r14428: Call fill_share_mode_entry with NO_OPLOCK instead of 0.
[samba.git] / source3 / locking / locking.c
index d42d041b79049c5ef42881e55490a1947269230c..5cb1f71555d6ac398dfcce4804476907e78c43be 100644 (file)
@@ -3,6 +3,7 @@
    Locking functions
    Copyright (C) Andrew Tridgell 1992-2000
    Copyright (C) Jeremy Allison 1992-2000
+   Copyright (C) Volker Lendecke 2005
    
    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
@@ -37,6 +38,9 @@
 #include "includes.h"
 uint16 global_smbpid;
 
+#undef DBGC_CLASS
+#define DBGC_CLASS DBGC_LOCKING
+
 /* the locking database handle */
 static TDB_CONTEXT *tdb;
 
@@ -51,27 +55,40 @@ static const char *lock_type_name(enum brl_type lock_type)
 
 /****************************************************************************
  Utility function called to see if a file region is locked.
- If check_self is True, then checks on our own fd with the same locking context
- are still made. If check_self is False, then checks are not made on our own fd
- with the same locking context are not made.
 ****************************************************************************/
 
 BOOL is_locked(files_struct *fsp,connection_struct *conn,
               SMB_BIG_UINT count,SMB_BIG_UINT offset, 
-              enum brl_type lock_type, BOOL check_self)
+              enum brl_type lock_type)
 {
        int snum = SNUM(conn);
+       int strict_locking = lp_strict_locking(snum);
        BOOL ret;
        
        if (count == 0)
                return(False);
 
-       if (!lp_locking(snum) || !lp_strict_locking(snum))
+       if (!lp_locking(snum) || !strict_locking)
                return(False);
 
-       ret = !brl_locktest(fsp->dev, fsp->inode, fsp->fnum,
-                            global_smbpid, sys_getpid(), conn->cnum, 
-                            offset, count, lock_type, check_self);
+       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;
+               } 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;
+               } else {
+                       ret = !brl_locktest(fsp->dev, fsp->inode, fsp->fnum,
+                                    global_smbpid, procid_self(), conn->cnum, 
+                                    offset, count, lock_type);
+               }
+       } else {
+               ret = !brl_locktest(fsp->dev, fsp->inode, fsp->fnum,
+                               global_smbpid, procid_self(), conn->cnum,
+                               offset, count, lock_type);
+       }
 
        DEBUG(10,("is_locked: brl start=%.0f len=%.0f %s for file %s\n",
                        (double)offset, (double)count, ret ? "locked" : "unlocked",
@@ -98,24 +115,23 @@ BOOL is_locked(files_struct *fsp,connection_struct *conn,
 ****************************************************************************/
 
 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)
+                SMB_BIG_UINT count,SMB_BIG_UINT offset,enum brl_type lock_type, BOOL *my_lock_ctx)
 {
-       NTSTATUS status;
+       NTSTATUS status = NT_STATUS_LOCK_NOT_GRANTED;
 
        if (!lp_locking(SNUM(conn)))
                return NT_STATUS_OK;
 
        /* NOTE! 0 byte long ranges ARE allowed and should be stored  */
 
-
        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, sys_getpid(), conn->cnum, 
+                                 lock_pid, procid_self(), conn->cnum, 
                                  offset, count, 
-                                 lock_type);
+                                 lock_type, my_lock_ctx);
 
                if (NT_STATUS_IS_OK(status) && lp_posix_locking(SNUM(conn))) {
 
@@ -126,14 +142,19 @@ static NTSTATUS do_lock(files_struct *fsp,connection_struct *conn, uint16 lock_p
                         */
 
                        if (!set_posix_lock(fsp, offset, count, lock_type)) {
-                               status = NT_STATUS_LOCK_NOT_GRANTED;
+                               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, sys_getpid(), conn->cnum, 
-                                                               offset, count);
+                                                               lock_pid, procid_self(), conn->cnum, 
+                                                               offset, count, False,
+                                                               NULL, NULL);
                        }
                }
        }
@@ -142,14 +163,14 @@ static NTSTATUS do_lock(files_struct *fsp,connection_struct *conn, uint16 lock_p
 }
 
 /****************************************************************************
- Utility function called by locking requests. This is *DISGISTING*. It also
+ 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.
 ****************************************************************************/
 
 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)
+                SMB_BIG_UINT count,SMB_BIG_UINT offset,enum brl_type lock_type, BOOL *my_lock_ctx)
 {
        int j, maxj = lp_lock_spin_count();
        int sleeptime = lp_lock_sleep_time();
@@ -161,7 +182,7 @@ NTSTATUS do_lock_spin(files_struct *fsp,connection_struct *conn, uint16 lock_pid
        ret = NT_STATUS_OK; /* to keep dumb compilers happy */
 
        for (j = 0; j < maxj; j++) {
-               status = do_lock(fsp, conn, lock_pid, count, offset, lock_type);
+               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;
@@ -169,6 +190,9 @@ NTSTATUS do_lock_spin(files_struct *fsp,connection_struct *conn, uint16 lock_pid
                /* 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);
@@ -176,6 +200,25 @@ NTSTATUS do_lock_spin(files_struct *fsp,connection_struct *conn, uint16 lock_pid
        return ret;
 }
 
+/* Struct passed to brl_unlock. */
+struct posix_unlock_data_struct {
+       files_struct *fsp;
+       SMB_BIG_UINT offset;
+       SMB_BIG_UINT count;
+};
+
+/****************************************************************************
+ Function passed to brl_unlock to allow POSIX unlock to be done first.
+****************************************************************************/
+
+static void posix_unlock(void *pre_data)
+{
+       struct posix_unlock_data_struct *pdata = (struct posix_unlock_data_struct *)pre_data;
+
+       if (lp_posix_locking(SNUM(pdata->fsp->conn)))
+               release_posix_lock(pdata->fsp, pdata->offset, pdata->count);
+}
+
 /****************************************************************************
  Utility function called by unlocking requests.
 ****************************************************************************/
@@ -184,6 +227,7 @@ NTSTATUS do_unlock(files_struct *fsp,connection_struct *conn, uint16 lock_pid,
                   SMB_BIG_UINT count,SMB_BIG_UINT offset)
 {
        BOOL ok = False;
+       struct posix_unlock_data_struct posix_data;
        
        if (!lp_locking(SNUM(conn)))
                return NT_STATUS_OK;
@@ -201,19 +245,18 @@ NTSTATUS do_unlock(files_struct *fsp,connection_struct *conn, uint16 lock_pid,
         * match then don't bother looking to remove POSIX locks.
         */
 
