted  = VIR_NETLINK_HANDLE_VALID;
    if (macaddr)
822
        virMacAddrSet(&srv->handles[r].macaddr, macaddr);
823
    else
824
        virMacAddrSetRaw(&srv->handles[r].macaddr,
E
Eric Blake 已提交
825
                         (unsigned char[VIR_MAC_BUFLEN]){0, 0, 0, 0, 0, 0});
826 827 828 829

    VIR_DEBUG("added client to loop slot: %d. with macaddr ptr=%p", r, macaddr);

    ret = nextWatch++;
830
 error:
831 832 833 834 835 836 837 838 839
    virNetlinkEventServerUnlock(srv);
    return ret;
}

/**
 * virNetlinkEventRemoveClient:
 *
 * @watch: watch whose handle to remove
 * @macaddr: macaddr whose handle to remove
840
 * @protocol: netlink protocol
841 842 843 844 845
 *
 * Unregister a callback from a netlink monitor.
 * The handler function referenced will no longer receive netlink messages.
 * Either watch or macaddr may be used, the other should be null.
 *
846
 * Returns -1 if the file handle was not registered, 0 upon success
847 848
 */
int
849
virNetlinkEventRemoveClient(int watch, const virMacAddr *macaddr,
850
                            unsigned int protocol)
851
{
852
    size_t i;
853
    int ret = -1;
854 855 856 857 858 859
    virNetlinkEventSrvPrivatePtr srv = NULL;

    if (protocol >= MAX_LINKS)
        return -EINVAL;

    srv = server[protocol];
860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875

    VIR_DEBUG("removing client watch=%d, mac=%p.", watch, macaddr);

    if (watch <= 0 && !macaddr) {
        VIR_WARN("Ignoring invalid netlink client id: %d", watch);
        return -1;
    }

    virNetlinkEventServerLock(srv);

    for (i = 0; i < srv->handlesCount; i++) {
        if (srv->handles[i].deleted != VIR_NETLINK_HANDLE_VALID)
            continue;

        if ((watch && srv->handles[i].watch == watch) ||
            (!watch &&
876
             virMacAddrCmp(macaddr, &srv->handles[i].macaddr) == 0)) {
877 878 879

            VIR_DEBUG("removed client: %d by %s.",
                      srv->handles[i].watch, watch ? "index" : "mac");
880
            virNetlinkEventRemoveClientPrimitive(i, protocol);
881 882 883 884 885 886
            ret = 0;
            goto cleanup;
        }
    }
    VIR_DEBUG("no client found to remove.");

887
 cleanup:
888 889 890 891
    virNetlinkEventServerUnlock(srv);
    return ret;
}

892 893
#else

894 895 896 897 898 899
# if defined(__linux)
static const char *unsupported = N_("libnl was not available at build time");
# else
static const char *unsupported = N_("not supported on non-linux platforms");
# endif

900 901 902 903 904 905 906 907 908 909 910 911
int
virNetlinkStartup(void)
{
    return 0;
}

void
virNetlinkShutdown(void)
{
    return;
}

912
int virNetlinkCommand(struct nl_msg *nl_msg ATTRIBUTE_UNUSED,
913
                      struct nlmsghdr **resp ATTRIBUTE_UNUSED,
914 915
                      unsigned int *respbuflen ATTRIBUTE_UNUSED,
                      uint32_t src_pid ATTRIBUTE_UNUSED,
916 917 918
                      uint32_t dst_pid ATTRIBUTE_UNUSED,
                      unsigned int protocol ATTRIBUTE_UNUSED,
                      unsigned int groups ATTRIBUTE_UNUSED)
919
{
920
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
921 922 923
    return -1;
}

924 925

int
926 927
virNetlinkDelLink(const char *ifname ATTRIBUTE_UNUSED,
                  virNetlinkDelLinkFallback fallback ATTRIBUTE_UNUSED)
928 929 930 931 932
{
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
    return -1;
}

933
/**
934 935
 * stopNetlinkEventServer: stop the monitor to receive netlink
 * messages for libvirtd
936
 */
937
int virNetlinkEventServiceStop(unsigned int protocol ATTRIBUTE_UNUSED)
938
{
939
    VIR_DEBUG("%s", _(unsupported));
940 941 942
    return 0;
}

943 944 945 946 947 948 949 950 951 952
/**
 * stopNetlinkEventServerAll: stop all the monitors to receive netlink
 * messages for libvirtd
 */
int virNetlinkEventServiceStopAll(void)
{
    VIR_DEBUG("%s", _(unsupported));
    return 0;
}

953
/**
954 955
 * startNetlinkEventServer: start a monitor to receive netlink
 * messages for libvirtd
956
 */
957 958
int virNetlinkEventServiceStart(unsigned int protocol ATTRIBUTE_UNUSED,
                                unsigned int groups ATTRIBUTE_UNUSED)
959
{
960
    VIR_DEBUG("%s", _(unsupported));
961 962 963 964
    return 0;
}

/**
965 966
 * virNetlinkEventServiceIsRunning: returns if the netlink event
 * service is running.
967
 */
968
bool virNetlinkEventServiceIsRunning(unsigned int protocol ATTRIBUTE_UNUSED)
969
{
970
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
971 972 973
    return 0;
}

974
int virNetlinkEventServiceLocalPid(unsigned int protocol ATTRIBUTE_UNUSED)
975
{
976
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
977 978 979
    return -1;
}

980
/**
981 982
 * virNetlinkEventAddClient: register a callback for handling of
 * netlink messages
983
 */
J
Jim Fehlig 已提交
984 985 986
int virNetlinkEventAddClient(virNetlinkEventHandleCallback handleCB ATTRIBUTE_UNUSED,
                             virNetlinkEventRemoveCallback removeCB ATTRIBUTE_UNUSED,
                             void *opaque ATTRIBUTE_UNUSED,
987
                             const virMacAddr *macaddr ATTRIBUTE_UNUSED,
988
                             unsigned int protocol ATTRIBUTE_UNUSED)
989
{
990
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
991 992 993 994 995 996
    return -1;
}

/**
 * virNetlinkEventRemoveClient: unregister a callback from a netlink monitor
 */
J
Jim Fehlig 已提交
997
int virNetlinkEventRemoveClient(int watch ATTRIBUTE_UNUSED,
998
                                const virMacAddr *macaddr ATTRIBUTE_UNUSED,
999
                                unsigned int protocol ATTRIBUTE_UNUSED)
1000
{
1001
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
1002 1003 1004
    return -1;
}

E
Eric Blake 已提交
1005 1006 1007 1008 1009 1010 1011 1012 1013

int
virNetlinkGetErrorCode(struct nlmsghdr *resp ATTRIBUTE_UNUSED,
                       unsigned int recvbuflen ATTRIBUTE_UNUSED)
{
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
    return -EINVAL;
}

1014
#endif /* __linux__ */