Herb's warning fixes. Also the POSIX locking fix.
[samba.git] / source3 / locking / posix.c
index 45d139b7342f9839518457812715a180b6ce35c8..fbdcf662fc9e2f8bcb2de2436df01a3e79714e06 100644 (file)
@@ -73,6 +73,8 @@ static TDB_DATA locking_key(SMB_DEV_T dev, SMB_INO_T inode)
 {
        static struct posix_lock_key key;
        TDB_DATA kbuf;
+
+       memset(&key, '\0', sizeof(key));
        key.device = dev;
        key.inode = inode;
        kbuf.dptr = (char *)&key;
@@ -202,7 +204,7 @@ int fd_close_posix(struct connection_struct *conn, files_struct *fsp)
                /*
                 * No POSIX to worry about, just close.
                 */
-               ret = conn->vfs_ops.close(fsp->fd);
+               ret = conn->vfs_ops.close(fsp,fsp->fd);
                fsp->fd = -1;
                return ret;
        }
@@ -257,7 +259,7 @@ int fd_close_posix(struct connection_struct *conn, files_struct *fsp)
                DEBUG(10,("fd_close_posix: doing close on %u fd's.\n", (unsigned int)count ));
 
                for(i = 0; i < count; i++) {
-                       if (conn->vfs_ops.close(fd_array[i]) == -1) {
+                       if (conn->vfs_ops.close(fsp,fd_array[i]) == -1) {
                                saved_errno = errno;
                        }
                }
@@ -277,7 +279,7 @@ int fd_close_posix(struct connection_struct *conn, files_struct *fsp)
         * Finally close the fd associated with this fsp.
         */
 
-       ret = conn->vfs_ops.close(fsp->fd);
+       ret = conn->vfs_ops.close(fsp,fsp->fd);
 
        if (saved_errno != 0) {
         errno = saved_errno;
@@ -299,46 +301,69 @@ static const char *posix_lock_type_name(int lock_type)
 }
 
 /****************************************************************************
Add an entry into the POSIX locking tdb. Returns the number of records that
match the given start and size, or -1 on error.
Delete a POSIX lock entry by index number. Used if the tdb add succeeds, but
then the POSIX fcntl lock fails.
 ****************************************************************************/
 
-static int add_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, int lock_type)
+static BOOL delete_posix_lock_entry_by_index(files_struct *fsp, size_t entry)
 {
        TDB_DATA kbuf = locking_key_fsp(fsp);
        TDB_DATA dbuf;
-       struct posix_lock pl;
-       struct posix_lock *entries;
-       size_t i, count;
-       int num_records = 0;
+       struct posix_lock *locks;
+       size_t count;
 
-       /*
-        * Windows is very strange. It allows read locks to be overlayed on 
-        * a write lock, but leaves the write lock in force until the first
-        * unlock. It also reference counts the locks. This means the following sequence :
-        *
-        * process1                                      process2
-        * ------------------------------------------------------------------------
-        * WRITE LOCK : start = 0, len = 10
-        *                                            READ LOCK: start =0, len = 10 - FAIL
-        * READ LOCK : start = 0, len = 10
-        *                                            READ LOCK: start =0, len = 10 - FAIL
-        * UNLOCK : start = 0, len = 10
-        *                                            READ LOCK: start =0, len = 10 - OK
-        *
-        * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but
-        * would leave a single read lock over the 0-10 region. In order to
-        * re-create Windows semantics mapped to POSIX locks, we create multiple TDB
-        * entries, one for each overlayed lock request. We are guarenteed by the brlock
-        * semantics that if a write lock is added, then it will be first in the array.
-        */
+       dbuf.dptr = NULL;
        
+       dbuf = tdb_fetch(posix_lock_tdb, kbuf);
+
+       if (!dbuf.dptr) {
+               DEBUG(10,("delete_posix_lock_entry_by_index: tdb_fetch failed !\n"));
+               goto fail;
+       }
+
+       count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
+       locks = (struct posix_lock *)dbuf.dptr;
+
+       if (count == 1) {
+               tdb_delete(posix_lock_tdb, kbuf);
+       } else {
+               if (entry < count-1) {
+                       memmove(&locks[entry], &locks[entry+1], sizeof(*locks)*((count-1) - entry));
+               }
+               dbuf.dsize -= sizeof(*locks);
+               tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
+       }
+
+       free(dbuf.dptr);
+
+       return True;
+
+ fail:
+    if (dbuf.dptr)
+               free(dbuf.dptr);
+    return False;
+}
+
+/****************************************************************************
+ Add an entry into the POSIX locking tdb. We return the index number of the
+ added lock (used in case we need to delete *exactly* this entry). Returns
+ False on fail, True on success.
+****************************************************************************/
+
+static BOOL add_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, int lock_type, size_t *pentry_num)
+{
+       TDB_DATA kbuf = locking_key_fsp(fsp);
+       TDB_DATA dbuf;
+       struct posix_lock pl;
+
        dbuf.dptr = NULL;
 
        dbuf = tdb_fetch(posix_lock_tdb, kbuf);
 
+       *pentry_num = (size_t)(dbuf.dsize / sizeof(pl));
+
        /*
-        * New record.
+        * Add new record.
         */
 
        pl.fd = fsp->fd;
@@ -355,19 +380,6 @@ static int add_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T si
        memcpy(dbuf.dptr + dbuf.dsize, &pl, sizeof(pl));
        dbuf.dsize += sizeof(pl);
 
