source3/lib/unix_msg/unix_msg.c

   1 /*
   2  * Unix SMB/CIFS implementation.
   3  * Copyright (C) Volker Lendecke 2013
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 3 of the License, or
   8  * (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  17  */
  18
  19 #include "replace.h"
  20 #include "unix_msg.h"
  21 #include "system/select.h"
  22 #include "system/time.h"
  23 #include "system/network.h"
  24 #include "dlinklist.h"
  25 #include "pthreadpool/pthreadpool.h"
  26 #include <fcntl.h>
  27
  28 /*
  29  * This file implements two abstractions: The "unix_dgram" functions implement
  30  * queueing for unix domain datagram sockets. You can send to a destination
  31  * socket, and if that has no free space available, it will fall back to an
  32  * anonymous socket that will poll for writability. "unix_dgram" expects the
  33  * data size not to exceed the system limit.
  34  *
  35  * The "unix_msg" functions implement the fragmentation of large messages on
  36  * top of "unix_dgram". This is what is exposed to the user of this API.
  37  */
  38
  39 struct unix_dgram_msg {
  40         struct unix_dgram_msg *prev, *next;
  41
  42         int sock;
  43         ssize_t sent;
  44         int sys_errno;
  45         size_t buflen;
  46         uint8_t buf[1];
  47 };
  48
  49 struct unix_dgram_send_queue {
  50         struct unix_dgram_send_queue *prev, *next;
  51         struct unix_dgram_ctx *ctx;
  52         int sock;
  53         struct unix_dgram_msg *msgs;
  54         char path[1];
  55 };
  56
  57 struct unix_dgram_ctx {
  58         int sock;
  59         pid_t created_pid;
  60         const struct poll_funcs *ev_funcs;
  61         size_t max_msg;
  62
  63         void (*recv_callback)(struct unix_dgram_ctx *ctx,
  64                               uint8_t *msg, size_t msg_len,
  65                               void *private_data);
  66         void *private_data;
  67
  68         struct poll_watch *sock_read_watch;
  69         struct unix_dgram_send_queue *send_queues;
  70
  71         struct pthreadpool *send_pool;
  72         struct poll_watch *pool_read_watch;
  73
  74         uint8_t *recv_buf;
  75         char path[1];
  76 };
  77
  78 static ssize_t iov_buflen(const struct iovec *iov, int iovlen);
  79 static void unix_dgram_recv_handler(struct poll_watch *w, int fd, short events,
  80                                     void *private_data);
  81
  82 /* Set socket non blocking. */
  83 static int prepare_socket_nonblock(int sock)
  84 {
  85         int flags;
  86 #ifdef O_NONBLOCK
  87 #define FLAG_TO_SET O_NONBLOCK
  88 #else
  89 #ifdef SYSV
  90 #define FLAG_TO_SET O_NDELAY
  91 #else /* BSD */
  92 #define FLAG_TO_SET FNDELAY
  93 #endif
  94 #endif
  95
  96         flags = fcntl(sock, F_GETFL);
  97         if (flags == -1) {
  98                 return errno;
  99         }
 100         flags |= FLAG_TO_SET;
 101         if (fcntl(sock, F_SETFL, flags) == -1) {
 102                 return errno;
 103         }
 104
 105 #undef FLAG_TO_SET
 106         return 0;
 107 }
 108
 109 /* Set socket close on exec. */
 110 static int prepare_socket_cloexec(int sock)
 111 {
 112 #ifdef FD_CLOEXEC
 113         int flags;
 114
 115         flags = fcntl(sock, F_GETFD, 0);
 116         if (flags == -1) {
 117                 return errno;
 118         }
 119         flags |= FD_CLOEXEC;
 120         if (fcntl(sock, F_SETFD, flags) == -1) {
 121                 return errno;
 122         }
 123 #endif
 124         return 0;
 125 }
 126
 127 /* Set socket non blocking and close on exec. */
 128 static int prepare_socket(int sock)
 129 {
 130         int ret = prepare_socket_nonblock(sock);
 131
 132         if (ret) {
 133                 return ret;
 134         }
