]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/kern/sysv_sem.c
xnu-792.1.5.tar.gz
[apple/xnu.git] / bsd / kern / sysv_sem.c
index 8c4c3d47c81e31f7188b664cf77211d9ae4e61f6..8f7b26537c25e1b34e3638dd26fd71ba171af894 100644 (file)
@@ -1,24 +1,21 @@
 /*
- * Copyright (c) 2000-2003 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved.
  *
  * @APPLE_LICENSE_HEADER_START@
  * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
+ * The contents of this file constitute Original Code as defined in and
+ * are subject to the Apple Public Source License Version 1.1 (the
+ * "License").  You may not use this file except in compliance with the
+ * License.  Please obtain a copy of the License at
+ * http://www.apple.com/publicsource and read it before using this file.
  * 
- * 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.
- * 
- * The Original Code and all software distributed under the License are
- * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * This Original Code and all software distributed under the License are
+ * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
- * Please see the License for the specific language governing rights and
- * limitations under the License.
+ * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
+ * License for the specific language governing rights and limitations
+ * under the License.
  * 
  * @APPLE_LICENSE_HEADER_END@
  */
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/kernel.h>
-#include <sys/proc.h>
-#include <sys/sem.h>
+#include <sys/proc_internal.h>
+#include <sys/kauth.h>
+#include <sys/sem_internal.h>
 #include <sys/malloc.h>
 #include <mach/mach_types.h>
 
 #include <sys/filedesc.h>
-#include <sys/file.h>
-#include <sys/kern_audit.h>
+#include <sys/file_internal.h>
 #include <sys/sysctl.h>
+#include <sys/ipcs.h>
+#include <sys/sysent.h>
+#include <sys/sysproto.h>
+
+#include <bsm/audit_kernel.h>
 
-/*#include <sys/sysproto.h>*/
-/*#include <sys/sysent.h>*/
 
 /* Uncomment this line to see the debugging output */
 /* #define SEM_DEBUG */
 
-/* Macros to deal with the semaphore subsystem lock.  The lock currently uses
- * the semlock_holder static variable as a mutex.  NULL means no lock, any
- * value other than NULL means locked.  semlock_holder is used because it was
- * present in the code before the Darwin port, and for no other reason.
- * When the time comes to relax the funnel requirements of the kernel only
- * these macros should need to be changed.  A spin lock would work well.
- */
-/* Aquire the lock */
-#define SUBSYSTEM_LOCK_AQUIRE(p) { sysv_sem_aquiring_threads++; \
-    while (semlock_holder != NULL) \
-        (void) tsleep((caddr_t)&semlock_holder, (PZERO - 4), "sysvsem", 0); \
-    semlock_holder = p; \
-    sysv_sem_aquiring_threads--; }
-
-/* Release the lock */
-#define SUBSYSTEM_LOCK_RELEASE { semlock_holder = NULL; wakeup((caddr_t)&semlock_holder); }
+#define M_SYSVSEM      M_TEMP
 
-/* Release the lock and return a value */
-#define UNLOCK_AND_RETURN(ret) { SUBSYSTEM_LOCK_RELEASE; return(ret); }
-
-#define M_SYSVSEM      M_SUBPROC
-
-#if 0
-static void seminit __P((void *));
-SYSINIT(sysv_sem, SI_SUB_SYSV_SEM, SI_ORDER_FIRST, seminit, NULL)
-#endif 0
 
 /* Hard system limits to avoid resource starvation / DOS attacks.
  * These are not needed if we can make the semaphore pages swappable.
@@ -112,27 +88,11 @@ struct seminfo seminfo = {
        SEMAEM  /* adjust on exit max value */
 };
 
-/* A counter so the module unload code knows when there are no more processes using
- * the sysv_sem code */
-static long sysv_sem_sleeping_threads = 0;
-static long sysv_sem_aquiring_threads = 0;
-
-struct semctl_args;
-int semctl __P((struct proc *p, struct semctl_args *uap, int *));
-struct semget_args;
-int semget __P((struct proc *p, struct semget_args *uap, int *));
-struct semop_args;
-int semop __P((struct proc *p, struct semop_args *uap, int *));
-struct semconfig_args;
-int semconfig __P((struct proc *p, struct semconfig_args *uap, int *));
 
-
-static struct sem_undo *semu_alloc __P((struct proc *p));
-static int semundo_adjust __P((struct proc *p, struct sem_undo **supptr, 
-               int semid, int semnum, int adjval));
-static void semundo_clear __P((int semid, int semnum));
-
-typedef int     sy_call_t __P((struct proc *, void *, int *));
+static struct sem_undo *semu_alloc(struct proc *p);
+static int semundo_adjust(struct proc *p, struct sem_undo **supptr, 
+               int semid, int semnum, int adjval);
+static void semundo_clear(int semid, int semnum);
 
 /* XXX casting to (sy_call_t *) is bogus, as usual. */
 static sy_call_t *semcalls[] = {
@@ -140,45 +100,97 @@ static sy_call_t *semcalls[] = {
        (sy_call_t *)semop, (sy_call_t *)semconfig
 };
 
-static int     semtot = 0;                     /* # of used semaphores */
-struct semid_ds *sema = NULL;                  /* semaphore id pool */
-struct sem *sem =  NULL;                       /* semaphore pool */
-static struct sem_undo *semu_list = NULL;   /* list of active undo structures */
-struct sem_undo *semu = NULL;                  /* semaphore undo pool */
+static int             semtot = 0;             /* # of used semaphores */
+struct user_semid_ds   *sema = NULL;           /* semaphore id pool */
+struct sem             *sem_pool =  NULL;      /* semaphore pool */
+static struct sem_undo *semu_list = NULL;      /* active undo structures */
+struct sem_undo                *semu = NULL;           /* semaphore undo pool */
 
-static struct proc *semlock_holder = NULL;
 
-/* seminit no longer needed.  The data structures are grown dynamically */
-void
-seminit()
+void sysv_sem_lock_init(void);
+static lck_grp_t       *sysv_sem_subsys_lck_grp;
+static lck_grp_attr_t  *sysv_sem_subsys_lck_grp_attr;
+static lck_attr_t      *sysv_sem_subsys_lck_attr;
+static lck_mtx_t        sysv_sem_subsys_mutex;
+
+#define SYSV_SEM_SUBSYS_LOCK() lck_mtx_lock(&sysv_sem_subsys_mutex)
+#define SYSV_SEM_SUBSYS_UNLOCK() lck_mtx_unlock(&sysv_sem_subsys_mutex)
+
+
+__private_extern__ void
+sysv_sem_lock_init( void )
+{
+
+    sysv_sem_subsys_lck_grp_attr = lck_grp_attr_alloc_init();
+    lck_grp_attr_setstat(sysv_sem_subsys_lck_grp_attr);
+
+    sysv_sem_subsys_lck_grp = lck_grp_alloc_init("sysv_shm_subsys_lock", sysv_sem_subsys_lck_grp_attr);
+
+    sysv_sem_subsys_lck_attr = lck_attr_alloc_init();
+    lck_attr_setdebug(sysv_sem_subsys_lck_attr); 
+    lck_mtx_init(&sysv_sem_subsys_mutex, sysv_sem_subsys_lck_grp, sysv_sem_subsys_lck_attr);
+}
+
+static __inline__ user_time_t
+sysv_semtime(void)
+{
+       struct timeval  tv;
+       microtime(&tv);
+       return (tv.tv_sec);
+}
+
+/*
+ * XXX conversion of internal user_time_t to external tume_t loses
+ * XXX precision; not an issue for us now, since we are only ever
+ * XXX setting 32 bits worth of time into it.
+ *
+ * pad field contents are not moved correspondingly; contents will be lost
+ *
+ * NOTE: Source and target may *NOT* overlap! (target is smaller)
+ */
+static void
+semid_ds_64to32(struct user_semid_ds *in, struct semid_ds *out)
 {
+       out->sem_perm = in->sem_perm;
+       out->sem_base = (__int32_t)in->sem_base;
+       out->sem_nsems = in->sem_nsems;
+       out->sem_otime = in->sem_otime;         /* XXX loses precision */
+       out->sem_ctime = in->sem_ctime;         /* XXX loses precision */
 }
 
