smbd: Add debugs to brlock.c
[mat/samba.git] / source3 / locking / brlock.c
index be2948c53113fd51218595c0d9e80b3ff02c6f7e..b5eebc8e0401984d1d618c99e20585b08690ad2f 100644 (file)
@@ -1,21 +1,21 @@
-/* 
+/*
    Unix SMB/CIFS implementation.
    byte range locking code
    Updated to handle range splits/merges.
 
    Copyright (C) Andrew Tridgell 1992-2000
    Copyright (C) Jeremy Allison 1992-2000
-   
+
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation; either version 3 of the License, or
    (at your option) any later version.
-   
+
    This program is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.
-   
+
    You should have received a copy of the GNU General Public License
    along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
    used. This allows us to provide the same semantics as NT */
 
 #include "includes.h"
+#include "system/filesys.h"
+#include "locking/proto.h"
+#include "smbd/globals.h"
+#include "dbwrap/dbwrap.h"
+#include "dbwrap/dbwrap_open.h"
+#include "serverid.h"
+#include "messages.h"
+#include "util_tdb.h"
 
 #undef DBGC_CLASS
 #define DBGC_CLASS DBGC_LOCKING
 
 static struct db_context *brlock_db;
 
+struct byte_range_lock {
+       struct files_struct *fsp;
+       unsigned int num_locks;
+       bool modified;
+       bool have_read_oplocks;
+       struct lock_struct *lock_data;
+       struct db_record *record;
+};
+
 /****************************************************************************
  Debug info at level 10 for lock struct.
 ****************************************************************************/
 
-static void print_lock_struct(unsigned int i, struct lock_struct *pls)
+static void print_lock_struct(unsigned int i, const struct lock_struct *pls)
 {
-       DEBUG(10,("[%u]: smbpid = %u, tid = %u, pid = %s, ",
+       DEBUG(10,("[%u]: smblctx = %llu, tid = %u, pid = %s, ",
                        i,
-                       (unsigned int)pls->context.smbpid,
+                       (unsigned long long)pls->context.smblctx,
                        (unsigned int)pls->context.tid,
-                       procid_str(debug_ctx(), &pls->context.pid) ));
-       
-       DEBUG(10,("start = %.0f, size = %.0f, fnum = %d, %s %s\n",
+                       server_id_str(talloc_tos(), &pls->context.pid) ));
+
+       DEBUG(10,("start = %.0f, size = %.0f, fnum = %llu, %s %s\n",
                (double)pls->start,
                (double)pls->size,
-               pls->fnum,
+               (unsigned long long)pls->fnum,
                lock_type_name(pls->lock_type),
                lock_flav_name(pls->lock_flav) ));
 }
 
+unsigned int brl_num_locks(const struct byte_range_lock *brl)
+{
+       return brl->num_locks;
+}
+
+struct files_struct *brl_fsp(struct byte_range_lock *brl)
+{
+       return brl->fsp;
+}
+
+bool brl_have_read_oplocks(const struct byte_range_lock *brl)
+{
+       return brl->have_read_oplocks;
+}
+
+void brl_set_have_read_oplocks(struct byte_range_lock *brl,
+                              bool have_read_oplocks)
+{
+       DEBUG(10, ("Setting have_read_oplocks to %s\n",
+                  have_read_oplocks ? "true" : "false"));
+       SMB_ASSERT(brl->record != NULL); /* otherwise we're readonly */
+       brl->have_read_oplocks = have_read_oplocks;
+       brl->modified = true;
+}
+
 /****************************************************************************
  See if two locking contexts are equal.
 ****************************************************************************/
 
-bool brl_same_context(const struct lock_context *ctx1, 
+static bool brl_same_context(const struct lock_context *ctx1,
                             const struct lock_context *ctx2)
 {
-       return (procid_equal(&ctx1->pid, &ctx2->pid) &&
-               (ctx1->smbpid == ctx2->smbpid) &&
+       return (serverid_equal(&ctx1->pid, &ctx2->pid) &&
+               (ctx1->smblctx == ctx2->smblctx) &&
                (ctx1->tid == ctx2->tid));
 }
 
@@ -75,7 +117,7 @@ static bool brl_overlap(const struct lock_struct *lck1,
                         const struct lock_struct *lck2)
 {
        /* XXX Remove for Win7 compatibility. */
-       /* this extra check is not redundent - it copes with locks
+       /* this extra check is not redundant - it copes with locks
           that go beyond the end of 64 bit file space */
        if (lck1->size != 0 &&
            lck1->start == lck2->start &&
@@ -94,7 +136,7 @@ static bool brl_overlap(const struct lock_struct *lck1,
  See if lock2 can be added when lock1 is in place.
 ****************************************************************************/
 
-static bool brl_conflict(const struct lock_struct *lck1, 
+static bool brl_conflict(const struct lock_struct *lck1,
                         const struct lock_struct *lck2)
 {
        /* Ignore PENDING locks. */
@@ -115,7 +157,7 @@ static bool brl_conflict(const struct lock_struct *lck1,
        }
 
        return brl_overlap(lck1, lck2);
-} 
+}
 
 /****************************************************************************
  See if lock2 can be added when lock1 is in place - when both locks are POSIX
@@ -123,7 +165,7 @@ static bool brl_conflict(const struct lock_struct *lck1,
  know already match.
 ****************************************************************************/
 
-static bool brl_conflict_posix(const struct lock_struct *lck1, 
+static bool brl_conflict_posix(const struct lock_struct *lck1,
                                const struct lock_struct *lck2)
 {
 #if defined(DEVELOPER)
@@ -148,10 +190,10 @@ static bool brl_conflict_posix(const struct lock_struct *lck1,
        /* One is read, the other write, or the context is different,
           do they overlap ? */
        return brl_overlap(lck1, lck2);
-} 
+}
 
 #if ZERO_ZERO
-static bool brl_conflict1(const struct lock_struct *lck1, 
+static bool brl_conflict1(const struct lock_struct *lck1,
                         const struct lock_struct *lck2)
 {
        if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
@@ -174,9 +216,9 @@ static bool brl_conflict1(const struct lock_struct *lck1,
            lck2->start >= (lck1->start + lck1->size)) {
                return False;
        }
-           
+
        return True;
-} 
+}
 #endif
 
 /****************************************************************************
@@ -190,7 +232,7 @@ static bool brl_conflict_other(const struct lock_struct *lck1, const struct lock
        if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
                return False;
 
-       if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) 
+       if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK)
                return False;
 
        /* POSIX flavour locks never conflict here - this is only called
@@ -211,7 +253,7 @@ static bool brl_conflict_other(const struct lock_struct *lck1, const struct lock
        }
 
        return brl_overlap(lck1, lck2);
-} 
+}
 
 /****************************************************************************
  Check if an unlock overlaps a pending lock.
@@ -232,7 +274,9 @@ static bool brl_pending_overlap(const struct lock_struct *lock, const struct loc
  app depends on this ?
 ****************************************************************************/
 
-NTSTATUS brl_lock_failed(files_struct *fsp, const struct lock_struct *lock, bool blocking_lock)
+static NTSTATUS brl_lock_failed(files_struct *fsp,
+                               const struct lock_struct *lock,
+                               bool blocking_lock)
 {
        if (lock->start >= 0xEF000000 && (lock->start >> 63) == 0) {
                /* amazing the little things you learn with a test
@@ -245,7 +289,7 @@ NTSTATUS brl_lock_failed(files_struct *fsp, const struct lock_struct *lock, bool
                return NT_STATUS_FILE_LOCK_CONFLICT;
        }
 
-       if (procid_equal(&lock->context.pid, &fsp->last_lock_failure.context.pid) &&
+       if (serverid_equal(&lock->context.pid, &fsp->last_lock_failure.context.pid) &&
                        lock->context.tid == fsp->last_lock_failure.context.tid &&
                        lock->fnum == fsp->last_lock_failure.fnum &&
                        lock->start == fsp->last_lock_failure.start) {
@@ -264,13 +308,27 @@ NTSTATUS brl_lock_failed(files_struct *fsp, const struct lock_struct *lock, bool
 
 void brl_init(bool read_only)
 {
+       int tdb_flags;
+
        if (brlock_db) {
                return;
        }
+
+       tdb_flags = TDB_DEFAULT|TDB_VOLATILE|TDB_CLEAR_IF_FIRST|TDB_INCOMPATIBLE_HASH;
+
+       if (!lp_clustering()) {
+               /*
+                * We can't use the SEQNUM trick to cache brlock
+                * entries in the clustering case because ctdb seqnum
+                * propagation has a delay.
+                */
+               tdb_flags |= TDB_SEQNUM;
+       }
+
        brlock_db = db_open(NULL, lock_path("brlock.tdb"),
-                           lp_open_files_db_hash_size(),
-                           TDB_DEFAULT|TDB_VOLATILE|TDB_CLEAR_IF_FIRST,
-                           read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644 );
+                           lp_open_files_db_hash_size(), tdb_flags,
+                           read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644,
+                           DBWRAP_LOCK_ORDER_2);
        if (!brlock_db) {
                DEBUG(0,("Failed to open byte range locking database %s\n",
                        lock_path("brlock.tdb")));
@@ -292,7 +350,7 @@ void brl_shutdown(void)
  Compare two locks for sorting.
 ****************************************************************************/
 
-static int lock_compare(const struct lock_struct *lck1, 
+static int lock_compare(const struct lock_struct *lck1,
                         const struct lock_struct *lck2)
 {
        if (lck1->start != lck2->start) {
@@ -319,15 +377,20 @@ NTSTATUS brl_lock_windows_default(struct byte_range_lock *br_lck,
 
        SMB_ASSERT(plock->lock_type != UNLOCK_LOCK);
 