+       posix_data.fsp = fsp;
+       posix_data.offset = offset;
+       posix_data.count = count;
+
        ok = brl_unlock(fsp->dev, fsp->inode, fsp->fnum,
-                       lock_pid, sys_getpid(), conn->cnum, offset, count);
+                       lock_pid, procid_self(), conn->cnum, offset, count,
+                       False, posix_unlock, (void *)&posix_data);
    
        if (!ok) {
                DEBUG(10,("do_unlock: returning ERRlock.\n" ));
                return NT_STATUS_RANGE_NOT_LOCKED;
        }
-
-       if (!lp_posix_locking(SNUM(conn)))
-               return NT_STATUS_OK;
-
-       (void)release_posix_lock(fsp, offset, count);
-
        return NT_STATUS_OK;
 }
 
@@ -223,7 +266,7 @@ NTSTATUS do_unlock(files_struct *fsp,connection_struct *conn, uint16 lock_pid,
 
 void locking_close_file(files_struct *fsp)
 {
-       pid_t pid = sys_getpid();
+       struct process_id pid = procid_self();
 
        if (!lp_locking(SNUM(fsp->conn)))
                return;
@@ -258,15 +301,15 @@ BOOL locking_init(int read_only)
                return True;
 
        tdb = tdb_open_log(lock_path("locking.tdb"), 
-                      0, TDB_DEFAULT|(read_only?0x0:TDB_CLEAR_IF_FIRST), 
-                      read_only?O_RDONLY:O_RDWR|O_CREAT,
-                      0644);
+                       SMB_OPEN_DATABASE_TDB_HASH_SIZE, TDB_DEFAULT|(read_only?0x0:TDB_CLEAR_IF_FIRST), 
+                       read_only?O_RDONLY:O_RDWR|O_CREAT,
+                       0644);
 
        if (!tdb) {
                DEBUG(0,("ERROR: Failed to initialise locking database\n"));
                return False;
        }
-       
+
        if (!posix_locking_init(read_only))
                return False;
 
@@ -281,21 +324,31 @@ BOOL locking_init(int read_only)
 
 BOOL locking_end(void)
 {
+       BOOL ret = True;
 
        brl_shutdown(open_read_only);
        if (tdb) {
-
                if (tdb_close(tdb) != 0)
-                       return False;
+                       ret = False;
        }
 
-       return True;
+       return ret;
 }
 
 /*******************************************************************
  Form a static locking key for a dev/inode pair.
 ******************************************************************/
 
+/* key and data records in the tdb locking database */
+struct locking_key {
+       SMB_DEV_T dev;
+       SMB_INO_T ino;
+};
+
+/*******************************************************************
+ Form a static locking key for a dev/inode pair.
+******************************************************************/
+
 static TDB_DATA locking_key(SMB_DEV_T dev, SMB_INO_T inode)
 {
        static struct locking_key key;
@@ -303,67 +356,30 @@ static TDB_DATA locking_key(SMB_DEV_T dev, SMB_INO_T inode)
 
        memset(&key, '\0', sizeof(key));
        key.dev = dev;
-       key.inode = inode;
+       key.ino = inode;
        kbuf.dptr = (char *)&key;
        kbuf.dsize = sizeof(key);
        return kbuf;
 }
 
-static TDB_DATA locking_key_fsp(files_struct *fsp)
-{
-       return locking_key(fsp->dev, fsp->inode);
-}
-
-/*******************************************************************
- Lock a hash bucket entry.
-******************************************************************/
-
-BOOL lock_share_entry(connection_struct *conn,
-                     SMB_DEV_T dev, SMB_INO_T inode)
-{
-       return tdb_chainlock(tdb, locking_key(dev, inode)) == 0;
-}
-
-/*******************************************************************
- Unlock a hash bucket entry.
-******************************************************************/
-
-void unlock_share_entry(connection_struct *conn,
-                       SMB_DEV_T dev, SMB_INO_T inode)
-{
-       tdb_chainunlock(tdb, locking_key(dev, inode));
-}
-
-/*******************************************************************
- Lock a hash bucket entry. use a fsp for convenience
-******************************************************************/
-
-BOOL lock_share_entry_fsp(files_struct *fsp)
-{
-       return tdb_chainlock(tdb, locking_key(fsp->dev, fsp->inode)) == 0;
-}
-
-/*******************************************************************
- Unlock a hash bucket entry.
-******************************************************************/
-
-void unlock_share_entry_fsp(files_struct *fsp)
-{
-       tdb_chainunlock(tdb, locking_key(fsp->dev, fsp->inode));
-}
-
 /*******************************************************************
  Print out a share mode.
 ********************************************************************/
 
-static char *share_mode_str(int num, share_mode_entry *e)
+char *share_mode_str(int num, struct share_mode_entry *e)
 {
        static pstring share_str;
 
-       slprintf(share_str, sizeof(share_str)-1, "share_mode_entry[%d]: \
-pid = %u, share_mode = 0x%x, desired_access = 0x%x, port = 0x%x, type= 0x%x, file_id = %lu, dev = 0x%x, inode = %.0f",
-       num, e->pid, e->share_mode, (unsigned int)e->desired_access, e->op_port, e->op_type, e->share_file_id,
-       (unsigned int)e->dev, (double)e->inode );
+       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",
+                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->dev, (double)e->inode );
 
        return share_str;
 }
@@ -374,13 +390,17 @@ pid = %u, share_mode = 0x%x, desired_access = 0x%x, port = 0x%x, type= 0x%x, fil
 
 static void print_share_mode_table(struct locking_data *data)
 {
-       int num_share_modes = data->u.num_share_mode_entries;
-       share_mode_entry *shares = (share_mode_entry *)(data + 1);
+       int num_share_modes = data->u.s.num_share_mode_entries;
+       struct share_mode_entry *shares =
+               (struct share_mode_entry *)(data + 1);
        int i;
 
        for (i = 0; i < num_share_modes; i++) {
-               share_mode_entry *entry_p = &shares[i];
-               DEBUG(10,("print_share_mode_table: %s\n", share_mode_str(i, entry_p) ));
+               struct share_mode_entry entry;
+
+               memcpy(&entry, &shares[i], sizeof(struct share_mode_entry));
+               DEBUG(10,("print_share_mode_table: %s\n",
+                         share_mode_str(i, &entry)));
        }
 }
 
