bsd/netinet/in_tclass.c

   1 /*
   2  * Copyright (c) 2009-2011 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28
  29 #include <sys/systm.h>
  30 #include <sys/kernel.h>
  31 #include <sys/types.h>
  32 #include <sys/filedesc.h>
  33 #include <sys/file_internal.h>
  34 #include <sys/proc.h>
  35 #include <sys/socket.h>
  36 #include <sys/socketvar.h>
  37 #include <sys/errno.h>
  38 #include <sys/protosw.h>
  39 #include <sys/domain.h>
  40 #include <sys/mbuf.h>
  41 #include <sys/queue.h>
  42
  43 #include <net/if.h>
  44 #include <net/route.h>
  45
  46 #include <netinet/in.h>
  47 #include <netinet/in_var.h>
  48 #include <netinet/in_pcb.h>
  49 #include <netinet/ip.h>
  50 #include <netinet/ip_var.h>
  51 #include <netinet/ip6.h>
  52 #include <netinet6/ip6_var.h>
  53 #include <netinet/udp.h>
  54 #include <netinet/udp_var.h>
  55 #include <netinet/tcp.h>
  56 #include <netinet/tcp_var.h>
  57 #include <netinet/tcp_cc.h>
  58
  59 extern char *proc_name_address(void *p);
  60
  61 static int tfp_count = 0;
  62
  63 static TAILQ_HEAD(, tclass_for_proc) tfp_head = TAILQ_HEAD_INITIALIZER(tfp_head);
  64
  65 struct tclass_for_proc {
  66         TAILQ_ENTRY(tclass_for_proc)    tfp_link;
  67         int                                                     tfp_class;
  68         pid_t                                                   tfp_pid;
  69         char                                                    tfp_pname[MAXCOMLEN + 1];
  70 };
  71
  72 extern void tcp_set_background_cc(struct socket *);
  73 extern void tcp_set_foreground_cc(struct socket *);
  74
  75 int dscp_code_from_mbuf_tclass(int );
  76
  77 static int get_pid_tclass(pid_t , int *);
  78 static int get_pname_tclass(const char * , int *);
  79 static int set_pid_tclass(pid_t , int );
  80 static int set_pname_tclass(const char * , int );
  81 static int purge_tclass_for_proc(void);
  82 static int flush_tclass_for_proc(void);
  83
  84
  85 static lck_grp_attr_t *tclass_lck_grp_attr = NULL;  /* mutex group attributes */
  86 static lck_grp_t *tclass_lck_grp = NULL;            /* mutex group definition */
  87 static lck_attr_t *tclass_lck_attr = NULL;          /* mutex attributes */
  88 static lck_mtx_t *tclass_lock = NULL;
  89
  90 /*
  91  * Must be called with tclass_lock held
  92  */
  93 static struct tclass_for_proc *
  94 find_tfp_by_pid(pid_t pid)
  95 {
  96         struct tclass_for_proc *tfp;
  97
  98         TAILQ_FOREACH(tfp, &tfp_head, tfp_link) {
  99                 if (tfp->tfp_pid == pid)
 100                         break;
 101         }
 102         return tfp;
 103 }
 104
 105 /*
 106  * Must be called with tclass_lock held
 107  */
 108 static struct tclass_for_proc *
 109 find_tfp_by_pname(const char *pname)
 110 {
 111         struct tclass_for_proc *tfp;
 112
 113         TAILQ_FOREACH(tfp, &tfp_head, tfp_link) {
 114                 if (strncmp(pname, tfp->tfp_pname, sizeof(tfp->tfp_pname)) == 0)
 115                         break;
 116         }
 117         return tfp;
 118 }
 119
 120 static int
 121 get_tclass_for_curr_proc(void)