 135         return prepare_socket_cloexec(sock);
 136 }
 137
 138 static int unix_dgram_init(const struct sockaddr_un *addr, size_t max_msg,
 139                            const struct poll_funcs *ev_funcs,
 140                            void (*recv_callback)(struct unix_dgram_ctx *ctx,
 141                                                  uint8_t *msg, size_t msg_len,
 142                                                  void *private_data),
 143                            void *private_data,
 144                            struct unix_dgram_ctx **result)
 145 {
 146         struct unix_dgram_ctx *ctx;
 147         size_t pathlen;
 148         int ret;
 149
 150         if (addr != NULL) {
 151                 pathlen = strlen(addr->sun_path)+1;
 152         } else {
 153                 pathlen = 1;
 154         }
 155
 156         ctx = malloc(offsetof(struct unix_dgram_ctx, path) + pathlen);
 157         if (ctx == NULL) {
 158                 return ENOMEM;
 159         }
 160         if (addr != NULL) {
 161                 memcpy(ctx->path, addr->sun_path, pathlen);
 162         } else {
 163                 ctx->path[0] = '\0';
 164         }
 165
 166         ctx->recv_buf = malloc(max_msg);
 167         if (ctx->recv_buf == NULL) {
 168                 free(ctx);
 169                 return ENOMEM;
 170         }
 171         ctx->max_msg = max_msg;
 172         ctx->ev_funcs = ev_funcs;
 173         ctx->recv_callback = recv_callback;
 174         ctx->private_data = private_data;
 175         ctx->sock_read_watch = NULL;
 176         ctx->send_pool = NULL;
 177         ctx->pool_read_watch = NULL;
 178         ctx->send_queues = NULL;
 179         ctx->created_pid = (pid_t)-1;
 180
 181         ctx->sock = socket(AF_UNIX, SOCK_DGRAM, 0);
 182         if (ctx->sock == -1) {
 183                 ret = errno;
 184                 goto fail_free;
 185         }
 186
 187         /* Set non-blocking and close-on-exec. */
 188         ret = prepare_socket(ctx->sock);
 189         if (ret != 0) {
 190                 goto fail_close;
 191         }
 192
 193         if (addr != NULL) {
 194                 ret = bind(ctx->sock,
 195                            (const struct sockaddr *)(const void *)addr,
 196                            sizeof(*addr));
 197                 if (ret == -1) {
 198                         ret = errno;
 199                         goto fail_close;
 200                 }
 201
 202                 ctx->created_pid = getpid();
 203
 204                 ctx->sock_read_watch = ctx->ev_funcs->watch_new(
 205                         ctx->ev_funcs, ctx->sock, POLLIN,
 206                         unix_dgram_recv_handler, ctx);
 207
 208                 if (ctx->sock_read_watch == NULL) {
 209                         ret = ENOMEM;
 210                         goto fail_close;
 211                 }
 212         }
 213
 214         *result = ctx;
 215         return 0;
 216
 217 fail_close:
 218         close(ctx->sock);
 219 fail_free:
 220         free(ctx->recv_buf);
 221         free(ctx);
 222         return ret;
 223 }
 224
 225 static void unix_dgram_recv_handler(struct poll_watch *w, int fd, short events,
 226                                     void *private_data)
 227 {
 228         struct unix_dgram_ctx *ctx = (struct unix_dgram_ctx *)private_data;
 229         ssize_t received;
 230         struct msghdr msg;
 231         struct iovec iov;
 232
 233         iov = (struct iovec) {
 234                 .iov_base = (void *)ctx->recv_buf,
 235                 .iov_len = ctx->max_msg,
 236         };
 237
 238         msg = (struct msghdr) {
 239                 .msg_iov = &iov,
 240                 .msg_iovlen = 1,
 241 #ifdef HAVE_STRUCT_MSGHDR_MSG_CONTROL
 242                 .msg_control = NULL,
