bsd/netinet/mp_pcb.c

   1 /*
   2  * Copyright (c) 2012-2017 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/param.h>
  30 #include <sys/systm.h>
  31 #include <sys/kernel.h>
  32 #include <sys/mbuf.h>
  33 #include <sys/mcache.h>
  34 #include <sys/syslog.h>
  35 #include <sys/socket.h>
  36 #include <sys/socketvar.h>
  37 #include <sys/protosw.h>
  38 #include <sys/proc_internal.h>
  39
  40 #include <mach/boolean.h>
  41 #include <kern/zalloc.h>
  42 #include <kern/locks.h>
  43
  44 #include <netinet/mp_pcb.h>
  45 #include <netinet/mptcp_var.h>
  46 #include <netinet6/in6_pcb.h>
  47
  48 static lck_grp_t        *mp_lock_grp;
  49 static lck_attr_t       *mp_lock_attr;
  50 static lck_grp_attr_t   *mp_lock_grp_attr;
  51 decl_lck_mtx_data(static, mp_lock);             /* global MULTIPATH lock */
  52 decl_lck_mtx_data(static, mp_timeout_lock);
  53
  54 static TAILQ_HEAD(, mppcbinfo) mppi_head = TAILQ_HEAD_INITIALIZER(mppi_head);
  55
  56 static boolean_t mp_timeout_run;        /* MP timer is scheduled to run */
  57 static boolean_t mp_garbage_collecting;
  58 static boolean_t mp_ticking;
  59 static void mp_sched_timeout(void);
  60 static void mp_timeout(void *);
  61
  62 void
  63 mp_pcbinit(void)
  64 {
  65         static int mp_initialized = 0;
  66
  67         VERIFY(!mp_initialized);
  68         mp_initialized = 1;
  69
  70         mp_lock_grp_attr = lck_grp_attr_alloc_init();
  71         mp_lock_grp = lck_grp_alloc_init("multipath", mp_lock_grp_attr);
  72         mp_lock_attr = lck_attr_alloc_init();
  73         lck_mtx_init(&mp_lock, mp_lock_grp, mp_lock_attr);
  74         lck_mtx_init(&mp_timeout_lock, mp_lock_grp, mp_lock_attr);
  75 }
  76
  77 static void
  78 mp_timeout(void *arg)
  79 {
  80 #pragma unused(arg)
  81         struct mppcbinfo *mppi;
  82         boolean_t t, gc;
  83         uint32_t t_act = 0;
  84         uint32_t gc_act = 0;
  85
  86         /*
  87          * Update coarse-grained networking timestamp (in sec.); the idea
  88          * is to piggy-back on the timeout callout to update the counter
  89          * returnable via net_uptime().
  90          */
  91         net_update_uptime();
  92
  93         lck_mtx_lock_spin(&mp_timeout_lock);
  94         gc = mp_garbage_collecting;
  95         mp_garbage_collecting = FALSE;
  96
  97         t = mp_ticking;
  98         mp_ticking = FALSE;
  99
 100         if (gc || t) {
 101                 lck_mtx_unlock(&mp_timeout_lock);
 102
 103                 lck_mtx_lock(&mp_lock);
 104                 TAILQ_FOREACH(mppi, &mppi_head, mppi_entry) {
 105                         if ((gc && mppi->mppi_gc != NULL) ||
 106                             (t && mppi->mppi_timer != NULL)) {
 107                                 lck_mtx_lock(&mppi->mppi_lock);
 108                                 if (gc && mppi->mppi_gc != NULL) {
 109                                         gc_act += mppi->mppi_gc(mppi);
 110                                 }
 111                                 if (t && mppi->mppi_timer != NULL) {
 112                                         t_act += mppi->mppi_timer(mppi);
 113                                 }
 114                                 lck_mtx_unlock(&mppi->mppi_lock);
 115                         }
 116                 }
 117                 lck_mtx_unlock(&mp_lock);
 118
 119                 lck_mtx_lock_spin(&mp_timeout_lock);
 120         }
 121
 122         /* lock was dropped above, so check first before overriding */
 123         if (!mp_garbage_collecting) {
 124                 mp_garbage_collecting = (gc_act != 0);
 125         }
 126         if (!mp_ticking) {
 127                 mp_ticking = (t_act != 0);