+       if ((plock->start + plock->size - 1 < plock->start) &&
+                       plock->size != 0) {
+               return NT_STATUS_INVALID_LOCK_RANGE;
+       }
+
        for (i=0; i < br_lck->num_locks; i++) {
                /* Do any Windows or POSIX locks conflict ? */
                if (brl_conflict(&locks[i], plock)) {
                        /* Remember who blocked us. */
-                       plock->context.smbpid = locks[i].context.smbpid;
+                       plock->context.smblctx = locks[i].context.smblctx;
                        return brl_lock_failed(fsp,plock,blocking_lock);
                }
 #if ZERO_ZERO
-               if (plock->start == 0 && plock->size == 0 && 
+               if (plock->start == 0 && plock->size == 0 &&
                                locks[i].size == 0) {
                        break;
                }
@@ -354,7 +417,7 @@ NTSTATUS brl_lock_windows_default(struct byte_range_lock *br_lck,
                                &errno_ret)) {
 
                        /* We don't know who blocked us. */
-                       plock->context.smbpid = 0xFFFFFFFF;
+                       plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL;
 
                        if (errno_ret == EACCES || errno_ret == EAGAIN) {
                                status = NT_STATUS_FILE_LOCK_CONFLICT;
@@ -367,7 +430,8 @@ NTSTATUS brl_lock_windows_default(struct byte_range_lock *br_lck,
        }
 
        /* no conflicts - add it to the list of locks */
-       locks = (struct lock_struct *)SMB_REALLOC(locks, (br_lck->num_locks + 1) * sizeof(*locks));
+       locks = talloc_realloc(br_lck, locks, struct lock_struct,
+                              (br_lck->num_locks + 1));
        if (!locks) {
                status = NT_STATUS_NO_MEMORY;
                goto fail;
@@ -390,10 +454,9 @@ NTSTATUS brl_lock_windows_default(struct byte_range_lock *br_lck,
  Cope with POSIX range splits and merges.
 ****************************************************************************/
 
-static unsigned int brlock_posix_split_merge(struct lock_struct *lck_arr,              /* Output array. */
-                                               const struct lock_struct *ex,           /* existing lock. */
-                                               const struct lock_struct *plock,        /* proposed lock. */
-                                               bool *lock_was_added)
+static unsigned int brlock_posix_split_merge(struct lock_struct *lck_arr,      /* Output array. */
+                                               struct lock_struct *ex,         /* existing lock. */
+                                               struct lock_struct *plock)      /* proposed lock. */
 {
        bool lock_types_differ = (ex->lock_type != plock->lock_type);
 
@@ -410,21 +473,23 @@ static unsigned int brlock_posix_split_merge(struct lock_struct *lck_arr,         /* Ou
        /* Did we overlap ? */
 
 /*********************************************
-                                             +---------+
-                                             | ex      |
-                                             +---------+
-                              +-------+
-                              | plock |
-                              +-------+
+                                        +---------+
+                                        | ex      |
+                                        +---------+
+                         +-------+
+                         | plock |
+                         +-------+
 OR....
-             +---------+
-             |  ex     |
-             +---------+
+        +---------+
+        |  ex     |
+        +---------+
 **********************************************/
 
        if ( (ex->start > (plock->start + plock->size)) ||
-                       (plock->start > (ex->start + ex->size))) {
+               (plock->start > (ex->start + ex->size))) {
+
                /* No overlap with this lock - copy existing. */
+
                memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
                return 1;
        }
@@ -436,26 +501,109 @@ OR....
         +---------------------------+
         |       plock               | -> replace with plock.
         +---------------------------+
+OR
+             +---------------+
+             |       ex      |
+             +---------------+
+        +---------------------------+
+        |       plock               | -> replace with plock.
+        +---------------------------+
+
 **********************************************/
 
        if ( (ex->start >= plock->start) &&
-                       (ex->start + ex->size <= plock->start + plock->size) ) {
-               memcpy(&lck_arr[0], plock, sizeof(struct lock_struct));
-               *lock_was_added = True;
-               return 1;
+               (ex->start + ex->size <= plock->start + plock->size) ) {
+
+               /* Replace - discard existing lock. */
+
+               return 0;
+       }
+
+/*********************************************
+Adjacent after.
+                        +-------+
+                        |  ex   |
+                        +-------+
+        +---------------+
+        |   plock       |
+        +---------------+
+
+BECOMES....
+        +---------------+-------+
+        |   plock       | ex    | - different lock types.
+        +---------------+-------+
+OR.... (merge)
+        +-----------------------+
+        |   plock               | - same lock type.
+        +-----------------------+
+**********************************************/
+
+       if (plock->start + plock->size == ex->start) {
+
+               /* If the lock types are the same, we merge, if different, we
+                  add the remainder of the old lock. */
+
+               if (lock_types_differ) {
+                       /* Add existing. */
+                       memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
+                       return 1;
+               } else {
+                       /* Merge - adjust incoming lock as we may have more
+                        * merging to come. */
+                       plock->size += ex->size;
+                       return 0;
+               }
        }
 
 /*********************************************
+Adjacent before.
+        +-------+
+        |  ex   |
+        +-------+
+                +---------------+
+                |   plock       |
+                +---------------+
+BECOMES....
+        +-------+---------------+
+        | ex    |   plock       | - different lock types
+        +-------+---------------+
+
+OR.... (merge)
+        +-----------------------+
+        |      plock            | - same lock type.
+        +-----------------------+
+
+**********************************************/
+
+       if (ex->start + ex->size == plock->start) {
+
+               /* If the lock types are the same, we merge, if different, we
+                  add the existing lock. */
+
+               if (lock_types_differ) {
+                       memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
+                       return 1;
+               } else {
+                       /* Merge - adjust incoming lock as we may have more
+                        * merging to come. */
+                       plock->start = ex->start;
+                       plock->size += ex->size;
+                       return 0;
+               }
+       }
+
+/*********************************************
+Overlap after.
         +-----------------------+
         |          ex           |
         +-----------------------+
         +---------------+
         |   plock       |
         +---------------+
-OR....
-                        +-------+
-                        |  ex   |
-                        +-------+
+OR
+               +----------------+
+               |       ex       |
+               +----------------+
         +---------------+
         |   plock       |
         +---------------+
@@ -466,60 +614,57 @@ BECOMES....
         +---------------+-------+
 OR.... (merge)
         +-----------------------+
-        |   ex                  | - same lock type.
+        |   plock               | - same lock type.
         +-----------------------+
 **********************************************/
 
        if ( (ex->start >= plock->start) &&
-                               (ex->start <= plock->start + plock->size) &&
-                               (ex->start + ex->size > plock->start + plock->size) ) {
-
-               *lock_was_added = True;
+               (ex->start <= plock->start + plock->size) &&
+               (ex->start + ex->size > plock->start + plock->size) ) {
 
                /* If the lock types are the same, we merge, if different, we
-                  add the new lock before the old. */
+                  add the remainder of the old lock. */
 
                if (lock_types_differ) {
-                       /* Add new. */
-                       memcpy(&lck_arr[0], plock, sizeof(struct lock_struct));
-                       memcpy(&lck_arr[1], ex, sizeof(struct lock_struct));
+                       /* Add remaining existing. */
+                       memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
                        /* Adjust existing start and size. */
-                       lck_arr[1].start = plock->start + plock->size;
-                       lck_arr[1].size = (ex->start + ex->size) - (plock->start + plock->size);
-                       return 2;
-               } else {
-                       /* Merge. */
-                       memcpy(&lck_arr[0], plock, sizeof(struct lock_struct));
-                       /* Set new start and size. */
-                       lck_arr[0].start = plock->start;
-                       lck_arr[0].size = (ex->start + ex->size) - plock->start;
+                       lck_arr[0].start = plock->start + plock->size;
+                       lck_arr[0].size = (ex->start + ex->size) - (plock->start + plock->size);
                        return 1;
+               } else {
+                       /* Merge - adjust incoming lock as we may have more
+                        * merging to come. */
+                       plock->size += (ex->start + ex->size) - (plock->start + plock->size);
+                       return 0;
                }
        }
 
 /*********************************************
-   +-----------------------+
-   |  ex                   |
-   +-----------------------+
-           +---------------+
-           |   plock       |
-           +---------------+
-OR....
-   +-------+        
-   |  ex   |
-   +-------+
-           +---------------+
-           |   plock       |
-           +---------------+
+Overlap before.
+        +-----------------------+
+        |  ex                   |
+        +-----------------------+
+                +---------------+
+                |   plock       |
+                +---------------+
+OR
+        +-------------+
+        |  ex         |
+        +-------------+
+                +---------------+
+                |   plock       |
+                +---------------+
+
 BECOMES....
-   +-------+---------------+
-   | ex    |   plock       | - different lock types
-   +-------+---------------+
+        +-------+---------------+
+        | ex    |   plock       | - different lock types
+        +-------+---------------+
 
 OR.... (merge)
-   +-----------------------+
-   | ex                    | - same lock type.
-   +-----------------------+
+        +-----------------------+
+        |      plock            | - same lock type.
+        +-----------------------+
 
 **********************************************/
 
@@ -527,27 +672,25 @@ OR.... (merge)
                        (ex->start + ex->size >= plock->start) &&
                        (ex->start + ex->size <= plock->start + plock->size) ) {
 
-               *lock_was_added = True;
-
                /* If the lock types are the same, we merge, if different, we
-                  add the new lock after the old. */
+                  add the truncated old lock. */
 
                if (lock_types_differ) {
                        memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
-                       memcpy(&lck_arr[1], plock, sizeof(struct lock_struct));
                        /* Adjust existing size. */
                        lck_arr[0].size = plock->start - ex->start;
-                       return 2;
-               } else {
-                       /* Merge. */
-                       memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
-                       /* Adjust existing size. */
-                       lck_arr[0].size = (plock->start + plock->size) - ex->start;
                        return 1;
+               } else {
+                       /* Merge - adjust incoming lock as we may have more
+                        * merging to come. MUST ADJUST plock SIZE FIRST ! */
+                       plock->size += (plock->start - ex->start);
+                       plock->start = ex->start;
+                       return 0;
                }
        }
 
