]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/ipc/ipc_mqueue.c
xnu-3248.60.10.tar.gz
[apple/xnu.git] / osfmk / ipc / ipc_mqueue.c
index d00ebe34f6f82c45055423a4c47d3070e485dc9a..1b5d82c194477545b0ee3ac6adb8a35c31261569 100644 (file)
@@ -1,16 +1,19 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2007 Apple Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
- * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
- * compliance with the License. Please obtain a copy of the License at
- * http://www.opensource.apple.com/apsl/ and read it before using this
- * file.
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
  * 
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
@@ -20,7 +23,7 @@
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @OSF_FREE_COPYRIGHT@
  *
  *     Functions to manipulate IPC message queues.
  */
+/*
+ * NOTICE: This file was modified by SPARTA, Inc. in 2006 to introduce
+ * support for mandatory and extensible security protections.  This notice
+ * is included in support of clause 2.2 (b) of the Apple Public License,
+ * Version 2.0.
+ */
+    
 
 #include <mach/port.h>
 #include <mach/message.h>
 #include <kern/counters.h>
 #include <kern/sched_prim.h>
 #include <kern/ipc_kobject.h>
+#include <kern/ipc_mig.h>      /* XXX - for mach_msg_receive_continue */
 #include <kern/misc_protos.h>
 #include <kern/task.h>
 #include <kern/thread.h>
-#include <kern/wait_queue.h>
+#include <kern/waitq.h>
 
 #include <ipc/ipc_mqueue.h>
 #include <ipc/ipc_kmsg.h>
 #include <ipc/ipc_pset.h>
 #include <ipc/ipc_space.h>
 
-#include <ddb/tr.h>
+#ifdef __LP64__
+#include <vm/vm_map.h>
+#endif
 
 int ipc_mqueue_full;           /* address is event for queue space */
 int ipc_mqueue_rcv;            /* address is event for message arrival */
 
-#define TR_ENABLE 0
+/* forward declarations */
+void ipc_mqueue_receive_results(wait_result_t result);
 
 /*
  *     Routine:        ipc_mqueue_init
@@ -94,12 +108,15 @@ int ipc_mqueue_rcv;                /* address is event for message arrival */
 void
 ipc_mqueue_init(
        ipc_mqueue_t    mqueue,
-       boolean_t       is_set)
+       boolean_t       is_set,
+       uint64_t        *reserved_link)
 {
        if (is_set) {
-               wait_queue_set_init(&mqueue->imq_set_queue, SYNC_POLICY_FIFO);
+               waitq_set_init(&mqueue->imq_set_queue,
+                              SYNC_POLICY_FIFO|SYNC_POLICY_PREPOST|SYNC_POLICY_DISABLE_IRQ,
+                              reserved_link);
        } else {
-               wait_queue_init(&mqueue->imq_wait_queue, SYNC_POLICY_FIFO);
+               waitq_init(&mqueue->imq_wait_queue, SYNC_POLICY_FIFO|SYNC_POLICY_DISABLE_IRQ);
                ipc_kmsg_queue_init(&mqueue->imq_messages);
                mqueue->imq_seqno = 0;
                mqueue->imq_msgcount = 0;
@@ -108,6 +125,53 @@ ipc_mqueue_init(
        }
 }
 
+void ipc_mqueue_deinit(
+       ipc_mqueue_t            mqueue)
+{
+       boolean_t is_set = imq_is_set(mqueue);
+
+       if (is_set)
+               waitq_set_deinit(&mqueue->imq_set_queue);
+       else
+               waitq_deinit(&mqueue->imq_wait_queue);
+}
+
+/*
+ *     Routine:        imq_reserve_and_lock
+ *     Purpose:
+ *             Atomically lock an ipc_mqueue_t object and reserve
+ *             an appropriate number of prepost linkage objects for
+ *             use in wakeup operations.
+ *     Conditions:
+ *             mq is unlocked
+ */
+void
+imq_reserve_and_lock(ipc_mqueue_t mq, uint64_t *reserved_prepost, spl_t *spl)
+{
+       *reserved_prepost = waitq_prepost_reserve(&mq->imq_wait_queue, 0,
+                                                 WAITQ_KEEP_LOCKED, spl);
+
+}
+
+
+/*
+ *     Routine:        imq_release_and_unlock
+ *     Purpose:
+ *             Unlock an ipc_mqueue_t object, re-enable interrupts,
+ *             and release any unused prepost object reservations.
+ *     Conditions:
+ *             mq is locked
+ */
+void
+imq_release_and_unlock(ipc_mqueue_t mq, uint64_t reserved_prepost, spl_t spl)
+{
+       assert(imq_held(mq));
+       waitq_unlock(&mq->imq_wait_queue);
+       splx(spl);
+       waitq_prepost_release_reserve(reserved_prepost);
+}
+
+
 /*
  *     Routine:        ipc_mqueue_member
  *     Purpose:
@@ -122,13 +186,13 @@ ipc_mqueue_init(
 
 boolean_t
 ipc_mqueue_member(
-       ipc_mqueue_t    port_mqueue,
-       ipc_mqueue_t    set_mqueue)
+       ipc_mqueue_t            port_mqueue,
+       ipc_mqueue_t            set_mqueue)
 {
-       wait_queue_t    port_waitq = &port_mqueue->imq_wait_queue;
-       wait_queue_t    set_waitq = &set_mqueue->imq_wait_queue;
+       struct waitq *port_waitq = &port_mqueue->imq_wait_queue;
+       struct waitq_set *set_waitq = &set_mqueue->imq_set_queue;
 
-       return (wait_queue_member(port_waitq, set_waitq));
+       return waitq_member(port_waitq, set_waitq);
 
 }
 
@@ -141,13 +205,13 @@ ipc_mqueue_member(
 
 kern_return_t
 ipc_mqueue_remove(
-       ipc_mqueue_t     mqueue,
-       ipc_mqueue_t     set_mqueue)
+       ipc_mqueue_t      mqueue,
+       ipc_mqueue_t      set_mqueue)
 {
-       wait_queue_t     mq_waitq = &mqueue->imq_wait_queue;
-       wait_queue_set_t set_waitq = &set_mqueue->imq_set_queue;
+       struct waitq *mq_waitq = &mqueue->imq_wait_queue;
+       struct waitq_set *set_waitq = &set_mqueue->imq_set_queue;
 
-       return wait_queue_unlink(mq_waitq, set_waitq);
+       return waitq_unlink(mq_waitq, set_waitq);
 }
 
 /*
@@ -158,12 +222,11 @@ ipc_mqueue_remove(
  *             Nothing locked.
  */
 void
-ipc_mqueue_remove_from_all(
-       ipc_mqueue_t    mqueue)
+ipc_mqueue_remove_from_all(ipc_mqueue_t        mqueue)
 {
-       wait_queue_t    mq_waitq = &mqueue->imq_wait_queue;
+       struct waitq *mq_waitq = &mqueue->imq_wait_queue;
 