-       count = (size_t)(dbuf.dsize / sizeof(pl));
-       entries = (struct posix_lock *)dbuf.dptr;
-
-       for (i = 0; i < count; i++) {
-               struct posix_lock *entry = &entries[i];
-
-               if (fsp->fd == entry->fd &&
-                       start == entry->start &&
-                       size == entry->size)
-                       num_records++;
-
-       }
-
        if (tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
                DEBUG(0,("add_posix_lock: Failed to add lock entry on file %s\n", fsp->fsp_name));
                goto fail;
@@ -375,21 +387,36 @@ static int add_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T si
 
     free(dbuf.dptr);
 
-       DEBUG(10,("add_posix_lock: File %s: type = %s: start=%.0f size=%.0f: num_records = %d : dev=%.0f inode=%.0f\n",
-                       fsp->fsp_name, posix_lock_type_name(lock_type), (double)start, (double)size, num_records,
+       DEBUG(10,("add_posix_lock: File %s: type = %s: start=%.0f size=%.0f: dev=%.0f inode=%.0f\n",
+                       fsp->fsp_name, posix_lock_type_name(lock_type), (double)start, (double)size,
                        (double)fsp->dev, (double)fsp->inode ));
 
-    return num_records;
+    return True;
 
  fail:
     if (dbuf.dptr)
                free(dbuf.dptr);
-    return -1;
+    return False;
+}
+
+/****************************************************************************
+ Calculate if locks have any overlap at all.
+****************************************************************************/
+
+static BOOL does_lock_overlap(SMB_OFF_T start1, SMB_OFF_T size1, SMB_OFF_T start2, SMB_OFF_T size2)
+{
+       if (start1 >= start2 && start1 <= start2 + size2)
+               return True;
+
+       if (start1 < start2 && start1 + size1 > start2)
+               return True;
+
+       return False;
 }
 
 /****************************************************************************
  Delete an entry from the POSIX locking tdb. Returns a copy of the entry being
- deleted and the number of remaining matching records, or -1 on error.
+ deleted and the number of records that are overlapped by this one, or -1 on error.
 ****************************************************************************/
 
 static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, struct posix_lock *pl)
@@ -398,7 +425,8 @@ static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T
        TDB_DATA dbuf;
        struct posix_lock *locks;
        size_t i, count;
-       int num_records = 0;
+       BOOL found = False;
+       int num_overlapping_records = 0;
 
        dbuf.dptr = NULL;
        
@@ -414,20 +442,10 @@ static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T
        count = (size_t)(dbuf.dsize / sizeof(*locks));
 
        /*
-        * Count the number of entries that match this
-        * unlock request.
+        * Search for and delete the first record that matches the
+        * unlock criteria.
         */
 
-       for (i = 0; i < count; i++) {
-               struct posix_lock *entry = &locks[i];
-
-               if (entry->fd == fsp->fd &&
-                       entry->start == start &&
-                       entry->size == size) {
-                               num_records++;
-               }
-       }
-
        for (i=0; i<count; i++) { 
                struct posix_lock *entry = &locks[i];
 
@@ -435,16 +453,10 @@ static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T
                        entry->start == start &&
                        entry->size == size) {
 
-                       num_records--; /* We're deleting one. */
-
                        /* Make a copy if requested. */
                        if (pl)
                                *pl = *entry;
 
-                       DEBUG(10,("delete_posix_lock_entry: type = %s: start=%.0f size=%.0f, num_records = %d\n",
-                                       posix_lock_type_name(entry->lock_type), (double)entry->start, (double)entry->size,
-                                       (unsigned int)num_records ));
-
                        /* Found it - delete it. */
                        if (count == 1) {
                                tdb_delete(posix_lock_tdb, kbuf);
@@ -455,13 +467,36 @@ static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T
                                dbuf.dsize -= sizeof(*locks);
                                tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
                        }
-
-                       free(dbuf.dptr);
-                       return num_records;
+                       count--;
+                       found = True;
+                       break;
                }
        }
 
-       /* We didn't find it. */
+       if (!found)
+               goto fail;
+
+       /*
+        * Count the number of entries that are
+        * overlapped by this unlock request.
+        */
+
+       for (i = 0; i < count; i++) {
+               struct posix_lock *entry = &locks[i];
+
+               if (fsp->fd == entry->fd &&
+                       does_lock_overlap( start, size, entry->start, entry->size))
+                               num_overlapping_records++;
+       }
+
+       DEBUG(10,("delete_posix_lock_entry: type = %s: start=%.0f size=%.0f, num_records = %d\n",
+                       posix_lock_type_name(pl->lock_type), (double)pl->start, (double)pl->size,
+                               (unsigned int)num_overlapping_records ));
+
+    if (dbuf.dptr)
+               free(dbuf.dptr);
+
+       return num_overlapping_records;
 
  fail:
     if (dbuf.dptr)
@@ -509,177 +544,81 @@ static int map_posix_lock_type( files_struct *fsp, enum brl_type lock_type)
 static BOOL posix_lock_in_range(SMB_OFF_T *offset_out, SMB_OFF_T *count_out,
                                                                SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
 {
-       SMB_OFF_T offset;
-       SMB_OFF_T count;
-
-#if defined(LARGE_SMB_OFF_T) && !defined(HAVE_BROKEN_FCNTL64_LOCKS)
-
-    SMB_OFF_T mask2 = ((SMB_OFF_T)0x4) << (SMB_OFF_T_BITS-4);
-    SMB_OFF_T mask = (mask2<<1);
-    SMB_OFF_T neg_mask = ~mask;
-
-       /*
-        * In this case SMB_OFF_T is 64 bits,
-        * and the underlying system can handle 64 bit signed locks.
-        * Cast to signed type.
-        */
-
-       offset = (SMB_OFF_T)u_offset;
-       count = (SMB_OFF_T)u_count;
+       SMB_OFF_T offset = (SMB_OFF_T)u_offset;
+       SMB_OFF_T count = (SMB_OFF_T)u_count;
 