 243                 .msg_controllen = 0,
 244 #endif
 245         };
 246
 247         received = recvmsg(fd, &msg, 0);
 248         if (received == -1) {
 249                 if ((errno == EAGAIN) ||
 250 #ifdef EWOULDBLOCK
 251                     (errno == EWOULDBLOCK) ||
 252 #endif
 253                     (errno == EINTR) || (errno == ENOMEM)) {
 254                         /* Not really an error - just try again. */
 255                         return;
 256                 }
 257                 /* Problem with the socket. Set it unreadable. */
 258                 ctx->ev_funcs->watch_update(w, 0);
 259                 return;
 260         }
 261         if (received > ctx->max_msg) {
 262                 /* More than we expected, not for us */
 263                 return;
 264         }
 265         ctx->recv_callback(ctx, ctx->recv_buf, received, ctx->private_data);
 266 }
 267
 268 static void unix_dgram_job_finished(struct poll_watch *w, int fd, short events,
 269                                     void *private_data);
 270
 271 static int unix_dgram_init_pthreadpool(struct unix_dgram_ctx *ctx)
 272 {
 273         int ret, signalfd;
 274
 275         if (ctx->send_pool != NULL) {
 276                 return 0;
 277         }
 278
 279         ret = pthreadpool_init(0, &ctx->send_pool);
 280         if (ret != 0) {
 281                 return ret;
 282         }
 283
 284         signalfd = pthreadpool_signal_fd(ctx->send_pool);
 285
 286         ctx->pool_read_watch = ctx->ev_funcs->watch_new(
 287                 ctx->ev_funcs, signalfd, POLLIN,
 288                 unix_dgram_job_finished, ctx);
 289         if (ctx->pool_read_watch == NULL) {
 290                 pthreadpool_destroy(ctx->send_pool);
 291                 ctx->send_pool = NULL;
 292                 return ENOMEM;
 293         }
 294
 295         return 0;
 296 }
 297
 298 static int unix_dgram_send_queue_init(
 299         struct unix_dgram_ctx *ctx, const struct sockaddr_un *dst,
 300         struct unix_dgram_send_queue **result)
 301 {
 302         struct unix_dgram_send_queue *q;
 303         size_t pathlen;
 304         int ret, err;
 305
 306         pathlen = strlen(dst->sun_path)+1;
 307
 308         q = malloc(offsetof(struct unix_dgram_send_queue, path) + pathlen);
 309         if (q == NULL) {
 310                 return ENOMEM;
 311         }
 312         q->ctx = ctx;
 313         q->msgs = NULL;
 314         memcpy(q->path, dst->sun_path, pathlen);
 315
 316         q->sock = socket(AF_UNIX, SOCK_DGRAM, 0);
 317         if (q->sock == -1) {
 318                 err = errno;
 319                 goto fail_free;
 320         }
 321
 322         err = prepare_socket_cloexec(q->sock);
 323         if (err != 0) {
 324                 goto fail_close;
 325         }
 326
 327         do {
 328                 ret = connect(q->sock,
 329                               (const struct sockaddr *)(const void *)dst,
 330                               sizeof(*dst));
 331         } while ((ret == -1) && (errno == EINTR));
 332
 333         if (ret == -1) {
 334                 err = errno;
 335                 goto fail_close;
 336         }
 337
 338         err = unix_dgram_init_pthreadpool(ctx);
 339         if (err != 0) {
 340                 goto fail_close;
 341         }
 342
 343         DLIST_ADD(ctx->send_queues, q);
 344
 345         *result = q;
 346         return 0;
 347
 348 fail_close:
 349         close(q->sock);
 350 fail_free:
 351         free(q);
 352         return err;
 353 }
 354
 355 static void unix_dgram_send_queue_free(struct unix_dgram_send_queue *q)
 356 {
 357         struct unix_dgram_ctx *ctx = q->ctx;
 358
 359         while (q->msgs != NULL) {
 360                 struct unix_dgram_msg *msg;
 361                 msg = q->msgs;
