r22846: Chunk one to replace message_send_pid with messaging_send: Deep inside
[sfrench/samba-autobuild/.git] / source / locking / locking.c
index 666988b5413e176612c5042b17c59c5c6bddbe04..28a7fab97d3fd2c0a30fc783adba92df404b77c2 100644 (file)
@@ -2,7 +2,7 @@
    Unix SMB/CIFS implementation.
    Locking functions
    Copyright (C) Andrew Tridgell 1992-2000
-   Copyright (C) Jeremy Allison 1992-2000
+   Copyright (C) Jeremy Allison 1992-2006
    Copyright (C) Volker Lendecke 2005
    
    This program is free software; you can redistribute it and/or modify
    rewrtten completely to use new tdb code. Tridge, Dec '99
 
    Added POSIX locking support. Jeremy Allison (jeremy@valinux.com), Apr. 2000.
+   Added Unix Extensions POSIX locking support. Jeremy Allison Mar 2006.
 */
 
 #include "includes.h"
-uint16 global_smbpid;
 
 #undef DBGC_CLASS
 #define DBGC_CLASS DBGC_LOCKING
@@ -44,237 +44,293 @@ uint16 global_smbpid;
 /* the locking database handle */
 static TDB_CONTEXT *tdb;
 
-struct locking_data {
-        union {
-               struct {
-                       int num_share_mode_entries;
-                       BOOL delete_on_close;
-                       BOOL initial_delete_on_close; /* Only set at NTCreateX if file was created. */
-                       uint32 delete_token_size; /* Only valid if either of
-                                                       the two previous fields 
-                                                       are True. */
-               } s;
-                struct share_mode_entry dummy; /* Needed for alignment. */
-        } u;
-        /* the following two entries are implicit
-           struct share_mode_entry modes[num_share_mode_entries];
-           char file_name[];
-        */
-};
-
 /****************************************************************************
- Debugging aid :-).
+ Debugging aids :-).
 ****************************************************************************/
 
-static const char *lock_type_name(enum brl_type lock_type)
+const char *lock_type_name(enum brl_type lock_type)
 {
-       return (lock_type == READ_LOCK) ? "READ" : "WRITE";
+       switch (lock_type) {
+               case READ_LOCK:
+                       return "READ";
+               case WRITE_LOCK:
+                       return "WRITE";
+               case PENDING_READ_LOCK:
+                       return "PENDING_READ";
+               case PENDING_WRITE_LOCK:
+                       return "PENDING_WRITE";
+               default:
+                       return "other";
+       }
+}
+
+const char *lock_flav_name(enum brl_flavour lock_flav)
+{
+       return (lock_flav == WINDOWS_LOCK) ? "WINDOWS_LOCK" : "POSIX_LOCK";
 }
 
 /****************************************************************************
  Utility function called to see if a file region is locked.
+ Called in the read/write codepath.
 ****************************************************************************/
 
-BOOL is_locked(files_struct *fsp,connection_struct *conn,
-              SMB_BIG_UINT count,SMB_BIG_UINT offset, 
-              enum brl_type lock_type)
+BOOL is_locked(files_struct *fsp,
+               uint32 smbpid,
+               SMB_BIG_UINT count,
+               SMB_BIG_UINT offset, 
+               enum brl_type lock_type)
 {
-       int snum = SNUM(conn);
-       int strict_locking = lp_strict_locking(snum);
-       BOOL ret;
+       int strict_locking = lp_strict_locking(fsp->conn->params);
+       enum brl_flavour lock_flav = lp_posix_cifsu_locktype(fsp);
+       BOOL ret = True;
        
-       if (count == 0)
-               return(False);
+       if (count == 0) {
+               return False;
+       }
 
-       if (!lp_locking(snum) || !strict_locking)
-               return(False);
+       if (!lp_locking(fsp->conn->params) || !strict_locking) {
+               return False;
+       }
 
        if (strict_locking == Auto) {
                if  (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && (lock_type == READ_LOCK || lock_type == WRITE_LOCK)) {
                        DEBUG(10,("is_locked: optimisation - exclusive oplock on file %s\n", fsp->fsp_name ));
-                       ret = 0;
+                       ret = False;
                } else if ((fsp->oplock_type == LEVEL_II_OPLOCK) &&
                           (lock_type == READ_LOCK)) {
                        DEBUG(10,("is_locked: optimisation - level II oplock on file %s\n", fsp->fsp_name ));
-                       ret = 0;
+                       ret = False;
                } else {
-                       ret = !brl_locktest(fsp->dev, fsp->inode, fsp->fnum,
-                                    global_smbpid, procid_self(), conn->cnum, 
-                                    offset, count, lock_type);
+                       struct byte_range_lock *br_lck = brl_get_locks_readonly(NULL, fsp);
+                       if (!br_lck) {
+                               return False;
+                       }
+                       ret = !brl_locktest(br_lck,
+                                       smbpid,
+                                       procid_self(),
+                                       offset,
+                                       count,
+                                       lock_type,
+                                       lock_flav);
+                       TALLOC_FREE(br_lck);
                }
        } else {
-               ret = !brl_locktest(fsp->dev, fsp->inode, fsp->fnum,
-                               global_smbpid, procid_self(), conn->cnum,
-                               offset, count, lock_type);
+               struct byte_range_lock *br_lck = brl_get_locks_readonly(NULL, fsp);
+               if (!br_lck) {
+                       return False;
+               }
+               ret = !brl_locktest(br_lck,
+                               smbpid,
+                               procid_self(),
+                               offset,
+                               count,
+                               lock_type,
+                               lock_flav);
+               TALLOC_FREE(br_lck);
        }
 