 122 {
 123         struct tclass_for_proc *tfp;
 124         int sotc = SO_TC_BE;
 125         proc_t p = current_proc();      /* Not ref counted */
 126         pid_t pid = proc_pid(p);
 127         char *pname = proc_name_address(p);
 128
 129         lck_mtx_lock(tclass_lock);
 130
 131         TAILQ_FOREACH(tfp, &tfp_head, tfp_link) {
 132                 if ((tfp->tfp_pid == pid) ||
 133                         (tfp->tfp_pid == -1 && strncmp(pname, tfp->tfp_pname, sizeof(tfp->tfp_pname)) == 0)) {
 134                         sotc = tfp->tfp_class;
 135                         break;
 136                 }
 137         }
 138
 139         lck_mtx_unlock(tclass_lock);
 140
 141         return sotc;
 142 }
 143
 144 /*
 145  * Purge entries with PIDs of exited processes
 146  */
 147 int
 148 purge_tclass_for_proc(void)
 149 {
 150         int error = 0;
 151         struct tclass_for_proc *tfp, *tvar;
 152
 153         lck_mtx_lock(tclass_lock);
 154
 155         TAILQ_FOREACH_SAFE(tfp, &tfp_head, tfp_link, tvar) {
 156                 proc_t p;
 157
 158                 if (tfp->tfp_pid == -1)
 159                         continue;
 160                 if ((p = proc_find(tfp->tfp_pid)) == NULL) {
 161                         tfp_count--;
 162                         TAILQ_REMOVE(&tfp_head, tfp, tfp_link);
 163
 164                         _FREE(tfp, M_TEMP);
 165                 } else {
 166                         proc_rele(p);
 167                 }
 168         }
 169
 170         lck_mtx_unlock(tclass_lock);
 171
 172         return error;
 173 }
 174
 175 /*
 176  * Remove one entry
 177  * Must be called with tclass_lock held
 178  */
 179 static void
 180 free_tclass_for_proc(struct tclass_for_proc *tfp)
 181 {
 182         if (tfp == NULL)
 183                 return;
 184         tfp_count--;
 185         TAILQ_REMOVE(&tfp_head, tfp, tfp_link);
 186         _FREE(tfp, M_TEMP);
 187 }
 188
 189 /*
 190  * Remove all entries
 191  */
 192 int
 193 flush_tclass_for_proc(void)
 194 {
 195         int error = 0;
 196         struct tclass_for_proc *tfp, *tvar;
 197
 198         lck_mtx_lock(tclass_lock);
 199
 200         TAILQ_FOREACH_SAFE(tfp, &tfp_head, tfp_link, tvar) {
 201                 free_tclass_for_proc(tfp);
 202         }
 203
 204         lck_mtx_unlock(tclass_lock);
 205
 206         return error;
 207
 208 }
 209
 210 /*
 211  * Must be called with tclass_lock held
 212  */
 213 static struct tclass_for_proc *
 214 alloc_tclass_for_proc(pid_t pid, const char *pname, int tclass)
 215 {
 216         struct tclass_for_proc *tfp;
 217
 218         if (pid == -1 && pname == NULL)
 219                 return NULL;
 220
 221         tfp = _MALLOC(sizeof(struct tclass_for_proc), M_TEMP, M_NOWAIT | M_ZERO);
 222         if (tfp == NULL)
 223                 return NULL;
 224
 225         tfp->tfp_pid = pid;
 226         tfp->tfp_class = tclass;
 227         /*
 228          * Add per pid entries before per proc name so we can find
 229          * a specific instance of a process before the general name base entry.
