xnu-6153.81.5.tar.gz

[apple/xnu.git] / bsd / nfs / nfs_lock.c

diff --git a/bsd/nfs/nfs_lock.c b/bsd/nfs/nfs_lock.c
index 762c140b0ca22dc32a946c411c4a2c50f869f69a..ab5f4f4c9b9da9b189149e31d0162aeb36cf076d 100644 (file)
--- a/bsd/nfs/nfs_lock.c
+++ b/bsd/nfs/nfs_lock.c
@@ -1,23 +1,29 @@
 /*
 /*

- * Copyright (c) 2002-2005 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2002-2016 Apple Inc. All rights reserved.

  *
  *

- * @APPLE_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 Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
+ *
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ *
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ *
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER

  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  * 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.
- * 
- * @APPLE_LICENSE_HEADER_END@
+ * 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_OSREFERENCE_LICENSE_HEADER_END@

  */
 /*-
  * Copyright (c) 1997 Berkeley Software Design, Inc. All rights reserved.
  */
 /*-
  * Copyright (c) 1997 Berkeley Software Design, Inc. All rights reserved.


@@ -53,13 +59,13 @@
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/fcntl.h>
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/fcntl.h>

-#include <sys/kernel.h>                /* for hz */
+#include <sys/kernel.h>         /* for hz */

 #include <sys/file_internal.h>
 #include <sys/malloc.h>
 #include <sys/file_internal.h>
 #include <sys/malloc.h>

-#include <sys/lockf.h>         /* for hz */ /* Must come after sys/malloc.h */
+#include <sys/lockf.h>          /* for hz */ /* Must come after sys/malloc.h */

 #include <sys/kpi_mbuf.h>
 #include <sys/mount_internal.h>
 #include <sys/kpi_mbuf.h>
 #include <sys/mount_internal.h>

-#include <sys/proc_internal.h> /* for p_start */
+#include <sys/proc_internal.h>  /* for p_start */

 #include <sys/kauth.h>
 #include <sys/resourcevar.h>
 #include <sys/socket.h>
 #include <sys/kauth.h>
 #include <sys/resourcevar.h>
 #include <sys/socket.h>


@@ -68,6 +74,7 @@
 #include <sys/vnode_internal.h>
 
 #include <kern/thread.h>
 #include <sys/vnode_internal.h>
 
 #include <kern/thread.h>

+#include <kern/host.h>

 
 #include <machine/limits.h>
 
 
 #include <machine/limits.h>
 


@@ -76,52 +83,38 @@
 #include <nfs/rpcv2.h>
 #include <nfs/nfsproto.h>
 #include <nfs/nfs.h>
 #include <nfs/rpcv2.h>
 #include <nfs/nfsproto.h>
 #include <nfs/nfs.h>

+#include <nfs/nfs_gss.h>

 #include <nfs/nfsmount.h>
 #include <nfs/nfsnode.h>
 #include <nfs/nfs_lock.h>
 
 #include <nfs/nfsmount.h>
 #include <nfs/nfsnode.h>
 #include <nfs/nfs_lock.h>
 

-#define OFF_MAX QUAD_MAX
+#include <mach/host_priv.h>
+#include <mach/mig_errors.h>
+#include <mach/host_special_ports.h>
+#include <lockd/lockd_mach.h>

 
 

-/*
- * globals for managing the lockd fifo
- */
-vnode_t nfslockdvnode = 0;
-int nfslockdwaiting = 0;
-time_t nfslockdstarttimeout = 0;
-int nfslockdfifolock = 0;
-#define NFSLOCKDFIFOLOCK_LOCKED        1
-#define NFSLOCKDFIFOLOCK_WANT  2
+extern void ipc_port_release_send(ipc_port_t);

 
 /*
  * pending lock request messages are kept in this queue which is
  * kept sorted by transaction ID (xid).
  */
 
 /*
  * pending lock request messages are kept in this queue which is
  * kept sorted by transaction ID (xid).
  */

-uint64_t nfs_lockxid = 0;
-LOCKD_MSG_QUEUE nfs_pendlockq;
+static uint64_t nfs_lockxid = 0;
+static LOCKD_MSG_QUEUE nfs_pendlockq;

 
 

-/*
- * This structure is used to identify processes which have acquired NFS locks.
- * Knowing which processes have ever acquired locks allows us to short-circuit
- * unlock requests for processes that have never had an NFS file lock.  Thus
- * avoiding a costly and unnecessary lockd request.
- */
-struct nfs_lock_pid {
-       TAILQ_ENTRY(nfs_lock_pid)       lp_lru;         /* LRU list */
-       LIST_ENTRY(nfs_lock_pid)        lp_hash;        /* hash chain */
-       int                             lp_valid;       /* valid entry? */
-       int                             lp_time;        /* last time seen valid */
-       pid_t                           lp_pid;         /* The process ID. */
-       struct timeval                  lp_pid_start;   /* Start time of process id */
-};
-
-#define NFS_LOCK_PID_HASH_SIZE         64      // XXX tune me
-#define        NFS_LOCK_PID_HASH(pid)  \
-       (&nfs_lock_pid_hash_tbl[(pid) & nfs_lock_pid_hash])
-LIST_HEAD(, nfs_lock_pid) *nfs_lock_pid_hash_tbl;
-TAILQ_HEAD(, nfs_lock_pid) nfs_lock_pid_lru;
-u_long nfs_lock_pid_hash;
-int nfs_lock_pid_lock;
+/* list of mounts that are (potentially) making lockd requests */
+TAILQ_HEAD(nfs_lockd_mount_list, nfsmount) nfs_lockd_mount_list;

 
 

+static lck_grp_t *nfs_lock_lck_grp;
+static lck_mtx_t *nfs_lock_mutex;
+
+void nfs_lockdmsg_enqueue(LOCKD_MSG_REQUEST *);
+void nfs_lockdmsg_dequeue(LOCKD_MSG_REQUEST *);
+int nfs_lockdmsg_compare_to_answer(LOCKD_MSG_REQUEST *, struct lockd_ans *);
+LOCKD_MSG_REQUEST *nfs_lockdmsg_find_by_answer(struct lockd_ans *);
+LOCKD_MSG_REQUEST *nfs_lockdmsg_find_by_xid(uint64_t);
+uint64_t nfs_lockxid_get(void);
+int nfs_lockd_send_request(LOCKD_MSG *, int);

 
 /*
  * initialize global nfs lock state
 
 /*
  * initialize global nfs lock state


@@ -130,16 +123,84 @@ void
 nfs_lockinit(void)
 {
        TAILQ_INIT(&nfs_pendlockq);
 nfs_lockinit(void)
 {
        TAILQ_INIT(&nfs_pendlockq);

-       nfs_lock_pid_lock = 0;
-       nfs_lock_pid_hash_tbl = hashinit(NFS_LOCK_PID_HASH_SIZE,
-                                        M_TEMP, &nfs_lock_pid_hash);
-       TAILQ_INIT(&nfs_lock_pid_lru);
+       TAILQ_INIT(&nfs_lockd_mount_list);
+
+       nfs_lock_lck_grp = lck_grp_alloc_init("nfs_lock", LCK_GRP_ATTR_NULL);
+       nfs_lock_mutex = lck_mtx_alloc_init(nfs_lock_lck_grp, LCK_ATTR_NULL);
+}
+
+/*
+ * Register a mount as (potentially) making lockd requests.
+ */
+void
+nfs_lockd_mount_register(struct nfsmount *nmp)
+{
+       lck_mtx_lock(nfs_lock_mutex);
+       TAILQ_INSERT_HEAD(&nfs_lockd_mount_list, nmp, nm_ldlink);
+       nfs_lockd_mounts++;
+       lck_mtx_unlock(nfs_lock_mutex);
+}
+
+/*
+ * Unregister a mount as (potentially) making lockd requests.
+ *
+ * When the lockd mount count drops to zero, then send a shutdown request to
+ * lockd if we've sent any requests to it.
+ */
+void
+nfs_lockd_mount_unregister(struct nfsmount *nmp)
+{
+       int send_shutdown;
+       mach_port_t lockd_port = IPC_PORT_NULL;
+       kern_return_t kr;
+
+       lck_mtx_lock(nfs_lock_mutex);
+       if (nmp->nm_ldlink.tqe_next == NFSNOLIST) {
+               lck_mtx_unlock(nfs_lock_mutex);
+               return;
+       }
+
+       TAILQ_REMOVE(&nfs_lockd_mount_list, nmp, nm_ldlink);
+       nmp->nm_ldlink.tqe_next = NFSNOLIST;
+
+       nfs_lockd_mounts--;
+
+       /* send a shutdown request if there are no more lockd mounts */
+       send_shutdown = ((nfs_lockd_mounts == 0) && nfs_lockd_request_sent);
+       if (send_shutdown) {
+               nfs_lockd_request_sent = 0;
+       }
+
+       lck_mtx_unlock(nfs_lock_mutex);
+
+       if (!send_shutdown) {
+               return;
+       }
+
+       /*
+        * Let lockd know that it is no longer needed for any NFS mounts
+        */
+       kr = host_get_lockd_port(host_priv_self(), &lockd_port);
+       if ((kr != KERN_SUCCESS) || !IPC_PORT_VALID(lockd_port)) {
+               printf("nfs_lockd_mount_change: shutdown couldn't get port, kr %d, port %s\n",
+                   kr, (lockd_port == IPC_PORT_NULL) ? "NULL" :
+                   (lockd_port == IPC_PORT_DEAD) ? "DEAD" : "VALID");
+               return;
+       }
+
+       kr = lockd_shutdown(lockd_port);
+       if (kr != KERN_SUCCESS) {
+               printf("nfs_lockd_mount_change: shutdown %d\n", kr);
+       }
+
+       ipc_port_release_send(lockd_port);