+/*
+ * pad field contents are not moved correspondingly; contents will be lost
+ *
+ * NOTE: Source and target may are permitted to overlap! (source is smaller);
+ * this works because we copy fields in order from the end of the struct to
+ * the beginning.
+ *
+ * XXX use CAST_USER_ADDR_T() for lack of a CAST_USER_TIME_T(); net effect
+ * XXX is the same.
+ */
+static void
+semid_ds_32to64(struct semid_ds *in, struct user_semid_ds *out)
+{
+       out->sem_ctime = in->sem_ctime;
+       out->sem_otime = in->sem_otime;
+       out->sem_nsems = in->sem_nsems;
+       out->sem_base = (void *)in->sem_base;
+       out->sem_perm = in->sem_perm;
+}
+
+
 /*
  * Entry point for all SEM calls
  *
  * In Darwin this is no longer the entry point.  It will be removed after
  *  the code has been tested better.
  */
-struct semsys_args {
-       u_int   which;
-       int     a2;
-       int     a3;
-       int     a4;
-       int     a5;
-};
+/* XXX actually varargs. */
 int
-semsys(p, uap, retval)
-       struct proc *p;
-       /* XXX actually varargs. */
-       struct semsys_args *uap;
-       register_t *retval;
+semsys(struct proc *p, struct semsys_args *uap, register_t *retval)
 {
 
        /* The individual calls handling the locking now */
-       /*while (semlock_holder != NULL && semlock_holder != p)
-               (void) tsleep((caddr_t)&semlock_holder, (PZERO - 4), "semsys", 0);
-        */
 
        if (uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
                return (EINVAL);
@@ -200,27 +212,18 @@ semsys(p, uap, retval)
  * in /dev/kmem.
  */
 
-#ifndef _SYS_SYSPROTO_H_
-struct semconfig_args {
-       semconfig_ctl_t flag;
-};
-#endif
-
 int
-semconfig(p, uap, retval)
-       struct proc *p;
-       struct semconfig_args *uap;
-       register_t *retval;
+semconfig(__unused struct proc *p, struct semconfig_args *uap, register_t *retval)
 {
        int eval = 0;
 
        switch (uap->flag) {
        case SEM_CONFIG_FREEZE:
-               SUBSYSTEM_LOCK_AQUIRE(p);
+               SYSV_SEM_SUBSYS_LOCK();
                break;
 
        case SEM_CONFIG_THAW:
-               SUBSYSTEM_LOCK_RELEASE;
+               SYSV_SEM_SUBSYS_UNLOCK();
                break;
 
        default:
@@ -234,19 +237,26 @@ semconfig(p, uap, retval)
        return(eval);
 }
 
-/* Expand the semu array to the given capacity.  If the expansion fails
+/*
+ * Expand the semu array to the given capacity.  If the expansion fails
  * return 0, otherwise return 1.
  *
  * Assumes we already have the subsystem lock.
  */
 static int
-grow_semu_array(newSize)
-       int newSize;
+grow_semu_array(int newSize)
 {
-       register int i, j;
+       register int i;
        register struct sem_undo *newSemu;
+       static boolean_t grow_semu_array_in_progress = FALSE;
+
+       while (grow_semu_array_in_progress) {
+               msleep(&grow_semu_array_in_progress, &sysv_sem_subsys_mutex,
+                      PPAUSE, "grow_semu_array", NULL);
+       }
+
        if (newSize <= seminfo.semmnu)
-               return 0;
+               return 1;
        if (newSize > limitseminfo.semmnu) /* enforce hard limit */
        {
 #ifdef SEM_DEBUG
@@ -261,8 +271,13 @@ grow_semu_array(newSize)
 #ifdef SEM_DEBUG
        printf("growing semu[] from %d to %d\n", seminfo.semmnu, newSize);
 #endif
-       MALLOC(newSemu, struct sem_undo*, sizeof(struct sem_undo)*newSize,
+       grow_semu_array_in_progress = TRUE;
+       SYSV_SEM_SUBSYS_UNLOCK();
+       MALLOC(newSemu, struct sem_undo*, sizeof(struct sem_undo) * newSize,
                M_SYSVSEM, M_WAITOK);
+       SYSV_SEM_SUBSYS_LOCK();
+       grow_semu_array_in_progress = FALSE;
+       wakeup((caddr_t) &grow_semu_array_in_progress);
        if (NULL == newSemu)
        {
 #ifdef SEM_DEBUG
@@ -275,8 +290,6 @@ grow_semu_array(newSize)
        for (i = 0; i < seminfo.semmnu; i++)
        {
                newSemu[i] = semu[i];
-               for(j = 0; j < SEMUME; j++)   /* Is this really needed? */
-                       newSemu[i].un_ent[j] = semu[i].un_ent[j];
        }
                for (i = seminfo.semmnu; i < newSize; i++)
         {
@@ -302,10 +315,9 @@ grow_semu_array(newSize)
  * Assumes we already have the subsystem lock.
  */
 static int
-grow_sema_array(newSize)
-       int newSize;
+grow_sema_array(int newSize)
 {
-       register struct semid_ds *newSema;
+       register struct user_semid_ds *newSema;
        register int i;
 
        if (newSize <= seminfo.semmni)
@@ -324,7 +336,7 @@ grow_sema_array(newSize)
 #ifdef SEM_DEBUG
        printf("growing sema[] from %d to %d\n", seminfo.semmni, newSize);
 #endif
-       MALLOC(newSema, struct semid_ds*, sizeof(struct semid_ds)*newSize,
+       MALLOC(newSema, struct user_semid_ds *, sizeof(struct user_semid_ds) * newSize,
                M_SYSVSEM, M_WAITOK);
        if (NULL == newSema)
        {
@@ -344,7 +356,7 @@ grow_sema_array(newSize)
                 * this with the existing code, so we wake up the
                 * process and let it do a lot of work to determine the
                 * semaphore set is really not available yet, and then
-                * sleep on the correct, reallocated semid_ds pointer.
+                * sleep on the correct, reallocated user_semid_ds pointer.
                 */
                if (sema[i].sem_perm.mode & SEM_ALLOC)
                        wakeup((caddr_t)&sema[i]);
@@ -352,7 +364,7 @@ grow_sema_array(newSize)
 
        for (i = seminfo.semmni; i < newSize; i++)
        {
-               newSema[i].sem_base = 0;
+               newSema[i].sem_base = NULL;
                newSema[i].sem_perm.mode = 0;
        }
 