-       wait_queue_unlink_all(mq_waitq);
+       waitq_unlink_all(mq_waitq);
        return;
 }
 
@@ -171,16 +234,15 @@ ipc_mqueue_remove_from_all(
  *     Routine:        ipc_mqueue_remove_all
  *     Purpose:
  *             Remove all the member queues from the specified set.
+ *             Also removes the queue from any containing sets.
  *     Conditions:
  *             Nothing locked.
  */
 void
-ipc_mqueue_remove_all(
-       ipc_mqueue_t    mqueue)
+ipc_mqueue_remove_all(ipc_mqueue_t     mqueue)
 {
-       wait_queue_set_t        mq_setq = &mqueue->imq_set_queue;
-
-       wait_queue_set_unlink_all(mq_setq);
+       struct waitq_set *mq_setq = &mqueue->imq_set_queue;
+       waitq_set_unlink_all(mq_setq);
        return;
 }
 
@@ -198,27 +260,39 @@ ipc_mqueue_remove_all(
  */
 kern_return_t
 ipc_mqueue_add(
-       ipc_mqueue_t     port_mqueue,
-       ipc_mqueue_t     set_mqueue)
+       ipc_mqueue_t    port_mqueue,
+       ipc_mqueue_t    set_mqueue,
+       uint64_t        *reserved_link,
+       uint64_t        *reserved_prepost)
 {
-       wait_queue_t     port_waitq = &port_mqueue->imq_wait_queue;
-       wait_queue_set_t set_waitq = &set_mqueue->imq_set_queue;
+       struct waitq     *port_waitq = &port_mqueue->imq_wait_queue;
+       struct waitq_set *set_waitq = &set_mqueue->imq_set_queue;
        ipc_kmsg_queue_t kmsgq;
        ipc_kmsg_t       kmsg, next;
        kern_return_t    kr;
        spl_t            s;
 
-       kr = wait_queue_link(port_waitq, set_waitq);
-       if (kr != KERN_SUCCESS)
+       assert(reserved_link && *reserved_link != 0);
+
+       s = splsched();
+       imq_lock(port_mqueue);
+
+       /*
+        * The link operation is now under the same lock-hold as
+        * message iteration and thread wakeup, but doesn't have to be...
+        */
+       kr = waitq_link(port_waitq, set_waitq, WAITQ_ALREADY_LOCKED, reserved_link);
+       if (kr != KERN_SUCCESS) {
+               imq_unlock(port_mqueue);
+               splx(s);
                return kr;
+       }
 
        /*
         * Now that the set has been added to the port, there may be
         * messages queued on the port and threads waiting on the set
         * waitq.  Lets get them together.
         */
-       s = splsched();
-       imq_lock(port_mqueue);
        kmsgq = &port_mqueue->imq_messages;
        for (kmsg = ipc_kmsg_queue_first(kmsgq);
             kmsg != IKM_NULL;
@@ -227,17 +301,31 @@ ipc_mqueue_add(
 
                for (;;) {
                        thread_t th;
+                       mach_msg_size_t msize;
+                       spl_t th_spl;
 
-                       th = wait_queue_wakeup64_identity_locked(
+                       th = waitq_wakeup64_identity_locked(
                                                port_waitq,
                                                IPC_MQUEUE_RECEIVE,
-                                               THREAD_AWAKENED,
-                                               FALSE);
+                                               THREAD_AWAKENED, &th_spl,
+                                               reserved_prepost, WAITQ_KEEP_LOCKED);
                        /* waitq/mqueue still locked, thread locked */
 
                        if (th == THREAD_NULL)
                                goto leave;
 
+                       /*
+                        * If the receiver waited with a facility not directly
+                        * related to Mach messaging, then it isn't prepared to get
+                        * handed the message directly.  Just set it running, and
+                        * go look for another thread that can.
+                        */
+                       if (th->ith_state != MACH_RCV_IN_PROGRESS) {
+                                 thread_unlock(th);
+                                 splx(th_spl);
+                                 continue;
+                       }
+
                        /*
                         * Found a receiver. see if they can handle the message
                         * correctly (the message is not too large for them, or
@@ -246,18 +334,20 @@ ipc_mqueue_add(
                         * the list and let them go back and figure it out and
                         * just move onto the next.
                         */
+                       msize = ipc_kmsg_copyout_size(kmsg, th->map);
                        if (th->ith_msize <
-                           kmsg->ikm_header.msgh_size +
-                           REQUESTED_TRAILER_SIZE(th->ith_option)) {
+                                       (msize + REQUESTED_TRAILER_SIZE(thread_is_64bit(th), th->ith_option))) {
                                th->ith_state = MACH_RCV_TOO_LARGE;
-                               th->ith_msize = kmsg->ikm_header.msgh_size;
+                               th->ith_msize = msize;
                                if (th->ith_option & MACH_RCV_LARGE) {
                                        /*
                                         * let him go without message
                                         */
+                                       th->ith_receiver_name = port_mqueue->imq_receiver_name;
                                        th->ith_kmsg = IKM_NULL;
                                        th->ith_seqno = 0;
                                        thread_unlock(th);
+                                       splx(th_spl);
                                        continue; /* find another thread */
                                }
                        } else {
@@ -268,14 +358,15 @@ ipc_mqueue_add(
                         * This thread is going to take this message,
                         * so give it to him.
                         */
-                       ipc_mqueue_release_msgcount(port_mqueue);
                        ipc_kmsg_rmqueue(kmsgq, kmsg);
+                       ipc_mqueue_release_msgcount(port_mqueue, IMQ_NULL);
+
                        th->ith_kmsg = kmsg;
                        th->ith_seqno = port_mqueue->imq_seqno++;
                        thread_unlock(th);
+                       splx(th_spl);
                        break;  /* go to next message */
                }
-                       
        }
  leave:
        imq_unlock(port_mqueue);
@@ -295,11 +386,12 @@ void
 ipc_mqueue_changed(
        ipc_mqueue_t            mqueue)
 {
-       wait_queue_wakeup64_all_locked(
-                               &mqueue->imq_wait_queue,
-                               IPC_MQUEUE_RECEIVE,
-                               THREAD_RESTART,
-                               FALSE);         /* unlock waitq? */
+       waitq_wakeup64_all_locked(&mqueue->imq_wait_queue,
+                                 IPC_MQUEUE_RECEIVE,
+                                 THREAD_RESTART,
+                                 NULL,
+                                 WAITQ_ALL_PRIORITIES,
+                                 WAITQ_KEEP_LOCKED);
 }
 