 }
 
 /*
  * insert a lock request message into the pending queue
 }
 
 /*
  * insert a lock request message into the pending queue

+ * (nfs_lock_mutex must be held)

  */
  */

-static inline void
+void

 nfs_lockdmsg_enqueue(LOCKD_MSG_REQUEST *msgreq)
 {
        LOCKD_MSG_REQUEST *mr;
 nfs_lockdmsg_enqueue(LOCKD_MSG_REQUEST *msgreq)
 {
        LOCKD_MSG_REQUEST *mr;


@@ -163,8 +224,9 @@ nfs_lockdmsg_enqueue(LOCKD_MSG_REQUEST *msgreq)
 
 /*
  * remove a lock request message from the pending queue
 
 /*
  * remove a lock request message from the pending queue

+ * (nfs_lock_mutex must be held)

  */
  */

-static inline void
+void

 nfs_lockdmsg_dequeue(LOCKD_MSG_REQUEST *msgreq)
 {
        TAILQ_REMOVE(&nfs_pendlockq, msgreq, lmr_next);
 nfs_lockdmsg_dequeue(LOCKD_MSG_REQUEST *msgreq)
 {
        TAILQ_REMOVE(&nfs_pendlockq, msgreq, lmr_next);


@@ -179,25 +241,29 @@ nfs_lockdmsg_dequeue(LOCKD_MSG_REQUEST *msgreq)
  * However, this may not be the case if there are blocked requests.  We may
  * want to move blocked requests to a separate queue (but that'll complicate
  * duplicate xid checking).
  * However, this may not be the case if there are blocked requests.  We may
  * want to move blocked requests to a separate queue (but that'll complicate
  * duplicate xid checking).

+ *
+ * (nfs_lock_mutex must be held)

  */
  */

-static inline LOCKD_MSG_REQUEST *
+LOCKD_MSG_REQUEST *

 nfs_lockdmsg_find_by_xid(uint64_t lockxid)
 {
        LOCKD_MSG_REQUEST *mr;
 
        TAILQ_FOREACH(mr, &nfs_pendlockq, lmr_next) {
 nfs_lockdmsg_find_by_xid(uint64_t lockxid)
 {
        LOCKD_MSG_REQUEST *mr;
 
        TAILQ_FOREACH(mr, &nfs_pendlockq, lmr_next) {

-               if (mr->lmr_msg.lm_xid == lockxid)
+               if (mr->lmr_msg.lm_xid == lockxid) {

                        return mr;
                        return mr;

-               if (mr->lmr_msg.lm_xid > lockxid)
+               }
+               if (mr->lmr_msg.lm_xid > lockxid) {

                        return NULL;
                        return NULL;

+               }

        }
        return mr;
 }
 
 /*
  * Because we can't depend on nlm_granted messages containing the same
        }
        return mr;
 }
 
 /*
  * Because we can't depend on nlm_granted messages containing the same

- * cookie we sent with the original lock request, we need code test if
- * an nlm_granted answer matches the lock request.  We also need code
+ * cookie we sent with the original lock request, we need code to test
+ * if an nlm_granted answer matches the lock request.  We also need code

  * that can find a lockd message based solely on the nlm_granted answer.
  */
 
  * that can find a lockd message based solely on the nlm_granted answer.
  */
 


@@ -206,21 +272,27 @@ nfs_lockdmsg_find_by_xid(uint64_t lockxid)
  *
  * returns 0 on equality and 1 if different
  */
  *
  * returns 0 on equality and 1 if different
  */

-static inline int
+int

 nfs_lockdmsg_compare_to_answer(LOCKD_MSG_REQUEST *msgreq, struct lockd_ans *ansp)
 {
 nfs_lockdmsg_compare_to_answer(LOCKD_MSG_REQUEST *msgreq, struct lockd_ans *ansp)
 {

-       if (!(ansp->la_flags & LOCKD_ANS_LOCK_INFO))
+       if (!(ansp->la_flags & LOCKD_ANS_LOCK_INFO)) {

                return 1;
                return 1;

-       if (msgreq->lmr_msg.lm_fl.l_pid != ansp->la_pid)
+       }
+       if (msgreq->lmr_msg.lm_fl.l_pid != ansp->la_pid) {

                return 1;
                return 1;

-       if (msgreq->lmr_msg.lm_fl.l_start != ansp->la_start)
+       }
+       if (msgreq->lmr_msg.lm_fl.l_start != ansp->la_start) {

                return 1;
                return 1;

-       if (msgreq->lmr_msg.lm_fl.l_len != ansp->la_len)
+       }
+       if (msgreq->lmr_msg.lm_fl.l_len != ansp->la_len) {

                return 1;
                return 1;

-       if (msgreq->lmr_msg.lm_fh_len != ansp->la_fh_len)
+       }
+       if (msgreq->lmr_msg.lm_fh_len != ansp->la_fh_len) {

                return 1;
                return 1;

-       if (bcmp(msgreq->lmr_msg.lm_fh, ansp->la_fh, ansp->la_fh_len))
+       }
+       if (bcmp(msgreq->lmr_msg.lm_fh, ansp->la_fh, ansp->la_fh_len)) {

                return 1;
                return 1;

+       }

        return 0;
 }
 
        return 0;
 }
 


@@ -236,25 +308,30 @@ nfs_lockdmsg_compare_to_answer(LOCKD_MSG_REQUEST *msgreq, struct lockd_ans *ansp
  * However, this may not be the case if there are blocked requests.  We may
  * want to move blocked requests to a separate queue (but that'll complicate
  * duplicate xid checking).
  * However, this may not be the case if there are blocked requests.  We may
  * want to move blocked requests to a separate queue (but that'll complicate
  * duplicate xid checking).

+ *
+ * (nfs_lock_mutex must be held)

  */
  */

-static inline LOCKD_MSG_REQUEST *
+LOCKD_MSG_REQUEST *

 nfs_lockdmsg_find_by_answer(struct lockd_ans *ansp)
 {
        LOCKD_MSG_REQUEST *mr;
 
 nfs_lockdmsg_find_by_answer(struct lockd_ans *ansp)
 {
        LOCKD_MSG_REQUEST *mr;
 

-       if (!(ansp->la_flags & LOCKD_ANS_LOCK_INFO))
+       if (!(ansp->la_flags & LOCKD_ANS_LOCK_INFO)) {

                return NULL;
                return NULL;

+       }

        TAILQ_FOREACH(mr, &nfs_pendlockq, lmr_next) {
        TAILQ_FOREACH(mr, &nfs_pendlockq, lmr_next) {

-               if (!nfs_lockdmsg_compare_to_answer(mr, ansp))
+               if (!nfs_lockdmsg_compare_to_answer(mr, ansp)) {

                        break;
                        break;

+               }

        }
        return mr;
 }
 
 /*
  * return the next unique lock request transaction ID
        }
        return mr;
 }
 
 /*
  * return the next unique lock request transaction ID

+ * (nfs_lock_mutex must be held)

  */
  */

-static inline uint64_t
+uint64_t

 nfs_lockxid_get(void)
 {
        LOCKD_MSG_REQUEST *mr;
 nfs_lockxid_get(void)
 {
        LOCKD_MSG_REQUEST *mr;


@@ -275,10 +352,11 @@ nfs_lockxid_get(void)
        /* make sure we get a unique xid */
        do {
                /* Skip zero xid if it should ever happen.  */
        /* make sure we get a unique xid */
        do {
                /* Skip zero xid if it should ever happen.  */

-               if (++nfs_lockxid == 0)
+               if (++nfs_lockxid == 0) {

                        nfs_lockxid++;
                        nfs_lockxid++;

+               }

                if (!(mr = TAILQ_LAST(&nfs_pendlockq, nfs_lock_msg_queue)) ||
                if (!(mr = TAILQ_LAST(&nfs_pendlockq, nfs_lock_msg_queue)) ||

-                    (mr->lmr_msg.lm_xid < nfs_lockxid)) {
+                   (mr->lmr_msg.lm_xid < nfs_lockxid)) {

                        /* fast path: empty queue or new largest xid */
                        break;
                }
                        /* fast path: empty queue or new largest xid */
                        break;
                }


@@ -288,406 +366,115 @@ nfs_lockxid_get(void)
        return nfs_lockxid;
 }
 
        return nfs_lockxid;
 }
 

+#define MACH_MAX_TRIES 3

 
 