 362                 DLIST_REMOVE(q->msgs, msg);
 363                 free(msg);
 364         }
 365         close(q->sock);
 366         DLIST_REMOVE(ctx->send_queues, q);
 367         free(q);
 368 }
 369
 370 static struct unix_dgram_send_queue *find_send_queue(
 371         struct unix_dgram_ctx *ctx, const char *dst_sock)
 372 {
 373         struct unix_dgram_send_queue *s;
 374
 375         for (s = ctx->send_queues; s != NULL; s = s->next) {
 376                 if (strcmp(s->path, dst_sock) == 0) {
 377                         return s;
 378                 }
 379         }
 380         return NULL;
 381 }
 382
 383 static int queue_msg(struct unix_dgram_send_queue *q,
 384                      const struct iovec *iov, int iovlen)
 385 {
 386         struct unix_dgram_msg *msg;
 387         ssize_t buflen;
 388         size_t msglen;
 389         int i;
 390
 391         buflen = iov_buflen(iov, iovlen);
 392         if (buflen == -1) {
 393                 return EINVAL;
 394         }
 395
 396         msglen = offsetof(struct unix_dgram_msg, buf) + buflen;
 397         if ((msglen < buflen) ||
 398             (msglen < offsetof(struct unix_dgram_msg, buf))) {
 399                 /* overflow */
 400                 return EINVAL;
 401         }
 402
 403         msg = malloc(msglen);
 404         if (msg == NULL) {
 405                 return ENOMEM;
 406         }
 407         msg->buflen = buflen;
 408         msg->sock = q->sock;
 409
 410         buflen = 0;
 411         for (i=0; i<iovlen; i++) {
 412                 memcpy(&msg->buf[buflen], iov[i].iov_base, iov[i].iov_len);
 413                 buflen += iov[i].iov_len;
 414         }
 415
 416         DLIST_ADD_END(q->msgs, msg, struct unix_dgram_msg);
 417         return 0;
 418 }
 419
 420 static void unix_dgram_send_job(void *private_data)
 421 {
 422         struct unix_dgram_msg *msg = private_data;
 423
 424         do {
 425                 msg->sent = send(msg->sock, msg->buf, msg->buflen, 0);
 426         } while ((msg->sent == -1) && (errno == EINTR));
 427 }
 428
 429 static void unix_dgram_job_finished(struct poll_watch *w, int fd, short events,
 430                                     void *private_data)
 431 {
 432         struct unix_dgram_ctx *ctx = private_data;
 433         struct unix_dgram_send_queue *q;
 434         struct unix_dgram_msg *msg;
 435         int ret, job;
 436
 437         ret = pthreadpool_finished_jobs(ctx->send_pool, &job, 1);
 438         if (ret != 1) {
 439                 return;
 440         }
 441
 442         for (q = ctx->send_queues; q != NULL; q = q->next) {
 443                 if (job == q->sock) {
 444                         break;
 445                 }
 446         }
 447
 448         if (q == NULL) {
 449                 /* Huh? Should not happen */
 450                 return;
 451         }
 452
 453         msg = q->msgs;
 454         DLIST_REMOVE(q->msgs, msg);
 455         free(msg);
 456
 457         if (q->msgs != NULL) {
 458                 ret = pthreadpool_add_job(ctx->send_pool, q->sock,
 459                                           unix_dgram_send_job, q->msgs);
 460                 if (ret == 0) {
 461                         return;
 462                 }
 463         }
 464
 465         unix_dgram_send_queue_free(q);
 466 }
 467
 468 static int unix_dgram_send(struct unix_dgram_ctx *ctx,
 469                            const struct sockaddr_un *dst,
 470                            const struct iovec *iov, int iovlen)
 471 {
 472         struct unix_dgram_send_queue *q;
 473         struct msghdr msg;
 474         int ret;
 475
 476         /*
 477          * To preserve message ordering, we have to queue a message when
 478          * others are waiting in line already.