 
@@ -316,7 +408,7 @@ ipc_mqueue_changed(
  *             the message and must do something with it.  If successful,
  *             the message is queued, given to a receiver, or destroyed.
  *     Conditions:
- *             Nothing locked.
+ *             mqueue is locked.
  *     Returns:
  *             MACH_MSG_SUCCESS        The message was accepted.
  *             MACH_SEND_TIMED_OUT     Caller still has message.
@@ -325,12 +417,12 @@ ipc_mqueue_changed(
 mach_msg_return_t
 ipc_mqueue_send(
        ipc_mqueue_t            mqueue,
-       ipc_kmsg_t                      kmsg,
+       ipc_kmsg_t              kmsg,
        mach_msg_option_t       option,
-       mach_msg_timeout_t      timeout)
+       mach_msg_timeout_t      send_timeout,
+       spl_t                   s)
 {
        int wresult;
-       spl_t s;
 
        /*
         *  Don't block if:
@@ -338,60 +430,74 @@ ipc_mqueue_send(
         *      2) Caller used the MACH_SEND_ALWAYS internal option.
         *      3) Message is sent to a send-once right.
         */
-       s = splsched();
-       imq_lock(mqueue);
-
        if (!imq_full(mqueue) ||
-               (option & MACH_SEND_ALWAYS) ||
-               (MACH_MSGH_BITS_REMOTE(kmsg->ikm_header.msgh_bits) ==
-                MACH_MSG_TYPE_PORT_SEND_ONCE)) {
+           (!imq_full_kernel(mqueue) && 
+            ((option & MACH_SEND_ALWAYS) ||
+             (MACH_MSGH_BITS_REMOTE(kmsg->ikm_header->msgh_bits) ==
+              MACH_MSG_TYPE_PORT_SEND_ONCE)))) {
                mqueue->imq_msgcount++;
+               assert(mqueue->imq_msgcount > 0);
                imq_unlock(mqueue);
                splx(s);
        } else {
+               thread_t cur_thread = current_thread();
+               uint64_t deadline;
 
                /* 
                 * We have to wait for space to be granted to us.
                 */
-               if ((option & MACH_SEND_TIMEOUT) && (timeout == 0)) {
+               if ((option & MACH_SEND_TIMEOUT) && (send_timeout == 0)) {
                        imq_unlock(mqueue);
                        splx(s);
                        return MACH_SEND_TIMED_OUT;
                }
+               if (imq_full_kernel(mqueue)) {
+                       imq_unlock(mqueue);
+                       splx(s);
+                       return MACH_SEND_NO_BUFFER;
+               }
                mqueue->imq_fullwaiters = TRUE;
-               wresult = wait_queue_assert_wait64_locked(
+               thread_lock(cur_thread);
+               if (option & MACH_SEND_TIMEOUT)
+                       clock_interval_to_deadline(send_timeout, 1000*NSEC_PER_USEC, &deadline);
+               else
+                       deadline = 0;
+               wresult = waitq_assert_wait64_locked(
                                                &mqueue->imq_wait_queue,
                                                IPC_MQUEUE_FULL,
                                                THREAD_ABORTSAFE,
-                                               TRUE); /* unlock? */
-               /* wait/mqueue is unlocked */
+                                               TIMEOUT_URGENCY_USER_NORMAL,
+                                               deadline, TIMEOUT_NO_LEEWAY,
+                                               cur_thread);
+               thread_unlock(cur_thread);
+               imq_unlock(mqueue);
                splx(s);
                
                if (wresult == THREAD_WAITING) {
-                       if (option & MACH_SEND_TIMEOUT) {
-                               thread_set_timer(timeout, 1000*NSEC_PER_USEC);
-                               wresult = thread_block(THREAD_CONTINUE_NULL);
-                               if (wresult != THREAD_TIMED_OUT)
-                                       thread_cancel_timer();
-                       } else {
-                               wresult = thread_block(THREAD_CONTINUE_NULL);
-                       }
+                       wresult = thread_block(THREAD_CONTINUE_NULL);
                        counter(c_ipc_mqueue_send_block++);
                }
                
                switch (wresult) {
+
+               case THREAD_AWAKENED:
+                       /* 
+                        * we can proceed - inherited msgcount from waker
+                        * or the message queue has been destroyed and the msgcount
+                        * has been reset to zero (will detect in ipc_mqueue_post()).
+                        */
+                       break;
+                       
                case THREAD_TIMED_OUT:
                        assert(option & MACH_SEND_TIMEOUT);
                        return MACH_SEND_TIMED_OUT;
                        
-               case THREAD_AWAKENED:
-                       /* we can proceed - inherited msgcount from waker */
-                       break;
-                       
                case THREAD_INTERRUPTED:
                        return MACH_SEND_INTERRUPTED;
                        
                case THREAD_RESTART:
+                       /* mqueue is being destroyed */
+                       return MACH_SEND_INVALID_DEST;
                default:
                        panic("ipc_mqueue_send");
                }
@@ -401,6 +507,7 @@ ipc_mqueue_send(
        return MACH_MSG_SUCCESS;
 }
 
+
 /*
  *     Routine:        ipc_mqueue_release_msgcount
  *     Purpose:
@@ -408,27 +515,45 @@ ipc_mqueue_send(
  *             found a waiter.
  *
  *     Conditions:
- *             The message queue is locked
+ *             The message queue is locked.
+ *             The message corresponding to this reference is off the queue.
+ *             There is no need to pass reserved preposts because this will
+ *             never prepost to anyone
  */
 void
-ipc_mqueue_release_msgcount(
-       ipc_mqueue_t mqueue)    
+ipc_mqueue_release_msgcount(ipc_mqueue_t port_mq, ipc_mqueue_t set_mq)
 {
-       assert(imq_held(mqueue));
-       assert(mqueue->imq_msgcount > 0);
+       (void)set_mq;
+       assert(imq_held(port_mq));
+       assert(port_mq->imq_msgcount > 1 || ipc_kmsg_queue_empty(&port_mq->imq_messages));
 
-       mqueue->imq_msgcount--;
-       if (!imq_full(mqueue) && mqueue->imq_fullwaiters) {
-               if (wait_queue_wakeup64_one_locked(
-                                               &mqueue->imq_wait_queue,
-                                               IPC_MQUEUE_FULL,
-                                               THREAD_AWAKENED,
-                                               FALSE) != KERN_SUCCESS) {
-                       mqueue->imq_fullwaiters = FALSE;
+       port_mq->imq_msgcount--;
+
+       if (!imq_full(port_mq) && port_mq->imq_fullwaiters) {
+               /*
+                * boost the priority of the awoken thread
+                * (WAITQ_PROMOTE_PRIORITY) to ensure it uses
+                * the message queue slot we've just reserved.
+                *
+                * NOTE: this will never prepost
+                */
+               if (waitq_wakeup64_one_locked(&port_mq->imq_wait_queue,
+                                             IPC_MQUEUE_FULL,
+                                             THREAD_AWAKENED,
+                                             NULL,
+                                             WAITQ_PROMOTE_PRIORITY,
+                                             WAITQ_KEEP_LOCKED) != KERN_SUCCESS) {
+                       port_mq->imq_fullwaiters = FALSE;
                } else {
-                       mqueue->imq_msgcount++;  /* gave it away */
+                       /* gave away our slot - add reference back */
+                       port_mq->imq_msgcount++;
                }
        }
+
+       if (ipc_kmsg_queue_empty(&port_mq->imq_messages)) {
+               /* no more msgs: invalidate the port's prepost object */
+               waitq_clear_prepost_locked(&port_mq->imq_wait_queue, NULL);
+       }
 }
 