 230          */
 231         if (pid != -1) {
 232                 TAILQ_INSERT_HEAD(&tfp_head, tfp, tfp_link);
 233         } else {
 234                 strlcpy(tfp->tfp_pname, pname, sizeof(tfp->tfp_pname));
 235                 TAILQ_INSERT_TAIL(&tfp_head, tfp, tfp_link);
 236         }
 237
 238         tfp_count++;
 239
 240         return tfp;
 241 }
 242
 243 /*
 244  * -1 for tclass means to remove the entry
 245  */
 246 int
 247 set_pid_tclass(pid_t pid, int tclass)
 248 {
 249         int error = EINVAL;
 250         proc_t p = NULL;
 251         struct filedesc *fdp;
 252         struct fileproc *fp;
 253         struct tclass_for_proc *tfp;
 254         int i;
 255
 256         p = proc_find(pid);
 257         if (p == NULL) {
 258                 printf("set_pid_tclass proc_find(%d) \n", pid);
 259                 goto done;
 260         }
 261
 262         /* Need a tfp */
 263         lck_mtx_lock(tclass_lock);
 264
 265         tfp = find_tfp_by_pid(pid);
 266         if (tclass == -1) {
 267                 if (tfp != NULL) {
 268                         free_tclass_for_proc(tfp);
 269                         error = 0;
 270                 }
 271                 lck_mtx_unlock(tclass_lock);
 272                 goto done;
 273         } else {
 274                 if (tfp == NULL) {
 275                         tfp = alloc_tclass_for_proc(pid, NULL, tclass);
 276                         if (tfp == NULL) {
 277                                 lck_mtx_unlock(tclass_lock);
 278                                 error = ENOBUFS;
 279                                 goto done;
 280                         }
 281                 } else {
 282                         tfp->tfp_class = tclass;
 283                 }
 284         }
 285         lck_mtx_unlock(tclass_lock);
 286
 287         if (tfp != NULL) {
 288                 proc_fdlock(p);
 289
 290                 fdp = p->p_fd;
 291                 for (i = 0; i < fdp->fd_nfiles; i++) {
 292                         struct socket *so;
 293
 294                         fp = fdp->fd_ofiles[i];
 295                         if (fp == NULL || (fdp->fd_ofileflags[i] & UF_RESERVED) != 0 ||
 296                                 fp->f_fglob->fg_type != DTYPE_SOCKET)
 297                                 continue;
 298
 299                         so = (struct socket *)fp->f_fglob->fg_data;
 300                         if (so->so_proto->pr_domain->dom_family != AF_INET &&
 301                                 so->so_proto->pr_domain->dom_family != AF_INET6)
 302                                 continue;
 303                         socket_lock(so, 1);
 304                         error = so_set_traffic_class(so, tclass != -1 ? tclass : SO_TC_BE);
 305                         socket_unlock(so, 1);
 306                         if (error != 0) {
 307                                 printf("set_pid_tclass so_set_traffic_class(%p, %d) failed %d\n", so, tclass, error);
 308                                 error = 0;
 309                         }
 310                 }
 311
 312                 proc_fdunlock(p);
 313         }
 314
 315         error = 0;
 316 done:
 317         if (p != NULL)
 318                 proc_rele(p);
 319
 320         return error;
 321 }
 322
 323 int
 324 set_pname_tclass(const char *pname, int tclass)
 325 {
 326         int error = EINVAL;
 327         struct tclass_for_proc *tfp;
 328
 329         lck_mtx_lock(tclass_lock);
 330
 331         tfp = find_tfp_by_pname(pname);
 332         if (tclass == -1) {
 333                 if (tfp != NULL)
 334                         free_tclass_for_proc(tfp);
 335         } else {
 336                 if (tfp == NULL) {
 337                         tfp = alloc_tclass_for_proc(-1, pname, tclass);
 338                         if (tfp == NULL) {
 339                                 lck_mtx_unlock(tclass_lock);
 340                                 error = ENOBUFS;
 341                                 goto done;
 342                         }
 343                 } else {
 344                         tfp->tfp_class = tclass;
 345                 }
 346         }
 347         lck_mtx_unlock(tclass_lock);
 348
 349         error = 0;
 350 done:
 351
 352         return error;
 353 }
 354
 355 int
 356 get_pid_tclass(pid_t pid, int *tclass)
 357 {
 358         int error = EINVAL;
 359         proc_t p = NULL;
 360         struct tclass_for_proc *tfp;
 361
 362         *tclass = -1; /* Means not set */
 363
 364         p = proc_find(pid);
 365         if (p == NULL) {
 366                 printf("get_pid_tclass proc_find(%d) \n", pid);
 367                 goto done;
 368         }
 369
 370         /* Need a tfp */
 371         lck_mtx_lock(tclass_lock);
 372
 373         tfp = find_tfp_by_pid(pid);
 374         if (tfp != NULL) {
 375                 *tclass = tfp->tfp_class ;
 376                 error = 0;
 377         }