@@ -369,38 +381,38 @@ grow_sema_array(newSize)
 }
 
 /*
- * Expand the sem array to the given capacity.  If the expansion fails
+ * Expand the sem_pool array to the given capacity.  If the expansion fails
  * we return 0 (fail), otherwise we return 1 (success).
  *
  * Assumes we already hold the subsystem lock.
  */
 static int
-grow_sem_array(newSize)
-               int newSize;
+grow_sem_pool(int new_pool_size)
 {
-       register struct sem *newSem = NULL;
-       register int i;
+       struct sem *new_sem_pool = NULL;
+       struct sem *sem_free;
+       int i;
 
-       if (newSize < semtot)
+       if (new_pool_size < semtot)
                return 0;
-       if (newSize > limitseminfo.semmns) /* enforce hard limit */
-       {
+       /* enforce hard limit */
+       if (new_pool_size > limitseminfo.semmns) {
 #ifdef SEM_DEBUG
                printf("semaphore hard limit of %d reached, requested %d\n",
-                       limitseminfo.semmns, newSize);
+                       limitseminfo.semmns, new_pool_size);
 #endif
                return 0;
        }
-       newSize = (newSize/SEMMNS_INC + 1) * SEMMNS_INC;
-       newSize = newSize > limitseminfo.semmns ? limitseminfo.semmns : newSize;
+
+       new_pool_size = (new_pool_size/SEMMNS_INC + 1) * SEMMNS_INC;
+       new_pool_size = new_pool_size > limitseminfo.semmns ? limitseminfo.semmns : new_pool_size;
 
 #ifdef SEM_DEBUG
-       printf("growing sem array from %d to %d\n", seminfo.semmns, newSize);
+       printf("growing sem_pool array from %d to %d\n", seminfo.semmns, new_pool_size);
 #endif
-       MALLOC(newSem, struct sem*, sizeof(struct sem)*newSize,
+       MALLOC(new_sem_pool, struct sem *, sizeof(struct sem) * new_pool_size,
                M_SYSVSEM, M_WAITOK);
-       if (NULL == newSem)
-       {
+       if (NULL == new_sem_pool) {
 #ifdef SEM_DEBUG
                printf("allocation failed.  no changes made.\n");
 #endif
@@ -408,26 +420,24 @@ grow_sem_array(newSize)
        }
 
        /* We have our new memory, now copy the old contents over */
-       if (sem)
+       if (sem_pool)
                for(i = 0; i < seminfo.semmns; i++)
-                       newSem[i] = sem[i];
+                       new_sem_pool[i] = sem_pool[i];
 
        /* Update our id structures to point to the new semaphores */
-       for(i = 0; i < seminfo.semmni; i++)
+       for(i = 0; i < seminfo.semmni; i++) {
                if (sema[i].sem_perm.mode & SEM_ALLOC)  /* ID in use */
-               {
-                       if (newSem > sem)
-                               sema[i].sem_base += newSem - sem;
-                       else
-                               sema[i].sem_base -= sem - newSem;
-               }
+                       sema[i].sem_base += (new_sem_pool - sem_pool);
+       }
+
+       sem_free = sem_pool;
+       sem_pool = new_sem_pool;
 
        /* clean up the old array */
-       if (sem)
-               FREE(sem, M_SYSVSEM);
+       if (sem_free != NULL)
+               FREE(sem_free, M_SYSVSEM);
 
-       sem = newSem;
-       seminfo.semmns = newSize;
+       seminfo.semmns = new_pool_size;
 #ifdef SEM_DEBUG
        printf("expansion complete\n");
 #endif
@@ -442,8 +452,7 @@ grow_sem_array(newSize)
  */
 
 static struct sem_undo *
-semu_alloc(p)
-       struct proc *p;
+semu_alloc(struct proc *p)
 {
        register int i;
        register struct sem_undo *suptr;
@@ -468,6 +477,7 @@ semu_alloc(p)
                                suptr->un_next = semu_list;
                                semu_list = suptr;
                                suptr->un_cnt = 0;
+                               suptr->un_ent = NULL;
                                suptr->un_proc = p;
                                return(suptr);
                        }
@@ -517,16 +527,12 @@ semu_alloc(p)
  *
  * Assumes we already hold the subsystem lock.
  */
-
 static int
-semundo_adjust(p, supptr, semid, semnum, adjval)
-       register struct proc *p;
-       struct sem_undo **supptr;
-       int semid, semnum;
-       int adjval;
+semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid,
+       int semnum, int adjval)
 {
        register struct sem_undo *suptr;
-       register struct undo *sunptr;
+       register struct undo *sueptr, **suepptr, *new_sueptr;
        int i;
 
        /* Look for and remember the sem_undo if the caller doesn't provide
@@ -555,31 +561,75 @@ semundo_adjust(p, supptr, semid, semnum, adjval)
         * Look for the requested entry and adjust it (delete if adjval becomes
         * 0).
         */
-       sunptr = &suptr->un_ent[0];
-       for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
-               if (sunptr->un_id != semid || sunptr->un_num != semnum)
+       new_sueptr = NULL;
+lookup:
+       for (i = 0, suepptr = &suptr->un_ent, sueptr = suptr->un_ent;
+            i < suptr->un_cnt;
+            i++, suepptr = &sueptr->une_next, sueptr = sueptr->une_next) {
+               if (sueptr->une_id != semid || sueptr->une_num != semnum)
                        continue;
                if (adjval == 0)
-                       sunptr->un_adjval = 0;
+                       sueptr->une_adjval = 0;
                else
-                       sunptr->un_adjval += adjval;
-               if (sunptr->un_adjval == 0) {
+                       sueptr->une_adjval += adjval;
+               if (sueptr->une_adjval == 0) {
                        suptr->un_cnt--;
-                       if (i < suptr->un_cnt)
-                               suptr->un_ent[i] =
-                                   suptr->un_ent[suptr->un_cnt];
+                       *suepptr = sueptr->une_next;
+                       FREE(sueptr, M_SYSVSEM);
+                       sueptr = NULL;
+               }
+               if (new_sueptr != NULL) {
+                       /*
+                        * We lost the race: free the "undo" entry we allocated
+                        * and use the one that won.
+                        */
+                       FREE(new_sueptr, M_SYSVSEM);
+                       new_sueptr = NULL;
                }
                return(0);
        }
 