 /*
@@ -439,6 +564,7 @@ ipc_mqueue_release_msgcount(
  *             the message queue.
  *
  *     Conditions:
+ *             mqueue is unlocked
  *             If we need to queue, our space in the message queue is reserved.
  */
 void
@@ -446,8 +572,8 @@ ipc_mqueue_post(
        register ipc_mqueue_t   mqueue,
        register ipc_kmsg_t             kmsg)
 {
-
        spl_t s;
+       uint64_t reserved_prepost = 0;
 
        /*
         *      While the msg queue     is locked, we have control of the
@@ -455,37 +581,65 @@ ipc_mqueue_post(
         *
         *      Check for a receiver for the message.
         */
-       s = splsched();
-       imq_lock(mqueue);
+       imq_reserve_and_lock(mqueue, &reserved_prepost, &s);
        for (;;) {
-               wait_queue_t waitq = &mqueue->imq_wait_queue;
+               struct waitq *waitq = &mqueue->imq_wait_queue;
+               spl_t th_spl;
                thread_t receiver;
-
-               receiver = wait_queue_wakeup64_identity_locked(
-                                                       waitq,
-                                                       IPC_MQUEUE_RECEIVE,
-                                                       THREAD_AWAKENED,
-                                                       FALSE);
+               mach_msg_size_t msize;
+
+               receiver = waitq_wakeup64_identity_locked(waitq,
+                                                         IPC_MQUEUE_RECEIVE,
+                                                         THREAD_AWAKENED,
+                                                         &th_spl,
+                                                         &reserved_prepost,
+                                                         WAITQ_KEEP_LOCKED);
                /* waitq still locked, thread locked */
 
                if (receiver == THREAD_NULL) {
+                       
                        /* 
-                        * no receivers; queue kmsg
+                        * no receivers; queue kmsg if space still reserved.
                         */
-                       assert(mqueue->imq_msgcount > 0);
-                       ipc_kmsg_enqueue_macro(&mqueue->imq_messages, kmsg);
-                       break;
+                       if (mqueue->imq_msgcount > 0) {
+                               ipc_kmsg_enqueue_macro(&mqueue->imq_messages, kmsg);
+                               break;
+                       }
+
+                       /*
+                        * Otherwise, the message queue must belong to an inactive
+                        * port, so just destroy the message and pretend it was posted.
+                        */
+                       /* clear the waitq boost we may have been given */
+                       waitq_clear_promotion_locked(waitq, current_thread());
+                       imq_release_and_unlock(mqueue, reserved_prepost, s);
+                       ipc_kmsg_destroy(kmsg);
+                       current_task()->messages_sent++;
+                       return;
                }
-               
+       
+               /*
+                * If the receiver waited with a facility not directly
+                * related to Mach messaging, then it isn't prepared to get
+                * handed the message directly.  Just set it running, and
+                * go look for another thread that can.
+                */
+               if (receiver->ith_state != MACH_RCV_IN_PROGRESS) {
+                                 thread_unlock(receiver);
+                                 splx(th_spl);
+                                 continue;
+               }
+
+       
                /*
                 * We found a waiting thread.
                 * If the message is too large or the scatter list is too small
                 * the thread we wake up will get that as its status.
                 */
+               msize = ipc_kmsg_copyout_size(kmsg, receiver->map);
                if (receiver->ith_msize <
-                   (kmsg->ikm_header.msgh_size) +
-                   REQUESTED_TRAILER_SIZE(receiver->ith_option)) {
-                       receiver->ith_msize = kmsg->ikm_header.msgh_size;
+                               (msize + REQUESTED_TRAILER_SIZE(thread_is_64bit(receiver), receiver->ith_option))) {
+                       receiver->ith_msize = msize;
                        receiver->ith_state = MACH_RCV_TOO_LARGE;
                } else {
                        receiver->ith_state = MACH_MSG_SUCCESS;
@@ -502,9 +656,10 @@ ipc_mqueue_post(
                        receiver->ith_kmsg = kmsg;
                        receiver->ith_seqno = mqueue->imq_seqno++;
                        thread_unlock(receiver);
+                       splx(th_spl);
 
                        /* we didn't need our reserved spot in the queue */
-                       ipc_mqueue_release_msgcount(mqueue);
+                       ipc_mqueue_release_msgcount(mqueue, IMQ_NULL);
                        break;
                }
 
@@ -513,26 +668,27 @@ ipc_mqueue_post(
                 * and handle its error without getting the message.  We
                 * need to go back and pick another one.
                 */
+               receiver->ith_receiver_name = mqueue->imq_receiver_name;
                receiver->ith_kmsg = IKM_NULL;
                receiver->ith_seqno = 0;
                thread_unlock(receiver);
+               splx(th_spl);
        }
 
-       imq_unlock(mqueue);
-       splx(s);
+       /* clear the waitq boost we may have been given */
+       waitq_clear_promotion_locked(&mqueue->imq_wait_queue, current_thread());
+       imq_release_and_unlock(mqueue, reserved_prepost, s);
        
        current_task()->messages_sent++;
        return;
 }
 
 
-kern_return_t
-ipc_mqueue_receive_results(void)
+/* static */ void
+ipc_mqueue_receive_results(wait_result_t saved_wait_result)
 {
        thread_t                self = current_thread();
        mach_msg_option_t       option = self->ith_option;
-       kern_return_t           saved_wait_result = self->wait_result;
-       kern_return_t           mr;
 
        /*
         * why did we wake up?
@@ -543,15 +699,11 @@ ipc_mqueue_receive_results(void)
                return;
 
        case THREAD_INTERRUPTED:
-               if (option & MACH_RCV_TIMEOUT)
-                       thread_cancel_timer();
                self->ith_state = MACH_RCV_INTERRUPTED;
                return;
 
        case THREAD_RESTART:
                /* something bad happened to the port/set */
-               if (option & MACH_RCV_TIMEOUT)
-                       thread_cancel_timer();
                self->ith_state = MACH_RCV_PORT_CHANGED;
                return;
 