 378         lck_mtx_unlock(tclass_lock);
 379 done:
 380         if (p != NULL)
 381                 proc_rele(p);
 382
 383         return error;
 384 }
 385
 386 int
 387 get_pname_tclass(const char *pname, int *tclass)
 388 {
 389         int error = EINVAL;
 390         struct tclass_for_proc *tfp;
 391
 392         *tclass = -1; /* Means not set */
 393
 394         /* Need a tfp */
 395         lck_mtx_lock(tclass_lock);
 396
 397         tfp = find_tfp_by_pname(pname);
 398         if (tfp != NULL) {
 399                 *tclass = tfp->tfp_class ;
 400                 error = 0;
 401         }
 402         lck_mtx_unlock(tclass_lock);
 403
 404         return error;
 405 }
 406
 407
 408
 409 /*
 410  * Setting options requires privileges
 411  */
 412 __private_extern__ int
 413 so_set_tcdbg(struct socket *so, struct so_tcdbg *so_tcdbg)
 414 {
 415         int error = 0;
 416
 417         if ((so->so_state & SS_PRIV) == 0)
 418                 return EPERM;
 419
 420         socket_unlock(so, 0);
 421
 422         switch (so_tcdbg->so_tcdbg_cmd) {
 423                 case SO_TCDBG_PID:
 424                         error = set_pid_tclass(so_tcdbg->so_tcdbg_pid, so_tcdbg->so_tcdbg_tclass);
 425                         break;
 426
 427                 case SO_TCDBG_PNAME:
 428                         error = set_pname_tclass(so_tcdbg->so_tcdbg_pname, so_tcdbg->so_tcdbg_tclass);
 429                         break;
 430
 431                 case SO_TCDBG_PURGE:
 432                         error = purge_tclass_for_proc();
 433                         break;
 434
 435                 case SO_TCDBG_FLUSH:
 436                         error = flush_tclass_for_proc();
 437                         break;
 438
 439                 default:
 440                         error = EINVAL;
 441                         break;
 442
 443         }
 444
 445         socket_lock(so, 0);
 446
 447         return error;
 448 }
 449
 450 /*
 451  * Not required to be privileged to get
 452  */
 453 __private_extern__ int
 454 sogetopt_tcdbg(struct socket *so, struct sockopt *sopt)
 455 {
 456         int error = 0;
 457         struct so_tcdbg so_tcdbg;
 458         void *buf = NULL;
 459         size_t len = sopt->sopt_valsize;
 460
 461         error = sooptcopyin(sopt, &so_tcdbg, sizeof(struct so_tcdbg), sizeof(struct so_tcdbg));
 462         if (error != 0)
 463                 return error;
 464
 465         sopt->sopt_valsize = len;
 466
 467         socket_unlock(so, 0);
 468
 469         switch (so_tcdbg.so_tcdbg_cmd) {
 470                 case SO_TCDBG_PID:
 471                         error = get_pid_tclass(so_tcdbg.so_tcdbg_pid, &so_tcdbg.so_tcdbg_tclass);
 472                         break;
 473
 474                 case SO_TCDBG_PNAME:
 475                         error = get_pname_tclass(so_tcdbg.so_tcdbg_pname, &so_tcdbg.so_tcdbg_tclass);
 476                         break;
 477
 478                 case SO_TCDBG_COUNT:
 479                         lck_mtx_lock(tclass_lock);
 480                         so_tcdbg.so_tcdbg_count = tfp_count;
 481                         lck_mtx_unlock(tclass_lock);
 482                         break;
 483
 484                 case SO_TCDBG_LIST: {
 485                         struct tclass_for_proc *tfp;
 486                         int n, alloc_count;
 487                         struct so_tcdbg *ptr;
 488
 489                         lck_mtx_lock(tclass_lock);
 490                         if ((alloc_count = tfp_count) == 0) {
 491                                 lck_mtx_unlock(tclass_lock);
 492                                 error = EINVAL;
 493                                 break;
 494                         }
 495                         len = alloc_count * sizeof(struct so_tcdbg);
 496                         lck_mtx_unlock(tclass_lock);
 497
 498                         buf = _MALLOC(len, M_TEMP, M_WAITOK | M_ZERO);
 499                         if (buf == NULL) {
 500                                 error = ENOBUFS;
 501                                 break;
 502                         }
 503
 504                         lck_mtx_lock(tclass_lock);
 505                         n = 0;
 506                         ptr = (struct so_tcdbg *)buf;
 507                         TAILQ_FOREACH(tfp, &tfp_head, tfp_link) {
 508                                 if (++n > alloc_count)
 509                                         break;
 510                                 if (tfp->tfp_pid != -1) {
 511                                         ptr->so_tcdbg_cmd = SO_TCDBG_PID;
 512                                         ptr->so_tcdbg_pid = tfp->tfp_pid;
 513                                 } else {
 514                                         ptr->so_tcdbg_cmd = SO_TCDBG_PNAME;
 515                                         ptr->so_tcdbg_pid = -1;