@@ -388,448 +408,823 @@ static void print_share_mode_table(struct locking_data *data)
  Get all share mode entries for a dev/inode pair.
 ********************************************************************/
 
-int get_share_modes(connection_struct *conn, 
-                   SMB_DEV_T dev, SMB_INO_T inode, 
-                   share_mode_entry **pp_shares)
+static BOOL parse_share_modes(TDB_DATA dbuf, struct share_mode_lock *lck)
 {
-       TDB_DATA dbuf;
        struct locking_data *data;
-       int num_share_modes;
-       share_mode_entry *shares = NULL;
-
-       *pp_shares = NULL;
+       int i;
 
-       dbuf = tdb_fetch(tdb, locking_key(dev, inode));
-       if (!dbuf.dptr)
-               return 0;
+       if (dbuf.dsize < sizeof(struct locking_data)) {
+               smb_panic("PANIC: parse_share_modes: buffer too short.\n");
+       }
 
        data = (struct locking_data *)dbuf.dptr;
-       num_share_modes = data->u.num_share_mode_entries;
-       if(num_share_modes) {
-               int i;
-               int del_count = 0;
 
-               shares = (share_mode_entry *)memdup(dbuf.dptr + sizeof(*data),  
-                                               num_share_modes * sizeof(share_mode_entry));
+       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)) {
+               DEBUG(0, ("invalid number of share modes: %d\n",
+                         lck->num_share_modes));
+               smb_panic("PANIC: invalid number of share modes");
+       }
+
+       lck->share_modes = NULL;
+       
+       if (lck->num_share_modes != 0) {
 
-               if (!shares) {
-                       SAFE_FREE(dbuf.dptr);
-                       return 0;
+               if (dbuf.dsize < (sizeof(struct locking_data) +
+                                 (lck->num_share_modes *
+                                  sizeof(struct share_mode_entry)))) {
+                       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));
 
-               /*
-                * Ensure that each entry has a real process attached.
-                */
+               if (lck->share_modes == NULL) {
+                       smb_panic("talloc failed\n");
+               }
+       }
 
-               for (i = 0; i < num_share_modes; ) {
-                       share_mode_entry *entry_p = &shares[i];
-                       if (process_exists(entry_p->pid)) {
-                               DEBUG(10,("get_share_modes: %s\n", share_mode_str(i, entry_p) ));
-                               i++;
-                       } else {
-                               DEBUG(10,("get_share_modes: deleted %s\n", share_mode_str(i, entry_p) ));
-                               memcpy( &shares[i], &shares[i+1],
-                                       sizeof(share_mode_entry) * (num_share_modes - i - 1));
-                               num_share_modes--;
-                               del_count++;
-                       }
+       /* Get any delete token. */
+       if (data->u.s.delete_token_size) {
+               char *p = dbuf.dptr + sizeof(*data) +
+                               (lck->num_share_modes *
+                               sizeof(struct share_mode_entry));
+
+               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");
                }
 
-               /* Did we delete any ? If so, re-store in tdb. */
-               if (del_count) {
-                       data->u.num_share_mode_entries = num_share_modes;
-                       
-                       if (num_share_modes)
-                               memcpy(dbuf.dptr + sizeof(*data), shares,
-                                               num_share_modes * sizeof(share_mode_entry));
-
-                       /* The record has shrunk a bit */
-                       dbuf.dsize -= del_count * sizeof(share_mode_entry);
-
-                       if (tdb_store(tdb, locking_key(dev, inode), dbuf, TDB_REPLACE) == -1) {
-                               SAFE_FREE(shares);
-                               SAFE_FREE(dbuf.dptr);
-                               return 0;
+               lck->delete_token = TALLOC_P(lck, UNIX_USER_TOKEN);
+               if (!lck->delete_token) {
+                       smb_panic("talloc failed\n");
+               }
+
+               /* Copy out the uid and gid. */
+               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 > (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,
+                                                       lck->delete_token->ngroups);
+                       if (!lck->delete_token) {
+                               smb_panic("talloc failed\n");
+                       }
+
+                       for (i = 0; i < lck->delete_token->ngroups; i++) {
+                               memcpy(&lck->delete_token->groups[i], p, sizeof(gid_t));
+                               p += sizeof(gid_t);
                        }
                }
+
+       } else {
+               lck->delete_token = NULL;
        }
 
-       SAFE_FREE(dbuf.dptr);
-       *pp_shares = shares;
-       return num_share_modes;
-}
+       /* Save off the associated service path and filename. */
+       lck->servicepath = talloc_strdup(lck, dbuf.dptr + sizeof(*data) +
+                                       (lck->num_share_modes *
+                                       sizeof(struct share_mode_entry)) +
+                                       data->u.s.delete_token_size );
 
-/*******************************************************************
- Fill a share mode entry.
-********************************************************************/
+       lck->filename = talloc_strdup(lck, dbuf.dptr + sizeof(*data) +
+                                       (lck->num_share_modes *
+                                       sizeof(struct share_mode_entry)) +
+                                       data->u.s.delete_token_size +
+                                       strlen(lck->servicepath) + 1 );
 
-static void fill_share_mode(char *p, files_struct *fsp, uint16 port, uint16 op_type)
+       /*
+        * Ensure that each entry has a real process attached.
+        */
+
+       for (i = 0; i < lck->num_share_modes; i++) {
+               struct share_mode_entry *entry_p = &lck->share_modes[i];
+               DEBUG(10,("parse_share_modes: %s\n",
+                         share_mode_str(i, entry_p) ));
+               if (!process_exists(entry_p->pid)) {
+                       DEBUG(10,("parse_share_modes: deleted %s\n",
+                                 share_mode_str(i, entry_p) ));
+                       entry_p->op_type = UNUSED_SHARE_MODE_ENTRY;
+                       lck->modified = True;
+               }
+       }
+
+       return True;
+}
+
+static TDB_DATA unparse_share_modes(struct share_mode_lock *lck)
 {
-       share_mode_entry *e = (share_mode_entry *)p;
-       void *x = &e->time; /* Needed to force alignment. p may not be aligned.... */
+       TDB_DATA result;
+       int num_valid = 0;
+       int i;
+       struct locking_data *data;
+       ssize_t offset;
+       ssize_t sp_len;
+       uint32 delete_token_size;
 