@@ -560,11 +712,6 @@ ipc_mqueue_receive_results(void)
                 * We do not need to go select a message, somebody
                 * handed us one (or a too-large indication).
                 */
-               if (option & MACH_RCV_TIMEOUT)
-                       thread_cancel_timer();
-               
-               mr = MACH_MSG_SUCCESS;
-
                switch (self->ith_state) {
                case MACH_RCV_SCATTER_SMALL:
                case MACH_RCV_TOO_LARGE:
@@ -592,9 +739,11 @@ ipc_mqueue_receive_results(void)
 }
 
 void
-ipc_mqueue_receive_continue(void)
+ipc_mqueue_receive_continue(
+       __unused void *param,
+       wait_result_t wresult)
 {
-       ipc_mqueue_receive_results();
+       ipc_mqueue_receive_results(wresult);
        mach_msg_receive_continue();  /* hard-coded for now */
 }
 
@@ -628,209 +777,444 @@ ipc_mqueue_receive_continue(void)
 
 void
 ipc_mqueue_receive(
-       ipc_mqueue_t            mqueue,
-       mach_msg_option_t       option,
-       mach_msg_size_t         max_size,
-       mach_msg_timeout_t      timeout,
-       int                     interruptible)
+       ipc_mqueue_t            mqueue,
+       mach_msg_option_t       option,
+       mach_msg_size_t         max_size,
+       mach_msg_timeout_t      rcv_timeout,
+       int                     interruptible)
 {
-       ipc_port_t              port;
-       mach_msg_return_t       mr, mr2;
-       ipc_kmsg_queue_t        kmsgs;
-       wait_result_t           wresult;
-       thread_t                self;
-       ipc_kmsg_t              *kmsgp;
-       mach_port_seqno_t       *seqnop;
-       spl_t s;
+       wait_result_t           wresult;
+        thread_t                self = current_thread();
+        
+        wresult = ipc_mqueue_receive_on_thread(mqueue, option, max_size,
+                                               rcv_timeout, interruptible,
+                                               self);
+        if (wresult == THREAD_NOT_WAITING)
+                return;
+
+       if (wresult == THREAD_WAITING) {
+               counter((interruptible == THREAD_ABORTSAFE) ? 
+                       c_ipc_mqueue_receive_block_user++ :
+                       c_ipc_mqueue_receive_block_kernel++);
+
+               if (self->ith_continuation)
+                       thread_block(ipc_mqueue_receive_continue);
+                       /* NOTREACHED */
+
+               wresult = thread_block(THREAD_CONTINUE_NULL);
+       }
+       ipc_mqueue_receive_results(wresult);
+}
+
+static int mqueue_process_prepost_receive(void *ctx, struct waitq *waitq,
+                                         struct waitq_set *wqset)
+{
+       ipc_mqueue_t     port_mq, *pmq_ptr;
+
+       (void)wqset;
+       port_mq = (ipc_mqueue_t)waitq;
+
+       /*
+        * If there are no messages on this queue, skip it and remove
+        * it from the prepost list
+        */
+       if (ipc_kmsg_queue_empty(&port_mq->imq_messages))
+               return WQ_ITERATE_INVALIDATE_CONTINUE;
+
+       /*
+        * There are messages waiting on this port.
+        * Instruct the prepost iteration logic to break, but keep the
+        * waitq locked.
+        */
+       pmq_ptr = (ipc_mqueue_t *)ctx;
+       if (pmq_ptr)
+               *pmq_ptr = port_mq;
+       return WQ_ITERATE_BREAK_KEEP_LOCKED;
+}
+
+wait_result_t
+ipc_mqueue_receive_on_thread(
+        ipc_mqueue_t            mqueue,
+       mach_msg_option_t       option,
+       mach_msg_size_t         max_size,
+       mach_msg_timeout_t      rcv_timeout,
+       int                     interruptible,
+       thread_t                thread)
+{
+       wait_result_t           wresult;
+       uint64_t                deadline;
+       spl_t                   s;
 
        s = splsched();
        imq_lock(mqueue);
+       /* no need to reserve anything: we never prepost to anyone */
        
        if (imq_is_set(mqueue)) {
-               wait_queue_link_t wql;
-               ipc_mqueue_t port_mq;
-               queue_t q;
+               ipc_mqueue_t port_mq = IMQ_NULL;
+               spl_t set_spl;
 
-               q = &mqueue->imq_setlinks;
-
-               /*
-                * If we are waiting on a portset mqueue, we need to see if
-                * any of the member ports have work for us.  If so, try to
-                * deliver one of those messages. By holding the portset's
-                * mqueue lock during the search, we tie up any attempts by
-                * mqueue_deliver or portset membership changes that may
-                * cross our path. But this is a lock order violation, so we
-                * have to do it "softly."  If we don't find a message waiting
-                * for us, we will assert our intention to wait while still
-                * holding that lock.  When we release the lock, the deliver/
-                * change will succeed and find us.
-                */
-       search_set:
-               queue_iterate(q, wql, wait_queue_link_t, wql_setlinks) {
-                       port_mq = (ipc_mqueue_t)wql->wql_queue;
-                       kmsgs = &port_mq->imq_messages;
-                       
-                       if (!imq_lock_try(port_mq)) {
-                               imq_unlock(mqueue);
-                               splx(s);
-                               delay(1);
-                               s = splsched();
-                               imq_lock(mqueue);
-                               goto search_set; /* start again at beginning - SMP */
-                       }
+               (void)waitq_set_iterate_preposts(&mqueue->imq_set_queue,
+                                                &port_mq,
+                                                mqueue_process_prepost_receive,
+                                                &set_spl);
 
+               if (port_mq != IMQ_NULL) {
                        /*
-                        * If there is still a message to be had, we will
-                        * try to select it (may not succeed because of size
-                        * and options).  In any case, we deliver those
-                        * results back to the user.
+                        * We get here if there is at least one message
+                        * waiting on port_mq. We have instructed the prepost
+                        * iteration logic to leave both the port_mq and the
+                        * set mqueue locked.
                         *
-                        * We also move the port's linkage to the tail of the
-                        * list for this set (fairness). Future versions will
-                        * sort by timestamp or priority.
+                        * TODO: previously, we would place this port at the
+                        *       back of the prepost list...
                         */
-                       if (ipc_kmsg_queue_first(kmsgs) == IKM_NULL) {
-                               imq_unlock(port_mq);
-                               continue;
-                       }
-                       queue_remove(q, wql, wait_queue_link_t, wql_setlinks);
-                       queue_enter(q, wql, wait_queue_link_t, wql_setlinks);
                        imq_unlock(mqueue);
 
