]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/vfs/vfs_quota.c
xnu-2422.1.72.tar.gz
[apple/xnu.git] / bsd / vfs / vfs_quota.c
index 36f4fb89aa2705a4fa61d1a6bf85d44de5b10228..4edc2115ba1b8f980051affa868b508d4ca8dd46 100644 (file)
@@ -1,23 +1,29 @@
 /*
- * Copyright (c) 2002 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2002-2006 Apple Computer, Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
- * 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. 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.
  * 
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * 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
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
- * License for the specific language governing rights and limitations
- * under the License.
+ * 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.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * Copyright (c) 1982, 1986, 1990, 1993, 1995
 #include <sys/kernel.h>
 #include <sys/systm.h>
 #include <sys/malloc.h>
-#include <sys/file.h>
-#include <sys/proc.h>
-#include <sys/vnode.h>
-#include <sys/mount.h>
+#include <sys/file_internal.h>
+#include <sys/proc_internal.h>
+#include <sys/vnode_internal.h>
+#include <sys/mount_internal.h>
 #include <sys/quota.h>
+#include <sys/uio_internal.h>
+
+#include <libkern/OSByteOrder.h>
+
+
+/* vars for quota file lock */
+lck_grp_t      * qf_lck_grp;
+lck_grp_attr_t * qf_lck_grp_attr;
+lck_attr_t     * qf_lck_attr;
+
+/* vars for quota list lock */
+lck_grp_t      * quota_list_lck_grp;
+lck_grp_attr_t * quota_list_lck_grp_attr;
+lck_attr_t     * quota_list_lck_attr;
+lck_mtx_t      * quota_list_mtx_lock;
+
+/* Routines to lock and unlock the quota global data */
+static int dq_list_lock(void);
+static void dq_list_unlock(void);
 
+static void dq_lock_internal(struct dquot *dq);
+static void dq_unlock_internal(struct dquot *dq);
 
 static u_int32_t quotamagic[MAXQUOTAS] = INITQMAGICS;
 
@@ -76,127 +103,388 @@ static u_int32_t quotamagic[MAXQUOTAS] = INITQMAGICS;
  * Code pertaining to management of the in-core dquot data structures.
  */
 #define DQHASH(dqvp, id) \
-       (&dqhashtbl[((((int)(dqvp)) >> 8) + id) & dqhash])
+       (&dqhashtbl[((((intptr_t)(dqvp)) >> 8) + id) & dqhash])
 LIST_HEAD(dqhash, dquot) *dqhashtbl;
 u_long dqhash;
 
+#define        DQUOTINC        5       /* minimum free dquots desired */
+long numdquot, desireddquot = DQUOTINC;
+
 /*
  * Dquot free list.
  */
-#define        DQUOTINC        5       /* minimum free dquots desired */
 TAILQ_HEAD(dqfreelist, dquot) dqfreelist;
-long numdquot, desireddquot = DQUOTINC;
+/*
+ * Dquot dirty orphans list
+ */
+TAILQ_HEAD(dqdirtylist, dquot) dqdirtylist;
+
 
+static int  dqlookup(struct quotafile *, u_int32_t, struct     dqblk *, u_int32_t *);
+static int  dqsync_locked(struct dquot *dq);
 
-static int dqlookup(struct quotafile *, u_long, struct dqblk *, u_int32_t *);
+static void qf_lock(struct quotafile *);
+static void qf_unlock(struct quotafile *);
+static int  qf_ref(struct quotafile *);
+static void qf_rele(struct quotafile *);
 
 
 /*
- * Initialize the quota system.
+ * Initialize locks for the quota system.
  */
 void
-dqinit()
+dqinit(void)
 {
+       /*
+        * Allocate quota list lock group attribute and group
+        */
+       quota_list_lck_grp_attr= lck_grp_attr_alloc_init();
+       quota_list_lck_grp = lck_grp_alloc_init("quota list",  quota_list_lck_grp_attr);
+       
+       /*
+        * Allocate qouta list lock attribute
+        */
+       quota_list_lck_attr = lck_attr_alloc_init();
+
+       /*
+        * Allocate quota list lock
+        */
+       quota_list_mtx_lock = lck_mtx_alloc_init(quota_list_lck_grp, quota_list_lck_attr);
+
+
+       /*
+        * allocate quota file lock group attribute and group
+        */
+       qf_lck_grp_attr= lck_grp_attr_alloc_init();
+       qf_lck_grp = lck_grp_alloc_init("quota file", qf_lck_grp_attr);
+
+       /*
+        * Allocate quota file lock attribute
+        */
+       qf_lck_attr = lck_attr_alloc_init();
+}
+
+/*
+ * Report whether dqhashinit has been run.
+ */
+int
+dqisinitialized(void)
+{
+       return (dqhashtbl != NULL);
+}
+
+/*
+ * Initialize hash table for dquot structures.
+ */
+void
+dqhashinit(void)
+{
+       dq_list_lock();
+       if (dqisinitialized())
+               goto out;
 