 /*********************************************
+Complete overlap.
         +---------------------------+
         |        ex                 |
         +---------------------------+
@@ -560,32 +703,31 @@ BECOMES.....
         +-------+---------+---------+
 OR
         +---------------------------+
-        |        ex                 | - same lock type.
+        |        plock              | - same lock type.
         +---------------------------+
 **********************************************/
 
        if ( (ex->start < plock->start) && (ex->start + ex->size > plock->start + plock->size) ) {
-               *lock_was_added = True;
 
                if (lock_types_differ) {
 
                        /* We have to split ex into two locks here. */
 
                        memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
-                       memcpy(&lck_arr[1], plock, sizeof(struct lock_struct));
-                       memcpy(&lck_arr[2], ex, sizeof(struct lock_struct));
+                       memcpy(&lck_arr[1], ex, sizeof(struct lock_struct));
 
                        /* Adjust first existing size. */
                        lck_arr[0].size = plock->start - ex->start;
 
                        /* Adjust second existing start and size. */
-                       lck_arr[2].start = plock->start + plock->size;
-                       lck_arr[2].size = (ex->start + ex->size) - (plock->start + plock->size);
-                       return 3;
+                       lck_arr[1].start = plock->start + plock->size;
+                       lck_arr[1].size = (ex->start + ex->size) - (plock->start + plock->size);
+                       return 2;
                } else {
-                       /* Just eat plock. */
-                       memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
-                       return 1;
+                       /* Just eat the existing locks, merge them into plock. */
+                       plock->start = ex->start;
+                       plock->size = ex->size;
+                       return 0;
                }
        }
 
@@ -609,7 +751,6 @@ static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
        unsigned int i, count, posix_count;
        struct lock_struct *locks = br_lck->lock_data;
        struct lock_struct *tp;
-       bool lock_was_added = False;
        bool signal_pending_read = False;
        bool break_oplocks = false;
        NTSTATUS status;
@@ -620,8 +761,7 @@ static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
        }
 
        /* Don't allow 64-bit lock wrap. */
-       if (plock->start + plock->size < plock->start ||
-                       plock->start + plock->size < plock->size) {
+       if (plock->start + plock->size - 1 < plock->start) {
                return NT_STATUS_INVALID_PARAMETER;
        }
 
@@ -629,12 +769,13 @@ static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
           existing POSIX lock range into two, and add our lock,
           so we need at most 2 more entries. */
 
-       tp = SMB_MALLOC_ARRAY(struct lock_struct, (br_lck->num_locks + 2));
+       tp = talloc_array(br_lck, struct lock_struct, br_lck->num_locks + 2);
        if (!tp) {
                return NT_STATUS_NO_MEMORY;
        }
-       
+
        count = posix_count = 0;
+
        for (i=0; i < br_lck->num_locks; i++) {
                struct lock_struct *curr_lock = &locks[i];
 
@@ -649,9 +790,9 @@ static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
                        /* Do any Windows flavour locks conflict ? */
                        if (brl_conflict(curr_lock, plock)) {
                                /* No games with error messages. */
-                               SAFE_FREE(tp);
+                               TALLOC_FREE(tp);
                                /* Remember who blocked us. */
-                               plock->context.smbpid = curr_lock->context.smbpid;
+                               plock->context.smblctx = curr_lock->context.smblctx;
                                return NT_STATUS_FILE_LOCK_CONFLICT;
                        }
                        /* Just copy the Windows lock into the new array. */
@@ -664,14 +805,14 @@ static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
                        if (brl_conflict_posix(curr_lock, plock)) {
                                /* Can't block ourselves with POSIX locks. */
                                /* No games with error messages. */
-                               SAFE_FREE(tp);
+                               TALLOC_FREE(tp);
                                /* Remember who blocked us. */
-                               plock->context.smbpid = curr_lock->context.smbpid;
+                               plock->context.smblctx = curr_lock->context.smblctx;
                                return NT_STATUS_FILE_LOCK_CONFLICT;
                        }
 
                        /* Work out overlaps. */
-                       tmp_count += brlock_posix_split_merge(&tp[count], curr_lock, plock, &lock_was_added);
+                       tmp_count += brlock_posix_split_merge(&tp[count], curr_lock, plock);
                        posix_count += tmp_count;
                        count += tmp_count;
                }
@@ -692,11 +833,22 @@ static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
                                             LEVEL2_CONTEND_POSIX_BRL);
        }
 
-       if (!lock_was_added) {
-               memcpy(&tp[count], plock, sizeof(struct lock_struct));
-               count++;
+       /* Try and add the lock in order, sorted by lock start. */
+       for (i=0; i < count; i++) {
+               struct lock_struct *curr_lock = &tp[i];
+
+               if (curr_lock->start <= plock->start) {
+                       continue;
+               }
        }
 
+       if (i < count) {
+               memmove(&tp[i+1], &tp[i],
+                       (count - i)*sizeof(struct lock_struct));
+       }
+       memcpy(&tp[i], plock, sizeof(struct lock_struct));
+       count++;
+
        /* We can get the POSIX lock, now see if it needs to
           be mapped into a lower level POSIX one, and if so can
           we get it ? */
@@ -715,28 +867,32 @@ static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
                                &errno_ret)) {
 
                        /* We don't know who blocked us. */
-                       plock->context.smbpid = 0xFFFFFFFF;
+                       plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL;
 
                        if (errno_ret == EACCES || errno_ret == EAGAIN) {
-                               SAFE_FREE(tp);
+                               TALLOC_FREE(tp);
                                status = NT_STATUS_FILE_LOCK_CONFLICT;
                                goto fail;
                        } else {
-                               SAFE_FREE(tp);
+                               TALLOC_FREE(tp);
                                status = map_nt_error_from_unix(errno);
                                goto fail;
                        }
                }
        }
 
-       /* Realloc so we don't leak entries per lock call. */
-       tp = (struct lock_struct *)SMB_REALLOC(tp, count * sizeof(*locks));
-       if (!tp) {
-               status = NT_STATUS_NO_MEMORY;
-               goto fail;
+       /* If we didn't use all the allocated size,
+        * Realloc so we don't leak entries per lock call. */
+       if (count < br_lck->num_locks + 2) {
+               tp = talloc_realloc(br_lck, tp, struct lock_struct, count);
+               if (!tp) {
+                       status = NT_STATUS_NO_MEMORY;
+                       goto fail;
+               }
        }
+
        br_lck->num_locks = count;
-       SAFE_FREE(br_lck->lock_data);
+       TALLOC_FREE(br_lck->lock_data);
        br_lck->lock_data = tp;
        locks = tp;
        br_lck->modified = True;
@@ -774,20 +930,31 @@ static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
        return status;
 }
 
+NTSTATUS smb_vfs_call_brl_lock_windows(struct vfs_handle_struct *handle,
+                                      struct byte_range_lock *br_lck,
+                                      struct lock_struct *plock,
+                                      bool blocking_lock,
+                                      struct blocking_lock_record *blr)
+{
+       VFS_FIND(brl_lock_windows);
+       return handle->fns->brl_lock_windows_fn(handle, br_lck, plock,
+                                               blocking_lock, blr);
+}
+
 /****************************************************************************
  Lock a range of bytes.
 ****************************************************************************/
 
 NTSTATUS brl_lock(struct messaging_context *msg_ctx,
                struct byte_range_lock *br_lck,
-               uint32 smbpid,
+               uint64_t smblctx,
                struct server_id pid,
                br_off start,
-               br_off size, 
+               br_off size,
                enum brl_type lock_type,
                enum brl_flavour lock_flav,
                bool blocking_lock,
-               uint32 *psmbpid,
+               uint64_t *psmblctx,
                struct blocking_lock_record *blr)
 {
        NTSTATUS ret;
@@ -801,10 +968,10 @@ NTSTATUS brl_lock(struct messaging_context *msg_ctx,
 
 #ifdef DEVELOPER
        /* Quieten valgrind on test. */
-       memset(&lock, '\0', sizeof(lock));
+       ZERO_STRUCT(lock);
 #endif
 
-       lock.context.smbpid = smbpid;
+       lock.context.smblctx = smblctx;
        lock.context.pid = pid;
        lock.context.tid = br_lck->fsp->conn->cnum;
        lock.start = start;
@@ -822,12 +989,12 @@ NTSTATUS brl_lock(struct messaging_context *msg_ctx,
 
 #if ZERO_ZERO
        /* sort the lock list */
-       qsort(br_lck->lock_data, (size_t)br_lck->num_locks, sizeof(lock), lock_compare);
+       TYPESAFE_QSORT(br_lck->lock_data, (size_t)br_lck->num_locks, lock_compare);
 #endif
 
        /* If we're returning an error, return who blocked us. */
-       if (!NT_STATUS_IS_OK(ret) && psmbpid) {
-               *psmbpid = lock.context.smbpid;
+       if (!NT_STATUS_IS_OK(ret) && psmblctx) {
+               *psmblctx = lock.context.smblctx;
        }
        return ret;
 }
@@ -875,6 +1042,10 @@ bool brl_unlock_windows_default(struct messaging_context *msg_ctx,
        for (i = 0; i < br_lck->num_locks; i++) {
                struct lock_struct *lock = &locks[i];
 
+               if (IS_PENDING_LOCK(lock->lock_type)) {
+                       continue;
+               }
+
                /* Only remove our own locks that match in start, size, and flavour. */
                if (brl_same_context(&lock->context, &plock->context) &&
                                        lock->fnum == plock->fnum &&
@@ -897,7 +1068,7 @@ bool brl_unlock_windows_default(struct messaging_context *msg_ctx,
 