        /*
-        * Deal with a very common case of count of all ones.
-        * (lock entire file).
+        * For the type of system we are, attempt to
+        * find the maximum positive lock offset as an SMB_OFF_T.
         */
 
-       if(count == (SMB_OFF_T)-1)
-               count &= ~mask;
+#if defined(LARGE_SMB_OFF_T) && !defined(HAVE_BROKEN_FCNTL64_LOCKS)
 
        /*
-        * POSIX lock ranges cannot be negative.
-        * Fail if any combination becomes negative.
+        * In this case SMB_OFF_T is 64 bits,
+        * and the underlying system can handle 64 bit signed locks.
         */
 
-       if(offset < 0 || count < 0 || (offset + count < 0)) {
-               DEBUG(10,("posix_lock_in_range: negative range: offset = %.0f, count = %.0f. Ignoring lock.\n",
-                               (double)offset, (double)count ));
-               return False;
-       }
-
-       /*
-        * In this case SMB_OFF_T is 64 bits, the offset and count
-        * fit within the positive range, and the underlying
-        * system can handle 64 bit locks. Just return as the
-        * cast values are ok.
-        */
+    SMB_OFF_T mask2 = ((SMB_OFF_T)0x4) << (SMB_OFF_T_BITS-4);
+    SMB_OFF_T mask = (mask2<<1);
+    SMB_OFF_T max_positive_lock_offset = ~mask;
 
 #else /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
 
        /*
         * In this case either SMB_OFF_T is 32 bits,
         * or the underlying system cannot handle 64 bit signed locks.
-        * Either way we have to try and mangle to fit within 31 bits.
-        * This is difficult.
+        * All offsets & counts must be 2^31 or less.
         */
 
-#if defined(HAVE_BROKEN_FCNTL64_LOCKS)
+    SMB_OFF_T max_positive_lock_offset = 0x7FFFFFFF;
 
-       /*
-        * SMB_OFF_T is 64 bits, but we need to use 31 bits due to
-        * broken large locking.
-        */
+#endif /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
 
        /*
-        * Deal with a very common case of count of all ones.
-        * (lock entire file).
+        * If the given offset was > max_positive_lock_offset then we cannot map this at all
+        * ignore this lock.
         */
 
-       if(u_count == (SMB_BIG_UINT)-1)
-               count = 0x7FFFFFFF;
-
-       if(((u_offset >> 32) & 0xFFFFFFFF) || ((u_count >> 32) & 0xFFFFFFFF)) {
-               DEBUG(10,("posix_lock_in_range: top 32 bits not zero. offset = %.0f, count = %.0f. Ignoring lock.\n",
-                               (double)u_offset, (double)u_count ));
-               /* Top 32 bits of offset or count were not zero. */
+       if (u_offset & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
+               DEBUG(10,("posix_lock_in_range: (offset = %.0f) offset > %.0f and we cannot handle this. Ignoring lock.\n",
+                               (double)u_offset, (double)((SMB_BIG_UINT)max_positive_lock_offset) ));
                return False;
        }
 
-       /* Cast from 64 bits unsigned to 64 bits signed. */
-       offset = (SMB_OFF_T)u_offset;
-       count = (SMB_OFF_T)u_count;
-
        /*
-        * Check if we are within the 2^31 range.
+        * We must truncate the offset and count to less than max_positive_lock_offset.
         */
 
-       {
-               int32 low_offset = (int32)offset;
-               int32 low_count = (int32)count;
+       offset &= max_positive_lock_offset;
+       count &= max_positive_lock_offset;
 
-               if(low_offset < 0 || low_count < 0 || (low_offset + low_count < 0)) {
-                       DEBUG(10,("posix_lock_in_range: not within 2^31 range. low_offset = %d, low_count = %d. Ignoring lock.\n",
-                                       low_offset, low_count ));
-                       return False;
-               }
-       }
-
-       /*
-        * Ok - we can map from a 64 bit number to a 31 bit lock.
-        */
-
-#else /* HAVE_BROKEN_FCNTL64_LOCKS */
-
-       /*
-        * SMB_OFF_T is 32 bits.
-        */
-
-#if defined(HAVE_LONGLONG)
-
-       /*
-        * SMB_BIG_UINT is 64 bits, we can do a 32 bit shift.
-        */
 
        /*
         * Deal with a very common case of count of all ones.
         * (lock entire file).
         */
 