-       DEBUG(10,("is_locked: brl start=%.0f len=%.0f %s for file %s\n",
+       DEBUG(10,("is_locked: flavour = %s brl start=%.0f len=%.0f %s for fnum %d file %s\n",
+                       lock_flav_name(lock_flav),
                        (double)offset, (double)count, ret ? "locked" : "unlocked",
-                       fsp->fsp_name ));
-
-       /*
-        * There is no lock held by an SMB daemon, check to
-        * see if there is a POSIX lock from a UNIX or NFS process.
-        */
-
-       if(!ret && lp_posix_locking(snum)) {
-               ret = is_posix_locked(fsp, offset, count, lock_type);
-
-               DEBUG(10,("is_locked: posix start=%.0f len=%.0f %s for file %s\n",
-                               (double)offset, (double)count, ret ? "locked" : "unlocked",
-                               fsp->fsp_name ));
-       }
+                       fsp->fnum, fsp->fsp_name ));
 
        return ret;
 }
 
 /****************************************************************************
Utility function called by locking requests.
Find out if a lock could be granted - return who is blocking us if we can't.
 ****************************************************************************/
 
-static NTSTATUS do_lock(files_struct *fsp,connection_struct *conn, uint16 lock_pid,
-                SMB_BIG_UINT count,SMB_BIG_UINT offset,enum brl_type lock_type, BOOL *my_lock_ctx)
+NTSTATUS query_lock(files_struct *fsp,
+                       uint32 *psmbpid,
+                       SMB_BIG_UINT *pcount,
+                       SMB_BIG_UINT *poffset,
+                       enum brl_type *plock_type,
+                       enum brl_flavour lock_flav)
 {
+       struct byte_range_lock *br_lck = NULL;
        NTSTATUS status = NT_STATUS_LOCK_NOT_GRANTED;
 
-       if (!lp_locking(SNUM(conn)))
-               return NT_STATUS_OK;
+       if (!fsp->can_lock) {
+               return fsp->is_directory ? NT_STATUS_INVALID_DEVICE_REQUEST : NT_STATUS_INVALID_HANDLE;
+       }
 
-       /* NOTE! 0 byte long ranges ARE allowed and should be stored  */
+       if (!lp_locking(fsp->conn->params)) {
+               return NT_STATUS_OK;
+       }
 
-       DEBUG(10,("do_lock: lock type %s start=%.0f len=%.0f requested for file %s\n",
-                 lock_type_name(lock_type), (double)offset, (double)count, fsp->fsp_name ));
-
-       if (OPEN_FSP(fsp) && fsp->can_lock && (fsp->conn == conn)) {
-               status = brl_lock(fsp->dev, fsp->inode, fsp->fnum,
-                                 lock_pid, procid_self(), conn->cnum, 
-                                 offset, count, 
-                                 lock_type, my_lock_ctx);
-
-               if (NT_STATUS_IS_OK(status) && lp_posix_locking(SNUM(conn))) {
-
-                       /*
-                        * Try and get a POSIX lock on this range.
-                        * Note that this is ok if it is a read lock
-                        * overlapping on a different fd. JRA.
-                        */
-
-                       if (!set_posix_lock(fsp, offset, count, lock_type)) {
-                               if (errno == EACCES || errno == EAGAIN)
-                                       status = NT_STATUS_FILE_LOCK_CONFLICT;
-                               else
-                                       status = map_nt_error_from_unix(errno);
-
-                               /*
-                                * We failed to map - we must now remove the brl
-                                * lock entry.
-                                */
-                               (void)brl_unlock(fsp->dev, fsp->inode, fsp->fnum,
-                                                               lock_pid, procid_self(), conn->cnum, 
-                                                               offset, count, False,
-                                                               NULL, NULL);
-                       }
-               }
+       br_lck = brl_get_locks_readonly(NULL, fsp);
+       if (!br_lck) {
+               return NT_STATUS_NO_MEMORY;
        }
 
+       status = brl_lockquery(br_lck,
+                       psmbpid,
+                       procid_self(),
+                       poffset,
+                       pcount,
+                       plock_type,
+                       lock_flav);
+
+       TALLOC_FREE(br_lck);
        return status;
 }
 
 /****************************************************************************
- Utility function called by locking requests. This is *DISGUSTING*. It also
- appears to be "What Windows Does" (tm). Andrew, ever wonder why Windows 2000
- is so slow on the locking tests...... ? This is the reason. Much though I hate
- it, we need this. JRA.
+ Utility function called by locking requests.
 ****************************************************************************/
 
-NTSTATUS do_lock_spin(files_struct *fsp,connection_struct *conn, uint16 lock_pid,
-                SMB_BIG_UINT count,SMB_BIG_UINT offset,enum brl_type lock_type, BOOL *my_lock_ctx)
+struct byte_range_lock *do_lock(struct messaging_context *msg_ctx,
+                       files_struct *fsp,
+                       uint32 lock_pid,
+                       SMB_BIG_UINT count,
+                       SMB_BIG_UINT offset,
+                       enum brl_type lock_type,
+                       enum brl_flavour lock_flav,
+                       BOOL blocking_lock,
+                       NTSTATUS *perr)
 {
-       int j, maxj = lp_lock_spin_count();
-       int sleeptime = lp_lock_sleep_time();
-       NTSTATUS status, ret;
+       struct byte_range_lock *br_lck = NULL;
+
+       if (!fsp->can_lock) {
+               *perr = fsp->is_directory ? NT_STATUS_INVALID_DEVICE_REQUEST : NT_STATUS_INVALID_HANDLE;
+               return NULL;
+       }
 
-       if (maxj <= 0)
-               maxj = 1;
+       if (!lp_locking(fsp->conn->params)) {
+               *perr = NT_STATUS_OK;
+               return NULL;
+       }
 
-       ret = NT_STATUS_OK; /* to keep dumb compilers happy */
+       /* NOTE! 0 byte long ranges ARE allowed and should be stored  */
 