-/*
- * Check the nfs_lock_pid hash table for an entry and, if requested,
- * add the entry if it is not found.
- *
- * (Also, if adding, try to clean up some stale entries.)
- */
-static int
-nfs_lock_pid_check(proc_t p, int addflag, vnode_t vp)
+int
+nfs_lockd_send_request(LOCKD_MSG *msg, int interruptable)

 {
 {

-       struct nfs_lock_pid *lp, *lplru, *lplru_next;
-       proc_t plru;
-       int error = 0;
-       struct timeval now;
+       kern_return_t kr;
+       int retries = 0;
+       mach_port_t lockd_port = IPC_PORT_NULL;

 
 

-       /* lock hash */
-loop:
-       if (nfs_lock_pid_lock) {
-               struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
-               while (nfs_lock_pid_lock) {
-                       nfs_lock_pid_lock = -1;
-                       tsleep(&nfs_lock_pid_lock, PCATCH, "nfslockpid", 0);
-                       if ((error = nfs_sigintr(nmp, NULL, p)))
-                               return (error);
-               }
-               goto loop;
+       kr = host_get_lockd_port(host_priv_self(), &lockd_port);
+       if (kr != KERN_SUCCESS || !IPC_PORT_VALID(lockd_port)) {
+               return ENOTSUP;

        }
        }

-       nfs_lock_pid_lock = 1;
-
-       /* Search hash chain */
-       error = ENOENT;
-       lp = NFS_LOCK_PID_HASH(proc_pid(p))->lh_first;
-       for (; lp != NULL; lp = lp->lp_hash.le_next)
-               if (lp->lp_pid == proc_pid(p)) {
-                       /* found pid... */
-                       if (timevalcmp(&lp->lp_pid_start, &p->p_stats->p_start, ==)) {
-                               /* ...and it's valid */
-                               /* move to tail of LRU */
-                               TAILQ_REMOVE(&nfs_lock_pid_lru, lp, lp_lru);
-                               microuptime(&now);
-                               lp->lp_time = now.tv_sec;
-                               TAILQ_INSERT_TAIL(&nfs_lock_pid_lru, lp, lp_lru);
-                               error = 0;
-                               break;
-                       }
-                       /* ...but it's no longer valid */
-                       /* remove from hash, invalidate, and move to lru head */
-                       LIST_REMOVE(lp, lp_hash);
-                       lp->lp_valid = 0;
-                       TAILQ_REMOVE(&nfs_lock_pid_lru, lp, lp_lru);
-                       TAILQ_INSERT_HEAD(&nfs_lock_pid_lru, lp, lp_lru);
-                       lp = NULL;
-                       break;
-               }

 
 

-       /* if we didn't find it (valid) and we've been asked to add it */
-       if ((error == ENOENT) && addflag) {
-               /* scan lru list for invalid, stale entries to reuse/free */
-               int lrucnt = 0;
-               microuptime(&now);
-               for (lplru = TAILQ_FIRST(&nfs_lock_pid_lru); lplru; lplru = lplru_next) {
-                       lplru_next = TAILQ_NEXT(lplru, lp_lru);
-                       if (lplru->lp_valid && (lplru->lp_time >= (now.tv_sec - 2))) {
-                               /*
-                                * If the oldest LRU entry is relatively new, then don't
-                                * bother scanning any further.
-                                */
-                               break;
-                       }
-                       /* remove entry from LRU, and check if it's still in use */
-                       TAILQ_REMOVE(&nfs_lock_pid_lru, lplru, lp_lru);
-                       if (!lplru->lp_valid || !(plru = pfind(lplru->lp_pid)) ||
-                           timevalcmp(&lplru->lp_pid_start, &plru->p_stats->p_start, !=)) {
-                               /* no longer in use */
-                               LIST_REMOVE(lplru, lp_hash);
-                               if (!lp) {
-                                       /* we'll reuse this one */
-                                       lp = lplru;
-                               } else {
-                                       /* we can free this one */
-                                       FREE(lplru, M_TEMP);
-                               }
-                       } else {
-                               /* still in use */
-                               lplru->lp_time = now.tv_sec;
-                               TAILQ_INSERT_TAIL(&nfs_lock_pid_lru, lplru, lp_lru);
+       do {
+               /* In the kernel all mach messaging is interruptable */
+               do {
+                       kr = lockd_request(
+                               lockd_port,
+                               msg->lm_version,
+                               msg->lm_flags,
+                               msg->lm_xid,
+                               msg->lm_fl.l_start,
+                               msg->lm_fl.l_len,
+                               msg->lm_fl.l_pid,
+                               msg->lm_fl.l_type,
+                               msg->lm_fl.l_whence,
+                               (uint32_t *)&msg->lm_addr,
+                               (uint32_t *)&msg->lm_cred,
+                               msg->lm_fh_len,
+                               msg->lm_fh);
+                       if (kr != KERN_SUCCESS) {
+                               printf("lockd_request received %d!\n", kr);

                        }
                        }

-                       /* don't check too many entries at once */
-                       if (++lrucnt > 8)
-                               break;
-               }
-               if (!lp) {
-                       /* we need to allocate a new one */
-                       MALLOC(lp, struct nfs_lock_pid *, sizeof(struct nfs_lock_pid),
-                               M_TEMP, M_WAITOK | M_ZERO);
-               }
-               if (!lp) {
-                       error = ENOMEM;
-               } else {
-                       /* (re)initialize nfs_lock_pid info */
-                       lp->lp_pid = proc_pid(p);
-                       lp->lp_pid_start = p->p_stats->p_start;
-                       /* insert pid in hash */
-                       LIST_INSERT_HEAD(NFS_LOCK_PID_HASH(lp->lp_pid), lp, lp_hash);
-                       lp->lp_valid = 1;
-                       lp->lp_time = now.tv_sec;
-                       TAILQ_INSERT_TAIL(&nfs_lock_pid_lru, lp, lp_lru);
-                       error = 0;
-               }
-       }
+               } while (!interruptable && kr == MACH_SEND_INTERRUPTED);
+       } while (kr == MIG_SERVER_DIED && retries++ < MACH_MAX_TRIES);

 
 

-       /* unlock hash */
-       if (nfs_lock_pid_lock < 0) {
-               nfs_lock_pid_lock = 0;
-               wakeup(&nfs_lock_pid_lock);
-       } else
-               nfs_lock_pid_lock = 0;
-
-       return (error);
+       ipc_port_release_send(lockd_port);
+       switch (kr) {
+       case MACH_SEND_INTERRUPTED:
+               return EINTR;
+       default:
+               /*
+                * Other MACH or MIG errors we will retry. Eventually
+                * we will call nfs_down and allow the user to disable
+                * locking.
+                */
+               return EAGAIN;
+       }

 }
 
 }
 

-

 /*
 /*

- * nfs_advlock --
- *      NFS advisory byte-level locks.
+ * NFS advisory byte-level locks (client)

  */
 int
  */
 int

-nfs_dolock(struct vnop_advlock_args *ap)
-/* struct vnop_advlock_args {
-       struct vnodeop_desc *a_desc;
-       vnode_t a_vp;
-       caddr_t a_id;
-       int a_op;
-       struct flock *a_fl;
-       int a_flags;
-       vfs_context_t a_context;
-}; */
+nfs3_lockd_request(
+       nfsnode_t np,
+       int type,
+       LOCKD_MSG_REQUEST *msgreq,
+       int flags,
+       thread_t thd)

 {
 {

-       LOCKD_MSG_REQUEST msgreq;
-       LOCKD_MSG *msg;
-       vnode_t vp, wvp;
-       struct nfsnode *np;
-       int error, error1;
-       struct flock *fl;
-       int fmode, ioflg;
+       LOCKD_MSG *msg = &msgreq->lmr_msg;
+       int error, error2;
+       int interruptable, slpflag;

        struct nfsmount *nmp;
        struct nfsmount *nmp;

-       struct nfs_vattr nvattr;
-       off_t start, end;

        struct timeval now;
        struct timeval now;

-       int timeo, endtime, lastmsg, wentdown = 0;
-       int lockpidcheck;
-       kauth_cred_t cred;
-       proc_t p;
+       int timeo, starttime, endtime, lastmsg, wentdown = 0;
+       struct timespec ts;

        struct sockaddr *saddr;
 
        struct sockaddr *saddr;
 