 516                                         strlcpy(ptr->so_tcdbg_pname, tfp->tfp_pname, sizeof(ptr->so_tcdbg_pname));
 517                                 }
 518                                 ptr->so_tcdbg_tclass = tfp->tfp_class;
 519                                 ptr++;
 520                         }
 521
 522                         lck_mtx_unlock(tclass_lock);
 523                         }
 524                         break;
 525
 526                 default:
 527                         error = EINVAL;
 528                         break;
 529
 530         }
 531
 532         socket_lock(so, 0);
 533
 534         if (error == 0) {
 535                 if (buf == NULL) {
 536                         error = sooptcopyout(sopt, &so_tcdbg, sizeof(struct so_tcdbg));
 537                 } else {
 538                         error = sooptcopyout(sopt, buf, len);
 539                         _FREE(buf, M_TEMP);
 540                 }
 541         }
 542         return error;
 543 }
 544
 545
 546 __private_extern__ int
 547 so_set_traffic_class(struct socket *so, int optval)
 548 {
 549         int error = 0;
 550
 551         if (optval < SO_TC_BE || optval > SO_TC_VO) {
 552                 error = EINVAL;
 553         } else {
 554                 so->so_traffic_class = optval;
 555
 556                 if ((INP_SOCKAF(so) == AF_INET || INP_SOCKAF(so) == AF_INET6) &&
 557                         INP_SOCKTYPE(so) == SOCK_STREAM) {
 558                         set_tcp_stream_priority(so);
 559                 }
 560         }
 561         return error;
 562 }
 563
 564 __private_extern__ void
 565 so_set_default_traffic_class(struct socket *so)
 566 {
 567         int sotc = SO_TC_BE;
 568
 569         if (tfp_count > 0 && (INP_SOCKAF(so) == AF_INET || INP_SOCKAF(so) == AF_INET6)) {
 570                 sotc = get_tclass_for_curr_proc();
 571         }
 572
 573         so->so_traffic_class = sotc;
 574
 575         return;
 576 }
 577
 578
 579 __private_extern__ int
 580 mbuf_traffic_class_from_control(struct mbuf *control)
 581 {
 582         struct cmsghdr *cm;
 583
 584         for (cm = M_FIRST_CMSGHDR(control);
 585                  cm != NULL;
 586                  cm = M_NXT_CMSGHDR(control, cm)) {
 587                 int tc;
 588
 589                 if (cm->cmsg_len < sizeof(struct cmsghdr))
 590                         break;
 591
 592                 if (cm->cmsg_level != SOL_SOCKET ||
 593                         cm->cmsg_type != SO_TRAFFIC_CLASS)
 594                         continue;
 595                 if (cm->cmsg_len != CMSG_LEN(sizeof(int)))
 596                         continue;
 597
 598                 tc = *(int *)CMSG_DATA(cm);
 599
 600                 switch (tc) {
 601                         case SO_TC_BE:
 602                                 return MBUF_TC_BE;
 603                         case SO_TC_BK:
 604                                 return MBUF_TC_BK;
 605                         case SO_TC_VI:
 606                                 return MBUF_TC_VI;
 607                         case SO_TC_VO:
 608                                 return MBUF_TC_VO;
 609                         default:
 610                                 break;
 611                 }
 612         }
 613
 614         return MBUF_TC_UNSPEC;
 615 }
 616
 617 __private_extern__  int
 618 dscp_code_from_mbuf_tclass(int mtc)
 619 {
 620         int dscp_code;
 621
 622         switch (mtc) {
 623                 default:
 624                 case MBUF_TC_BE:
 625                         dscp_code = 0;
 626                         break;
 627                 case MBUF_TC_BK:
 628                         dscp_code = 0x08;
 629                         break;
 630                 case MBUF_TC_VI:
 631                         dscp_code = 0x20;
 632                         break;
 633                 case MBUF_TC_VO:
 634                         dscp_code = 0x30;
 635                         break;
 636         }
 637
 638         return dscp_code;
 639 }
 640
 641 __private_extern__ void
 642 so_recv_data_stat(struct socket *so, struct mbuf *m, size_t off)
 643 {
 644         uint32_t sotc = m->m_pkthdr.prio;
 645
 646         if (sotc >= SO_TC_STATS_MAX)
 647                 sotc = SO_TC_BE;
 648
 649         so->so_tc_stats[sotc].rxpackets += 1;
 650         so->so_tc_stats[sotc].rxbytes += ((m->m_flags & M_PKTHDR) ? m->m_pkthdr.len : 0) + off;
 651
 652         return;
 653 }
 654
 655 __private_extern__ void
 656 set_tcp_stream_priority(struct socket *so)
 657 {
 658         struct tcpcb *tp = intotcpcb(sotoinpcb(so));
 659
 660         /* If the socket was marked as a background socket or if the
 661          * traffic class is set to background with traffic class socket
 662          * option then make both send and recv side of the stream to be
 663          * background. The variable sotcdb which can be set with sysctl
 664          * is used to disable these settings for testing.