-       if(u_count == (SMB_BIG_UINT)-1)
-               count = 0x7FFFFFFF;
-
-       if(((u_offset >> 32) & 0xFFFFFFFF) || ((u_count >> 32) & 0xFFFFFFFF)) {
-               DEBUG(10,("posix_lock_in_range: top 32 bits not zero. u_offset = %.0f, u_count = %.0f. Ignoring lock.\n",
-                               (double)u_offset, (double)u_count ));
-               return False;
-       }
-
-       /* Cast from 64 bits unsigned to 32 bits signed. */
-       offset = (SMB_OFF_T)u_offset;
-       count = (SMB_OFF_T)u_count;
-
-       /*
-        * Check if we are within the 2^31 range.
-        */
-
-       if(offset < 0 || count < 0 || (offset + count < 0)) {
-               DEBUG(10,("posix_lock_in_range: not within 2^31 range. offset = %d, count = %d. Ignoring lock.\n",
-                               (int)offset, (int)count ));
-               return False;
-       }
-
-#else /* HAVE_LONGLONG */
-
-       /*
-        * SMB_BIG_UINT and SMB_OFF_T are both 32 bits,
-        * just cast.
-        */
+       if(count == (SMB_OFF_T)-1)
+               count = max_positive_lock_offset;
 
        /*
-        * Deal with a very common case of count of all ones.
-        * (lock entire file).
+        * Truncate count to end at max lock offset.
         */
 
-       if(u_count == (SMB_BIG_UINT)-1)
-               count = 0x7FFFFFFF;
-
-       /* Cast from 32 bits unsigned to 32 bits signed. */
-       offset = (SMB_OFF_T)u_offset;
-       count = (SMB_OFF_T)u_count;
+       if (offset + count < 0 || offset + count > max_positive_lock_offset)
+               count = max_positive_lock_offset - offset;
 
        /*
-        * Check if we are within the 2^31 range.
+        * If we ate all the count, ignore this lock.
         */
 
-       if(offset < 0 || count < 0 || (offset + count < 0)) {
-               DEBUG(10,("posix_lock_in_range: not within 2^31 range. offset = %d, count = %d. Ignoring lock.\n",
-                               (int)offset, (int)count ));
+       if (count == 0) {
+               DEBUG(10,("posix_lock_in_range: Count = 0. Ignoring lock u_offset = %.0f, u_count = %.0f\n",
+                               (double)u_offset, (double)u_count ));
                return False;
        }
 
-#endif /* HAVE_LONGLONG */
-#endif /* LARGE_SMB_OFF_T */
-#endif /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
-
        /*
         * The mapping was successful.
         */
@@ -749,7 +688,7 @@ static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OF
 
        DEBUG(8,("posix_fcntl_lock %d %d %.0f %.0f %d\n",fsp->fd,op,(double)offset,(double)count,type));
 
-       ret = conn->vfs_ops.lock(fsp->fd,op,offset,count,type);
+       ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
 
        if (!ret && (errno == EFBIG)) {
                if( DEBUGLVL( 0 )) {
@@ -760,7 +699,7 @@ static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OF
                /* 32 bit NFS file system, retry with smaller offset */
                errno = 0;
                count &= 0x7fffffff;
-               ret = conn->vfs_ops.lock(fsp->fd,op,offset,count,type);
+               ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
        }
 
        /* A lock query - just return. */
@@ -788,7 +727,7 @@ static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OF
 
                                count = (orig_count & 0x7FFFFFFF);
                                offset = (SMB_OFF_T)map_lock_offset(off_high, off_low);
-                               ret = conn->vfs_ops.lock(fsp->fd,op,offset,count,type);
+                               ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
                                if (!ret) {
                                        if (errno == EINVAL) {
                                                DEBUG(3,("posix_fcntl_lock: locking not supported? returning True\n"));
@@ -844,77 +783,25 @@ BOOL is_posix_locked(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_co
        return posix_fcntl_lock(fsp,SMB_F_GETLK,offset,count,posix_lock_type);
 }
 
-/****************************************************************************
- POSIX function to acquire a lock. Returns True if the
- lock could be granted, False if not.
-****************************************************************************/
-
-BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
-{
-       SMB_OFF_T offset;
-       SMB_OFF_T count;
-       BOOL ret = True;
-       int posix_lock_type = map_posix_lock_type(fsp,lock_type);
-       int ref_count;
-
-       DEBUG(5,("set_posix_lock: File %s, offset = %.0f, count = %.0f, type = %s\n",
-                       fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
-
-       /*
-        * If the requested lock won't fit in the POSIX range, we will
-        * pretend it was successful.
-        */
-
-       if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
-               return True;
-
-       /*
-        * Note that setting multiple overlapping locks on different
-        * file descriptors will not be held separately by the kernel (POSIX
-        * braindamage), but will be merged into one continuous lock
-        * range. We cope with this case in the release_posix_lock code
-        * below. We need to add the posix lock entry into the tdb before
-        * doing the real posix lock call to deal with the locking overlay
-        * case described above in add_posix_lock_entry().
-        */
-
-       ref_count = add_posix_lock_entry(fsp,offset,count,posix_lock_type);
-
-       if (ref_count == 1) {
-               /*
-                * First lock entry created. Do a real POSIX lock.
-                */
-           ret = posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type);
-
-               /*
-                * Oops, POSIX lock failed, delete the tdb entry.
-                */
-               if (!ret)
-                       delete_posix_lock_entry(fsp,offset,count,NULL);
-       }
-
-       return ret;
-}
-
 /*
  * Structure used when splitting a lock range
  * into a POSIX lock range. Doubly linked list.
  */
 
-struct unlock_list {
-    struct unlock_list *next;
-    struct unlock_list *prev;
+struct lock_list {
+    struct lock_list *next;
+    struct lock_list *prev;
     SMB_OFF_T start;
     SMB_OFF_T size;
 };
 