-       for (j = 0; j < maxj; j++) {
-               status = do_lock(fsp, conn, lock_pid, count, offset, lock_type, my_lock_ctx);
-               if (!NT_STATUS_EQUAL(status, NT_STATUS_LOCK_NOT_GRANTED) &&
-                   !NT_STATUS_EQUAL(status, NT_STATUS_FILE_LOCK_CONFLICT)) {
-                       return status;
-               }
-               /* if we do fail then return the first error code we got */
-               if (j == 0) {
-                       ret = status;
-                       /* Don't spin if we blocked ourselves. */
-                       if (*my_lock_ctx)
-                               return ret;
-               }
-               if (sleeptime)
-                       sys_usleep(sleeptime);
+       DEBUG(10,("do_lock: lock flavour %s lock type %s start=%.0f len=%.0f requested for fnum %d file %s\n",
+               lock_flav_name(lock_flav), lock_type_name(lock_type),
+               (double)offset, (double)count, fsp->fnum, fsp->fsp_name ));
+
+       br_lck = brl_get_locks(NULL, fsp);
+       if (!br_lck) {
+               *perr = NT_STATUS_NO_MEMORY;
+               return NULL;
        }
-       return ret;
-}
 
-/* Struct passed to brl_unlock. */
-struct posix_unlock_data_struct {
-       files_struct *fsp;
-       SMB_BIG_UINT offset;
-       SMB_BIG_UINT count;
-};
+       *perr = brl_lock(msg_ctx,
+                       br_lck,
+                       lock_pid,
+                       procid_self(),
+                       offset,
+                       count, 
+                       lock_type,
+                       lock_flav,
+                       blocking_lock);
 
-/****************************************************************************
- Function passed to brl_unlock to allow POSIX unlock to be done first.
-****************************************************************************/
+       /* blocking ie. pending, locks also count here,
+        * as this is an efficiency counter to avoid checking
+        * the lock db. on close. JRA. */
 
-static void posix_unlock(void *pre_data)
-{
-       struct posix_unlock_data_struct *pdata = (struct posix_unlock_data_struct *)pre_data;
+       fsp->current_lock_count++;
 
-       if (lp_posix_locking(SNUM(pdata->fsp->conn)))
-               release_posix_lock(pdata->fsp, pdata->offset, pdata->count);
+       return br_lck;
 }
 
 /****************************************************************************
  Utility function called by unlocking requests.
 ****************************************************************************/
 
-NTSTATUS do_unlock(files_struct *fsp,connection_struct *conn, uint16 lock_pid,
-                  SMB_BIG_UINT count,SMB_BIG_UINT offset)
+NTSTATUS do_unlock(struct messaging_context *msg_ctx,
+                       files_struct *fsp,
+                       uint32 lock_pid,
+                       SMB_BIG_UINT count,
+                       SMB_BIG_UINT offset,
+                       enum brl_flavour lock_flav)
 {
        BOOL ok = False;
-       struct posix_unlock_data_struct posix_data;
+       struct byte_range_lock *br_lck = NULL;
        
-       if (!lp_locking(SNUM(conn)))
-               return NT_STATUS_OK;
+       if (!fsp->can_lock) {
+               return fsp->is_directory ? NT_STATUS_INVALID_DEVICE_REQUEST : NT_STATUS_INVALID_HANDLE;
+       }
        
-       if (!OPEN_FSP(fsp) || !fsp->can_lock || (fsp->conn != conn)) {
-               return NT_STATUS_INVALID_HANDLE;
+       if (!lp_locking(fsp->conn->params)) {
+               return NT_STATUS_OK;
        }
        
-       DEBUG(10,("do_unlock: unlock start=%.0f len=%.0f requested for file %s\n",
-                 (double)offset, (double)count, fsp->fsp_name ));
+       DEBUG(10,("do_unlock: unlock start=%.0f len=%.0f requested for fnum %d file %s\n",
+                 (double)offset, (double)count, fsp->fnum, fsp->fsp_name ));
 
-       /*
-        * Remove the existing lock record from the tdb lockdb
-        * before looking at POSIX locks. If this record doesn't
-        * match then don't bother looking to remove POSIX locks.
-        */
-
-       posix_data.fsp = fsp;
-       posix_data.offset = offset;
-       posix_data.count = count;
+       br_lck = brl_get_locks(NULL, fsp);
+       if (!br_lck) {
+               return NT_STATUS_NO_MEMORY;
+       }
 
-       ok = brl_unlock(fsp->dev, fsp->inode, fsp->fnum,
-                       lock_pid, procid_self(), conn->cnum, offset, count,
-                       False, posix_unlock, (void *)&posix_data);
+       ok = brl_unlock(msg_ctx,
+                       br_lck,
+                       lock_pid,
+                       procid_self(),
+                       offset,
+                       count,
+                       lock_flav);
    
+       TALLOC_FREE(br_lck);
+
        if (!ok) {
                DEBUG(10,("do_unlock: returning ERRlock.\n" ));
                return NT_STATUS_RANGE_NOT_LOCKED;
        }
+
+       SMB_ASSERT(fsp->current_lock_count > 0);
+       fsp->current_lock_count--;
+
+       return NT_STATUS_OK;
+}
+
+/****************************************************************************
+ Cancel any pending blocked locks.
+****************************************************************************/
+
+NTSTATUS do_lock_cancel(files_struct *fsp,
+                       uint32 lock_pid,
+                       SMB_BIG_UINT count,
+                       SMB_BIG_UINT offset,
+                       enum brl_flavour lock_flav)
+{
+       BOOL ok = False;
+       struct byte_range_lock *br_lck = NULL;
+       
+       if (!fsp->can_lock) {
+               return fsp->is_directory ?
+                       NT_STATUS_INVALID_DEVICE_REQUEST : NT_STATUS_INVALID_HANDLE;
+       }
+       
+       if (!lp_locking(fsp->conn->params)) {
+               return NT_STATUS_DOS(ERRDOS, ERRcancelviolation);
+       }
+
+       DEBUG(10,("do_lock_cancel: cancel start=%.0f len=%.0f requested for fnum %d file %s\n",
+                 (double)offset, (double)count, fsp->fnum, fsp->fsp_name ));
+
+       br_lck = brl_get_locks(NULL, fsp);
+       if (!br_lck) {
+               return NT_STATUS_NO_MEMORY;
+       }
+
+       ok = brl_lock_cancel(br_lck,
+                       lock_pid,
+                       procid_self(),
+                       offset,
+                       count,
+                       lock_flav);
+   
+       TALLOC_FREE(br_lck);
+
+       if (!ok) {
+               DEBUG(10,("do_lock_cancel: returning ERRcancelviolation.\n" ));
+               return NT_STATUS_DOS(ERRDOS, ERRcancelviolation);
+       }
+
+       SMB_ASSERT(fsp->current_lock_count > 0);
+       fsp->current_lock_count--;
+
        return NT_STATUS_OK;
 }
 