        /* Actually delete the lock. */
        if (i < br_lck->num_locks - 1) {
-               memmove(&locks[i], &locks[i+1], 
+               memmove(&locks[i], &locks[i+1],
                        sizeof(*locks)*((br_lck->num_locks-1) - i));
        }
 
@@ -944,9 +1115,9 @@ bool brl_unlock_windows_default(struct messaging_context *msg_ctx,
 
 static bool brl_unlock_posix(struct messaging_context *msg_ctx,
                             struct byte_range_lock *br_lck,
-                            const struct lock_struct *plock)
+                            struct lock_struct *plock)
 {
-       unsigned int i, j, count, posix_count;
+       unsigned int i, j, count;
        struct lock_struct *tp;
        struct lock_struct *locks = br_lck->lock_data;
        bool overlap_found = False;
@@ -967,17 +1138,15 @@ static bool brl_unlock_posix(struct messaging_context *msg_ctx,
           existing POSIX lock range into two, so we need at most
           1 more entry. */
 
-       tp = SMB_MALLOC_ARRAY(struct lock_struct, (br_lck->num_locks + 1));
+       tp = talloc_array(br_lck, struct lock_struct, br_lck->num_locks + 1);
        if (!tp) {
                DEBUG(10,("brl_unlock_posix: malloc fail\n"));
                return False;
        }
 
-       count = posix_count = 0;
+       count = 0;
        for (i = 0; i < br_lck->num_locks; i++) {
                struct lock_struct *lock = &locks[i];
-               struct lock_struct tmp_lock[3];
-               bool lock_was_added = False;
                unsigned int tmp_count;
 
                /* Only remove our own locks - ignore fnum. */
@@ -988,73 +1157,55 @@ static bool brl_unlock_posix(struct messaging_context *msg_ctx,
                        continue;
                }
 
-               /* Work out overlaps. */
-               tmp_count = brlock_posix_split_merge(&tmp_lock[0], &locks[i], plock, &lock_was_added);
-
-               if (tmp_count == 1) {
-                       /* Ether the locks didn't overlap, or the unlock completely
-                          overlapped this lock. If it didn't overlap, then there's
-                          no change in the locks. */
-                       if (tmp_lock[0].lock_type != UNLOCK_LOCK) {
-                               SMB_ASSERT(tmp_lock[0].lock_type == locks[i].lock_type);
-                               /* No change in this lock. */
-                               memcpy(&tp[count], &tmp_lock[0], sizeof(struct lock_struct));
-                               count++;
-                               posix_count++;
-                       } else {
-                               SMB_ASSERT(tmp_lock[0].lock_type == UNLOCK_LOCK);
-                               overlap_found = True;
-                       }
-                       continue;
-               } else if (tmp_count == 2) {
-                       /* The unlock overlapped an existing lock. Copy the truncated
-                          lock into the lock array. */
-                       if (tmp_lock[0].lock_type != UNLOCK_LOCK) {
-                               SMB_ASSERT(tmp_lock[0].lock_type == locks[i].lock_type);
-                               SMB_ASSERT(tmp_lock[1].lock_type == UNLOCK_LOCK);
-                               memcpy(&tp[count], &tmp_lock[0], sizeof(struct lock_struct));
-                               if (tmp_lock[0].size != locks[i].size) {
-                                       overlap_found = True;
-                               }
-                       } else {
-                               SMB_ASSERT(tmp_lock[0].lock_type == UNLOCK_LOCK);
-                               SMB_ASSERT(tmp_lock[1].lock_type == locks[i].lock_type);
-                               memcpy(&tp[count], &tmp_lock[1], sizeof(struct lock_struct));
-                               if (tmp_lock[1].start != locks[i].start) {
-                                       overlap_found = True;
-                               }
+               if (lock->lock_flav == WINDOWS_LOCK) {
+                       /* Do any Windows flavour locks conflict ? */
+                       if (brl_conflict(lock, plock)) {
+                               TALLOC_FREE(tp);
+                               return false;
                        }
+                       /* Just copy the Windows lock into the new array. */
+                       memcpy(&tp[count], lock, sizeof(struct lock_struct));
                        count++;
-                       posix_count++;
                        continue;
-               } else {
-                       /* tmp_count == 3 - (we split a lock range in two). */
-                       SMB_ASSERT(tmp_lock[0].lock_type == locks[i].lock_type);
-                       SMB_ASSERT(tmp_lock[1].lock_type == UNLOCK_LOCK);
-                       SMB_ASSERT(tmp_lock[2].lock_type == locks[i].lock_type);
+               }
+
+               /* Work out overlaps. */
+               tmp_count = brlock_posix_split_merge(&tp[count], lock, plock);
+
+               if (tmp_count == 0) {
+                       /* plock overlapped the existing lock completely,
+                          or replaced it. Don't copy the existing lock. */
+                       overlap_found = true;
+               } else if (tmp_count == 1) {
+                       /* Either no overlap, (simple copy of existing lock) or
+                        * an overlap of an existing lock. */
+                       /* If the lock changed size, we had an overlap. */
+                       if (tp[count].size != lock->size) {
+                               overlap_found = true;
+                       }
+                       count += tmp_count;
+               } else if (tmp_count == 2) {
+                       /* We split a lock range in two. */
+                       overlap_found = true;
+                       count += tmp_count;
 
-                       memcpy(&tp[count], &tmp_lock[0], sizeof(struct lock_struct));
-                       count++;
-                       posix_count++;
-                       memcpy(&tp[count], &tmp_lock[2], sizeof(struct lock_struct));
-                       count++;
-                       posix_count++;
-                       overlap_found = True;
                        /* Optimisation... */
                        /* We know we're finished here as we can't overlap any
                           more POSIX locks. Copy the rest of the lock array. */
+
                        if (i < br_lck->num_locks - 1) {
-                               memcpy(&tp[count], &locks[i+1], 
+                               memcpy(&tp[count], &locks[i+1],
                                        sizeof(*locks)*((br_lck->num_locks-1) - i));
                                count += ((br_lck->num_locks-1) - i);
                        }
                        break;
                }
+
        }
 
        if (!overlap_found) {
                /* Just ignore - no change. */
-               SAFE_FREE(tp);
+               TALLOC_FREE(tp);
                DEBUG(10,("brl_unlock_posix: No overlap - unlocked.\n"));
                return True;
        }
@@ -1071,24 +1222,22 @@ static bool brl_unlock_posix(struct messaging_context *msg_ctx,
 
        /* Realloc so we don't leak entries per unlock call. */
        if (count) {
-               tp = (struct lock_struct *)SMB_REALLOC(tp, count * sizeof(*locks));
+               tp = talloc_realloc(br_lck, tp, struct lock_struct, count);
                if (!tp) {
                        DEBUG(10,("brl_unlock_posix: realloc fail\n"));
                        return False;
                }
        } else {
                /* We deleted the last lock. */
-               SAFE_FREE(tp);
+               TALLOC_FREE(tp);
                tp = NULL;
        }
 
-       if (posix_count == 0) {
-               contend_level2_oplocks_end(br_lck->fsp,
-                                          LEVEL2_CONTEND_POSIX_BRL);
-       }
+       contend_level2_oplocks_end(br_lck->fsp,
+                                  LEVEL2_CONTEND_POSIX_BRL);
 
        br_lck->num_locks = count;
-       SAFE_FREE(br_lck->lock_data);
+       TALLOC_FREE(br_lck->lock_data);
        locks = tp;
        br_lck->lock_data = tp;
        br_lck->modified = True;
@@ -1116,13 +1265,23 @@ static bool brl_unlock_posix(struct messaging_context *msg_ctx,
        return True;
 }
 
+bool smb_vfs_call_brl_unlock_windows(struct vfs_handle_struct *handle,
+                                    struct messaging_context *msg_ctx,
+                                    struct byte_range_lock *br_lck,
+                                    const struct lock_struct *plock)
+{
+       VFS_FIND(brl_unlock_windows);
+       return handle->fns->brl_unlock_windows_fn(handle, msg_ctx, br_lck,
+                                                 plock);
+}
+
 /****************************************************************************
  Unlock a range of bytes.
 ****************************************************************************/
 
 bool brl_unlock(struct messaging_context *msg_ctx,
                struct byte_range_lock *br_lck,
-               uint32 smbpid,
+               uint64_t smblctx,
                struct server_id pid,
                br_off start,
                br_off size,
@@ -1130,7 +1289,7 @@ bool brl_unlock(struct messaging_context *msg_ctx,
 {
        struct lock_struct lock;
 
-       lock.context.smbpid = smbpid;
+       lock.context.smblctx = smblctx;
        lock.context.pid = pid;
        lock.context.tid = br_lck->fsp->conn->cnum;
        lock.start = start;
@@ -1153,10 +1312,10 @@ bool brl_unlock(struct messaging_context *msg_ctx,
 ****************************************************************************/
 
 bool brl_locktest(struct byte_range_lock *br_lck,
-               uint32 smbpid,
+               uint64_t smblctx,
                struct server_id pid,
                br_off start,
-               br_off size, 
+               br_off size,
                enum brl_type lock_type,
                enum brl_flavour lock_flav)
 {
@@ -1166,7 +1325,7 @@ bool brl_locktest(struct byte_range_lock *br_lck,
        const struct lock_struct *locks = br_lck->lock_data;
        files_struct *fsp = br_lck->fsp;
 
-       lock.context.smbpid = smbpid;
+       lock.context.smblctx = smblctx;
        lock.context.pid = pid;
        lock.context.tid = br_lck->fsp->conn->cnum;
        lock.start = start;
@@ -1194,9 +1353,9 @@ bool brl_locktest(struct byte_range_lock *br_lck,
        if(lp_posix_locking(fsp->conn->params) && (lock_flav == WINDOWS_LOCK)) {
                ret = is_posix_locked(fsp, &start, &size, &lock_type, WINDOWS_LOCK);
 