-       memset(e, '\0', sizeof(share_mode_entry));
-       e->pid = sys_getpid();
-       e->share_mode = fsp->share_mode;
-       e->desired_access = fsp->desired_access;
-       e->op_port = port;
-       e->op_type = op_type;
-       memcpy(x, &fsp->open_time, sizeof(struct timeval));
-       e->share_file_id = fsp->file_id;
-       e->dev = fsp->dev;
-       e->inode = fsp->inode;
+       result.dptr = NULL;
+       result.dsize = 0;
+
+       for (i=0; i<lck->num_share_modes; i++) {
+               if (!is_unused_share_mode_entry(&lck->share_modes[i])) {
+                       num_valid += 1;
+               }
+       }
+
+       if (num_valid == 0) {
+               return result;
+       }
+
+       sp_len = strlen(lck->servicepath);
+       delete_token_size = (lck->delete_token ?
+                       (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);
+
+       if (result.dptr == NULL) {
+               smb_panic("talloc failed\n");
+       }
+
+       data = (struct locking_data *)result.dptr;
+       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",
+               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,
+              sizeof(struct share_mode_entry)*lck->num_share_modes);
+       offset = sizeof(*data) +
+               sizeof(struct share_mode_entry)*lck->num_share_modes;
+
+       /* Store any delete on close token. */
+       if (lck->delete_token) {
+               char *p = result.dptr + offset;
+
+               memcpy(p, &lck->delete_token->uid, sizeof(uid_t));
+               p += sizeof(uid_t);
+
+               memcpy(p, &lck->delete_token->gid, sizeof(gid_t));
+
+               for (i = 0; i < lck->delete_token->ngroups; i++) {
+                       memcpy(p, &lck->delete_token->groups[i], sizeof(gid_t));
+                       p += sizeof(gid_t);
+               }
+               offset = p - result.dptr;
+       }
+
+       safe_strcpy(result.dptr + offset, lck->servicepath,
+                   result.dsize - offset - 1);
+       offset += sp_len + 1;
+       safe_strcpy(result.dptr + offset, lck->filename,
+                   result.dsize - offset - 1);
+
+       if (DEBUGLEVEL >= 10) {
+               print_share_mode_table(data);
+       }
+
+       return result;
 }
 
-/*******************************************************************
- Check if two share mode entries are identical, ignoring oplock 
- and port info and desired_access.
-********************************************************************/
+static int share_mode_lock_destructor(void *p)
+{
+       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;
 
-BOOL share_modes_identical( share_mode_entry *e1, share_mode_entry *e2)
+       if (!lck->modified) {
+               goto done;
+       }
+
+       data = unparse_share_modes(lck);
+
+       if (data.dptr == NULL) {
+               if (!lck->fresh) {
+                       /* There has been an entry before, delete it */
+                       if (tdb_delete(tdb, key) == -1) {
+                               smb_panic("Could not delete share entry\n");
+                       }
+               }
+               goto done;
+       }
+
+       if (tdb_store(tdb, key, data, TDB_REPLACE) == -1) {
+               smb_panic("Could not store share mode entry\n");
+       }
+
+ done:
+       tdb_chainunlock(tdb, key);
+
+       return 0;
+}
+
+struct share_mode_lock *get_share_mode_lock(TALLOC_CTX *mem_ctx,
+                                               SMB_DEV_T dev, SMB_INO_T ino,
+                                               const char *servicepath,
+                                               const char *fname)
 {
-#if 1 /* JRA PARANOIA TEST - REMOVE LATER */
-       if (e1->pid == e2->pid &&
-               e1->share_file_id == e2->share_file_id &&
-               e1->dev == e2->dev &&
-               e1->inode == e2->inode &&
-               (e1->share_mode & ~DELETE_ON_CLOSE_FLAG) != (e2->share_mode & ~DELETE_ON_CLOSE_FLAG)) {
-                       DEBUG(0,("PANIC: share_modes_identical: share_mode missmatch (e1 = %u, e2 = %u). Logic error.\n",
-                               (unsigned int)(e1->share_mode & ~DELETE_ON_CLOSE_FLAG),
-                               (unsigned int)(e2->share_mode & ~DELETE_ON_CLOSE_FLAG) ));
-               smb_panic("PANIC: share_modes_identical logic error.\n");
+       struct share_mode_lock *lck;
+       TDB_DATA key = locking_key(dev, ino);
+       TDB_DATA data;
+
+       lck = TALLOC_P(mem_ctx, struct share_mode_lock);
+       if (lck == NULL) {
+               DEBUG(0, ("talloc failed\n"));
+               return NULL;
        }
-#endif
 
-       return (e1->pid == e2->pid &&
-               (e1->share_mode & ~DELETE_ON_CLOSE_FLAG) == (e2->share_mode & ~DELETE_ON_CLOSE_FLAG) &&
-               e1->dev == e2->dev &&
-               e1->inode == e2->inode &&
-               e1->share_file_id == e2->share_file_id );
+       /* Ensure we set every field here as the destructor must be
+          valid even if parse_share_modes fails. */
+
+       lck->servicepath = NULL;
+       lck->filename = NULL;
+       lck->dev = dev;
+       lck->ino = ino;
+       lck->num_share_modes = 0;
+       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);
+               return NULL;
+       }
+
+       /* We must set the destructor immediately after the chainlock
+          ensure the lock is cleaned up on any of the error return
+          paths below. */
+
+       talloc_set_destructor(lck, share_mode_lock_destructor);
+
+       data = tdb_fetch(tdb, key);
+       lck->fresh = (data.dptr == NULL);
+
+       if (lck->fresh) {
+
+               if (fname == NULL || servicepath == NULL) {
+                       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);
+                       return NULL;
+               }
+       } else {
+               if (!parse_share_modes(data, lck)) {
+                       DEBUG(0, ("Could not parse share modes\n"));
+                       TALLOC_FREE(lck);
+                       SAFE_FREE(data.dptr);
+                       return NULL;
+               }
+       }
+
+       SAFE_FREE(data.dptr);
+
+       return lck;
 }
 
 /*******************************************************************
- Delete a specific share mode. Return the number
- of entries left, and a memdup'ed copy of the entry deleted (if required).
- Ignore if no entry deleted.
+ Sets the service name and filename for rename.
+ At this point we emit "file renamed" messages to all
+ process id's that have this file open.
+ Based on an initial code idea from SATOH Fumiyasu <fumiya@samba.gr.jp>
 ********************************************************************/
 