 /****************************************************************************
  Create a list of lock ranges that don't overlap a given range. Used in calculating
- POSIX lock unlocks. This is a difficult function that requires ASCII art to
+ POSIX locks and unlocks. This is a difficult function that requires ASCII art to
  understand it :-).
 ****************************************************************************/
 
-static struct unlock_list *posix_unlock_list(TALLOC_CTX *ctx, struct unlock_list *ulhead, files_struct *fsp)
+static struct lock_list *posix_lock_list(TALLOC_CTX *ctx, struct lock_list *lhead, files_struct *fsp)
 {
        TDB_DATA kbuf = locking_key_fsp(fsp);
        TDB_DATA dbuf;
@@ -925,9 +812,8 @@ static struct unlock_list *posix_unlock_list(TALLOC_CTX *ctx, struct unlock_list
 
        dbuf = tdb_fetch(posix_lock_tdb, kbuf);
 
-       if (!dbuf.dptr) {
-               return ulhead;
-       }
+       if (!dbuf.dptr)
+               return lhead;
        
        locks = (struct posix_lock *)dbuf.dptr;
        num_locks = (size_t)(dbuf.dsize / sizeof(*locks));
@@ -937,48 +823,48 @@ static struct unlock_list *posix_unlock_list(TALLOC_CTX *ctx, struct unlock_list
         * Quit if the list is deleted.
         */
 
-       DEBUG(10,("posix_unlock_list: curr: start=%.0f,size=%.0f\n",
-               (double)ulhead->start, (double)ulhead->size ));
+       DEBUG(10,("posix_lock_list: curr: start=%.0f,size=%.0f\n",
+               (double)lhead->start, (double)lhead->size ));
 
-       for (i=0; i<num_locks && ulhead; i++) {
+       for (i=0; i<num_locks && lhead; i++) {
 
                struct posix_lock *lock = &locks[i];
-               struct unlock_list *ul_curr;
+               struct lock_list *l_curr;
 
                /*
-                * Walk the unlock list, checking for overlaps. Note that
-                * the unlock list can expand within this loop if the current
+                * Walk the lock list, checking for overlaps. Note that
+                * the lock list can expand within this loop if the current
                 * range being examined needs to be split.
                 */
 
-               for (ul_curr = ulhead; ul_curr;) {
+               for (l_curr = lhead; l_curr;) {
 
-                       DEBUG(10,("posix_unlock_list: lock: start=%.0f,size=%.0f:",
-                               (double)lock->start, (double)lock->size ));
+                       DEBUG(10,("posix_lock_list: lock: fd=%d: start=%.0f,size=%.0f:type=%s", lock->fd,
+                               (double)lock->start, (double)lock->size, posix_lock_type_name(lock->lock_type) ));
 
-                       if ( (ul_curr->start >= (lock->start + lock->size)) ||
-                                (lock->start >= (ul_curr->start + ul_curr->size))) {
+                       if ( (l_curr->start >= (lock->start + lock->size)) ||
+                                (lock->start >= (l_curr->start + l_curr->size))) {
 
                                /* No overlap with this lock - leave this range alone. */
 /*********************************************
                                              +---------+
-                                             | ul_curr |
+                                             | l_curr  |
                                              +---------+
                                 +-------+
                                 | lock  |
                                 +-------+
 OR....
              +---------+
-             | ul_curr |
+             |  l_curr |
              +---------+
 **********************************************/
 
                                DEBUG(10,("no overlap case.\n" ));
 
-                               ul_curr = ul_curr->next;
+                               l_curr = l_curr->next;
 
-                       } else if ( (ul_curr->start >= lock->start) &&
-                                               (ul_curr->start + ul_curr->size <= lock->start + lock->size) ) {
+                       } else if ( (l_curr->start >= lock->start) &&
+                                               (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
 
                                /*
                                 * This unlock is completely overlapped by this existing lock range
@@ -986,26 +872,26 @@ OR....
                                 */
 /*********************************************
                 +---------+
-                | ul_curr |
+                |  l_curr |
                 +---------+
         +---------------------------+
         |       lock                |
         +---------------------------+
 **********************************************/
                                /* Save the next pointer */
-                               struct unlock_list *ul_next = ul_curr->next;
+                               struct lock_list *ul_next = l_curr->next;
 
                                DEBUG(10,("delete case.\n" ));
 
-                               DLIST_REMOVE(ulhead, ul_curr);
-                               if(ulhead == NULL)
+                               DLIST_REMOVE(lhead, l_curr);
+                               if(lhead == NULL)
                                        break; /* No more list... */
 
-                               ul_curr = ul_next;
+                               l_curr = ul_next;
                                
-                       } else if ( (ul_curr->start >= lock->start) &&
-                                               (ul_curr->start < lock->start + lock->size) &&
-                                               (ul_curr->start + ul_curr->size > lock->start + lock->size) ) {
+                       } else if ( (l_curr->start >= lock->start) &&
+                                               (l_curr->start < lock->start + lock->size) &&
+                                               (l_curr->start + l_curr->size > lock->start + lock->size) ) {
 
                                /*
                                 * This unlock overlaps the existing lock range at the high end.
@@ -1013,27 +899,28 @@ OR....
                                 */
 /*********************************************
                 +---------------+
-                | ul_curr       |
+                |  l_curr       |
                 +---------------+
         +---------------+
         |    lock       |
         +---------------+
 BECOMES....
                         +-------+
-                        |ul_curr|
+                        | l_curr|
                         +-------+
 **********************************************/
 