-               DEBUG(10,("brl_locktest: posix start=%.0f len=%.0f %s for fnum %d file %s\n",
+               DEBUG(10,("brl_locktest: posix start=%.0f len=%.0f %s for %s file %s\n",
                        (double)start, (double)size, ret ? "locked" : "unlocked",
-                       fsp->fnum, fsp->fsp_name ));
+                       fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
 
                /* We need to return the inverse of is_posix_locked. */
                ret = !ret;
@@ -1211,10 +1370,10 @@ bool brl_locktest(struct byte_range_lock *br_lck,
 ****************************************************************************/
 
 NTSTATUS brl_lockquery(struct byte_range_lock *br_lck,
-               uint32 *psmbpid,
+               uint64_t *psmblctx,
                struct server_id pid,
                br_off *pstart,
-               br_off *psize, 
+               br_off *psize,
                enum brl_type *plock_type,
                enum brl_flavour lock_flav)
 {
@@ -1223,7 +1382,7 @@ NTSTATUS brl_lockquery(struct byte_range_lock *br_lck,
        const struct lock_struct *locks = br_lck->lock_data;
        files_struct *fsp = br_lck->fsp;
 
-       lock.context.smbpid = *psmbpid;
+       lock.context.smblctx = *psmblctx;
        lock.context.pid = pid;
        lock.context.tid = br_lck->fsp->conn->cnum;
        lock.start = *pstart;
@@ -1239,12 +1398,12 @@ NTSTATUS brl_lockquery(struct byte_range_lock *br_lck,
 
                if (exlock->lock_flav == WINDOWS_LOCK) {
                        conflict = brl_conflict(exlock, &lock);
-               } else {        
+               } else {
                        conflict = brl_conflict_posix(exlock, &lock);
                }
 
                if (conflict) {
-                       *psmbpid = exlock->context.smbpid;
+                       *psmblctx = exlock->context.smblctx;
                        *pstart = exlock->start;
                        *psize = exlock->size;
                        *plock_type = exlock->lock_type;
@@ -1260,13 +1419,13 @@ NTSTATUS brl_lockquery(struct byte_range_lock *br_lck,
        if(lp_posix_locking(fsp->conn->params)) {
                bool ret = is_posix_locked(fsp, pstart, psize, plock_type, POSIX_LOCK);
 
-               DEBUG(10,("brl_lockquery: posix start=%.0f len=%.0f %s for fnum %d file %s\n",
+               DEBUG(10,("brl_lockquery: posix start=%.0f len=%.0f %s for %s file %s\n",
                        (double)*pstart, (double)*psize, ret ? "locked" : "unlocked",
-                       fsp->fnum, fsp->fsp_name ));
+                       fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
 
                if (ret) {
-                       /* Hmmm. No clue what to set smbpid to - use -1. */
-                       *psmbpid = 0xFFFF;
+                       /* Hmmm. No clue what to set smblctx to - use -1. */
+                       *psmblctx = 0xFFFFFFFFFFFFFFFFLL;
                        return NT_STATUS_LOCK_NOT_GRANTED;
                }
         }
@@ -1274,11 +1433,21 @@ NTSTATUS brl_lockquery(struct byte_range_lock *br_lck,
        return NT_STATUS_OK;
 }
 
+
+bool smb_vfs_call_brl_cancel_windows(struct vfs_handle_struct *handle,
+                                    struct byte_range_lock *br_lck,
+                                    struct lock_struct *plock,
+                                    struct blocking_lock_record *blr)
+{
+       VFS_FIND(brl_cancel_windows);
+       return handle->fns->brl_cancel_windows_fn(handle, br_lck, plock, blr);
+}
+
 /****************************************************************************
  Remove a particular pending lock.
 ****************************************************************************/
 bool brl_lock_cancel(struct byte_range_lock *br_lck,
-               uint32 smbpid,
+               uint64_t smblctx,
                struct server_id pid,
                br_off start,
                br_off size,
@@ -1288,7 +1457,7 @@ bool brl_lock_cancel(struct byte_range_lock *br_lck,
        bool ret;
        struct lock_struct lock;
 
-       lock.context.smbpid = smbpid;
+       lock.context.smblctx = smblctx;
        lock.context.pid = pid;
        lock.context.tid = br_lck->fsp->conn->cnum;
        lock.start = start;
@@ -1336,7 +1505,7 @@ bool brl_lock_cancel_default(struct byte_range_lock *br_lck,
 
        if (i < br_lck->num_locks - 1) {
                /* Found this particular pending lock - delete it */
-               memmove(&locks[i], &locks[i+1], 
+               memmove(&locks[i], &locks[i+1],
                        sizeof(*locks)*((br_lck->num_locks-1) - i));
        }
 
@@ -1355,166 +1524,239 @@ void brl_close_fnum(struct messaging_context *msg_ctx,
                    struct byte_range_lock *br_lck)
 {
        files_struct *fsp = br_lck->fsp;
-       uint16 tid = fsp->conn->cnum;
-       int fnum = fsp->fnum;
-       unsigned int i, j, dcount=0;
-       int num_deleted_windows_locks = 0;
+       uint32_t tid = fsp->conn->cnum;
+       uint64_t fnum = fsp->fnum;
+       unsigned int i;
        struct lock_struct *locks = br_lck->lock_data;
-       struct server_id pid = procid_self();
-       bool unlock_individually = False;
-       bool posix_level2_contention_ended = false;
+       struct server_id pid = messaging_server_id(fsp->conn->sconn->msg_ctx);
+       struct lock_struct *locks_copy;
+       unsigned int num_locks_copy;
+
+       /* Copy the current lock array. */
+       if (br_lck->num_locks) {
+               locks_copy = (struct lock_struct *)talloc_memdup(br_lck, locks, br_lck->num_locks * sizeof(struct lock_struct));
+               if (!locks_copy) {
+                       smb_panic("brl_close_fnum: talloc failed");
+                       }
+       } else {
+               locks_copy = NULL;
+       }
 
-       if(lp_posix_locking(fsp->conn->params)) {
+       num_locks_copy = br_lck->num_locks;
 
-               /* Check if there are any Windows locks associated with this dev/ino
-                  pair that are not this fnum. If so we need to call unlock on each
-                  one in order to release the system POSIX locks correctly. */
+       for (i=0; i < num_locks_copy; i++) {
+               struct lock_struct *lock = &locks_copy[i];
 
-               for (i=0; i < br_lck->num_locks; i++) {
-                       struct lock_struct *lock = &locks[i];
+               if (lock->context.tid == tid && serverid_equal(&lock->context.pid, &pid) &&
+                               (lock->fnum == fnum)) {
+                       brl_unlock(msg_ctx,
+                               br_lck,
+                               lock->context.smblctx,
+                               pid,
+                               lock->start,
+                               lock->size,
+                               lock->lock_flav);
+               }
+       }
+}
 
-                       if (!procid_equal(&lock->context.pid, &pid)) {
-                               continue;
-                       }
+bool brl_mark_disconnected(struct files_struct *fsp)
+{
+       uint32_t tid = fsp->conn->cnum;
+       uint64_t smblctx = fsp->op->global->open_persistent_id;
+       uint64_t fnum = fsp->fnum;
+       unsigned int i;
+       struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx);
+       struct byte_range_lock *br_lck = NULL;
 
-                       if (lock->lock_type != READ_LOCK && lock->lock_type != WRITE_LOCK) {
-                               continue; /* Ignore pending. */
-                       }
+       if (!fsp->op->global->durable) {
+               return false;
+       }
 
-                       if (lock->context.tid != tid || lock->fnum != fnum) {
-                               unlock_individually = True;
-                               break;
-                       }
-               }
+       if (fsp->current_lock_count == 0) {
+               return true;
+       }
 
-               if (unlock_individually) {
-                       struct lock_struct *locks_copy;
-                       unsigned int num_locks_copy;
-
-                       /* Copy the current lock array. */
-                       if (br_lck->num_locks) {
-                               locks_copy = (struct lock_struct *)TALLOC_MEMDUP(br_lck, locks, br_lck->num_locks * sizeof(struct lock_struct));
-                               if (!locks_copy) {
-                                       smb_panic("brl_close_fnum: talloc failed");
-                               }
-                       } else {        
-                               locks_copy = NULL;
-                       }
+       br_lck = brl_get_locks(talloc_tos(), fsp);
+       if (br_lck == NULL) {
+               return false;
+       }
 
-                       num_locks_copy = br_lck->num_locks;
+       for (i=0; i < br_lck->num_locks; i++) {
+               struct lock_struct *lock = &br_lck->lock_data[i];
 
-                       for (i=0; i < num_locks_copy; i++) {
-                               struct lock_struct *lock = &locks_copy[i];
+               /*
+                * as this is a durable handle, we only expect locks
+                * of the current file handle!
+                */
 
-                               if (lock->context.tid == tid && procid_equal(&lock->context.pid, &pid) &&
-                                               (lock->fnum == fnum)) {
-                                       brl_unlock(msg_ctx,
-                                               br_lck,
-                                               lock->context.smbpid,
-                                               pid,
-                                               lock->start,
-                                               lock->size,
-                                               lock->lock_flav);
-                               }
-                       }
-                       return;
+               if (lock->context.smblctx != smblctx) {
+                       TALLOC_FREE(br_lck);
+                       return false;
+               }
+
+               if (lock->context.tid != tid) {
+                       TALLOC_FREE(br_lck);
+                       return false;
                }
+
+               if (!serverid_equal(&lock->context.pid, &self)) {
+                       TALLOC_FREE(br_lck);
+                       return false;
+               }
+
+               if (lock->fnum != fnum) {
+                       TALLOC_FREE(br_lck);
+                       return false;
+               }
+
+               server_id_set_disconnected(&lock->context.pid);
+               lock->context.tid = TID_FIELD_INVALID;
+               lock->fnum = FNUM_FIELD_INVALID;
        }
 