        /* Didn't find the right entry - create it */
-       if (adjval == 0)
+       if (adjval == 0) {
+               if (new_sueptr != NULL) {
+                       FREE(new_sueptr, M_SYSVSEM);
+                       new_sueptr = NULL;
+               }
                return(0);
-       if (suptr->un_cnt != limitseminfo.semume) {
-               sunptr = &suptr->un_ent[suptr->un_cnt];
+       }
+
+       if (new_sueptr != NULL) {
+               /*
+                * Use the new "undo" entry we allocated in the previous pass
+                */
+               new_sueptr->une_next = suptr->un_ent;
+               suptr->un_ent = new_sueptr;
                suptr->un_cnt++;
-               sunptr->un_adjval = adjval;
-               sunptr->un_id = semid; sunptr->un_num = semnum;
+               new_sueptr->une_adjval = adjval;
+               new_sueptr->une_id = semid;
+               new_sueptr->une_num = semnum;
+               return 0;
+       }
+
+       if (suptr->un_cnt != limitseminfo.semume) {
+               SYSV_SEM_SUBSYS_UNLOCK();
+               /*
+                * Unlocking opens the door to race conditions.  Someone else
+                * could be trying to allocate the same thing at this point,
+                * so we'll have to check if we lost the race.
+                */
+               MALLOC(new_sueptr, struct undo *, sizeof (struct undo),
+                      M_SYSVSEM, M_WAITOK);
+               SYSV_SEM_SUBSYS_LOCK();
+               if (new_sueptr == NULL) {
+                       return ENOMEM;
+               }
+               /*
+                * There might be other threads doing the same thing for this
+                * process, so check again if an "undo" entry exists for that
+                * semaphore.
+                */
+               goto lookup;
        } else
                return(EINVAL);
        return(0);
@@ -588,94 +638,96 @@ semundo_adjust(p, supptr, semid, semnum, adjval)
 /* Assumes we already hold the subsystem lock.
  */
 static void
-semundo_clear(semid, semnum)
-       int semid, semnum;
+semundo_clear(int semid, int semnum)
 {
-       register struct sem_undo *suptr;
+       struct sem_undo *suptr;
 
        for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
-               register struct undo *sunptr = &suptr->un_ent[0];
-               register int i = 0;
+               struct undo *sueptr;
+               struct undo **suepptr;
+               int i = 0;
 
+               sueptr = suptr->un_ent;
+               suepptr = &suptr->un_ent;
                while (i < suptr->un_cnt) {
-                       if (sunptr->un_id == semid) {
-                               if (semnum == -1 || sunptr->un_num == semnum) {
+                       if (sueptr->une_id == semid) {
+                               if (semnum == -1 || sueptr->une_num == semnum) {
                                        suptr->un_cnt--;
-                                       if (i < suptr->un_cnt) {
-                                               suptr->un_ent[i] =
-                                                 suptr->un_ent[suptr->un_cnt];
-                                               continue;
-                                       }
+                                       *suepptr = sueptr->une_next;
+                                       FREE(sueptr, M_SYSVSEM);
+                                       sueptr = *suepptr;
+                                       continue;
                                }
                                if (semnum != -1)
                                        break;
                        }
-                       i++, sunptr++;
+                       i++;
+                       suepptr = &sueptr->une_next;
+                       sueptr = sueptr->une_next;
                }
        }
 }
 
 /*
- * Note that the user-mode half of this passes a union, not a pointer
+ * Note that the user-mode half of this passes a union coerced to a
+ * user_addr_t.  The union contains either an int or a pointer, and
+ * so we have to coerce it back, variant on whether the calling
+ * process is 64 bit or not.  The coercion works for the 'val' element
+ * because the alignment is the same in user and kernel space.
  */
-#ifndef _SYS_SYSPROTO_H_
-struct semctl_args {
-       int     semid;
-       int     semnum;
-       int     cmd;
-       union   semun arg;
-};
-#endif
-
 int
-semctl(p, uap, retval)
-       struct proc *p;
-       register struct semctl_args *uap;
-       register_t *retval;
+semctl(struct proc *p, struct semctl_args *uap, register_t *retval)
 {
        int semid = uap->semid;
        int semnum = uap->semnum;
        int cmd = uap->cmd;
-       union semun arg = uap->arg;
-       union semun real_arg;
-       struct ucred *cred = p->p_ucred;
+       user_semun_t user_arg = (user_semun_t)uap->arg;
+       kauth_cred_t cred = kauth_cred_get();
        int i, rval, eval;
-       struct semid_ds sbuf;
-       register struct semid_ds *semaptr;
+       struct user_semid_ds sbuf;
+       struct user_semid_ds *semaptr;
+       struct user_semid_ds uds;
+       
 
        AUDIT_ARG(svipc_cmd, cmd);
        AUDIT_ARG(svipc_id, semid);
-       SUBSYSTEM_LOCK_AQUIRE(p);
+
+       SYSV_SEM_SUBSYS_LOCK();
+
 #ifdef SEM_DEBUG
-       printf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
+       printf("call to semctl(%d, %d, %d, 0x%qx)\n", semid, semnum, cmd, user_arg);
 #endif
 
        semid = IPCID_TO_IX(semid);
-       if (semid < 0 || semid >= seminfo.semmni)
-{
+
+       if (semid < 0 || semid >= seminfo.semmni) {
 #ifdef SEM_DEBUG
                printf("Invalid semid\n");
 #endif
-               UNLOCK_AND_RETURN(EINVAL);
-}
+               eval = EINVAL;
+               goto semctlout;
+       }
 
        semaptr = &sema[semid];
        if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
-           semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
-               UNLOCK_AND_RETURN(EINVAL);
+           semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
+               eval = EINVAL;
+               goto semctlout;
+       }
 
        eval = 0;
        rval = 0;
 
        switch (cmd) {
        case IPC_RMID:
-               if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_M)))
-                       UNLOCK_AND_RETURN(eval);
-               semaptr->sem_perm.cuid = cred->cr_uid;
-               semaptr->sem_perm.uid = cred->cr_uid;
+               if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_M))) 
+                       goto semctlout;
+
+               semaptr->sem_perm.cuid = kauth_cred_getuid(cred);
+               semaptr->sem_perm.uid = kauth_cred_getuid(cred);
                semtot -= semaptr->sem_nsems;
-               for (i = semaptr->sem_base - sem; i < semtot; i++)
-                       sem[i] = sem[i + semaptr->sem_nsems];
+               for (i = semaptr->sem_base - sem_pool; i < semtot; i++)
+                       sem_pool[i] = sem_pool[i + semaptr->sem_nsems];
                for (i = 0; i < seminfo.semmni; i++) {
                        if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
                            sema[i].sem_base > semaptr->sem_base)
@@ -688,60 +740,84 @@ semctl(p, uap, retval)
 
