bsd/netinet/mp_pcb.c

   1 /*
   2  * Copyright (c) 2012-2016 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
  47 static lck_grp_t        *mp_lock_grp;
  48 static lck_attr_t       *mp_lock_attr;
  49 static lck_grp_attr_t   *mp_lock_grp_attr;
  50 decl_lck_mtx_data(static, mp_lock);             /* global MULTIPATH lock */
  51 decl_lck_mtx_data(static, mp_timeout_lock);
  52
  53 static TAILQ_HEAD(, mppcbinfo) mppi_head = TAILQ_HEAD_INITIALIZER(mppi_head);
  54
  55 static boolean_t mp_timeout_run;        /* MP timer is scheduled to run */
  56 static boolean_t mp_garbage_collecting;
  57 static boolean_t mp_ticking;
  58 static void mp_sched_timeout(void);
  59 static void mp_timeout(void *);
  60
  61 void
  62 mp_pcbinit(void)
  63 {
  64         static int mp_initialized = 0;
  65
  66         VERIFY(!mp_initialized);
  67         mp_initialized = 1;
  68
  69         mp_lock_grp_attr = lck_grp_attr_alloc_init();
  70         mp_lock_grp = lck_grp_alloc_init("multipath", mp_lock_grp_attr);
  71         mp_lock_attr = lck_attr_alloc_init();
  72         lck_mtx_init(&mp_lock, mp_lock_grp, mp_lock_attr);
  73         lck_mtx_init(&mp_timeout_lock, mp_lock_grp, mp_lock_attr);
  74 }
  75
  76 static void
  77 mp_timeout(void *arg)
  78 {
  79 #pragma unused(arg)
  80         struct mppcbinfo *mppi;
  81         boolean_t t, gc;
  82         uint32_t t_act = 0;
  83         uint32_t gc_act = 0;
  84
  85         /*
  86          * Update coarse-grained networking timestamp (in sec.); the idea
  87          * is to piggy-back on the timeout callout to update the counter
  88          * returnable via net_uptime().
  89          */
  90         net_update_uptime();
  91
  92         lck_mtx_lock_spin(&mp_timeout_lock);
  93         gc = mp_garbage_collecting;
  94         mp_garbage_collecting = FALSE;
  95
  96         t = mp_ticking;
  97         mp_ticking = FALSE;
  98
  99         if (gc || t) {
 100                 lck_mtx_unlock(&mp_timeout_lock);
 101
 102                 lck_mtx_lock(&mp_lock);
 103                 TAILQ_FOREACH(mppi, &mppi_head, mppi_entry) {
 104                         if ((gc && mppi->mppi_gc != NULL) ||
 105                             (t && mppi->mppi_timer != NULL)) {
 106                                 lck_mtx_lock(&mppi->mppi_lock);
 107                                 if (gc && mppi->mppi_gc != NULL)
 108                                         gc_act += mppi->mppi_gc(mppi);
 109                                 if (t && mppi->mppi_timer != NULL)
 110                                         t_act += mppi->mppi_timer(mppi);
 111                                 lck_mtx_unlock(&mppi->mppi_lock);
 112                         }
 113                 }
 114                 lck_mtx_unlock(&mp_lock);
 115
 116                 lck_mtx_lock_spin(&mp_timeout_lock);
 117         }
 118
 119         /* lock was dropped above, so check first before overriding */
 120         if (!mp_garbage_collecting)
 121                 mp_garbage_collecting = (gc_act != 0);
 122         if (!mp_ticking)
 123                 mp_ticking = (t_act != 0);
 124
 125         /* re-arm the timer if there's work to do */
 126         mp_timeout_run = FALSE;
 127         mp_sched_timeout();
 128         lck_mtx_unlock(&mp_timeout_lock);
 129 }
 130
 131 static void
 132 mp_sched_timeout(void)
 133 {
 134         lck_mtx_assert(&mp_timeout_lock, LCK_MTX_ASSERT_OWNED);
 135
 136         if (!mp_timeout_run && (mp_garbage_collecting || mp_ticking)) {
 137                 lck_mtx_convert_spin(&mp_timeout_lock);
 138                 mp_timeout_run = TRUE;
 139                 timeout(mp_timeout, NULL, hz);
 140         }
 141 }
 142
 143 void
 144 mp_gc_sched(void)
 145 {
 146         lck_mtx_lock_spin(&mp_timeout_lock);
 147         mp_garbage_collecting = TRUE;
 148         mp_sched_timeout();
 149         lck_mtx_unlock(&mp_timeout_lock);
 150 }
 151
 152 void
 153 mptcp_timer_sched(void)
 154 {
 155         lck_mtx_lock_spin(&mp_timeout_lock);
 156         mp_ticking = TRUE;
 157         mp_sched_timeout();
 158         lck_mtx_unlock(&mp_timeout_lock);
 159 }
 160
 161 void
 162 mp_pcbinfo_attach(struct mppcbinfo *mppi)
 163 {
 164         struct mppcbinfo *mppi0;
 165
 166         lck_mtx_lock(&mp_lock);