-       p = vfs_context_proc(ap->a_context);
-       cred = vfs_context_ucred(ap->a_context);
-
-       vp = ap->a_vp;
-       fl = ap->a_fl;
-       np = VTONFS(vp);
-
-       nmp = VFSTONFS(vnode_mount(vp));
-       if (!nmp)
-               return (ENXIO);
-       if (nmp->nm_flag & NFSMNT_NOLOCKS)
-               return (ENOTSUP);
-
-       /*
-        * The NLM protocol doesn't allow the server to return an error
-        * on ranges, so we do it.  Pre LFS (Large File Summit)
-        * standards required EINVAL for the range errors.  More recent
-        * standards use EOVERFLOW, but their EINVAL wording still
-        * encompasses these errors.
-        * Any code sensitive to this is either:
-        *  1) written pre-LFS and so can handle only EINVAL, or
-        *  2) written post-LFS and thus ought to be tolerant of pre-LFS
-        *     implementations.
-        * Since returning EOVERFLOW certainly breaks 1), we return EINVAL.
-        */
-       if (fl->l_whence != SEEK_END) {
-               if ((fl->l_whence != SEEK_CUR && fl->l_whence != SEEK_SET) ||
-                   fl->l_start < 0 ||
-                   (fl->l_len > 0 && fl->l_len - 1 > OFF_MAX - fl->l_start) ||
-                   (fl->l_len < 0 && fl->l_start + fl->l_len < 0))
-                       return (EINVAL);
-       }
-       /*
-        * If daemon is running take a ref on its fifo vnode
-        */
-       if (!(wvp = nfslockdvnode)) {
-               if (!nfslockdwaiting && !nfslockdstarttimeout)
-                       return (ENOTSUP);
-               /*
-                * Don't wake lock daemon if it hasn't been started yet and
-                * this is an unlock request (since we couldn't possibly
-                * actually have a lock on the file).  This could be an
-                * uninformed unlock request due to closef()'s behavior of doing
-                * unlocks on all files if a process has had a lock on ANY file.
-                */
-               if (!nfslockdvnode && (fl->l_type == F_UNLCK))
-                       return (EINVAL);
-               microuptime(&now);
-               if (nfslockdwaiting) {
-                       /* wake up lock daemon */
-                       nfslockdstarttimeout = now.tv_sec + 60;
-                       (void)wakeup((void *)&nfslockdwaiting);
-               }
-               /* wait on nfslockdvnode for a while to allow daemon to start */
-               while (!nfslockdvnode && (now.tv_sec < nfslockdstarttimeout)) {
-                       error = tsleep((void *)&nfslockdvnode, PCATCH | PUSER, "lockdstart", 2*hz);
-                       if (error && (error != EWOULDBLOCK))
-                               return (error);
-                       /* check that we still have our mount... */
-                       /* ...and that we still support locks */
-                       nmp = VFSTONFS(vnode_mount(vp));
-                       if (!nmp)
-                               return (ENXIO);
-                       if (nmp->nm_flag & NFSMNT_NOLOCKS)
-                               return (ENOTSUP);
-                       if (!error)
-                               break;
-                       microuptime(&now);
-               }
-               /*
-                * check for nfslockdvnode
-                * If it hasn't started by now, there's a problem.
-                */
-               if (!(wvp = nfslockdvnode))
-                       return (ENOTSUP);
-       }
-       error = vnode_getwithref(wvp);
-       if (error)
-               return (ENOTSUP);
-       error = vnode_ref(wvp);
-       if (error) {
-               vnode_put(wvp);
-               return (ENOTSUP);
+       nmp = NFSTONMP(np);
+       if (!nmp || !nmp->nm_saddr) {
+               return ENXIO;

        }
 
        }
 

-       /*
-        * Need to check if this process has successfully acquired an NFS lock before.
-        * If not, and this is an unlock request we can simply return success here.
-        */
-       lockpidcheck = nfs_lock_pid_check(p, 0, vp);
-       if (lockpidcheck) {
-               if (lockpidcheck != ENOENT) {
-                       vnode_rele(wvp);
-                       vnode_put(wvp);
-                       return (lockpidcheck);
-               }
-               if (ap->a_op == F_UNLCK) {
-                       vnode_rele(wvp);
-                       vnode_put(wvp);
-                       return (0);
-               }
-       }
-
-       /*
-        * The NFS Lock Manager protocol doesn't directly handle
-        * negative lengths or SEEK_END, so we need to normalize
-        * things here where we have all the info.
-        * (Note: SEEK_CUR is already adjusted for at this point)
-        */
-       /* Convert the flock structure into a start and end. */
-       switch (fl->l_whence) {
-       case SEEK_SET:
-       case SEEK_CUR:
-               /*
-                * Caller is responsible for adding any necessary offset
-                * to fl->l_start when SEEK_CUR is used.
-                */
-               start = fl->l_start;
-               break;
-       case SEEK_END:
-               /* need to flush, and refetch attributes to make */
-               /* sure we have the correct end of file offset   */
-               if (np->n_flag & NMODIFIED) {
-                       NATTRINVALIDATE(np);
-                       error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1);
-                       if (error) {
-                               vnode_rele(wvp);
-                               vnode_put(wvp);
-                               return (error);
-                       }
-               }
-               NATTRINVALIDATE(np);
-
-               error = nfs_getattr(vp, &nvattr, cred, p);
-               if (error) {
-                       vnode_rele(wvp);
-                       vnode_put(wvp);
-                       return (error);
-               }
-               start = np->n_size + fl->l_start;
-               break;
-       default:
-               vnode_rele(wvp);
-               vnode_put(wvp);
-               return (EINVAL);
-       }
-       if (fl->l_len == 0)
-               end = -1;
-       else if (fl->l_len > 0)
-               end = start + fl->l_len - 1;
-       else { /* l_len is negative */
-               end = start - 1;
-               start += fl->l_len;
-       }
-       if (start < 0) {
-               vnode_rele(wvp);
-               vnode_put(wvp);
-               return (EINVAL);
-       }
-       if (!NFS_ISV3(vp) &&
-           ((start >= 0x80000000) || (end >= 0x80000000))) {
-               vnode_rele(wvp);
-               vnode_put(wvp);
-               return (EINVAL);
+       lck_mtx_lock(&nmp->nm_lock);
+       saddr = nmp->nm_saddr;
+       bcopy(saddr, &msg->lm_addr, min(sizeof msg->lm_addr, saddr->sa_len));
+       if (nmp->nm_vers == NFS_VER3) {
+               msg->lm_flags |= LOCKD_MSG_NFSV3;

        }
 
        }
 

-       /*
-        * Fill in the information structure.
-        */
-       msgreq.lmr_answered = 0;
-       msgreq.lmr_errno = 0;
-       msgreq.lmr_saved_errno = 0;
-       msg = &msgreq.lmr_msg;
-       msg->lm_version = LOCKD_MSG_VERSION;
-       msg->lm_flags = 0;
-
-       msg->lm_fl = *fl;
-       msg->lm_fl.l_start = start;
-       if (end != -1)
-               msg->lm_fl.l_len = end - start + 1;
-       msg->lm_fl.l_pid = proc_pid(p);
-
-       if (ap->a_flags & F_WAIT)
-               msg->lm_flags |= LOCKD_MSG_BLOCK;
-       if (ap->a_op == F_GETLK)
-               msg->lm_flags |= LOCKD_MSG_TEST;
-
-       nmp = VFSTONFS(vnode_mount(vp));
-       if (!nmp) {
-               vnode_rele(wvp);
-               vnode_put(wvp);
-               return (ENXIO);
+       if (nmp->nm_sotype != SOCK_DGRAM) {
+               msg->lm_flags |= LOCKD_MSG_TCP;

        }
 
        }
 

-       saddr = mbuf_data(nmp->nm_nam);
-       bcopy(saddr, &msg->lm_addr, min(sizeof msg->lm_addr, saddr->sa_len));
-       msg->lm_fh_len = NFS_ISV3(vp) ? VTONFS(vp)->n_fhsize : NFSX_V2FH;
-       bcopy(VTONFS(vp)->n_fhp, msg->lm_fh, msg->lm_fh_len);
-       if (NFS_ISV3(vp))
-               msg->lm_flags |= LOCKD_MSG_NFSV3;
-       cru2x(cred, &msg->lm_cred);
-

        microuptime(&now);
        microuptime(&now);

+       starttime = now.tv_sec;

        lastmsg = now.tv_sec - ((nmp->nm_tprintf_delay) - (nmp->nm_tprintf_initial_delay));
        lastmsg = now.tv_sec - ((nmp->nm_tprintf_delay) - (nmp->nm_tprintf_initial_delay));

+       interruptable = NMFLAG(nmp, INTR);
+       lck_mtx_unlock(&nmp->nm_lock);

 
 

-       fmode = FFLAGS(O_WRONLY);
-       if ((error = VNOP_OPEN(wvp, fmode, ap->a_context))) {
-               vnode_rele(wvp);
-               vnode_put(wvp);
-               return (error);
-       }
-       vnode_lock(wvp);
-       ++wvp->v_writecount;
-       vnode_unlock(wvp);
+       lck_mtx_lock(nfs_lock_mutex);

 
        /* allocate unique xid */
        msg->lm_xid = nfs_lockxid_get();
 
        /* allocate unique xid */
        msg->lm_xid = nfs_lockxid_get();

-       nfs_lockdmsg_enqueue(&msgreq);
-
-       timeo = 2*hz;
-#define IO_NOMACCHECK 0;
-       ioflg = IO_UNIT | IO_NOMACCHECK;
-       for (;;) {
-               error = 0;
-               while (nfslockdfifolock & NFSLOCKDFIFOLOCK_LOCKED) {
-                       nfslockdfifolock |= NFSLOCKDFIFOLOCK_WANT;
-                       error = tsleep((void *)&nfslockdfifolock,
-                                       PCATCH | PUSER, "lockdfifo", 20*hz);
-                       if (error)
-                               break;
-               }
-               if (error)
-                       break;
-               nfslockdfifolock |= NFSLOCKDFIFOLOCK_LOCKED;
+       nfs_lockdmsg_enqueue(msgreq);