        case IPC_SET:
                if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_M)))
-                       UNLOCK_AND_RETURN(eval);
-               /*if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
-                       UNLOCK_AND_RETURN(eval);*/
-               if ((eval = copyin(arg.buf, (caddr_t)&sbuf,
-                   sizeof(sbuf))) != 0)
-                       UNLOCK_AND_RETURN(eval);
+                               goto semctlout;
+
+               SYSV_SEM_SUBSYS_UNLOCK();
+
+               if (IS_64BIT_PROCESS(p)) {
+                       eval = copyin(user_arg.buf, &sbuf, sizeof(struct user_semid_ds));
+               } else {
+                       eval = copyin(user_arg.buf, &sbuf, sizeof(struct semid_ds));
+                       /* convert in place; ugly, but safe */
+                       semid_ds_32to64((struct semid_ds *)&sbuf, &sbuf);
+               }
+               
+               if (eval != 0)
+                       return(eval);
+
+               SYSV_SEM_SUBSYS_LOCK();
+
                semaptr->sem_perm.uid = sbuf.sem_perm.uid;
                semaptr->sem_perm.gid = sbuf.sem_perm.gid;
                semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
                    (sbuf.sem_perm.mode & 0777);
-               semaptr->sem_ctime = time_second;
+               semaptr->sem_ctime = sysv_semtime();
                break;
 
        case IPC_STAT:
                if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
-                       UNLOCK_AND_RETURN(eval);
-               /*if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
-                       UNLOCK_AND_RETURN(eval);*/
-               eval = copyout((caddr_t)semaptr, arg.buf,
-                   sizeof(struct semid_ds));
+                               goto semctlout;
+               bcopy(semaptr, &uds, sizeof(struct user_semid_ds));
+               SYSV_SEM_SUBSYS_UNLOCK();
+               if (IS_64BIT_PROCESS(p)) {
+                       eval = copyout(&uds, user_arg.buf, sizeof(struct user_semid_ds));
+               } else {
+                       struct semid_ds semid_ds32;
+                       semid_ds_64to32(&uds, &semid_ds32);
+                       eval = copyout(&semid_ds32, user_arg.buf, sizeof(struct semid_ds));
+               }
+               SYSV_SEM_SUBSYS_LOCK();
                break;
 
        case GETNCNT:
                if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
-                       UNLOCK_AND_RETURN(eval);
-               if (semnum < 0 || semnum >= semaptr->sem_nsems)
-                       UNLOCK_AND_RETURN(EINVAL);
+                               goto semctlout;
+               if (semnum < 0 || semnum >= semaptr->sem_nsems) {
+                       eval = EINVAL;
+                       goto semctlout;
+               }
                rval = semaptr->sem_base[semnum].semncnt;
                break;
 
        case GETPID:
                if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
-                       UNLOCK_AND_RETURN(eval);
-               if (semnum < 0 || semnum >= semaptr->sem_nsems)
-                       UNLOCK_AND_RETURN(EINVAL);
+                               goto semctlout;
+               if (semnum < 0 || semnum >= semaptr->sem_nsems) {
+                       eval = EINVAL;
+                       goto semctlout;
+               }
                rval = semaptr->sem_base[semnum].sempid;
                break;
 
        case GETVAL:
                if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
-                       UNLOCK_AND_RETURN(eval);
-               if (semnum < 0 || semnum >= semaptr->sem_nsems)
-                       UNLOCK_AND_RETURN(EINVAL);
+                               goto semctlout;
+               if (semnum < 0 || semnum >= semaptr->sem_nsems) {
+                       eval = EINVAL;
+                       goto semctlout;
+               }
                rval = semaptr->sem_base[semnum].semval;
                break;
 
        case GETALL:
                if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
-                       UNLOCK_AND_RETURN(eval);
-               /*if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
-                       UNLOCK_AND_RETURN(eval);*/
+                               goto semctlout;
+/* XXXXXXXXXXXXXXXX TBD XXXXXXXXXXXXXXXX */
                for (i = 0; i < semaptr->sem_nsems; i++) {
+                       /* XXX could be done in one go... */
                        eval = copyout((caddr_t)&semaptr->sem_base[i].semval,
-                           &arg.array[i], sizeof(arg.array[0]));
+                           user_arg.array + (i * sizeof(unsigned short)),
+                           sizeof(unsigned short));
                        if (eval != 0)
                                break;
                }
@@ -749,9 +825,11 @@ semctl(p, uap, retval)
 
        case GETZCNT:
                if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
-                       UNLOCK_AND_RETURN(eval);
-               if (semnum < 0 || semnum >= semaptr->sem_nsems)
-                       UNLOCK_AND_RETURN(EINVAL);
+                               goto semctlout;
+               if (semnum < 0 || semnum >= semaptr->sem_nsems) {
+                       eval = EINVAL;
+                       goto semctlout;
+               }
                rval = semaptr->sem_base[semnum].semzcnt;
                break;
 
@@ -761,36 +839,35 @@ semctl(p, uap, retval)
 #ifdef SEM_DEBUG
                        printf("Invalid credentials for write\n");
 #endif
-                       UNLOCK_AND_RETURN(eval);
+                               goto semctlout;
                }
                if (semnum < 0 || semnum >= semaptr->sem_nsems)
                {
 #ifdef SEM_DEBUG
                        printf("Invalid number out of range for set\n");
 #endif
-                       UNLOCK_AND_RETURN(EINVAL);
+                       eval = EINVAL;
+                       goto semctlout;
                }
-               /*if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
-               {
-#ifdef SEM_DEBUG
-                       printf("Error during value copyin\n");
-#endif
-                       UNLOCK_AND_RETURN(eval);
-               }*/
-               semaptr->sem_base[semnum].semval = arg.val;
+               /*
+                * Cast down a pointer instead of using 'val' member directly
+                * to avoid introducing endieness and a pad field into the
+                * header file.  Ugly, but it works.
+                */
+               semaptr->sem_base[semnum].semval = CAST_DOWN(int,user_arg.buf);
                semundo_clear(semid, semnum);
                wakeup((caddr_t)semaptr);
                break;
 
        case SETALL:
                if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_W)))
-                       UNLOCK_AND_RETURN(eval);
-               /*if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
-                       UNLOCK_AND_RETURN(eval);*/
+                               goto semctlout;
+/*** XXXXXXXXXXXX TBD ********/
                for (i = 0; i < semaptr->sem_nsems; i++) {
-                       eval = copyin(&arg.array[i],
+                       /* XXX could be done in one go... */
+                       eval = copyin(user_arg.array + (i * sizeof(unsigned short)),
                            (caddr_t)&semaptr->sem_base[i].semval,
-                           sizeof(arg.array[0]));
+                           sizeof(unsigned short));
                        if (eval != 0)
                                break;
                }
@@ -799,41 +876,36 @@ semctl(p, uap, retval)
                break;
 
        default:
-               UNLOCK_AND_RETURN(EINVAL);
+                       eval = EINVAL;
+                       goto semctlout;
        }
 
        if (eval == 0)
                *retval = rval;
-       UNLOCK_AND_RETURN(eval);
+semctlout:
+       SYSV_SEM_SUBSYS_UNLOCK();
+       return(eval);
 }
 