-ssize_t del_share_entry( SMB_DEV_T dev, SMB_INO_T inode,
-                       share_mode_entry *entry, share_mode_entry **ppse)
+BOOL rename_share_filename(struct share_mode_lock *lck,
+                       const char *servicepath,
+                       const char *newname)
 {
-       TDB_DATA dbuf;
-       struct locking_data *data;
-       int i, del_count=0;
-       share_mode_entry *shares;
-       ssize_t count = 0;
-
-       if (ppse)
-               *ppse = NULL;
+       size_t sp_len;
+       size_t fn_len;
+       size_t msg_len;
+       char *frm = NULL;
+       int i;
 
-       /* read in the existing share modes */
-       dbuf = tdb_fetch(tdb, locking_key(dev, inode));
-       if (!dbuf.dptr)
-               return -1;
+       if (!lck) {
+               return False;
+       }
 
-       data = (struct locking_data *)dbuf.dptr;
-       shares = (share_mode_entry *)(dbuf.dptr + sizeof(*data));
+       DEBUG(10, ("rename_share_filename: servicepath %s newname %s\n",
+               servicepath, newname));
 
        /*
-        * Find any with this pid and delete it
-        * by overwriting with the rest of the data 
-        * from the record.
+        * rename_internal_fsp() and rename_internals() add './' to
+        * head of newname if newname does not contain a '/'.
         */
+       while (newname[0] && newname[1] && newname[0] == '.' && newname[1] == '/') {
+               newname += 2;
+       }
 
-       DEBUG(10,("del_share_entry: num_share_modes = %d\n", data->u.num_share_mode_entries ));
+       lck->servicepath = talloc_strdup(lck, servicepath);
+       lck->filename = talloc_strdup(lck, newname);
+       if (lck->filename == NULL || lck->servicepath == NULL) {
+               DEBUG(0, ("rename_share_filename: talloc failed\n"));
+               return False;
+       }
+       lck->modified = True;
 
-       for (i=0;i<data->u.num_share_mode_entries;) {
-               if (share_modes_identical(&shares[i], entry)) {
-                       DEBUG(10,("del_share_entry: deleted %s\n",
-                               share_mode_str(i, &shares[i]) ));
-                       if (ppse)
-                               *ppse = memdup(&shares[i], sizeof(*shares));
-                       data->u.num_share_mode_entries--;
-                       memmove(&shares[i], &shares[i+1], 
-                               dbuf.dsize - (sizeof(*data) + (i+1)*sizeof(*shares)));
-                       del_count++;
+       sp_len = strlen(lck->servicepath);
+       fn_len = strlen(lck->filename);
 
-                       DEBUG(10,("del_share_entry: deleting entry %d\n", i ));
+       msg_len = MSG_FILE_RENAMED_MIN_SIZE + sp_len + 1 + fn_len + 1;
 
-               } else {
-                       i++;
-               }
+       /* Set up the name changed message. */
+       frm = TALLOC(lck, msg_len);
+       if (!frm) {
+               return False;
        }
 
-       if (del_count) {
-               /* the record may have shrunk a bit */
-               dbuf.dsize -= del_count * sizeof(*shares);
+       SDEV_T_VAL(frm,0,lck->dev);
+       SINO_T_VAL(frm,8,lck->ino);
 
-               count = (ssize_t)data->u.num_share_mode_entries;
+       DEBUG(10,("rename_share_filename: msg_len = %d\n", msg_len ));
 
-               /* store it back in the database */
-               if (data->u.num_share_mode_entries == 0) {
-                       if (tdb_delete(tdb, locking_key(dev, inode)) == -1)
-                               count = -1;
-               } else {
-                       if (tdb_store(tdb, locking_key(dev, inode), dbuf, TDB_REPLACE) == -1)
-                               count = -1;
+       safe_strcpy(&frm[16], lck->servicepath, sp_len);
+       safe_strcpy(&frm[16 + sp_len + 1], lck->filename, fn_len);
+
+       /* Send the messages. */
+       for (i=0; i<lck->num_share_modes; i++) {
+               struct share_mode_entry *se = &lck->share_modes[i];
+               if (!is_valid_share_mode_entry(se)) {
+                       continue;
                }
+               /* But not to ourselves... */
+               if (procid_is_me(&se->pid)) {
+                       continue;
+               }
+
+               DEBUG(10,("rename_share_filename: sending rename message to pid %u "
+                       "dev %x, inode  %.0f sharepath %s newname %s\n",
+                       (unsigned int)procid_to_pid(&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();
        }
-       DEBUG(10,("del_share_entry: Remaining table.\n"));
-       print_share_mode_table((struct locking_data *)dbuf.dptr);
-       SAFE_FREE(dbuf.dptr);
-       return count;
+
+       return True;
 }
 
-/*******************************************************************
- Del the share mode of a file for this process. Return the number
- of entries left, and a memdup'ed copy of the entry deleted.
-********************************************************************/
+BOOL get_delete_on_close_flag(SMB_DEV_T dev, SMB_INO_T inode)
+{
+       BOOL result;
+       struct share_mode_lock *lck = get_share_mode_lock(NULL, dev, inode, NULL, NULL);
+       if (!lck) {
+               return False;
+       }
+       result = lck->delete_on_close;
+       TALLOC_FREE(lck);
+       return result;
+}
 
-ssize_t del_share_mode(files_struct *fsp, share_mode_entry **ppse)
+BOOL is_valid_share_mode_entry(const struct share_mode_entry *e)
 {
-       share_mode_entry entry;
+       int num_props = 0;
 
-       /*
-        * Fake up a share_mode_entry for comparisons.
-        */
+       num_props += ((e->op_type == NO_OPLOCK) ? 1 : 0);
+       num_props += (EXCLUSIVE_OPLOCK_TYPE(e->op_type) ? 1 : 0);
+       num_props += (LEVEL_II_OPLOCK_TYPE(e->op_type) ? 1 : 0);
+
+       SMB_ASSERT(num_props <= 1);
+       return (num_props != 0);
+}
+
+BOOL is_deferred_open_entry(const struct share_mode_entry *e)
+{
+       return (e->op_type == DEFERRED_OPEN_ENTRY);
+}
 
-       fill_share_mode((char *)&entry, fsp, 0, 0);
-       return del_share_entry(fsp->dev, fsp->inode, &entry, ppse);
+BOOL is_unused_share_mode_entry(const struct share_mode_entry *e)
+{
+       return (e->op_type == UNUSED_SHARE_MODE_ENTRY);
 }
 