 
 

-               error = vn_rdwr(UIO_WRITE, wvp, (caddr_t)msg, sizeof(*msg), 0,
-                   UIO_SYSSPACE32, ioflg, proc_ucred(kernproc), NULL, p);
+       timeo = 4;

 
 

-               nfslockdfifolock &= ~NFSLOCKDFIFOLOCK_LOCKED;
-               if (nfslockdfifolock & NFSLOCKDFIFOLOCK_WANT) {
-                       nfslockdfifolock &= ~NFSLOCKDFIFOLOCK_WANT;
-                       wakeup((void *)&nfslockdfifolock);
-               }
+       for (;;) {
+               nfs_lockd_request_sent = 1;

 
 

-               if (error && (((ioflg & IO_NDELAY) == 0) || error != EAGAIN)) {
+               /* need to drop nfs_lock_mutex while calling nfs_lockd_send_request() */
+               lck_mtx_unlock(nfs_lock_mutex);
+               error = nfs_lockd_send_request(msg, interruptable);
+               lck_mtx_lock(nfs_lock_mutex);
+               if (error && error != EAGAIN) {

                        break;
                }
 
                        break;
                }
 


@@ -700,22 +487,26 @@ nfs_dolock(struct vnop_advlock_args *ap)
                 * Retry if it takes too long to get a response.
                 *
                 * The timeout numbers were picked out of thin air... they start
                 * Retry if it takes too long to get a response.
                 *
                 * The timeout numbers were picked out of thin air... they start

-                * at 2 and double each timeout with a max of 60 seconds.
+                * at 4 and double each timeout with a max of 30 seconds.

                 *
                 * In order to maintain responsiveness, we pass a small timeout
                 *
                 * In order to maintain responsiveness, we pass a small timeout

-                * to tsleep and calculate the timeouts ourselves.  This allows
+                * to msleep and calculate the timeouts ourselves.  This allows

                 * us to pick up on mount changes quicker.
                 */
 wait_for_granted:
                error = EWOULDBLOCK;
                 * us to pick up on mount changes quicker.
                 */
 wait_for_granted:
                error = EWOULDBLOCK;

+               slpflag = (interruptable && (type != F_UNLCK)) ? PCATCH : 0;
+               ts.tv_sec = 2;
+               ts.tv_nsec = 0;

                microuptime(&now);
                microuptime(&now);

-               if ((timeo/hz) > 0)
-                       endtime = now.tv_sec + timeo/hz;
-               else
-                       endtime = now.tv_sec + 1;
+               endtime = now.tv_sec + timeo;

                while (now.tv_sec < endtime) {
                while (now.tv_sec < endtime) {

-                       error = tsleep((void *)&msgreq, PCATCH | PUSER, "lockd", 2*hz);
-                       if (msgreq.lmr_answered) {
+                       error = error2 = 0;
+                       if (!msgreq->lmr_answered) {
+                               error = msleep(msgreq, nfs_lock_mutex, slpflag | PUSER, "lockd", &ts);
+                               slpflag = 0;
+                       }
+                       if (msgreq->lmr_answered) {

                                /*
                                 * Note: it's possible to have a lock granted at
                                 * essentially the same time that we get interrupted.
                                /*
                                 * Note: it's possible to have a lock granted at
                                 * essentially the same time that we get interrupted.


@@ -723,102 +514,130 @@ wait_for_granted:
                                 * error from this request or we might not unlock the
                                 * lock that's been granted.
                                 */
                                 * error from this request or we might not unlock the
                                 * lock that's been granted.
                                 */

-                               error = 0;
+                               nmp = NFSTONMP(np);
+                               if ((msgreq->lmr_errno == ENOTSUP) && nmp &&
+                                   (nmp->nm_state & NFSSTA_LOCKSWORK)) {
+                                       /*
+                                        * We have evidence that locks work, yet lockd
+                                        * returned ENOTSUP.  This is probably because
+                                        * it was unable to contact the server's lockd
+                                        * to send it the request.
+                                        *
+                                        * Because we know locks work, we'll consider
+                                        * this failure to be a timeout.
+                                        */
+                                       error = EWOULDBLOCK;
+                               } else {
+                                       error = 0;
+                               }

                                break;
                        }
                                break;
                        }

-                       if (error != EWOULDBLOCK)
+                       if (error != EWOULDBLOCK) {

                                break;
                                break;

+                       }

                        /* check that we still have our mount... */
                        /* ...and that we still support locks */
                        /* check that we still have our mount... */
                        /* ...and that we still support locks */

-                       nmp = VFSTONFS(vnode_mount(vp));
-                       if (!nmp || (nmp->nm_flag & NFSMNT_NOLOCKS))
+                       /* ...and that there isn't a recovery pending */
+                       nmp = NFSTONMP(np);
+                       if ((error2 = nfs_sigintr(nmp, NULL, NULL, 0))) {
+                               error = error2;
+                               if (type == F_UNLCK) {
+                                       printf("nfs3_lockd_request: aborting unlock request, error %d\n", error);
+                               }

                                break;
                                break;

-                       /*
-                        * If the mount is hung and we've requested not to hang
-                        * on remote filesystems, then bail now.
-                        */
-                       if ((p != NULL) && ((proc_noremotehang(p)) != 0) &&
-                           ((nmp->nm_state & (NFSSTA_TIMEO|NFSSTA_LOCKTIMEO)) != 0)) {
-                               if (fl->l_type == F_UNLCK)
-                                       printf("nfs_dolock: aborting unlock request "
-                                           "due to timeout (noremotehang)\n");
-                               error = EIO;
+                       }
+                       lck_mtx_lock(&nmp->nm_lock);
+                       if (nmp->nm_lockmode == NFS_LOCK_MODE_DISABLED) {
+                               lck_mtx_unlock(&nmp->nm_lock);
+                               break;
+                       }
+                       if ((nmp->nm_state & NFSSTA_RECOVER) && !(flags & R_RECOVER)) {
+                               /* recovery pending... return an error that'll get this operation restarted */
+                               error = NFSERR_GRACE;
+                               lck_mtx_unlock(&nmp->nm_lock);

                                break;
                        }
                                break;
                        }

+                       interruptable = NMFLAG(nmp, INTR);
+                       lck_mtx_unlock(&nmp->nm_lock);

                        microuptime(&now);
                }
                if (error) {
                        /* check that we still have our mount... */
                        microuptime(&now);
                }
                if (error) {
                        /* check that we still have our mount... */

-                       nmp = VFSTONFS(vnode_mount(vp));
-                       if (!nmp) {
-                               if (error == EWOULDBLOCK)
-                                       error = ENXIO;
-                               break;
+                       nmp = NFSTONMP(np);
+                       if ((error2 = nfs_sigintr(nmp, NULL, NULL, 0))) {
+                               error = error2;
+                               if (error2 != EINTR) {
+                                       if (type == F_UNLCK) {
+                                               printf("nfs3_lockd_request: aborting unlock request, error %d\n", error);
+                                       }
+                                       break;
+                               }

                        }
                        /* ...and that we still support locks */
                        }
                        /* ...and that we still support locks */

-                       if (nmp->nm_flag & NFSMNT_NOLOCKS) {
-                               if (error == EWOULDBLOCK)
+                       lck_mtx_lock(&nmp->nm_lock);
+                       if (nmp->nm_lockmode == NFS_LOCK_MODE_DISABLED) {
+                               if (error == EWOULDBLOCK) {

                                        error = ENOTSUP;
                                        error = ENOTSUP;

+                               }
+                               lck_mtx_unlock(&nmp->nm_lock);

                                break;
                        }
                                break;
                        }

-                       if ((error == ENOTSUP) &&
-                           (nmp->nm_state & NFSSTA_LOCKSWORK)) {
-                               /*
-                                * We have evidence that locks work, yet lockd
-                                * returned ENOTSUP.  This is probably because
-                                * it was unable to contact the server's lockd to
-                                * send it the request.
-                                *
-                                * Because we know locks work, we'll consider
-                                * this failure to be a timeout.
-                                */
-                               error = EWOULDBLOCK;
+                       /* ...and that there isn't a recovery pending */
+                       if ((error == EWOULDBLOCK) && (nmp->nm_state & NFSSTA_RECOVER) && !(flags & R_RECOVER)) {
+                               /* recovery pending... return to allow recovery to occur */
+                               error = NFSERR_DENIED;
+                               lck_mtx_unlock(&nmp->nm_lock);
+                               break;

                        }
                        }

-                       if (error != EWOULDBLOCK) {
+                       interruptable = NMFLAG(nmp, INTR);
+                       if ((error != EWOULDBLOCK) ||
+                           ((nmp->nm_state & NFSSTA_RECOVER) && !(flags & R_RECOVER)) ||
+                           ((flags & R_RECOVER) && ((now.tv_sec - starttime) > 30))) {
+                               if ((error == EWOULDBLOCK) && (flags & R_RECOVER)) {
+                                       /* give up if this is for recovery and taking too long */
+                                       error = ETIMEDOUT;
+                               } else if ((nmp->nm_state & NFSSTA_RECOVER) && !(flags & R_RECOVER)) {
+                                       /* recovery pending... return an error that'll get this operation restarted */
+                                       error = NFSERR_GRACE;
+                               }
+                               lck_mtx_unlock(&nmp->nm_lock);