-#ifndef _SYS_SYSPROTO_H_
-struct semget_args {
-       key_t   key;
-       int     nsems;
-       int     semflg;
-};
-#endif
-
 int
-semget(p, uap, retval)
-       struct proc *p;
-       register struct semget_args *uap;
-       register_t *retval;
+semget(__unused struct proc *p, struct semget_args *uap, register_t *retval)
 {
        int semid, eval;
        int key = uap->key;
        int nsems = uap->nsems;
        int semflg = uap->semflg;
-       struct ucred *cred = p->p_ucred;
+       kauth_cred_t cred = kauth_cred_get();
 
-       SUBSYSTEM_LOCK_AQUIRE(p);
 #ifdef SEM_DEBUG
        if (key != IPC_PRIVATE)
                printf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
        else
                printf("semget(IPC_PRIVATE, %d, 0%o)\n", nsems, semflg);
 #endif
+
+
+               SYSV_SEM_SUBSYS_LOCK();
+
     
        if (key != IPC_PRIVATE) {
                for (semid = 0; semid < seminfo.semmni; semid++) {
@@ -847,18 +919,20 @@ semget(p, uap, retval)
 #endif
                        if ((eval = ipcperm(cred, &sema[semid].sem_perm,
                            semflg & 0700)))
-                               UNLOCK_AND_RETURN(eval);
-                       if (nsems > 0 && sema[semid].sem_nsems < nsems) {
+                               goto semgetout;
+                       if (nsems < 0 || sema[semid].sem_nsems < nsems) {
 #ifdef SEM_DEBUG
                                printf("too small\n");
 #endif
-                               UNLOCK_AND_RETURN(EINVAL);
+                               eval = EINVAL;
+                               goto semgetout;
                        }
                        if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
 #ifdef SEM_DEBUG
                                printf("not exclusive\n");
 #endif
-                               UNLOCK_AND_RETURN(EEXIST);
+                               eval = EEXIST;
+                               goto semgetout;
                        }
                        goto found;
                }
@@ -873,19 +947,20 @@ semget(p, uap, retval)
                        printf("nsems out of range (0<%d<=%d)\n", nsems,
                            seminfo.semmsl);
 #endif
-                       UNLOCK_AND_RETURN(EINVAL);
+                       eval = EINVAL;
+                       goto semgetout;
                }
                if (nsems > seminfo.semmns - semtot) {
 #ifdef SEM_DEBUG
                        printf("not enough semaphores left (need %d, got %d)\n",
                            nsems, seminfo.semmns - semtot);
 #endif
-                       if (!grow_sem_array(semtot + nsems))
-                       {
+                       if (!grow_sem_pool(semtot + nsems)) {
 #ifdef SEM_DEBUG
                                printf("failed to grow the sem array\n");
 #endif
-                               UNLOCK_AND_RETURN(ENOSPC);
+                               eval = ENOSPC;
+                               goto semgetout;
                        }
                }
                for (semid = 0; semid < seminfo.semmni; semid++) {
@@ -901,15 +976,16 @@ semget(p, uap, retval)
 #ifdef SEM_DEBUG
                                printf("failed to grow sema array\n");
 #endif
-                               UNLOCK_AND_RETURN(ENOSPC);
+                               eval = ENOSPC;
+                               goto semgetout;
                        }
                }
 #ifdef SEM_DEBUG
                printf("semid %d is available\n", semid);
 #endif
                sema[semid].sem_perm.key = key;
-               sema[semid].sem_perm.cuid = cred->cr_uid;
-               sema[semid].sem_perm.uid = cred->cr_uid;
+               sema[semid].sem_perm.cuid = kauth_cred_getuid(cred);
+               sema[semid].sem_perm.uid = kauth_cred_getuid(cred);
                sema[semid].sem_perm.cgid = cred->cr_gid;
                sema[semid].sem_perm.gid = cred->cr_gid;
                sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
@@ -917,20 +993,21 @@ semget(p, uap, retval)
                    (sema[semid].sem_perm.seq + 1) & 0x7fff;
                sema[semid].sem_nsems = nsems;
                sema[semid].sem_otime = 0;
-               sema[semid].sem_ctime = time_second;
-               sema[semid].sem_base = &sem[semtot];
+               sema[semid].sem_ctime = sysv_semtime();
+               sema[semid].sem_base = &sem_pool[semtot];
                semtot += nsems;
                bzero(sema[semid].sem_base,
                    sizeof(sema[semid].sem_base[0])*nsems);
 #ifdef SEM_DEBUG
                printf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base,
-                   &sem[semtot]);
+                   &sem_pool[semtot]);
 #endif
        } else {
 #ifdef SEM_DEBUG
                printf("didn't find it and wasn't asked to create it\n");
 #endif
-               UNLOCK_AND_RETURN(ENOENT);
+               eval = ENOENT;
+               goto semgetout;
        }
 
 found:
@@ -939,72 +1016,73 @@ found:
 #ifdef SEM_DEBUG
        printf("semget is done, returning %d\n", *retval);
 #endif
-       SUBSYSTEM_LOCK_RELEASE;
-       return(0);
-}
+       eval = 0;
 
-#ifndef _SYS_SYSPROTO_H_
-struct semop_args {
-       int     semid;
-       struct  sembuf *sops;
-       int     nsops;
-};
-#endif
+semgetout:
+       SYSV_SEM_SUBSYS_UNLOCK();
+       return(eval);
+}
 
 int
-semop(p, uap, retval)
-       struct proc *p;
-       register struct semop_args *uap;
-       register_t *retval;
+semop(struct proc *p, struct semop_args *uap, register_t *retval)
 {
        int semid = uap->semid;
        int nsops = uap->nsops;
        struct sembuf sops[MAX_SOPS];
-       register struct semid_ds *semaptr;
-       register struct sembuf *sopptr;
-       register struct sem *semptr;
+       register struct user_semid_ds *semaptr;
+       register struct sembuf *sopptr = NULL;  /* protected by 'semptr' */
+       register struct sem *semptr = NULL;     /* protected by 'if' */
        struct sem_undo *suptr = NULL;
-       struct ucred *cred = p->p_ucred;
        int i, j, eval;
        int do_wakeup, do_undos;
 
        AUDIT_ARG(svipc_id, uap->semid);
-       SUBSYSTEM_LOCK_AQUIRE(p);
+
+       SYSV_SEM_SUBSYS_LOCK();
+
 #ifdef SEM_DEBUG
        printf("call to semop(%d, 0x%x, %d)\n", semid, sops, nsops);
 #endif
 
        semid = IPCID_TO_IX(semid);     /* Convert back to zero origin */
 
-       if (semid < 0 || semid >= seminfo.semmni)
-               UNLOCK_AND_RETURN(EINVAL);
+       if (semid < 0 || semid >= seminfo.semmni) {
+               eval = EINVAL;
+               goto semopout;
+       }
 
        semaptr = &sema[semid];
-       if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
-               UNLOCK_AND_RETURN(EINVAL);
-       if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
-               UNLOCK_AND_RETURN(EINVAL);
+       if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) {
+               eval = EINVAL;
+               goto semopout;
+       }
+       if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
+               eval = EINVAL;
+               goto semopout;
+       }
 