@@ -282,26 +338,29 @@ NTSTATUS do_unlock(files_struct *fsp,connection_struct *conn, uint16 lock_pid,
  Remove any locks on this fd. Called from file_close().
 ****************************************************************************/
 
-void locking_close_file(files_struct *fsp)
+void locking_close_file(struct messaging_context *msg_ctx,
+                       files_struct *fsp)
 {
-       struct process_id pid = procid_self();
+       struct byte_range_lock *br_lck;
 
-       if (!lp_locking(SNUM(fsp->conn)))
+       if (!lp_locking(fsp->conn->params)) {
                return;
+       }
 
-       /*
-        * Just release all the brl locks, no need to release individually.
+       /* If we have not outstanding locks or pending
+        * locks then we don't need to look in the lock db.
         */
 
-       brl_close(fsp->dev, fsp->inode, pid, fsp->conn->cnum, fsp->fnum);
-
-       if(lp_posix_locking(SNUM(fsp->conn))) {
+       if (fsp->current_lock_count == 0) {
+               return;
+       }
 
-               /* 
-                * Release all the POSIX locks.
-                */
-               posix_locking_close_file(fsp);
+       br_lck = brl_get_locks(NULL,fsp);
 
+       if (br_lck) {
+               cancel_pending_lock_requests_by_fid(fsp, br_lck);
+               brl_close_fnum(msg_ctx, br_lck);
+               TALLOC_FREE(br_lck);
        }
 }
 
@@ -319,7 +378,8 @@ BOOL locking_init(int read_only)
                return True;
 
        tdb = tdb_open_log(lock_path("locking.tdb"), 
-                       SMB_OPEN_DATABASE_TDB_HASH_SIZE, TDB_DEFAULT|(read_only?0x0:TDB_CLEAR_IF_FIRST), 
+                       lp_open_files_db_hash_size(),
+                       TDB_DEFAULT|(read_only?0x0:TDB_CLEAR_IF_FIRST), 
                        read_only?O_RDONLY:O_RDWR|O_CREAT,
                        0644);
 
@@ -328,6 +388,9 @@ BOOL locking_init(int read_only)
                return False;
        }
 
+       /* Activate the per-hashchain freelist */
+       tdb_set_max_dead(tdb, 5);
+
        if (!posix_locking_init(read_only))
                return False;
 
@@ -375,7 +438,7 @@ static TDB_DATA locking_key(SMB_DEV_T dev, SMB_INO_T inode)
        memset(&key, '\0', sizeof(key));
        key.dev = dev;
        key.ino = inode;
-       kbuf.dptr = (char *)&key;
+       kbuf.dptr = (uint8 *)&key;
        kbuf.dsize = sizeof(key);
        return kbuf;
 }
@@ -391,12 +454,13 @@ char *share_mode_str(int num, struct share_mode_entry *e)
        slprintf(share_str, sizeof(share_str)-1, "share_mode_entry[%d]: %s "
                 "pid = %s, share_access = 0x%x, private_options = 0x%x, "
                 "access_mask = 0x%x, mid = 0x%x, type= 0x%x, file_id = %lu, "
-                "dev = 0x%x, inode = %.0f",
+                "uid = %u, flags = %u, dev = 0x%x, inode = %.0f",
                 num,
                 e->op_type == UNUSED_SHARE_MODE_ENTRY ? "UNUSED" : "",
                 procid_str_static(&e->pid),
                 e->share_access, e->private_options,
                 e->access_mask, e->op_mid, e->op_type, e->share_file_id,
+                (unsigned int)e->uid, (unsigned int)e->flags,
                 (unsigned int)e->dev, (double)e->inode );
 
        return share_str;
@@ -438,14 +502,11 @@ static BOOL parse_share_modes(TDB_DATA dbuf, struct share_mode_lock *lck)
        data = (struct locking_data *)dbuf.dptr;
 
        lck->delete_on_close = data->u.s.delete_on_close;
-       lck->initial_delete_on_close = data->u.s.initial_delete_on_close;
        lck->num_share_modes = data->u.s.num_share_mode_entries;
 
        DEBUG(10, ("parse_share_modes: delete_on_close: %d, "
-                  "initial_delete_on_close: %d, "
                   "num_share_modes: %d\n",
                lck->delete_on_close,
-               lck->initial_delete_on_close,
                lck->num_share_modes));
 
        if ((lck->num_share_modes < 0) || (lck->num_share_modes > 1000000)) {
@@ -464,9 +525,10 @@ static BOOL parse_share_modes(TDB_DATA dbuf, struct share_mode_lock *lck)
                        smb_panic("PANIC: parse_share_modes: buffer too short.\n");
                }
                                  
-               lck->share_modes = talloc_memdup(lck, dbuf.dptr+sizeof(*data),
-                                                lck->num_share_modes *
-                                                sizeof(struct share_mode_entry));
+               lck->share_modes = (struct share_mode_entry *)
+                       TALLOC_MEMDUP(lck, dbuf.dptr+sizeof(*data),
+                                     lck->num_share_modes *
+                                     sizeof(struct share_mode_entry));
 
                if (lck->share_modes == NULL) {
                        smb_panic("talloc failed\n");
@@ -475,13 +537,12 @@ static BOOL parse_share_modes(TDB_DATA dbuf, struct share_mode_lock *lck)
 
        /* Get any delete token. */
        if (data->u.s.delete_token_size) {
-               /* Each uid/gid is stored as a 4 byte value. */
-               uint32 val;
-               uint32 *p = (uint32 *)(dbuf.dptr + sizeof(*data) +
+               uint8 *p = dbuf.dptr + sizeof(*data) +
                                (lck->num_share_modes *
-                               sizeof(struct share_mode_entry)));
+                               sizeof(struct share_mode_entry));
 