 167         TAILQ_FOREACH(mppi0, &mppi_head, mppi_entry) {
 168                 if (mppi0 == mppi) {
 169                         panic("%s: mppi %p already in the list\n",
 170                             __func__, mppi);
 171                         /* NOTREACHED */
 172                 }
 173         }
 174         TAILQ_INSERT_TAIL(&mppi_head, mppi, mppi_entry);
 175         lck_mtx_unlock(&mp_lock);
 176 }
 177
 178 int
 179 mp_pcbinfo_detach(struct mppcbinfo *mppi)
 180 {
 181         struct mppcbinfo *mppi0;
 182         int error = 0;
 183
 184         lck_mtx_lock(&mp_lock);
 185         TAILQ_FOREACH(mppi0, &mppi_head, mppi_entry) {
 186                 if (mppi0 == mppi)
 187                         break;
 188         }
 189         if (mppi0 != NULL)
 190                 TAILQ_REMOVE(&mppi_head, mppi0, mppi_entry);
 191         else
 192                 error = ENXIO;
 193         lck_mtx_unlock(&mp_lock);
 194
 195         return (error);
 196 }
 197
 198 int
 199 mp_pcballoc(struct socket *so, struct mppcbinfo *mppi)
 200 {
 201         struct mppcb *mpp = NULL;
 202
 203         VERIFY(sotomppcb(so) == NULL);
 204
 205         lck_mtx_lock(&mppi->mppi_lock);
 206         if (mppi->mppi_count >= mptcp_socket_limit) {
 207                 lck_mtx_unlock(&mppi->mppi_lock);
 208                 mptcplog((LOG_ERR, "MPTCP Socket: Reached MPTCP socket limit."),
 209                     MPTCP_SOCKET_DBG, MPTCP_LOGLVL_ERR);
 210                 /*
 211                  * This limit may be reached either because of
 212                  * a leak or a transient condition where
 213                  * MPTCP connections are not released fast
 214                  * enough.
 215                  * We return EAFNOSUPPORT here to have user
 216                  * space library fallback to TCP.
 217                  * XXX We need to revist this when we get rid
 218                  * of the current low limit imposed on MPTCP.
 219                  */
 220                 return (EAFNOSUPPORT);
 221         }
 222         lck_mtx_unlock(&mppi->mppi_lock);
 223
 224         mpp = zalloc(mppi->mppi_zone);
 225         if (mpp == NULL) {
 226                 return (ENOBUFS);
 227         }
 228
 229         bzero(mpp, mppi->mppi_size);
 230         lck_mtx_init(&mpp->mpp_lock, mppi->mppi_lock_grp, mppi->mppi_lock_attr);
 231         mpp->mpp_pcbinfo = mppi;
 232         mpp->mpp_state = MPPCB_STATE_INUSE;
 233         mpp->mpp_socket = so;
 234         so->so_pcb = mpp;
 235
 236         if (NULL == mppi->mppi_pcbe_create(so, mpp)) {
 237                 lck_mtx_destroy(&mpp->mpp_lock, mppi->mppi_lock_grp);
 238                 zfree(mppi->mppi_zone, mpp);
 239                 return (ENOBUFS);
 240         }
 241
 242         lck_mtx_lock(&mppi->mppi_lock);
 243         mpp->mpp_flags |= MPP_ATTACHED;
 244         TAILQ_INSERT_TAIL(&mppi->mppi_pcbs, mpp, mpp_entry);
 245         mppi->mppi_count++;
 246         lck_mtx_unlock(&mppi->mppi_lock);
 247
 248         return (0);
 249 }
 250
 251 void
 252 mp_pcbdetach(struct mppcb *mpp)
 253 {
 254         struct socket *so = mpp->mpp_socket;
 255
 256         VERIFY(so->so_pcb == mpp);
 257
 258         mpp->mpp_state = MPPCB_STATE_DEAD;
 259         if (!(so->so_flags & SOF_PCBCLEARING))
 260                 so->so_flags |= SOF_PCBCLEARING;
 261
 262         mp_gc_sched();
 263 }
 264
 265 void
 266 mp_pcbdispose(struct mppcb *mpp)
 267 {
 268         struct mppcbinfo *mppi = mpp->mpp_pcbinfo;
 269
 270         VERIFY(mppi != NULL);
 271
 272         lck_mtx_assert(&mppi->mppi_lock, LCK_MTX_ASSERT_OWNED);
 273         lck_mtx_assert(&mpp->mpp_lock, LCK_MTX_ASSERT_OWNED);
 274
 275         VERIFY(mpp->mpp_state == MPPCB_STATE_DEAD);
 276
 277         VERIFY(mpp->mpp_flags & MPP_ATTACHED);
 278         mpp->mpp_flags &= ~MPP_ATTACHED;
 279         TAILQ_REMOVE(&mppi->mppi_pcbs, mpp, mpp_entry);
 280         VERIFY(mppi->mppi_count != 0);
 281         mppi->mppi_count--;
 282
 283         VERIFY(mpp->mpp_socket != NULL);
 284         VERIFY(mpp->mpp_socket->so_usecount == 0);
 285         mpp->mpp_socket->so_pcb = NULL;
 286         mpp->mpp_socket = NULL;
 287
 288         lck_mtx_unlock(&mpp->mpp_lock);
 289         lck_mtx_destroy(&mpp->mpp_lock, mppi->mppi_lock_grp);
 290         zfree(mppi->mppi_zone, mpp);
 291 }