-       if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_W))) {
+       if ((eval = ipcperm(kauth_cred_get(), &semaptr->sem_perm, IPC_W))) {
 #ifdef SEM_DEBUG
                printf("eval = %d from ipaccess\n", eval);
 #endif
-               UNLOCK_AND_RETURN(eval);
+               goto semopout;
        }
 
-       if (nsops > MAX_SOPS) {
+       if (nsops < 0 || nsops > MAX_SOPS) {
 #ifdef SEM_DEBUG
                printf("too many sops (max=%d, nsops=%d)\n", MAX_SOPS, nsops);
 #endif
-               UNLOCK_AND_RETURN(E2BIG);
+               eval = E2BIG;
+               goto semopout;
        }
 
-       if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) {
+       /*  OK for LP64, since sizeof(struct sembuf) is currently invariant */
+       if ((eval = copyin(uap->sops, &sops, nsops * sizeof(struct sembuf))) != 0) {
 #ifdef SEM_DEBUG
                printf("eval = %d from copyin(%08x, %08x, %ld)\n", eval,
-                   uap->sops, &sops, nsops * sizeof(sops[0]));
+                   uap->sops, &sops, nsops * sizeof(struct sembuf));
 #endif
-               UNLOCK_AND_RETURN(eval);
+               goto semopout;
        }
 
        /*
@@ -1024,8 +1102,10 @@ semop(p, uap, retval)
                for (i = 0; i < nsops; i++) {
                        sopptr = &sops[i];
 
-                       if (sopptr->sem_num >= semaptr->sem_nsems)
-                               UNLOCK_AND_RETURN(EFBIG);
+                       if (sopptr->sem_num >= semaptr->sem_nsems) {
+                               eval = EFBIG;
+                               goto semopout;
+                       }
 
                        semptr = &semaptr->sem_base[sopptr->sem_num];
 
@@ -1086,8 +1166,10 @@ semop(p, uap, retval)
                 * If the request that we couldn't satisfy has the
                 * NOWAIT flag set then return with EAGAIN.
                 */
-               if (sopptr->sem_flg & IPC_NOWAIT)
-                       UNLOCK_AND_RETURN(EAGAIN);
+               if (sopptr->sem_flg & IPC_NOWAIT) {
+                       eval = EAGAIN;
+                       goto semopout;
+               }
 
                if (sopptr->sem_op == 0)
                        semptr->semzcnt++;
@@ -1102,23 +1184,23 @@ semop(p, uap, retval)
                 * waiting for. We will get the lock back after we
                 * wake up.
                 */
-               SUBSYSTEM_LOCK_RELEASE;
-                sysv_sem_sleeping_threads++;
-               eval = tsleep((caddr_t)semaptr, (PZERO - 4) | PCATCH,
+               eval = msleep((caddr_t)semaptr, &sysv_sem_subsys_mutex , (PZERO - 4) | PCATCH,
                    "semwait", 0);
-                sysv_sem_sleeping_threads--;
                 
 #ifdef SEM_DEBUG
                printf("semop:  good morning (eval=%d)!\n", eval);
 #endif
-               /* There is no need to get the lock if we are just
-                * going to return without performing more semaphore
-                * operations.
-                */
-               if (eval != 0)
-                       return(EINTR);
+               /* we need the lock here due to mods on semptr */
+               if (eval != 0) {
+                       if (sopptr->sem_op == 0)
+                               semptr->semzcnt--;
+                       else
+                               semptr->semncnt--;
+
+                       eval = EINTR;
+                       goto semopout;
+               }
 
-               SUBSYSTEM_LOCK_AQUIRE(p);       /* Get it back */
                suptr = NULL;   /* sem_undo may have been reallocated */
                semaptr = &sema[semid];    /* sema may have been reallocated */
 
@@ -1134,11 +1216,16 @@ semop(p, uap, retval)
                    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
                        /* The man page says to return EIDRM. */
                        /* Unfortunately, BSD doesn't define that code! */
+                       if (sopptr->sem_op == 0)
+                               semptr->semzcnt--;
+                       else
+                               semptr->semncnt--;
 #ifdef EIDRM
-                       UNLOCK_AND_RETURN(EIDRM);
+               eval = EIDRM;
 #else
-                       UNLOCK_AND_RETURN(EINVAL);
+               eval = EINVAL;
 #endif
+               goto semopout;
                }
 
                /*
@@ -1203,7 +1290,7 @@ done:
 #ifdef SEM_DEBUG
                        printf("eval = %d from semundo_adjust\n", eval);
 #endif
-                       UNLOCK_AND_RETURN(eval);
+                       goto semopout;
                } /* loop through the sops */
        } /* if (do_undos) */
 
@@ -1214,16 +1301,6 @@ done:
                semptr->sempid = p->p_pid;
        }
 
-       /* Do a wakeup if any semaphore was up'd.
-        *  we will release our lock on the semaphore subsystem before
-        *  we wakeup other processes to prevent a little thrashing.
-        *  Note that this is fine because we are done using the
-        *  semaphore structures at this point in time.  We only use
-        *  a local variable pointer value, and the retval
-        *  parameter.
-        *  Note 2: Future use of sem_wakeup may reqiure the lock.
-        */
-       SUBSYSTEM_LOCK_RELEASE;
        if (do_wakeup) {
 #ifdef SEM_DEBUG
                printf("semop:  doing wakeup\n");
@@ -1241,7 +1318,10 @@ done:
        printf("semop:  done\n");
 #endif
        *retval = 0;
-       return(0);
+       eval = 0;
+semopout:
+       SYSV_SEM_SUBSYS_UNLOCK();
+       return(eval);
 }
 
 /*
@@ -1249,8 +1329,7 @@ done:
  * semaphores.
  */
 void
-semexit(p)
-       struct proc *p;
+semexit(struct proc *p)
 {
        register struct sem_undo *suptr;
        register struct sem_undo **supptr;
@@ -1260,10 +1339,11 @@ semexit(p)
         * anything to undo, but we need the lock to prevent
         * dynamic memory race conditions.
         */
-       SUBSYSTEM_LOCK_AQUIRE(p);
-       if (!sem)
+       SYSV_SEM_SUBSYS_LOCK();
+
+       if (!sem_pool)
        {
-               SUBSYSTEM_LOCK_RELEASE;
+               SYSV_SEM_SUBSYS_UNLOCK();
                return;
        }
        did_something = 0;
@@ -1291,13 +1371,17 @@ semexit(p)
         * If there are any active undo elements then process them.
         */
        if (suptr->un_cnt > 0) {
-               int ix;
+               while (suptr->un_ent != NULL) {
+                       struct undo *sueptr;
+                       int semid;
+                       int semnum;
+                       int adjval;
+                       struct user_semid_ds *semaptr;
 
-               for (ix = 0; ix < suptr->un_cnt; ix++) {
-                       int semid = suptr->un_ent[ix].un_id;
-                       int semnum = suptr->un_ent[ix].un_num;
-                       int adjval = suptr->un_ent[ix].un_adjval;
-                       struct semid_ds *semaptr;
+                       sueptr = suptr->un_ent;
+                       semid = sueptr->une_id;
+                       semnum = sueptr->une_num;
+                       adjval = sueptr->une_adjval;
 