-                               ul_curr->size = (ul_curr->start + ul_curr->size) - (lock->start + lock->size);
-                               ul_curr->start = lock->start + lock->size;
+                               l_curr->size = (l_curr->start + l_curr->size) - (lock->start + lock->size);
+                               l_curr->start = lock->start + lock->size;
 
                                DEBUG(10,("truncate high case: start=%.0f,size=%.0f\n",
-                                                               (double)ul_curr->start, (double)ul_curr->size ));
+                                                               (double)l_curr->start, (double)l_curr->size ));
 
-                               ul_curr = ul_curr->next;
+                               l_curr = l_curr->next;
 
-                       } else if ( (ul_curr->start < lock->start) &&
-                                               (ul_curr->start + ul_curr->size > lock->start) ) {
+                       } else if ( (l_curr->start < lock->start) &&
+                                               (l_curr->start + l_curr->size > lock->start) &&
+                                               (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
 
                                /*
                                 * This unlock overlaps the existing lock range at the low end.
@@ -1041,26 +928,26 @@ BECOMES....
                                 */
 /*********************************************
    +---------------+
-   | ul_curr       |
+   |  l_curr       |
    +---------------+
            +---------------+
            |    lock       |
            +---------------+
 BECOMES....
    +-------+
-   |ul_curr|
+   | l_curr|
    +-------+
 **********************************************/
 
-                               ul_curr->size = lock->start - ul_curr->start;
+                               l_curr->size = lock->start - l_curr->start;
 
                                DEBUG(10,("truncate low case: start=%.0f,size=%.0f\n",
-                                                               (double)ul_curr->start, (double)ul_curr->size ));
+                                                               (double)l_curr->start, (double)l_curr->size ));
 
-                               ul_curr = ul_curr->next;
+                               l_curr = l_curr->next;
                
-                       } else if ( (ul_curr->start < lock->start) &&
-                                               (ul_curr->start + ul_curr->size > lock->start + lock->size) ) {
+                       } else if ( (l_curr->start < lock->start) &&
+                                               (l_curr->start + l_curr->size > lock->start + lock->size) ) {
                                /*
                                 * Worst case scenario. Unlock request completely overlaps an existing
                                 * lock range. Split the request into two, push the new (upper) request
@@ -1069,39 +956,46 @@ BECOMES....
                                 */
 /*********************************************
         +---------------------------+
-        |       ul_curr             |
+        |        l_curr             |
         +---------------------------+
                 +---------+
                 | lock    |
                 +---------+
 BECOMES.....
         +-------+         +---------+
-        |ul_curr|         |ul_new   |
+        | l_curr|         | l_new   |
         +-------+         +---------+
 **********************************************/
-                               struct unlock_list *ul_new = (struct unlock_list *)talloc(ctx,
-                                                                                                       sizeof(struct unlock_list));
+                               struct lock_list *l_new = (struct lock_list *)talloc(ctx,
+                                                                                                       sizeof(struct lock_list));
 
-                               if(ul_new == NULL) {
-                                       DEBUG(0,("posix_unlock_list: talloc fail.\n"));
+                               if(l_new == NULL) {
+                                       DEBUG(0,("posix_lock_list: talloc fail.\n"));
                                        return NULL; /* The talloc_destroy takes care of cleanup. */
                                }
 
-                               ZERO_STRUCTP(ul_new);
-                               ul_new->start = lock->start + lock->size;
-                               ul_new->size = ul_curr->start + ul_curr->size - ul_new->start;
+                               ZERO_STRUCTP(l_new);
+                               l_new->start = lock->start + lock->size;
+                               l_new->size = l_curr->start + l_curr->size - l_new->start;
 
-                               /* Add into the dlink list after the ul_curr point - NOT at ulhead. */
-                               DLIST_ADD(ul_curr, ul_new);
-
-                               /* Truncate the ul_curr. */
-                               ul_curr->size = lock->start - ul_curr->start;
+                               /* Truncate the l_curr. */
+                               l_curr->size = lock->start - l_curr->start;
 
                                DEBUG(10,("split case: curr: start=%.0f,size=%.0f \
-new: start=%.0f,size=%.0f\n", (double)ul_curr->start, (double)ul_curr->size,
-                                                               (double)ul_new->start, (double)ul_new->size ));
+new: start=%.0f,size=%.0f\n", (double)l_curr->start, (double)l_curr->size,
+                                                               (double)l_new->start, (double)l_new->size ));
+
+                               /*
+                                * Add into the dlink list after the l_curr point - NOT at lhead. 
+                                * Note we can't use DLINK_ADD here as this inserts at the head of the given list.
+                                */
+
+                               l_new->prev = l_curr;
+                               l_new->next = l_curr->next;
+                               l_curr->next = l_new;
 
-                               ul_curr = ul_new->next;
+                               /* And move after the link we added. */
+                               l_curr = l_new->next;
 
                        } else {
 
@@ -1111,18 +1005,159 @@ new: start=%.0f,size=%.0f\n", (double)ul_curr->start, (double)ul_curr->size,
                                 */
                                pstring msg;
 