                                /*
                                 * We're going to bail on this request.
                                 * If we were a blocked lock request, send a cancel.
                                 */
                                /*
                                 * We're going to bail on this request.
                                 * If we were a blocked lock request, send a cancel.
                                 */

-                               if ((msgreq.lmr_errno == EINPROGRESS) &&
+                               if ((msgreq->lmr_errno == EINPROGRESS) &&

                                    !(msg->lm_flags & LOCKD_MSG_CANCEL)) {
                                        /* set this request up as a cancel */
                                        msg->lm_flags |= LOCKD_MSG_CANCEL;
                                    !(msg->lm_flags & LOCKD_MSG_CANCEL)) {
                                        /* set this request up as a cancel */
                                        msg->lm_flags |= LOCKD_MSG_CANCEL;

-                                       nfs_lockdmsg_dequeue(&msgreq);
+                                       nfs_lockdmsg_dequeue(msgreq);

                                        msg->lm_xid = nfs_lockxid_get();
                                        msg->lm_xid = nfs_lockxid_get();

-                                       nfs_lockdmsg_enqueue(&msgreq);
-                                       msgreq.lmr_saved_errno = error;
-                                       msgreq.lmr_errno = 0;
-                                       msgreq.lmr_answered = 0;
+                                       nfs_lockdmsg_enqueue(msgreq);
+                                       msgreq->lmr_saved_errno = error;
+                                       msgreq->lmr_errno = 0;
+                                       msgreq->lmr_answered = 0;

                                        /* reset timeout */
                                        /* reset timeout */

-                                       timeo = 2*hz;
+                                       timeo = 2;

                                        /* send cancel request */
                                        continue;
                                }
                                break;
                        }
 
                                        /* send cancel request */
                                        continue;
                                }
                                break;
                        }
 

-                       /*
-                        * If the mount is hung and we've requested not to hang
-                        * on remote filesystems, then bail now.
-                        */
-                       if ((p != NULL) && ((proc_noremotehang(p)) != 0) &&
-                           ((nmp->nm_state & (NFSSTA_TIMEO|NFSSTA_LOCKTIMEO)) != 0)) {
-                               if (fl->l_type == F_UNLCK)
-                                       printf("nfs_dolock: aborting unlock request "
-                                           "due to timeout (noremotehang)\n");
-                               error = EIO;
-                               break;
-                       }

                        /* warn if we're not getting any response */
                        microuptime(&now);
                        /* warn if we're not getting any response */
                        microuptime(&now);

-                       if ((msgreq.lmr_errno != EINPROGRESS) &&
+                       if ((msgreq->lmr_errno != EINPROGRESS) &&
+                           !(msg->lm_flags & LOCKD_MSG_DENIED_GRACE) &&

                            (nmp->nm_tprintf_initial_delay != 0) &&
                            ((lastmsg + nmp->nm_tprintf_delay) < now.tv_sec)) {
                            (nmp->nm_tprintf_initial_delay != 0) &&
                            ((lastmsg + nmp->nm_tprintf_delay) < now.tv_sec)) {

+                               lck_mtx_unlock(&nmp->nm_lock);

                                lastmsg = now.tv_sec;
                                lastmsg = now.tv_sec;

-                               nfs_down(nmp, p, 0, NFSSTA_LOCKTIMEO, "lockd not responding");
+                               nfs_down(nmp, thd, 0, NFSSTA_LOCKTIMEO, "lockd not responding", 1);

                                wentdown = 1;
                                wentdown = 1;

+                       } else {
+                               lck_mtx_unlock(&nmp->nm_lock);

                        }
                        }

-                       if (msgreq.lmr_errno == EINPROGRESS) {
+
+                       if (msgreq->lmr_errno == EINPROGRESS) {

                                /*
                                 * We've got a blocked lock request that we are
                                 * going to retry.  First, we'll want to try to
                                /*
                                 * We've got a blocked lock request that we are
                                 * going to retry.  First, we'll want to try to


@@ -831,70 +650,65 @@ wait_for_granted:
                                 * it is NLM_BLOCKED).
                                 */
                                msg->lm_flags |= LOCKD_MSG_CANCEL;
                                 * it is NLM_BLOCKED).
                                 */
                                msg->lm_flags |= LOCKD_MSG_CANCEL;

-                               nfs_lockdmsg_dequeue(&msgreq);
+                               nfs_lockdmsg_dequeue(msgreq);

                                msg->lm_xid = nfs_lockxid_get();
                                msg->lm_xid = nfs_lockxid_get();

-                               nfs_lockdmsg_enqueue(&msgreq);
-                               msgreq.lmr_saved_errno = msgreq.lmr_errno;
-                               msgreq.lmr_errno = 0;
-                               msgreq.lmr_answered = 0;
-                               timeo = 2*hz;
+                               nfs_lockdmsg_enqueue(msgreq);
+                               msgreq->lmr_saved_errno = msgreq->lmr_errno;
+                               msgreq->lmr_errno = 0;
+                               msgreq->lmr_answered = 0;
+                               timeo = 2;

                                /* send cancel then resend request */
                                continue;
                        }
                                /* send cancel then resend request */
                                continue;
                        }

+

                        /*
                        /*

-                        * We timed out, so we will rewrite the request
-                        * to the fifo, but only if it isn't already full.
+                        * We timed out, so we will resend the request.

                         */
                         */

-                       ioflg |= IO_NDELAY;
-                       timeo *= 2;
-                       if (timeo > 60*hz)
-                               timeo = 60*hz;
+                       if (!(flags & R_RECOVER)) {
+                               timeo *= 2;
+                       }
+                       if (timeo > 30) {
+                               timeo = 30;
+                       }

                        /* resend request */
                        continue;
                }
 
                /* we got a reponse, so the server's lockd is OK */
                        /* resend request */
                        continue;
                }
 
                /* we got a reponse, so the server's lockd is OK */

-               nfs_up(VFSTONFS(vnode_mount(vp)), p, NFSSTA_LOCKTIMEO,
-                       wentdown ? "lockd alive again" : NULL);
+               nfs_up(NFSTONMP(np), thd, NFSSTA_LOCKTIMEO,
+                   wentdown ? "lockd alive again" : NULL);

                wentdown = 0;
 
                wentdown = 0;
 

-               if (msgreq.lmr_errno == EINPROGRESS) {
+               if (msgreq->lmr_answered && (msg->lm_flags & LOCKD_MSG_DENIED_GRACE)) {
+                       /*
+                        * The lock request was denied because the server lockd is
+                        * still in its grace period.  So, we need to try the
+                        * request again in a little bit.  Return the GRACE error so
+                        * the higher levels can perform the retry.
+                        */
+                       msgreq->lmr_saved_errno = msgreq->lmr_errno = error = NFSERR_GRACE;
+               }
+
+               if (msgreq->lmr_errno == EINPROGRESS) {

                        /* got NLM_BLOCKED response */
                        /* need to wait for NLM_GRANTED */
                        /* got NLM_BLOCKED response */
                        /* need to wait for NLM_GRANTED */

-                       timeo = 60*hz;
-                       msgreq.lmr_answered = 0;
+                       timeo = 30;
+                       msgreq->lmr_answered = 0;

                        goto wait_for_granted;
                }
 
                if ((msg->lm_flags & LOCKD_MSG_CANCEL) &&
                        goto wait_for_granted;
                }
 
                if ((msg->lm_flags & LOCKD_MSG_CANCEL) &&

-                   (msgreq.lmr_saved_errno == EINPROGRESS)) {
+                   (msgreq->lmr_saved_errno == EINPROGRESS)) {

                        /*
                         * We just got a successful reply to the
                         * cancel of the previous blocked lock request.
                        /*
                         * We just got a successful reply to the
                         * cancel of the previous blocked lock request.

-                        * Now, go ahead and resend the request.
+                        * Now, go ahead and return a DENIED error so the
+                        * higher levels can resend the request.