                        semaptr = &sema[semid];
                        if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
@@ -1307,10 +1391,11 @@ semexit(p)
 
 #ifdef SEM_DEBUG
                        printf("semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
-                           suptr->un_proc, suptr->un_ent[ix].un_id,
-                           suptr->un_ent[ix].un_num,
-                           suptr->un_ent[ix].un_adjval,
-                           semaptr->sem_base[semnum].semval);
+                              suptr->un_proc,
+                              semid,
+                              semnum,
+                              adjval,
+                              semaptr->sem_base[semnum].semval);
 #endif
 
                        if (adjval < 0) {
@@ -1338,6 +1423,10 @@ semexit(p)
 #ifdef SEM_DEBUG
                        printf("semexit:  back from wakeup\n");
 #endif
+                       suptr->un_cnt--;
+                       suptr->un_ent = sueptr->une_next;
+                       FREE(sueptr, M_SYSVSEM);
+                       sueptr = NULL;
                }
        }
 
@@ -1370,32 +1459,36 @@ unlock:
          * same leaky semaphore problem.
          */
 
-       SUBSYSTEM_LOCK_RELEASE;
+       SYSV_SEM_SUBSYS_UNLOCK();
 }
+
+
 /* (struct sysctl_oid *oidp, void *arg1, int arg2, \
         struct sysctl_req *req) */
 static int
-sysctl_seminfo SYSCTL_HANDLER_ARGS
+sysctl_seminfo(__unused struct sysctl_oid *oidp, void *arg1,
+       __unused int arg2, struct sysctl_req *req)
 {
        int error = 0;
 
        error = SYSCTL_OUT(req, arg1, sizeof(int));
-       if (error || !req->newptr)
+       if (error || req->newptr == USER_ADDR_NULL)
                return(error);
 
-       SUBSYSTEM_LOCK_AQUIRE(current_proc());
+       SYSV_SEM_SUBSYS_LOCK();
+
        /* Set the values only if shared memory is not initialised */
-       if ((sem == (struct sem *) 0) && 
-               (sema == (struct semid_ds *) 0) && 
-               (semu == (struct semid_ds *) 0) && 
-               (semu_list == (struct sem_undo *) 0)) {
-                       if (error = SYSCTL_IN(req, arg1, sizeof(int))) {
+       if ((sem_pool == NULL) && 
+               (sema == NULL) && 
+               (semu == NULL) && 
+               (semu_list == NULL)) {
+                       if ((error = SYSCTL_IN(req, arg1, sizeof(int)))) {
                                goto out;
                        }
        } else 
                error = EINVAL;
 out:
-       SUBSYSTEM_LOCK_RELEASE;
+       SYSV_SEM_SUBSYS_UNLOCK();
        return(error);
        
 }
@@ -1418,3 +1511,103 @@ SYSCTL_PROC(_kern_sysv, KSYSV_SEMUNE, semume, CTLTYPE_INT | CTLFLAG_RW,
     &limitseminfo.semume, 0, &sysctl_seminfo ,"I","semume");
 
 
+static int
+IPCS_sem_sysctl(__unused struct sysctl_oid *oidp, __unused void *arg1,
+       __unused int arg2, struct sysctl_req *req)
+{
+       int error;
+       int cursor;
+       union {
+               struct IPCS_command u32;
+               struct user_IPCS_command u64;
+       } ipcs;
+       struct semid_ds semid_ds32;     /* post conversion, 32 bit version */
+       void *semid_dsp;
+       size_t ipcs_sz = sizeof(struct user_IPCS_command);
+       size_t semid_ds_sz = sizeof(struct user_semid_ds);
+       struct proc *p = current_proc();
+
+       /* Copy in the command structure */
+       if ((error = SYSCTL_IN(req, &ipcs, ipcs_sz)) != 0) {
+               return(error);
+       }
+
+       if (!IS_64BIT_PROCESS(p)) {
+               ipcs_sz = sizeof(struct IPCS_command);
+               semid_ds_sz = sizeof(struct semid_ds);
+       }
+
+       /* Let us version this interface... */
+       if (ipcs.u64.ipcs_magic != IPCS_MAGIC) {
+               return(EINVAL);
+       }
+
+       SYSV_SEM_SUBSYS_LOCK();
+       switch(ipcs.u64.ipcs_op) {
+       case IPCS_SEM_CONF:     /* Obtain global configuration data */
+               if (ipcs.u64.ipcs_datalen != sizeof(struct seminfo)) {
+                       error = ERANGE;
+                       break;
+               }
+               if (ipcs.u64.ipcs_cursor != 0) {        /* fwd. compat. */
+                       error = EINVAL;
+                       break;
+               }
+               SYSV_SEM_SUBSYS_UNLOCK();
+               error = copyout(&seminfo, ipcs.u64.ipcs_data, ipcs.u64.ipcs_datalen);
+               SYSV_SEM_SUBSYS_LOCK();
+               break;
+
+       case IPCS_SEM_ITER:     /* Iterate over existing segments */
+               cursor = ipcs.u64.ipcs_cursor;
+               if (cursor < 0 || cursor >= seminfo.semmni) {
+                       error = ERANGE;
+                       break;
+               }
+               if (ipcs.u64.ipcs_datalen != (int)semid_ds_sz ) {
+                       error = EINVAL;
+                       break;
+               }
+               for( ; cursor < seminfo.semmni; cursor++) {
+                       if (sema[cursor].sem_perm.mode & SEM_ALLOC)
+                               break;
+                       continue;
+               }
+               if (cursor == seminfo.semmni) {
+                       error = ENOENT;
+                       break;
+               }
+
+               semid_dsp = &sema[cursor];      /* default: 64 bit */
+
+               /*
+                * If necessary, convert the 64 bit kernel segment
+                * descriptor to a 32 bit user one.
+                */
+               if (!IS_64BIT_PROCESS(p)) {
+                       semid_ds_64to32(semid_dsp, &semid_ds32);
+                       semid_dsp = &semid_ds32;
+               }
+               SYSV_SEM_SUBSYS_UNLOCK();
+               error = copyout(semid_dsp, ipcs.u64.ipcs_data, ipcs.u64.ipcs_datalen);
+               if (!error) {
+                       /* update cursor */
+                       ipcs.u64.ipcs_cursor = cursor + 1;
+                       error = SYSCTL_OUT(req, &ipcs, ipcs_sz);
+               }
+               SYSV_SEM_SUBSYS_LOCK();
+               break;
+
+       default:
+               error = EINVAL;
+               break;
+       }
+       SYSV_SEM_SUBSYS_UNLOCK();
+       return(error);
+}
+
+SYSCTL_DECL(_kern_sysv_ipcs);
+SYSCTL_PROC(_kern_sysv_ipcs, OID_AUTO, sem, CTLFLAG_RW|CTLFLAG_ANYBODY,
+       0, 0, IPCS_sem_sysctl,
+       "S,IPCS_sem_command",
+       "ipcs sem command interface");