posix.c

samba-3.0.22.tar.gz 编译smb服务器的源码

				DEBUG(10,("truncate low case: start=%.0f,size=%.0f\n",
								(double)l_curr->start, (double)l_curr->size ));

				l_curr = l_curr->next;
		
			} 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
				 * into the dlink list, and continue with the entry after ul_new (as we
				 * know that ul_new will not overlap with this lock).
				 */
/*********************************************
        +---------------------------+
        |        l_curr             |
        +---------------------------+
                +---------+
                | lock    |
                +---------+
BECOMES.....
        +-------+         +---------+
        | l_curr|         | l_new   |
        +-------+         +---------+
**********************************************/
				struct lock_list *l_new = TALLOC_P(ctx, struct lock_list);

				if(l_new == NULL) {
					DEBUG(0,("posix_lock_list: talloc fail.\n"));
					return NULL; /* The talloc_destroy takes care of cleanup. */
				}

				ZERO_STRUCTP(l_new);
				l_new->start = lock->start + lock->size;
				l_new->size = l_curr->start + l_curr->size - l_new->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)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;

				/* And move after the link we added. */
				l_curr = l_new->next;

			} else {

				/*
				 * This logic case should never happen. Ensure this is the
				 * case by forcing an abort.... Remove in production.
				 */
				pstring msg;

				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 ( l_curr = lhead; l_curr;) */
	} /* end for (i=0; i<num_locks && ul_head; i++) */

	SAFE_FREE(dbuf.dptr);
	
	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("set_posix_lock")) == NULL) {
		DEBUG(0,("set_posix_lock: unable to init talloc context.\n"));
		return True; /* Not a fatal error. */
	}

	if ((ll = TALLOC_P(l_ctx, 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. Errno = %s\n",
				posix_lock_type_name(posix_lock_type), (double)offset, (double)count, strerror(errno) ));
			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;
}

/****************************************************************************
 POSIX function to release a lock. Returns True if the
 lock could be released, False if not.
****************************************************************************/

BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
{
	SMB_OFF_T offset;
	SMB_OFF_T count;
	BOOL ret = True;
	TALLOC_CTX *ul_ctx = NULL;
	struct lock_list *ulist = NULL;
	struct lock_list *ul = NULL;
	struct posix_lock deleted_lock;
	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 ));

	/*
	 * 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;

	/*
	 * We treat this as one unlock request for POSIX accounting purposes even
	 * 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_overlapped_entries = delete_posix_lock_entry(fsp, offset, count, &deleted_lock);

	if (num_overlapped_entries == -1) {
		smb_panic("release_posix_lock: unable find entry to delete !\n");
	}

	/*
	 * 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_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 with error %s !\n", strerror(errno) ));
			return False;
		}
	}

	if ((ul_ctx = talloc_init("release_posix_lock")) == NULL) {
		DEBUG(0,("release_posix_lock: unable to init talloc context.\n"));
		return True; /* Not a fatal error. */
	}

	if ((ul = TALLOC_P(ul_ctx, 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. */
	}

	/*
	 * Create the initial list entry containing the
	 * lock we want to remove.
	 */

	ZERO_STRUCTP(ul);
	ul->start = offset;
	ul->size = count;

	DLIST_ADD(ulist, ul);

	/*
	 * The following call calculates if there are any
	 * overlapping locks held by this process on
	 * fd's open on the same file and creates a
	 * list of unlock ranges that will allow
	 * POSIX lock ranges to remain on the file whilst the
	 * unlocks are performed.
	 */

	ulist = posix_lock_list(ul_ctx, ulist, fsp);

	/*
	 * Release the POSIX locks on the list of ranges returned.
	 */

	for(; ulist; ulist = ulist->next) {
		offset = ulist->start;
		count = ulist->size;

		DEBUG(5,("release_posix_lock: Real unlock: offset = %.0f, count = %.0f\n",
			(double)offset, (double)count ));

		if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK))
			ret = False;
	}

	talloc_destroy(ul_ctx);

	return ret;
}

/****************************************************************************
 Remove all lock entries for a specific dev/inode pair from the tdb.
****************************************************************************/

static void delete_posix_lock_entries(files_struct *fsp)
{
	TDB_DATA kbuf = locking_key_fsp(fsp);

	if (tdb_delete(posix_lock_tdb, kbuf) == -1)
		DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
}

/****************************************************************************
 Debug function.
****************************************************************************/

static void dump_entry(struct posix_lock *pl)
{
	DEBUG(10,("entry: start=%.0f, size=%.0f, type=%d, fd=%i\n",
		(double)pl->start, (double)pl->size, (int)pl->lock_type, pl->fd ));
}

/****************************************************************************
 Remove any locks on this fd. Called from file_close().
****************************************************************************/

void posix_locking_close_file(files_struct *fsp)
{
	struct posix_lock *entries = NULL;
	size_t count, i;

	/*
	 * Optimization for the common case where we are the only
	 * opener of a file. If all fd entries are our own, we don't
	 * need to explicitly release all the locks via the POSIX functions,
	 * we can just remove all the entries in the tdb and allow the
	 * close to remove the real locks.
	 */

	count = get_posix_lock_entries(fsp, &entries);

	if (count == 0) {
		DEBUG(10,("posix_locking_close_file: file %s has no outstanding locks.\n", fsp->fsp_name ));
		return;
	}

	for (i = 0; i < count; i++) {
		if (entries[i].fd != fsp->fh->fd )
			break;

		dump_entry(&entries[i]);
	}

	if (i == count) {
		/* All locks are ours. */
		DEBUG(10,("posix_locking_close_file: file %s has %u outstanding locks, but all on one fd.\n", 
			fsp->fsp_name, (unsigned int)count ));
		SAFE_FREE(entries);
		delete_posix_lock_entries(fsp);
		return;
	}

	/*
	 * Difficult case. We need to delete all our locks, whilst leaving
	 * all other POSIX locks in place.
	 */

	for (i = 0; i < count; i++) {
		struct posix_lock *pl = &entries[i];
		if (pl->fd == fsp->fh->fd)
			release_posix_lock(fsp, (SMB_BIG_UINT)pl->start, (SMB_BIG_UINT)pl->size );
	}
	SAFE_FREE(entries);
}

/*******************************************************************
 Create the in-memory POSIX lock databases.
********************************************************************/

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_log(NULL, 0, TDB_INTERNAL,
					  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_log(NULL, 0, TDB_INTERNAL,
						   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;
}

/*******************************************************************
 Delete the in-memory POSIX lock databases.
********************************************************************/

BOOL posix_locking_end(void)
{
    if (posix_lock_tdb && tdb_close(posix_lock_tdb) != 0)
		return False;
    if (posix_pending_close_tdb && tdb_close(posix_pending_close_tdb) != 0)
		return False;
	return True;
}