                         */
                        msg->lm_flags &= ~LOCKD_MSG_CANCEL;
                         */
                        msg->lm_flags &= ~LOCKD_MSG_CANCEL;

-                       nfs_lockdmsg_dequeue(&msgreq);
-                       msg->lm_xid = nfs_lockxid_get();
-                       nfs_lockdmsg_enqueue(&msgreq);
-                       msgreq.lmr_saved_errno = 0;
-                       msgreq.lmr_errno = 0;
-                       msgreq.lmr_answered = 0;
-                       timeo = 2*hz;
-                       /* resend request */
-                       continue;
-               }
-
-               if ((msg->lm_flags & LOCKD_MSG_TEST) && msgreq.lmr_errno == 0) {
-                       if (msg->lm_fl.l_type != F_UNLCK) {
-                               fl->l_type = msg->lm_fl.l_type;
-                               fl->l_pid = msg->lm_fl.l_pid;
-                               fl->l_start = msg->lm_fl.l_start;
-                               fl->l_len = msg->lm_fl.l_len;
-                               fl->l_whence = SEEK_SET;
-                       } else {
-                               fl->l_type = F_UNLCK;
-                       }
+                       error = NFSERR_DENIED;
+                       /* Will dequeue msgreq after the following break at the end of this routine */
+                       break;

                }
 
                /*
                }
 
                /*


@@ -904,35 +718,202 @@ wait_for_granted:
                 */
                if (msg->lm_flags & LOCKD_MSG_CANCEL) {
                        msg->lm_flags &= ~LOCKD_MSG_CANCEL;
                 */
                if (msg->lm_flags & LOCKD_MSG_CANCEL) {
                        msg->lm_flags &= ~LOCKD_MSG_CANCEL;

-                       error = msgreq.lmr_saved_errno;
-               } else
-                       error = msgreq.lmr_errno;
+                       error = msgreq->lmr_saved_errno;
+               } else {
+                       error = msgreq->lmr_errno;
+               }

 
 

-               if (!error) {
-                       /* record that NFS file locking has worked on this mount */
-                       nmp = VFSTONFS(vnode_mount(vp));
-                       if (nmp && !(nmp->nm_state & NFSSTA_LOCKSWORK))
-                               nmp->nm_state |= NFSSTA_LOCKSWORK;
+               nmp = NFSTONMP(np);
+               if ((error == ENOTSUP) && nmp && !(nmp->nm_state & NFSSTA_LOCKSWORK)) {

                        /*
                        /*

-                        * If we successfully acquired a lock, make sure this pid
-                        * is in the nfs_lock_pid hash table so we know we can't
-                        * short-circuit unlock requests.
+                        * We have NO evidence that locks work and lockd
+                        * returned ENOTSUP.  Let's take this as a hint
+                        * that locks aren't supported and disable them
+                        * for this mount.

                         */
                         */

-                       if ((lockpidcheck == ENOENT) &&
-                           ((ap->a_op == F_SETLK) || (ap->a_op == F_SETLKW)))
-                               nfs_lock_pid_check(p, 1, vp);
-       
+                       nfs_lockdmsg_dequeue(msgreq);
+                       lck_mtx_unlock(nfs_lock_mutex);
+                       lck_mtx_lock(&nmp->nm_lock);
+                       if (nmp->nm_lockmode == NFS_LOCK_MODE_ENABLED) {
+                               nmp->nm_lockmode = NFS_LOCK_MODE_DISABLED;
+                               nfs_lockd_mount_unregister(nmp);
+                       }
+                       nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
+                       lck_mtx_unlock(&nmp->nm_lock);
+                       printf("lockd returned ENOTSUP, disabling locks for nfs server: %s\n",
+                           vfs_statfs(nmp->nm_mountp)->f_mntfromname);
+                       return error;
+               }
+               if (!error) {
+                       /* record that NFS file locking has worked on this mount */
+                       if (nmp) {
+                               lck_mtx_lock(&nmp->nm_lock);
+                               if (!(nmp->nm_state & NFSSTA_LOCKSWORK)) {
+                                       nmp->nm_state |= NFSSTA_LOCKSWORK;
+                               }
+                               lck_mtx_unlock(&nmp->nm_lock);
+                       }

                }
                break;
        }
 
                }
                break;
        }
 

-       nfs_lockdmsg_dequeue(&msgreq);
+       nfs_lockdmsg_dequeue(msgreq);
+
+       lck_mtx_unlock(nfs_lock_mutex);
+
+       return error;
+}
+
+/*
+ * Send an NLM LOCK message to the server
+ */
+int
+nfs3_setlock_rpc(
+       nfsnode_t np,
+       struct nfs_open_file *nofp,
+       struct nfs_file_lock *nflp,
+       int reclaim,
+       int flags,
+       thread_t thd,
+       kauth_cred_t cred)
+{
+       struct nfs_lock_owner *nlop = nflp->nfl_owner;
+       struct nfsmount *nmp;
+       int error;
+       LOCKD_MSG_REQUEST msgreq;
+       LOCKD_MSG *msg;
+
+       nmp = NFSTONMP(np);
+       if (nfs_mount_gone(nmp)) {
+               return ENXIO;
+       }
+
+       if (!nlop->nlo_open_owner) {
+               nfs_open_owner_ref(nofp->nof_owner);
+               nlop->nlo_open_owner = nofp->nof_owner;
+       }
+       if ((error = nfs_lock_owner_set_busy(nlop, thd))) {
+               return error;
+       }
+
+       /* set up lock message request structure */
+       bzero(&msgreq, sizeof(msgreq));
+       msg = &msgreq.lmr_msg;
+       msg->lm_version = LOCKD_MSG_VERSION;
+       if ((nflp->nfl_flags & NFS_FILE_LOCK_WAIT) && !reclaim) {
+               msg->lm_flags |= LOCKD_MSG_BLOCK;
+       }
+       if (reclaim) {
+               msg->lm_flags |= LOCKD_MSG_RECLAIM;
+       }
+       msg->lm_fh_len = (nmp->nm_vers == NFS_VER2) ? NFSX_V2FH : np->n_fhsize;
+       bcopy(np->n_fhp, msg->lm_fh, msg->lm_fh_len);
+       cru2x(cred, &msg->lm_cred);
+
+       msg->lm_fl.l_whence = SEEK_SET;
+       msg->lm_fl.l_start = nflp->nfl_start;
+       msg->lm_fl.l_len = NFS_FLOCK_LENGTH(nflp->nfl_start, nflp->nfl_end);
+       msg->lm_fl.l_type = nflp->nfl_type;
+       msg->lm_fl.l_pid = nlop->nlo_pid;
+
+       error = nfs3_lockd_request(np, 0, &msgreq, flags, thd);

 
 

-       error1 = VNOP_CLOSE(wvp, FWRITE, ap->a_context);
-       vnode_rele(wvp);
-       vnode_put(wvp);
-       /* prefer any previous 'error' to our vn_close 'error1'. */
-       return (error != 0 ? error : error1);
+       nfs_lock_owner_clear_busy(nlop);
+       return error;
+}
+
+/*
+ * Send an NLM UNLOCK message to the server
+ */
+int
+nfs3_unlock_rpc(
+       nfsnode_t np,
+       struct nfs_lock_owner *nlop,
+       __unused int type,
+       uint64_t start,
+       uint64_t end,
+       int flags,
+       thread_t thd,
+       kauth_cred_t cred)
+{
+       struct nfsmount *nmp;
+       LOCKD_MSG_REQUEST msgreq;
+       LOCKD_MSG *msg;
+
+       nmp = NFSTONMP(np);
+       if (!nmp) {
+               return ENXIO;
+       }
+
+       /* set up lock message request structure */
+       bzero(&msgreq, sizeof(msgreq));
+       msg = &msgreq.lmr_msg;
+       msg->lm_version = LOCKD_MSG_VERSION;
+       msg->lm_fh_len = (nmp->nm_vers == NFS_VER2) ? NFSX_V2FH : np->n_fhsize;
+       bcopy(np->n_fhp, msg->lm_fh, msg->lm_fh_len);
+       cru2x(cred, &msg->lm_cred);
+
+       msg->lm_fl.l_whence = SEEK_SET;
+       msg->lm_fl.l_start = start;
+       msg->lm_fl.l_len = NFS_FLOCK_LENGTH(start, end);
+       msg->lm_fl.l_type = F_UNLCK;
+       msg->lm_fl.l_pid = nlop->nlo_pid;
+
+       return nfs3_lockd_request(np, F_UNLCK, &msgreq, flags, thd);
+}
+
+/*
+ * Send an NLM LOCK TEST message to the server
+ */
+int
+nfs3_getlock_rpc(
+       nfsnode_t np,
+       struct nfs_lock_owner *nlop,
+       struct flock *fl,
+       uint64_t start,
+       uint64_t end,
+       vfs_context_t ctx)
+{
+       struct nfsmount *nmp;
+       int error;
+       LOCKD_MSG_REQUEST msgreq;
+       LOCKD_MSG *msg;
+
+       nmp = NFSTONMP(np);
+       if (nfs_mount_gone(nmp)) {
+               return ENXIO;
+       }
+
+       /* set up lock message request structure */
+       bzero(&msgreq, sizeof(msgreq));
+       msg = &msgreq.lmr_msg;
+       msg->lm_version = LOCKD_MSG_VERSION;
+       msg->lm_flags |= LOCKD_MSG_TEST;
+       msg->lm_fh_len = (nmp->nm_vers == NFS_VER2) ? NFSX_V2FH : np->n_fhsize;
+       bcopy(np->n_fhp, msg->lm_fh, msg->lm_fh_len);
+       cru2x(vfs_context_ucred(ctx), &msg->lm_cred);
+
+       msg->lm_fl.l_whence = SEEK_SET;
+       msg->lm_fl.l_start = start;
+       msg->lm_fl.l_len = NFS_FLOCK_LENGTH(start, end);
+       msg->lm_fl.l_type = fl->l_type;
+       msg->lm_fl.l_pid = nlop->nlo_pid;
+
+       error = nfs3_lockd_request(np, 0, &msgreq, 0, vfs_context_thread(ctx));
+
+       if (!error && (msg->lm_flags & LOCKD_MSG_TEST) && !msgreq.lmr_errno) {
+               if (msg->lm_fl.l_type != F_UNLCK) {
+                       fl->l_type = msg->lm_fl.l_type;
+                       fl->l_pid = msg->lm_fl.l_pid;
+                       fl->l_start = msg->lm_fl.l_start;
+                       fl->l_len = msg->lm_fl.l_len;
+                       fl->l_whence = SEEK_SET;
+               } else {
+                       fl->l_type = F_UNLCK;
+               }
+       }
+
+       return error;