 128         }
 129
 130         /* re-arm the timer if there's work to do */
 131         mp_timeout_run = FALSE;
 132         mp_sched_timeout();
 133         lck_mtx_unlock(&mp_timeout_lock);
 134 }
 135
 136 static void
 137 mp_sched_timeout(void)
 138 {
 139         LCK_MTX_ASSERT(&mp_timeout_lock, LCK_MTX_ASSERT_OWNED);
 140
 141         if (!mp_timeout_run && (mp_garbage_collecting || mp_ticking)) {
 142                 lck_mtx_convert_spin(&mp_timeout_lock);
 143                 mp_timeout_run = TRUE;
 144                 timeout(mp_timeout, NULL, hz);
 145         }
 146 }
 147
 148 void
 149 mp_gc_sched(void)
 150 {
 151         lck_mtx_lock_spin(&mp_timeout_lock);
 152         mp_garbage_collecting = TRUE;
 153         mp_sched_timeout();
 154         lck_mtx_unlock(&mp_timeout_lock);
 155 }
 156
 157 void
 158 mptcp_timer_sched(void)
 159 {
 160         lck_mtx_lock_spin(&mp_timeout_lock);
 161         mp_ticking = TRUE;
 162         mp_sched_timeout();
 163         lck_mtx_unlock(&mp_timeout_lock);
 164 }
 165
 166 void
 167 mp_pcbinfo_attach(struct mppcbinfo *mppi)
 168 {
 169         struct mppcbinfo *mppi0;
 170
 171         lck_mtx_lock(&mp_lock);
 172         TAILQ_FOREACH(mppi0, &mppi_head, mppi_entry) {
 173                 if (mppi0 == mppi) {
 174                         panic("%s: mppi %p already in the list\n",
 175                             __func__, mppi);
 176                         /* NOTREACHED */
 177                 }
 178         }
 179         TAILQ_INSERT_TAIL(&mppi_head, mppi, mppi_entry);
 180         lck_mtx_unlock(&mp_lock);
 181 }
 182
 183 int
 184 mp_pcbinfo_detach(struct mppcbinfo *mppi)
 185 {
 186         struct mppcbinfo *mppi0;
 187         int error = 0;
 188
 189         lck_mtx_lock(&mp_lock);
 190         TAILQ_FOREACH(mppi0, &mppi_head, mppi_entry) {
 191                 if (mppi0 == mppi) {
 192                         break;
 193                 }
 194         }
 195         if (mppi0 != NULL) {
 196                 TAILQ_REMOVE(&mppi_head, mppi0, mppi_entry);
 197         } else {
 198                 error = ENXIO;
 199         }
 200         lck_mtx_unlock(&mp_lock);
 201
 202         return error;
 203 }
 204
 205 int
 206 mp_pcballoc(struct socket *so, struct mppcbinfo *mppi)
 207 {
 208         struct mppcb *mpp = NULL;
 209         int error;
 210
 211         VERIFY(mpsotomppcb(so) == NULL);
 212
 213         mpp = zalloc(mppi->mppi_zone);
 214         if (mpp == NULL) {
 215                 return ENOBUFS;
 216         }
 217
 218         bzero(mpp, mppi->mppi_size);
 219         lck_mtx_init(&mpp->mpp_lock, mppi->mppi_lock_grp, mppi->mppi_lock_attr);
 220         mpp->mpp_pcbinfo = mppi;
 221         mpp->mpp_state = MPPCB_STATE_INUSE;
 222         mpp->mpp_socket = so;
 223         so->so_pcb = mpp;
 224
 225         error = mptcp_sescreate(mpp);
 226         if (error) {
 227                 lck_mtx_destroy(&mpp->mpp_lock, mppi->mppi_lock_grp);
 228                 zfree(mppi->mppi_zone, mpp);
 229                 return error;
 230         }
 231
 232         lck_mtx_lock(&mppi->mppi_lock);
 233         mpp->mpp_flags |= MPP_ATTACHED;
 234         TAILQ_INSERT_TAIL(&mppi->mppi_pcbs, mpp, mpp_entry);
 235         mppi->mppi_count++;
 236         lck_mtx_unlock(&mppi->mppi_lock);
 237
 238         return 0;
 239 }
 240
 241 void
 242 mp_pcbdetach(struct socket *mp_so)
 243 {
 244         struct mppcb *mpp = mpsotomppcb(mp_so);
 245
 246         mpp->mpp_state = MPPCB_STATE_DEAD;
 247         if (!(mp_so->so_flags & SOF_PCBCLEARING)) {
 248                 mp_so->so_flags |= SOF_PCBCLEARING;
 249         }
 250
 251         mp_gc_sched();
 252 }
 253