 479          */
 480         q = find_send_queue(ctx, dst->sun_path);
 481         if (q != NULL) {
 482                 return queue_msg(q, iov, iovlen);
 483         }
 484
 485         /*
 486          * Try a cheap nonblocking send
 487          */
 488
 489         msg.msg_name = discard_const_p(struct sockaddr_un, dst);
 490         msg.msg_namelen = sizeof(*dst);
 491         msg.msg_iov = discard_const_p(struct iovec, iov);
 492         msg.msg_iovlen = iovlen;
 493 #ifdef HAVE_STRUCT_MSGHDR_MSG_CONTROL
 494         msg.msg_control = NULL;
 495         msg.msg_controllen = 0;
 496 #endif
 497         msg.msg_flags = 0;
 498
 499         ret = sendmsg(ctx->sock, &msg, 0);
 500         if (ret >= 0) {
 501                 return 0;
 502         }
 503 #ifdef EWOULDBLOCK
 504         if ((errno != EWOULDBLOCK) && (errno != EAGAIN) && (errno != EINTR)) {
 505 #else
 506         if ((errno != EAGAIN) && (errno != EINTR)) {
 507 #endif
 508                 return errno;
 509         }
 510
 511         ret = unix_dgram_send_queue_init(ctx, dst, &q);
 512         if (ret != 0) {
 513                 return ret;
 514         }
 515         ret = queue_msg(q, iov, iovlen);
 516         if (ret != 0) {
 517                 unix_dgram_send_queue_free(q);
 518                 return ret;
 519         }
 520         ret = pthreadpool_add_job(ctx->send_pool, q->sock,
 521                                   unix_dgram_send_job, q->msgs);
 522         if (ret != 0) {
 523                 unix_dgram_send_queue_free(q);
 524                 return ret;
 525         }
 526         return 0;
 527 }
 528
 529 static int unix_dgram_sock(struct unix_dgram_ctx *ctx)
 530 {
 531         return ctx->sock;
 532 }
 533
 534 static int unix_dgram_free(struct unix_dgram_ctx *ctx)
 535 {
 536         if (ctx->send_queues != NULL) {
 537                 return EBUSY;
 538         }
 539
 540         if (ctx->send_pool != NULL) {
 541                 int ret = pthreadpool_destroy(ctx->send_pool);
 542                 if (ret != 0) {
 543                         return ret;
 544                 }
 545                 ctx->ev_funcs->watch_free(ctx->pool_read_watch);
 546         }
 547
 548         ctx->ev_funcs->watch_free(ctx->sock_read_watch);
 549
 550         if (getpid() == ctx->created_pid) {
 551                 /* If we created it, unlink. Otherwise someone else might
 552                  * still have it open */
 553                 unlink(ctx->path);
 554         }
 555
 556         close(ctx->sock);
 557         free(ctx->recv_buf);
 558         free(ctx);
 559         return 0;
 560 }
 561
 562 /*
 563  * Every message starts with a uint64_t cookie.
 564  *
 565  * A value of 0 indicates a single-fragment message which is complete in
 566  * itself. The data immediately follows the cookie.
 567  *
 568  * Every multi-fragment message has a cookie != 0 and starts with a cookie
 569  * followed by a struct unix_msg_header and then the data. The pid and sock
 570  * fields are used to assure uniqueness on the receiver side.
 571  */
 572
 573 struct unix_msg_hdr {
 574         size_t msglen;
 575         pid_t pid;
 576         int sock;
 577 };
 578
 579 struct unix_msg {
 580         struct unix_msg *prev, *next;
 581         size_t msglen;
 582         size_t received;
 583         pid_t sender_pid;
 584         int sender_sock;
 585         uint64_t cookie;
 586         uint8_t buf[1];
 587 };
 588
 589 struct unix_msg_ctx {
 590         struct unix_dgram_ctx *dgram;
 591         size_t fragment_len;
 592         uint64_t cookie;
 593
 594         void (*recv_callback)(struct unix_msg_ctx *ctx,
 595                               uint8_t *msg, size_t msg_len,
 596                               void *private_data);
 597         void *private_data;
 598
 599         struct unix_msg *msgs;
 600 };
 601
 602 static void unix_msg_recv(struct unix_dgram_ctx *ctx,
 603                           uint8_t *msg, size_t msg_len,
 604                           void *private_data);
 605