-       /* We can bulk delete - any POSIX locks will be removed when the fd closes. */
+       br_lck->modified = true;
+       TALLOC_FREE(br_lck);
+       return true;
+}
+
+bool brl_reconnect_disconnected(struct files_struct *fsp)
+{
+       uint32_t tid = fsp->conn->cnum;
+       uint64_t smblctx = fsp->op->global->open_persistent_id;
+       uint64_t fnum = fsp->fnum;
+       unsigned int i;
+       struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx);
+       struct byte_range_lock *br_lck = NULL;
 
-       /* Remove any existing locks for this fnum (or any fnum if they're POSIX). */
+       if (!fsp->op->global->durable) {
+               return false;
+       }
+
+       /*
+        * When reconnecting, we do not want to validate the brlock entries
+        * and thereby remove our own (disconnected) entries but reactivate
+        * them instead.
+        */
+       fsp->lockdb_clean = true;
+
+       br_lck = brl_get_locks(talloc_tos(), fsp);
+       if (br_lck == NULL) {
+               return false;
+       }
+
+       if (br_lck->num_locks == 0) {
+               TALLOC_FREE(br_lck);
+               return true;
+       }
 
        for (i=0; i < br_lck->num_locks; i++) {
-               struct lock_struct *lock = &locks[i];
-               bool del_this_lock = False;
-
-               if (lock->context.tid == tid && procid_equal(&lock->context.pid, &pid)) {
-                       if ((lock->lock_flav == WINDOWS_LOCK) && (lock->fnum == fnum)) {
-                               del_this_lock = True;
-                               num_deleted_windows_locks++;
-                               contend_level2_oplocks_end(br_lck->fsp,
-                                   LEVEL2_CONTEND_WINDOWS_BRL);
-                       } else if (lock->lock_flav == POSIX_LOCK) {
-                               del_this_lock = True;
-
-                               /* Only end level2 contention once for posix */
-                               if (!posix_level2_contention_ended) {
-                                       posix_level2_contention_ended = true;
-                                       contend_level2_oplocks_end(br_lck->fsp,
-                                           LEVEL2_CONTEND_POSIX_BRL);
-                               }
-                       }
-               }
+               struct lock_struct *lock = &br_lck->lock_data[i];
 
-               if (del_this_lock) {
-                       /* Send unlock messages to any pending waiters that overlap. */
-                       for (j=0; j < br_lck->num_locks; j++) {
-                               struct lock_struct *pend_lock = &locks[j];
+               /*
+                * as this is a durable handle we only expect locks
+                * of the current file handle!
+                */
 
-                               /* Ignore our own or non-pending locks. */
-                               if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
-                                       continue;
-                               }
+               if (lock->context.smblctx != smblctx) {
+                       TALLOC_FREE(br_lck);
+                       return false;
+               }
 
-                               /* Optimisation - don't send to this fnum as we're
-                                  closing it. */
-                               if (pend_lock->context.tid == tid &&
-                                   procid_equal(&pend_lock->context.pid, &pid) &&
-                                   pend_lock->fnum == fnum) {
-                                       continue;
-                               }
+               if (lock->context.tid != TID_FIELD_INVALID) {
+                       TALLOC_FREE(br_lck);
+                       return false;
+               }
 
-                               /* We could send specific lock info here... */
-                               if (brl_pending_overlap(lock, pend_lock)) {
-                                       messaging_send(msg_ctx, pend_lock->context.pid,
-                                                      MSG_SMB_UNLOCK, &data_blob_null);
-                               }
-                       }
+               if (!server_id_is_disconnected(&lock->context.pid)) {
+                       TALLOC_FREE(br_lck);
+                       return false;
+               }
 
-                       /* found it - delete it */
-                       if (br_lck->num_locks > 1 && i < br_lck->num_locks - 1) {
-                               memmove(&locks[i], &locks[i+1], 
-                                       sizeof(*locks)*((br_lck->num_locks-1) - i));
-                       }
-                       br_lck->num_locks--;
-                       br_lck->modified = True;
-                       i--;
-                       dcount++;
+               if (lock->fnum != FNUM_FIELD_INVALID) {
+                       TALLOC_FREE(br_lck);
+                       return false;
                }
-       }
 
-       if(lp_posix_locking(fsp->conn->params) && num_deleted_windows_locks) {
-               /* Reduce the Windows lock POSIX reference count on this dev/ino pair. */
-               reduce_windows_lock_ref_count(fsp, num_deleted_windows_locks);
+               lock->context.pid = self;
+               lock->context.tid = tid;
+               lock->fnum = fnum;
        }
+
+       fsp->current_lock_count = br_lck->num_locks;
+       br_lck->modified = true;
+       TALLOC_FREE(br_lck);
+       return true;
 }
 
 /****************************************************************************
  Ensure this set of lock entries is valid.
 ****************************************************************************/
-static bool validate_lock_entries(unsigned int *pnum_entries, struct lock_struct **pplocks)
+static bool validate_lock_entries(TALLOC_CTX *mem_ctx,
+                                 unsigned int *pnum_entries, struct lock_struct **pplocks,
+                                 bool keep_disconnected)
 {
        unsigned int i;
        unsigned int num_valid_entries = 0;
        struct lock_struct *locks = *pplocks;
+       TALLOC_CTX *frame = talloc_stackframe();
+       struct server_id *ids;
+       bool *exists;
+
+       ids = talloc_array(frame, struct server_id, *pnum_entries);
+       if (ids == NULL) {
+               DEBUG(0, ("validate_lock_entries: "
+                         "talloc_array(struct server_id, %u) failed\n",
+                         *pnum_entries));
+               talloc_free(frame);
+               return false;
+       }
+
+       exists = talloc_array(frame, bool, *pnum_entries);
+       if (exists == NULL) {
+               DEBUG(0, ("validate_lock_entries: "
+                         "talloc_array(bool, %u) failed\n",
+                         *pnum_entries));
+               talloc_free(frame);
+               return false;
+       }
 
        for (i = 0; i < *pnum_entries; i++) {
-               struct lock_struct *lock_data = &locks[i];
-               if (!process_exists(lock_data->context.pid)) {
-                       /* This process no longer exists - mark this
-                          entry as invalid by zeroing it. */
-                       ZERO_STRUCTP(lock_data);
-               } else {
+               ids[i] = locks[i].context.pid;
+       }
+
+       if (!serverids_exist(ids, *pnum_entries, exists)) {
+               DEBUG(3, ("validate_lock_entries: serverids_exists failed\n"));
+               talloc_free(frame);
+               return false;
+       }
+
+       for (i = 0; i < *pnum_entries; i++) {
+               if (exists[i]) {
                        num_valid_entries++;
+                       continue;
+               }
+
+               if (keep_disconnected &&
+                   server_id_is_disconnected(&ids[i]))
+               {
+                       num_valid_entries++;
+                       continue;
                }
+
+               /* This process no longer exists - mark this
+                  entry as invalid by zeroing it. */
+               ZERO_STRUCTP(&locks[i]);
        }
+       TALLOC_FREE(frame);
 
        if (num_valid_entries != *pnum_entries) {
                struct lock_struct *new_lock_data = NULL;
 
                if (num_valid_entries) {
-                       new_lock_data = SMB_MALLOC_ARRAY(struct lock_struct, num_valid_entries);
+                       new_lock_data = talloc_array(
+                               mem_ctx, struct lock_struct,
+                               num_valid_entries);
                        if (!new_lock_data) {
                                DEBUG(3, ("malloc fail\n"));
                                return False;
@@ -1523,7 +1765,7 @@ static bool validate_lock_entries(unsigned int *pnum_entries, struct lock_struct
                        num_valid_entries = 0;
                        for (i = 0; i < *pnum_entries; i++) {
                                struct lock_struct *lock_data = &locks[i];
-                               if (lock_data->context.smbpid &&
+                               if (lock_data->context.smblctx &&
                                                lock_data->context.tid) {
                                        /* Valid (nonzero) entry - copy it. */
                                        memcpy(&new_lock_data[num_valid_entries],
@@ -1533,7 +1775,7 @@ static bool validate_lock_entries(unsigned int *pnum_entries, struct lock_struct
                        }
                }
 
-               SAFE_FREE(*pplocks);
+               TALLOC_FREE(*pplocks);
                *pplocks = new_lock_data;
                *pnum_entries = num_valid_entries;
        }
@@ -1555,7 +1797,7 @@ struct brl_forall_cb {
  on each lock.
 ****************************************************************************/
 
-static int traverse_fn(struct db_record *rec, void *state)
+static int brl_traverse_fn(struct db_record *rec, void *state)
 {
        struct brl_forall_cb *cb = (struct brl_forall_cb *)state;
        struct lock_struct *locks;
@@ -1563,23 +1805,28 @@ static int traverse_fn(struct db_record *rec, void *state)
        unsigned int i;
        unsigned int num_locks = 0;
        unsigned int orig_num_locks = 0;
+       TDB_DATA dbkey;
+       TDB_DATA value;
+
+       dbkey = dbwrap_record_get_key(rec);
+       value = dbwrap_record_get_value(rec);
 
        /* In a traverse function we must make a copy of
           dbuf before modifying it. */
 
-       locks = (struct lock_struct *)memdup(rec->value.dptr,
-                                            rec->value.dsize);
+       locks = (struct lock_struct *)talloc_memdup(
+               talloc_tos(), value.dptr, value.dsize);
        if (!locks) {
                return -1; /* Terminate traversal. */
        }
 
-       key = (struct file_id *)rec->key.dptr;
-       orig_num_locks = num_locks = rec->value.dsize/sizeof(*locks);
+       key = (struct file_id *)dbkey.dptr;
+       orig_num_locks = num_locks = value.dsize/sizeof(*locks);
 
        /* Ensure the lock db is clean of entries from invalid processes. */
 
-       if (!validate_lock_entries(&num_locks, &locks)) {
-               SAFE_FREE(locks);
+       if (!validate_lock_entries(talloc_tos(), &num_locks, &locks, true)) {
+               TALLOC_FREE(locks);
                return -1; /* Terminate traversal */
        }
 