-       dqhashtbl = hashinit(desiredvnodes, M_DQUOT, &dqhash);
        TAILQ_INIT(&dqfreelist);
+       TAILQ_INIT(&dqdirtylist);
+       dqhashtbl = hashinit(desiredvnodes, M_DQUOT, &dqhash);
+out:
+       dq_list_unlock();
+}
+
+
+static volatile int dq_list_lock_cnt = 0;
+
+static int
+dq_list_lock(void)
+{
+       lck_mtx_lock(quota_list_mtx_lock);
+       return ++dq_list_lock_cnt;
+}
+
+static int
+dq_list_lock_changed(int oldval) {
+       return (dq_list_lock_cnt != oldval);
+}
+
+static int
+dq_list_lock_val(void) {
+       return dq_list_lock_cnt;
+}
+
+void
+dq_list_unlock(void)
+{
+       lck_mtx_unlock(quota_list_mtx_lock);
+}
+
+
+/*
+ * must be called with the quota_list_lock held
+ */
+void
+dq_lock_internal(struct dquot *dq)
+{
+        while (dq->dq_lflags & DQ_LLOCK) {
+               dq->dq_lflags |= DQ_LWANT;
+               msleep(&dq->dq_lflags, quota_list_mtx_lock, PVFS, "dq_lock_internal", NULL);
+       }
+       dq->dq_lflags |= DQ_LLOCK;
+}
+
+/*
+ * must be called with the quota_list_lock held
+ */
+void
+dq_unlock_internal(struct dquot *dq)
+{
+        int wanted = dq->dq_lflags & DQ_LWANT;
+
+       dq->dq_lflags &= ~(DQ_LLOCK | DQ_LWANT);
+
+       if (wanted)
+               wakeup(&dq->dq_lflags);
+}
+
+void
+dqlock(struct dquot *dq) {
+
+       lck_mtx_lock(quota_list_mtx_lock);
+
+       dq_lock_internal(dq);
+
+       lck_mtx_unlock(quota_list_mtx_lock);
+}
+
+void
+dqunlock(struct dquot *dq) {
+
+       lck_mtx_lock(quota_list_mtx_lock);
+
+       dq_unlock_internal(dq);
+
+       lck_mtx_unlock(quota_list_mtx_lock);
+}
+
+
+
+int
+qf_get(struct quotafile *qfp, int type)
+{
+        int error = 0;
+
+        dq_list_lock();
+  
+        switch (type) {
+
+       case QTF_OPENING:
+               while ( (qfp->qf_qflags & (QTF_OPENING | QTF_CLOSING)) ) {
+                       if ( (qfp->qf_qflags & QTF_OPENING) ) {
+                               error = EBUSY;
+                               break;
+                       }
+                       if ( (qfp->qf_qflags & QTF_CLOSING) ) {
+                               qfp->qf_qflags |= QTF_WANTED;
+                               msleep(&qfp->qf_qflags, quota_list_mtx_lock, PVFS, "qf_get", NULL);
+                       }
+               }
+               if (qfp->qf_vp != NULLVP)
+                       error = EBUSY;
+               if (error == 0)
+                       qfp->qf_qflags |= QTF_OPENING;
+               break;
+
+       case QTF_CLOSING:
+               if ( (qfp->qf_qflags & QTF_CLOSING) ) {
+                       error = EBUSY;
+                       break;
+               }
+               qfp->qf_qflags |= QTF_CLOSING;
+
+               while ( (qfp->qf_qflags & QTF_OPENING) || qfp->qf_refcnt ) {
+                       qfp->qf_qflags |= QTF_WANTED;
+                       msleep(&qfp->qf_qflags, quota_list_mtx_lock, PVFS, "qf_get", NULL);
+               }
+               if (qfp->qf_vp == NULLVP) {
+                       qfp->qf_qflags &= ~QTF_CLOSING;
+                       error = EBUSY;
+               }
+               break;
+       }
+       dq_list_unlock();
+
+       return (error);
+}
+
+void
+qf_put(struct quotafile *qfp, int type)
+{
+
+        dq_list_lock();
+
+        switch (type) {
+
+       case QTF_OPENING:
+       case QTF_CLOSING:
+               qfp->qf_qflags &= ~type;
+               break;
+       }
+       if ( (qfp->qf_qflags & QTF_WANTED) ) {
+               qfp->qf_qflags &= ~QTF_WANTED;
+               wakeup(&qfp->qf_qflags);
+       }
+       dq_list_unlock();
+}
+
+
+static void
+qf_lock(struct quotafile *qfp)
+{
+       lck_mtx_lock(&qfp->qf_lock);
+}
+
+static void
+qf_unlock(struct quotafile *qfp)
+{
+       lck_mtx_unlock(&qfp->qf_lock);
+}
+
+
+/*
+ * take a reference on the quota file while we're
+ * in dqget... this will prevent a quota_off from
+ * occurring while we're potentially playing with
+ * the quota file... the quota_off will stall until
+ * all the current references 'die'... once we start
+ * into quoto_off, all new references will be rejected
+ * we also don't want any dqgets being processed while
+ * we're in the middle of the quota_on... once we've
+ * actually got the quota file open and the associated
+ * struct quotafile inited, we can let them come through
+ *
+ * quota list lock must be held on entry
+ */
+static int
+qf_ref(struct quotafile *qfp)
+{
+        int error = 0;
+
+       if ( (qfp->qf_qflags & (QTF_OPENING | QTF_CLOSING)) || (qfp->qf_vp == NULLVP) )
+               error = EINVAL;
+       else
+               qfp->qf_refcnt++;
+
+       return (error);
+}
+
+/*
+ * drop our reference and wakeup any waiters if 
+ * we were the last one holding a ref
+ *
+ * quota list lock must be held on entry
+ */
+static void
+qf_rele(struct quotafile *qfp)
+{
+       qfp->qf_refcnt--;
+
+       if ( (qfp->qf_qflags & QTF_WANTED) && qfp->qf_refcnt == 0) {
+               qfp->qf_qflags &= ~QTF_WANTED;
+               wakeup(&qfp->qf_qflags);
+       }
+}
+
+
+void
+dqfileinit(struct quotafile *qfp)
+{
+        qfp->qf_vp = NULLVP;
+       qfp->qf_qflags = 0;
+
+       lck_mtx_init(&qfp->qf_lock, qf_lck_grp, qf_lck_attr);
 }
 