 665          */
 666         if (soisbackground(so) || so->so_traffic_class == SO_TC_BK) {
 667                 if ((sotcdb & SOTCDB_NO_SENDTCPBG) != 0) {
 668                         if (tp->tcp_cc_index == TCP_CC_ALGO_BACKGROUND_INDEX)
 669                                 tcp_set_foreground_cc(so);
 670                 } else {
 671                         if (tp->tcp_cc_index != TCP_CC_ALGO_BACKGROUND_INDEX)
 672                                 tcp_set_background_cc(so);
 673                 }
 674
 675                 /* Set receive side background flags */
 676                 if ((sotcdb & SOTCDB_NO_RECVTCPBG) != 0) {
 677                         so->so_traffic_mgt_flags &= ~(TRAFFIC_MGT_TCP_RECVBG);
 678                 } else {
 679                         so->so_traffic_mgt_flags |= TRAFFIC_MGT_TCP_RECVBG;
 680                 }
 681         } else {
 682                 so->so_traffic_mgt_flags &= ~(TRAFFIC_MGT_TCP_RECVBG);
 683                 if (tp->tcp_cc_index == TCP_CC_ALGO_BACKGROUND_INDEX)
 684                         tcp_set_foreground_cc(so);
 685         }
 686         return;
 687 }
 688
 689 /*
 690  * Set traffic class to an IPv4 or IPv6 packet
 691  * - mark the mbuf
 692  * - set the DSCP code following the WMM mapping
 693  */
 694 __private_extern__ void
 695 set_packet_tclass(struct mbuf *m, struct socket *so, int in_mtc, int isipv6)
 696 {
 697         int mtc = MBUF_TC_BE; /* Best effort by default */
 698         struct inpcb *inp = sotoinpcb(so);       /* in6pcb and inpcb are the same */
 699         struct ip *ip = mtod(m, struct ip *);
 700 #if INET6
 701         struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
 702 #endif /* INET6 */
 703
 704         if (!(m->m_flags & M_PKTHDR))
 705                 return;
 706
 707         /*
 708          * Here is the precedence:
 709          * 1) TRAFFIC_MGT_SO_BACKGROUND trumps all
 710          * 2) Traffic class passed via ancillary data to sendmsdg(2)
 711          * 3) Traffic class socket option last
 712          */
 713         if (soisbackground(so)) {
 714                 mtc = MBUF_TC_BK;
 715         } else if (in_mtc != MBUF_TC_UNSPEC) {
 716                 if (in_mtc >= MBUF_TC_BE && in_mtc <= MBUF_TC_VO)
 717                         mtc = in_mtc;
 718         } else {
 719                 switch (so->so_traffic_class) {
 720                         case SO_TC_BE:
 721                                 mtc = MBUF_TC_BE;
 722                                 break;
 723                         case SO_TC_BK:
 724                                 mtc = MBUF_TC_BK;
 725                                 break;
 726                         case SO_TC_VI:
 727                                 mtc = MBUF_TC_VI;
 728                                 break;
 729                         case SO_TC_VO:
 730                                 mtc = MBUF_TC_VO;
 731                                 break;
 732                         default:
 733                                 break;
 734                 }
 735         }
 736
 737         /*
 738          * Set the traffic class in the mbuf packet header prio field
 739          */
 740         if ((sotcdb & SOTCDB_NO_MTC))
 741                 goto no_mbtc;
 742         m->m_pkthdr.prio = mtc;
 743
 744 no_mbtc:
 745         /*
 746          * Quick exit when best effort
 747          */
 748         if (mtc == MBUF_TC_BE)
 749                 goto no_dscp;
 750         /*
 751          * Now let set the DSCP code in IPv4 or IPv6 header
 752          * By default do this only for local traffic if a code is not already set
 753          */
 754         if ((sotcdb & SOTCDB_NO_DSCP))
 755                 goto no_dscp;
 756