@@ -1588,9 +1835,9 @@ static int traverse_fn(struct db_record *rec, void *state)
                        TDB_DATA data;
                        data.dptr = (uint8_t *)locks;
                        data.dsize = num_locks*sizeof(struct lock_struct);
-                       rec->store(rec, data, TDB_REPLACE);
+                       dbwrap_record_store(rec, data, TDB_REPLACE);
                } else {
-                       rec->delete_rec(rec);
+                       dbwrap_record_delete(rec);
                }
        }
 
@@ -1606,7 +1853,7 @@ static int traverse_fn(struct db_record *rec, void *state)
                }
        }
 
-       SAFE_FREE(locks);
+       TALLOC_FREE(locks);
        return 0;
 }
 
@@ -1622,13 +1869,21 @@ int brl_forall(void (*fn)(struct file_id id, struct server_id pid,
               void *private_data)
 {
        struct brl_forall_cb cb;
+       NTSTATUS status;
+       int count = 0;
 
        if (!brlock_db) {
                return 0;
        }
        cb.fn = fn;
        cb.private_data = private_data;
-       return brlock_db->traverse(brlock_db, traverse_fn, &cb);
+       status = dbwrap_traverse(brlock_db, brl_traverse_fn, &cb, &count);
+
+       if (!NT_STATUS_IS_OK(status)) {
+               return -1;
+       } else {
+               return count;
+       }
 }
 
 /*******************************************************************
@@ -1637,19 +1892,25 @@ int brl_forall(void (*fn)(struct file_id id, struct server_id pid,
  Unlock the record.
 ********************************************************************/
 
-static int byte_range_lock_destructor(struct byte_range_lock *br_lck)
+static void byte_range_lock_flush(struct byte_range_lock *br_lck)
 {
-       if (br_lck->read_only) {
-               SMB_ASSERT(!br_lck->modified);
-       }
-
+       size_t data_len;
        if (!br_lck->modified) {
+               DEBUG(10, ("br_lck not modified\n"));
                goto done;
        }
 
-       if (br_lck->num_locks == 0) {
+       data_len = br_lck->num_locks * sizeof(struct lock_struct);
+
+       if (br_lck->have_read_oplocks) {
+               data_len += 1;
+       }
+
+       DEBUG(10, ("data_len=%d\n", (int)data_len));
+
+       if (data_len == 0) {
                /* No locks - delete this entry. */
-               NTSTATUS status = br_lck->record->delete_rec(br_lck->record);
+               NTSTATUS status = dbwrap_record_delete(br_lck->record);
                if (!NT_STATUS_IS_OK(status)) {
                        DEBUG(0, ("delete_rec returned %s\n",
                                  nt_errstr(status)));
@@ -1659,21 +1920,35 @@ static int byte_range_lock_destructor(struct byte_range_lock *br_lck)
                TDB_DATA data;
                NTSTATUS status;
 
-               data.dptr = (uint8 *)br_lck->lock_data;
-               data.dsize = br_lck->num_locks * sizeof(struct lock_struct);
+               data.dsize = data_len;
+               data.dptr = talloc_array(talloc_tos(), uint8_t, data_len);
+               SMB_ASSERT(data.dptr != NULL);
+
+               memcpy(data.dptr, br_lck->lock_data,
+                      br_lck->num_locks * sizeof(struct lock_struct));
 
-               status = br_lck->record->store(br_lck->record, data,
-                                              TDB_REPLACE);
+               if (br_lck->have_read_oplocks) {
+                       data.dptr[data_len-1] = 1;
+               }
+
+               status = dbwrap_record_store(br_lck->record, data, TDB_REPLACE);
+               TALLOC_FREE(data.dptr);
                if (!NT_STATUS_IS_OK(status)) {
                        DEBUG(0, ("store returned %s\n", nt_errstr(status)));
                        smb_panic("Could not store byte range mode entry");
                }
        }
 
- done:
+       DEBUG(10, ("seqnum=%d\n", dbwrap_get_seqnum(brlock_db)));
 
-       SAFE_FREE(br_lck->lock_data);
+ done:
+       br_lck->modified = false;
        TALLOC_FREE(br_lck->record);
+}
+
+static int byte_range_lock_destructor(struct byte_range_lock *br_lck)
+{
+       byte_range_lock_flush(br_lck);
        return 0;
 }
 
@@ -1683,11 +1958,10 @@ static int byte_range_lock_destructor(struct byte_range_lock *br_lck)
  TALLOC_FREE(brl) will release the lock in the destructor.
 ********************************************************************/
 
-static struct byte_range_lock *brl_get_locks_internal(TALLOC_CTX *mem_ctx,
-                                       files_struct *fsp, bool read_only)
+struct byte_range_lock *brl_get_locks(TALLOC_CTX *mem_ctx, files_struct *fsp)
 {
        TDB_DATA key, data;
-       struct byte_range_lock *br_lck = TALLOC_P(mem_ctx, struct byte_range_lock);
+       struct byte_range_lock *br_lck = talloc(mem_ctx, struct byte_range_lock);
 
        if (br_lck == NULL) {
                return NULL;
@@ -1695,40 +1969,22 @@ static struct byte_range_lock *brl_get_locks_internal(TALLOC_CTX *mem_ctx,
 
        br_lck->fsp = fsp;
        br_lck->num_locks = 0;
+       br_lck->have_read_oplocks = false;
        br_lck->modified = False;
-       memset(&br_lck->key, '\0', sizeof(struct file_id));
-       br_lck->key = fsp->file_id;
 
-       key.dptr = (uint8 *)&br_lck->key;
+       key.dptr = (uint8 *)&fsp->file_id;
        key.dsize = sizeof(struct file_id);
 
-       if (!fsp->lockdb_clean) {
-               /* We must be read/write to clean
-                  the dead entries. */
-               read_only = False;
-       }
+       br_lck->record = dbwrap_fetch_locked(brlock_db, br_lck, key);
 
-       if (read_only) {
-               if (brlock_db->fetch(brlock_db, br_lck, key, &data) == -1) {
-                       DEBUG(3, ("Could not fetch byte range lock record\n"));
-                       TALLOC_FREE(br_lck);
-                       return NULL;
-               }
-               br_lck->record = NULL;
+       if (br_lck->record == NULL) {
+               DEBUG(3, ("Could not lock byte range lock entry\n"));
+               TALLOC_FREE(br_lck);
+               return NULL;
        }
-       else {
-               br_lck->record = brlock_db->fetch_locked(brlock_db, br_lck, key);
-
-               if (br_lck->record == NULL) {
-                       DEBUG(3, ("Could not lock byte range lock entry\n"));
-                       TALLOC_FREE(br_lck);
-                       return NULL;
-               }
 
-               data = br_lck->record->value;
-       }
+       data = dbwrap_record_get_value(br_lck->record);
 
-       br_lck->read_only = read_only;
        br_lck->lock_data = NULL;
 
        talloc_set_destructor(br_lck, byte_range_lock_destructor);
@@ -1736,27 +1992,42 @@ static struct byte_range_lock *brl_get_locks_internal(TALLOC_CTX *mem_ctx,
        br_lck->num_locks = data.dsize / sizeof(struct lock_struct);
 
        if (br_lck->num_locks != 0) {
-               br_lck->lock_data = SMB_MALLOC_ARRAY(struct lock_struct,
-                                                    br_lck->num_locks);
+               br_lck->lock_data = talloc_array(
+                       br_lck, struct lock_struct, br_lck->num_locks);
                if (br_lck->lock_data == NULL) {
                        DEBUG(0, ("malloc failed\n"));
                        TALLOC_FREE(br_lck);
                        return NULL;
                }
 
-               memcpy(br_lck->lock_data, data.dptr, data.dsize);
+               memcpy(br_lck->lock_data, data.dptr,
+                      talloc_get_size(br_lck->lock_data));
        }
-       
+
+       DEBUG(10, ("data.dsize=%d\n", (int)data.dsize));
+
+       if ((data.dsize % sizeof(struct lock_struct)) == 1) {
+               br_lck->have_read_oplocks = (data.dptr[data.dsize-1] == 1);
+       }
+
        if (!fsp->lockdb_clean) {
                int orig_num_locks = br_lck->num_locks;
 
-               /* This is the first time we've accessed this. */
-               /* Go through and ensure all entries exist - remove any that don't. */
-               /* Makes the lockdb self cleaning at low cost. */
+               /*
+                * This is the first time we access the byte range lock
+                * record with this fsp. Go through and ensure all entries
+                * are valid - remove any that don't.
+                * This makes the lockdb self cleaning at low cost.
+                *
+                * Note: Disconnected entries belong to disconnected
+                * durable handles. So at this point, we have a new
+                * handle on the file and the disconnected durable has
+                * already been closed (we are not a durable reconnect).
+                * So we need to clean the disconnected brl entry.
+                */
 
-               if (!validate_lock_entries(&br_lck->num_locks,
-                                          &br_lck->lock_data)) {
-                       SAFE_FREE(br_lck->lock_data);
+               if (!validate_lock_entries(br_lck, &br_lck->num_locks,
+                                          &br_lck->lock_data, false)) {
                        TALLOC_FREE(br_lck);
                        return NULL;
                }
@@ -1780,19 +2051,137 @@ static struct byte_range_lock *brl_get_locks_internal(TALLOC_CTX *mem_ctx,
                        print_lock_struct(i, &locks[i]);
                }
        }
+
        return br_lck;
 }
 