 
 /*
  * Initialize a quota file
+ *
+ * must be called with the quota file lock held
  */
 int
-dqfileopen(qfp, type)
-       struct quotafile *qfp;
-       int type;
+dqfileopen(struct quotafile *qfp, int type)
 {
        struct dqfilehdr header;
-       struct vattr vattr;
-       struct iovec aiov;
-       struct uio auio;
-       int error;
+       struct vfs_context context;
+       off_t file_size;
+       uio_t auio;
+       int error = 0;
+       char uio_buf[ UIO_SIZEOF(1) ];
 
+       context.vc_thread = current_thread();
+       context.vc_ucred = qfp->qf_cred;
+       
        /* Obtain the file size */
-       error = VOP_GETATTR(qfp->qf_vp, &vattr, qfp->qf_cred, current_proc());
-       if (error)
-               return (error);
+       if ((error = vnode_size(qfp->qf_vp, &file_size, &context)) != 0)
+               goto out;
 
        /* Read the file header */
-       auio.uio_iov = &aiov;
-       auio.uio_iovcnt = 1;
-       aiov.iov_base = (caddr_t)&header;
-       aiov.iov_len = sizeof (header);
-       auio.uio_resid = sizeof (header);
-       auio.uio_offset = (off_t)(0);
-       auio.uio_segflg = UIO_SYSSPACE;
-       auio.uio_rw = UIO_READ;
-       auio.uio_procp = (struct proc *)0;
-       error = VOP_READ(qfp->qf_vp, &auio, 0, qfp->qf_cred);
+       auio = uio_createwithbuffer(1, 0, UIO_SYSSPACE, UIO_READ, 
+                                   &uio_buf[0], sizeof(uio_buf));
+       uio_addiov(auio, CAST_USER_ADDR_T(&header), sizeof (header));
+       error = VNOP_READ(qfp->qf_vp, auio, 0, &context);
        if (error)
-               return (error);
-       else if (auio.uio_resid)
-               return (EINVAL);
-
+               goto out;
+       else if (uio_resid(auio)) {
+               error = EINVAL;
+               goto out;
+       }
        /* Sanity check the quota file header. */
-       if ((header.dqh_magic != quotamagic[type]) ||
-           (header.dqh_version > QF_VERSION) ||
-           (!powerof2(header.dqh_maxentries)) ||
-           (header.dqh_maxentries > (vattr.va_size / sizeof(struct dqblk))))
-               return (EINVAL);
-
+       if ((OSSwapBigToHostInt32(header.dqh_magic) != quotamagic[type]) ||
+           (OSSwapBigToHostInt32(header.dqh_version) > QF_VERSION) ||
+           (!powerof2(OSSwapBigToHostInt32(header.dqh_maxentries))) ||
+           (OSSwapBigToHostInt32(header.dqh_maxentries) > (file_size / sizeof(struct dqblk)))) {
+               error = EINVAL;
+               goto out;
+       }
        /* Set up the time limits for this quota. */
-       if (header.dqh_btime > 0)
-               qfp->qf_btime = header.dqh_btime;
+       if (header.dqh_btime != 0)
+               qfp->qf_btime = OSSwapBigToHostInt32(header.dqh_btime);
        else
                qfp->qf_btime = MAX_DQ_TIME;
-       if (header.dqh_itime > 0)
-               qfp->qf_itime = header.dqh_itime;
+       if (header.dqh_itime != 0)
+               qfp->qf_itime = OSSwapBigToHostInt32(header.dqh_itime);
        else
                qfp->qf_itime = MAX_IQ_TIME;
 
        /* Calculate the hash table constants. */
-       qfp->qf_maxentries = header.dqh_maxentries;
-       qfp->qf_entrycnt = header.dqh_entrycnt;
-       qfp->qf_shift = dqhashshift(header.dqh_maxentries);
-
-       return (0);
+       qfp->qf_maxentries = OSSwapBigToHostInt32(header.dqh_maxentries);
+       qfp->qf_entrycnt = OSSwapBigToHostInt32(header.dqh_entrycnt);
+       qfp->qf_shift = dqhashshift(qfp->qf_maxentries);
+out:
+       return (error);
 }
 
 /*
  * Close down a quota file
  */
 void
-dqfileclose(qfp, type)
-       struct quotafile *qfp;
-       int type;
+dqfileclose(struct quotafile *qfp, __unused int type)
 {
        struct dqfilehdr header;
-       struct iovec aiov;
-       struct uio auio;
-
-       auio.uio_iov = &aiov;
-       auio.uio_iovcnt = 1;
-       aiov.iov_base = (caddr_t)&header;
-       aiov.iov_len = sizeof (header);
-       auio.uio_resid = sizeof (header);
-       auio.uio_offset = (off_t)(0);
-       auio.uio_segflg = UIO_SYSSPACE;
-       auio.uio_rw = UIO_READ;
-       auio.uio_procp = (struct proc *)0;
-       if (VOP_READ(qfp->qf_vp, &auio, 0, qfp->qf_cred) == 0) {
-               header.dqh_entrycnt = qfp->qf_entrycnt;
-
-               auio.uio_iov = &aiov;
-               auio.uio_iovcnt = 1;
-               aiov.iov_base = (caddr_t)&header;
-               aiov.iov_len = sizeof (header);
-               auio.uio_resid = sizeof (header);
-               auio.uio_offset = (off_t)(0);
-               auio.uio_segflg = UIO_SYSSPACE;
-               auio.uio_rw = UIO_WRITE;
-               auio.uio_procp = (struct proc *)0;
-               (void) VOP_WRITE(qfp->qf_vp, &auio, 0, qfp->qf_cred);
+       struct vfs_context context;
+       uio_t auio;
+       char uio_buf[ UIO_SIZEOF(1) ];
+
+       auio = uio_createwithbuffer(1, 0, UIO_SYSSPACE, UIO_READ, 
+                                                                 &uio_buf[0], sizeof(uio_buf));
+       uio_addiov(auio, CAST_USER_ADDR_T(&header), sizeof (header));
+
+       context.vc_thread = current_thread();
+       context.vc_ucred = qfp->qf_cred;
+       
+       if (VNOP_READ(qfp->qf_vp, auio, 0, &context) == 0) {
+               header.dqh_entrycnt = OSSwapHostToBigInt32(qfp->qf_entrycnt);
+               uio_reset(auio, 0, UIO_SYSSPACE, UIO_WRITE);
+               uio_addiov(auio, CAST_USER_ADDR_T(&header), sizeof (header));
+               (void) VNOP_WRITE(qfp->qf_vp, auio, 0, &context);
        }
 }
 