 254 void
 255 mp_pcbdispose(struct mppcb *mpp)
 256 {
 257         struct mppcbinfo *mppi = mpp->mpp_pcbinfo;
 258
 259         VERIFY(mppi != NULL);
 260
 261         LCK_MTX_ASSERT(&mppi->mppi_lock, LCK_MTX_ASSERT_OWNED);
 262         mpp_lock_assert_held(mpp);
 263
 264         VERIFY(mpp->mpp_state == MPPCB_STATE_DEAD);
 265         VERIFY(mpp->mpp_flags & MPP_ATTACHED);
 266
 267         mpp->mpp_flags &= ~MPP_ATTACHED;
 268         TAILQ_REMOVE(&mppi->mppi_pcbs, mpp, mpp_entry);
 269         VERIFY(mppi->mppi_count != 0);
 270         mppi->mppi_count--;
 271
 272         mpp_unlock(mpp);
 273
 274 #if NECP
 275         necp_mppcb_dispose(mpp);
 276 #endif /* NECP */
 277
 278         lck_mtx_destroy(&mpp->mpp_lock, mppi->mppi_lock_grp);
 279
 280         VERIFY(mpp->mpp_socket != NULL);
 281         VERIFY(mpp->mpp_socket->so_usecount == 0);
 282         mpp->mpp_socket->so_pcb = NULL;
 283         mpp->mpp_socket = NULL;
 284
 285         zfree(mppi->mppi_zone, mpp);
 286 }
 287
 288 static int
 289 mp_getaddr_v4(struct socket *mp_so, struct sockaddr **nam, boolean_t peer)
 290 {
 291         struct mptses *mpte = mpsotompte(mp_so);
 292         struct sockaddr_in *sin;
 293
 294         /*
 295          * Do the malloc first in case it blocks.
 296          */
 297         MALLOC(sin, struct sockaddr_in *, sizeof(*sin), M_SONAME, M_WAITOK);
 298         if (sin == NULL) {
 299                 return ENOBUFS;
 300         }
 301         bzero(sin, sizeof(*sin));
 302         sin->sin_family = AF_INET;
 303         sin->sin_len = sizeof(*sin);
 304
 305         if (!peer) {
 306                 sin->sin_port = mpte->__mpte_src_v4.sin_port;
 307                 sin->sin_addr = mpte->__mpte_src_v4.sin_addr;
 308         } else {
 309                 sin->sin_port = mpte->__mpte_dst_v4.sin_port;
 310                 sin->sin_addr = mpte->__mpte_dst_v4.sin_addr;
 311         }
 312
 313         *nam = (struct sockaddr *)sin;
 314         return 0;
 315 }
 316
 317 static int
 318 mp_getaddr_v6(struct socket *mp_so, struct sockaddr **nam, boolean_t peer)
 319 {
 320         struct mptses *mpte = mpsotompte(mp_so);
 321         struct in6_addr addr;
 322         in_port_t port;
 323
 324         if (!peer) {
 325                 port = mpte->__mpte_src_v6.sin6_port;
 326                 addr = mpte->__mpte_src_v6.sin6_addr;
 327         } else {
 328                 port = mpte->__mpte_dst_v6.sin6_port;
 329                 addr = mpte->__mpte_dst_v6.sin6_addr;
 330         }
 331
 332         *nam = in6_sockaddr(port, &addr);
 333         if (*nam == NULL) {
 334                 return ENOBUFS;
 335         }
 336
 337         return 0;
 338 }
 339
 340 int
 341 mp_getsockaddr(struct socket *mp_so, struct sockaddr **nam)
 342 {
 343         struct mptses *mpte = mpsotompte(mp_so);
 344
 345         if (mpte->mpte_src.sa_family == AF_INET || mpte->mpte_src.sa_family == 0) {
 346                 return mp_getaddr_v4(mp_so, nam, false);
 347         } else if (mpte->mpte_src.sa_family == AF_INET6) {
 348                 return mp_getaddr_v6(mp_so, nam, false);
 349         } else {
 350                 return EINVAL;
 351         }
 352 }
 353
 354 int
 355 mp_getpeeraddr(struct socket *mp_so, struct sockaddr **nam)
 356 {
 357         struct mptses *mpte = mpsotompte(mp_so);
 358
 359         if (mpte->mpte_src.sa_family == AF_INET || mpte->mpte_src.sa_family == 0) {
 360                 return mp_getaddr_v4(mp_so, nam, true);
 361         } else if (mpte->mpte_src.sa_family == AF_INET6) {
 362                 return mp_getaddr_v6(mp_so, nam, true);
 363         } else {
 364                 return EINVAL;
 365         }
 366 }