-               if ((data->u.s.delete_token_size < 8) || (data->u.s.delete_token_size % 4) != 0) {
+               if ((data->u.s.delete_token_size < sizeof(uid_t) + sizeof(gid_t)) ||
+                               ((data->u.s.delete_token_size - sizeof(uid_t)) % sizeof(gid_t)) != 0) {
                        DEBUG(0, ("parse_share_modes: invalid token size %d\n",
                                data->u.s.delete_token_size));
                        smb_panic("parse_share_modes: invalid token size\n");
@@ -493,14 +554,16 @@ static BOOL parse_share_modes(TDB_DATA dbuf, struct share_mode_lock *lck)
                }
 
                /* Copy out the uid and gid. */
-               memcpy(&val, p++, 4);
-               lck->delete_token->uid = (uid_t)val;
-               memcpy(&val, p++, 4);
-               lck->delete_token->gid = (gid_t)val;
+               memcpy(&lck->delete_token->uid, p, sizeof(uid_t));
+               p += sizeof(uid_t);
+               memcpy(&lck->delete_token->gid, p, sizeof(gid_t));
+               p += sizeof(gid_t);
 
                /* Any supplementary groups ? */
-               lck->delete_token->ngroups = (data->u.s.delete_token_size > 8) ?
-                                               ((data->u.s.delete_token_size - 8)/4) : 0;
+               lck->delete_token->ngroups = (data->u.s.delete_token_size > (sizeof(uid_t) + sizeof(gid_t))) ?
+                                       ((data->u.s.delete_token_size -
+                                               (sizeof(uid_t) + sizeof(gid_t)))/sizeof(gid_t)) : 0;
+
                if (lck->delete_token->ngroups) {
                        /* Make this a talloc child of lck->delete_token. */
                        lck->delete_token->groups = TALLOC_ARRAY(lck->delete_token, gid_t,
@@ -510,8 +573,8 @@ static BOOL parse_share_modes(TDB_DATA dbuf, struct share_mode_lock *lck)
                        }
 
                        for (i = 0; i < lck->delete_token->ngroups; i++) {
-                               memcpy(&val, p++, 4);
-                               lck->delete_token->groups[i] = (gid_t)val;
+                               memcpy(&lck->delete_token->groups[i], p, sizeof(gid_t));
+                               p += sizeof(gid_t);
                        }
                }
 
@@ -520,16 +583,22 @@ static BOOL parse_share_modes(TDB_DATA dbuf, struct share_mode_lock *lck)
        }
 
        /* Save off the associated service path and filename. */
-       lck->servicepath = talloc_strdup(lck, dbuf.dptr + sizeof(*data) +
+       lck->servicepath = talloc_strdup(lck, (const char *)dbuf.dptr + sizeof(*data) +
                                        (lck->num_share_modes *
                                        sizeof(struct share_mode_entry)) +
                                        data->u.s.delete_token_size );
+       if (lck->servicepath == NULL) {
+               smb_panic("talloc_strdup failed\n");
+       }
 
-       lck->filename = talloc_strdup(lck, dbuf.dptr + sizeof(*data) +
+       lck->filename = talloc_strdup(lck, (const char *)dbuf.dptr + sizeof(*data) +
                                        (lck->num_share_modes *
                                        sizeof(struct share_mode_entry)) +
                                        data->u.s.delete_token_size +
                                        strlen(lck->servicepath) + 1 );
+       if (lck->filename == NULL) {
+               smb_panic("talloc_strdup failed\n");
+       }
 
        /*
         * Ensure that each entry has a real process attached.
@@ -575,14 +644,14 @@ static TDB_DATA unparse_share_modes(struct share_mode_lock *lck)
 
        sp_len = strlen(lck->servicepath);
        delete_token_size = (lck->delete_token ?
-                       (8 + (lck->delete_token->ngroups*4)) : 0);
+                       (sizeof(uid_t) + sizeof(gid_t) + (lck->delete_token->ngroups*sizeof(gid_t))) : 0);
 
        result.dsize = sizeof(*data) +
                lck->num_share_modes * sizeof(struct share_mode_entry) +
                delete_token_size +
                sp_len + 1 +
                strlen(lck->filename) + 1;
-       result.dptr = talloc_size(lck, result.dsize);
+       result.dptr = TALLOC_ARRAY(lck, uint8, result.dsize);
 
        if (result.dptr == NULL) {
                smb_panic("talloc failed\n");
@@ -592,11 +661,9 @@ static TDB_DATA unparse_share_modes(struct share_mode_lock *lck)
        ZERO_STRUCTP(data);
        data->u.s.num_share_mode_entries = lck->num_share_modes;
        data->u.s.delete_on_close = lck->delete_on_close;
-       data->u.s.initial_delete_on_close = lck->initial_delete_on_close;
        data->u.s.delete_token_size = delete_token_size;
-       DEBUG(10, ("unparse_share_modes: del: %d, initial del %d, tok = %u, num: %d\n",
+       DEBUG(10, ("unparse_share_modes: del: %d, tok = %u, num: %d\n",
                data->u.s.delete_on_close,
-               data->u.s.initial_delete_on_close,
                (unsigned int)data->u.s.delete_token_size,
                data->u.s.num_share_mode_entries));
        memcpy(result.dptr + sizeof(*data), lck->share_modes,
@@ -606,26 +673,25 @@ static TDB_DATA unparse_share_modes(struct share_mode_lock *lck)
 
        /* Store any delete on close token. */
        if (lck->delete_token) {
-               uint32 val;
-               uint32 *p = (uint32 *)(result.dptr + offset);
+               uint8 *p = result.dptr + offset;
 
-               val = (uint32)lck->delete_token->uid;
-               memcpy(p++, &val, 4);
+               memcpy(p, &lck->delete_token->uid, sizeof(uid_t));
+               p += sizeof(uid_t);
 