@@ -206,88 +494,262 @@ dqfileclose(qfp, type)
  * reading the information from the file if necessary.
  */
 int
-dqget(vp, id, qfp, type, dqp)
-       struct vnode *vp;
-       u_long id;
-       struct quotafile *qfp;
-       register int type;
-       struct dquot **dqp;
+dqget(u_int32_t id, struct quotafile *qfp, int type, struct dquot **dqp)
 {
-       struct proc *p = current_proc();                /* XXX */
        struct dquot *dq;
+       struct dquot *ndq = NULL;
+       struct dquot *fdq = NULL;
        struct dqhash *dqh;
        struct vnode *dqvp;
        int error = 0;
+       int listlockval = 0;
 
-       dqvp = qfp->qf_vp;
-       if (id == 0 || dqvp == NULLVP || (qfp->qf_qflags & QTF_CLOSING)) {
+       if (!dqisinitialized()) {
                *dqp = NODQUOT;
                return (EINVAL);
        }
+
+       if ( id == 0 || qfp->qf_vp == NULLVP ) {
+               *dqp = NODQUOT;
+               return (EINVAL);
+       }
+       dq_list_lock();
+
+       if ( (qf_ref(qfp)) ) {
+               dq_list_unlock();
+
+               *dqp = NODQUOT;
+               return (EINVAL);
+       }
+       if ( (dqvp = qfp->qf_vp) == NULLVP ) {
+               qf_rele(qfp);
+               dq_list_unlock();
+
+               *dqp = NODQUOT;
+               return (EINVAL);
+       }
+       dqh = DQHASH(dqvp, id);
+
+relookup:
+       listlockval = dq_list_lock_val();
+
        /*
         * Check the cache first.
         */
-       dqh = DQHASH(dqvp, id);
        for (dq = dqh->lh_first; dq; dq = dq->dq_hash.le_next) {
                if (dq->dq_id != id ||
                    dq->dq_qfile->qf_vp != dqvp)
                        continue;
+
+               dq_lock_internal(dq);
+               if (dq_list_lock_changed(listlockval)) {
+                       dq_unlock_internal(dq);
+                       goto relookup;
+               }
+
+               /*
+                * dq_lock_internal may drop the quota_list_lock to msleep, so
+                * we need to re-evaluate the identity of this dq
+                */
+               if (dq->dq_id != id || dq->dq_qfile == NULL ||
+                   dq->dq_qfile->qf_vp != dqvp) {
+                       dq_unlock_internal(dq);
+                       goto relookup;
+               }
                /*
                 * Cache hit with no references.  Take
                 * the structure off the free list.
                 */
-               if (dq->dq_cnt == 0)
-                       TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
-               DQREF(dq);
+               if (dq->dq_cnt++ == 0) {
+                       if (dq->dq_flags & DQ_MOD)
+                               TAILQ_REMOVE(&dqdirtylist, dq, dq_freelist);
+                       else
+                               TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
+               }
+               dq_unlock_internal(dq);
+
+               if (fdq != NULL) {
+                       /*
+                        * we grabbed this from the free list in the first pass
+                        * but we found the dq we were looking for in
+                        * the cache the 2nd time through
+                        * so stick it back on the free list and return the cached entry
+                        */
+                       TAILQ_INSERT_HEAD(&dqfreelist, fdq, dq_freelist);
+               }
+               qf_rele(qfp);
+               dq_list_unlock();
+               
+               if (ndq != NULL) {
+                       /*
+                        * we allocated this in the first pass
+                        * but we found the dq we were looking for in
+                        * the cache the 2nd time through so free it
+                        */
+                       _FREE(ndq, M_DQUOT);
+               }
                *dqp = dq;
+
                return (0);
        }
        /*
         * Not in cache, allocate a new one.
         */
-       if (dqfreelist.tqh_first == NODQUOT &&
+       if (TAILQ_EMPTY(&dqfreelist) &&
            numdquot < MAXQUOTAS * desiredvnodes)
                desireddquot += DQUOTINC;
-       if (numdquot < desireddquot) {
-               dq = (struct dquot *)_MALLOC(sizeof *dq, M_DQUOT, M_WAITOK);
-               bzero((char *)dq, sizeof *dq);
-               numdquot++;
+
+       if (fdq != NULL) {
+               /*
+                * we captured this from the free list
+                * in the first pass through, so go
+                * ahead and use it
+                */
+               dq = fdq;
+               fdq = NULL;
+       } else if (numdquot < desireddquot) {
+               if (ndq == NULL) {
+                       /*
+                        * drop the quota list lock since MALLOC may block
+                        */
+                       dq_list_unlock();
+
+                       ndq = (struct dquot *)_MALLOC(sizeof *dq, M_DQUOT, M_WAITOK);
+                       bzero((char *)ndq, sizeof *dq);
+
+                       listlockval = dq_list_lock();
+                       /*
+                        * need to look for the entry again in the cache
+                        * since we dropped the quota list lock and
+                        * someone else may have beaten us to creating it
+                        */
+                       goto relookup;
+               } else {
+                       /*
+                        * we allocated this in the first pass through
+                        * and we're still under out target, so go
+                        * ahead and use it
+                        */
