diff --git a/sys/ntfs/TODO b/sys/ntfs/TODO
new file mode 100644
index 000000000000..afc54e2433d2
--- /dev/null
+++ b/sys/ntfs/TODO
@@ -0,0 +1,38 @@
+NTFS - list of things to do in no particular order:
+
+- handle Unicode chars more inteligently - currently only iso-8859-1
+  chars are supported, it should be possible to get the filename
+  converted into koi-8, iso-8859-2 & utf-8 at least - there is
+  some Unicode support present under FreeBSD, would be worth considering
+  for adoption (it's needed to properly support Juliet CDs anyway)
+  FreeBSD info: http://triaez.kaisei.org/~mzaki/joliet/
+  * not very hard
+
+- better support for writing; tricky is crash recovery
+
+- cleanup the mess with several *readattr*() functions if possible
+
+- avoid allocation temporary buffer to hold result of read()/data
+  to be written in ntfs_read() & ntfs_write(): it's very easy
+  to panic the kernel that way, with simple
+	read(open(...), NULL, some_very_big_number);
+  The data should be transfered directly to the user buffer
+	(from Chuq)
+  * finishing testing, even gained about 3-5% performance improvement
+    for simple cat file >/dev/null test
+
+- ntfs_ntvattrget() identify if it's worth the efford to use some hashing
+  for the list of ntnodes, so that it's not all scanned each time; it would
+  be fine to migrate the found ntnode towards the start of the list too;
+  the list search code is duplicated on two places, eliminate the duplicity
+  somehow
+  
+- convert to code to do caching in buffer cache indexed by
+  file-vnode/file-offset so that NTFS would take advantage of UBC;
+  it should also improve performance somewhat, because if it's necessary
+  to find some offset in a file, it's possible to check if the data
+  are available in the cache before doing the bmap operation first
+	(from Chuq)
+
+- working VOP_BALLOC()
+	(from Chuq)