-                       ipc_mqueue_select(port_mq, option, max_size);
+                       /* TODO: if/when port mqueues become non irq safe,
+                        *       we won't need this spl, and we should be
+                        *       able to call splx(s) (if that's even
+                        *       necessary).
+                        * For now, we've still disabled interrupts via
+                        * imq_reserve_and_lock();
+                        */
+                       splx(set_spl);
+
+                       /*
+                        * Continue on to handling the message with just
+                        * the port mqueue locked.
+                        */
+                       ipc_mqueue_select_on_thread(port_mq, mqueue, option,
+                                                   max_size, thread);
+
                        imq_unlock(port_mq);
                        splx(s);
-                       return;
-                       
+                       return THREAD_NOT_WAITING;
                }
-
        } else {
+               ipc_kmsg_queue_t kmsgs;
 
                /*
                 * Receive on a single port. Just try to get the messages.
                 */
                kmsgs = &mqueue->imq_messages;
                if (ipc_kmsg_queue_first(kmsgs) != IKM_NULL) {
-                       ipc_mqueue_select(mqueue, option, max_size);
+                       ipc_mqueue_select_on_thread(mqueue, IMQ_NULL, option,
+                                                   max_size, thread);
                        imq_unlock(mqueue);
                        splx(s);
-                       return;
+                       return THREAD_NOT_WAITING;
                }
        }
-               
+       
        /*
         * Looks like we'll have to block.  The mqueue we will
         * block on (whether the set's or the local port's) is
         * still locked.
         */
-       self = current_thread();
        if (option & MACH_RCV_TIMEOUT) {
-               if (timeout == 0) {
+               if (rcv_timeout == 0) {
                        imq_unlock(mqueue);
                        splx(s);
-                       self->ith_state = MACH_RCV_TIMED_OUT;
-                       return;
+                       thread->ith_state = MACH_RCV_TIMED_OUT;
+                       return THREAD_NOT_WAITING;
                }
        }
 
-       self->ith_state = MACH_RCV_IN_PROGRESS;
-       self->ith_option = option;
-       self->ith_msize = max_size;
-               
-       wresult = wait_queue_assert_wait64_locked(&mqueue->imq_wait_queue,
-                                               IPC_MQUEUE_RECEIVE,
-                                               interruptible,
-                                               TRUE); /* unlock? */
-       /* mqueue/waitq is unlocked */
+       /* NOTE: need splsched() here if mqueue no longer needs irq disabled */
+       thread_lock(thread);
+       thread->ith_state = MACH_RCV_IN_PROGRESS;
+       thread->ith_option = option;
+       thread->ith_msize = max_size;
+
+       if (option & MACH_RCV_TIMEOUT)
+               clock_interval_to_deadline(rcv_timeout, 1000*NSEC_PER_USEC, &deadline);
+       else
+               deadline = 0;
+
+       wresult = waitq_assert_wait64_locked(&mqueue->imq_wait_queue,
+                                            IPC_MQUEUE_RECEIVE,
+                                            interruptible,
+                                            TIMEOUT_URGENCY_USER_NORMAL,
+                                            deadline,
+                                            TIMEOUT_NO_LEEWAY,
+                                            thread);
+       /* preposts should be detected above, not here */
+       if (wresult == THREAD_AWAKENED)
+               panic("ipc_mqueue_receive_on_thread: sleep walking");
+
+       thread_unlock(thread);
+       imq_unlock(mqueue);
        splx(s);
-
-       if (wresult == THREAD_WAITING) {
-               if (option & MACH_RCV_TIMEOUT)
-                       thread_set_timer(timeout, 1000*NSEC_PER_USEC);
-
-               if (interruptible == THREAD_ABORTSAFE)
-                       counter(c_ipc_mqueue_receive_block_user++);
-               else
-                       counter(c_ipc_mqueue_receive_block_kernel++);
-
-               if (self->ith_continuation)
-                       thread_block(ipc_mqueue_receive_continue);
-                       /* NOTREACHED */
-
-               thread_block(THREAD_CONTINUE_NULL);
-       }
-       ipc_mqueue_receive_results();
+       return wresult;
 }
 
 
 /*
- *     Routine:        ipc_mqueue_select
+ *     Routine:        ipc_mqueue_select_on_thread
  *     Purpose:
  *             A receiver discovered that there was a message on the queue
  *             before he had to block.  Pick the message off the queue and
- *             "post" it to himself.
+ *             "post" it to thread.
  *     Conditions:
  *             mqueue locked.
+ *              thread not locked.
  *             There is a message.
+ *             No need to reserve prepost objects - it will never prepost
+ *
  *     Returns:
  *             MACH_MSG_SUCCESS        Actually selected a message for ourselves.
  *             MACH_RCV_TOO_LARGE  May or may not have pull it, but it is large
  */
 void
-ipc_mqueue_select(
-       ipc_mqueue_t            mqueue,
+ipc_mqueue_select_on_thread(
+       ipc_mqueue_t            port_mq,
+       ipc_mqueue_t            set_mq,
        mach_msg_option_t       option,
-       mach_msg_size_t         max_size)
+       mach_msg_size_t         max_size,
+       thread_t                thread)
 {
-       thread_t self = current_thread();
        ipc_kmsg_t kmsg;
-       mach_port_seqno_t seqno;
-       mach_msg_return_t mr;
-
-       mr = MACH_MSG_SUCCESS;
-
+       mach_msg_return_t mr = MACH_MSG_SUCCESS;
+       mach_msg_size_t rcv_size;
 
        /*
         * Do some sanity checking of our ability to receive
         * before pulling the message off the queue.
         */
-       kmsg = ipc_kmsg_queue_first(&mqueue->imq_messages);
-
+       kmsg = ipc_kmsg_queue_first(&port_mq->imq_messages);
        assert(kmsg != IKM_NULL);
 
-       if (kmsg->ikm_header.msgh_size + 
-               REQUESTED_TRAILER_SIZE(option) > max_size) {
-               mr = MACH_RCV_TOO_LARGE;
-       } 
-
        /*
         * If we really can't receive it, but we had the
         * MACH_RCV_LARGE option set, then don't take it off
         * the queue, instead return the appropriate error
         * (and size needed).
         */
-       if ((mr == MACH_RCV_TOO_LARGE) && (option & MACH_RCV_LARGE)) {
-               self->ith_kmsg = IKM_NULL;
-               self->ith_msize = kmsg->ikm_header.msgh_size;
-               self->ith_seqno = 0;
-               self->ith_state = mr;
-               return;
+       rcv_size = ipc_kmsg_copyout_size(kmsg, thread->map);
+       if (rcv_size + REQUESTED_TRAILER_SIZE(thread_is_64bit(thread), option) > max_size) {
+               mr = MACH_RCV_TOO_LARGE;
+               if (option & MACH_RCV_LARGE) {
+                       thread->ith_receiver_name = port_mq->imq_receiver_name;
+                       thread->ith_kmsg = IKM_NULL;
+                       thread->ith_msize = rcv_size;
+                       thread->ith_seqno = 0;
+                       thread->ith_state = mr;
+                       return;
+               }
        }
 