-                               slprintf(msg, sizeof(msg)-1, "logic flaw in cases: ul_curr: start = %.0f, size = %.0f : \
-lock: start = %.0f, size = %.0f\n", (double)ul_curr->start, (double)ul_curr->size, (double)lock->start, (double)lock->size );
+                               slprintf(msg, sizeof(msg)-1, "logic flaw in cases: l_curr: start = %.0f, size = %.0f : \
+lock: start = %.0f, size = %.0f\n", (double)l_curr->start, (double)l_curr->size, (double)lock->start, (double)lock->size );
 
                                smb_panic(msg);
                        }
-               } /* end for ( ul_curr = ulhead; ul_curr;) */
+               } /* end for ( l_curr = lhead; l_curr;) */
        } /* end for (i=0; i<num_locks && ul_head; i++) */
 
        if (dbuf.dptr)
                free(dbuf.dptr);
        
-       return ulhead;
+       return lhead;
+}
+
+/****************************************************************************
+ POSIX function to acquire a lock. Returns True if the
+ lock could be granted, False if not.
+****************************************************************************/
+
+BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
+{
+       SMB_OFF_T offset;
+       SMB_OFF_T count;
+       BOOL ret = True;
+       size_t entry_num = 0;
+       size_t lock_count;
+       TALLOC_CTX *l_ctx = NULL;
+       struct lock_list *llist = NULL;
+       struct lock_list *ll = NULL;
+       int posix_lock_type = map_posix_lock_type(fsp,lock_type);
+
+       DEBUG(5,("set_posix_lock: File %s, offset = %.0f, count = %.0f, type = %s\n",
+                       fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
+
+       /*
+        * If the requested lock won't fit in the POSIX range, we will
+        * pretend it was successful.
+        */
+
+       if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
+               return True;
+
+       /*
+        * Windows is very strange. It allows read locks to be overlayed
+        * (even over a write lock), but leaves the write lock in force until the first
+        * unlock. It also reference counts the locks. This means the following sequence :
+        *
+        * process1                                      process2
+        * ------------------------------------------------------------------------
+        * WRITE LOCK : start = 2, len = 10
+        *                                            READ LOCK: start =0, len = 10 - FAIL
+        * READ LOCK : start = 0, len = 14 
+        *                                            READ LOCK: start =0, len = 10 - FAIL
+        * UNLOCK : start = 2, len = 10
+        *                                            READ LOCK: start =0, len = 10 - OK
+        *
+        * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but
+        * would leave a single read lock over the 0-14 region. In order to
+        * re-create Windows semantics mapped to POSIX locks, we create multiple TDB
+        * entries, one for each overlayed lock request. We are guarenteed by the brlock
+        * semantics that if a write lock is added, then it will be first in the array.
+        */
+       
+       if ((l_ctx = talloc_init()) == NULL) {
+               DEBUG(0,("set_posix_lock: unable to init talloc context.\n"));
+               return True; /* Not a fatal error. */
+       }
+
+       if ((ll = (struct lock_list *)talloc(l_ctx, sizeof(struct lock_list))) == NULL) {
+               DEBUG(0,("set_posix_lock: unable to talloc unlock list.\n"));
+               talloc_destroy(l_ctx);
+               return True; /* Not a fatal error. */
+       }
+
+       /*
+        * Create the initial list entry containing the
+        * lock we want to add.
+        */
+
+       ZERO_STRUCTP(ll);
+       ll->start = offset;
+       ll->size = count;
+
+       DLIST_ADD(llist, ll);
+
+       /*
+        * The following call calculates if there are any
+        * overlapping locks held by this process on
+        * fd's open on the same file and splits this list
+        * into a list of lock ranges that do not overlap with existing
+        * POSIX locks.
+        */
+
+       llist = posix_lock_list(l_ctx, llist, fsp);
+
+       /*
+        * Now we have the list of ranges to lock it is safe to add the
+        * entry into the POSIX lock tdb. We take note of the entry we
+        * added here in case we have to remove it on POSIX lock fail.
+        */
+
+       if (!add_posix_lock_entry(fsp,offset,count,posix_lock_type,&entry_num)) {
+               DEBUG(0,("set_posix_lock: Unable to create posix lock entry !\n"));
+               talloc_destroy(l_ctx);
+               return False;
+       }
+
+       /*
+        * Add the POSIX locks on the list of ranges returned.
+        * As the lock is supposed to be added atomically, we need to
+        * back out all the locks if any one of these calls fail.
+        */
+
+       for (lock_count = 0, ll = llist; ll; ll = ll->next, lock_count++) {
+               offset = ll->start;
+               count = ll->size;
+
+               DEBUG(5,("set_posix_lock: Real lock: Type = %s: offset = %.0f, count = %.0f\n",
+                       posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
+
+               if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type)) {
+                       DEBUG(5,("set_posix_lock: Lock fail !: Type = %s: offset = %.0f, count = %.0f\n",
+                               posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
+                       ret = False;
+                       break;
+               }
+       }
+
+       if (!ret) {
+
+               /*
+                * Back out all the POSIX locks we have on fail.
+                */
+
+               for (ll = llist; lock_count; ll = ll->next, lock_count--) {
+                       offset = ll->start;
+                       count = ll->size;
+
+                       DEBUG(5,("set_posix_lock: Backing out locks: Type = %s: offset = %.0f, count = %.0f\n",
+                               posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
+
+                       posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK);
+               }
+
+               /*
+                * Remove the tdb entry for this lock.
+                */
+
+               delete_posix_lock_entry_by_index(fsp,entry_num);
+       }
+
+       talloc_destroy(l_ctx);
+       return ret;
 }
 