+                       dq = ndq;
+                       ndq = NULL;
+                       numdquot++;
+               }
        } else {
-               if ((dq = dqfreelist.tqh_first) == NULL) {
+               if (TAILQ_EMPTY(&dqfreelist)) {
+                       qf_rele(qfp);
+                       dq_list_unlock();
+
+                       if (ndq) {
+                               /*
+                                * we allocated this in the first pass through
+                                * but we're now at the limit of our cache size
+                                * so free it
+                                */
+                               _FREE(ndq, M_DQUOT);
+                       }
                        tablefull("dquot");
                        *dqp = NODQUOT;
                        return (EUSERS);
                }
-               if (dq->dq_cnt || (dq->dq_flags & DQ_MOD))
-                       panic("free dquot isn't");
+               dq = TAILQ_FIRST(&dqfreelist);
+
+               dq_lock_internal(dq);
+
+               if (dq_list_lock_changed(listlockval) || dq->dq_cnt || (dq->dq_flags & DQ_MOD)) {
+                       /*
+                        * we lost the race while we weren't holding
+                        * the quota list lock... dq_lock_internal
+                        * will drop it to msleep... this dq has been
+                        * reclaimed... go find another
+                        */
+                       dq_unlock_internal(dq);
+
+                       /*
+                        * need to look for the entry again in the cache
+                        * since we dropped the quota list lock and
+                        * someone else may have beaten us to creating it
+                        */
+                       goto relookup;
+               }
                TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
-               LIST_REMOVE(dq, dq_hash);
+
+               if (dq->dq_qfile != NULL) {
+                       LIST_REMOVE(dq, dq_hash);
+                       dq->dq_qfile = NULL;
+                       dq->dq_id = 0;
+               }
+               dq_unlock_internal(dq);
+
+               /*
+                * because we may have dropped the quota list lock
+                * in the call to dq_lock_internal, we need to 
+                * relookup in the hash in case someone else
+                * caused a dq with this identity to be created...
+                * if we don't find it, we'll use this one
+                */
+               fdq = dq;
+               goto relookup;
        }
+       /*
+        * we've either freshly allocated a dq
+        * or we've atomically pulled it out of
+        * the hash and freelists... no one else
+        * can have a reference, which means no
+        * one else can be trying to use this dq
+        */
+       dq_lock_internal(dq);
+       if (dq_list_lock_changed(listlockval)) {
+               dq_unlock_internal(dq);
+               goto relookup;
+       }
+
        /*
         * Initialize the contents of the dquot structure.
         */
-       if (vp != dqvp)
-               vn_lock(dqvp, LK_EXCLUSIVE | LK_RETRY, p);
-       LIST_INSERT_HEAD(dqh, dq, dq_hash);
-       DQREF(dq);
-       dq->dq_flags = DQ_LOCK;
+       dq->dq_cnt = 1;
+       dq->dq_flags = 0;
        dq->dq_id = id;
        dq->dq_qfile = qfp;
        dq->dq_type = type;
+       /*
+        * once we insert it in the hash and
+        * drop the quota_list_lock, it can be
+        * 'found'... however, we're still holding
+        * the dq_lock which will keep us from doing
+        * anything with it until we've finished
+        * initializing it...
+        */
+       LIST_INSERT_HEAD(dqh, dq, dq_hash);
+       dq_list_unlock();
+
+       if (ndq) {
+               /*
+                * we allocated this in the first pass through
+                * but we didn't need it, so free it after
+                * we've droped the quota list lock
+                */
+               _FREE(ndq, M_DQUOT);
+       }
+
        error = dqlookup(qfp, id, &dq->dq_dqb, &dq->dq_index);
 
-       if (vp != dqvp)
-               VOP_UNLOCK(dqvp, 0, p);
-       if (dq->dq_flags & DQ_WANT)
-               wakeup((caddr_t)dq);
-       dq->dq_flags = 0;
        /*
         * I/O error in reading quota file, release
         * quota structure and reflect problem to caller.
         */
        if (error) {
+               dq_list_lock();
+
+               dq->dq_id = 0;
+               dq->dq_qfile = NULL;
                LIST_REMOVE(dq, dq_hash);
-               dqrele(vp, dq);
+
+               dq_unlock_internal(dq);
+               qf_rele(qfp);
+               dq_list_unlock();
+
+               dqrele(dq);
+
                *dqp = NODQUOT;
                return (error);
        }
@@ -299,44 +761,46 @@ dqget(vp, id, qfp, type, dqp)
            dq->dq_ihardlimit == 0 && dq->dq_bhardlimit == 0)
                dq->dq_flags |= DQ_FAKE;
        if (dq->dq_id != 0) {
+               struct timeval tv;
+
+               microtime(&tv);
                if (dq->dq_btime == 0)
-                       dq->dq_btime = time.tv_sec + qfp->qf_btime;
+                       dq->dq_btime = tv.tv_sec + qfp->qf_btime;
                if (dq->dq_itime == 0)
-                       dq->dq_itime = time.tv_sec + qfp->qf_itime;
+                       dq->dq_itime = tv.tv_sec + qfp->qf_itime;
        }
+       dq_list_lock();
+       dq_unlock_internal(dq);
+       qf_rele(qfp);
+       dq_list_unlock();
+
        *dqp = dq;
        return (0);
 }
 