 /*******************************************************************
Set the share mode of a file. Return False on fail, True on success.
Fill a share mode entry.
 ********************************************************************/
 
-BOOL set_share_mode(files_struct *fsp, uint16 port, uint16 op_type)
+static void fill_share_mode_entry(struct share_mode_entry *e,
+                                 files_struct *fsp,
+                                 uint16 mid, uint16 op_type)
 {
-       TDB_DATA dbuf;
-       struct locking_data *data;
-       char *p=NULL;
-       int size;
-       BOOL ret = True;
-               
-       /* read in the existing share modes if any */
-       dbuf = tdb_fetch(tdb, locking_key_fsp(fsp));
-       if (!dbuf.dptr) {
-               /* we'll need to create a new record */
-               pstring fname;
-
-               pstrcpy(fname, fsp->conn->connectpath);
-               pstrcat(fname, "/");
-               pstrcat(fname, fsp->fsp_name);
-
-               size = sizeof(*data) + sizeof(share_mode_entry) + strlen(fname) + 1;
-               p = (char *)malloc(size);
-               if (!p)
-                       return False;
-               data = (struct locking_data *)p;
-               data->u.num_share_mode_entries = 1;
-       
-               DEBUG(10,("set_share_mode: creating entry for file %s. num_share_modes = 1\n",
-                       fsp->fsp_name ));
+       ZERO_STRUCTP(e);
+       e->pid = procid_self();
+       e->share_access = fsp->share_access;
+       e->private_options = fsp->fh->private_options;
+       e->access_mask = fsp->access_mask;
+       e->op_mid = mid;
+       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;
+}
 
-               pstrcpy(p + sizeof(*data) + sizeof(share_mode_entry), fname);
-               fill_share_mode(p + sizeof(*data), fsp, port, op_type);
-               dbuf.dptr = p;
-               dbuf.dsize = size;
-               if (tdb_store(tdb, locking_key_fsp(fsp), dbuf, TDB_REPLACE) == -1)
-                       ret = False;
+static void fill_deferred_open_entry(struct share_mode_entry *e,
+                                    const struct timeval request_time,
+                                    SMB_DEV_T dev, SMB_INO_T ino, uint16 mid)
+{
+       ZERO_STRUCTP(e);
+       e->pid = procid_self();
+       e->op_mid = mid;
+       e->op_type = DEFERRED_OPEN_ENTRY;
+       e->time.tv_sec = request_time.tv_sec;
+       e->time.tv_usec = request_time.tv_usec;
+       e->dev = dev;
+       e->inode = ino;
+}
 
-               print_share_mode_table((struct locking_data *)p);
+static void add_share_mode_entry(struct share_mode_lock *lck,
+                                const struct share_mode_entry *entry)
+{
+       int i;
 
-               SAFE_FREE(p);
-               return ret;
+       for (i=0; i<lck->num_share_modes; i++) {
+               struct share_mode_entry *e = &lck->share_modes[i];
+               if (is_unused_share_mode_entry(e)) {
+                       *e = *entry;
+                       break;
+               }
        }
 
-       /* we're adding to an existing entry - this is a bit fiddly */
-       data = (struct locking_data *)dbuf.dptr;
+       if (i == lck->num_share_modes) {
+               /* No unused entry found */
+               ADD_TO_ARRAY(lck, struct share_mode_entry, *entry,
+                            &lck->share_modes, &lck->num_share_modes);
+       }
+       lck->modified = True;
+}
 
-       data->u.num_share_mode_entries++;
-       
-       DEBUG(10,("set_share_mode: adding entry for file %s. new num_share_modes = %d\n",
-               fsp->fsp_name, data->u.num_share_mode_entries ));
+void set_share_mode(struct share_mode_lock *lck, files_struct *fsp,
+                   uint16 mid, uint16 op_type)
+{
+       struct share_mode_entry entry;
+       fill_share_mode_entry(&entry, fsp, mid, op_type);
+       add_share_mode_entry(lck, &entry);
+}
 
-       size = dbuf.dsize + sizeof(share_mode_entry);
-       p = malloc(size);
-       if (!p)
-               return False;
-       memcpy(p, dbuf.dptr, sizeof(*data));
-       fill_share_mode(p + sizeof(*data), fsp, port, op_type);
-       memcpy(p + sizeof(*data) + sizeof(share_mode_entry), dbuf.dptr + sizeof(*data),
-              dbuf.dsize - sizeof(*data));
-       SAFE_FREE(dbuf.dptr);
-       dbuf.dptr = p;
-       dbuf.dsize = size;
-       if (tdb_store(tdb, locking_key_fsp(fsp), dbuf, TDB_REPLACE) == -1)
-               ret = False;
-       print_share_mode_table((struct locking_data *)p);
-       SAFE_FREE(p);
-       return ret;
+void add_deferred_open(struct share_mode_lock *lck, uint16 mid,
+                      struct timeval request_time,
+                      SMB_DEV_T dev, SMB_INO_T ino)
+{
+       struct share_mode_entry entry;
+       fill_deferred_open_entry(&entry, request_time, dev, ino, mid);
+       add_share_mode_entry(lck, &entry);
 }
 
 /*******************************************************************
- A generic in-place modification call for share mode entries.
+ Check if two share mode entries are identical, ignoring oplock 
+ and mid info and desired_access.
 ********************************************************************/
 
-static BOOL mod_share_mode( SMB_DEV_T dev, SMB_INO_T inode, share_mode_entry *entry,
-                          void (*mod_fn)(share_mode_entry *, SMB_DEV_T, SMB_INO_T, void *),
-                          void *param)
+static BOOL share_modes_identical(struct share_mode_entry *e1,
+                                 struct share_mode_entry *e2)
 {
-       TDB_DATA dbuf;
-       struct locking_data *data;
-       int i;
-       share_mode_entry *shares;
-       BOOL need_store=False;
-       BOOL ret = True;
+#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
 
-       /* read in the existing share modes */
-       dbuf = tdb_fetch(tdb, locking_key(dev, inode));
-       if (!dbuf.dptr)
-               return False;
+       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 );
+}
 
-       data = (struct locking_data *)dbuf.dptr;
-       shares = (share_mode_entry *)(dbuf.dptr + sizeof(*data));
+static BOOL deferred_open_identical(struct share_mode_entry *e1,
+                                   struct share_mode_entry *e2)
+{
+       return (procid_equal(&e1->pid, &e2->pid) &&
+               (e1->op_mid == e2->op_mid) &&
+               (e1->dev == e2->dev) &&
+               (e1->inode == e2->inode));
+}
 