-               val = (uint32)lck->delete_token->uid;
-               memcpy(p++, &val, 4);
+               memcpy(p, &lck->delete_token->gid, sizeof(gid_t));
+               p += sizeof(gid_t);
 
                for (i = 0; i < lck->delete_token->ngroups; i++) {
-                       val = (uint32)lck->delete_token->groups[i];
-                       memcpy(p++, &val, 4);
+                       memcpy(p, &lck->delete_token->groups[i], sizeof(gid_t));
+                       p += sizeof(gid_t);
                }
-               offset = ((char *)p - result.dptr);
+               offset = p - result.dptr;
        }
 
-       safe_strcpy(result.dptr + offset, lck->servicepath,
+       safe_strcpy((char *)result.dptr + offset, lck->servicepath,
                    result.dsize - offset - 1);
        offset += sp_len + 1;
-       safe_strcpy(result.dptr + offset, lck->filename,
+       safe_strcpy((char *)result.dptr + offset, lck->filename,
                    result.dsize - offset - 1);
 
        if (DEBUGLEVEL >= 10) {
@@ -635,10 +701,8 @@ static TDB_DATA unparse_share_modes(struct share_mode_lock *lck)
        return result;
 }
 
-static int share_mode_lock_destructor(void *p)
+static int share_mode_lock_destructor(struct share_mode_lock *lck)
 {
-       struct share_mode_lock *lck =
-               talloc_get_type_abort(p, struct share_mode_lock);
        TDB_DATA key = locking_key(lck->dev, lck->ino);
        TDB_DATA data;
 
@@ -694,13 +758,12 @@ struct share_mode_lock *get_share_mode_lock(TALLOC_CTX *mem_ctx,
        lck->share_modes = NULL;
        lck->delete_token = NULL;
        lck->delete_on_close = False;
-       lck->initial_delete_on_close = False;
        lck->fresh = False;
        lck->modified = False;
 
        if (tdb_chainlock(tdb, key) != 0) {
                DEBUG(3, ("Could not lock share entry\n"));
-               talloc_free(lck);
+               TALLOC_FREE(lck);
                return NULL;
        }
 
@@ -716,20 +779,20 @@ struct share_mode_lock *get_share_mode_lock(TALLOC_CTX *mem_ctx,
        if (lck->fresh) {
 
                if (fname == NULL || servicepath == NULL) {
-                       talloc_free(lck);
+                       TALLOC_FREE(lck);
                        return NULL;
                }
                lck->filename = talloc_strdup(lck, fname);
                lck->servicepath = talloc_strdup(lck, servicepath);
                if (lck->filename == NULL || lck->servicepath == NULL) {
                        DEBUG(0, ("talloc failed\n"));
-                       talloc_free(lck);
+                       TALLOC_FREE(lck);
                        return NULL;
                }
        } else {
                if (!parse_share_modes(data, lck)) {
                        DEBUG(0, ("Could not parse share modes\n"));
-                       talloc_free(lck);
+                       TALLOC_FREE(lck);
                        SAFE_FREE(data.dptr);
                        return NULL;
                }
@@ -747,7 +810,8 @@ struct share_mode_lock *get_share_mode_lock(TALLOC_CTX *mem_ctx,
  Based on an initial code idea from SATOH Fumiyasu <fumiya@samba.gr.jp>
 ********************************************************************/
 
-BOOL rename_share_filename(struct share_mode_lock *lck,
+BOOL rename_share_filename(struct messaging_context *msg_ctx,
+                       struct share_mode_lock *lck,
                        const char *servicepath,
                        const char *newname)
 {
@@ -756,6 +820,7 @@ BOOL rename_share_filename(struct share_mode_lock *lck,
        size_t msg_len;
        char *frm = NULL;
        int i;
+       DATA_BLOB msg;
 
        if (!lck) {
                return False;
@@ -786,7 +851,7 @@ BOOL rename_share_filename(struct share_mode_lock *lck,
        msg_len = MSG_FILE_RENAMED_MIN_SIZE + sp_len + 1 + fn_len + 1;
 
        /* Set up the name changed message. */
-       frm = TALLOC(lck, msg_len);
+       frm = TALLOC_ARRAY(lck, char, msg_len);
        if (!frm) {
                return False;
        }
@@ -794,11 +859,13 @@ BOOL rename_share_filename(struct share_mode_lock *lck,
        SDEV_T_VAL(frm,0,lck->dev);
        SINO_T_VAL(frm,8,lck->ino);
 
-       DEBUG(10,("rename_share_filename: msg_len = %d\n", msg_len ));
+       DEBUG(10,("rename_share_filename: msg_len = %u\n", (unsigned int)msg_len ));
 
        safe_strcpy(&frm[16], lck->servicepath, sp_len);
        safe_strcpy(&frm[16 + sp_len + 1], lck->filename, fn_len);
 
+       msg = data_blob_const(frm, msg_len);
+
        /* Send the messages. */
        for (i=0; i<lck->num_share_modes; i++) {
                struct share_mode_entry *se = &lck->share_modes[i];
@@ -810,30 +877,41 @@ BOOL rename_share_filename(struct share_mode_lock *lck,
                        continue;
                }
 
-               DEBUG(10,("rename_share_filename: sending rename message to pid %u "
+               DEBUG(10,("rename_share_filename: sending rename message to pid %s "
                        "dev %x, inode  %.0f sharepath %s newname %s\n",
-                       (unsigned int)procid_to_pid(&se->pid),
+                       procid_str_static(&se->pid),
                        (unsigned int)lck->dev, (double)lck->ino,
                        lck->servicepath, lck->filename ));
 