 757         /*
 758          * Test if a IP TOS or IPV6 TCLASS has already been set on the socket or the raw packet
 759          */
 760         if ((sotcdb & SOTCDB_NO_DSCPTST) == 0) {
 761 #if INET6
 762                 if (isipv6)
 763                 {
 764                         if ((so->so_type == SOCK_RAW && (ip6->ip6_flow & htonl(0xff << 20)) != 0) ||
 765                             (inp->in6p_outputopts && inp->in6p_outputopts->ip6po_tclass != -1))
 766                                 goto no_dscp;
 767                 }
 768                 else
 769 #endif /* INET6 */
 770                 {
 771                         if ((so->so_type == SOCK_RAW && (inp->inp_flags & INP_HDRINCL)) ||
 772                                 inp->inp_ip_tos != 0)
 773                                 goto no_dscp;
 774                 }
 775         }
 776
 777         /*
 778          * Test if destination is local
 779          */
 780         if ((sotcdb & SOTCDB_NO_LCLTST) == 0) {
 781                 int islocal = 0;
 782                 struct route *ro = &inp->inp_route;
 783
 784                 if (so->so_type == SOCK_STREAM) {
 785                         struct tcpcb *tp = intotcpcb(inp);
 786
 787                         if ((tp->t_flags & TF_LOCAL))
 788                                 islocal = 1;
 789                 }
 790                 else
 791 #if INET6
 792                 if (isipv6)
 793                 {
 794                         if ((ro != NULL && ro->ro_rt != NULL &&
 795                                  (ro->ro_rt->rt_gateway->sa_family == AF_LINK ||
 796                                   (ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK))) ||
 797                                  in6addr_local(&ip6->ip6_dst))
 798                                 islocal = 1;
 799                 }
 800                 else
 801 #endif /* INET6 */
 802                 {
 803                         if ((ro != NULL && ro->ro_rt != NULL &&
 804                                  (ro->ro_rt->rt_gateway->sa_family == AF_LINK ||
 805                                   (ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK))) ||
 806                                  inaddr_local(ip->ip_dst))
 807                                 islocal = 1;
 808                 }
 809                 if (islocal == 0)
 810                         goto no_dscp;
 811         }
 812
 813 #if INET6
 814         if (isipv6)
 815                 ip6->ip6_flow |=
 816                         htonl(dscp_code_from_mbuf_tclass(m->m_pkthdr.prio) << 20);
 817         else
 818 #endif /* INET6 */
 819                 ip->ip_tos |= dscp_code_from_mbuf_tclass(m->m_pkthdr.prio) << 2;
 820
 821 no_dscp:
 822         /*
 823          * For TCP with background traffic class switch CC algo based on sysctl
 824          */
 825         if (so->so_type == SOCK_STREAM) {
 826                 set_tcp_stream_priority(so);
 827         }
 828
 829         /*
 830          * Assume socket and mbuf traffic class values are the same
 831          * Also assume the socket lock is held
 832          */
 833         so->so_tc_stats[mtc].txpackets += 1;
 834         so->so_tc_stats[mtc].txbytes += m->m_pkthdr.len;
 835
 836         return;
 837 }
 838
 839 __private_extern__ void
 840 socket_tclass_init(void)
 841 {
 842         tclass_lck_grp_attr = lck_grp_attr_alloc_init();
 843         tclass_lck_grp = lck_grp_alloc_init("tclass", tclass_lck_grp_attr);
 844         tclass_lck_attr = lck_attr_alloc_init();
 845         if ((tclass_lock = lck_mtx_alloc_init(tclass_lck_grp, tclass_lck_attr)) == NULL) {
 846                         panic("failed to allocate memory for tclass\n");
 847         }
 848 }
 849
 850