-       /* find any with our pid and call the supplied function */
-       for (i=0;i<data->u.num_share_mode_entries;i++) {
-               if (share_modes_identical(entry, &shares[i])) {
-                       mod_fn(&shares[i], dev, inode, param);
-                       need_store=True;
-               }
-       }
+static struct share_mode_entry *find_share_mode_entry(struct share_mode_lock *lck,
+                                                     struct share_mode_entry *entry)
+{
+       int i;
 
-       /* if the mod fn was called then store it back */
-       if (need_store) {
-               if (data->u.num_share_mode_entries == 0) {
-                       if (tdb_delete(tdb, locking_key(dev, inode)) == -1)
-                               ret = False;
-               } else {
-                       if (tdb_store(tdb, locking_key(dev, inode), dbuf, TDB_REPLACE) == -1)
-                               ret = False;
+       for (i=0; i<lck->num_share_modes; i++) {
+               struct share_mode_entry *e = &lck->share_modes[i];
+               if (is_valid_share_mode_entry(entry) &&
+                   is_valid_share_mode_entry(e) &&
+                   share_modes_identical(e, entry)) {
+                       return e;
+               }
+               if (is_deferred_open_entry(entry) &&
+                   is_deferred_open_entry(e) &&
+                   deferred_open_identical(e, entry)) {
+                       return e;
                }
        }
-
-       SAFE_FREE(dbuf.dptr);
-       return ret;
+       return NULL;
 }
 
 /*******************************************************************
- Static function that actually does the work for the generic function
below.
+ Del the share mode of a file for this process. Return the number of
entries left.
 ********************************************************************/
 
-static void remove_share_oplock_fn(share_mode_entry *entry, SMB_DEV_T dev, SMB_INO_T inode, 
-                                   void *param)
+BOOL del_share_mode(struct share_mode_lock *lck, files_struct *fsp)
 {
-       DEBUG(10,("remove_share_oplock_fn: removing oplock info for entry dev=%x ino=%.0f\n",
-                 (unsigned int)dev, (double)inode ));
-       /* Delete the oplock info. */
-       entry->op_port = 0;
-       entry->op_type = NO_OPLOCK;
-}
+       struct share_mode_entry entry, *e;
 
-/*******************************************************************
- Remove an oplock port and mode entry from a share mode.
-********************************************************************/
+       fill_share_mode_entry(&entry, fsp, 0, NO_OPLOCK);
 
-BOOL remove_share_oplock(files_struct *fsp)
+       e = find_share_mode_entry(lck, &entry);
+       if (e == NULL) {
+               return False;
+       }
+
+       e->op_type = UNUSED_SHARE_MODE_ENTRY;
+       lck->modified = True;
+       return True;
+}
+
+void del_deferred_open_entry(struct share_mode_lock *lck, uint16 mid)
 {
-       share_mode_entry entry;
-       /*
-        * Fake up an entry for comparisons...
-        */
-       fill_share_mode((char *)&entry, fsp, 0, 0);
-       return mod_share_mode(fsp->dev, fsp->inode, &entry, remove_share_oplock_fn, NULL);
+       struct share_mode_entry entry, *e;
+
+       fill_deferred_open_entry(&entry, timeval_zero(),
+                                lck->dev, lck->ino, mid);
+
+       e = find_share_mode_entry(lck, &entry);
+       if (e == NULL) {
+               return;
+       }
+
+       e->op_type = UNUSED_SHARE_MODE_ENTRY;
+       lck->modified = True;
 }
 
 /*******************************************************************
- Static function that actually does the work for the generic function
- below.
+ Remove an oplock mid and mode entry from a share mode.
 ********************************************************************/
 
-static void downgrade_share_oplock_fn(share_mode_entry *entry, SMB_DEV_T dev, SMB_INO_T inode, 
-                                   void *param)
+BOOL remove_share_oplock(struct share_mode_lock *lck, files_struct *fsp)
 {
-       DEBUG(10,("downgrade_share_oplock_fn: downgrading oplock info for entry dev=%x ino=%.0f\n",
-                 (unsigned int)dev, (double)inode ));
-       entry->op_type = LEVEL_II_OPLOCK;
+       struct share_mode_entry entry, *e;
+
+       fill_share_mode_entry(&entry, fsp, 0, NO_OPLOCK);
+
+       e = find_share_mode_entry(lck, &entry);
+       if (e == NULL) {
+               return False;
+       }
+
+       e->op_mid = 0;
+       e->op_type = NO_OPLOCK;
+       lck->modified = True;
+       return True;
 }
 
 /*******************************************************************
  Downgrade a oplock type from exclusive to level II.
 ********************************************************************/
 
-BOOL downgrade_share_oplock(files_struct *fsp)
+BOOL downgrade_share_oplock(struct share_mode_lock *lck, files_struct *fsp)
 {
-       share_mode_entry entry;
+       struct share_mode_entry entry, *e;
+
+       fill_share_mode_entry(&entry, fsp, 0, NO_OPLOCK);
+
+       e = find_share_mode_entry(lck, &entry);
+       if (e == NULL) {
+               return False;
+       }
+
+       e->op_type = LEVEL_II_OPLOCK;
+       lck->modified = True;
+       return True;
+}
+
+/****************************************************************************
+ 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.
+****************************************************************************/
+
+NTSTATUS can_set_delete_on_close(files_struct *fsp, BOOL delete_on_close,
+                                uint32 dosmode)
+{
+       if (!delete_on_close) {
+               return NT_STATUS_OK;
+       }
+
+       /*
+        * Only allow delete on close for writable files.
+        */
+
+       if ((dosmode & aRONLY) &&
+           !lp_delete_readonly(SNUM(fsp->conn))) {
+               DEBUG(10,("can_set_delete_on_close: file %s delete on close "
+                         "flag set but file attribute is readonly.\n",
+                         fsp->fsp_name ));
+               return NT_STATUS_CANNOT_DELETE;
+       }
+
+       /*
+        * Only allow delete on close for writable shares.
+        */
+
+       if (!CAN_WRITE(fsp->conn)) {
+               DEBUG(10,("can_set_delete_on_close: file %s delete on "
+                         "close flag set but write access denied on share.\n",
+                         fsp->fsp_name ));
+               return NT_STATUS_ACCESS_DENIED;
+       }
+
        /*
-        * Fake up an entry for comparisons...
+        * Only allow delete on close for files/directories opened with delete
+        * intent.
         */
-       fill_share_mode((char *)&entry, fsp, 0, 0);
-       return mod_share_mode(fsp->dev, fsp->inode, &entry, downgrade_share_oplock_fn, NULL);
+
+       if (!(fsp->access_mask & DELETE_ACCESS)) {
+               DEBUG(10,("can_set_delete_on_close: file %s delete on "
+                         "close flag set but delete access denied.\n",
+                         fsp->fsp_name ));
+               return NT_STATUS_ACCESS_DENIED;
+       }
+
+       return NT_STATUS_OK;
 }
 