 /*
  * Lookup a dqblk structure for the specified identifier and
- * quota file.  If there is no enetry for this identifier then
+ * quota file.  If there is no entry for this identifier then
  * one is inserted.  The actual hash table index is returned.
  */
 static int
-dqlookup(qfp, id, dqb, index)
-       struct quotafile *qfp;
-       u_long id;
-       struct  dqblk *dqb;
-       u_int32_t *index;
+dqlookup(struct quotafile *qfp, u_int32_t id, struct dqblk *dqb, uint32_t *index)
 {
        struct vnode *dqvp;
-       struct ucred *cred;
-       struct iovec aiov;
-       struct uio auio;
+       struct vfs_context context;
+       uio_t auio;
        int i, skip, last;
-       u_long mask;
+       u_int32_t mask;
        int error = 0;
+       char uio_buf[ UIO_SIZEOF(1) ];
+
+
+       qf_lock(qfp);
 
-       if (id == 0)
-               return (EINVAL);
        dqvp = qfp->qf_vp;
-       cred = qfp->qf_cred;
 
-       auio.uio_iov = &aiov;
-       auio.uio_iovcnt = 1;
-       auio.uio_segflg = UIO_SYSSPACE;
-       auio.uio_procp = (struct proc *)0;
+       context.vc_thread = current_thread();
+       context.vc_ucred = qfp->qf_cred;
 
        mask = qfp->qf_maxentries - 1;
        i = dqhash1(id, qfp->qf_shift, mask);
@@ -345,19 +809,16 @@ dqlookup(qfp, id, dqb, index)
        for (last = (i + (qfp->qf_maxentries-1) * skip) & mask;
             i != last;
             i = (i + skip) & mask) {
-
-               aiov.iov_base = (caddr_t)dqb;
-               aiov.iov_len = sizeof (struct dqblk);
-               auio.uio_resid = sizeof (struct dqblk);
-               auio.uio_offset = (off_t)dqoffset(i);
-               auio.uio_rw = UIO_READ;
-               error = VOP_READ(dqvp, &auio, 0, cred);
+               auio = uio_createwithbuffer(1, dqoffset(i), UIO_SYSSPACE, UIO_READ, 
+                                                                         &uio_buf[0], sizeof(uio_buf));
+               uio_addiov(auio, CAST_USER_ADDR_T(dqb), sizeof (struct dqblk));
+               error = VNOP_READ(dqvp, auio, 0, &context);
                if (error) {
-                       printf("dqlookup: error %d looking up id %d at index %d\n", error, id, i);
+                       printf("dqlookup: error %d looking up id %u at index %d\n", error, id, i);
                        break;
-               } else if (auio.uio_resid) {
+               } else if (uio_resid(auio)) {
                        error = EIO;
-                       printf("dqlookup: error looking up id %d at index %d\n", id, i);
+                       printf("dqlookup: error looking up id %u at index %d\n", id, i);
                        break;
                }
                /*
@@ -367,115 +828,220 @@ dqlookup(qfp, id, dqb, index)
                 */
                if (dqb->dqb_id == 0) {
                        bzero(dqb, sizeof(struct dqblk));
-                       dqb->dqb_id = id;
+                       dqb->dqb_id = OSSwapHostToBigInt32(id);
                        /*
                         * Write back to reserve entry for this id
                         */
-                       aiov.iov_base = (caddr_t)dqb;
-                       aiov.iov_len = sizeof (struct dqblk);
-                       auio.uio_resid = sizeof (struct dqblk);
-                       auio.uio_offset = (off_t)dqoffset(i);
-                       auio.uio_rw = UIO_WRITE;
-                       error = VOP_WRITE(dqvp, &auio, 0, cred);
-                       if (auio.uio_resid && error == 0)
+                       uio_reset(auio, dqoffset(i), UIO_SYSSPACE, UIO_WRITE);
+                       uio_addiov(auio, CAST_USER_ADDR_T(dqb), sizeof (struct dqblk));
+                       error = VNOP_WRITE(dqvp, auio, 0, &context);
+                       if (uio_resid(auio) && error == 0)
                                error = EIO;
                        if (error == 0)
                                ++qfp->qf_entrycnt;
+                       dqb->dqb_id = id;
                        break;
                }
                /* An id match means an entry was found. */
-               if (dqb->dqb_id == id)
+               if (OSSwapBigToHostInt32(dqb->dqb_id) == id) {
+                       dqb->dqb_bhardlimit = OSSwapBigToHostInt64(dqb->dqb_bhardlimit);
+                       dqb->dqb_bsoftlimit = OSSwapBigToHostInt64(dqb->dqb_bsoftlimit);
+                       dqb->dqb_curbytes   = OSSwapBigToHostInt64(dqb->dqb_curbytes);
+                       dqb->dqb_ihardlimit = OSSwapBigToHostInt32(dqb->dqb_ihardlimit);
+                       dqb->dqb_isoftlimit = OSSwapBigToHostInt32(dqb->dqb_isoftlimit);
+                       dqb->dqb_curinodes  = OSSwapBigToHostInt32(dqb->dqb_curinodes);
+                       dqb->dqb_btime      = OSSwapBigToHostInt32(dqb->dqb_btime);
+                       dqb->dqb_itime      = OSSwapBigToHostInt32(dqb->dqb_itime);
+                       dqb->dqb_id         = OSSwapBigToHostInt32(dqb->dqb_id);
                        break;
+               }
        }
-       
+       qf_unlock(qfp);
+
        *index = i;  /* remember index so we don't have to recompute it later */
+
        return (error);
 }
 