 606 int unix_msg_init(const char *path, const struct poll_funcs *ev_funcs,
 607                   size_t fragment_len, uint64_t cookie,
 608                   void (*recv_callback)(struct unix_msg_ctx *ctx,
 609                                         uint8_t *msg, size_t msg_len,
 610                                         void *private_data),
 611                   void *private_data,
 612                   struct unix_msg_ctx **result)
 613 {
 614         struct unix_msg_ctx *ctx;
 615         struct sockaddr_un addr;
 616         struct sockaddr_un *paddr = NULL;
 617         int ret;
 618
 619         ctx = malloc(sizeof(*ctx));
 620         if (ctx == NULL) {
 621                 return ENOMEM;
 622         }
 623
 624         if (path != NULL) {
 625                 size_t pathlen = strlen(path)+1;
 626
 627                 if (pathlen > sizeof(addr.sun_path)) {
 628                         return ENAMETOOLONG;
 629                 }
 630                 addr = (struct sockaddr_un) { .sun_family = AF_UNIX };
 631                 memcpy(addr.sun_path, path, pathlen);
 632                 paddr = &addr;
 633         }
 634
 635         ret = unix_dgram_init(paddr, fragment_len, ev_funcs,
 636                               unix_msg_recv, ctx, &ctx->dgram);
 637         if (ret != 0) {
 638                 free(ctx);
 639                 return ret;
 640         }
 641
 642         ctx->fragment_len = fragment_len;
 643         ctx->cookie = cookie;
 644         ctx->recv_callback = recv_callback;
 645         ctx->private_data = private_data;
 646         ctx->msgs = NULL;
 647
 648         *result = ctx;
 649         return 0;
 650 }
 651
 652 int unix_msg_send(struct unix_msg_ctx *ctx, const char *dst_sock,
 653                   const struct iovec *iov, int iovlen)
 654 {
 655         ssize_t msglen;
 656         size_t sent;
 657         int ret = 0;
 658         struct iovec *iov_copy;
 659         struct unix_msg_hdr hdr;
 660         struct iovec src_iov;
 661         struct sockaddr_un dst;
 662         size_t dst_len;
 663
 664         dst_len = strlen(dst_sock);
 665         if (dst_len >= sizeof(dst.sun_path)) {
 666                 return ENAMETOOLONG;
 667         }
 668         dst = (struct sockaddr_un) { .sun_family = AF_UNIX };
 669         memcpy(dst.sun_path, dst_sock, dst_len);
 670
 671         if (iovlen < 0) {
 672                 return EINVAL;
 673         }
 674
 675         msglen = iov_buflen(iov, iovlen);
 676         if (msglen == -1) {
 677                 return EINVAL;
 678         }
 679
 680         if (msglen <= (ctx->fragment_len - sizeof(uint64_t))) {
 681                 struct iovec tmp_iov[iovlen+1];
 682                 uint64_t cookie = 0;
 683
 684                 tmp_iov[0].iov_base = &cookie;
 685                 tmp_iov[0].iov_len = sizeof(cookie);
 686                 if (iovlen > 0) {
 687                         memcpy(&tmp_iov[1], iov,
 688                                sizeof(struct iovec) * iovlen);
 689                 }
 690
 691                 return unix_dgram_send(ctx->dgram, &dst, tmp_iov, iovlen+1);
 692         }
 693
 694         hdr.msglen = msglen;
 695         hdr.pid = getpid();
 696         hdr.sock = unix_dgram_sock(ctx->dgram);
 697
 698         iov_copy = malloc(sizeof(struct iovec) * (iovlen + 2));
 699         if (iov_copy == NULL) {
 700                 return ENOMEM;
 701         }
 702         iov_copy[0].iov_base = &ctx->cookie;
 703         iov_copy[0].iov_len = sizeof(ctx->cookie);
 704         iov_copy[1].iov_base = &hdr;
 705         iov_copy[1].iov_len = sizeof(hdr);
 706
 707         sent = 0;
 708         src_iov = iov[0];
 709
 710         /*
 711          * The following write loop sends the user message in pieces. We have
 712          * filled the first two iovecs above with "cookie" and "hdr". In the
 713          * following loops we pull message chunks from the user iov array and
 714          * fill iov_copy piece by piece, possibly truncating chunks from the
 715          * caller's iov array. Ugly, but hopefully efficient.