-/*******************************************************************
Get/Set the delete on close flag in a set of share modes.
Return False on fail, True on success.
-********************************************************************/
+/*************************************************************************
Return a talloced copy of a UNIX_USER_TOKEN. NULL on fail.
(Should this be in locking.c.... ?).
+*************************************************************************/
 
-BOOL modify_delete_flag( SMB_DEV_T dev, SMB_INO_T inode, BOOL delete_on_close)
+static UNIX_USER_TOKEN *copy_unix_token(TALLOC_CTX *ctx, UNIX_USER_TOKEN *tok)
 {
-       TDB_DATA dbuf;
-       struct locking_data *data;
-       int i;
-       share_mode_entry *shares;
-
-       /* read in the existing share modes */
-       dbuf = tdb_fetch(tdb, locking_key(dev, inode));
-       if (!dbuf.dptr)
-               return False;
+       UNIX_USER_TOKEN *cpy;
 
-       data = (struct locking_data *)dbuf.dptr;
-       shares = (share_mode_entry *)(dbuf.dptr + sizeof(*data));
+       if (tok == NULL) {
+               return NULL;
+       }
 
-       /* Set/Unset the delete on close element. */
-       for (i=0;i<data->u.num_share_mode_entries;i++,shares++) {
-               shares->share_mode = (delete_on_close ?
-                            (shares->share_mode | DELETE_ON_CLOSE_FLAG) :
-                            (shares->share_mode & ~DELETE_ON_CLOSE_FLAG) );
+       cpy = TALLOC_P(ctx, UNIX_USER_TOKEN);
+       if (!cpy) {
+               return NULL;
        }
 
-       /* store it back */
-       if (data->u.num_share_mode_entries) {
-               if (tdb_store(tdb, locking_key(dev,inode), dbuf, TDB_REPLACE)==-1) {
-                       SAFE_FREE(dbuf.dptr);
-                       return False;
+       cpy->uid = tok->uid;
+       cpy->gid = tok->gid;
+       cpy->ngroups = tok->ngroups;
+       if (tok->ngroups) {
+               /* Make this a talloc child of cpy. */
+               cpy->groups = TALLOC_ARRAY(cpy, gid_t, tok->ngroups);
+               if (!cpy->groups) {
+                       return NULL;
                }
+               memcpy(cpy->groups, tok->groups, tok->ngroups * sizeof(gid_t));
        }
+       return cpy;
+}
 
-       SAFE_FREE(dbuf.dptr);
-       return True;
+/****************************************************************************
+ Replace the delete on close token.
+****************************************************************************/
+
+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... */
+               lck->delete_token = NULL;
+       }
+
+       /* Copy the new token (can be NULL). */
+       lck->delete_token = copy_unix_token(lck, tok);
+       lck->modified = True;
 }
 
 /****************************************************************************
- Traverse the whole database with this function, calling traverse_callback
- on each share mode
+ Sets the delete on close flag over all share modes on this file.
+ Modify the share mode entry for all files open
+ on this device and inode to tell other smbds we have
+ 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.
 ****************************************************************************/
 
+BOOL set_delete_on_close(files_struct *fsp, BOOL delete_on_close, UNIX_USER_TOKEN *tok)
+{
+       struct share_mode_lock *lck;
+       
+       DEBUG(10,("set_delete_on_close: %s delete on close flag for "
+                 "fnum = %d, file %s\n",
+                 delete_on_close ? "Adding" : "Removing", fsp->fnum,
+                 fsp->fsp_name ));
+
+       if (fsp->is_stat) {
+               return True;
+       }
+
+       lck = get_share_mode_lock(NULL, fsp->dev, fsp->inode, NULL, NULL);
+       if (lck == NULL) {
+               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;
+       }
+
+       if (lck->initial_delete_on_close) {
+               lck->initial_delete_on_close = False;
+               lck->modified = True;
+       }
+
+       TALLOC_FREE(lck);
+       return True;
+}
+
 static int traverse_fn(TDB_CONTEXT *the_tdb, TDB_DATA kbuf, TDB_DATA dbuf, 
-                       voidstate)
+                       void *state)
 {
        struct locking_data *data;
-       share_mode_entry *shares;
-       char *name;
+       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;
 
-       SHAREMODE_FN(traverse_callback) = (SHAREMODE_FN_CAST())state;
+       /* Ensure this is a locking_key record. */
+       if (kbuf.dsize != sizeof(struct locking_key))
+               return 0;
 
        data = (struct locking_data *)dbuf.dptr;
-       shares = (share_mode_entry *)(dbuf.dptr + sizeof(*data));
-       name = dbuf.dptr + sizeof(*data) + data->u.num_share_mode_entries*sizeof(*shares);
-
-       for (i=0;i<data->u.num_share_mode_entries;i++) {
-               traverse_callback(&shares[i], name);
+       shares = (struct share_mode_entry *)(dbuf.dptr + sizeof(*data));
+       sharepath = dbuf.dptr + sizeof(*data) +
+               data->u.s.num_share_mode_entries*sizeof(*shares) +
+               data->u.s.delete_token_size;
+       fname = 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);
        }
        return 0;
 }
@@ -839,9 +1234,9 @@ static int traverse_fn(TDB_CONTEXT *the_tdb, TDB_DATA kbuf, TDB_DATA dbuf,
  share mode system.
 ********************************************************************/
 
-int share_mode_forall(SHAREMODE_FN(fn))
+int share_mode_forall(void (*fn)(const struct share_mode_entry *, const char *, const char *))
 {
-       if (!tdb)
+       if (tdb == NULL)
                return 0;
-       return tdb_traverse(tdb, traverse_fn, (void*)fn);
+       return tdb_traverse(tdb, traverse_fn, fn);
 }