+
 /*
- * Obtain a reference to a dquot.
+ * Release a reference to a dquot.
  */
 void
-dqref(dq)
-       struct dquot *dq;
+dqrele(struct dquot *dq)
 {
 
-       dq->dq_cnt++;
+       if (dq == NODQUOT)
+               return;
+       dqlock(dq);
+
+       if (dq->dq_cnt > 1) {
+               dq->dq_cnt--;
+
+               dqunlock(dq);
+               return;
+       }
+       if (dq->dq_flags & DQ_MOD)
+               (void) dqsync_locked(dq);
+       dq->dq_cnt--;
+
+       dq_list_lock();
+       TAILQ_INSERT_TAIL(&dqfreelist, dq, dq_freelist);
+        dq_unlock_internal(dq);
+       dq_list_unlock();
 }
 
 /*
- * Release a reference to a dquot.
+ * Release a reference to a dquot but don't do any I/O.
  */
 void
-dqrele(vp, dq)
-       struct vnode *vp;
-       register struct dquot *dq;
+dqreclaim(struct dquot *dq)
 {
 
        if (dq == NODQUOT)
                return;
-       if (dq->dq_cnt > 1) {
-               dq->dq_cnt--;
+
+       dq_list_lock();
+       dq_lock_internal(dq);
+
+       if (--dq->dq_cnt > 0) {
+               dq_unlock_internal(dq);
+               dq_list_unlock();
                return;
        }
        if (dq->dq_flags & DQ_MOD)
-               (void) dqsync(vp, dq);
-       if (--dq->dq_cnt > 0)
-               return;
-       TAILQ_INSERT_TAIL(&dqfreelist, dq, dq_freelist);
+               TAILQ_INSERT_TAIL(&dqdirtylist, dq, dq_freelist);
+       else
+               TAILQ_INSERT_TAIL(&dqfreelist, dq, dq_freelist);
+
+       dq_unlock_internal(dq);
+       dq_list_unlock();
 }
 
+/*
+ * Update a quota file's orphaned disk quotas.
+ */
+void
+dqsync_orphans(struct quotafile *qfp)
+{
+       struct dquot *dq;
+       
+       dq_list_lock();
+  loop:
+       TAILQ_FOREACH(dq, &dqdirtylist, dq_freelist) {
+               if (dq->dq_qfile != qfp)
+                       continue;
+
+               dq_lock_internal(dq);
+
+               if (dq->dq_qfile != qfp) {
+                       /*
+                        * the identity of this dq changed while
+                        * the quota_list_lock was dropped
+                        * dq_lock_internal can drop it to msleep
+                        */
+                       dq_unlock_internal(dq);
+                       goto loop;
+               }
+               if ((dq->dq_flags & DQ_MOD) == 0) {
+                       /*
+                        * someone cleaned and removed this from
+                        * the dq from the dirty list while the
+                        * quota_list_lock was dropped
+                        */
+                       dq_unlock_internal(dq);
+                       goto loop;
+               }
+               if (dq->dq_cnt != 0)
+                       panic("dqsync_orphans: dquot in use");
+
+               TAILQ_REMOVE(&dqdirtylist, dq, dq_freelist);
+
+               dq_list_unlock();
+               /*
+                * we're still holding the dqlock at this point
+                * with the reference count == 0
+                * we shouldn't be able
+                * to pick up another one since we hold dqlock
+                */
+               (void) dqsync_locked(dq);
+                       
+               dq_list_lock();
+
+               TAILQ_INSERT_TAIL(&dqfreelist, dq, dq_freelist);
+
+               dq_unlock_internal(dq);
+               goto loop;
+       }
+       dq_list_unlock();
+}
+
+int
+dqsync(struct dquot *dq)
+{
+        int error = 0;
+
+       if (dq != NODQUOT) {
+               dqlock(dq);
+
+               if ( (dq->dq_flags & DQ_MOD) )
+               error = dqsync_locked(dq);
+
+               dqunlock(dq);
+       }
+       return (error);
+}
+
+
 /*
  * Update the disk quota in the quota file.
  */
 int
-dqsync(vp, dq)
-       struct vnode *vp;
-       struct dquot *dq;
+dqsync_locked(struct dquot *dq)
 {
-       struct proc *p = current_proc();                /* XXX */
+       struct vfs_context context;
        struct vnode *dqvp;
-       struct iovec aiov;
-       struct uio auio;
+       struct dqblk dqb, *dqblkp;
+       uio_t auio;
        int error;
+       char uio_buf[ UIO_SIZEOF(1) ];
 