-struct byte_range_lock *brl_get_locks(TALLOC_CTX *mem_ctx,
-                                       files_struct *fsp)
+struct brl_get_locks_readonly_state {
+       TALLOC_CTX *mem_ctx;
+       struct byte_range_lock **br_lock;
+};
+
+static void brl_get_locks_readonly_parser(TDB_DATA key, TDB_DATA data,
+                                         void *private_data)
 {
-       return brl_get_locks_internal(mem_ctx, fsp, False);
+       struct brl_get_locks_readonly_state *state =
+               (struct brl_get_locks_readonly_state *)private_data;
+       struct byte_range_lock *br_lock;
+
+       br_lock = talloc_pooled_object(
+               state->mem_ctx, struct byte_range_lock, 1, data.dsize);
+       if (br_lock == NULL) {
+               *state->br_lock = NULL;
+               return;
+       }
+       br_lock->lock_data = (struct lock_struct *)talloc_memdup(
+               br_lock, data.dptr, data.dsize);
+       br_lock->num_locks = data.dsize / sizeof(struct lock_struct);
+
+       if ((data.dsize % sizeof(struct lock_struct)) == 1) {
+               br_lock->have_read_oplocks = (data.dptr[data.dsize-1] == 1);
+       }
+
+       DEBUG(10, ("Got %d bytes, have_read_oplocks: %s\n", (int)data.dsize,
+                  br_lock->have_read_oplocks ? "true" : "false"));
+
+       *state->br_lock = br_lock;
 }
 
-struct byte_range_lock *brl_get_locks_readonly(TALLOC_CTX *mem_ctx,
-                                       files_struct *fsp)
+struct byte_range_lock *brl_get_locks_readonly(files_struct *fsp)
 {
-       return brl_get_locks_internal(mem_ctx, fsp, True);
+       struct byte_range_lock *br_lock = NULL;
+       struct byte_range_lock *rw = NULL;
+
+       DEBUG(10, ("seqnum=%d, fsp->brlock_seqnum=%d\n",
+                  dbwrap_get_seqnum(brlock_db), fsp->brlock_seqnum));
+
+       if ((fsp->brlock_rec != NULL)
+           && (dbwrap_get_seqnum(brlock_db) == fsp->brlock_seqnum)) {
+               /*
+                * We have cached the brlock_rec and the database did not
+                * change.
+                */
+               return fsp->brlock_rec;
+       }
+
+       if (!fsp->lockdb_clean) {
+               /*
+                * Fetch the record in R/W mode to give validate_lock_entries
+                * a chance to kick in once.
+                */
+               rw = brl_get_locks(talloc_tos(), fsp);
+               if (rw == NULL) {
+                       return NULL;
+               }
+               fsp->lockdb_clean = true;
+       }
+
+       if (rw != NULL) {
+               size_t lock_data_size;
+
+               /*
+                * Make a copy of the already retrieved and sanitized rw record
+                */
+               lock_data_size = rw->num_locks * sizeof(struct lock_struct);
+               br_lock = talloc_pooled_object(
+                       fsp, struct byte_range_lock, 1, lock_data_size);
+               if (br_lock == NULL) {
+                       goto fail;
+               }
+               br_lock->have_read_oplocks = rw->have_read_oplocks;
+               br_lock->num_locks = rw->num_locks;
+               br_lock->lock_data = (struct lock_struct *)talloc_memdup(
+                       br_lock, rw->lock_data, lock_data_size);
+       } else {
+               struct brl_get_locks_readonly_state state;
+               NTSTATUS status;
+
+               /*
+                * Parse the record fresh from the database
+                */
+
+               state.mem_ctx = fsp;
+               state.br_lock = &br_lock;
+
+               status = dbwrap_parse_record(
+                       brlock_db,
+                       make_tdb_data((uint8_t *)&fsp->file_id,
+                                     sizeof(fsp->file_id)),
+                       brl_get_locks_readonly_parser, &state);
+               if (!NT_STATUS_IS_OK(status)) {
+                       DEBUG(3, ("Could not parse byte range lock record: "
+                                 "%s\n", nt_errstr(status)));
+                       goto fail;
+               }
+               if (br_lock == NULL) {
+                       goto fail;
+               }
+       }
+
+       br_lock->fsp = fsp;
+       br_lock->modified = false;
+       br_lock->record = NULL;
+
+       if (lp_clustering()) {
+               /*
+                * In the cluster case we can't cache the brlock struct
+                * because dbwrap_get_seqnum does not work reliably over
+                * ctdb. Thus we have to throw away the brlock struct soon.
+                */
+               talloc_steal(talloc_tos(), br_lock);
+       } else {
+               /*
+                * Cache the brlock struct, invalidated when the dbwrap_seqnum
+                * changes. See beginning of this routine.
+                */
+               TALLOC_FREE(fsp->brlock_rec);
+               fsp->brlock_rec = br_lock;
+               fsp->brlock_seqnum = dbwrap_get_seqnum(brlock_db);
+       }
+
+fail:
+       TALLOC_FREE(rw);
+       return br_lock;
 }
 
 struct brl_revalidate_state {
@@ -1829,8 +2218,8 @@ static void brl_revalidate_collect(struct file_id id, struct server_id pid,
 
 static int compare_procids(const void *p1, const void *p2)
 {
-       const struct server_id *i1 = (struct server_id *)p1;
-       const struct server_id *i2 = (struct server_id *)p2;
+       const struct server_id *i1 = (const struct server_id *)p1;
+       const struct server_id *i2 = (const struct server_id *)p2;
 
        if (i1->pid < i2->pid) return -1;
        if (i2->pid > i2->pid) return 1;
@@ -1846,17 +2235,17 @@ static int compare_procids(const void *p1, const void *p2)
  * array, then qsort that array and only send to non-dupes.
  */
 
-static void brl_revalidate(struct messaging_context *msg_ctx,
-                          void *private_data,
-                          uint32_t msg_type,
-                          struct server_id server_id,
-                          DATA_BLOB *data)
+void brl_revalidate(struct messaging_context *msg_ctx,
+                   void *private_data,
+                   uint32_t msg_type,
+                   struct server_id server_id,
+                   DATA_BLOB *data)
 {
        struct brl_revalidate_state *state;
        uint32 i;
        struct server_id last_pid;
 
-       if (!(state = TALLOC_ZERO_P(NULL, struct brl_revalidate_state))) {
+       if (!(state = talloc_zero(NULL, struct brl_revalidate_state))) {
                DEBUG(0, ("talloc failed\n"));
                return;
        }
@@ -1872,13 +2261,12 @@ static void brl_revalidate(struct messaging_context *msg_ctx,
                goto done;
        }
 
-       qsort(state->pids, state->num_pids, sizeof(state->pids[0]),
-             compare_procids);
+       TYPESAFE_QSORT(state->pids, state->num_pids, compare_procids);
 
        ZERO_STRUCT(last_pid);
 
        for (i=0; i<state->num_pids; i++) {
-               if (procid_equal(&last_pid, &state->pids[i])) {
+               if (serverid_equal(&last_pid, &state->pids[i])) {
                        /*
                         * We've seen that one already
                         */
@@ -1895,8 +2283,74 @@ static void brl_revalidate(struct messaging_context *msg_ctx,
        return;
 }
 
-void brl_register_msgs(struct messaging_context *msg_ctx)
+bool brl_cleanup_disconnected(struct file_id fid, uint64_t open_persistent_id)
 {
-       messaging_register(msg_ctx, NULL, MSG_SMB_BRL_VALIDATE,
-                          brl_revalidate);
+       bool ret = false;
+       TALLOC_CTX *frame = talloc_stackframe();
+       TDB_DATA key, val;
+       struct db_record *rec;
+       struct lock_struct *lock;
+       unsigned n, num;
+       NTSTATUS status;
+
+       key = make_tdb_data((void*)&fid, sizeof(fid));
+
+       rec = dbwrap_fetch_locked(brlock_db, frame, key);
+       if (rec == NULL) {
+               DEBUG(5, ("brl_cleanup_disconnected: failed to fetch record "
+                         "for file %s\n", file_id_string(frame, &fid)));
+               goto done;
+       }
+
+       val = dbwrap_record_get_value(rec);
+       lock = (struct lock_struct*)val.dptr;
+       num = val.dsize / sizeof(struct lock_struct);
+       if (lock == NULL) {
+               DEBUG(10, ("brl_cleanup_disconnected: no byte range locks for "
+                          "file %s\n", file_id_string(frame, &fid)));
+               ret = true;
+               goto done;
+       }
+
+       for (n=0; n<num; n++) {
+               struct lock_context *ctx = &lock[n].context;
+
+               if (!server_id_is_disconnected(&ctx->pid)) {
+                       DEBUG(5, ("brl_cleanup_disconnected: byte range lock "
+                                 "%s used by server %s, do not cleanup\n",
+                                 file_id_string(frame, &fid),
+                                 server_id_str(frame, &ctx->pid)));
+                       goto done;
+               }
+
+               if (ctx->smblctx != open_persistent_id) {
+                       DEBUG(5, ("brl_cleanup_disconnected: byte range lock "
+                                 "%s expected smblctx %llu but found %llu"
+                                 ", do not cleanup\n",
+                                 file_id_string(frame, &fid),
+                                 (unsigned long long)open_persistent_id,
+                                 (unsigned long long)ctx->smblctx));
+                       goto done;
+               }
+       }
+
+       status = dbwrap_record_delete(rec);
+       if (!NT_STATUS_IS_OK(status)) {
+               DEBUG(5, ("brl_cleanup_disconnected: failed to delete record "
+                         "for file %s from %s, open %llu: %s\n",
+                         file_id_string(frame, &fid), dbwrap_name(brlock_db),
+                         (unsigned long long)open_persistent_id,
+                         nt_errstr(status)));
+               goto done;
+       }
+
+       DEBUG(10, ("brl_cleanup_disconnected: "
+                  "file %s cleaned up %u entries from open %llu\n",
+                  file_id_string(frame, &fid), num,
+                  (unsigned long long)open_persistent_id));
+
+       ret = true;
+done:
+       talloc_free(frame);
+       return ret;
 }