-       ipc_kmsg_rmqueue_first_macro(&mqueue->imq_messages, kmsg);
-       ipc_mqueue_release_msgcount(mqueue);
-       self->ith_seqno = mqueue->imq_seqno++;
-       self->ith_kmsg = kmsg;
-       self->ith_state = mr;
+       ipc_kmsg_rmqueue_first_macro(&port_mq->imq_messages, kmsg);
+       ipc_mqueue_release_msgcount(port_mq, set_mq);
+       thread->ith_seqno = port_mq->imq_seqno++;
+       thread->ith_kmsg = kmsg;
+       thread->ith_state = mr;
 
        current_task()->messages_received++;
        return;
 }
 
+/*
+ *     Routine:        ipc_mqueue_peek
+ *     Purpose:
+ *             Peek at a (non-set) message queue to see if it has a message
+ *             matching the sequence number provided (if zero, then the
+ *             first message in the queue) and return vital info about the
+ *             message.
+ *
+ *     Conditions:
+ *             Locks may be held by callers, so this routine cannot block.
+ *             Caller holds reference on the message queue.
+ */
+unsigned
+ipc_mqueue_peek(ipc_mqueue_t mq,
+                mach_port_seqno_t * seqnop,
+                mach_msg_size_t * msg_sizep,
+                mach_msg_id_t * msg_idp,
+                mach_msg_max_trailer_t * msg_trailerp)
+{
+       ipc_kmsg_queue_t kmsgq;
+       ipc_kmsg_t kmsg;
+       mach_port_seqno_t seqno, msgoff;
+       int res = 0;
+       spl_t s;
+
+       assert(!imq_is_set(mq));
+
+       s = splsched();
+       imq_lock(mq);
+
+       seqno = 0;
+       if (seqnop != NULL)
+               seqno = *seqnop;
+
+       if (seqno == 0) {
+               seqno = mq->imq_seqno;
+               msgoff = 0;
+       } else if (seqno >= mq->imq_seqno && 
+                  seqno < mq->imq_seqno + mq->imq_msgcount) {
+               msgoff = seqno - mq->imq_seqno;
+       } else
+               goto out;
+
+       /* look for the message that would match that seqno */
+       kmsgq = &mq->imq_messages;
+       kmsg = ipc_kmsg_queue_first(kmsgq);
+       while (msgoff-- && kmsg != IKM_NULL) {
+               kmsg = ipc_kmsg_queue_next(kmsgq, kmsg);
+       }
+       if (kmsg == IKM_NULL)
+               goto out;
+
+       /* found one - return the requested info */
+       if (seqnop != NULL)
+               *seqnop = seqno;
+       if (msg_sizep != NULL)
+               *msg_sizep = kmsg->ikm_header->msgh_size;
+       if (msg_idp != NULL)
+               *msg_idp = kmsg->ikm_header->msgh_id;
+       if (msg_trailerp != NULL)
+               memcpy(msg_trailerp, 
+                      (mach_msg_max_trailer_t *)((vm_offset_t)kmsg->ikm_header +
+                                                 round_msg(kmsg->ikm_header->msgh_size)),
+                      sizeof(mach_msg_max_trailer_t));
+       res = 1;
+
+ out:
+       imq_unlock(mq);
+       splx(s);
+       return res;
+}
+
+
+/*
+ * peek at the contained port message queues, break prepost iteration as soon
+ * as we spot a message on one of the message queues referenced by the set's
+ * prepost list.  No need to lock each message queue, as only the head of each
+ * queue is checked. If a message wasn't there before we entered here, no need
+ * to find it (if we do, great).
+ */
+static int mqueue_peek_iterator(void *ctx, struct waitq *waitq,
+                               struct waitq_set *wqset)
+{
+       ipc_mqueue_t port_mq = (ipc_mqueue_t)waitq;
+       ipc_kmsg_queue_t kmsgs = &port_mq->imq_messages;
+
+       (void)ctx;
+       (void)wqset;
+               
+       if (ipc_kmsg_queue_first(kmsgs) != IKM_NULL)
+               return WQ_ITERATE_BREAK; /* break out of the prepost iteration */
+
+       return WQ_ITERATE_CONTINUE;
+}
+
+/*
+ *     Routine:        ipc_mqueue_set_peek
+ *     Purpose:
+ *             Peek at a message queue set to see if it has any ports
+ *             with messages.
+ *
+ *     Conditions:
+ *             Locks may be held by callers, so this routine cannot block.
+ *             Caller holds reference on the message queue.
+ */
+unsigned
+ipc_mqueue_set_peek(ipc_mqueue_t mq)
+{
+       spl_t s;
+       int ret;
+
+       assert(imq_is_set(mq));
+
+       s = splsched();
+       imq_lock(mq);
+
+       ret = waitq_set_iterate_preposts(&mq->imq_set_queue, NULL,
+                                        mqueue_peek_iterator, NULL);
+
+       imq_unlock(mq);
+       splx(s);
+       return (ret == WQ_ITERATE_BREAK);
+}
+
+/*
+ *     Routine:        ipc_mqueue_set_gather_member_names
+ *     Purpose:
+ *             Discover all ports which are members of a given port set.
+ *             Because the waitq linkage mechanism was redesigned to save
+ *             significan amounts of memory, it no longer keeps back-pointers
+ *             from a port set to a port. Therefore, we must iterate over all
+ *             ports within a given IPC space and individually query them to
+ *             see if they are members of the given set. Port names of ports
+ *             found to be members of the given set will be gathered into the
+ *             provided 'names' array.  Actual returned names are limited to
+ *             maxnames entries, but we keep counting the actual number of
+ *             members to let the caller decide to retry if necessary.
+ *
+ *     Conditions:
+ *             Locks may be held by callers, so this routine cannot block.
+ *             Caller holds reference on the message queue (via port set).
+ */
+void
+ipc_mqueue_set_gather_member_names(
+       ipc_space_t space,
+       ipc_mqueue_t set_mq,
+       ipc_entry_num_t maxnames,
+       mach_port_name_t *names,
+       ipc_entry_num_t *actualp)
+{
+       ipc_entry_t table;
+       ipc_entry_num_t tsize;
+       struct waitq_set *wqset;
+       ipc_entry_num_t actual = 0;
+
+       assert(set_mq != IMQ_NULL);
+       wqset = &set_mq->imq_set_queue;
+
+       assert(space != IS_NULL);
+       is_read_lock(space);
+       if (!is_active(space)) {
+               is_read_unlock(space);
+               goto out;
+       }
+
+       if (!waitq_set_is_valid(wqset)) {
+               is_read_unlock(space);
+               goto out;
+       }
+
+       table = space->is_table;
+       tsize = space->is_table_size;
+       for (ipc_entry_num_t idx = 0; idx < tsize; idx++) {
+               ipc_entry_t entry = &table[idx];
+
+               /* only receive rights can be members of port sets */
+               if ((entry->ie_bits & MACH_PORT_TYPE_RECEIVE) != MACH_PORT_TYPE_NONE) {
+                       __IGNORE_WCASTALIGN(ipc_port_t port = (ipc_port_t)entry->ie_object);
+                       ipc_mqueue_t mq = &port->ip_messages;
+
+                       assert(IP_VALID(port));
+                       if (ip_active(port) &&
+                           waitq_member(&mq->imq_wait_queue, wqset)) {
+                               if (actual < maxnames)
+                                       names[actual] = mq->imq_receiver_name;
+                               actual++;
+                       }
+               }
+       }
+
+       is_read_unlock(space);
+
+out:
+       *actualp = actual;
+}
+
+
 /*
  *     Routine:        ipc_mqueue_destroy
  *     Purpose:
- *             Destroy a message queue.  Set any blocked senders running.
+ *             Destroy a (non-set) message queue.
+ *             Set any blocked senders running.
  *             Destroy the kmsgs in the queue.
  *     Conditions:
  *             Nothing locked.
@@ -838,38 +1222,71 @@ ipc_mqueue_select(
  */
 void
 ipc_mqueue_destroy(
-       ipc_mqueue_t    mqueue) 
+       ipc_mqueue_t    mqueue)
 {
        ipc_kmsg_queue_t kmqueue;
        ipc_kmsg_t kmsg;
+       boolean_t reap = FALSE;
        spl_t s;
 