-               become_root();
-               message_send_pid(se->pid, MSG_SMB_FILE_RENAME,
-                               frm, msg_len, True);
-               unbecome_root();
+               messaging_send(msg_ctx, se->pid, MSG_SMB_FILE_RENAME, &msg);
        }
 
        return True;
 }
 
-BOOL get_delete_on_close_flag(SMB_DEV_T dev, SMB_INO_T inode)
+static int pull_delete_on_close_flag(TDB_DATA key, TDB_DATA dbuf,
+                                    void *private_data)
 {
-       BOOL result;
-       struct share_mode_lock *lck = get_share_mode_lock(NULL, dev, inode, NULL, NULL);
-       if (!lck) {
-               return False;
+       BOOL *result = (BOOL *)private_data;
+       struct locking_data *data;
+
+       if (dbuf.dsize < sizeof(struct locking_data)) {
+               smb_panic("PANIC: parse_share_modes: buffer too short.\n");
        }
-       result = lck->delete_on_close;
-       talloc_free(lck);
+
+       data = (struct locking_data *)dbuf.dptr;
+
+       *result = data->u.s.delete_on_close;
+       return 0;
+}
+
+BOOL get_delete_on_close_flag(SMB_DEV_T dev, SMB_INO_T inode)
+{
+       TDB_DATA key = locking_key(dev, inode);
+       BOOL result = False;
+
+       tdb_parse_record(tdb, key, pull_delete_on_close_flag,
+                        (void *)&result);
        return result;
 }
 
@@ -865,7 +943,7 @@ BOOL is_unused_share_mode_entry(const struct share_mode_entry *e)
 
 static void fill_share_mode_entry(struct share_mode_entry *e,
                                  files_struct *fsp,
-                                 uint16 mid, uint16 op_type)
+                                 uid_t uid, uint16 mid, uint16 op_type)
 {
        ZERO_STRUCTP(e);
        e->pid = procid_self();
@@ -876,9 +954,11 @@ static void fill_share_mode_entry(struct share_mode_entry *e,
        e->op_type = op_type;
        e->time.tv_sec = fsp->open_time.tv_sec;
        e->time.tv_usec = fsp->open_time.tv_usec;
-       e->share_file_id = fsp->file_id;
        e->dev = fsp->dev;
        e->inode = fsp->inode;
+       e->share_file_id = fsp->fh->file_id;
+       e->uid = (uint32)uid;
+       e->flags = fsp->posix_open ? SHARE_MODE_FLAG_POSIX_OPEN : 0;
 }
 
 static void fill_deferred_open_entry(struct share_mode_entry *e,
@@ -893,6 +973,8 @@ static void fill_deferred_open_entry(struct share_mode_entry *e,
        e->time.tv_usec = request_time.tv_usec;
        e->dev = dev;
        e->inode = ino;
+       e->uid = (uint32)-1;
+       e->flags = 0;
 }
 
 static void add_share_mode_entry(struct share_mode_lock *lck,
@@ -917,10 +999,10 @@ static void add_share_mode_entry(struct share_mode_lock *lck,
 }
 
 void set_share_mode(struct share_mode_lock *lck, files_struct *fsp,
-                   uint16 mid, uint16 op_type)
+                       uid_t uid, uint16 mid, uint16 op_type)
 {
        struct share_mode_entry entry;
-       fill_share_mode_entry(&entry, fsp, mid, op_type);
+       fill_share_mode_entry(&entry, fsp, uid, mid, op_type);
        add_share_mode_entry(lck, &entry);
 }
 
@@ -935,28 +1017,19 @@ void add_deferred_open(struct share_mode_lock *lck, uint16 mid,
 
 /*******************************************************************
  Check if two share mode entries are identical, ignoring oplock 
- and mid info and desired_access.
+ and mid info and desired_access. (Removed paranoia test - it's
+ not automatically a logic error if they are identical. JRA.)
 ********************************************************************/
 
 static BOOL share_modes_identical(struct share_mode_entry *e1,
                                  struct share_mode_entry *e2)
 {
-#if 1 /* JRA PARANOIA TEST - REMOVE LATER */
-       if (procid_equal(&e1->pid, &e2->pid) &&
-           e1->share_file_id == e2->share_file_id &&
-           e1->dev == e2->dev &&
-           e1->inode == e2->inode &&
-           (e1->share_access) != (e2->share_access)) {
-               DEBUG(0,("PANIC: share_modes_identical: share_mode "
-                        "mismatch (e1 = 0x%x, e2 = 0x%x). Logic error.\n",
-                        (unsigned int)e1->share_access,
-                        (unsigned int)e2->share_access ));
-               smb_panic("PANIC: share_modes_identical logic error.\n");
-       }
-#endif
+       /* We used to check for e1->share_access == e2->share_access here
+          as well as the other fields but 2 different DOS or FCB opens
+          sharing the same share mode entry may validly differ in
+          fsp->share_access field. */
 
        return (procid_equal(&e1->pid, &e2->pid) &&
-               (e1->share_access) == (e2->share_access) &&
                e1->dev == e2->dev &&
                e1->inode == e2->inode &&
                e1->share_file_id == e2->share_file_id );
@@ -1001,7 +1074,8 @@ BOOL del_share_mode(struct share_mode_lock *lck, files_struct *fsp)
 {
        struct share_mode_entry entry, *e;
 
-       fill_share_mode_entry(&entry, fsp, 0, 0);
+       /* Don't care about the pid owner being correct here - just a search. */
+       fill_share_mode_entry(&entry, fsp, (uid_t)-1, 0, NO_OPLOCK);
 
        e = find_share_mode_entry(lck, &entry);
        if (e == NULL) {
@@ -1037,7 +1111,8 @@ BOOL remove_share_oplock(struct share_mode_lock *lck, files_struct *fsp)
 {
        struct share_mode_entry entry, *e;
 
-       fill_share_mode_entry(&entry, fsp, 0, 0);
+       /* Don't care about the pid owner being correct here - just a search. */
+       fill_share_mode_entry(&entry, fsp, (uid_t)-1, 0, NO_OPLOCK);
 
        e = find_share_mode_entry(lck, &entry);
        if (e == NULL) {
@@ -1058,7 +1133,8 @@ BOOL downgrade_share_oplock(struct share_mode_lock *lck, files_struct *fsp)
 {
        struct share_mode_entry entry, *e;
 