 /****************************************************************************
@@ -1136,10 +1171,10 @@ BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u
        SMB_OFF_T count;
        BOOL ret = True;
        TALLOC_CTX *ul_ctx = NULL;
-       struct unlock_list *ulist = NULL;
-       struct unlock_list *ul = NULL;
+       struct lock_list *ulist = NULL;
+       struct lock_list *ul = NULL;
        struct posix_lock deleted_lock;
-       int num_entries;
+       int num_overlapped_entries;
 
        DEBUG(5,("release_posix_lock: File %s, offset = %.0f, count = %.0f\n",
                fsp->fsp_name, (double)u_offset, (double)u_count ));
@@ -1154,40 +1189,37 @@ BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u
 
        /*
         * We treat this as one unlock request for POSIX accounting purposes even
-        * if it may have been split into multiple smaller POSIX unlock ranges.
+        * if it may later be split into multiple smaller POSIX unlock ranges.
+        * num_overlapped_entries is the number of existing locks that have any
+        * overlap with this unlock request.
         */ 
 
-       num_entries = delete_posix_lock_entry(fsp, offset, count, &deleted_lock);
+       num_overlapped_entries = delete_posix_lock_entry(fsp, offset, count, &deleted_lock);
 
-       if (num_entries == -1) {
+       if (num_overlapped_entries == -1) {
         smb_panic("release_posix_lock: unable find entry to delete !\n");
        }
 
        /*
-        * If num_entries is > 0, and the lock_type we just deleted from the tdb was
-        * a POSIX write lock, then rather than doing an unlock we need to downgrade
-        * the POSIX lock to a read lock.
-        */
-
-       if (num_entries > 0 && deleted_lock.lock_type == F_WRLCK) {
-               return posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK);
-       }
-
-       /*
-        * Only do the POSIX unlock when the num_entries is now zero.
+        * If num_overlapped_entries is > 0, and the lock_type we just deleted from the tdb was
+        * a POSIX write lock, then before doing the unlock we need to downgrade
+        * the POSIX lock to a read lock. This allows any overlapping read locks
+        * to be atomically maintained.
         */
 
-       if (num_entries > 0) {
-               DEBUG(10, ("release_posix_lock: num_entries = %d\n", num_entries ));
-               return True;
+       if (num_overlapped_entries > 0 && deleted_lock.lock_type == F_WRLCK) {
+               if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK)) {
+                       DEBUG(0,("release_posix_lock: downgrade of lock failed !\n"));
+                       return False;
+               }
        }
 
        if ((ul_ctx = talloc_init()) == NULL) {
-        DEBUG(0,("release_posix_lock: unable to init talloc context.\n"));
+               DEBUG(0,("release_posix_lock: unable to init talloc context.\n"));
                return True; /* Not a fatal error. */
        }
 
-       if ((ul = (struct unlock_list *)talloc(ul_ctx, sizeof(struct unlock_list))) == NULL) {
+       if ((ul = (struct lock_list *)talloc(ul_ctx, sizeof(struct lock_list))) == NULL) {
                DEBUG(0,("release_posix_lock: unable to talloc unlock list.\n"));
                talloc_destroy(ul_ctx);
                return True; /* Not a fatal error. */
@@ -1213,7 +1245,7 @@ BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u
         * unlocks are performed.
         */
 
-       ulist = posix_unlock_list(ul_ctx, ulist, fsp);
+       ulist = posix_lock_list(ul_ctx, ulist, fsp);
 
        /*
         * Release the POSIX locks on the list of ranges returned.
@@ -1223,16 +1255,6 @@ BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u
                offset = ulist->start;
                count = ulist->size;
 
-               if(u_count == 0) {
-
-                       /*
-                        * This lock must overlap with an existing lock.
-                        * Don't do any POSIX call.
-                        */
-
-                       continue;
-               }
-
                DEBUG(5,("release_posix_lock: Real unlock: offset = %.0f, count = %.0f\n",
                        (double)offset, (double)count ));
 
@@ -1324,25 +1346,25 @@ void posix_locking_close_file(files_struct *fsp)
  Create the in-memory POSIX lock databases.
 ********************************************************************/
 
-BOOL posix_locking_init(void)
+BOOL posix_locking_init(int read_only)
 {
        if (posix_lock_tdb && posix_pending_close_tdb)
                return True;
-
+       
        if (!posix_lock_tdb)
                posix_lock_tdb = tdb_open(NULL, 0, TDB_INTERNAL,
-                                         O_RDWR|O_CREAT, 0644);
-    if (!posix_lock_tdb) {
-        DEBUG(0,("Failed to open POSIX byte range locking database.\n"));
+                                         read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
+       if (!posix_lock_tdb) {
+               DEBUG(0,("Failed to open POSIX byte range locking database.\n"));
                return False;
-    }
+       }
        if (!posix_pending_close_tdb)
                posix_pending_close_tdb = tdb_open(NULL, 0, TDB_INTERNAL,
-                   O_RDWR|O_CREAT, 0644);
-    if (!posix_pending_close_tdb) {
-        DEBUG(0,("Failed to open POSIX pending close database.\n"));
+                                                  read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
+       if (!posix_pending_close_tdb) {
+               DEBUG(0,("Failed to open POSIX pending close database.\n"));
                return False;
-    }
+       }
 
        return True;
 }