+       assert(!imq_is_set(mqueue));
 
        s = splsched();
        imq_lock(mqueue);
+
        /*
         *      rouse all blocked senders
+        *      (don't boost anyone - we're tearing this queue down)
+        *      (never preposts)
         */
        mqueue->imq_fullwaiters = FALSE;
-       wait_queue_wakeup64_all_locked(
-                               &mqueue->imq_wait_queue,
-                               IPC_MQUEUE_FULL,
-                               THREAD_AWAKENED,
-                               FALSE);
+       waitq_wakeup64_all_locked(&mqueue->imq_wait_queue,
+                                 IPC_MQUEUE_FULL,
+                                 THREAD_RESTART,
+                                 NULL,
+                                 WAITQ_ALL_PRIORITIES,
+                                 WAITQ_KEEP_LOCKED);
 
+       /*
+        * Move messages from the specified queue to the per-thread
+        * clean/drain queue while we have the mqueue lock.
+        */
        kmqueue = &mqueue->imq_messages;
-
        while ((kmsg = ipc_kmsg_dequeue(kmqueue)) != IKM_NULL) {
-               imq_unlock(mqueue);
-               splx(s);
+               boolean_t first;
+               first = ipc_kmsg_delayed_destroy(kmsg);
+               if (first)
+                       reap = first;
+       }
 
-               ipc_kmsg_destroy_dest(kmsg);
+       /*
+        * Wipe out message count, both for messages about to be
+        * reaped and for reserved space for (previously) woken senders.
+        * This is the indication to them that their reserved space is gone
+        * (the mqueue was destroyed).
+        */
+       mqueue->imq_msgcount = 0;
+
+       /* clear out any preposting we may have done */
+       waitq_clear_prepost_locked(&mqueue->imq_wait_queue, &s);
 
-               s = splsched();
-               imq_lock(mqueue);
-       }
        imq_unlock(mqueue);
        splx(s);
+
+       /*
+        * assert that we're destroying a queue that's not a
+        * member of any other queue
+        */
+       assert(mqueue->imq_wait_queue.waitq_prepost_id == 0);
+       assert(mqueue->imq_wait_queue.waitq_set_id == 0);
+
+
+       /*
+        * Destroy the messages we enqueued if we aren't nested
+        * inside some other attempt to drain the same queue.
+        */
+       if (reap)
+               ipc_kmsg_reap_delayed();
 }
 
 /*
@@ -888,6 +1305,8 @@ ipc_mqueue_set_qlimit(
 {
         spl_t s;
 
+        assert(qlimit <= MACH_PORT_QLIMIT_MAX);
+
         /* wake up senders allowed by the new qlimit */
         s = splsched();
         imq_lock(mqueue);
@@ -898,16 +1317,25 @@ ipc_mqueue_set_qlimit(
                 wakeup = qlimit - mqueue->imq_qlimit;
 
                 for (i = 0; i < wakeup; i++) {
-                        if (wait_queue_wakeup64_one_locked(
-                                                       &mqueue->imq_wait_queue,
-                                                       IPC_MQUEUE_FULL,
-                                                       THREAD_AWAKENED,
-                                                       FALSE) == KERN_NOT_WAITING) {
-                                        mqueue->imq_fullwaiters = FALSE;
-                                        break;
-                        }
+                       /*
+                        * boost the priority of the awoken thread
+                        * (WAITQ_PROMOTE_PRIORITY) to ensure it uses
+                        * the message queue slot we've just reserved.
+                        *
+                        * NOTE: this will never prepost
+                        */
+                       if (waitq_wakeup64_one_locked(&mqueue->imq_wait_queue,
+                                                     IPC_MQUEUE_FULL,
+                                                     THREAD_AWAKENED,
+                                                     NULL,
+                                                     WAITQ_PROMOTE_PRIORITY,
+                                                     WAITQ_KEEP_LOCKED) == KERN_NOT_WAITING) {
+                               mqueue->imq_fullwaiters = FALSE;
+                               break;
+                       }
+                       mqueue->imq_msgcount++;  /* give it to the awakened thread */
                 }
-        }
+       }
        mqueue->imq_qlimit = qlimit;
        imq_unlock(mqueue);
        splx(s);
@@ -964,7 +1392,7 @@ ipc_mqueue_copyin(
        ipc_mqueue_t mqueue;
 
        is_read_lock(space);
-       if (!space->is_active) {
+       if (!is_active(space)) {
                is_read_unlock(space);
                return MACH_RCV_INVALID_NAME;
        }
@@ -979,9 +1407,8 @@ ipc_mqueue_copyin(
 
        if (entry->ie_bits & MACH_PORT_TYPE_RECEIVE) {
                ipc_port_t port;
-               ipc_pset_t pset;
 
-               port = (ipc_port_t) object;
+               __IGNORE_WCASTALIGN(port = (ipc_port_t) object);
                assert(port != IP_NULL);
 
                ip_lock(port);
@@ -994,7 +1421,7 @@ ipc_mqueue_copyin(
        } else if (entry->ie_bits & MACH_PORT_TYPE_PORT_SET) {
                ipc_pset_t pset;
 
-               pset = (ipc_pset_t) object;
+               __IGNORE_WCASTALIGN(pset = (ipc_pset_t) object);
                assert(pset != IPS_NULL);
 
                ips_lock(pset);
@@ -1020,4 +1447,3 @@ ipc_mqueue_copyin(
        *mqueuep = mqueue;
        return MACH_MSG_SUCCESS;
 }
-