-       fill_share_mode_entry(&entry, fsp, 0, 0);
+       /* Don't care about the pid owner being correct here - just a search. */
+       fill_share_mode_entry(&entry, fsp, (uid_t)-1, 0, NO_OPLOCK);
 
        e = find_share_mode_entry(lck, &entry);
        if (e == NULL) {
@@ -1073,7 +1149,7 @@ BOOL downgrade_share_oplock(struct share_mode_lock *lck, files_struct *fsp)
 /****************************************************************************
  Deal with the internal needs of setting the delete on close flag. Note that
  as the tdb locking is recursive, it is safe to call this from within 
- open_file_shared. JRA.
+ open_file_ntcreate. JRA.
 ****************************************************************************/
 
 NTSTATUS can_set_delete_on_close(files_struct *fsp, BOOL delete_on_close,
@@ -1118,6 +1194,11 @@ NTSTATUS can_set_delete_on_close(files_struct *fsp, BOOL delete_on_close,
                return NT_STATUS_ACCESS_DENIED;
        }
 
+       /* Don't allow delete on close for non-empty directories. */
+       if (fsp->is_directory) {
+               return can_delete_directory(fsp->conn, fsp->fsp_name);
+       }
+
        return NT_STATUS_OK;
 }
 
@@ -1161,7 +1242,7 @@ void set_delete_on_close_token(struct share_mode_lock *lck, UNIX_USER_TOKEN *tok
 {
        /* Ensure there's no token. */
        if (lck->delete_token) {
-               talloc_free(lck->delete_token); /* Also deletes groups... */
+               TALLOC_FREE(lck->delete_token); /* Also deletes groups... */
                lck->delete_token = NULL;
        }
 
@@ -1177,11 +1258,22 @@ void set_delete_on_close_token(struct share_mode_lock *lck, UNIX_USER_TOKEN *tok
  changed the delete on close flag. This will be noticed
  in the close code, the last closer will delete the file
  if flag is set.
- Note that setting this to any value clears the initial_delete_on_close flag.
- If delete_on_close is True this makes a copy of any UNIX_USER_TOKEN into the
- lck entry.
+ This makes a copy of any UNIX_USER_TOKEN into the
+ lck entry. This function is used when the lock is already granted.
 ****************************************************************************/
 
+void set_delete_on_close_lck(struct share_mode_lock *lck, BOOL delete_on_close, UNIX_USER_TOKEN *tok)
+{
+       if (lck->delete_on_close != delete_on_close) {
+               set_delete_on_close_token(lck, tok);
+               lck->delete_on_close = delete_on_close;
+               if (delete_on_close) {
+                       SMB_ASSERT(lck->delete_token != NULL);
+               }
+               lck->modified = True;
+       }
+}
+
 BOOL set_delete_on_close(files_struct *fsp, BOOL delete_on_close, UNIX_USER_TOKEN *tok)
 {
        struct share_mode_lock *lck;
@@ -1200,33 +1292,33 @@ BOOL set_delete_on_close(files_struct *fsp, BOOL delete_on_close, UNIX_USER_TOKE
                return False;
        }
 
-       if (lck->delete_on_close != delete_on_close) {
-               set_delete_on_close_token(lck, tok);
-               lck->delete_on_close = delete_on_close;
-               if (delete_on_close) {
-                       SMB_ASSERT(lck->delete_token != NULL);
-               }
-               lck->modified = True;
-       }
+       set_delete_on_close_lck(lck, delete_on_close, tok);
 
-       if (lck->initial_delete_on_close) {
-               lck->initial_delete_on_close = False;
-               lck->modified = True;
+       if (fsp->is_directory) {
+               send_stat_cache_delete_message(fsp->fsp_name);
        }
 
-       talloc_free(lck);
+       TALLOC_FREE(lck);
        return True;
 }
 
+struct forall_state {
+       void (*fn)(const struct share_mode_entry *entry,
+                  const char *sharepath,
+                  const char *fname,
+                  void *private_data);
+       void *private_data;
+};
+
 static int traverse_fn(TDB_CONTEXT *the_tdb, TDB_DATA kbuf, TDB_DATA dbuf, 
-                       void *state)
+                       void *_state)
 {
+       struct forall_state *state = (struct forall_state *)_state;
        struct locking_data *data;
        struct share_mode_entry *shares;
        const char *sharepath;
        const char *fname;
        int i;
-       void (*traverse_callback)(struct share_mode_entry *, const char *, const char *) = state;
 
        /* Ensure this is a locking_key record. */
        if (kbuf.dsize != sizeof(struct locking_key))
@@ -1234,16 +1326,17 @@ static int traverse_fn(TDB_CONTEXT *the_tdb, TDB_DATA kbuf, TDB_DATA dbuf,
 
        data = (struct locking_data *)dbuf.dptr;
        shares = (struct share_mode_entry *)(dbuf.dptr + sizeof(*data));
-       sharepath = dbuf.dptr + sizeof(*data) +
+       sharepath = (const char *)dbuf.dptr + sizeof(*data) +
                data->u.s.num_share_mode_entries*sizeof(*shares) +
                data->u.s.delete_token_size;
-       fname = dbuf.dptr + sizeof(*data) +
+       fname = (const char *)dbuf.dptr + sizeof(*data) +
                data->u.s.num_share_mode_entries*sizeof(*shares) +
                data->u.s.delete_token_size +
                strlen(sharepath) + 1;
 
        for (i=0;i<data->u.s.num_share_mode_entries;i++) {
-               traverse_callback(&shares[i], sharepath, fname);
+               state->fn(&shares[i], sharepath, fname,
+                         state->private_data);
        }
        return 0;
 }
@@ -1253,9 +1346,17 @@ static int traverse_fn(TDB_CONTEXT *the_tdb, TDB_DATA kbuf, TDB_DATA dbuf,
  share mode system.
 ********************************************************************/
 
-int share_mode_forall(void (*fn)(const struct share_mode_entry *, const char *, const char *))
+int share_mode_forall(void (*fn)(const struct share_mode_entry *, const char *,
+                                const char *, void *),
+                     void *private_data)
 {
+       struct forall_state state;
+
        if (tdb == NULL)
                return 0;
-       return tdb_traverse(tdb, traverse_fn, fn);
+
+       state.fn = fn;
+       state.private_data = private_data;
+
+       return tdb_traverse(tdb, traverse_fn, (void *)&state);
 }