 }
 
 /*
 }
 
 /*


@@ -947,12 +928,16 @@ nfslockdans(proc_t p, struct lockd_ans *ansp)
 
        /* Let root make this call. */
        error = proc_suser(p);
 
        /* Let root make this call. */
        error = proc_suser(p);

-       if (error)
-               return (error);
+       if (error) {
+               return error;
+       }

 
        /* the version should match, or we're out of sync */
 
        /* the version should match, or we're out of sync */

-       if (ansp->la_version != LOCKD_ANS_VERSION)
-               return (EINVAL);
+       if (ansp->la_version != LOCKD_ANS_VERSION) {
+               return EINVAL;
+       }
+
+       lck_mtx_lock(nfs_lock_mutex);

 
        /* try to find the lockd message by transaction id (cookie) */
        msgreq = nfs_lockdmsg_find_by_xid(ansp->la_xid);
 
        /* try to find the lockd message by transaction id (cookie) */
        msgreq = nfs_lockdmsg_find_by_xid(ansp->la_xid);


@@ -963,25 +948,30 @@ nfslockdans(proc_t p, struct lockd_ans *ansp)
                 * If no message was found or it doesn't match the answer,
                 * we look for the lockd message by the answer's lock info.
                 */
                 * If no message was found or it doesn't match the answer,
                 * we look for the lockd message by the answer's lock info.
                 */

-               if (!msgreq || nfs_lockdmsg_compare_to_answer(msgreq, ansp))
+               if (!msgreq || nfs_lockdmsg_compare_to_answer(msgreq, ansp)) {

                        msgreq = nfs_lockdmsg_find_by_answer(ansp);
                        msgreq = nfs_lockdmsg_find_by_answer(ansp);

+               }

                /*
                 * We need to make sure this request isn't being cancelled
                 * If it is, we don't want to accept the granted message.
                 */
                /*
                 * We need to make sure this request isn't being cancelled
                 * If it is, we don't want to accept the granted message.
                 */

-               if (msgreq && (msgreq->lmr_msg.lm_flags & LOCKD_MSG_CANCEL))
+               if (msgreq && (msgreq->lmr_msg.lm_flags & LOCKD_MSG_CANCEL)) {

                        msgreq = NULL;
                        msgreq = NULL;

+               }
+       }
+       if (!msgreq) {
+               lck_mtx_unlock(nfs_lock_mutex);
+               return EPIPE;

        }
        }

-       if (!msgreq)
-               return (EPIPE);

 
        msgreq->lmr_errno = ansp->la_errno;
        if ((msgreq->lmr_msg.lm_flags & LOCKD_MSG_TEST) && msgreq->lmr_errno == 0) {
                if (ansp->la_flags & LOCKD_ANS_LOCK_INFO) {
 
        msgreq->lmr_errno = ansp->la_errno;
        if ((msgreq->lmr_msg.lm_flags & LOCKD_MSG_TEST) && msgreq->lmr_errno == 0) {
                if (ansp->la_flags & LOCKD_ANS_LOCK_INFO) {

-                       if (ansp->la_flags & LOCKD_ANS_LOCK_EXCL)
+                       if (ansp->la_flags & LOCKD_ANS_LOCK_EXCL) {

                                msgreq->lmr_msg.lm_fl.l_type = F_WRLCK;
                                msgreq->lmr_msg.lm_fl.l_type = F_WRLCK;

-                       else
+                       } else {

                                msgreq->lmr_msg.lm_fl.l_type = F_RDLCK;
                                msgreq->lmr_msg.lm_fl.l_type = F_RDLCK;

+                       }

                        msgreq->lmr_msg.lm_fl.l_pid = ansp->la_pid;
                        msgreq->lmr_msg.lm_fl.l_start = ansp->la_start;
                        msgreq->lmr_msg.lm_fl.l_len = ansp->la_len;
                        msgreq->lmr_msg.lm_fl.l_pid = ansp->la_pid;
                        msgreq->lmr_msg.lm_fl.l_start = ansp->la_start;
                        msgreq->lmr_msg.lm_fl.l_len = ansp->la_len;


@@ -989,72 +979,74 @@ nfslockdans(proc_t p, struct lockd_ans *ansp)
                        msgreq->lmr_msg.lm_fl.l_type = F_UNLCK;
                }
        }
                        msgreq->lmr_msg.lm_fl.l_type = F_UNLCK;
                }
        }

+       if (ansp->la_flags & LOCKD_ANS_DENIED_GRACE) {
+               msgreq->lmr_msg.lm_flags |= LOCKD_MSG_DENIED_GRACE;
+       }

 
        msgreq->lmr_answered = 1;
 
        msgreq->lmr_answered = 1;

-       (void)wakeup((void *)msgreq);
+       lck_mtx_unlock(nfs_lock_mutex);
+       wakeup(msgreq);

 
 

-       return (0);
+       return 0;

 }
 
 /*
 }
 
 /*

- * nfslockdfd --
- *      NFS advisory byte-level locks: fifo file# from the lock daemon.
+ * nfslockdnotify --
+ *      NFS host restart notification from the lock daemon.
+ *
+ * Used to initiate reclaiming of held locks when a server we
+ * have mounted reboots.

  */
 int
  */
 int

-nfslockdfd(proc_t p, int fd)
+nfslockdnotify(proc_t p, user_addr_t argp)

 {
 {

-       int error;
-       vnode_t vp, oldvp;
+       int error, i, headsize;
+       struct lockd_notify ln;
+       struct nfsmount *nmp;
+       struct sockaddr *saddr;

 
 

+       /* Let root make this call. */

        error = proc_suser(p);
        error = proc_suser(p);

-       if (error)
-               return (error);
-       if (fd < 0) {
-               vp = NULL;
-       } else {
-               error = file_vnode(fd, &vp);
-               if (error)
-                       return (error);
-               error = vnode_getwithref(vp);
-               if (error)
-                       return (error);
-               error = vnode_ref(vp);
-               if (error) {
-                       vnode_put(vp);
-                       return (error);
-               }
+       if (error) {
+               return error;

        }
        }

-       oldvp = nfslockdvnode;
-       nfslockdvnode = vp;
-       if (oldvp) {
-               vnode_rele(oldvp);
+
+       headsize = (char*)&ln.ln_addr[0] - (char*)&ln.ln_version;
+       error = copyin(argp, &ln, headsize);
+       if (error) {
+               return error;

        }
        }

-       (void)wakeup((void *)&nfslockdvnode);
-       if (vp) {
-               vnode_put(vp);
+       if (ln.ln_version != LOCKD_NOTIFY_VERSION) {
+               return EINVAL;

        }
        }

-       return (0);
-}
+       if ((ln.ln_addrcount < 1) || (ln.ln_addrcount > 128)) {
+               return EINVAL;
+       }
+       argp += headsize;
+       saddr = (struct sockaddr *)&ln.ln_addr[0];

 
 

-/*
- * nfslockdwait --
- *      lock daemon waiting for lock request
- */
-int
-nfslockdwait(proc_t p)
-{
-       int error;
+       lck_mtx_lock(nfs_lock_mutex);

 
 

-       error = proc_suser(p);
-       if (error)
-               return (error);
-       if (nfslockdwaiting || nfslockdvnode)
-               return (EBUSY);
+       for (i = 0; i < ln.ln_addrcount; i++) {
+               error = copyin(argp, &ln.ln_addr[0], sizeof(ln.ln_addr[0]));
+               if (error) {
+                       break;
+               }
+               argp += sizeof(ln.ln_addr[0]);
+               /* scan lockd mount list for match to this address */
+               TAILQ_FOREACH(nmp, &nfs_lockd_mount_list, nm_ldlink) {
+                       /* check if address matches this mount's server address */
+                       if (!nmp->nm_saddr || nfs_sockaddr_cmp(saddr, nmp->nm_saddr)) {
+                               continue;
+                       }
+                       /* We have a match!  Mark it as needing recovery. */
+                       lck_mtx_lock(&nmp->nm_lock);
+                       nfs_need_recover(nmp, 0);
+                       lck_mtx_unlock(&nmp->nm_lock);
+               }
+       }

 
 

-       nfslockdstarttimeout = 0;
-       nfslockdwaiting = 1;
-       tsleep((void *)&nfslockdwaiting, PCATCH | PUSER, "lockd", 0);
-       nfslockdwaiting = 0;
+       lck_mtx_unlock(nfs_lock_mutex);

 
 

-       return (0);
+       return error;

 }
 }