 716          */
 717
 718         while (sent < msglen) {
 719                 size_t fragment_len;
 720                 size_t iov_index = 2;
 721
 722                 fragment_len = sizeof(ctx->cookie) + sizeof(hdr);
 723
 724                 while (fragment_len < ctx->fragment_len) {
 725                         size_t space, chunk;
 726
 727                         space = ctx->fragment_len - fragment_len;
 728                         chunk = MIN(space, src_iov.iov_len);
 729
 730                         iov_copy[iov_index].iov_base = src_iov.iov_base;
 731                         iov_copy[iov_index].iov_len = chunk;
 732                         iov_index += 1;
 733
 734                         src_iov.iov_base = (char *)src_iov.iov_base + chunk;
 735                         src_iov.iov_len -= chunk;
 736                         fragment_len += chunk;
 737
 738                         if (src_iov.iov_len == 0) {
 739                                 iov += 1;
 740                                 iovlen -= 1;
 741                                 if (iovlen == 0) {
 742                                         break;
 743                                 }
 744                                 src_iov = iov[0];
 745                         }
 746                 }
 747                 sent += (fragment_len - sizeof(ctx->cookie) - sizeof(hdr));
 748
 749                 ret = unix_dgram_send(ctx->dgram, &dst, iov_copy, iov_index);
 750                 if (ret != 0) {
 751                         break;
 752                 }
 753         }
 754
 755         free(iov_copy);
 756
 757         ctx->cookie += 1;
 758         if (ctx->cookie == 0) {
 759                 ctx->cookie += 1;
 760         }
 761
 762         return ret;
 763 }
 764
 765 static void unix_msg_recv(struct unix_dgram_ctx *dgram_ctx,
 766                           uint8_t *buf, size_t buflen,
 767                           void *private_data)
 768 {
 769         struct unix_msg_ctx *ctx = (struct unix_msg_ctx *)private_data;
 770         struct unix_msg_hdr hdr;
 771         struct unix_msg *msg;
 772         size_t space;
 773         uint64_t cookie;
 774
 775         if (buflen < sizeof(cookie)) {
 776                 return;
 777         }
 778         memcpy(&cookie, buf, sizeof(cookie));
 779
 780         buf += sizeof(cookie);
 781         buflen -= sizeof(cookie);
 782
 783         if (cookie == 0) {
 784                 ctx->recv_callback(ctx, buf, buflen, ctx->private_data);
 785                 return;
 786         }
 787
 788         if (buflen < sizeof(hdr)) {
 789                 return;
 790         }
 791         memcpy(&hdr, buf, sizeof(hdr));
 792
 793         buf += sizeof(hdr);
 794         buflen -= sizeof(hdr);
 795
 796         for (msg = ctx->msgs; msg != NULL; msg = msg->next) {
 797                 if ((msg->sender_pid == hdr.pid) &&
 798                     (msg->sender_sock == hdr.sock)) {
 799                         break;
 800                 }
 801         }
 802
 803         if ((msg != NULL) && (msg->cookie != cookie)) {
 804                 DLIST_REMOVE(ctx->msgs, msg);
 805                 free(msg);
 806                 msg = NULL;
 807         }
 808
 809         if (msg == NULL) {
 810                 msg = malloc(offsetof(struct unix_msg, buf) + hdr.msglen);
 811                 if (msg == NULL) {
 812                         return;
 813                 }
 814                 msg->msglen = hdr.msglen;
 815                 msg->received = 0;
 816                 msg->sender_pid = hdr.pid;
 817                 msg->sender_sock = hdr.sock;
 818                 msg->cookie = cookie;
 819                 DLIST_ADD(ctx->msgs, msg);
 820         }
 821
 822         space = msg->msglen - msg->received;
 823         if (buflen > space) {
 824                 return;
 825         }
 826
 827         memcpy(msg->buf + msg->received, buf, buflen);
 828         msg->received += buflen;
 829
 830         if (msg->received < msg->msglen) {
 831                 return;
 832         }
 833
 834         DLIST_REMOVE(ctx->msgs, msg);
 835         ctx->recv_callback(ctx, msg->buf, msg->msglen, ctx->private_data);
 836         free(msg);
 837 }
 838
 839 int unix_msg_free(struct unix_msg_ctx *ctx)
 840 {
 841         int ret;
 842
 843         ret = unix_dgram_free(ctx->dgram);
 844         if (ret != 0) {
 845                 return ret;
 846         }
 847
 848         while (ctx->msgs != NULL) {
 849                 struct unix_msg *msg = ctx->msgs;
 850                 DLIST_REMOVE(ctx->msgs, msg);
 851                 free(msg);
 852         }
 853
 854         free(ctx);
 855         return 0;
 856 }
 857
 858 static ssize_t iov_buflen(const struct iovec *iov, int iovlen)
 859 {
 860         size_t buflen = 0;
 861         int i;
 862
 863         for (i=0; i<iovlen; i++) {
 864                 size_t thislen = iov[i].iov_len;
 865                 size_t tmp = buflen + thislen;
 866
 867                 if ((tmp < buflen) || (tmp < thislen)) {
 868                         /* overflow */
 869                         return -1;
 870                 }
 871                 buflen = tmp;
 872         }
 873         return buflen;
 874 }