-       if (dq == NODQUOT)
-               panic("dqsync: dquot");
-       if ((dq->dq_flags & DQ_MOD) == 0)
+       if (dq->dq_id == 0) {
+               dq->dq_flags &= ~DQ_MOD;
                return (0);
-       if (dq->dq_id == 0)
-               return(0);
-       if ((dqvp = dq->dq_qfile->qf_vp) == NULLVP)
-               panic("dqsync: file");
-       if (vp != dqvp)
-               vn_lock(dqvp, LK_EXCLUSIVE | LK_RETRY, p);
-       while (dq->dq_flags & DQ_LOCK) {
-               dq->dq_flags |= DQ_WANT;
-               sleep((caddr_t)dq, PINOD+2);
-               if ((dq->dq_flags & DQ_MOD) == 0) {
-                       if (vp != dqvp)
-                               VOP_UNLOCK(dqvp, 0, p);
-                       return (0);
-               }
        }
-       dq->dq_flags |= DQ_LOCK;
-       auio.uio_iov = &aiov;
-       auio.uio_iovcnt = 1;
-       aiov.iov_base = (caddr_t)&dq->dq_dqb;
-       aiov.iov_len = sizeof (struct dqblk);
-       auio.uio_resid = sizeof (struct dqblk);
-       auio.uio_offset = (off_t)dqoffset(dq->dq_index);
-       auio.uio_segflg = UIO_SYSSPACE;
-       auio.uio_rw = UIO_WRITE;
-       auio.uio_procp = (struct proc *)0;
-       error = VOP_WRITE(dqvp, &auio, 0, dq->dq_qfile->qf_cred);
-       if (auio.uio_resid && error == 0)
+       if (dq->dq_qfile == NULL)
+               panic("dqsync: NULL dq_qfile");
+       if ((dqvp = dq->dq_qfile->qf_vp) == NULLVP)
+               panic("dqsync: NULL qf_vp");
+
+       auio = uio_createwithbuffer(1, dqoffset(dq->dq_index), UIO_SYSSPACE, 
+                                                                 UIO_WRITE, &uio_buf[0], sizeof(uio_buf));
+       uio_addiov(auio, CAST_USER_ADDR_T(&dqb), sizeof (struct dqblk));
+
+       context.vc_thread = current_thread();   /* XXX */
+       context.vc_ucred = dq->dq_qfile->qf_cred;
+
+       dqblkp = &dq->dq_dqb;
+       dqb.dqb_bhardlimit = OSSwapHostToBigInt64(dqblkp->dqb_bhardlimit);
+       dqb.dqb_bsoftlimit = OSSwapHostToBigInt64(dqblkp->dqb_bsoftlimit);
+       dqb.dqb_curbytes   = OSSwapHostToBigInt64(dqblkp->dqb_curbytes);
+       dqb.dqb_ihardlimit = OSSwapHostToBigInt32(dqblkp->dqb_ihardlimit);
+       dqb.dqb_isoftlimit = OSSwapHostToBigInt32(dqblkp->dqb_isoftlimit);
+       dqb.dqb_curinodes  = OSSwapHostToBigInt32(dqblkp->dqb_curinodes);
+       dqb.dqb_btime      = OSSwapHostToBigInt32(dqblkp->dqb_btime);
+       dqb.dqb_itime      = OSSwapHostToBigInt32(dqblkp->dqb_itime);
+       dqb.dqb_id         = OSSwapHostToBigInt32(dqblkp->dqb_id);
+       dqb.dqb_spare[0]   = 0;
+       dqb.dqb_spare[1]   = 0;
+       dqb.dqb_spare[2]   = 0;
+       dqb.dqb_spare[3]   = 0;
+
+       error = VNOP_WRITE(dqvp, auio, 0, &context);
+       if (uio_resid(auio) && error == 0)
                error = EIO;
-       if (dq->dq_flags & DQ_WANT)
-               wakeup((caddr_t)dq);
-       dq->dq_flags &= ~(DQ_MOD|DQ_LOCK|DQ_WANT);
-       if (vp != dqvp)
-               VOP_UNLOCK(dqvp, 0, p);
+       dq->dq_flags &= ~DQ_MOD;
+
        return (error);
 }
 
@@ -483,17 +1049,21 @@ dqsync(vp, dq)
  * Flush all entries from the cache for a particular vnode.
  */
 void
-dqflush(vp)
-       register struct vnode *vp;
+dqflush(struct vnode *vp)
 {
-       register struct dquot *dq, *nextdq;
+       struct dquot *dq, *nextdq;
        struct dqhash *dqh;
 
+       if (!dqisinitialized())
+               return;
+
        /*
         * Move all dquot's that used to refer to this quota
         * file off their hash chains (they will eventually
         * fall off the head of the free list and be re-used).
         */
+       dq_list_lock();
+
        for (dqh = &dqhashtbl[dqhash]; dqh >= dqhashtbl; dqh--) {
                for (dq = dqh->lh_first; dq; dq = nextdq) {
                        nextdq = dq->dq_hash.le_next;
@@ -502,7 +1072,32 @@ dqflush(vp)
                        if (dq->dq_cnt)
                                panic("dqflush: stray dquot");
                        LIST_REMOVE(dq, dq_hash);
-                       dq->dq_qfile = (struct quotafile *)0;
+                       dq->dq_qfile = NULL;
                }
        }
+       dq_list_unlock();
+}
+
+/*
+ * LP64 support for munging dqblk structure.
+ * XXX conversion of user_time_t to time_t loses precision; not an issue for 
+ * XXX us now, since we are only ever setting 32 bits worth of time into it.
+ */
+__private_extern__ void 
+munge_dqblk(struct dqblk *dqblkp, struct user_dqblk *user_dqblkp, boolean_t to64)
+{
+       if (to64) {
+               /* munge kernel (32 bit) dqblk into user (64 bit) dqblk */
+               bcopy((caddr_t)dqblkp, (caddr_t)user_dqblkp, offsetof(struct dqblk, dqb_btime));
+               user_dqblkp->dqb_id = dqblkp->dqb_id;
+               user_dqblkp->dqb_itime = dqblkp->dqb_itime;
+               user_dqblkp->dqb_btime = dqblkp->dqb_btime;
+       }
+       else {
+               /* munge user (64 bit) dqblk into kernel (32 bit) dqblk */
+               bcopy((caddr_t)user_dqblkp, (caddr_t)dqblkp, offsetof(struct dqblk, dqb_btime));
+               dqblkp->dqb_id = user_dqblkp->dqb_id;
+               dqblkp->dqb_itime = user_dqblkp->dqb_itime;     /* XXX - lose precision */
+               dqblkp->dqb_btime = user_dqblkp->dqb_btime;     /* XXX - lose precision */
+       }
 }